Learning in Carbon Accounting and Energy Efficiency

Learning in carbon accounting and energy efficiency

Cover design: Martin Weiss

Printing: Proefschriftmaken.nl

ISBN: 978-90-8891-130-9 Learning in carbon accounting and energy efficiency

Leren in het monitoren van koolstofstromen en energie-efficiëntie (met een samenvatting in het Nederlands)

Lernen auf den Gebieten der Kohlenstoffbilanzierung und Energieeffizienz (mit einer Zusammenfassung auf Deutsch)

Proefschrift ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van de rector magnificus, prof. dr. J.C. Stoof, ingevolge het besluit van het college voor promoties in het openbaar te verdedigen op 7 december 2009 des namiddags te 12.45 uur

door

Martin Weiss

geboren op 10 december 1976 te Suhl, Duitsland Promotor: Prof. dr. K. Blok Co-promotoren: Dr. M.K. Patel Dr. M. Junginger

Für meine Eltern Elke und Berthold

Contents

1 Introduction...... 1 1.1 Industrial manufacturing and energy consumption ...... 1 1.2 Non-energy use of fossil fuels and resulting emissions ...... 5 1.3 Technological learning in energy demand technologies ...... 8 1.4 Scope and outline of this thesis ...... 12

2 Applying bottom-up analysis to identify the system boundaries of non-energy use data in international energy statistics ...... 13 Abstract...... 13 2.1 Introduction ...... 14 2.2 Background information...... 15 2.3 Methodology and data sources ...... 19 2.4 Results ...... 24 2.5 Discussion...... 31 2.6 Conclusions ...... 34 Acknowledgements...... 35 Appendix...... 35

3 Non-energy use of fossil fuels and resulting carbon dioxide emissions: bottom-up estimates for the world as a whole and for major developing countries...... 39 Abstract...... 39 3.1 Introduction ...... 40 3.2 Methodology and data sources ...... 41 3.3 Results ...... 48 3.4 Discussion...... 55 3.5 Conclusions ...... 58 Acknowledgements...... 58 Appendix...... 59

4 Non-energy use and related carbon dioxide emission in Germany: a carbon flow analysis with the NEAT model for the period of 1990-2003 ...... 63 Abstract...... 63 4.1 Introduction ...... 64 4.2 Methodology and data sources ...... 66 4.3 Results ...... 71 4.4 Discussion...... 82

vii Contents

Conclusions...... 85 Acknowledgements...... 86 Appendix...... 87

5 Market diffusion, technological learning, and cost-benefit dynamics of condensing gas boilers in the Netherlands...... 91 Abstract...... 91 5.1 Introduction ...... 92 5.2 A short history of condensing gas boilers in the Netherlands ...... 93 5.3 Methodology and data sources ...... 95 5.4 Results ...... 99 5.5 Discussion...... 109 5.6 Conclusions ...... 112 Acknowledgements...... 113 Appendix...... 114

6 Analyzing price and efficiency dynamics of large appliances with the experience curve approach ...... 117 Abstract...... 117 6.1 Introduction ...... 118 6.2 Methodology and data sources ...... 120 6.3 Results ...... 124 6.4 Discussion...... 130 6.5 Conclusions ...... 137 Acknowledgements...... 138 Appendix...... 139

7 A review of experience curve analyses for energy demand technologies...... 141 Abstract...... 141 7.1 Introduction ...... 142 7.2 Methodology...... 143 7.3 Results ...... 146 7.4 Discussion...... 152 7.5 Conclusions ...... 161 Acknowledgements...... 162 Appendix...... 163

viii Contents

8 Discussion...... 167 8.1 Scope of this thesis ...... 167 8.2 Non-energy use of fossil fuels...... 167 8.3 Technological learning in energy demand technologies ...... 172

9 References...... 177

Summary...... 195 Samenvatting ...... 201 Zusammenfassung...... 207 List of abbreviations and units ...... 213 Acknowledgements ...... 215 Curriculum vitae...... 217

1 Introduction

1.1 Industrial manufacturing and energy consumption At the beginning of the twenty-first century, fossil fuels are still a main driver of the global economy. Whether one considers food production, industrial manufacturing, the provision of services, transportation, or private household consumption - the vast majority of economic activities relies either directly or indirectly on affordable energy, mainly derived from fossil resources such as coal, lignite, crude oil, or natural gas. Our dependency on fossil fuels can be traced back to the early nineteenth century when industrial development was first enabled and later accelerated by the availability of inexpensive fossil energy. Since then, fossil fuel use has increased more than 150-fold, rising from a share of 12% in the global total primary energy supply (TPES) in 1850 to a share of 81% in 2006 (Holdren, 2008; IEA, 2008a). Simultaneously, the global economy has grown exponentially, driven by a self-enhancing feedback cycle of cheap fossil energy, factor substitution, economies of scale, and technological progress (Ayres et al., 2003; Ayres and van den Bergh, 2005; de Vries, 2006).

Although this feedback cycle has contributed to a tremendous improvement of the quality of life in industrialized societies, the fossil fuel-based economy was associated from the beginning with major shortcomings. Fundamental problems in the pursuit of a sustainable energy system arise because fossil energy resources are depletable and unequally distributed around the globe. This leads to concerns about accessibility and security of energy supply (Fischedick et al., 2002; EU, 2006b). Extraction, conversion, and end-use of fossil energy causes human health problems and progressively deteriorates the environment on a local, regional, and global scale (Caldeira and Wickett, 2003; Kaiser, 2005; UNEP, 2007; Holdren, 2008). In particular, the incineration of fossil fuels is a major source for anthropogenic carbon dioxide (CO2) emissions, which in turn are regarded as main driver of global climate change (IPCC, 2007a). Atmospheric carbon dioxide concentration increased from pre-industrial levels of 280 ppm to 387 ppm in 2009 (NOAA, 2009). The average global surface temperature rose by 0.75 ± 0.20 oC in the past century and is expected to continue to increase by another 1.8-4.0 oC before the end of this century, resulting in sea level rise and an increased frequency of extreme weather events such as droughts, heavy precipitation, heat waves, and tropical cyclones (IPCC, 2007b).

The increasing concerns about anthropogenic climate change have triggered an international policy response. In 1988, the World Meteorological Organization (WMO) and the United Nations Environment Programme (UNEP) jointly established the Intergovernmental Panel on Climate Change (IPCC) to assess scientific knowledge about the physical basis, impacts, adaptation, and mitigation of climate change. The work of the IPCC contributed to the establishment of the United Nations Framework Convention on Climate Change (UNFCCC) in 1992. The UNFCCC is an international environmental treaty that aims at stabilizing greenhouse gas (GHG) concentrations in the atmosphere at

1 Chapter 1 levels that would prevent dangerous anthropogenic interference with the climate system (UN, 1992). As a milestone in international climate policy, the third Conference of Parties of the UNFCCC passed the Kyoto Protocol in 1997, which sets mandatory emission targets for industrialized countries for the period from 2008 to 2012. The Kyoto protocol entered into force in 2005; as of January 2009, 183 parties had ratified the protocol (UNFCCC, 2009).

These global policy initiatives are embedded in and supplemented by climate and energy policy on a regional and national level. The main pillar of the European Union’s climate policy is the climate-energy legislation package, which was adopted by the European Council in 2009 and defines the following targets for 2020: (i) to reach shares of 20% of renewable energy in the EU’s final energy consumption and 10% of renewable energy in transport energy consumption (EU, 2009a) (ii) to revise the current EU emissions trading system, which currently covers more than 10,000 installations in the energy and industry sector, thereby achieving emission reductions of 21% in the energy and manufacturing industry compared to 2005 (EU, 2008b, 2009b) (iii) to cut economy-wide greenhouse gas emissions unconditionally by at least 20% compared to 1990 levels and to commit to an emissions reduction target of at least 30%, if other industrialized countries sign comparable agreements (EU, 2009c) (iv) to cut GHG emissions of fuel supply by 6% over the entire life cycle of fuel products compared to 1990 (EU, 2009d)

The climate-energy legislation package is complemented by (i) the European Union’s action plan that aims at improving energy efficiency by 20% until 2020 (EU, 2006a) and (i) the European energy service directive that aims at a cost-effective improvement of end-use energy efficiency (EU, 2006c). Policies addressing specific areas of the energy-environment-economy problem include: (i) the mandatory energy labelling for energy consuming goods as it was introduced and successively expanded within the European Union since 1992 (EU, 1992) (ii) the EU directive on the landfill of waste (EU, 1999) (iii) the EU directive on the energy performance of buildings (EU, 2003a) (iv) the EU eco-design directive, which obliges manufacturers to decrease life cycle energy consumption and environmental impacts of their products already at the design stage (EU, 2005)

In the context of climate and energy policy, the manufacturing industry plays a dual role, namely (i) as a consumer of energy, and (ii) as a producer of energy conversion and end-use technologies. Industrial manufacturing itself uses one third of the global total primary energy supply (including feedstock use for chemicals) and is responsible for 40% of global fossil GHG emissions (IEA, 2008a, b, d, 2009). Furthermore, the manufacturing industry produces energy consuming equipment and technologies, which are employed in other sectors of the economy such as for energy conversion, transport, agriculture,

2 Introduction forestry, services, or in households. The application of energy consuming technologies in these sectors accounts together for the other two thirds of global total primary energy supply and fossil GHG emissions.

Industrial manufacturing offers considerable potentials for energy and emission savings. The International Energy Agency (IEA) estimates that, on a global scale, the manufacturing industry could already save today up to one quarter of its primary energy use and the related emissions by applying best practice technologies (IEA, 2008d). In manufacturing, the largest potentials for energy efficiency improvements can be realized in the bulk materials industry (i.e., in the production of iron and steel, cement, chemicals, as well as pulp and paper), which accounts for more than three quarters of industrial energy use (Bernstein et al., 2007; IEA, 2008d). However, even larger energy efficiency potentials are currently offered by energy consuming technologies, which are employed outside the manufacturing industry in other sectors of the economy (Blok and de Visser, 2005; Barker et al., 2007; IEA, 2008c, 2009). In the residential and commercial building sector, for example, more than one third of the current energy use and up to 85% of current GHG emissions could be saved until 2050 (IEA, 2008c). Technical energy saving potentials of new houses might exceed even 100%, if buildings function as net energy producers (Weizsäcker et al., 1997; Torcellini et al., 2006). Furthermore, the transportation sector offers potential fuel savings per kilometre of up to 50%, if manufacturers implement novel technologies and light-weight materials (IEA, 2008c).

The dual function of the manufacturing industry in the context of energy use and GHG emissions has been addressed by a large body of scientific work. Attention was, in particular, paid to product and material innovation (Lintona and Walsh, 2008; Baker, 2008), technological progress (e.g., IEA, 2000; Junginger et al., 2008), efficiency improvements (e.g., IEA, 2008c, d; Worrell et al., 2009), as well as the monitoring of energy use, CO2 emissions, and waste streams (e.g., Patel, 1999; IEA, 2008d; Neelis, 2008). However, to date, we still lack a detailed quantitative understanding of several important key elements of industrial manufacturing.

In the context of industry as an energy consumer, existing knowledge gaps refer to the accuracy of energy statistics and GHG emission inventories. Energy statistics form the basis for economy-wide monitoring of energy flows, energy efficiency improvements, GHG emissions, as well as uncountable research activities in related fields. However, so far, energy statistics have received very little attention in the scientific literature (Neelis, 2008). Limited timeliness, quality, and comparability of data in the energy statistics of many countries still pose an obstacle for developing policy-relevant energy indicators. The IEA (2008d) identifies a substantial need to increase efforts in collecting reliable energy data across countries and sectors. Greenhouse gas emission inventories face similar problems.

Although global and national emission inventories are subject to continuous improvement and expert review, crude assumptions, data gaps, and erroneous emissions accounting continue to introduce uncertainty into GHG emission estimates (UNFCCC, 2005c, 2006c; Olivier and Peters, 2002; Olivier et al., 2009). In the first part of this thesis,

3 Chapter 1 we address these knowledge gaps by analyzing fossil fuels, which are used for so called non-energy purposes (mainly in the chemical industry) and the resulting carbon dioxide emissions (Figure 1.1). We justify this choice as follows: first, the chemical industry is the largest and fastest growing sector within the manufacturing industry. Its absolute and relative importance is likely to continue to increase in the future (IEA, 2008c, d). Second, accounting for non-energy use and related emissions in energy statistics and GHG inventories has been and still is subject to major uncertainties (Olivier and Peters, 2002; Patel et al., 2005).

Related to the role of industry as a producer of energy consuming technologies, knowledge gaps exist regarding technological learning and more specifically regarding the rate at which costs decline for novel energy technologies. Studying potentials for cost decline is particularly important because high initial costs of novel energy technologies often present a barrier for successful market diffusion. The bulk of the literature has focused so far on quantifying cost decline for energy supply technologies, in particular for renewable energy supply technologies (Kahouli-Brahmi, 2008; Junginger et al., 2008; Neij, 2008). However, our knowledge regarding technological learning and cost decline in energy demand technologies is still limited, although demand-side energy efficiency improvements might present the quickest and cheapest means for reducing energy consumption and GHG emissions (Holdren, 2008; IEA, 2008c, 2009). In the second part of this thesis, we address this knowledge gap by analyzing technological learning of energy demand technologies (Figure 1.1).

Environment Emissions and waste Agriculture and Forestry

Construction

Services (2)(2) Energy demand technologies Electricity use

Fuel use Recycling for energy Manufacturing

Materials Fuel use for non- (1) energy purposes IndustryIndustry

Energy Private and commercial conversion consumption Disposal

Resource Anthroposphere extraction Emissions and waste

Figure 1.1: Simplified scheme of anthroposphere-environment interactions; the system boundaries of our analysis are indicated by dotted lines: (1) non-energy use of fossil fuels and related emissions, (2) technological learning in energy demand technologies

4 Introduction

Both research objectives are important elements in supporting climate and energy policies that aim to (i) reduce fossil energy use and (ii) mitigate anthropogenic GHG emissions. In the next two sections, we introduce first non-energy use of fossil fuels and then technological learning of energy demand technologies as two important aspects of industrial manufacturing.

1.2 Non-energy use of fossil fuels and resulting emissions Industry accounts for more than one third of global fossil fuel use (IEA, 2008a). Fossil fuels are, however, not only consumed for energy but also for so called non-energy purposes such as the production of bitumen, lubricants, or polymers. In this thesis, we refer to non-energy use as the parts of fossil fuels, which are used to manufacture materials. Non-energy use consists of two principle components: (i) the consumption of fossil fuels as feedstock in the chemical industry (e.g., the use of naphtha and ethane for olefins production in steam crackers or the consumption of natural gas for the production of ammonia) (ii) the consumption of refinery and coke oven products for non-energy purposes as well as the use of other solid carbon for the production of non-ferrous metals, ferroalloys, and inorganic chemicals (e.g., the use of lubricants and hydraulic oils, the use of bitumen in the building sector, or the consumption of electrodes for aluminium production)

We uniformly exclude the consumption of coke and coal for iron and steel making in blast furnaces from non-energy use. This approach follows the general practice in energy statistics, where coke and coal used for iron and steel making are accounted for under energy conversions.

On a global scale, the share of non-energy use in the total primary energy supply (TPES) has increased from 4.3% in 1973 to 6.3% in 2006, driven by a disproportionate growth of chemicals production (PE, 2007; IEA, 2008b). For countries with a relatively large chemical industry, these shares can be substantially higher. Examples are the Netherlands and Belgium where non-energy use accounted for 14.4% and 11.7% of the TPES in 2006, respectively (IEA, 2008b).

Accounting for non-energy use is complicated by the relatively complex inter- linkages of material and energy flows in the chemical industry. A particular problem arises because in chemical processes, parts of the feedstock are, strictly speaking, not used for non-energy purposes but to fuel processes (e.g., production of hydrogen via steam reforming for the synthesis of ammonia or production of olefins in steam crackers). This complicates the exact allocation of fossil energy carriers (e.g., naphtha, natural gas) to either energy or non-energy use. Non-energy use can therefore be defined in pure net terms that entirely exclude fuel use of feedstock (e.g., based on the calorific value of final products) or in pure gross terms that include fuel use of feedstock in non-energy use. An example is the allocation of natural gas and heavy fuel oil, which are partially oxidized to provide heat in the endothermic formation of hydrogen for the synthesis of ammonia,

5 Chapter 1 uniformly to either non-energy use or energy use. A partial net or gross definition of non- energy use is also possible in cases where the system boundaries of feedstock use versus fuel use are not uniformly defined for individual energy carriers (e.g., allocating the fuel use of natural gas in ammonia production to energy use but accounting the use of heavy fuel oil for the same purpose under non-energy use).

Parts of non-energy use carbon remain stored in products with lifetimes ranging up to several decades. However, other parts are also emitted via the following pathways: (i) oxidation of feedstock parts in the production of basic chemicals (e.g., during steam reforming of natural gas for ammonia production) (ii) oxidation of parts of basic and intermediate chemicals during chemical conversion processes (e.g., oxidation of parts of ethylene and propylene during the production of ethylene oxide and acrylonitrile) (iii) flaring of non-specified chemical by-products within the chemical industry (iv) oxidation of electrodes, coke, and coal during the production of electric arc furnace (EAF) steel as well as non-ferrous metals, ferroalloys, and inorganic chemicals (v) oxidation and partial oxidation of fossil carbon contained in solvents, lubricants, pesticides, urea fertilizers, and other substances during product use (vi) oxidation of surfactants and other fossil carbon containing substances during wastewater treatment (vii) oxidation of fossil carbon during waste incineration

We uniformly exclude emissions from waste incineration from our non-energy use emission estimates because the former are generally regarded as energy use emissions in GHG inventories (IPCC, 2006). Excluding these carbon quantities from non-energy use emissions prevents potential double counting of emissions resulting from waste treatment.

Non-energy use emissions consist almost entirely of CO2, whereas smaller fractions resulting from product use and wastewater treatment are also composed of methane and non-methane volatile organic compounds (NMVOCs). Non-energy use emissions are growing in both absolute and relative terms, but still constitute only a relatively small fraction of total fossil fuel use emissions. Globally, the share of non- energy use emissions in total fossil fuel use emissions increased from 1% in 1970 to 3% in 2005. In the context of the manufacturing industry, non-energy use accounted for about 4% (1970) and 13% (2005) of global industrial fuel use emissions, respectively (PBL, 2009; IEA, 2008f). Due to the relatively small amounts, non-energy use emission accounting received far less attention than the accounting of emissions from energy use (Patel et al., 2005).

In 2006, the IPCC published updated guidelines to improve the accuracy and consistency of international GHG emissions accounting. In contrast to the previous 1996 guidelines (IPCC, 1997), the new guidelines quantify non-energy use emissions only by the detailed IPCC-Sectoral Approach (IPCC-SA). The IPCC-SA is a bottom-up method for calculating non-energy use emissions based on three methodologies (Tier 1-Tier 3),

6 Introduction which differ regarding their level of detail and accuracy. At the least accurate level (Tier 1), emissions are calculated by multiplying activity data with average IPCC default emission factors. With Tier 2, emissions are estimated based on country-specific emission factors that replace the IPCC default values. According to Tier 3, either detailed emission models or direct emission measurements at individual plants should be used to generate emission estimates (IPCC, 2006). Complete emissions accounting under the IPCC-SA requires that non-energy use emissions of the various activities are reported under several different emission source categories, i.e., industrial processes, product use, agriculture, and waste (IPCC, 2006). Among all activities, industrial processes are by far the most dominant source of non-energy use emissions. Currently, countries make use of their right to continue to use the old 1996 IPCC guidelines until the end of the Kyoto-period in 2012 (UBA, 2006). The 1996 guidelines provide a mechanism for crosschecks by accounting for non-energy use emissions with a relatively simple top-down methodology, i.e., the IPCC-Reference Approach (IPCC-RA). In this approach, data from energy statistics are multiplied by fuel-specific carbon storage fractions to arrive at estimates of domestic non- energy use emissions (IPCC, 1997). According to the new 2006 guidelines, such crosschecks are no longer possible because carbon that does not lead to fuel use emissions is systematically excluded from the IPCC-RA (IPCC, 2006).

The accounting of non-energy use in energy statistics and of the related emissions according to IPCC-SA and IPCC-RA is complicated and associated with major uncertainties. Several in-depth studies found errors and inconsistencies in the system boundaries of non-energy use data as stated by Dutch, South Korean, and German energy statistics (Neelis et al., 2005a; Park, 2005; Patel et al., 2005). In addition, errors and data gaps were identified in the national GHG inventories of the Netherlands, South Korea, Germany, and Italy (Neelis et al., 2005a; Park, 2005; UBA, 2003; La Motta et al., 2005). In particular, the application of the IPCC-SA for non-energy use emissions accounting often suffers from incompleteness as a consequence of the multitude of relevant emission sources. Inaccuracies result also from the application of process-specific IPCC default emission factors, which might not always correctly represent the situation in individual countries. Losing the ability to check non-energy use emissions as stated by the IPCC-SA with estimates from the IPCC-RA introduces additional uncertainty into non-energy use emission estimates once the new 2006 guidelines are applied by the UNFCCC parties in the post-Kyoto period after 2012.

The problems associated with the accounting of non-energy use and related emissions demand in-depth research. To address existing knowledge gaps, scientists, energy statisticians, greenhouse gas inventory makers, and experts from the chemical and petrochemical industry established the non-energy use and CO2 emissions (NEU-CO2) network in 1999 (Patel, 2005). Our research on non-energy use was performed within the activities of this network. In view of the current accounting of non-energy use and related emissions, we address in the first part of this thesis the following research question: How can we improve accuracy and completeness in the accounting of emissions caused by the non-energy use of fossil fuels? In Chapters 2-4, we separately address knowledge gaps that are directly related to this research question by:

7 Chapter 1

(i) scrutinizing non-energy use data in international energy statistics, thereby contributing to a more accurate accounting of non-energy use (ii) providing the first detailed estimates of non-energy use emissions for the world as a whole and for major developing countries, thereby contributing to the establishment of reliable GHG inventories in these countries (iii) providing, as a case study, detailed estimates for non-energy use and resulting emissions of Germany during the period from 1990 to 2003, thereby improving the methodology and accuracy of the German GHG inventory

Our research on non-energy use addresses the manufacturing industry as consumer of energy resources. In the next section, we take a different perspective and focus on the manufacturing industry as a producer of energy consuming technologies.

1.3 Technological learning in energy demand technologies The second part of this thesis focuses on energy demand technologies, which we define as technologies that serve the primary purpose of end-use energy conversion. Energy demand technologies are principally used for transportation (e.g., internal combustion engines in cars and trucks) or to provide services such as space and water heating, lighting, managing information, or entertaining. Among all sectors of the economy, the residential and commercial building sector offers the largest potentials for energy efficiency improvements and GHG emission reductions (Blok and de Visser, 2005; Barker et al., 2007; IEA, 2008d, e). Whether these potentials will be exploited depends on (i) the enhanced market diffusion of already existing technologies and (ii) the market introduction of novel energy demand technologies. However, consumer investments into novel and efficient energy demand technologies are often discouraged by high initial costs, insufficient information, principal-agent dilemmas, or other market distortions in favor of incumbent technologies. In particular, high initial investment costs often present a key barrier for the successful diffusion of novel and efficient energy demand technologies. In view of the manufacturing industry, the question is how fast producers can decrease manufacturing costs of novel and efficient energy demand technologies. A rapid decline of costs potentially speeds up market implementation and thus demand-side energy efficiency improvements.

For the manufacturing industry as well as for policy makers and scientists, it is of interest to obtain quantitative insight into the prospects for cost decline to analyze potentials of technologies and to identify the possible benefits of policy intervention. Cost decline in manufacturing typically occurs due to various learning mechanisms (e.g., learning by doing), economies of scale, factor substitution, process innovation, and indirect effects from other sectors of the economy (Arrow, 1962; Dutton and Thomas, 1984; Argote and Epple, 1990). The combination of several or all of these factors is generally referred to as technological learning. In this thesis, we define technological learning as the combination of all mechanisms and effects that lead to a systematic decline of production costs and improvement of performance of products and services. Technological learning is the result of factors interacting with each other at various levels

8 Introduction of the economy. At the macro level, technological learning is both the driver and the result of increasing productivity (Figure 1.2).

Market competition Increasing spendable income

Declining unit price

Declining unit Increasing Increasing costs productivity consumer demand

Technological learning: - Learning mechanisms - Economies of scale - Substitution of production factors - Indirect effects from other sectors

Technological spillovers Availability of from other sectors production factors Figure 1.2: Technological learning in a competitive market economy (adapted from: Ayres et al., 2003)

Focusing on the level of individual manufacturers, technological learning is the outcome of complex multi-level interactions of factors, which are both endogenous and exogenous to the manufacturing process. Of major importance for technological learning are various learning mechanisms: (i) learning by searching - technology and process improvements occur mainly due to research and development; the most important mechanism in the stage of technology invention and innovation (Kouvaritakis et al., 2000) (ii) learning by doing - technology and process improvements occur in the course of the manufacturing; an important mechanism in the stages of innovation and pervasive market diffusion (Arrow, 1962; Schoots et. al., 2008) (iii) learning by using - technology improvements occurs due to product use; important mechanisms after niche market introduction of technologies, which builds upon feedback between users and producers of technologies (Rosenberg, 1982) (iv) learning by interacting - network interactions between producers, suppliers, and customers improve the communication of knowledge and is important in all stages of product life cycle (Kamp, 2002; Lundvall, 1988)

Technological learning and its effect on production costs and technology improvement are important along the entire life cycle of products. Approaches to quantify the effects of technological learning date back to the 1930s, when Wright (1936) found that unit labor costs in airframe manufacturing decline at a constant rate with each doubling of cumulative production. He noted the particular relevance of his finding for the investigation of future cost developments in manufacturing. The graphical representation of Wright’s discovery is nowadays referred to as learning curve, which applies to the

9 Chapter 1 effects of learning-by-doing, i.e., the decline in labor costs due to a decrease of working time requirements for manufacturing. Arrow (1962) introduced the notion that declining labor costs are a result of growing experience. The Boston Consulting Group extended the analysis by modeling the dynamics of total production costs as a function of cumulative production, which serves as proxy for technological learning (BCG, 1972). The modeling of total production costs as a black-box function of cumulative production is generally referred to as the experience curve approach. Dutton and Thomas (1984) differentiate experience curves and progress curves; the former represent average production costs of multiple manufacturers, whereas the latter represent production costs at the level of individual firms. Here, we focus on industry-wide average prices and costs of technologies. Hence, we uniformly use the term experience curve throughout this thesis.

The experience curve approach is an empirical concept that models production costs of a technology as a power-law function of cumulative production. The approach thereby assumes that production costs decline at a constant rate with each doubling of cumulative production, resulting in a linear relationship, if both parameters are plotted on logarithmic scales (Figure 1.3).

80 60

40 First doubling: -19% /klm 30 Second doubling: -19% 2006 Third doubling: -19% 20 Fourth doubling: -19%

10 8 6

2 4 R = 0.91 3 PR = 81 ± 4% Average CFL price in EUR LR = 19 ± 4% 2 Period: 1988-2006 Price = (178 ± 40)×Cum. production(−0.30±0.04)

100200 500 1000 2000 5000 10000 Cumulative worldwide CFL production in Glm Figure 1.3: Experience curve for compact fluorescent light bulbs (CFLs) showing a cost decline of 19 ± 4% with each doubling of cumulative production; prices in real Euro deflated to the base year of 2006 [EUR2006] per kilolumen [klm]; cumulative worldwide CFL production in gigalumen [Glm]; error bars indicate the standard deviation of price data in each year (adapted from: Weiss et al., 2008c)

In experience curve analysis, cost decline is typically quantified either by the progress ratio (PR) or the learning rate (LR). The progress ratio indicates the percentage of initial costs after one doubling of cumulative production; the learning rate specifies the

10 Introduction percentage at which costs decline after each doubling of cumulative production. Compact fluorescent light bulbs in Figure 1.3 show a cost decline of 19 ± 4% with each doubling of cumulative production; the progress ratio is hence 81 ± 4% and the learning rate 19 ± 4%. In reality, learning rates show a wide spectrum, frequently occurring in the interval of 10-30% (Dutton and Thomas, 1984; Junginger et al., 2008).

In the 1970s and 1980s, experience curves were primarily applied as a management tool and for strategic planning in manufacturing because they allow tracing cost decline closer to its drivers than simple time-series or static bottom-up engineering technology analysis (Dutton and Thomas, 1984; Argote and Epple, 1990). Since the early 1990s, experience curves gained importance as an instrument to forecast costs and diffusion rates of energy supply technologies in energy and CO2 emission scenarios (Wene et al., 2000; IEA, 2000; Kahouli-Brahmi, 2008). In particular, experience curves have been extensively applied to renewable and non-renewable energy supply technologies (Junginger et al., 2004, 2005, 2006, 2008; McDonald and Schrattenholzer, 2001; Neij, 2008; Nemet, 2009). In the context of renewable energy technologies, experience curves were and still are used in particular to estimate learning investments, i.e., investment costs that are necessary to reach cost-break-even with established technologies as well as cumulative break-even production at which cost break-even will be reached. The time period necessary to reach the break-even point depends on the deployment rate of a technology and can be actively influenced by policy measures.

In the past decade, several attempts were made to conceptually extend the experience curve approach. These include the modeling of specific energy consumption (i) in the manufacturing of ammonia and urea (Ramírez and Worrell, 2006) as well as (ii) in the production of ethanol from corn (Hettinga et al., 2009) as a function of cumulative production. Klaassen et al. (2005) and Jamasb (2007) model costs as a function of both cumulative production and cumulative spending for research and development (R&D) by so-called two-factor experience curves. Junginger et al. (2004) and Ferioli et al. (2009) regard technology costs as an aggregate of individual cost components. Such an approach allows for differentiating learning rates for individual cost components.

Although the conventional experience curve approach has been frequently applied to energy supply technologies, its application to energy demand technologies has been far less common. Today, we therefore lack a detailed quantitative understanding of technological learning and the rate of cost decline of efficient energy demand technologies. In view of this knowledge gap, we address in the second part of this thesis the following research question: At which rates does technological learning occur for energy demand technologies? In Chapters 5-7, we address this research question by: (i) applying the experience curve approach to condensing gas boilers and large appliances, thereby quantifying technological learning for these energy demand technologies (ii) identifying average learning rates and the distribution of learning rates for a wider range of energy demand technologies, thereby exploring the potentials for and limitations of devising technology-specific learning rates for energy policy and in energy-economy-environment models

11 Chapter 1

1.4 Scope and outline of this thesis In the next three chapters, we focus our research primarily on the chemical industry for which we develop and apply models to estimate non-energy use of fossil fuels and resulting emissions. In Chapter 2, we develop a simple bottom-up model to quantify total non-energy use and major components thereof. We apply this model to crosscheck official non-energy use data as published in international energy statistics (IEA, 2005a, b) for the world as a whole and for the 50 countries with the highest consumption of fossil fuels for non-energy purposes.

In Chapter 3, we develop and apply an extended model to estimate emissions related to non-energy use. We apply our model to the world as a whole, to the total of Annex I and non-Annex I countries, as well as to major developing countries. Chapter 4 provides an in-depth analysis of non-energy use. Here, we estimate in a case study non- energy use and related CO2 emissions of Germany during the period from 1990 to 2003. We extend the original NEAT model (Neelis et al., 2005b) by modules that allow us to estimate emissions from (i) conversion losses in the chemical industry, (ii) product use, (iii) agriculture, and (iv) waste.

In the following three chapters, we shift our analysis and focus on the manufacturing industry as a producer of energy consuming technologies. Chapter 5 addresses technological learning of condensing gas boilers in the Netherlands. We first provide a short overview of the market diffusion of this energy-efficient heating technology. Afterwards, we construct experience curves to quantify technological learning and the rate of price decline. In addition, we analyze costs and benefits of condensing gas boilers from both a consumer and a governmental perspective. In particular, we are interested in quantifying the importance of technological learning as a driver for market diffusion and increasing consumer benefits.

In Chapter 6, we apply the experience curve approach to analyze long-term price and energy efficiency dynamics of three wet appliances (washing machines, laundry dryers, and dishwashers) and two cold appliances (refrigerators and freezers). Here, we extend the conventional experience curve approach by analyzing the dynamics of specific energy consumption as a function of cumulative production. Drawing on the insight gained in the previous two chapters, we provide in Chapter 7 a literature review of experience curve studies on energy demand technologies. We estimate average as well as technology-specific learning rates, and we compare our findings with results for energy supply technologies and manufacturing in general. One important motivation for this research is to identify potentials and limitations of devising technology-specific ex-ante learning rates for novel energy demand technologies, which are not yet available on the market.

In Chapter 8, we discuss the strengths and limitations of our research, subjects of future research, as well as implications of our findings for energy statisticians, inventory experts, scientists, and policy makers.

2 Applying bottom-up analysis to identify the system boundaries of non-energy use data in international energy statistics

Martin Weiss, Maarten L. Neelis, Matthijs Zuidberg, and Martin K. Patel Published slightly adapted in: Energy 33 (2008), pp. 1609-1622.

Abstract Data on the non-energy use of fossil fuels in energy statistics are subject to major uncertainties. Here, we apply a simple bottom-up methodology to recalculate non-energy use for the world as a whole and for the 50 countries with the highest consumption of fossil fuels for non-energy purposes. We quantify worldwide non-energy use in the year 2000 to be 20 ± 2 EJ (exajoules), thereby accounting for 5% of the global total primary energy supply. Our bottom-up estimates are in line with data from international energy statistics for 12 individual countries. Our estimates are lower than official data in the case of 26 countries and for the world as a whole, whereas they exceed official non-energy use data for another 12 countries. The partial or complete inclusion of fuel use of feedstock into non-energy use data as reported by international energy statistics can explain parts of the observed deviations. We regard our bottom-up methodology as reliable, albeit being subject to uncertainties. We recommend that energy statisticians use our method to generate a shortlist of countries, for which efforts should be made to clarify and improve the quality of non-energy use data in national and international energy statistics.

13 Chapter 2

2.1 Introduction Energy statistics are the single most important data source for national greenhouse gas (GHG) inventories. Reliable data reporting in energy statistics is hence of vital importance for the correct accounting of anthropogenic GHG emissions. One element in both national and international energy statistics that is subject to major uncertainties is the non-energy use of fossil fuels, i.e., the use of feedstock in the chemical industry and the consumption of refinery and coke oven products for non-energy purposes. Non-energy use is a source of CO2 emissions that has been increasing substantially in the past three decades (Patel et al., 2005; IEA 2005a, b). However, as a consequence of the relatively complex inter-linkages between material and energy flows in refineries and in the chemical industry, non-energy use data as stated in energy statistics often suffer from errors and inconsistencies in the definition of system boundaries1. Patel et al. (2005) identified considerable deviations regarding the system boundaries of non-energy use when summarizing detailed country studies for Italy, Japan, South Korea, the Netherlands, and the USA. Inconsistencies and errors in non-energy use data were also found within the national energy statistics of Germany (Weiss et al., 2008b).

According to the Intergovernmental Panel on Climate Change, emissions from non-energy use of fossil fuels are calculated within national GHG inventories by two independent methods: (i) the relatively detailed bottom-up IPCC-Sectoral Approach (IPCC-SA) and (ii) the simple top-down IPCC-Reference Approach (IPCC-RA)2 (IPCC, 1997). The general difficulties associated with the calculation of non-energy use and related CO2 emissions were and still are reflected by the national GHG inventories of, e.g., Japan, Italy, the Netherlands, and Germany, where estimates of non-energy use emissions were found to be erroneous and incomplete (Gielen and Yagita, 2002; La Motta et al., 2005; Neelis et al, 2005a; Weiss et al, 2008b). The observed errors can lead to false conclusions when evaluating both the effectiveness of national GHG emission mitigation policies and the compliance with emission targets as agreed upon under the Kyoto protocol.

To analyse non-energy use and related emissions in detail, the non-energy use and 3 CO2 emissions (NEU-CO2) network was established . Since 1999, the network has provided substantial contributions to a better understanding of non-energy use and related CO2 emissions (Patel et al., 2005; Neelis et al, 2005b). However, uncertainties and errors in official non-energy use data have been quantified so far only by a few detailed country analyses (La Motta et al., 2005; Neelis, 2005a; Park, 2005; Weiss et al., 2008b).

1 We use the term system boundary here to distinguish fossil fuels that are consumed for non-energy purposes (these are within the boundaries of the system) from the ones that are used for energy (these are outside of the boundary of the system; see dotted line in Figure 2.3). We discuss the problems related to the definition of non-energy use, i.e., to the definition of its system boundary, in Section 2.2. 2 For a more detailed discussion of the accounting of non-energy use emissions with the IPCC-SA and IPCC-RA, we refer the reader to the next section. 3 The NEU-CO2 network is funded by the European Commission, Directorate General for Research, under the ENRICH (European Network for Research into Global Change) program. It brings together scientists, greenhouse gas inventory makers, and experts from the chemical industry to gain a deeper insight into the non-energy use of fossil fuels and the associated CO2 emissions (Patel, 2005).

14 Non-energy use in international energy statistics

This challenged us to develop a simple and easily applicable bottom-up methodology4 that allows for crosschecking non-energy use data as stated by national and international energy statistics for any country in the world.

In this chapter, we apply such a bottom-up methodology for the year 2000 to the world as a whole and to the 50 countries with the highest consumption of fossil fuels for non-energy purposes5. The primary goal of our research is to assess the quality of international energy statistics by reproducing non-energy use data with a simple methodology that can easily be applied by international energy statisticians and inventory experts without the need for in-depth data research. We thereby aim at (i) obtaining insight into the system boundaries of official non-energy use data and (ii) identifying obvious errors within international energy statistics as published by the International Energy Agency (IEA, 2005a, b). With our results, we intend to support energy statisticians and GHG inventory makers at both national and international level in evaluating the reliability of their data and in generating a shortlist of countries for which additional efforts should be made to improve the quality of non-energy use estimates. We furthermore regard our research as contribution to an international harmonization of the system boundaries of non-energy use data.

The chapter continues with background information and a short explanation of relevant terminology (Section 2.2). In Section 2.3, we describe methodology, assumptions, and data sources used for our analysis. We present the results of our bottom-up methodology in Section 2.4. Afterward, we discuss methodology and results and we provide recommendations for energy statisticians (Section 2.5). The chapter ends with general conclusions in Section 2.6.

2.2 Background information We define non-energy use according to IEA (2005a, b) as the sum of two components: (i) the consumption of fossil fuels as feedstock in the chemical industry (e.g., the use of naphtha and ethane for olefins production in steam crackers or the consumption of natural gas for the production of ammonia) (ii) the consumption of refinery and coke oven products for non-energy purposes as well as the use of other solid carbon for the production of non-ferrous metals, ferroalloys, and inorganic chemicals (e.g., the use of lubricants and hydraulic oils, the use of bitumen in the building sector, or the consumption of electrodes for aluminium production)

4 We define bottom-up methodology as a methodology that aims at quantifying features of a system (e.g., its energy use) based on estimates of its individual components. The complementary approach to a bottom-up methodology is the top-down methodology, which quantifies the components of a system based on aggregated data of the entire system. 5 Although system boundaries of non-energy use data in international energy statistics might not be consistent (IEA, 2005a, b), we base the selection of countries on this source as it provides the only complete worldwide overview of non-energy use data for individual countries.

15 Chapter 2

The worldwide share of non-energy use as a percentage of total primary energy supply (TPES) has grown from 4.3% in 1971 to 5.8% in 2000 (IEA, 2005a). Both, the share of non-energy use in the TPES as well as the shares of the two principal components in the total non-energy use vary greatly between individual countries (Figures 2.1 and 2.2). For countries with a large chemical industry, non-energy use is more important than indicated by the world’s average. Examples are the Netherlands and Belgium where non- energy use accounts for 13.5% and 10.2% of the TPES in the year 2000, respectively (IEA, 2005b).

Kuwait Bulgaria Year 1971 Sweden Year 2000 Austria Hungary Slovak Republic Israel Colombia Algeria Czech Republic Romania Norway Malaysia Belarus Uzbekistan Venezuela Portugal Qatar New Zealand Pakistan Libya Trinidad and Tobago Egypt Singapore Turkey Thailand Australia Poland Argentina Iran South Africa Belgium Indonesia Mexico China (Taiwan) Italy Spain Saudi Arabia The Netherlands United Kingdom Brazil India France Canada Russia Germany South Korea Japan China (excl. Taiwan) USA World

010205051525 55 Share of non-energy use on the TPES in % Figure 2.1: Share of non-energy use in the total primary energy supply for the world as a whole and for 50 selected countries; in ascending order of magnitude of non- energy use in the years 1971 and 2000 (data sources: IEA, 2005a, b)

16 Non-energy use in international energy statistics

Kuwait Bulgaria Sweden Austria Hungary Slovak Republic Israel Colombia Algeria Czech Republic Romania Norway Malaysia Belarus Uzbekistan Venezuela Portugal Qatar New Zealand Pakistan Libya Trinidad and Tobago Egypt Singapore Turkey Thailand Australia Poland Argentina Iran South Africa Belgium Indonesia Mexico China (Taiwan) Italy Spain Saudi Arabia The Netherlands United Kingdom Brazil India France Canada Russia Germany South Korea Japan China (excl. Taiwan) USA World

0 20406080100 Share of principal components on non-energy use in %

Feedstock use Non-energy use of refinery products, coke oven products, and other solid carbon

Figure 2.2: Share of feedstock use and non-energy use refinery products in the total non- energy use for the world as a whole and for 50 selected countries; in ascending order of magnitude of non-energy use in the year 2000 (data sources: IEA, 2005a, b)

17 Chapter 2

One part of the carbon that is contained in non-energy use remains stored in products with life times ranging up to several decades. However, the other part becomes oxidized and subsequently emitted to the atmosphere. Non-energy use of fossil fuels leads to CO2 emissions via several pathways: (i) feedstock oxidation in industrial processes (e.g., during steam cracking or ammonia production) (ii) oxidation of parts of basic and intermediate chemicals during chemical conversion processes (e.g., oxidation of parts of propylene during the production of acrylonitrile) (iii) use of solvents, lubricants, and other products that partially or fully oxidize during their use phase (iv) application of pesticides and urea fertilizers in agriculture (v) wastewater treatment (e.g., oxidation of surfactants and other fossil carbon contained in wastewater) (vi) incineration of products at the end of their life cycle or flaring of non- specified chemical by-products within the chemical industry6

According to IPCC-RA estimates, non-energy use emissions account in 2000 on average for roughly 2% of the total fossil-based CO2 emissions in Annex-I countries (UNFCCC, 2006a). Again, this share can be substantially higher for countries with a large chemical industry. The accounting for both non-energy use and related CO2 emissions is complicated by relatively complex inter-linkages of material and energy flows within the chemical industry (Figure 2.3). In industrial processes (e.g., production of olefins in steam crackers), parts of the total input of fossil energy carriers is not used as feedstock but as fuel to sustain endothermic reactions. This complicates the exact allocation of fossil energy carriers (e.g., naphtha, natural gas) to either non-energy use or energy use. The definition of non-energy use can, therefore, vary between a pure net definition that entirely excludes fuel use of feedstock and a pure gross definition that includes the fuel use of feedstock into non-energy use data. We illustrate the problem related to the definition of non-energy use with two examples (see also Figure 2.3): (i) The internal use of steam cracking products (e.g., mainly hydrogen and methane) for fuel purposes can either be included in or excluded from official non-energy use data. (ii) For ammonia production, natural gas is first converted to hydrogen and carbon monoxide via endothermic steam reforming. The necessary heat of reaction is supplied by burning parts of the natural gas feedstock. The fraction of natural gas used for fuel purposes is roughly 30% (Neelis et al., 2005b) and can be either reported as energy or as non-energy use.

6 Emissions from incineration with energy recovery are included under the source category of energy in GHG inventories and are thus excluded from the accounting of non-energy use emissions (IPCC, 1997). We refer here to waste incineration as well as flaring of off-gases without energy recovery.

18 Non-energy use in international energy statistics

Energy production Coal, Lignite Coke oven Reducing agents for the production of non-ferrous metals, ferroalloys, and Tars, Pitch inorganic chemicals

Production of non-ferrous Electrodes metals, ferroalloys, and inorganic chemicals

Aromatics Petroleum cokes Pygas Import Ethane Organic Ethylene chemicals Naphtha Steam Crude oil production Refinery Propylene Lubricants cracker Other C Export 4 Bitumen Internal fuel use

Liquid petroleum gas (LPG)

Heavy oils

Gasoline, Kerosene, Diesel

Natural gas Energy production

Figure 2.3: Schematic overview: the consumption of fossil fuels and the system boundary of non-energy use according to a gross definition (dashed bold line)

To compile international energy statistics, the IEA asks the recipients of its questionnaires to apply a net definition for non-energy use (IEA, 2006)7. However, only very limited guidance is provided by the IEA on the level of individual chemical and refinery processes to assure compliance with the proposed system boundary. Given the complex material and energy flows within the chemical and refinery industry, we would, therefore, expect that official non-energy use data are not per se harmonized with respect to their system boundaries. The complex streams of fuels and chemicals within the petrochemical industry, furthermore, raise additional questions about the correct accounting (of all individual components) of non-energy use by national and international energy statistics.

2.3 Methodology and data sources We apply a bottom-up methodology to estimate non-energy use independently from energy statistics as the sum of two components:

= + NEU i NEU F ,i NEU R,i (2.1)

7 The IEA (2006) specifies in its oil questionnaire: “Report quantities of oil used in the petrochemical sector for the purpose of producing ethylene, propylene, butene, synthesis gas, aromatics, butadiene and other hydrocarbon-based raw materials in processes such as steam cracking, aromatics plants and steam reforming. Exclude amounts of oil used for fuel purposes.”

19 Chapter 2

where NEUi [PJ] represents non-energy use, NEUF,i [PJ] stands for the consumption of feedstock in the chemical industry, and NEUR,i [PJ] for the consumption of non-energy use refinery products, coke oven products, and other solid carbon in country i. We estimate the consumption of feedstock use as the sum of its four elements:

= + + + NEU F ,i FS ,i FA,i FM ,i FC,i (2.2) where FS,i [PJ] stands for the feedstock consumption in steam crackers and FA,i [PJ], FM,i [PJ], and FC,i [PJ] for the feedstock consumption during the production of ammonia, methanol, and carbon black, respectively. We estimate the consumption of non-energy use refinery products, coke oven products, and other solid carbon by calculating the sum of four activities:

= + + + NEU R,i C A,i CE,i CB,i CL,i (2.3) where CA,i [PJ] represents the consumption of refinery aromatics for non-energy purposes, CE,i [PJ] stands for the consumption of electrodes and other solid carbon in the 8 manufacturing of non-ferrous metals, ferroalloy, and inorganic chemicals , and CB,i [PJ] and CL,i [PJ] for the consumption of bitumen and lubricants, respectively.

We now give more detailed explanations on the estimation of individual components of non-energy use. We estimate feedstock consumption in steam crackers (FS,i) based on ethylene production (Table 2.1). Unlike for other chemical processes, we differentiate various types of feedstock by using information as given by the Oil and Gas Journal (2001). In cases where no specific information on feedstock consumption is given, we assume naphtha to be used as feedstock for ethylene production. Exceptions to this are Canada and the USA, for which we assume ethane to be used as feedstock because (unlike in Western Europe) ethane is the most important feedstock for steam cracking in North America (OGJ, 2001; Weissermel and Arpe, 2003).

We calculate feedstock requirements for steam cracking and for all other chemical processes by multiplying production data with process-specific feedstock requirements (Table 2.1). We uniformly exclude backflows from steam crackers to refineries because these are generally not part of non-energy use in international energy statistics (IEA, 2005c). We differentiate between net and gross feedstock requirements, i.e., we either exclude (net feedstock requirements) or include (gross feedstock requirements) feedstock shares that are used for energy purposes in the various chemical processes. We furthermore estimate error intervals, which represent the 95% confidence interval of the assumed average process-specific gross feedstock requirements (see Table 2.1). We approximate actual chemicals production (i.e., production of ethylene, ammonia, methanol, and carbon black) primarily by production capacities, assuming an average

8 We exclude here the consumption of coke and coal for iron and steel production because these items are generally accounted for under energy conversions in energy statistics. The consumption of coke and coal in the iron and steel industry is therefore not part of non-energy use.

20 Non-energy use in international energy statistics capacity utilization of 90 ± 10% (Neelis et al., 2005a; Weiss et al., 2008b; OGJ, 2001; CW, 2000, 2001, 2005).

We approximate parts of the consumption of refinery products (NEUR,i) for non- energy purposes, i.e., the use of aromatics, lubricants, and bitumen, by using openly available data, which generally refer to production capacities. We assume an average capacity utilization rate of 90 ± 10% based on information as given by OGJ (1999).

Electrodes and other solid carbon are used as reducing agents for the production of a wide variety of non-ferrous metals, ferroalloys, and inorganic chemicals. A substantial part of electrodes is consumed in aluminium production for which production data are generally available. For the bulk of other non-ferrous metals, ferroalloys, and inorganic chemicals, production data are, however, often unavailable. We, therefore, make use of two considerations: (i) The availability of cheap electricity is one important precondition for both aluminium production and the manufacturing of other non-ferrous metals, ferroalloys, and inorganic chemicals. If a country produces aluminium, it can hence be expected that it also produces other non-ferrous metals, ferroalloys, and inorganic chemicals. (ii) Countries with a large absolute gross domestic product (GDP) are likely to have a larger non-ferrous metals industry than countries with a small GDP.

Based on these considerations, we would expect a positive relationship between electrodes consumption for aluminium production, absolute GDP, and the consumption of electrodes and other solid carbon sources for the total production of non-ferrous metals, ferroalloys, and inorganic chemicals. We hence estimate the total consumption of electrodes and other solid carbon sources in the production of non-ferrous metals, ferroalloys, and other inorganic chemicals (CE,i) based on multiple regression analysis as it was performed by Weiss et al. (2009a)9:

= ± × + ± × CE,i (1.7 0.4) C Al,i (1.2 0.3) GDPi (2.4) where CAl,i [PJ] stands for the consumption of electrodes in aluminium production and 12 GDPi [10 purchasing power-corrected constant 1995 $] for the absolute GDP in country i.

9 The multiple regression analysis as it was conducted by Weiss et al. (2009a) results in a coefficient of determination of R2 = 0.86 and includes the following countries and years (in parentheses): Australia (2000), Austria (2000), Belgium (2001), Denmark (2000), Finland (1997), France (2000), Germany (2000), Greece (2000), Iceland (1997), India (2000), Italy (2001), Korea (2000), The Netherlands (1999), Norway (1997), Poland (2003), Russia (2000), South Africa (2000), Spain (1997), Sweden (1997), United Kingdom (2001), and USA (2003).

22 Chapter 2

Table 2.1: Data sources and assumptions used in our bottom-up methodology for calculating non-energy use in the year 2000 Input data used to Principal Individual Capacity Feedstock for the Gross feedstock Net feedstock estimate individual components of components of utilization rate production of chemicals requirements (FR ) requirements (FR ) components of non- G N non-energy use non-energy use in % and refinery products in GJ/t producta,i in GJ/t productb,i energy use (Source) Ethanec 59 ± 9 40 Steam cracking Propanec 88 ± 13 67 Ethylene production 90 ± 10 Butanec 91 ± 14 69 d capacity (OGJ, 2001) c (FS,i) Naphtha 136 ± 20 100 Gas oilc 142 ± 21 105 Feedstock use Ammonia Ammonia production Natural gas, heavy oil, e 90 ± 10 35 ± 7 24 (NEUF,i) Production (FA,i) capacity (CW, 2000) coal Methanol Methanol production Natural gas, heavy oil, e 90 ± 10 35 ± 7 27 production (FM,i) capacity (CW, 2001) coal Carbon black Carbon black production capacity 90 ± 10 Heavy oil, natural gas 85 ± 17 56 production (F )f C,i (CW, 2005) Aromatics Aromatics production 90 ± 10 Oil, coal 40 consumption (C ) capacity (OGJ, 1999) Non-energy use A,i Electrodes and Aluminium production of refinery other solid carbon (UN, 2000), GDP (WB, - Oil, coal 18 products coke use (C )h 2004) oven products, E,i Bitumen Bitumen production and other solid 90 ± 10 Oil 40 consumption (C ) capacity (OGJ, 1999) carbon (NEU )g B,i R,i Lubricants Lubricants production 90 ± 10 Oil 40 consumption (CL,i) capacity (OGJ, 1999) a We state here (i) the specific feedstock requirements in gigajoules per tonne of product for the production of chemicals (excluding backflows to refineries) and (ii) the lower heating values of non-energy use refinery products. We assume that feedstock requirements show a Gaussian distribution; the implemented uncertainty intervals represent approximately the 95% confidence interval (two standard deviations) of data on gross feedstock requirements as presented by Worrell (1994), Phylipsen (2000), Groenenberg (2002), and Neelis et al. (2005b). b We determine net feedstock requirements based on Neelis et al. (2005a, b). We assume the following shares of feedstock being used for energy purposes in steam cracking: 37% of ethane, 24% of propane, 24% of butane, 26% of naphtha, and 28% of gas oil. We assume furthermore that 30% of the feedstock for ammonia production and 22% and 34% of the feedstock for the production of methanol and carbon black, respectively, is consumed for fuel purposes. c We use data on feedstock distribution as given by OGJ (2001).

d We base the specific energy requirements for steam cracking of individual feedstocks on IPTS (2003) and Neelis et al. (2005a, b). We assume gross feedstock requirements as of average Western European steam crackers. To account for variability in the severity (i.e., propylene to ethylene ratio) at which steam crackers are operated worldwide, we introduce a 25% uncertainty interval to the total gross and net feedstock use of steam crackers (Ullmann, 2007; not shown in Table 2.1). e We base gross and net feedstock requirements for ammonia and methanol production on Worrell (1994), Goenenberg (2002), and Neelis et al. (2005a, b). We approximate production capacities for methanol in the year 2000 based on data available for the year 2001 (CW, 2001) by assuming that methanol production grows at 2% per year. f We base gross and net feedstock requirements for carbon black production on Neelis et al. (2005a, b). We approximate production capacities for carbon black in the year 2000 based on available data for the year 2005 (CW, 2005) by assuming that carbon black production grows at 2% per year. 23 g We include here only production in aromatics plants at refinery sites. We approximate the production of aromatics in steam crackers (and the related feedstock requirements) by ethylene production under steam cracking. h We calculate total requirements of electrodes and other solid carbon for the manufacturing of non-ferrous metals, ferroalloys, and other inorganic chemicals based on regression analysis as outlined in Equation (2.4). We refer here to the electrode consumption for aluminium production. We assume (i) electrodes to be produced to 16% from pitch (lower heating value of 37.5 GJ/t, carbon content of 93%) and to 84% from petroleum coke (lower heating value of 40.2 GJ/t, carbon content of 97%), (ii) 5% carbon losses during electrode production, and (iii) 99% carbon content of electrodes. Our calculation results in feedstock requirements of 42.9 GJ/t of electrodes. Assuming electrode consumption of 0.41 t/t aluminium (Sjardin, 2003) yields electrode requirements of roughly 18 GJ/t of aluminium. i We do not differentiate between gross and net feedstock requirements for refinery products because the shares of refinery feedstock that are used for energy energystatistics Non-energy useininternational purposes in the production of refinery products are part of the energy conversion section in energy statistics and are therefore excluded from non-energy use.

Chapter 2

With the outlined bottom-up methodology, we estimate non-energy use for the year 2000 in the first instance based on a net definition. We apply our methodology to the world as a whole and to the 50 selected countries, which show the highest consumption of fossil fuels for non-energy purposes. Taking the uncertainties of individual components into account, we apply standard (i.e., Gaussian) error propagation rules to quantify the total uncertainties (i.e., the 95% confidence interval) of our bottom-up non-energy use estimates (see Appendix of this chapter). We compare our results with data from international energy statistics (IEA, 2005a, b) by calculating deviations such as:

(NEU − NEU ) = i IEA,i × Di 100% (2.5) NEU IEA,i where Di represents the deviation between our estimate and the official IEA value and NEUIEA,i the non-energy use as stated by IEA (2005a, b) for country i. In cases where our results are more than two standard deviations lower than IEA values, we calculate in a second step the non-energy use based on a gross definition (see Table 2.1) to identify the effect of the chosen system boundary on the deviations between our bottom-up estimates and IEA data.

2.4 Results According to our analysis, worldwide net non-energy use amounts to 20 ± 2 EJ (exajoules). This value is 19% lower than the official IEA (2005a) estimate of 24 EJ. We find that the USA (4.3 ± 0.5 EJ), China (1.4 ± 0.1 EJ), Russia (1.3 ± 0.1 EJ), and Japan (1.2 ± 0.2 EJ) are the largest consumers of fossil fuels for non-energy purposes. The 50 countries included in our analysis consume roughly 93% of global non-energy use. The data from international energy statistics (IEA, 2005a, b) are within the uncertainty intervals of our estimates for only 12 countries (Figure 2.4). Our bottom-up estimates and their uncertainty intervals are lower than IEA data in the case of 26 countries and for the world as a whole. For 12 countries, our results and the related uncertainty intervals exceed official IEA data.

With our bottom-up approach, we follow a net definition of non-energy use, thereby excluding feedstock parts that are used for energy purposes in chemicals production. We would hence expect IEA values to be located within the uncertainty intervals of our estimates. The good compliance between our estimates and the IEA (2005a, b) data for 12 individual countries indicates that these countries might indeed follow a net definition of non-energy use. However, finding that our net non-energy use data underestimate non-energy use in 26 countries and in the world as a whole shows that official non-energy use data might not be harmonized with regard to their system boundaries (i.e., countries might either partially or completely include feedstock that is used for energy into their non-energy use data). In these 27 cases, the application of a partial or complete gross definition might offer an explanation for the observed deviations.

24 Non-energy use in international energy statistics

We therefore conduct for 26 countries and for the world as a whole also an analysis, for which we apply a gross definition of non-energy use, thereby adding the fuel use of feedstock to our non-energy use estimates.

World USA China (excl. Taiwan) Japan South Korea Germany Russia Canada France India Brazil United Kingdom The Netherlands Saudi Arabia Spain Italy China (Taiwan) Mexico Indonesia Belgium South Africa Iran Argentina Poland Australia Thailand Turkey Singapore Egypt Trinidad and Tobago Lybia Pakistan New Zealand Qatar Portugal Venezuela Uzbekistan Belarus Malaysia Norway Romania Czech Republic Algeria Colombia Israel Slovak Republic Hungary Austria Sweden Bulgaria Kuwait

-100 -50 0 50 100 150 200 Relative deviation of non-energy use estimates in % Figure 2.4: Relative deviations between net non-energy use as estimated with our bottom-up methodology and as given by IEA (2005a, b); positive deviations indicate that our estimates are higher than IEA data; negative deviations imply that our estimates are lower than IEA data10

10 It is important to note that, for example, a positive deviation of 100% indicates that our estimate exceeds the IEA value by a factor of two. On the other hand, if our estimate is a factor of two smaller than the official IEA value, we find a negative deviation of only 50%. A deviation of -100% would indicate that our bottom- up estimate for non-energy use is zero.

25 Chapter 2

Although we lack detailed insight into the statistical practice of most countries, the results given in Table 2.2 provide support for our previous arguments. If we include fuel use of feedstock, our non-energy use estimates are in line with IEA (2005a, b) data for the world as a whole and for 13 out of the 26 countries. The deviations between our estimates and IEA data become smaller, but IEA data are still located outside of the uncertainty interval of our results for a remaining 13 countries. Detailed insight into the energy statistics of Germany and the Netherlands has been gained in the course of previous analyses (Neelis et al., 2005a; Neelis and Pouwelse; 2008; Weiss et al., 2008b). These analyses reveal that the Netherlands apply indeed a net definition for non-energy use. By contrast, Germany applies a net definition only for the feedstock use of natural gas but a gross definition for coal and oil feedstock (Weiss et al., 2008b). Furthermore, deviations between our bottom-up estimates and official non-energy use data are for both countries caused to some extent by errors in energy statistics11.

We have shown that negative deviations can be explained for some countries in a rather straightforward way by the fact that we apply a net definition of non-energy use whereas IEA (2005a, b) data might refer to a partial or complete gross definition. However, explaining negative deviations is more difficult in the case of the 13 countries, for which our gross non-energy use estimates still underestimate non-energy use in comparison to IEA (2005a, b). In the case of South Korea, we explain the negative deviation of 29% between our gross non-energy use estimate and the IEA (2005a) value to some extent with erroneous energy statistics, i.e., double counting of naphtha consumption in steam crackers as it was identified by Park (2005). Further explanations for remaining negative deviations (see Table 2.2) refer to a possible under-estimation of specific feedstock requirements in our bottom-up calculations for countries, which produce chemicals via alternative process routes. South Africa might be an example for such a case, where basic chemicals such as ethylene or propylene are produced from coal-based feedstock (i.e., via liquefaction of coal), which is less efficient than olefins production from oil via naphtha.

Explaining positive deviations for the 12 countries for which our results are higher than IEA data (see Figure 2.4) requires detailed insight into the statistical practice of individual countries. One possible reason for positive deviations might be that parts of solid carbon, which are used for the production of non-ferrous metals, ferroalloys, and inorganic chemicals are assigned to energy use (and thus excluded from non-energy use) in international energy statistics. Another reason for positive deviations might be that we neglect trade of refinery products in our bottom-up estimates. This might lead to over- or under-estimation of actual non-energy use. Furthermore, errors in energy statistics can cause substantial over- and under-estimation of non-energy use, as the cases of the Netherlands and South Korea have shown.

11 Neelis (2006), for example, quantifies errors related to non-energy use data in Dutch energy statistics with 103 PJ. These errors account for roughly 17% of the total non-energy use as stated by IEA (2005b). They are caused by the erroneously exclusion of aromatics and other chemical grade products from non-energy use.

Table 2.2: Non-energy use of the world as a whole and of selected countries as stated by IEA (2005a, b) and as calculated based on a net and gross definition of system boundaries Country Non-energy use in PJ Deviation in % Are IEA data [5, 6] within the IEA Bottom-up, Bottom-up, Uncertainty intervala Bottom-up, net Bottom-up, gross uncertainty interval of our net (2005a,b) net gross Bottom-up, gross vs. IEA vs. IEA bottom-up estimates? World 24,183 19570 23954 3251 -19 -1 Yes 27 USA 5,096 4282 5060 692 -16 -1 Yes China (excl. Taiwan) 2,147 1415 1910 288 -34 -11 Yes Japan 1,824 1214 1454 252 -33 -20 No South Korea 1,149 652 816 173 -43 -29 No France 681 528 651 120 -22 -4 Yes Brazil 599 367 457 81 -39 -24 No Spain 399 297 340 53 -26 -15 No Indonesia 273 200 276 44 -27 1 Yes

South Africa 234 85 103 11 -64 -56 No energystatistics Non-energy useininternational Argentina 182 113 139 16 -38 -24 No Poland 182 150 188 25 -18 3 Yes Thailand 164 129 172 42 -22 5 Yes Turkey 150 98 111 17 -35 -26 No Egypt 137 105 137 18 -24 0 Yes Libya 125 40 51 13 -68 -59 No Pakistan 119 79 107 21 -33 -10 Yes New Zealand 101 77 93 17 -24 -8 Yes Qatar 100 67 94 14 -32 -6 Yes Portugal 98 60 70 13 -39 -28 No Uzbekistan 93 81 101 15 -13 9 Yes Belarus 91 67 82 10 -26 -10 Yes Algeria 79 34 47 9 -57 -40 No Colombia 76 14 17 2 -82 -77 No Israel 70 20 25 6 -71 -64 No Slovak Republic 67 61 70 8 -10 3 Yes Austria 65 28 36 10 -57 -44 No a We estimate uncertainty intervals based on assumptions as specified in Table 2.1.

Chapter 2

To obtain more detailed insight into the structure of official non-energy use data, we disaggregate the deviations shown in Figure 2.4 into two parts: (i) feedstock use in the chemical industry (NEUF,i, see Figure 2.5) and (ii) non-energy use of refinery products, coke oven products, and other solid carbon (NEUR,i, see Figure 2.6).

Uzbekistan Belarus Malaysia Norway Romania Czech Republic Algeria

Israel Slovak Republic Hungary Austria Sweden Bulgaria

-100 -50 0 50 100 150 200 Relative deviation of feedstock use estimates in % Figure 2.5: Relative deviations between net feedstock use as estimated with our bottom-up methodology and as given by IEA (2005a, b); Note that deviations for Venezuela, Colombia, and Kuwait are excluded here because feedstock use as given by IEA (2005a) is zero for these countries.

28 Non-energy use in international energy statistics

-1000 100 200 300 400 500 600 700 800 Relative deviation of estimates on non-energy use of refinery products, etc. in % Figure 2.6: Relative deviations between non-energy use of refinery products, coke oven products, and other solid carbon as estimated with our bottom-up methodology and as given by IEA (2005a, b)

We first focus on feedstock use in the chemical industry (NEUF,i). IEA (2005a, b) data are within the uncertainty intervals of our results for 11 individual countries (Figure 2.5). Our results (including their 95% confidence interval) are lower than IEA (2005a, b) data for the world as a whole and in the case of 27 countries. Our estimates exceed IEA (2005a, b) data in the case of 9 countries. The feedstock use as given by IEA (2005a) for Colombia, Venezuela, and Kuwait is zero. By contrast, we estimate feedstock requirements for these countries based on ethylene production capacities of 100, 600, and 800 kt, respectively, as given by OGJ (2001). We hence find a clear inconsistency, which is most likely caused by errors and thus underreporting of feedstock use in international energy statistics. Another example is the Ukraine, which was not selected for this analysis

29 Chapter 2 because this country is not among the 50 largest consumers of fossil fuels for non-energy purposes according to IEA (2005a, b). International energy statistics report zero feedstock use for the Ukraine in the year 2000 (IEA, 2005a) whereas CW (2000) states ammonia production capacities of about 4.2 Mt. This translates into 145 PJ (gross) feedstock use and ranks the Ukraine among the 30 largest consumers of fossil fuels for non-energy purposes in the world.

We now focus on non-energy use related to the consumption of refinery products, coke oven products, and other solid carbon (NEUR,i; Figure 2.6). IEA (2005a, b) data are within the uncertainty intervals of our estimates for the world as a whole and for 15 individual countries. Our estimates (and their uncertainty intervals) are higher and lower than IEA (2005a, b) data in the case of 17 and 18 countries, respectively. The magnitude of deviations with regard to the consumption of refinery products, coke oven products, and other solid carbon is higher than for feedstock use (compare Figure 2.5 and Figure 2.6). In the case of Trinidad and Tobago, IEA (2005a) reports only 0.4 PJ of non-energy use of refinery products, coke oven products, and other solid carbon. However, OGJ (1999) reports lubricant production of roughly 77 kt resulting in a non-energy use of 3 PJ in our bottom-up calculations and causing a positive deviation of 670% between our estimate and the data from international energy statistics (IEA, 2005a) (Figure 2.6). We also find substantial positive deviations for, e.g., Singapore, Mexico, and South Korea, where our estimates exceed official IEA (2005a, b) data by roughly 420%, 170%, and 120%, respectively. Conversely, we estimate the non-energy use of refinery products to be zero in Algeria and Kuwait based on available data on production capacities (OGJ, 1999). We under-estimate the same item by more than 90% in the case of Austria and Colombia. These deviations might potentially be caused by our methodology that accounts for reasons of simplicity only for the production but not for the consumption of refinery products. Our methodology might therefore lead to under- or over-estimation of non- energy use of refinery products, coke oven products, and other solid carbon in cases where substantial net trade flows occur (see also Section 2.5.1, below).

The findings of our disaggregated analysis reveal another problem associated with the data comparison displayed in Figure 2.4. The correspondence between our estimates and the IEA (2005a, b) data can be an artefact of data aggregation, if positive and negative deviations for the two components of non-energy use cancel each other out. The results presented in Figure 2.5 and Figure 2.6 indicate that this is to some extent the case for, e.g., China for which we find a negative deviation of 1% for NEUF,i and a negative deviation of 79% for NEUR,i resulting in an overall negative deviation of only 34% between IEA (2005a) data and our bottom-up net non-energy use estimate.

30 Non-energy use in international energy statistics

2.5 Discussion 2.5.1 Discussion of methodology The overall uncertainties of our net non-energy use estimates range for individual countries between 8% and 24% of the final result. Our implemented uncertainty intervals represent the 95% confidence interval related to the capacity utilization of chemical plants and refineries as well as feedstock requirements for chemicals production in the various countries. The implemented uncertainty intervals, furthermore, account for regression errors that result from the application of multiple regression analysis to estimate the consumption of electrodes and other solid carbon for the production of non-ferrous metals, ferroalloys, and inorganic chemicals (i.e., CE,i). However, due to limited availability of country-specific data, we are unable to quantify several sources of uncertainty such as the effect of net trade of refinery products. An extended uncertainty analysis, which is performed for each country individually on the basis of detailed data, would go beyond the scope of this research. Such analysis might potentially increase the uncertainty intervals of our results, leading to the situation that our bottom-up estimates become compliant with IEA (2005a, b) data for additional countries (e.g., India, Malaysia, or the Slovak Republic).

In the following, we qualitatively discuss uncertainties, which are not covered by our uncertainty intervals. As for any bottom-up approach, the quality of our results depends on (i) the accuracy of input data and (ii) the reliability of the applied modelling approach. Addressing the first point, i.e., the accuracy of model input data we base our bottom-up calculations for the non-energy use of refinery products (i.e., aromatics, bitumen, and lubricants) on production capacities of the respective countries. We thereby neglect trade of refinery products that is accounted for by the IEA (2005a, b), i.e., as exports and imports of secondary energy carriers. The related error is small for large countries where net trade flows are generally negligible relative to domestic production. We might, however, substantially over- or under-estimate non-energy use of refinery products for small countries where the net trade flows of refinery products can be substantial (e.g., as it might be the case for Trinidad and Tobago, Mexico, or Austria).

With respect to our bottom-up methodology, uncertainties relate to the accounting of refinery aromatics12 (i.e., aromatics that are produced in aromatics plants at refinery sites). Here, the international energy statistics (IEA, 2005a, b) remain vague in two points: (i) It is unclear whether pure aromatics are regarded as energy carriers or as chemicals. It, therefore, remains ambiguous whether pure refinery aromatics are part of energy statistics (i.e., being accounted for as other petroleum products or whether they are regarded as chemicals, for which only feedstock use is reported in IEA energy statistics (IEA, 2005a, b). (ii) Assuming that pure refinery aromatics are included under other petroleum products in IEA energy statistics (IEA, 2005a, b), we encounter a second

12 Chemical grade aromatics are produced in aromatics plants at refinery sites and as by products in steam crackers. In our discussion of uncertainties, we refer only to refinery aromatics. The non-energy use that is related to the production of aromatics in steam crackers is part of the feedstock requirements for ethylene production (see Table 2.2) and accounted for under feedstock use (NEUF).

31 Chapter 2

problem, i.e., the question whether refinery aromatics are accounted for under feedstock use of the petrochemical industry or under non-energy use in the industry, transformation, and energy sector (i.e., as non-energy use of refinery products).

Information given by IEA energy statisticians (IEA, 2007a) could not entirely clarify these points. Additional uncertainties are related to our estimates on the consumption of electrodes and other solid carbon in the non-ferrous metals, ferroalloys, 13 and inorganic chemicals industry (CE,i) . In the following, we explain each of them separately. First, uncertainties refer to the applied multiple regression analysis in which we include data from 21, mainly industrialized, countries (Weiss et al., 2009a). The under- representation of developing countries might introduce bias into our regression analysis and the functional relationship established by Equation (2.4). Second, uncertainties refer to the fact that our regression analysis is based on CO2 emissions related to the consumption of electrodes and other carbon sources. We thereby assume that the carbon content of all reducing agents is equal. This is, however, not entirely true14. Third, uncertainties result from our assumption that electrodes consumption in a particular country equals electrodes production. In general, this assumption is justified as electrodes for metallurgical processes are usually produced in spatial proximity to the place of their consumption. This assumption might nevertheless introduce an error into our non-energy use estimates, if net trade of electrodes is substantial15.

Finally, uncertainties are related to the inclusion of other solid carbon that is used next to electrodes as reducing agent in the production of non-ferrous metals, ferroalloys, and inorganic chemicals. These carbon quantities might be excluded from non-energy use in the energy statistics of some countries. However, we regard the combined effect of uncertainties related to the estimation of electrodes and other solid carbon use (CE,i) minor, given the small share of this activity on the non-energy use of most countries. A final factor of uncertainty is related to the assumed gross feedstock requirements of chemical processes (i.e., steam cracking, production of ammonia, methanol, and carbon black). Although we introduce error intervals in our bottom-up calculation, our data might not exactly reproduce the actual feedstock requirements in each of the 50 countries that are included in our analysis. This refers in first instance to countries where, e.g., abundant hard coal is used as feedstock for ammonia and methanol production. The use of coal (instead of natural gas or heavy oils) results in substantially increased process-specific gross feedstock requirements.

13 This activity only accounts on average for 4% of non-energy use. However, in individual cases such as Norway and Australia, electrodes and other solid carbon use in the non-ferrous metals, ferroalloys, and inorganic chemicals industry is responsible for 35% and 27% of non-energy use, respectively. 14 Based on IPCC (1995), we determine the following carbon content for the relevant substances: electrodes (88.9 kg CO2/GJ), coke (108.2 kg CO2/GJ), and coal (94.6 kg CO2/GJ). Although differences in the carbon content of the various reducing agents are relatively small, variations in the share of carbon sources used for the production of non-ferrous metals, ferroalloys, and inorganic chemicals from one country to another can lead to over- or under-estimation of actual non-energy use, if it is expressed in energy units (i.e., PJ). 15 In the case of Germany, net electrode exports amount to 63 kt in 2000. This quantity accounts, however, for only 0.1% of the total German gross non-energy use in the same year.

32 Non-energy use in international energy statistics

2.5.2 Discussion of results For the world as a whole and half of the analysed countries, our bottom-up estimates are either in line with IEA (2005a, b) data or deviations can be explained easily by inconsistencies in the system boundaries of non-energy use as given by international energy statistics. However, positive deviations for 12 countries as well as negative deviations for another 13 countries cannot be explained in a straightforward way. These deviations might to some extent indicate erroneous accounting of non-energy use in international energy statistics. To clarify this point, we recommend more detailed analyses, e.g., by using country-specific data on plant efficiencies and on the consumption of refinery products or by conducting in-depth analyses with the Non-energy use Emission Accounting Tables (NEAT) model (Neelis, 2005b).

Our results demonstrate the usefulness of the applied bottom-up methodology (i) for obtaining insight into the system boundaries of non-energy use data and (ii) for identifying possible errors in the international energy statistics (IEA, 2005a, b). Applying our methodology would allow energy statisticians, inventory experts, and energy researchers (i) to evaluate the accuracy of non-energy use data for countries that are so far excluded from our analysis and (ii) to create a shortlist of countries for which non-energy use data should be verified by more detailed analyses. Efforts regarding the latter point should focus in first instance on: (i) clarifying the system boundaries of non-energy use data for, e.g., the USA, China, France, Indonesia, and Poland (ii) correcting potential errors and incompleteness of official non-energy use data as we identified in the case of Ukraine, Venezuela, Kuwait, and Colombia

Our results also emphasize the need to improve clarity and scope of the annual questionnaires sent out by the IEA to refineries and chemical companies. This is true regarding the system boundary of non-energy use data and, in more particular, the accounting of aromatics, for which guidance in the current IEA (2006) oil questionnaires is still ambiguous. Prerequisites for consistent reporting of non-energy use have been defined by Neelis and Pouwelse (2008) and include a clear statement (i) on how to include the various chemical conversion processes into energy statistics and (ii) on the status of products, i.e., whether or not to regard hydrocarbons as chemicals or energy carriers. These requirements are to date not entirely fulfilled, rendering the questionnaires uncertain regarding both system boundaries of non-energy use and the treatment of energy conversions (Neelis and Pouwelse, 2008). We hence find additional support for our conclusion that errors in official non-energy use data might provide one important explanation for the observed deviations between our estimates and IEA (2005a, b) data. The deviations that are caused by errors in the reporting of non-energy use potentially increase, if the absolute non-energy use of a country is small because only one chemical plant that is omitted or that reports erroneous data can have a tremendous effect on the overall non-energy use as it is reported in energy statistics.

Energy statistics serve as the single most important data source for GHG inventories. The requirements for accurate and precise data in energy statistics are growing and are expected to do so in the future due to an increasing need for reliable and more

33 Chapter 2 detailed GHG emission data. This is especially true, if emission estimates from GHG inventories are used to evaluate the effectiveness of GHG emission mitigation policies. Reliable non-energy use data in national and international energy statistics are also important for the monitoring of energy efficiency developments in industry. Continuous improvement of energy efficiency is regarded as one important strategy for stabilizing global energy demand.

The requirements for both GHG inventories and the monitoring of energy efficiency challenge energy statisticians to continuously improve the quality of their data. The process of quality improvement is especially complicated with respect to non-energy use data because fossil fuel quantities used for non-energy purposes are generally small and often within the uncertainty intervals of fuel use data in energy statistics (DIW, 2007). Improved questionnaires and better guidance of refineries and chemical companies can be one measure to increase the quality of non-energy use data in energy statistics. Applying a simple model tool to continuously crosscheck data consistency (as it is proposed in this research) can be a second measure. Apart from the issue of data quality, the bottom-up methodology outlined in this chapter can help inventory makers to gain insight into the system boundaries of non-energy use data and assists them in the preparation of correct and reliable GHG inventories.

2.6 Conclusions In this chapter, we apply a simple bottom-up methodology to estimate non-energy use of fossil fuels for the world as a whole and for 50 individual countries, which are the largest consumers of fossil fuels for non-energy purposes. We quantify worldwide non-energy use to be 20 ± 2 EJ. Our estimates are in line with data from IEA (2005a, b) energy statistics for 12 countries. Our estimates are lower than IEA (2005a, b) data for the world as a whole and for 26 individual countries, whereas our estimates exceed official IEA data in the case of 12 countries. Fourteen out of the 27 cases in which our estimates are lower than IEA (2005a, b) might be entirely explained by the application of a gross or partial gross definition for non-energy use in international energy statistics. Positive as well as unexplained negative deviations between our estimates and official IEA data might be caused to some extent by errors and incompleteness in the accounting of non-energy use by official IEA data.

However, disregarding trade of refinery products in our bottom-up methodology might also provide an explanation for deviations between our estimates and official IEA data for small countries with substantial net trade flows. Accounting for both, the overall uncertainties of non-energy use data in international energy statistics and the scope of our analysis, we regard our approach suitable and the uncertainties of our results acceptable. Our method can hence be used by both energy statisticians and GHG inventory experts to crosscheck non-energy use data for any country in the world. Our bottom-up approach is furthermore suitable and to create an international short list of countries for which additional efforts should be made to verify data from energy statistics. Efforts should thereby focus on (i) clarifying the system boundaries of non-energy use data as it might be

34 Non-energy use in international energy statistics necessary for, e.g., the USA, China, France, Indonesia, and Poland and (ii) correcting potential errors in official non-energy use data as we identified in the case of Ukraine, Venezuela, Kuwait and Colombia. Based on the presented research, we conclude that this chapter makes a contribution to a more reliable accounting of fossil fuels that are used for non-energy purposes.

Acknowledgements This research was funded by the European Commission under the 6th framework programme on ‘Sustainable development, global change, and ecosystems’, contract number FP6-50345 (International network non-energy use and CO2 emissions (NEU-CO2, phase III)). The authors would like to thank all network participants and the three anonymous reviewers for their valuable contributions to this chapter.

Appendix As outlined in Section 2.3, we calculate uncertainty intervals for our bottom-up non- energy use estimates. We apply standard (i.e., Gaussian) error propagation rules to arrive at estimates for the total uncertainty based on uncertainties related to the individual components of non-energy use (see Table 2.1). We explain now our methodology in greater detail. The variables that are used in the following equations are the same as introduced in Section 2.3. The error interval, i.e., the uncertainty of each of the components is indicated by the symbol .

We calculate the uncertainty interval of our non-energy use estimates based on the uncertainties of each of its two principal components as:

Δ = Δ 2 + Δ 2 NEU i ( NEU F,i ) ( NEU R,i ) (A2.1)

The uncertainty related to the consumption of feedstock in the chemical industry (NEUF,i) is calculated based on the uncertainties related to each of the individual components of feedstock use as:

Δ = Δ 2 + Δ 2 + Δ 2 + Δ 2 NEU F ,i ( FS ,i ) ( FA,i ) ( FM ,i ) ( FC,i ) (A2.2)

The uncertainty interval of feedstock use in each individual chemical process is calculated based on two components, (i) the uncertainty related to the capacity utilization, i.e., 90 ± 10% and (ii) the uncertainty related to the process-specific gross feedstock requirements (see Table 2.1). In the example of ammonia production, we calculate the uncertainty interval related to feedstock use as:

35 Chapter 2

2 2 C ΔP S C ΔFR S Δ = D A,i T + D GA T × FA,i D T D T FA,i (A2.3) E PA,i U E FRGA U where PA,i stands for ammonia production in country i and FRGA for the gross feedstock requirements of ammonia production. The calculation of error intervals for gross feedstock use in the manufacturing of methanol, and carbon black follows in analogy.

In Table 2.1 we differentiate various types of feedstock for ethylene production in steam crackers. Our uncertainty calculation for ethylene production in steam crackers involves hence the calculation of uncertainties on a more disaggregated level of individual feedstocks such as:

Δ = Δ 2 + Δ 2 + Δ 2 + Δ 2 + Δ 2 FS,i ( FSE,i ) ( FSP,i ) ( FSB,i ) ( FSN,i ) ( FSG,i ) (A2.4) where FSE,i, FSP,i, FSB,i, FSN,i, and FSG,i refer to the uncertainty related to the steam cracking ethane, propane, butane, naphtha, and gas oil, respectively. For each of these types of feedstock, we calculate uncertainty intervals in analogy to the example given below for ethane as follows:

2 2 C ΔP S C ΔFR S Δ = D EE,i T + D GEE T × FSE,i D T D T FSE,i (A2.5) E PEE,i U E FRGEE U where PEE,i stands for ethylene production in country i based on ethane feedstock and FRGEE for the gross feedstock requirements for ethylene production based on ethane feedstock.

We calculate the uncertainty interval related to the consumption of non-energy use of refinery products, coke oven products, and other solid carbon (NEUF,i) as:

Δ = Δ 2 + Δ 2 + Δ 2 + Δ 2 NEU R,i ( C A,i ) ( CE,i ) ( CB,i ) ( CL,i ) (A2.6)

We approximate the consumption of aromatics (CA,i), bitumen (CB,i), and lubricants (CL,i) by openly available data on production capacities. The uncertainties related to the disregarding of trade are country specific (see Section 2.5.1.) and are therefore excluded from our uncertainty analysis. For calculating uncertainty intervals related to the consumption of aromatics, bitumen, and lubricants, we include only uncertainties that result from the approximation of actual production by production capacities, i.e., by assuming capacity utilization of 90 ± 10% (see Table 2.1). Uncertainties related to consumption of aromatics are calculated as:

36 Non-energy use in international energy statistics

2 C ΔP S Δ = D A,i T × C A,i D T C A,i (A2.7) E PA,i U where PA,i represents the production of aromatics in refineries of country i. The calculation of uncertainties related to bitumen and lubricants follow in analogy.

Uncertainties related to the consumption of electrodes and other solid carbon sources in the production of non-ferrous metals, ferroalloys, and other inorganic chemicals (CE,i) are calculated based on the uncertainty interval of the regression analysis as:

Δ = × + × CE,i 0.4 C Al,i 0.3 GDPi (A2.8)

We disregard uncertainties that might be related to data on CAL,i and GDPi because these are generally negligible compared to the other uncertainties of our regression analysis.

3 Non-energy use of fossil fuels and resulting carbon dioxide emissions: bottom-up estimates for the world as a whole and for major developing countries

Martin Weiss, Maarten L. Neelis, Kornelis Blok, and Martin K. Patel Published slightly adapted in: Climatic Change 95 (2009), pp. 369-394.

Abstract Fossil fuels are mainly used for energy but also for so-called non-energy purposes, e.g., as feedstock in the chemical industry and for the production of bitumen and lubricants in refineries. Data on non-energy use of fossil fuels and resulting carbon dioxide (CO2) emissions are in general subject to uncertainty. Here, we address this problem by applying a simple bottom-up model for estimating non-energy use and resulting CO2 emissions for the year 2000: (i) to the world as a whole, (ii) to the aggregate of Annex I countries and non-Annex I countries, and (iii) to the ten non-Annex I countries with the highest consumption of fossil fuels for non-energy purposes. We find that worldwide non-energy use is equivalent to 1,670 ± 120 Mt CO2 and leads to 700 ± 90 Mt CO2 emissions. Around 75% of non-energy use emissions are related to industrial processes. The remainder is attributed to the source categories of solvent and other product use, agriculture, and waste. Annex I countries account for 51% (360 ± 50 Mt CO2) and non-Annex I countries for 49% (340 ± 70 Mt CO2) of worldwide non-energy use emissions. Among non-Annex I countries, China is by far the largest emitter of non-energy use emissions (122 ± 18 Mt CO2). Our research deepens the understanding of non-energy use and related CO2 emissions in countries for which detailed emission inventories do not yet exist. Despite existing model uncertainties, we recommend our bottom-up model to inventory experts for preparing accurate non-energy use emission estimates for any country in the world.

39 Chapter 3

3.1 Introduction Fossil fuels are used for energy generation but also for so-called non-energy purposes, e.g., as feedstock in the chemical industry or for the production of bitumen and lubricants in refineries. The worldwide share of non-energy use in the total primary energy supply (TPES) has increased from 4.3% in 1971 to 5.8% in the year 2000 (IEA, 2005a) and 1 represents an important potential source for carbon dioxide (CO2) emissions . Estimating non-energy use and related CO2 emissions is complicated by the complex connections between energy and material flows within the chemical industry (Patel et al., 2005). As a consequence, non-energy use data as published by both national and international energy statistics lack a consistent definition of system boundaries (Neelis, 2006; Weiss et al., 2008a, b). Estimates of non-energy use emissions in national greenhouse gas (GHG) inventories suffer from errors or incompleteness and are generally subject to major uncertainties (La Motta et al., 2005; Neelis et al., 2005a; Park, 2005; Weiss et al., 2008b). Global emission inventories such as compiled by ORNL (Oak Ridge National Laboratory; Marland et al., 2006) or EDGAR (Emission Database for Global Atmospheric Research; Olivier, 2005a, b) account for non-energy use emissions by applying crude assumptions, which might not always be correct2. Within global emission inventories, the quality of non-energy use data proves to be an important source of uncertainty. Olivier and Peters (2002) determine deviations of 55% between individual estimates of non-energy use emissions, if calculated with different versions of EDGAR solely based on uncertainties related to underlying non-energy use data.

3 To overcome existing shortcomings, scientists involved in the NEU-CO2 network developed the Non-energy Use Emission Accounting Tables (NEAT) model. The application of NEAT enabled the identification of major errors and data gaps regarding non-energy use emission data in the national GHG inventories of Germany, Italy, the Netherlands, and South Korea (Weiss et al., 2008b; La Motta et al., 2005; Neelis et al., 2005a; Park, 2005). NEAT, however, requires extensive information on the production and trade of chemicals as well as detailed insight into the structure of the chemical industry. It is therefore impossible to apply the NEAT model to countries with limited data availability.

1 For countries with a large chemical industry, this share can be substantially higher. Examples in Europe are Belgium and the Netherlands where non-energy use accounts for 10.2% and 13.5% of TPES in the year 2000, respectively (IEA, 2005a). 2 Olivier (2005a, b) estimates non-energy use emissions in the EDGAR-3 database by multiplying total non- energy use as reported by IEA (2005a, b) with default carbon storage fractions as given by IPCC (1997). This approach is problematic because the IPCC (1997) default carbon storage fractions do not account for country specifics regarding (i) system boundaries of non-energy use data in energy statistics and (ii) trade of chemicals. 3 The NEU-CO2 (Non-Energy Use and CO2 Emissions) network was funded by the European Commission, DG-Research, under the ENRICH program (European Network for Research into Global Change). The network brings together scientists, greenhouse gas inventory makers, and experts from the chemical and petrochemical industry to analyze non-energy use and associated CO2 emissions (Patel, 2005).

40 Non-energy use and related emissions in major developing countries

We hence see the need for a simple model, suitable for calculating worldwide and country-specific non-energy use and resulting CO2 emissions correctly and independently from official energy statistics. The aim of this chapter is to develop a simplified version of the detailed NEAT model (i.e., NEAT-SIMP) and to apply this model for the year 2000 (i) to the world as a whole, (ii) to the aggregate of Annex I countries and non-Annex I countries4 and (iii) to the ten non-Annex I countries with the highest consumption of fossil fuels for non-energy purposes. The selected non-Annex I countries together account for 28% of the worldwide non-energy use and for 76% of the non-energy use of all non- Annex I countries in the year 2000 (IEA, 2005b)5. In addition, we apply NEAT-SIMP to three selected Annex I countries (i.e., Germany, Italy, and the Netherlands) to crosscheck our estimates with data from both the detailed NEAT model and the respective national GHG inventory.

Our research provides in the first instance a comprehensive overview of non- energy use and related CO2 emissions worldwide and in the most important non-Annex I countries for the year 2000. It furthermore reveals insights into the system boundaries of non-energy use data as stated by IEA (2005a, b) and contributes to the improvement of national and international energy statistics. Ultimately, NEAT-SIMP estimates can provide inventory makers with building blocks for preparing and crosschecking emission estimates in national GHG inventories.

The chapter continues with a short description of the NEAT-SIMP model and the data sources used for our calculations (Section 3.2). In Section 3.3, we present results and we compare our findings with data from three sources, i.e., the international energy statistics (IEA, 2005a, b), the national GHG inventories (UNFCCC, 2006a), and the detailed NEAT model (Weiss et al., 2008b; La Motta et al., 2005; Neelis et al., 2005a). In the fourth and fifth section of this chapter, we discuss assumptions and uncertainties related to our NEAT-SIMP estimates, we give recommendations for future research, and we draw conclusions.

3.2 Methodology and data sources 3.2.1 Definitions and background We define non-energy use consistently with IEA (2005a, b) as the sum of two components: (i) the consumption of fossil fuels as feedstock in the chemical industry (e.g., the use of naphtha for olefins production in steam crackers or the consumption of natural gas for the production of ammonia) (ii) the consumption of refinery products, coke oven products, and other solid carbon for non-energy purposes (e.g., the use of lubricants for transportation,

4 We refer here to countries, which are not included under Annex I of UNFCCC (2006a). Non-Annex I countries do not submit yearly GHG inventories to the UNFCCC. 5 Despite the non-harmonized system boundaries of non-energy use data as stated by IEA (2005a, b), we base the selection of countries on this source as it provides the only complete worldwide overview of energy data for individual countries. The following non-Annex I countries are included in our analysis: China, South Korea, India, Brazil, Saudi Arabia, Mexico, Indonesia, South Africa, Iran, and Argentina.

41 Chapter 3

the use of bitumen in the building sector, or the consumption of electrodes and other solid carbon in the manufacturing of non-ferrous metals, ferroalloys, and inorganic chemicals)6

One part of non-energy use remains stored in long-life products such as polymers or bitumen, whereas the other part is oxidized and subsequently released into the atmosphere. Both components of non-energy use lead to CO2 emissions in each of the following two IPCC emission source categories (IPCC 1997): (i) industrial processes, e.g., emissions from feedstock use during the production of ammonia, or from electrodes use in the non-ferrous metals industry7 (ii) solvent and other product use, e.g., emissions from the consumption of solvents (produced from chemical feedstock) or from the use of lubricants (made in refineries)

The first component of non-energy use, i.e., feedstock use in the chemical industry, also leads to CO2 emissions in the IPCC emission source categories agriculture (e.g., via the application of urea fertilizers) and waste8 (e.g., via the oxidation of surfactants during wastewater treatment; IPCC, 1997). Before continuing, it is important to define the system boundary for non-energy use because in many industrial processes, parts of the feedstock are not used for non-energy purposes but as fuel for heat raising in furnaces. Thus, estimates for non-energy use can vary between either a gross definition that includes fuel use of feedstock or a net definition that excludes fuel use of feedstock from non-energy use. In NEAT-SIMP, we follow a gross definition of non-energy use, i.e., we include the parts of feedstock that are used for fuel in chemical processes but we exclude backflows from steam crackers to refineries. We hence expect NEAT-SIMP estimates to be in the upper value range of IEA (2005a, b) non-energy use data as the latter are not entirely harmonized, i.e., their system boundary might follow a gross, partial net, or pure net definition (Weiss et al. 2008b)9. In the next section, we explain the NEAT-SIMP

6 We exclude the consumption of coke and coal for iron and steel production from non-energy use because these items are generally accounted for under energy conversions in energy statistics. 7 Emissions from the incineration of fuel-grade by-products of chemical conversions are generally regarded as energy-related emissions in national GHG inventories and are thus excluded from non-energy use emission estimates. On exemption from this rule are fuel-grade by products in steam crackers (i.e., hydrogen and methane), which are accounted for as industrial process emissions in cases where a gross definition of non-energy use is applied. 8 Potential sources for non-energy use emissions in the source category waste include also landfilling and post consumer waste incineration (with and without energy recovery). The oxidation of fossil-based carbon in landfills is generally negligible and thus excluded from our estimates (see, e.g., UBA, 2004). Emissions from waste incineration are often regarded as secondary fuel use emissions and are thus accounted for under the source category energy. We completely exclude fossil based CO2 emissions that result from waste incineration from our estimates; thereby potentially neglecting emissions that result from waste incineration without energy recovery (see also discussion in Section 3.4). 9 For compiling international energy statistics, IEA asks recipients of its questionnaires to apply a net definition for the system boundary of non-energy use (IEA, 2006). However, only very limited guidance is provided by IEA to assure compliance on the level of individual chemical processes. Given the complexity of material and energy flows within the chemical and refinery industry, we would, therefore, expect that official non-energy use data are not harmonized with respect to their system boundaries. The findings of Weiss et al. (2008b) strongly support this conclusion.

42 Non-energy use and related emissions in major developing countries methodology for calculating non-energy use. Afterwards, we describe the calculation of non-energy use emissions separately for the four relevant source categories.

3.2.2 Estimating total non-energy use In NEAT-SIMP, we estimate non-energy use (NEUk) [Mt CO2 equivalents] for country or region k as:

FVFV3 4 =++++++ NEUkikSkAkMkCkNFkikGWGW()(BB PC,,,,,, E E E E E )() CR , (3.1) HXHXi=1 i=1 where PCi,k [Mt CO2 equivalents] stands for the physical production of basic chemical i (i.e., chemicals produced in steam crackers as well as methanol and carbon black), and ES,k, EA,k, EM,k, EC,k, ENF,k [Mt CO2 equivalents] for emissions from steam cracking, the production of ammonia, methanol, carbon black, and the use of electrodes and other solid carbon in manufacturing non-ferrous metals, ferroalloys, and inorganic chemicals, respectively. CRi,k [Mt CO2 equivalents] represents the domestic consumption of refinery products (i.e., refinery aromatics, bitumen, lubricants, as well as waxes and paraffins). By contrast to what is done in official energy statistics, we do not present total non-energy use in energy units, i.e., PJ or ktoe but as carbon content in Mt CO2 equivalents.

We quantify individual parameters by multiplying activity data with (i) specific emission factors or (ii) data on specific carbon content of chemicals and refinery products. Due to limited data availability, we uniformly approximate the consumption of refinery products by domestic production (Table 3.1). For the same reason, we estimate parts of the production data for both chemicals and refinery products based on production capacities, assuming a capacity utilization rate of 90 ± 10 % based on information given by OGJ (1999) and CW (2000). The uncertainty ranges of individual parameters represent the 95% confidence interval of possible values. We use standard error propagation rules to quantify the uncertainties of our final results. For analyzing non-energy use, we use model input data from OGJ (1999, 2001), UN (2000), Destatis (1990–2003a, b), IFA (2004), CBS (2006), and CW (2000, 2005).

43 Chapter 3

Table 3.1: Input data and carbon content of feedstock and refinery products as applied in NEAT-SIMP (based on Neelis et al. (2005b)) Carbon content Input data Feedstock in kg CO2/kg product Naphtha 10.0 ± 1.0 Steam cracking: Gas oil 12.6 ± 1.3 production of Ethane 3.6 ± 0.4 ethylenea Propane 6.3 ± 0.6 Butane 6.6 ± 0.7 Consumption of Others 10.0 ± 1.0 Gas oil, Diesel, Heavy fuel oil 2.8 ± 0.3 fossil fuels as Production of Natural gas 1.8 ± 0.2 feedstock ammoniab Coal 4.3 ± 0.4 Gas oil, Diesel, Heavy fuel oil 3.1 ± 0.3 Production of Natural gas 1.0 ± 0.1 methanolb Soft coal 6.2 ± 0.6 Production of Coal tar, Heavy fuel oil 5.8 ± 0.6 carbon blackb Natural gas 0.5 ± 0.1 Bitumen production - 3.1 and trade Consumption of Lubricants - 3.1 refinery production and trade products, coke Production of - 3.4 oven products, refinery aromaticsc and other solid Production of carbon aluminium, GDP Regression analysis based on total GDP and emissions from (gross domestic aluminium production product) a Specific feedstock requirements for steam cracking are derived from the detailed NEAT model and refer to average plant efficiencies in Western Europe. We assume feedstock requirements to be higher than these averages by 25% in China, by 20% in India, Iran, and Indonesia, by 15% for the aggregate of non-Annex I countries, by 10% for the world’s total and in Brazil, Mexico, Saudi Arabia, South Africa, Argentina, and by 5% for the aggregate of Annex I countries. We assume that feedstock requirements in South Korea are 10% lower compared to Western European averages (based on Phylipsen (2000) and Groenenberg (2002)). b We derive specific feedstock requirements for the production of ammonia, methanol, and carbon black from the detailed NEAT model. We account here for the fact that chemical plants in other parts of the world are less efficient than in Western Europe (based on Groenenberg (2002), Worrell (1994), and Kuramochi (2006)) c The feedstock requirements for aromatics produced in steam crackers are already accounted for under steam cracking.

3.2.3 Estimating industrial process emissions We estimate industrial process emissions in NEAT-SIMP with two approaches. The first approach is used for processes where detailed data on production and production capacities are available (e.g., for ethylene production in steam crackers and the production of ammonia, methanol, carbon black, and aluminium). Here, we multiply production data with process- and feedstock-specific emission factors (Table 3.2).

44 Non-energy use and related emissions in major developing countries

Table 3.2: Emission factors as applied in NEAT-SIMP for calculating gross industrial process emissions (based on Neelis et al. (2005b)) Emission factor in Process Feedstock kg CO2/kg product Naphtha 1.9 ± 0.2 Gas oil 2.4 ± 0.2 Steam crackinga Ethane 0.8 ± 0.1 Propane 1.1 ± 0.1 Butane 1.2 ± 0.1 Others 1.9 ± 0.2 Gas oil, Diesel, Heavy fuel oil 2.8 ± 0.3 Production of ammoniab,c Natural gas 1.8 ± 0.2 Coal 4.3 ± 0.4 d Production of urea CO2 -0.7 Gas oil, Diesel, Heavy fuel oil 1.5 ± 0.1 Production of methanolb Natural gas 0.4 ± 0.04 Soft coal 3.2 ± 0.3 Production of carbon blackb Coal tar, Heavy fuel oil 2.0 ± 0.2 Natural gas 0.2 ± 0.02 Production of aluminium Carbon Electrodes 1.5 ± 0.2 a Emission factors for steam cracking are derived from the detailed NEAT model and refer to average plant efficiencies in Western Europe. We assume emissions to be higher than these averages by 25% in China, by 20% in India, Iran, and Indonesia, by 15% for the aggregate of non-Annex I countries, by 10% for the world’s total and in Brazil, Mexico, Saudi Arabia, South Africa, and Argentina, and by 5% for the aggregate of Annex I countries. We assume emissions to be 10% lower in South Korea compared to Western European averages (based on Phylipsen (2000) and Groenenberg (2002)). b We derive specific emission factors for the production of ammonia, methanol, and carbon black from the detailed NEAT model. We correct emission factors for the fact that chemical plants in other parts of the world are less efficient than in Western Europe (based on Groenenberg (2002), Worrell (1994), and Kuramochi (2006)). c The emission factors presented here for ammonia production do not account for CO2 sequestration during the production of urea. d Urea is produced from ammonia and CO2. We deduct here the amount of CO2 that is sequestered for urea production.

In the case of ammonia production, we deduct the amount of CO2 that is sequestered for urea production from industrial process emissions. We base emission factors on plant efficiencies in Western Europe and apply correction factors to account for less efficient chemicals production in other regions of the world (see footnotes below Table 3.2). We obtain data on feedstock distribution for steam cracking as well as for the production of ammonia, methanol, and carbon black from OGJ (2001), IPTS (2004), and Neelis et al. (2005b).

The second approach for calculating industrial process emissions is used for estimating emissions from the consumption of electrodes and other solid carbon (e.g., coke and coal) that are used as reducing agents in the production of non-ferrous metals (e.g., zinc, lead), ferroalloys (e.g., ferrosilicon, ferromanganese), and inorganic chemicals (i.e., carbides). The availability of production data is generally limited for these materials. For this reason, we estimate emissions based on two considerations: (i) countries with a large absolute gross domestic product (GDP) are likely to have industrial production of nonferrous metals, ferroalloys, and inorganic chemicals

45 Chapter 3

(ii) the availability of cheap electricity is a precondition for both aluminium production and the manufacturing of other non-ferrous metals, ferroalloys, and inorganic chemicals.

Therefore, we expect a positive correlation between absolute GDP, emissions from aluminium production, and emissions from the total production of non-ferrous metals, ferroalloys, and inorganic chemicals. In NEAT-SIMP, we estimate total emissions from the production of nonferrous metals, ferroalloys, and inorganic chemicals based on regression analysis using detailed bottom-up data for 21 countries based on UN (2000), IPTS (2001), and WBMS (2002)10. In a first approach, we plot individually total emissions from non-ferrous metals, ferroalloys, and inorganic chemicals production as a function of (i) absolute GDP (WB, 2004) and (ii) emissions from aluminium production. Multiple least square regression, analyzing total emissions from the production of non-ferrous metals, ferroalloys, and inorganic chemicals as a function of both absolute GDP and emissions from aluminium production emissions provided, however, by far the best fit to our empirical data set. We find a clear linear relationship between total GDP, emissions from the production of aluminium, and emissions from the manufacturing of total nonferrous metals, ferroalloys, and inorganic chemicals (adjusted multiple coefficient of 2 11 determination of R = 0.79) . Hence, we estimate in NEAT-SIMP CO2 emissions from the production of non-ferrous metals, ferroalloys, and inorganic chemicals in country or region k with the following equation:

= ± × + ± × ENF ,k (1.7 0.8) E Al,k (1.2 0.5) GDPk (3.2) where EAl,k [Mt CO2 equivalents] represents the emissions from aluminium production 12 and GDPk the total gross domestic product [10 purchasing power parity-corrected constant (1995) $]. The error intervals in Equation 3.2 represent the 95% confidence interval of slope coefficients for the functional relationship expressed here.

3.2.4 Estimating solvent and other product use emissions In NEAT-SIMP, we apply a simple bottom–up approach to estimate emissions from three key sources, i.e., the consumption of solvents, lubricants, waxes, and paraffins. Depending on data availability, we estimate emissions for each key source based on either activity data or proxies. We estimate emissions from lubricant consumption by multiplying activity data as given by OGJ (1999) with an emission factor of 0.94 ± 0.63 kg CO2/kg lubricants (Patel et al., 1999; Trischler, 1997).

10 We include data from the following countries and years (in parentheses) in our regression analysis: Australia (2000), Austria (2000), Belgium (2001), Denmark (2000), Finland (1997), France (2000), Germany (2000), Greece (2000), Iceland (1997), India (2000), Italy (2001), Korea (2000), The Netherlands (1999), Norway (1997), Poland (2003), Russia (2000), South Africa (2000), Spain (1997), Sweden (1997), United Kingdom (2001), and USA (2003). We calculate CO2 emissions based on production data provided by producer associations, NEU-CO2 network partners, IPTS (2001), and UN (2000) as well as process- specific emission factors as given by Neelis et al. (2005b). 11 The regression coefficients for our two predictor variables are: (i) ! = 0.46, - = 0.002 for aluminum production and (ii) !=0.55 (p<0.001) for absolute GDP.

46 Non-energy use and related emissions in major developing countries

We calculate solvent use emissions based on linear regression analysis performed with detailed data on bottom–up emissions and absolute GDP of six industrialized countries as12:

= ± × ESO ,k (1.03 0.26) GDPk (3.3) where ESO,k represents solvent use emissions [Mt CO2 equivalents] in country or region k. The limited data availability for our regression analysis (coefficient of determination of R2 = 0.99) yields an error range of the slope coefficient, which we regard as too small given the overall uncertainty related to estimates for solvent use emissions. We therefore assume the error of the slope coefficient to be 25% based on our findings for waxes, paraffins, and surfactants (see below).

Similar to solvent use emissions, we also estimate emissions from the use of waxes and paraffins based on linear regression analysis. We identify in the first instance data on the consumption of waxes and paraffins in 31, mainly industrialized, countries and the 13 world as a whole . We apply an emission factor of 2.20 ± 0.63 kg CO2/kg of waxes and paraffins for calculating CO2 emissions (Patel et al., 1999; Ullmann, 1997; Weissermel and Arpe, 2003). In a second step, we plot CO2 emissions from the use of waxes and paraffins as a linear function of absolute GDP. Based on this analysis, we calculate with NEAT-SIMP emissions from the consumption of waxes and paraffins in country or region k as:

= ± × EWA,k (0.27 0.08) GDPk (3.4) where EWA,k [Mt CO2 equivalents] represents total emissions resulting from the consumption of waxes and paraffins. The coefficient of determination (R2) of the established relationship equals 0.81. The error of the slope coefficient represents the 95% confidence interval related to the oxidation rate of waxes and paraffins.

3.2.5 Estimating emissions from agriculture Non-energy use emissions in the source category agriculture result from the application of pesticides and urea fertilizers. We estimate emissions from pesticide use by multiplying activity data as given by FAO (2006) with an emission factor of 0.77 ± 0.45 kg CO2/kg of pesticide (e.g., Sonnenberg and Sietz, 2007). The error accounts for the 95% confidence

12 We include the following countries and years (in parentheses) Austria (2001), Germany (2001), France (2003), Italy (2002), The Netherlands (2003), and USA (2001). We uniformly convert solvent use emissions as given in NMVOC equivalents into CO2 equivalents by applying a conversion factor of 2.31 kg CO2/kg NMVOC (Schmidt-Stejskal et al., 2004). One exception is the USA, for which we apply a country specific conversion factor of 2.06 kg CO2/kg NMVOC (Freed et al., 2005). 13 Based on data provided by various NEU-CO2 network partners and UN (2000), we include in our analysis the following countries and years (in parentheses) Argentina (2001), Australia (1999), Austria (2000), Belgium (2000), Chile (2001), China (1999), Croatia (2001), Czech Republic (2001), Denmark (2000), Estonia (2000), Finland (2000), France (2001), Germany (2000), India (2000), Indonesia (2001), Ireland (2000), Italy (1997), Japan (1996), Lithuania (2000), Luxembourg (2000), Poland (2001), Romania (2000), South Korea (2000), The Netherlands (1999), Russia (1999), South Africa (2001), Spain (2000), Sweden (2000), Turkey (2001), USA (1999), and Venezuela (2001).

47 Chapter 3 interval of fossil carbon content (i.e., 35 ± 10%) and oxidation rate (i.e., 60 ± 30%) of pesticides. We estimate emissions from the application of urea fertilizers based on urea consumption data as given by IFA (2004). We assume a fossil carbon content of 0.73 kg CO2/kg of urea and a carbon oxidation rate of 100%.

3.2.6 Estimating emissions from waste Non-energy use emissions in the source category waste result mainly from the oxidation of surfactants during wastewater treatment14. We exclude from our analysis emissions from waste incineration with energy recovery because these emissions are accounted for under the emission source category energy. We estimate with NEAT-SIMP emissions from the use of surfactants based on linear regression analysis. We conduct in a first step bottom– up analysis of surfactant consumption in 19, mainly industrialized, countries15. We uniformly assume an average fossil carbon content of 27 ± 6% and a carbon oxidation rate of 100%, resulting in an emission factor of 0.99 ± 0.21 kg CO2/kg of surfactants (Patel, 1999). In a second step, we plot emissions from surfactant use in the various countries versus absolute GDP. Based on the results, we calculate with NEAT-SIMP emissions from the oxidation of surfactants in country or region k as:

= ± × ESU ,k (0.43 0.09) GDPk (3.5) where ESU,k [Mt CO2 equivalents] represents total emissions from surfactant use. The coefficient of determination (R2) of the established relationship equals 0.97. The error of the slope coefficient represents the 95% confidence interval of the average fossil carbon content of surfactants.

3.3 Results 3.3.1 Total non-energy use We quantify worldwide non-energy use in the year 2000 to be equivalent to 1,670 ± 120 Mt CO2. This result is consistent with official IEA (2005b) data (1,717 Mt CO2). Of this total, 64% (1,070 ± 80 Mt CO2) is consumed by Annex I countries and 36% (600 ± 100 Mt CO2) by non-Annex I countries. NEAT-SIMP estimates for non- energy use in the selected ten non-Annex I countries range from 175 ± 19 Mt CO2 in China (exceeding second-place South Korea by more than a factor of two) to 7 ± 1 Mt CO2 in Argentina (Figure 3.1).

14 The position of emissions from surfactant use is unclear in the current 1997 IPCC inventory guidelines (IPCC, 1997). Emissions can either be allocated to the source category solvent and other product use or waste. For reasons of clarity, we include emissions from surfactant use here as waste emissions. 15 Based on data provided by various NEU-CO2 network partners and UN (2000), we include in our analysis the following countries and years (in parentheses) Austria (2003), Belgium (2000), Denmark (2000), Finland (2000), France (2000), Germany (2003), India (2000), Ireland (2000), Italy (2000), South Korea (2000), Luxembourg (2000), The Netherlands (2003), Portugal (2000), Russia (2003), South Africa (1996), Spain (2000), Sweden (2000), UK (2000), and the USA (2004).

48 Non-energy use and related emissions in major developing countries

China (2000) South Korea (2000) India (2000) Brazil (2000) Saudi Arabia (2000) Mexico (2000) IEA NEAT-SIMP Indonesia (2000) IPCC-RA South Africa (2000) NEAT-Model Iran (2000) Argentina (2000) Germany (2000) Italy (1997) The Netherlands (1999)

0 50 100 150 200 Total non-energy use in Mt CO 2 Figure 3.1: Total non-energy use as calculated with NEAT-SIMP, the detailed NEAT model, and as reported by IEA (2005a, b) and IPCC-RA (UNFCCC, 2006a); numbers in parentheses indicate the year of analysis

For individual countries, differences between NEAT-SIMP estimates and official non-energy use data as stated by IEA (2005a, b) and UNFCCC (2006a) can be substantial (Figure 3.2). Recall that we expect NEAT-SIMP estimates to be in the upper range of IEA data (2005a, b). In other words, we expect countries to be either on or right of the zero line in Figure 3.2 because IEA data potentially follow a partial net or a pure net definition of non-energy use (i.e., they might partially or fully exclude fuel use of feedstock from non- energy use). Inconsistencies in the system boundary of non-energy use data may explain, for example, the positive deviations for Mexico and Iran (Figure 3.2). Excluding fuel use from the total non-energy use of both countries reduces the differences between NEAT- SIMP estimates and official IEA data from 52% and 30% to 31% and 14%, respectively. Earlier analyses with the detailed NEAT model reveal that the Netherlands and Germany apply a net and a partial net definition of non-energy use (Neelis et al., 2005a; Weiss et al. 2008b)16. For both countries, we explain deviations also with errors in energy statistics that lead to a substantial underestimation of non-energy use (Neelis 2006; Weiss et al. 2008b)17. Furthermore, data stated by IEA (2005a) and the national GHG inventory of the Netherlands deviate from each other by roughly 19% because methodologies applied to

16 The system boundaries of non-energy use in the German energy statistics are not defined consistently for individual feedstocks. A gross definition is applied for coal- and oil-based feedstock, whereas a net definition is used for natural gas (AGE, 2007). 17 Neelis (2006) quantifies errors for the Netherlands to be 53 PJ of pure chemical grade aromatics and 50 PJ of other chemical grade products that are erroneously excluded from non-energy use. Using conversion factors of 83 t CO2/GJ for chemical grade aromatics and 73 t CO2/GJ for other chemical grade products, we estimate the total error of non-energy use data in energy statistics to be 8.1 Mt CO2 equivalents, i.e., 30%.

49 Chapter 3 calculate non-energy use are partly incompatible (Figure 3.1)18. Park (2005) identified errors in earlier versions of the South Korean energy statistics. Here, double counting of process energy and/or backflows to refineries proved to be a source of major inconsistencies.

World (2000) Total non-Annex I countries (2000) Total Annex I countries (2000) China (2000) South Korea (2000) India (2000) Brazil (2000) Saudi Arabia (2000) Mexico (2000) Indonesia (2000) South Africa (2000) Iran (2000) Argentina (2000) Germany (2000) Italy (2000) The Netherlands (2000)

-60 -40 -20 0 20 40 60 80 100 Relative deviation in % Figure 3.2: Relative deviations of total non-energy use as estimated with NEAT-SIMP and as stated by IEA (2005a, b); positive deviations indicate that NEAT-SIMP estimates are higher than IEA data; negative deviations indicate that NEAT-SIMP estimates are lower than IEA data; numbers in parentheses indicate the year of analysis

We argue that differences in the system boundaries of non-energy use and to some extent errors in energy statistics are likely to explain large parts of the deviations observed in Figures 3.1 and 3.2. Detailed knowledge about the reporting of non-energy use in each individual country is, however, necessary to entirely explain deviations between NEAT- SIMP estimates and official IEA (2005a, b) data. This is also true for countries where IEA (2005b) data exceed NEAT-SIMP estimates (i.e., Brazil, South Africa, and Argentina; see Figure 3.2). One possible explanation for negative deviations is the potential underestimation of specific energy consumption of industrial processes in these countries by NEAT-SIMP. This might especially be the case for South Africa where basic chemicals such as ethylene and propylene are produced from coal-based, rather than from oil-based feedstock. Specific feedstock requirements for coal-based chemical production are

18 IEA (2005a) applies a pure net definition for non-energy use in the Netherlands with regard to industrial processes, whereas the IPCC-RA uses a net definition for feedstock consumption in the production of ammonia and methanol but a gross definition for feedstock use in steam crackers.

50 Non-energy use and related emissions in major developing countries considerably higher than for oil-based steam cracking, hence causing a potential underestimation of non-energy use in NEAT-SIMP by 30–50%.

3.3.2 Total non-energy use emissions As mentioned in Section 3.2.1, some parts of non-energy use remain stored in long-life products, while others become oxidized in industrial processes, during solvent and other product use, pesticide and fertilizer application, or wastewater treatment. We now present total non-energy use emissions. Afterward, we show NEAT-SIMP emission estimates on a more disaggregated level for all relevant source categories.

Worldwide non-energy use emissions as calculated with NEAT-SIMP (700 ± 90 Mt CO2) account for 1.5% of global GHG emissions and 2.6% of global CO2 emissions from fuel combustion in the year 2000. Our estimates are considerably lower than the data stated by the EDGAR emissions database (916 Mt CO2; Olivier, 2005b). We explain the differences as resulting from (i) deviations of underlying non-energy use data and (ii) methodological differences (see Footnote 2). In particular, the EDGAR methodology leads most likely to errors because for the majority of countries the system boundaries of non- energy use data and of the applied carbon storage fractions might be inconsistent. We therefore regard our results as more reliable estimates for global non-energy use emissions than the EDGAR data (Olivier 2005b).

According to the IPCC (1997) guidelines, non-energy use emissions are reported in national GHG inventories based on two independent methods, i.e., the simple top-down IPCC-Reference Approach (IPCC-RA) and the disaggregated and more detailed bottom- up IPCC-Sectoral Approach (IPCC-SA)19. Regarding non-energy use emissions in Annex I countries, we find relatively good agreement between NEAT-SIMP estimates (360 ± 50 Mt CO2 equivalents) and data from the IPCC-RA (386 Mt CO2; UNFCCC, 2006a). However, our estimates exceed data according to the IPCC-SA (179 Mt CO2) by a factor of two. This result provides an indication that NEAT-SIMP generates reliable estimates in comparison with IPCC-RA data. Based on previous insights obtained for Germany, Italy and the Netherlands (Weiss et al., 2008b; La Motta et al., 2005; Neelis et al., 2005a), we argue that IPCC-SA estimates are incomplete with regard to non-energy use emissions.

19 The IPCC-RA is a top–down approach using energy supply data (as stated by official energy statistics) mainly for calculating emissions from the combustion of fossil fuels (IPCC, 1997). The IPCC-RA is intended as a straightforward method for crosschecking emissions from energy and non-energy use as stated under the various source categories of the more detailed bottom-up IPCC-SA. In 2006, revised guidelines for GHG inventories were issued (IPCC, 2006), which will be adopted by most countries only in the post-Kyoto period after 2008-2012. The 1997 IPCC guidelines are therefore relevant for the discussion of non-energy use emissions in this chapter. The new guidelines do not allow for crosschecking non-energy use emissions between IPCC-RA and IPCC-SA any more because fuel-specific carbon storage fractions, which previously have been multiplied with non-energy use data for calculating non-energy use emissions in the IPCC-RA, are generally set to 100% (IPCC, 2006). Non-energy use emissions are therefore regarded as storage in the IPCC-RA and are only dealt with under the relevant source categories in the IPCC-SA. This makes verification of non-energy use emissions by means of internal quality checks within national GHG inventories virtually impossible.

51 Chapter 3

It is interesting to note that non-Annex I countries account for only 36% of global non-energy use but for 49% (340 ± 70 Mt CO2) of global non-energy use emissions. On the contrary, Annex I countries consume 64% of all fossil fuels used worldwide for non- energy purposes but release only 51% (360 ± 60 Mt CO2) of related CO2 emissions. Non- energy use emissions in China (122 ± 18 Mt CO2) exceed the ones in all other non-Annex I countries by far (Figure 3.3).

Among non-Annex I countries, also India, South Korea, and Brazil are major emitters of CO2 through non-energy use. Despite considerably lower non-energy use, India exceeds South Korean non-energy use emissions by 24 ± 5 Mt CO2. Our NEAT-SIMP results for the three selected Annex I countries show comparatively small deviations from detailed NEAT emission estimates. However, NEAT-SIMP results differ considerably from IPCC-SA estimates in GHG inventories (UNFCCC, 2006a). We explain these deviations to a large extent by incomplete emission estimates in the GHG inventories of the selected countries (see, e.g., La Motta et al., 2005; Neelis et al., 2005a; Weiss et al., 2008b). In the next section, we obtain deeper insight into non-energy use emissions by disaggregating NEAT-SIMP estimates according to the various relevant emission source categories.

China (2000) South Korea (2000) India (2000) Brazil (2000)

Saudi Arabia (2000) NEAT-SIMP EDGAR Mexico (2000) NEAT model Indonesia (2000) IPCC-RA IPCC-SA South Africa (2000) Iran (2000) Argentina (2000) Germany (2000) Italy (1997) The Netherlands (1999)

0 20 40 60 80 100 120 140 Total non-energy use emissions in Mt CO 2 Figure 3.3: Total yearly non-energy use CO2 emissions as modeled with NEAT-SIMP, the detailed NEAT model, EDGAR, and as given in the IPCC-RA and IPCC-SA in national GHG inventories (UNFCCC, 2006a); numbers in parentheses indicate the year of analysis

52 Non-energy use and related emissions in major developing countries

3.3.3 Breakdown of non-energy use emissions Industrial processes contribute worldwide 522 ± 84 Mt CO2, i.e., 74% to non-energy use emissions. Solvent and other product use accounts for 13% of worldwide non-energy use emissions, whereas emissions from agriculture and waste constitute shares of only 10% and 3%, respectively (Figure 3.4).

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Industrial processes - steam cracking Industrial processes - ammonia production Industrial processes - methanol production Industrial processes - carbon black production Industrial processes - aluminium production Industrial processes - electrodes and other solid carbon use (excluding aluminium production) Solvent and other product use - consumption of solvents Solvent and other product use - consumption of lubricants Solvent and other product use - consumption of w axes and paraffins Agriculture - application of pesticides Agriculture - application of urea fertilizers Waste - consumption of surfactants

Figure 3.4: Shares of individual source categories in total non-energy use emissions; numbers in parentheses indicate the year of analysis

Total industrial process emissions in Annex I countries and non-Annex I countries are estimated to be 279 ± 46 Mt CO2 and 243 ± 74 Mt CO2, respectively. Steam cracking and ammonia production are by far the largest sources of industrial process emissions20. The overall shares of emissions resulting from the production of methanol and carbon black, as well as from the use of electrodes and other solid carbon in the manufacturing of

20 An exception is Brazil, where electrodes and other solid carbon use in the manufacturing of non-ferrous metals, ferroalloys, and inorganic chemicals is the major source for industrial process emissions (4.6 ± 2.1 Mt CO2).

53 Chapter 3 non-ferrous metals, ferroalloys, and inorganic chemicals are relatively similar in Annex I and non-Annex I countries (Figure 3.4). This is, however, not the case for steam cracking and ammonia production. In Annex I countries, steam cracking and ammonia production account for 28% and 39% of industrial process emissions, respectively. In non-Annex I countries, the shares are 17% for steam cracking and 52% for ammonia production. We explain this finding with specific differences in the structure of the economy, in particular, the chemical industry. Feedstock use for steam cracking accounts for a larger proportion of total non-energy use in Annex I countries than in non-Annex I countries. Propylene and ethylene produced in steam crackers are used as building blocks for the synthesis of polymers, which do not oxidize during use and hence remain stored in the anthroposphere. By contrast, in ammonia production the feedstock carbon (typically contained in natural gas or heavy oils) becomes entirely oxidized released as industrial process emissions. Parts of these process emissions are sequestered for urea production. The carbon contained in urea is ultimately emitted during the application of urea fertilizers (as emissions in agriculture, see below) in the relatively large agricultural sector of non-Annex I countries. Furthermore, carbon sequestration for urea production in Annex I countries and trade of urea fertilizers to non-Annex I countries reduce industrial process emissions from ammonia production in Annex I countries but add to the non-energy use emissions in non- Annex I countries. These findings also explain, why non-Annex I countries consume only one third of worldwide non-energy use but are responsible for about half of the global non-energy use emissions.

Our estimates for industrial process emissions in individual non-Annex I countries range from 91 ± 18 Mt CO2 (China) to 3 ± 1 Mt CO2 (Argentina). Regarding the sources of industrial process emissions, we find large variations between individual countries (Table A3.1 in the Appendix of this chapter). The reasoning presented above provides not only an explanation for differences between non-Annex I and Annex I countries but also for the observed variation among individual non-Annex I countries, e.g., for the relatively high industrial process emissions in India and the relatively low industrial process emissions in South Korea (see Figures 3.3 and 3.4).

For the whole of Annex I countries, NEAT-SIMP estimates are 79% higher than official IPCC-SA estimates. This deviation is, to a large extent, caused by underestimation of industrial process emissions in official GHG inventories (UNFCCC, 2006a). We explain differences between NEAT-SIMP and NEAT regarding emission estimates for steam cracking and the production of ammonia, methanol, and carbon black with deviations in the assumed system boundaries of non-energy use (see Section 3.2.1). NEAT-SIMP results differ considerably from NEAT estimates with respect to emissions from the use of electrodes and other solid carbon for the manufacturing of non-ferrous metals, ferroalloys, and inorganic chemicals. We attribute these differences to the relatively crude method used in NEAT-SIMP for estimating emissions from this source category based on aluminium production and GDP.

Worldwide emissions from solvent and other product use amount to 91 ± 26 Mt CO2 equivalents (Table A3.2 in the Appendix of this chapter). Annex I countries account for 63% of global solvent and other product use emissions. Among non-

54 Non-energy use and related emissions in major developing countries

Annex I countries, China is the largest emitter of CO2 from solvent and other product use (8.2 ± 2.0 Mt), whereas Saudi Arabia is the smallest one (0.5 ± 0.1 Mt). Emissions from the application of pesticides and urea fertilizers in agriculture amount worldwide to 72 ± 7 Mt CO2. Non-Annex I countries contribute 80% to this total. We estimate global waste emissions from the oxidation of surfactants during wastewater treatment to be 19 ± 9 Mt CO2. Annex I countries and non-Annex I countries contribute 57% and 43%, respectively to this total (see Table A3.2 in the Appendix of this chapter).

Disaggregating non-energy use emissions according to the various source categories indicates (i) a shift of feedstock use from ammonia production towards steam cracking in the industrialized Annex I countries and (ii) the importance of the agricultural sector as a consumer of chemicals and as an emitter of CO2 from non-energy use in non- Annex I countries.

3.4 Discussion NEAT-SIMP is a simple yet useful model for providing an overview of non-energy use and related emissions. For our model estimations we make several assumptions, which we explain in Section 3.2. As for any other model approach, the quality of NEAT-SIMP results depends on (i) the accuracy of model input data and (ii) the reliability of intrinsic model assumptions. We now discuss both aspects separately from each other.

Addressing the first point (i.e., the accuracy of model input data), for most of the non-Annex I countries, complete and reliable data sets for chemicals production are not available to us. We therefore approximate production wherever necessary (e.g., for ethylene, methanol, and carbon black) by production capacities, assuming a capacity utilization rate of 90 ± 10%. For modeling non-energy use of refinery products (e.g., bitumen and lubricants), we approximate consumption by domestic production capacities. This approach neglects trade of refinery products, which is accounted for in official IEA energy statistics (IEA 2005a, b). Disregarding trade might cause under- or over-estimation of non-energy use and resulting CO2 emissions for small countries where net trade flows can be substantial. Neglecting trade of refinery products can hence be expected to introduce only minor uncertainties to NEAT-SIMP estimates for the ten, relatively large, non-Annex I countries analyzed here. Furthermore, feedstock distribution for ammonia and methanol production is a particular point of uncertainty. Unless more detailed information is available (i.e., for China, India, Korea, South Africa, Germany, Italy, and the Netherlands), we use world averages for the feedstock distribution of ammonia production21 and we assume natural gas to be the only feedstock for methanol production. The accuracy and precision of NEAT-SIMP estimates could be improved, if detailed country-specific data on actual production, consumption, and feedstock distribution were used as model input.

21 We assume that the feedstock for ammonia production consists to 84% of natural gas, 8% of heavy fuel oils, and 8% of coal (IPTS, 2004).

55 Chapter 3

Apart from the quality of model input data, intrinsic model assumptions made in NEAT-SIMP are subject to uncertainties. This applies specifically to: (i) the emission factors applied for chemical processes, (ii) the division of carbon fractions, which are oxidized versus non-oxidized during solvent and other product use, and (iii) the estimation of parts of non-energy use and related emissions based on regression analysis.

NEAT-SIMP might under- or over-estimate feedstock use and resulting CO2 emissions for countries (e.g., South Africa), where ethylene is not produced in steam crackers from oil-derived feedstock. Underestimation of feedstock use results, for example, if ethylene production (i) is based on coal-derived naphtha or (ii) is made via separation from coke oven gas. Overestimation, on the other hand, might be a consequence, if ethylene is produced via dehydration of bio-based ethanol. This production route still provides a supplement to conventional ethylene streams in many African and South American countries (Weissermel and Arpe, 2003). Knowledge regarding the system boundaries of non-energy use is critical for correct emissions accounting. We explain the differences between NEAT-SIMP and NEAT regarding industrial process emissions (see Table A3.1, in the Appendix of this chapter) with differences in the system boundaries applied for non-energy use. Acquiring insight into the system boundaries of non-energy use as stated in the energy statistics of individual countries and adapting NEAT-SIMP accordingly can improve the accuracy of emission estimates with minimal effort. The issue of system boundaries also relates to the accounting of emissions from waste incineration. We exclude these emissions here because they are generally accounted for under the source category energy, if waste is used as fuel. However, both waste incineration without energy recovery and open burning of waste might be sources for substantial amounts of CO2 emissions in non-Annex I countries. We recommend additional research on this particular point of uncertainty.

The division between fractions of carbon oxidized versus carbon stored during product use is rather clear-cut for, e.g., solvents and surfactants because 100% of the carbon originally contained in these products is oxidized during product use or wastewater treatment. This partitioning is, however, less straightforward for lubricants. In NEAT- SIMP we assume 30 ± 20% of all lubricants to be oxidized during the use phase based on a literature study done for Germany (Trischler, 1997; Patel et al., 1999). The error range implemented in NEAT-SIMP accounts for regional variability, i.e., assuming lubricant oxidation rates in non-Annex I countries to be higher than in Germany due to the abundance of two-stroke engines and a larger share of lubricants being attributed to losses, leakages, or dissipative use. Detailed bottom-up analysis is required to determine the exact share of carbon that becomes oxidized during lubricant use in each individual country. By assuming the extreme case of 50% lubricant oxidation in Annex I countries and 100% lubricant oxidation in non-Annex I countries, worldwide solvent and other product use emissions would increase by 43% and total non-energy use emissions would increase by 6%. Based on this finding, we strongly recommend detailed analysis on the use and fate of lubricants at the level of individual countries.

56 Non-energy use and related emissions in major developing countries

Estimating non-energy use and related CO2 emissions based on regression analysis, as it is done for (i) electrodes and other solid carbon use in the manufacturing of non- ferrous metals, ferroalloys, and inorganic chemicals and (ii) the consumption of solvents, waxes, paraffins, as well as surfactants adds uncertainty to our results. Crosschecking NEAT-SIMP estimates indicates relatively good correspondence with estimates from IEA (2007b), which identified worldwide use of solvents and surfactants in the year 2004 to be 20 Mt (46.2 Mt CO2 equivalents) and 15 Mt (34.9 Mt CO2 equivalents), respectively. However, the identified linear relationships rely on data for mainly industrialized countries. Therefore, NEAT-SIMP might not always estimate emissions correctly for individual non-Annex I countries22. We argue that NEAT-SIMP estimates in the source categories solvent and other product use and waste could be improved, if detailed consumption data for individual countries were available.

A review of GHG inventories submitted in 2006 by Annex I countries reveals a clear need to improve official estimates on solvent and other product use emissions (UNFCCC, 2006a). To this end, detailed bottom–up studies are urgently needed. Such analyses are, however, very time and resource intensive and were done in the past only for a few Annex I countries, e.g., Austria and Germany (Schmidt- Stejskal et al., 2004; Theloke et al., 2000). With NEAT-SIMP we provide an important step in that direction by (i) calculating solvent and other product use emissions disaggregated according to the three most important source categories and by (ii) identifying and quantifying other sources of product use emissions than solvent use alone. We furthermore quantify additional sources of non-energy use emissions, i.e., in the categories agriculture and waste, thereby contributing to a more complete accounting of non-energy use emissions.

On average, 5% of total non-energy use and 21% of non-energy use emissions are calculated in NEAT-SIMP based on regression analysis and not by means of activity data. Whereas this share is rather small in the case of total non-energy use, it constitutes one fifth of non-energy use emissions and on average 71% of total emissions in the categories of solvent and other product use and waste. Given the overall uncertainties related to the calculation of non-energy use emissions and taking the scope of this research into account, we regard our model approach as justified. With NEAT-SIMP, we first provide bottom-up estimates of non-energy use and related CO2 emissions for the world as a whole, for the total of Annex I and non-Annex I countries, as well as for selected non-Annex I countries individually with an average total uncertainty of 13%. We are confident that our results represent non-energy use and associated emissions correctly albeit with varying uncertainties depending on individual countries and source categories analyzed.

22 Note that the relationships identified by our regression analyses are strictly valid only for a defined time period. This is especially true for our analysis on solvent use emissions. Despite continuous growth of GDP, emissions from solvent use decreased substantially in the past 15 years in many Annex I countries as a consequence of technological progress, material substitution, and successful mitigation policies. The modeling of a linear relationship between solvent use emissions and GDP is hence a simplification of the actual relationship between both parameters. This shortcoming can be solved in the first instance by re- estimating regression coefficients for each particular time period analyzed or by conducting direct bottom-up studies on solvent consumption in the country of interest.

57 Chapter 3

3.5 Conclusions In this chapter, we develop and apply a simple bottom-up model (NEAT-SIMP) to estimate total non-energy use and associated CO2 emissions for the world as a whole and for selected non-Annex I and Annex I countries. NEAT-SIMP produces reliable estimates for total non-energy use and for non-energy use emissions resulting from the most important industrial processes, i.e., steam cracking and the production of ammonia, methanol, and carbon black. The model has its limitations when estimating emissions from (i) the production of non-ferrous metals, ferroalloys, and inorganic chemicals and (ii) the use of solvents, waxes, paraffins, and surfactants. We estimate emissions for these sources based on linear regression analysis, using data mainly from industrialized countries. Uncertainties also arise from the differentiation of carbon fractions, which are oxidized versus non-oxidized during lubricants consumption. These limitations can be overcome by detailed and hence more resource and data intensive bottom–up analyses. The difficulties we face with NEAT-SIMP in the accounting of non-energy use emissions are also reflected by the data in official GHG inventories (UNFCCC, 2006a), which are often erroneous and incomplete. Based on the insight gained in the course of our research, we conclude that NEAT-SIMP is useful (i) for estimating non-energy use emissions for non- Annex I countries, which have not yet developed detailed GHG inventories and (ii) for crosschecking existing inventory data in Annex I countries. Our research deepens the understanding of worldwide and country-specific non-energy use and related CO2 emissions, thereby contributing to the preparation of correct, consistent, and comprehensive GHG emission inventories. We recommend NEAT-SIMP to inventory experts as a valuable and easily applicable model for assessing non-energy use and related CO2 emissions for any country in the world.

Acknowledgements This research was funded by the European Commission under the 6th framework program on ‘Sustainable development, global change, and ecosystems’; contract number FP6- 50345 (International network non-energy use and CO2 emissions (NEU-CO2), phase III). The authors would like to thank all network participants for their support regarding data collection and methodological improvements of NEAT-SIMP. We are grateful to Jos Olivier (Netherlands Environmental Assessment Agency, Bilthoven, the Netherlands) and Alexandra Newman (Colorado School of Mines, Golden, USA) for their comments on earlier drafts of this chapter.

Appendix Table A3.1: Industrial process emissions as calculated with NEAT-SIMP, the detailed NEAT model, and as given according to IPCC-SA (UNFCCC, 2006a); numbers in parentheses indicate the year of analysis

Emissions in Mt CO2 Other non-ferrous metals, Region (Year) Estimate Steam Ammonia Methanol Carbon black Aluminium ferroalloy, and inorganic Total cracking productiona production production production chemicals production Non-energy use and related emissionsinmajordevelopingcountries Non-energy useandrelated World (2000) NEAT-SIMP 119 ± 18 233 ± 61 23 ± 14 30 ± 12 38 ± 4 79 ± 52 522 ± 84 59 Non-Annex I countries (2000) NEAT-SIMP 41 ± 18 125 ± 64 14 ± 8 16 ± 9 15 ± 4 33 ± 30 243 ± 74 Annex I countries (2000) NEAT-SIMP 78 ± 16 108 ± 24 9 ± 6 15 ± 6 24 ± 2 46 ± 31 279 ± 44 IPCC-SA - 85 2 5 29 15 156d China (2000) NEAT-SIMP 8.3 ± 1.7 63.3 ± 16.9 5.0 ± 1.6 1.8 ± 0.7 4.2 ± 0.4 8.3 ± 5.6 91.0 ± 17.9 Korea (2000) NEAT-SIMP 9.5 ± 0.9 0.2 ± 0.1 0.0 1.0 ± 0.4 0.0 0.8 ± 0.3 11.5 ± 1.0 NEATb 9.5 0.3 0.0 0.9 0.0 0.5 11.1 India (2000) NEAT-SIMP 3.5 ± 0.5 10.4 ± 4.7 0.2 ± 0.1 0.8 ± 0.4 0.9 ± 0.1 3.4 ± 1.8 19.2 ± 5.1 Brazil (2000)c NEAT-SIMP 3.2 ± 0.6 2.0 ± 0.5 0.0 1.0 ± 0.4 1.9 ± 0.2 2.7 ± 2.1 10.9 ± 2.3 Saudi Arabia (2000) NEAT-SIMP 4.0 ± 0.8 2.0 ± 0.8 1.8 ± 1.3 0.0 0.0 0.3 ± 0.1 8.2 ± 1.8 Mexico (2000) NEAT-SIMP 2.4 ± 0.5 4.2 ± 0.8 0.2 ± 0.1 0.4 ± 0.2 0.9 ± 0.1 1.0 ± 0.5 8.3 ± 1.1 Indonesia (2000) NEAT-SIMP 1.0 ± 0.2 5.4 ± 1.9 0.5 ± 0.3 0.4 ± 0.2 0.29 ± 0.03 0.9 ± 0.5 8.5 ± 2.1 South Africa (2000) NEAT-SIMP 0.3 ± 0.1 4.0 ± 0.8 0.0 0.2 ± 0.1 1.0 ± 0.1 1.1 ± 1.0 6.7 ± 1.3 Iran (2000) NEAT-SIMP 1.1 ± 0.2 1.7 ± 0.5 0.4 ± 0.3 0.3 ± 0.1 0.21 ± 0.02 0.6 ± 0.3 4.2 ± 0.7 Argentina (2000) NEAT-SIMP 0.3 ± 0.1 1.5 ± 0.3 0.0 0.3 ± 0.1 0.39 ± 0.04 0.8 ± 0.5 3.2 ± 0.6

60 Chapter 3

Table A3.1 (cont.): Industrial process emissions as calculated with NEAT-SIMP, the detailed NEAT model, and as given according to IPCC-SA (UNFCCC, 2006a); numbers in parentheses indicate the year of analysis

Region (Year) Emissions in Mt CO2 Other non-ferrous metals, Estimate Steam Ammonia Methanol Carbon black Aluminium ferroalloy, and inorganic Total cracking Productiona) production production production chemicals production Germany (2000) NEAT-SIMP 9.2 ± 0.8 5.1 ± 0.6 2.4 ± 0.2 1.0 ± 0.3 1.0 ± 0.1 3.1 ± 1.8 21.7 ± 2.0 NEATb 8.4 3.9 2.1 0.6 0.9 2.4 18.2 IPCC-SA - 4.8 2.1 0.6 0.9 0.02 8.4 Italy (1997) NEAT-SIMP 3.7 ± 0.3 0.6 ± 0.1 0.0 0.5 ± 0.1 0.28 ± 0.03 1.7 ± 0.8 6.8 ± 0.9 NEATb 3.3 0.5 - 0.4 0.3 0.4 4.8 IPCC-SA - 0.5 - 0.5 0.3 0.4 1.7 the Netherlands (1999) NEAT-SIMP 5.1 ± 0.5 5.0 ± 0.6 0.4 ± 0.1 0.3 ± 0.1 0.43 ± 0.04 0.8 ± 0.5 12.0 ± 1.0 NEATb 0.0 2.9 0.0 0.0 0.4 0.7 4.0 IPCC-SA - 3.6 - - 0.4 0.1 4.1 a For NEAT-SIMP, we report gross emissions from ammonia production, including emissions from fuel use of feedstock but excluding CO2 sequestered for urea production. b NEAT estimates follow the definition of non-energy use as applied in national energy statistics. The differences between NEAT-SIMP and NEAT emission estimates for steam cracking, as well as for the production of ammonia, methanol, and carbon black are caused (i) by deviations regarding the applied system boundaries for non-energy use and (ii) by assuming conservative rather then average emission factors in the detailed NEAT model. c There is no methanol production from fossil feedstock in Brazil. d For the total of Annex I countries, we find in the IPCC-SA roughly 20 Mt CO2 emissions reported under industrial processes, which cannot be allocated to either of the source categories specified in this table. For this reason, the IPCC-SA total as given here in the last column is not equal to the sum of emissions reported for the individual source categories.

Table A3.2: Emissions from solvent and other product use, agriculture, and waste as calculated with NEAT-SIMP, the detailed NEAT model, and as given according to IPCC-SA (UNFCCC, 2006a); numbers in parentheses indicate the year of analysis Region (Year) Emissions from solvents and other product use Emissions from agriculture Emissions from waste in Mt CO a in Mt CO in Mt CO Totals Estimate 2 2 2 Waxes and Urea in Mt CO Solvents Lubricants Pesticides Surfactants 2 paraffins fertilizers World (2000) NEAT-SIMP 44 ± 11 35 ± 23 11 ± 3 2 ± 1 69 ± 7 18 ± 4 181 ± 27

Non-Annex I countries (2000) NEAT-SIMP 19 ± 7 9 ± 6 5 ± 2 1.5 ± 0.9 56 ± 6 8 ± 3 99 ± 12 emissionsinmajordevelopingcountries Non-energy useandrelated NEAT-SIMP 25 ± 6 26 ± 17 7 ± 2 0.9 ± 0.5 13 ± 1 11 ± 2 82 ± 19 Annex I countries (2000) IPCC-SA1) 23b - - -c -c -c - China (2000) NEAT-SIMP 4.6 ± 1.2 2.4 ± 1.6 1.2 ± 0.3 1.0 ± 0.6 20.4 ± 2.0 1.9 ± 0.4 31.5 ± 3.0 NEAT-SIMP 0.7 ± 0.2 0.8 ± 0.5 0.17 ± 0.05 0.02 ± 0.01 0.8 ± 0.1 0.3 ± 0.1 2.7 ± 0.6 Korea (2000) NEAT - 0.86 0.08 - 0.76 - 9.4d India (2000) NEAT-SIMP 2.3 ± 0.6 1.2 ± 0.8 0.6 ± 0.2 0.03 ± 0.02 14.1 ± 1.4 1.0 ± 0.2 19.2 ± 1.7 Brazil (2000) NEAT-SIMP 1.2 ± 0.3 0.7 ± 0.4 0.31 ± 0.09 0.10 ± 0.06 1.6 ± 0.2 0.5 ± 0.1 4.4 ± 0.6 Saudi Arabia (2000) NEAT-SIMP 0.25 ± 0.06 0.15 ± 0.10 0.06 ± 0.02 0.003 ± 0.002 0.33 ± 0.03 0.10 ± 0.02 0.9 ± 0.1 Mexico (2000) NEAT-SIMP 0.8 ± 0.2 0.7 ± 0.5 0.21 ± 0.06 0.04 ± 0.02 1.0 ± 0.1 0.34 ± 0.07 3.1 ± 0.5 Indonesia (2000) NEAT-SIMP 0.6 ± 0.2 0.4 ± 0.2 0.15 ± 0.04 0.001 ± 0.001 2.8 ± 0.3 0.25 ± 0.05 4.2 ± 0.4 South Africa (2000) NEAT-SIMP 0.4 ± 0.1 0.3 ± 0.2 0.10 ± 0.03 0.04 ± 0.02 0.37 ± 0.04 0.17 ± 0.05 1.4 ± 0.2 Iran (2000) NEAT-SIMP 0.4 ± 0.1 0.7 ± 0.4 0.09 ± 0.03 0.008 ± 0.005 1.1 ± 0.1 0.15 ± 0.03 2.4 ± 0.5 Argentina (2000) NEAT-SIMP 0.4 ± 0.1 0.22 ± 0.15 0.11 ± 0.03 0.10 ± 0.06 0.52 ± 0.05 0.18 ± 0.04 1.6 ± 0.2 NEAT-SIMP 2.0 ± 0.5 1.0 ± 0.7 0.5 ± 0.2 0.04 ± 0.02 0.7 ± 0.1 0.8 ± 0.2 5.1 ± 0.9 Germany (2000) NEAT 1.7 ± 0.3 1.0 ± 0.2 1.0 ± 0.2 0.04 ± 0.01 0.7 ± 0.1 0.9 ± 0.3 5.3 ± 0.4 IPCC-SA 2.0b - - - 0.6 -e - NEAT-SIMP 1.3 ± 0.3 0.5 ± 0.3 0.3 ± 0.1 0.07 ± 0.04 0.8 ± 0.1 0.5 ± 0.1 3.5 ± 0.5 Italy (1997) NEAT - 0.6 1.1 - 0.8 - 4.6d IPCC-SA 1.3b - - - -e -e - 61 NEAT-SIMP 0.4 ± 0.1 0.2 ± 0.1 0.10 ± 0.03 0.009 ± 0.005 0.20 ± 0.02 0.17 ± 0.04 1.1 ± 0.2 the Netherlands (1999) NEAT - 0.2 0.2 - 0.1 - 2.7d IPCC-SA 0.4b - - - -e -e - a b For conversion of NMVOC emissions as given in the IPCC-SA into CO2 equivalents see Footnote 12 Specified as emissions from solvent and other product use c Not uniformly specified among Annex I countries d Total ‘oxidize during use’ emissions e Not specified, included elsewhere

4 Non-energy use and related carbon dioxide emission in Germany: a carbon flow analysis with the NEAT model for the period of 1990-2003

Martin Weiss, Maarten L. Neelis, Kornelis Blok, and Martin K. Patel Published in: Resources, Conservation and Recycling 52 (2008), pp. 1252-1265.

Abstract Non-energy use of fossil fuels accounts for 7% of the total primary energy supply of Germany and potentially represents an important source of carbon dioxide (CO2) emissions. To gain a better understanding of emissions associated with non-energy use in Germany, we conduct a bottom-up carbon flow analysis with the Non-energy use Emission Accounting Tables (NEAT) model for the period from 1990 to 2003. We calculate average yearly non-energy use emissions to be 25 ± 2 Mt CO2, of which 77% are related to industrial processes, 17% to solvent and other product use, 2% to the use of urea fertilizers and pesticides in agriculture, and 4% to wastewater treatment. The comparison of our NEAT estimates with official data reveals gaps and errors in the German greenhouse gas (GHG) inventory. Our research highlights the difficulties associated with the accounting of non-energy use emissions not only in Germany but in other countries as well. To ensure correct calculation of non-energy use emissions, we recommend that inventory experts (i) obtain detailed insight into the system boundaries of non-energy use data as stated in national energy statistics, (ii) allocate accordingly non-energy use emissions to the relevant emission source categories (i.e., energy, industrial processes, solvent and other product use, agriculture, or waste), (iii) ensure completeness of emission estimates, and (iv) be cautious with the use of default emission factors as given by the Intergovernmental Panel on Climate Change.

63 Chapter 4

4.1 Introduction In the greenhouse gas (GHG) inventory of Germany (UNFCCC, 2005a, c), most attention has been paid to CO2 emissions originating from fossil fuel combustion. In Germany, approximately 7% of all fossil fuels are, however, not used for energy but for non-energy purposes, e.g., as feedstock in the chemical industry or for the production of lubricants and bitumen in refineries (IEA, 2005b). Non-energy use is therefore an important potential source of carbon dioxide (CO2) emissions. Moreover, the importance of non-energy use in Germany has increased substantially and is expected to grow further mainly due to the expansion of polymer production (more than 75% growth between 1990 and 2003). In this chapter, we define non-energy use as the sum of two components1: (i) The consumption of fossil fuels as feedstock in the chemical industry (e.g., the use of naphtha for olefins and aromatics production in steam crackers or the consumption of natural gas for the production of ammonia). (ii) The consumption of refinery and coke oven products as well as other solid carbon for non-energy purposes (e.g., the use of lubricants for transportation, the use of bitumen in the building sector, or the consumption of electrodes for aluminium production).

The non-energy use of fossil fuels leads to non-energy use emissions (mainly in the form of CO2) in various ways: (i) due to partial or complete oxidation of feedstock, electrodes, and other solid carbon during production processes in the chemical and non-ferrous metal industry (ii) due to product use of, e.g., solvents or lubricants (iii) due to the application of urea fertilizers and pesticides in agriculture (iv) due to the oxidation of surfactants in the course of wastewater treatment2

Estimating non-energy use emissions is not straightforward because only parts of the carbon initially contained in non-energy use are emitted during production, consumption, and disposal of materials, whereas a remainder is stored in products with lifetimes ranging from years to decades and longer. Various parallel and subsequent conversion steps in the chemical industry as well as multiple forms of chemicals’ use and product life cycles further complicate the accurate accounting of non-energy use emissions.

According to the inventory guidelines issued by the Intergovernmental Panel on Climate Change in 1997, non-energy use emissions are calculated by two methods that are

1 The consumption of coal and cokes in blast furnaces for pig iron production is part of the energy conversion sector in German energy statistics. We follow this practice and exclude these items from the non- energy use of fossil fuels. 2 Potential sources of non-energy use emissions from waste treatment also include landfilling and waste incineration (with and without energy recovery). In line with UBA (2004), we regard the oxidation of fossil- based carbon in landfills as negligible. Waste incineration without energy recovery does not take place in Germany. Emissions from waste incineration with energy recovery are reported as secondary fuel use emissions under the source category of energy in the German GHG inventory (UNFCCC, 2005a, c). We therefore exclude these emissions from our non-energy use emission estimates.

64 Non-energy use and related emissions in Germany the relatively simple top-down Reference Approach (IPCC-RA) and the more detailed bottom-up Sectoral Approach (IPCC-SA; IPCC, 1997)3. In 2006, the IPCC issued new guidelines for national GHG inventories in which (i) the accounting of non-energy use emissions in the IPCC-RA was practically abolished and (ii) the IPCC-SA has been refined (IPCC, 2006). Nevertheless, many countries, among them Germany, make use of their right to report their GHG emissions according to the 1997 IPCC guidelines until the end of the first Kyoto period in 2008-2012. The 1997 IPCC guidelines are therefore relevant for the discussion of non-energy use emissions in this chapter4.

For monitoring GHG emissions in Germany (and in all other Annex I countries), the IPCC-SA is the standard approach, whereas the IPCC-RA plays an important role as crosscheck. By comparing non-energy use emissions as calculated according to IPCC-RA and IPCC-SA, inventory makers discovered substantial inconsistencies within the German GHG inventory (UBA, 2003, 2004). Our recalculation of these inconsistencies is shown in Figure 4.1.

The results according to the IPCC-RA exceeded the ones calculated with the IPCC- SA by a factor 3-5 (UNFCCC, 2004, 2005a). This raised questions about the real level of non-energy use emissions in Germany. An international review team appointed by the UNFCCC also recognized inconsistencies in the German GHG inventory and pinpointed incomplete emission estimates for various source categories in the IPCC-SA, among them for chemical processes (UNFCCC, 2005b).

Given these problems, the Federal Environmental Agency of Germany (Umweltbundesamt, UBA) commissioned Utrecht University to apply its NEAT model to arrive at independent estimates of yearly non-energy use emissions for Germany in the period of 1990–2003 (UBA, 2003)5. This gave us the interesting though challenging opportunity to analyze in detail the strengths and weaknesses of the IPCC-RA and IPCC- SA as used in the German GHG inventory.

3 As a top-down method, the IPCC-RA calculates fuel-specific non-energy use emissions by multiplying non-energy use data from energy statistics and fuel-specific carbon storage fractions. The more detailed IPCC-SA, in turn, is a bottom-up method for calculating non-energy use emissions based on three methodologies (Tier 1 to Tier 3) differing in their level of detail and accuracy. In the simplest case (Tier 1), emissions are calculated, e.g., by multiplying activity data with average default emission factors. Non- energy use emissions are reported in the IPCC-SA according to various different source categories (i.e., industrial processes, solvent and other product use, agriculture, and waste). 4 We make exemptions from this general rule by calculating emissions that result from (i) industrial processes, (ii) solvent and other product use, (iii) the application of urea fertilizers in agriculture, and (iv) wastewater treatment at a greater level of detail than required by IPCC (1997). 5 Earlier model versions were applied to Italy, South Korea, and the Netherlands and provided the necessary insight to substantially improve the quality of national GHG inventories in these countries (La Motta et al., 2005; Park, 2005; Neelis et al., 2005a).

65 Chapter 4

25 IPCC-RA (2005) IPCC-SA (2005) 20

2 15

in Mt CO in Mt 10

5 Yearly non-energy use emissions use emissions non-energy Yearly

0 19901991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Figure 4.1: Yearly non-energy use emissions as calculated according to IPCC-RA and IPCC- SA of the German GHG inventory for the period between 1990 and 2003 (data source: UNFCCC (2005a))6

The objective of this chapter is to apply an improved version of the Non-energy use Emission Accounting Tables (NEAT) model for calculating non-energy use and related emissions and to compare our results with emission estimates according to the German GHG inventory. Based on our results, we give detailed recommendations on how to improve data consistency between the IPCC-RA and the IPCC-SA. Our research is not only valuable for improving the quality of the German GHG inventory but it also provides critical knowledge for assuring consistent emissions reporting in the GHG inventories of any other Annex-I country.

This chapter is structured as follows: In the next section, we explain methodology and data sources of our NEAT model. In Section 4.3, we present our model results and compare them with official emission data according to IPCC-RA and IPCC-SA of the German GHG inventory. In Section 4.4, we discuss our findings, address model uncertainties, and give advice on critical aspects of non-energy use emissions accounting. The chapter ends with conclusions and general recommendations for inventory experts.

4.2 Methodology and data sources

NEAT is a spreadsheet-based model, for estimating non-energy use and related CO2 emissions based on a carbon flow and mass balance approach. The NEAT model

6 We include here solvent and other product use emissions. For this, we multiply emission estimates given according to the IPCC-SA in non-methane volatile organic compound (NMVOC) equivalents with a conversion factor of 2.31 kg CO2/kg NMVOC equivalents (Schmidt-Stejskal et al., 2004).

66 Non-energy use and related emissions in Germany calculations are based mainly on official production and trade statistics and are, to a large extent, independent of data published in national energy statistics. The non-energy use emissions calculated with NEAT can be allocated to four principle IPCC-SA source categories: (i) industrial processes (ii) solvent and other product use (iii) agriculture (iv) waste

A detailed description of the NEAT model (NEAT 2.0) can be found in Neelis et al. (2005b). Here, we only explain key features and model adaptations made in the new model version of NEAT that is used for our analysis. In comparison to NEAT 2.0, we track in the improved version of NEAT non-energy use emissions closer to their actual source7 thereby calculating more reliable emission estimates. We account for uncertainties related to (i) production and trade data, (ii) emission factors, (iii) feedstock distribution, (iv) the carbon content of chemicals, and (v) the fractions of carbon that become oxidized during product use and wastewater treatment. We assume that the implemented uncertainty intervals represent the 95% confidence interval of possible values. We calculate uncertainty intervals for total non-energy use, carbon storage, total non-energy use emissions, as well as for each source category of non-energy use emissions individually by uniformly applying standard error propagation rules.

4.2.1 Estimating emissions from industrial processes Both components of non-energy use (i.e., feedstock use and the consumption of non- energy use refinery products, coke oven products, and other solid carbon) lead to industrial process emissions. NEAT calculates industrial process emissions for the following processes: (i) steam cracking, (ii) the production of ammonia, methanol, and carbon black8, (iii) chemical conversions9 (including 36 chemical conversion processes; see Appendix of this chapter), and (iv) electrodes and other solid carbon use in the manufacturing of 19 non-ferrous metals, ferroalloys, and other inorganic chemicals10. We calculate total industrial process emissions as the sum of emissions from individual processes based on process-specific emission factors and production data as:

7 This refers to the calculation of emissions that result from chemical conversion processes (i.e., chemical conversion losses), product use, the application of urea fertilizers in agriculture, and wastewater treatment. 8 We account for the various types of feedstock used (i) in steam cracking (i.e., naphtha, gas oil, ethane, butane, propane) and (ii) in the production of ammonia and methanol (i.e., natural gas, lignite, heavy fuel oils). 9 We define chemical conversion processes as conversions of basic and intermediate chemicals within the chemical industry (e.g., production of styrene from ethylbenzene). The use of energy carriers as feedstock and their subsequent conversion to basic chemicals (e.g., ethylene, propylene, methanol, or carbon black) is, therefore, excluded from the category of chemical conversions. 10 We include emissions from the production of: primary aluminium, electric arc furnace steel, white phosphorus, titanium dioxide, ferrosilicon, calcium carbide, silicon carbide, silicon, ferromanganese, silicon manganese, ferrochromium, ferrochromium-silicon, chromium, primary and secondary lead, magnesium, nickel, tin, and zinc.

67 Chapter 4

= × EIP ,k B(Pi,k EFi,k ) (4.1) i where EIP,k [Mt CO2 equivalents] represents the total yearly emissions from industrial processes, Pi,k [Mt] the physical production of chemicals, non-ferrous metals, ferroalloys, and inorganic chemicals in process i, EFi,k the process-specific emission factor [t CO2 equivalents/t product], and k the index for the year of study.

We derive (i) production data from Destatis (1990-2003a), Consultic (1990-2003), and GDA (2007)11 and (ii) process- and feedstock-specific emission factors from a variety of different sources as described in detail by Neelis et al. (2005b). We estimate feedstock composition for steam cracking, ammonia, and methanol production based on VCI (2004a) and interviews with several industry experts.

In the earlier versions of NEAT, Neelis et al. (2005a, b) assume all chemical conversion processes to be 100% carbon efficient, i.e., they model the total amount of carbon initially contained in basic and intermediate chemicals as being incorporated in final products (e.g., polymers). This simplification, however, neglects carbon losses that occur in the various chemical conversion processes due to partial feedstock oxidation, leakages, and the generation of non-specified by-products. In this research, we extend NEAT by a module that allows us to estimate CO2 emissions resulting from conversion processes of 36 basic and intermediate chemicals (e.g., production of ethylene dichloride, acrylonitrile, and polyvinyl chloride)12. We identify chemical conversion processes for Germany based on Neelis et al. (2007), Patel et al. (1999), Ullmann (1997), Weissermel and Arpe (2003) and miscellaneous expert interviews. We multiply production data of the various chemicals with process-specific carbon losses as determined by Neelis et al. (2007) (see Appendix of this chapter)13.

The accounting of industrial process emissions is complicated by the fact that for many processes (the most prominent being ammonia production and steam cracking), parts of the hydrocarbon input are strictly speaking not used as feedstock but as fuel to sustain chemical reactions. This situation makes the exact allocation of fossil fuels (e.g., naphtha, natural gas) to either non-energy use or to energy use very difficult. Detailed investigations revealed that German energy statistics (DIW, 2005; MWV, 2005) do not

11 We take data for polymer and aluminium production from Consultic (1990–2003) and GDA (2007), respectively because we identified inconsistencies in the data sets as stated by the official German production statistics (i.e., Destatis, 1990–2003a). 12 Our estimates of emissions from chemical conversion processes in principle exclude emissions from the combustion of fuel-grade by-products. There is, however, a small risk of double counting emissions, if NEAT results are used directly in the IPCC-SA because emissions can potentially be reported twice, once under industrial processes and again under the category of energy. Detailed insight into both the German energy statistics and the GHG inventory gives strong indication that this is, however, not the case (Weiss et al., 2007). 13 The calculation of carbon losses for 36 chemical conversion processes with NEAT exceeds the degree of detail as specified by both IPCC (1997) and IPCC (2006). The 1997 guidelines (that are followed by the German GHG inventory) do not specify any of the chemical conversion processes that are included in NEAT as potential sources for CO2 emissions. The 2006 IPPC guidelines point out only the production of ethylene dichloride, ethylene oxide, and acrylonitrile as potential emission sources.

68 Non-energy use and related emissions in Germany possess uniform system boundaries for the non-energy use of the various types of fossil fuels: (i) For coal- and oil-based hydrocarbons (e.g., coke, hard coal, lignite, coal oils and tars, fuel oils, naphtha) a gross definition of non-energy use is applied. Here, the total fossil hydrocarbon input into industrial processes (including hydrocarbons used as fuel) is regarded as non-energy use. (ii) For natural gas, a net definition of non-energy use is applied, i.e., the parts of natural gas that are consumed for fuel purposes are excluded from non- energy use and reported as fuel.

With NEAT, we calculate non-energy use and related industrial process emissions according to the system boundaries as applied in German energy statistics to ensure comparability with data from the official German GHG inventory (which are largely based on German energy statistics)14.

4.2.2 Estimating emissions from solvent and other product use In contrast to previous NEAT studies, we apply a simple bottom-up approach to estimate emissions from solvent and other product use separately for three relevant key sources, i.e., the consumption of (i) solvents, (ii) lubricants, and (iii) waxes and paraffins. We refer to this method as the key sources approach. We calculate total solvent and other product use emissions as:

= × × ESPU ,k B(Ci,k FCi SCi ) (4.2) i where ESPU,k [Mt CO2 equivalents] represents the total yearly emissions from solvent and other product use, Ci,k [Mt] the consumption of product i in year k, FCi the product- specific fossil carbon content [kg CO2 equivalents/kg product], and SCi [%] the product- specific shares of fossil carbon that become oxidized during product use15. The principal data sources for our calculations are (Destatis, 1990-2003a) and (Destatis, 1990–2003b), UBA (2007), FAO (2006), Theloke et al. (2000), and Jepsen et al. (2004). For calculating emissions, we multiply consumption data with specific emission factors as given in Table 4.1. To account for uncertainty in activity data, we assume an uncertainty interval of 15% for the consumption of solvents and 10% for the consumption of lubricants, waxes, and paraffins.

14 Ensuring consistency of system boundaries is critical for calculating industrial process emissions because depending on the definition chosen for non-energy use in energy statistics, emissions are reported according to the IPCC-SA either under the source categories energy or industrial processes. 15 For calculating emissions from solvent use, we make use of detailed bottom-up emission studies as conducted by Theloke et al. (2000) and Jepsen et al. (2004). Both studies are used as principal data sources for solvent and other product use emission estimates in the German GHG inventory (UNFCCC, 2007a, b).

69 Chapter 4

Table 4.1: Emission factors as applied in the key sources approach for calculating emissions from solvent and other product use Fossil carbon content Emission factor Carbon oxidation Producta in kg CO equivalents/ in kg CO equivalents/ 2 rate in % 2 kg product kg product Solvents(1) 2.31 ± 0.23 100 2.31 ± 0.23b Lubricants(2) 3.15 30 ± 20 0.95 ± 0.63 Waxes and paraffins(3,4,5) 3.15 43 ± 14 1.36 ± 0.44 a The indices given below refer to the following sources: (1) Schmidt-Stejskal et al. (2004); (2) Trischler (1997); (3) Patel et al. (1999); (4) Ullmann (1997); (5) Weissermel and Arpe (2003) b Conversion factor for re-calculating solvent use emissions from NMVOC equivalents into CO2 equivalents

It is rather straightforward to calculate the fossil carbon content of lubricants, waxes, and paraffins because their chemical composition is relatively homogenous. Solvents, however, comprise a relatively large group of substances, making our estimates for the fossil carbon content of these substances more uncertain (see first column in Table 4.1). Depending on the application, solvent use leads to emissions of the entire carbon initially contained in products within rather short time periods. This is not the case for the other key sources where substances serve as intermediate chemicals (i.e., waxes and paraffins) and are recycled or eventually combusted with energy recovery (i.e., lubricants; see second column in Table 4.1). We neglect other emission sources such as the use of chemical auxiliaries in the textile, paper, and leather industry because of their minor share in total solvent and other product use emissions.

4.2.3 Estimating emissions from agriculture Non-energy use emissions in agriculture (EAC,k) result from the application of urea fertilizers and pesticides. We estimate yearly CO2 emissions from the application of fertilizers based on consumption data for urea containing fertilizers as provided by Yara (2008). We assume (i) that 50% of the nitrogen contained in ammonia-nitrate-urea solutions is derived from urea and (ii) that sulphur fertilizers account for 5% of the total urea consumption in agriculture. We assume a carbon oxidation rate of 100% and a carbon content of 0.73 kg CO2/kg urea. We estimate emissions from pesticide use by assuming an emission factor of 1.03 ± 0.23 kg CO2/kg pesticides based on a fossil carbon content of 1.28 ± 0.37 kg CO2/kg pesticides and a carbon oxidation rate of 80 ± 20% (Theloke et al., 2000; UBA, 2007; Sonnenberg and Sietz, 2007).

4.2.4 Estimating emissions from waste Non-energy use emission in the source category waste result from the oxidation of surfactants and other fossil carbon containing substances during wastewater treatment. We estimate fossil CO2 emissions from wastewater treatment (EW,k) based on domestic surfactants consumption of roughly 670 kt per year in Germany (i.e., approximately 8 kg of surfactants per capita and year). We assume the fossil carbon content in surfactants to be 1.32 kg CO2/kg surfactant and a carbon oxidation rate of 100% (Patel, 1999; Patel et al., 1999).

70 Non-energy use and related emissions in Germany

4.2.5 Estimating non-energy use and carbon storage As stated in the introduction, the non-energy use of fossil fuels consists of two components, (i) the consumption of feedstock in the chemical industry and (ii) the consumption of refinery products, coke oven products, and other solid carbon for non- energy purposes. We account for the carbon that is contained (i) in the physical production of 15 basic chemicals (including urea)16, (ii) in emissions that result from the production processes of basic chemicals as well as from electrode and other solid carbon use in the production of non-ferrous metals, ferroalloys, and inorganic chemicals, and (iii) in the consumption of refinery products for non-energy purposes17. Non-energy use is then calculated as:

FVFV16 3 =++++++ NEUkikSkAkMkCkNFkikGWGW()(BB PC,,,,,, E E E E E )() CR , (4.3) HXHXi=1 i=1 where NEUk [Mt CO2 equivalents] represents non-energy use in year k, PCi,k [Mt CO2 equivalents] the production of chemical i, ES,k, EA,k, EM,k, EC,k, ENF,k [Mt CO2 equivalents] emissions from steam cracking, the production of ammonia, methanol, carbon black, and the manufacturing of non-ferrous metals, ferroalloys, and inorganic chemicals, respectively, and CRi,k the domestic consumption of refinery and coke oven products for non-energy use applications.

Carbon storage is calculated in NEAT as the carbon that is initially contained in the non-energy use of fossil fuels minus all further downstream emissions:

= − + + + CS k NEU k (EIP,k ESPU ,k E AC ,k EW ,k ) (4.4) where CSk [Mt CO2 equivalents] represents the total carbon storage in year k. Note that EIP,k equals the sum of ES,k, EA,k, EM,k, EC,k, ENF,k. For further details on the NEAT methodology, we refer the reader to Neelis et al. (2005b).

4.3 Results In this section, we present the results of our NEAT calculations and we compare our findings with emission estimates from the German GHG inventory as submitted to the UNFCCC in the years 2005, 2006, and 2007. It is surprising that NEAT estimates on total non-energy use emissions substantially exceed the values as calculated according to both IPCC-RA and IPCC-SA of the German GHG inventory (Figure 4.2). Emission estimates according to the IPCC-RA are essentially identical in the GHG inventory submissions of 2006 and 2007 and denoted as “IPCC-RA (2006, 2007)” in the legend of Figure 4.2. The same is true for emission estimates as stated according to the IPCC-SA (denoted as

16 We include the following basic chemicals: acetylene, benzene, butane, butadiene, carbon black, carbon monoxide, ethylene, methanol, naphthalene, propylene, toluene, ortho-, meta-, para-xylene, and urea. 17 We include the following refinery and coke oven products: creosote oil, bitumen, lubricants, waxes, and paraffins.

71 Chapter 4

“IPCC-SA (2006, 2007)” in Figure 4.2). The GHG inventory submission of 2005, however, differs from the submissions for 2006 and 2007 because in later years, important adaptations were made that also include the use of parts of our NEAT results.

IPCC-RA (2005) 35 IPCC-RA (2006, 2007) IPCC-SA (2005) 30 IPCC-SA (2006, 2007) NEAT

25 2

in Mt CO 15

Yearly non-energy use emissions emissions use non-energy Yearly 5

0 19901991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Figure 4.2: Yearly non-energy use emissions as estimated by NEAT and as calculated according to IPCC-RA and IPCC-SA of the German GHG inventory (submissions to the UNFCCC in the years 2005, 2006, and 2007; UNFCCC, 2005a, 2006b, 2007a)

We first present and discuss our NEAT results in comparison to inventory estimates according to the IPCC-SA at the level of individual source categories. Later we focus on total non-energy use emissions, total non-energy use, and carbon storage and we compare our results to estimates from the IPCC-RA.

4.3.1 Emissions from industrial processes Total yearly industrial process emissions as calculated with NEAT range from 16.0 ± 1.4 Mt CO2 in 1990 to 22.6 ± 2.1 Mt CO2 in 2003 and show an increasing trend (on average 2.7% per year) in the period from 1990 to 2003 (Figure 4.3). NEAT results exceed the values calculated according to IPCC-SA on average by 640% when compared to data from the 2005 inventory submission and by 77% when compared to the submissions of the years 2006 and 2007 (Figure 4.3)18.

18 Electrodes and other solid carbon are used for the production of non-ferrous metals, ferroalloys, and inorganic chemicals. The IPCC-SA reports emissions from these processes under the following source categories: (i) aluminium production, (ii) carbide production, and (iii) ferroalloy production.

72 Non-energy use and related emissions in Germany

11 2

10 IPCC-SA (2005) IPCC (2006,2007) 9 NEAT

1 Yearly industrial process emissions in MtCO 0 1990 1995 2000 2003 1990 2000 2003 1990 1995 2000 2003 1990 1995 2000 2003 1990 1995 2000 2003 1990 1995 2000 2003 1995 Steam Ammonia Methanol Carbon black Chemical conversion Electrodes and other cracking production production production processes solid carbon use Figure 4.3: Yearly industrial process emissions as estimated with NEAT and as calculated according to the IPCC-SA (UNFCCC, 2005a, 2006b, 2007a)

We explain the differences between NEAT and the IPCC-SA as follows19: (i) CO2 emissions from steam cracking are not reported according to the IPCC- SA, whereas we calculate yearly emissions of 5.8 ± 1.2 Mt CO2 to 8.6 ± 1.7 Mt CO2 for the period of 1990–2003. This fact explains 43% and even 87% of the difference between NEAT and the IPCC-SA of the inventory submission 2005 and the inventory submissions 2006 and 2007, respectively. (ii) The IPCC-SA of the 2005 inventory submission excludes emissions from the production of methanol and carbon black (together representing 14% of the difference) as well as losses from chemical conversion processes (accounting for 19% of the difference between NEAT and IPCC-SA). In the inventory submissions of 2006 and 2007, NEAT results are used to fill these data gaps in the IPCC-SA. (iii) NEAT and IPCC-SA differ with respect to emission estimates for ammonia production due to incompatible emission factors (discussed below). The observed deviations account on average for 12% and -10% of the total differences between NEAT and IPCC-SA of the inventory submissions in 2005 and 2006, 2007, respectively. (iv) The IPCC-SA accounts only incompletely for emissions from electrodes and other solid carbon use in the production of non-ferrous metals, ferroalloys, and inorganic chemicals (deviations explain 12% and 23% of the total differences between NEAT results and the IPCC-SA of the inventory submissions in 2005 and 2006, 2007, respectively).

19 The differences specified for the following source categories explain 100% of the total deviations between NEAT and IPCC-SA with respect to industrial process emissions. In the case of the inventory submissions of 2006 and 2007, we determine negative deviations of roughly 10% for ammonia production because estimates according to IPCC-SA exceed NEAT results.

73 Chapter 4

As indicated above, NEAT results were used to improve the completeness of the IPCC-SA (in the inventory submissions of 2006 and 2007) with respect to emissions resulting from the production of methanol, carbon black, as well as losses of chemical conversion processes (Figure 4.4). Gaps, however, still remain in the IPCC-SA (UNFCCC, 2006b, 2007a). This refers in first instance to emissions from steam cracking. According to our investigations, these are part of non-energy use emissions as stated according to the IPCC-RA of the German GHG inventory and should therefore be reported as industrial process emissions in the IPCC-SA. Omitting this emission source in the IPCC-SA probably results in substantial underreporting of emissions (Figure 4.4)20.

20 Electrodes and other solid carbon use

Ammonia 2 Steam cracking production 15 8.4 Mt CO Mt 8.4 2 Electrodes and other solid carbon use 10 Chemical conversion processes 2 in Mt CO in Mt Carbon black production Methanol production 8.3 Mt CO Mt 8.3 5 Ammonia production Average yearly industrial process emissions industrial yearly Average 0 IPCC-SA (2005) IPCC-SA (2006,2007) NEAT Figure 4.4: Average yearly industrial process emissions as estimated with NEAT and as calculated according to the IPCC-SA (UNFCCC, 2005a, 2006b, 2007a) for the period between 1990 and 2003

NEAT emission estimates for ammonia production differ from IPCC-SA data as stated in the German GHG inventories submitted to the UNFCCC in the period of 2005– 2007. According to the IPCC-SA (inventory submission 2005) emissions are 33–58% lower than our NEAT results. This deviation is entirely caused by the IPCC-SA emission factor (i.e., 0.84 kg CO2/kg nitrogen contained in ammonia) that substantially underestimates actual ammonia production emissions.

20 Given the insight obtained in the course of our research, we consider it unlikely that parts of these emissions are reported according to the IPCC-SA under the source category of energy. Indication is given, e.g., by the fact that the entry for ethylene production (in steam crackers) in the IPCC-SA is labeled as NO (not occurring) instead of IE (included elsewhere). However, further research is recommended to entirely clarify this point.

74 Non-energy use and related emissions in Germany

This shortcoming was also addressed by an external review team (UNFCCC, 2005b)21 and has been corrected in the inventory submissions of the years 2006 and 2007. In these submissions, the IPCC default emission factor of 1.5 kg CO2/kg ammonia is applied (IPCC, 1997). This results in emission estimates that are higher than our NEAT results. We argue that the use of the IPCC-SA default emission factor does not correctly account for emissions from ammonia production in Germany because it neglects several country-specific features: (i) The IPCC default emission factor assumes natural gas to be used as only feedstock for ammonia production and neglects that roughly 30% of ammonia is produced from oil-based feedstock in Germany. (ii) The IPCC default emission factor does not account for the fractions of CO2 sequestered for the production of urea (i.e., 13–26% of process emissions from ammonia production in the various years). This leads to double counting of emissions, once under the source category of industrial processes and again under the source category of agriculture. (iii) The IPCC default emission factor does not account for the system boundaries of the non-energy use of natural gas and oil-based feedstock as applied in the German energy statistics. Consequently, emissions from the fuel use of oil- based feedstock are erroneously excluded from the emission estimates for ammonia production in the IPCC-SA.

Combining these three points, we argue that the IPCC default emission factor of 1.5 kg CO2/kg ammonia overestimates actual emissions from ammonia production in Germany. The emission factors as implemented in NEAT (1.2–1.4 kg CO2/kg ammonia) assume efficient ammonia plants and account for both (i) the sequestration of process CO2 for urea production22 and (ii) the system boundaries for the non-energy use of natural gas and heavy oil in the German energy statistics (see Section 4.2.1).

According to the IPCC-SA, CO2 emissions from the use of electrodes and other solid carbon comprise three principle source categories, i.e., production of aluminium, carbides, and ferroalloys. NEAT, furthermore, covers the manufacturing of other non- ferrous metals and inorganic chemicals (e.g., silicon, lead, zinc). Both NEAT and the IPCC-SA use the same production data and emission factors to estimate emissions from aluminium production. For the production of other non-ferrous metals, ferroalloys, and inorganic chemicals, NEAT arrives at clearly higher estimates than the IPCC-SA (the difference is on average 1 Mt CO2). This deviation is partly explained by the NEAT results, which exceed IPCC-SA emission estimates for the production of carbides by, on average, 0.5 Mt CO2. Emissions from ferroalloy production are only calculated in the IPCC-SA of the inventory submissions for the years 2006 and 2007. Here, emission

21 The external review team states: “Emissions from ammonia production are estimated using an EF [emission factor] that is lower than the IPCC default and the lowest of all reporting parties, and is not well documented. The ERT [external review team] noted that Germany has planned to begin using the IPCC default value, which is recommended in the future” (UNFCCC, 2005b). 22 The fast majority of urea is used for fertilizer production, whereas a small share becomes incorporated into urea resins. The carbon that is initially contained in urea is thereby either stored in urea resins or it becomes emitted after the application of fertilizers in agriculture. The resulting emissions are accounted for under the source category agriculture in the IPCC-SA (IPCC, 1997).

75 Chapter 4 estimates are substantially lower than our NEAT results. For both emission sources (i.e., production of carbides and ferroalloys) activity data and emission factors are not stated in the German GHG inventory. However, based on (i) the emission quantities calculated according to the IPCC-SA and (ii) the information provided by the German inventory report (UNFCCC, 2007b), we argue that total IPCC-SA emission estimates for electrodes and other solid carbon use in the production of non-ferrous metals, ferroalloys, and inorganic chemicals are incomplete because relevant emission sources (e.g., production of lead, magnesium, or zinc) are neglected.

4.3.2 Emissions from solvent and other product use With our key sources approach, we identify a decrease in yearly solvent and other product use emissions from 4.7 ± 1.1 Mt CO2 in 1990 to 3.5 ± 0.8 Mt CO2 in 2003 (Figure 4.5). The German GHG inventory reports emissions from solvent and other product use in kt NMVOC equivalents23. To make IPCC-SA results comparable with our estimates, we 24 apply a conversion factor of 2.31 kg CO2/kg NMVOC (Schmidt-Stejskal et al., 2004) . The IPCC-SA estimates regarding total solvent and other product use emissions are only about half of our NEAT values. The differences are explained by the fact that emissions as calculated according to the IPCC-SA include only solvents use25 but exclude emissions from the consumption of other relevant products. Disregarding lubricant use and the consumption of waxes and paraffins leads to considerable underestimation of yearly solvent and other product use emissions in the IPCC-SA by roughly 2 Mt CO2 equivalents.

23 Note that solvent and other product use emissions are covered in the German GHG inventory following IPCC (1997) methodology. Emissions are typically NMVOCs that quickly become oxidized to CO2 once released to the atmosphere. Solvent and other product use emissions are nevertheless not converted to CO2 equivalents in the German GHG inventory. They remain excluded from the estimate of total national GHG emissions that is relevant as reference for the Kyoto target (UNFCCC, 2007b). 24 We show here only results based on the IPCC-SA of the German GHG inventory submitted in the year 2007. Data of earlier inventory submissions (i.e., for the years 2005 and 2006) differ only marginally from the ones presented in Figure 4.5. 25 Both our estimates on solvent use emissions and the results of the IPCC-SA on emissions from total solvent and other product use are based on detailed bottom-up analyses of solvent consumption in Germany as conducted by Theloke et al. (2000) and Jepsen et al. (2004). The minor differences depicted in Figure 4.5 for solvent use emissions in the year 2000 might be attributed to data adaptations in the German GHG inventory that are not communicated in the bottom-up analyses.

76 Non-energy use and related emissions in Germany

7 Total (IPCC-SA, 2007) Total (NEAT) Solvent use (NEAT) 6 Lubricant use (NEAT) Use of waxes and paraffins (NEAT)

4 equivalents 2 3

in Mt CO in 2

Yearly solvent and other product use emissions use product other and solvent Yearly 0 19901991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Figure 4.5: Yearly solvent and other product use emissions as estimated based on our key sources approach and as calculated according to the IPCC-SA (UNFCCC, 2007a)

4.3.3 Emissions from agriculture Yearly fossil CO2 emissions from the application of urea fertilizers and pesticides in agriculture increase from 0.47 ± 0.07 Mt CO2 in 1990 to 0.74 ± 0.11 Mt CO2 in 2003. Emissions from the application of urea fertilizers account for 95% of this total, rendering emissions from pesticide use negligible. The IPCC-SA of the German GHG inventory accounts for the first time in the submission of 2007 for emissions from this source category. The estimates of the 2007 inventory submission are within the uncertainty ranges of our results.

4.3.4 Emissions from waste We quantify yearly fossil CO2 emissions from the oxidation of surfactants during wastewater treatment to be 0.89 ± 0.27 Mt CO2. The IPCC-SA of the German GHG inventory does not calculate fossil-based CO2 emissions from wastewater treatment. We regard our results as a very rough first estimate that might serve as a benchmark for a more detailed calculation of emissions from this source category.

4.3.5 Total non-energy use emissions, carbon storage, and non-energy use of fossil fuels In this section, we compare total yearly non-energy use emissions as calculated by NEAT with data from the IPCC-RA. Our NEAT estimates range from 22 ± 2 Mt CO2 (1990) to 28 ± 2 Mt CO2 (2003). This is 4–8 Mt CO2/a higher than the non-energy use emissions as calculated with the IPCC-RA of the German GHG inventory (submission 2007; Figure 4.6). The observed differences are caused per definition by deviations in either total non- energy use (NEUk) or carbon storage (CSk) as calculated with NEAT and IPCC-RA.

77 Chapter 4

60 /a 2 50

Mt CO 40

0 19901991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year

Non-energy use (IPCC-RA, 2007) Non-energy use (NEAT) Carbon storage (IPCC-RA, 2007) Carbon storage (NEAT) Non-energy use emissions (IPCC-RA, 2007) Non-energy use emissions (NEAT)

Figure 4.6: Total yearly non-energy use, carbon storage, and non-energy use emissions as estimated with NEAT and as calculated according to the IPCC-RA; inventory submission 2007 (UNFCCC, 2007a)

Addressing the first parameter, total yearly non-energy use as calculated with NEAT ranges between 67 ± 5 Mt CO2 (1991) and 86 ± 5 Mt CO2 (2000) and shows an overall increasing trend in the period of 1990–2003. With the exception of the years 199026 and 1991, non-energy use as calculated with NEAT exceeds the values reported according to the IPCC-RA (inventory submission 2007), which originate from German energy statistics (AGE, 2007) by 2–18%27. The differences are mainly caused by the following factors: (i) The estimates for non-energy use of coal products according to the IPCC-RA are incomplete (by roughly 1 Mt CO2), as they do not cover all feedstock

26 Due to the reunification of Germany in 1990, official production data (Destatis, 1990–2003a) are particularly uncertain for this year as they might account for production in the former western part of Germany only. It was beyond the scope of this research to entirely clarify this uncertainty. The problem we encounter here is, however, also acknowledged as a major source of uncertainty within the German GHG inventory (UNFCCC, 2005c). 27 The deviation of 18% in the year 1995 is exceptionally high. It is caused by the comparatively low non- energy use as reported by the German energy statistics. Unlike for all other years, the energy statistics do not report non-energy use of residual fuel oils, leading to roughly 10% lower non-energy use than reported for the years 1994 and 1996. This might be an error in the energy statistics that should be corrected in the future.

78 Non-energy use and related emissions in Germany

requirements for the domestic production of coal-derived tars and crude benzene. (ii) In addition to electrodes, NEAT estimates also include the amounts of other solid carbon such as coal and cokes that are used as reducing agents for metallurgical processes (i.e., the manufacturing of non-ferrous metals, ferroalloys, and inorganic chemicals). Within the IPCC-RA, this carbon (1.1-1.8 Mt CO2) might be accounted for under the source category energy and could have therefore been excluded from non-energy use. (iii) Chemical grade refinery propylene (roughly 1.8 Mt CO2 equivalents) is excluded and refinery butadiene and aromatics are likely to be excluded from non-energy use according to the IPCC-RA. All three chemicals are, however, included in NEAT because they are consumed for non-energy purposes by the chemical industry. (iv) Butene produced in steam crackers (0.7–1.2 Mt CO2) can be used for both the production of polymers and as a gasoline additive. It remains, however, unclear if and to what extent the fractions of butene consumed for fuel additives enter both the German energy statistics (and subsequently the IPCC-RA) and official production statistics (see also Section 4.4.1). (v) The non-energy use of natural gas for carbon black production is included in NEAT (0.15–0.18 Mt CO2) but excluded from the IPCC-RA (VCI, 2004b).

Combining these factors and leaving the uncertain position of butene aside, we estimate that average yearly non-energy use is potentially underestimated within the IPCC-RA by roughly 4 Mt CO2. If we apply these corrections, the differences between NEAT and the values calculated according to the IPCC-RA (inventory submission 2007) are substantially reduced28.

We now address the second parameter that is relevant for explaining the differences between our NEAT results and the IPCC-RA with respect to non-energy use emissions, i.e., carbon storage. The carbon storage as estimated with NEAT is generally in line with official values from the IPCC-RA (inventory submission 2007). However, the good agreement is an artefact that results from deviations in both non-energy use and carbon storage fractions. Carbon storage is calculated in the IPCC-RA by multiplying the non- energy use of fossil fuels with fuel-specific carbon storage fractions. The average carbon storage fractions according to the IPCC-RA were determined by Prognos (2000) and vary between 70 and 79%. They are higher than our NEAT carbon storage fractions (66 ± 10% to 69 ± 8%). The differences are caused by deviations in the system boundaries of the applied carbon storage fractions. The carbon storage fractions as applied in the IPCC-RA are neither entirely consistent with the system boundaries of non-energy use data nor with the aim of the IPCC-RA as outlined by IPCC (1997). The IPCC-RA carbon storage fractions only account for (i) emissions resulting from the combustion of feedstock and (ii) direct CO2 emissions from solvent and other product use (Prognos, 2000). They hence

28 An exception is the year 1990 for which the uncorrected IPCC-RA value already exceeds the NEAT estimate. After correction, the values according to IPCC-RA are outside the NEAT uncertainty ranges for the years 1990, 1995, and 2003. If the data correction is applied, deviations between IPCC-RA and NEAT for the two latter years reduce considerably.

79 Chapter 4 treat parts of the industrial process emissions (e.g., emissions resulting from ammonia production) as well as NMVOC emissions from solvent and other product use as storage29. We argue that this methodological flaw alone leads to an underestimation of actual non- energy use emissions in the IPCC-RA (inventory submission 2007) by on average 2 Mt CO2. In line with the system boundaries of non-energy use data in German energy statistics (AGE, 2007), NEAT carbon storage fractions account for the entire amount of non-energy use emissions resulting from the various source categories as presented previously (Sections 4.3.1-4.3.4).

Although both NEAT and the IPCC-RA (based on carbon storage fractions as calculated by Prognos (2000)) result in similar estimates for carbon storage, the IPCC-RA underestimates non-energy use emissions. This finding indicates that the identified gap between the IPCC-RA and IPCC-SA methods (i.e., 9-18 Mt CO2 per year; see Figure 4.1) is even larger due to (i) incomplete non-energy use data and (ii) the application of carbon storage fractions that only insufficiently account for non-energy use emissions in the IPCC-RA. In the following section, we show at the example of average yearly non-energy use emissions in the period of 1990–2003 how NEAT results can help removing this inconsistency from the German GHG inventory.

4.3.6 Comparison between IPCC-RA and IPCC-SA emission estimates One core objective of this research is to improve data consistency between IPCC-RA (i.e., average yearly non-energy use emissions of 19.6 Mt CO2 in the period of 1990-2003) and IPCC-SA (i.e., average yearly non-energy use emissions of 4.9 Mt CO2) (inventory submission 2005) (Figure 4.7). We have already discussed in Section 4.3.1 that NEAT emission estimates for methanol and carbon black production as well as for losses from chemical conversion processes have been used within the IPCC-SA of the German GHG inventory submissions in the years 2006 and 2007 to correct for data gaps in the IPCC-SA of the 2005 inventory submission. Furthermore, the emission factor for ammonia production has been adapted and emissions from urea application in agriculture were estimated for the first time in the inventory submission of 2007. Combining these adaptations reduces the differences between IPCC-RA (submission 2005) and IPCC-SA results (submission 2007) by almost 60% (Figure 4.7).

If we include now emission sources that are currently not covered by the IPCC-SA method (i.e., steam cracking, additional non-ferrous metals, ferroalloy, and inorganic chemicals production, product use, wastewater treatment) and if we correct the IPCC (1997) default emission factor currently applied for ammonia production, the resulting non-energy use emissions exceed the estimates according to the IPCC-RA by 5.3 Mt CO2. The gap between IPCC-RA, IPCC-SA, and NEAT can, however, be closed up to a difference of 0.4 Mt CO2, if two adaptations are made. First, we add 4 Mt CO2 to the results of the IPCC-RA to account for the underestimation of non-energy use in the German energy statistics (see Section 4.3.5).

29 For a more detailed discussion on assumptions, system boundaries, and methodology applied to calculate NEAT and IPCC-RA storage fractions, we refer to Weiss et al. (2007).

80 Non-energy use and related emissions in Germany

-0.4 Mt CO2 +5.3 Mt CO2 25

-14.7 Mt CO2 -6.0 Mt CO2 20 2

15 in Mt CO

5 Average yearly non-energy use emissions emissions use non-energy yearly Average 0 IPCC-RA IPCC-SA IPCC-SA NEAT IPCC-RA (2005) (2005) (2007) (2005, adapted)

Wastewater treatment Use of urea fertilizers and presticides in agri Solvent and other product use Chemical conversion processes Carbon black production Methanol production Steam cracking Electrodes and other solid carbon use (excl. aluminium production) Aluminium production Carbide production Ammonia production

Figure 4.7: Overview: closing the gap between IPCC-RA and IPCC-SA (UNFCCC, 2005a, 2006b, 2007a) by applying NEAT emission estimates

For the second adaptation, we recall our discussion regarding the system boundaries of the Prognos (2000) carbon storage fractions as applied in the IPCC-RA. These were designed to account for emissions from the combustion of feedstock and therefore neglect a large part of emissions from, e.g., industrial processes and solvent and other product use. To correct for this, we can calculate adapted IPCC-RA non-energy use emissions by subtracting NEAT carbon storage from the adapted non-energy use as stated according to the IPCC-RA30. The average difference between NEAT and the adapted IPCC-RA emission estimates is within the uncertainty range of our results (see Figure 4.7). The difference can however be higher than shown in Figure 4.7 for individual years (e.g., 1990, 1991, or 2003) for which deviations between NEAT and official non-

30 Note that carbon storage is independent from the system boundaries of non-energy use and therefore does not depend on choices regarding a net versus gross definition of feedstock use.

81 Chapter 4 energy use data in the German energy statistics are not fully explained by the factors described in Section 4.3.5.

4.4 Discussion We first provide a critical discussion of the NEAT approach that has been used for calculating non-energy use and related emissions. In the second part, we give recommendations for both German and international inventory experts and we highlight general aspects, being critical for the correct reporting of non-energy use emissions in the national GHG inventories of Annex I countries.

4.4.1 Discussion of NEAT methodology Total non-energy use emissions as calculated with NEAT are subject to uncertainties in the range of 10% of the final result. Given both the scope of this research and the general difficulties related to the complete and reliable accounting of non-energy use emission, we regard our model uncertainties as acceptable. We nevertheless see potentials for reducing uncertainties by future research. This refers in the first instance to the calculation of industrial process emissions that account for 70% of all uncertainties related to total non- energy use emissions as calculated with NEAT.

We estimate industrial process emissions in NEAT by assuming efficient to very efficient plants for steam cracking as well as for the production of ammonia, methanol, and carbon black. This results in CO2 emission factors, which are at the lower end of their possible value range31. The uncertainty intervals of NEAT results (i.e., ranging from 10 to 25% for individual processes) may be reduced by applying country-specific emission factors based on detailed company surveys. The uncertainty intervals of industrial process emissions can be further reduced by applying more detailed estimation approaches according to Tier 3 methodology as outlined by IPCC (1997).

Model uncertainties are also related to emissions from chemical conversion processes that are based in NEAT on process-specific emission factors from open literature (Neelis et al., 2007). Our results represent average estimates for the most important conversion processes in the chemical industry but they do not, however, account for specific settings on the level of individual plants. The estimates of CO2 emissions from chemical conversion processes refer to reaction losses and exclude both the generation of fuel-grade by-products as well as energy use of feedstock in conversion processes. However, uncertainties result because only some parts of the losses are directly emitted as CO2, whereas other parts might be either flared with or without additional fuel input and with or without energy recovery (Neelis et al., 2007). Within the scope of this research project, it was not possible to elaborate in greater detail on the exact fate of carbon losses from the various chemical conversion processes. Further research is recommended to reduce uncertainties of our NEAT estimates (which we quantify with 12–16%) for this emission source category. A source of uncertainty, which is excluded from the uncertainty

31 For a detailed comparison of NEAT emission factors with data from literature, we refer the reader to Neelis et al. (2003).

82 Non-energy use and related emissions in Germany intervals, refers to the assumed chemical conversion routes. Identifying chemical conversion routes is by no means straightforward and requires detailed insight into the structure of the chemical industry in Germany. Various alternative production routes for individual chemicals exist (e.g., phenol is produced from cumene in Germany, whereas it is produced entirely from toluene in the Netherlands). Companies tend to give only vague information due to confidentiality reasons. Further research is recommended to improve the reliability of information on chemical conversion routes that are operated in Germany.

With regard to emissions from solvent and other product use, we highly recommend more detailed bottom-up analyses on the fate and oxidation of lubricants, waxes and paraffins. Such research could substantially reduce the uncertainty intervals (i.e., around 40%) that are currently related to our results.

Our estimates for yearly non-energy use (and therefore also the ones for yearly carbon storage) are associated with additional uncertainties that are linked to the reliability of production and trade data used as model input. We argue that trade data for basic chemicals and refinery products as published by Destatis (1990–2003b) can generally be considered reliable (because they are used for taxation purposes). This is, however, not necessarily the case for production data. We identified major inconsistencies in the production data as stated by the Federal Statistical Office of Germany (Destatis, 1990- 2003a) (e.g., for bitumen, lubricants, basic chemicals like butadiene and toluene).

Further uncertainties of non-energy use and carbon storage as calculated with NEAT relate to (i) the unclear position of butene in German production and energy statistics and (ii) the system boundaries for coal- and lignite-based non-energy use in German energy statistics. Intensive discussions with experts preparing the energy statistics for Germany did not allow clarification of the exact definition of system boundaries for coal-, lignite-, and coke-based non-energy use beyond any doubts. The non-energy use of these energy carriers, which accounts on average for 7% of total non-energy use, is therefore particularly uncertain.

4.4.2 Recommendations for inventory experts and energy statisticians Our NEAT model results provide insight into non-energy use and related CO2 emissions and allow us to derive recommendations for energy statisticians as well as GHG inventory experts. One outcome of this research is the identification of system boundaries applied to non-energy use in German energy statistics. The system boundaries for non-energy use in energy statistics are not uniform for the various types of fuels (i.e., a gross definition for coal-, lignite-, and oil-derived feedstock and a net definition for natural gas is applied, see Section 4.2.1).

83 Chapter 4

We explain this inconsistency with different levels of insight obtained by data suppliers into the consumption structure of non-energy use32. We recommend harmonizing the system boundaries of non-energy use. Ideally, a uniform approach should be agreed upon internationally. This would not only simplify the calculation of non-energy use emissions for the various relevant source categories, it would also allow international comparisons. From this research, we therefore recommend to link the processes of preparing both GHG inventories and national energy statistics more closely to each other.

Insight into energy statistics is also important for the application of IPCC default carbon storage fractions in the IPCC-RA (IPCC, 1997) because these storage fractions are often inconsistent with the system boundaries of non-energy use data. To simplify the calculation of emissions in the IPCC-RA, the improved 2006 IPCC inventory guidelines generally recommend applying storage fractions of 100%, thereby completely removing non-energy use emissions from the IPCC-RA and only considering them in the relevant source categories of the IPCC-SA.

Although solving the difficulties regarding the IPCC-RA storage fractions, the 2006 inventory guidelines no longer allow for crosschecking non-energy use emissions within GHG inventories (IPCC, 2006). Such data comparison can, however, be extremely useful for reliable and complete emissions accounting based on the IPCC-SA method, as the data gaps in the German GHG inventory have demonstrated (see Figure 4.1). We therefore recommend inventory experts to apply independent model tools such as NEAT or NEAT-SIMP (Weiss et al., 2009a) to check the completeness of IPCC-SA emission estimates. NEAT-SIMP is a simplified version of the NEAT model that avoids the detailed mass balance calculations of NEAT. NEAT-SIMP requires considerably less input data but still generates reliable emission estimates for the various source categories. NEAT- SIMP can, however, not be used for estimating process-specific emissions resulting from conversion processes in the chemical industry.

Due to their high level of detail, NEAT results allow us to identify and reduce gaps regarding industrial process emissions in the IPCC-SA of the German GHG inventory. The accounting of CO2 emissions from chemical conversion processes in the current German IPCC-SA goes beyond the reporting requirements of both IPCC (1997) and IPCC (2006). We acknowledge this fact but point to remaining gaps and uncertainties within the German GHG inventory. Emissions from steam cracking are still not included under industrial processes in the 2007 inventory submission. The comparison between NEAT

32 For example, non-energy use data for oil-based feedstock and refinery products are provided in Germany by individual refineries. They lack the detailed insight into the final consumption of their deliveries within the chemical sector and can, therefore, only report less detailed gross deliveries. The non-energy use data of natural gas, on the other hand, are provided by the Association of the German Chemical Industry (VCI - Verband der Chemischen Industrie e.V.). This association has relatively good insight into the consumption of natural gas by the various chemical companies and can thus distinguish the fractions consumed either as feedstock or as fuel. The VCI delivers, hence, net data for non-energy use, excluding the fractions of natural gas used for combustion purposes in chemical processes. Another factor that explains the current non-energy use reporting in Germany (and possibly also in other countries) is that natural gas is used directly as fuel for, e.g., ammonia production, whereas naphtha and other oil products are only indirectly consumed as fuel via the production of, e.g., hydrogen and other waste gases.

84 Non-energy use and related emissions in Germany and IPCC-SA furthermore revealed that the applied IPCC default emission factor for ammonia production is incorrect for Germany given feedstock distribution, system boundaries of non-energy use, and CO2 sequestration for urea production. Moreover, emission estimates for the production of non-ferrous metals, ferroalloys, and inorganic chemicals are incomplete in the IPCC-SA (2007 inventory submission). For these cases, we recommend adapting the emission factors and completing the IPCC-SA calculations by using our NEAT model results.

Although the current reporting of solvent and other product use emissions in the German GHG inventory is consistent with the requirements stated by IPCC (1997), the improved IPCC (2006) guidelines request more detailed calculations. We therefore recommend complementing emission estimates from solvent use (Theloke, 2000; Jepsen, 2004) by emission data for the most prominent sources of product use emissions, i.e., the consumption of lubricants, waxes, and paraffins. We furthermore suggest using our estimates on emissions from wastewater treatment as a benchmark for further, more detailed IPCC-SA analysis of emissions from this source category.

4.5 Conclusions

In this chapter, we apply the NEAT model to estimate non-energy use and related CO2 emissions for Germany in the period from 1990 to 2003. NEAT calculates both parameters independently from energy statistics and national GHG inventories, providing hence an important consistency check for official data. We regard our estimates to be reliable and useful for filling data gaps in the respective source categories of the IPCC-SA. A draw- back with respect to the applicability of NEAT is the requirement for large numbers of production and trade data as well as detailed insight into the German chemical industry for calculating in particular emissions from chemical conversion processes, carbon storage, and total non-energy use. To avoid extensive data collection, a simplified version of NEAT (i.e., NEAT-SIMP) can be used for estimating non-energy use and related emissions for the most important source categories.

The NEAT model helped to identify and reduce inconsistencies in the German GHG inventory to a large extent. By implementing the recommendations given in this chapter, most of the shortcomings that remain in the German GHG inventory submission of 2007 can be addressed in a satisfactory manner. Uncertainties that deserve special attention in the future are firstly related to the definition of non-energy use and secondly to emission estimates for (i) the use of electrodes and other solid carbon in the production non-ferrous metals, ferroalloy, and inorganic chemicals and (ii) the consumption of lubricants, waxes, and paraffins.

We finally conclude that applying the NEAT model to Germany has made an important contribution to a more accurate and reliable accounting of GHG emissions. The insight gained by this research is not only relevant for Germany but more generally for the accounting of non-energy use and related emissions in any other country. We summarize our conclusions with the following recommendations for inventory experts:

85 Chapter 4

(i) Identify the system boundaries of non-energy use data as applied in national energy statistics and allocate emissions accordingly to the various relevant source categories outlined by the IPCC-SA (i.e., either energy or industrial processes, solvent and other product use, agriculture, or waste). (ii) Ensure completeness of source categories relevant for non-energy use emissions. (iii) Be cautious with the use of default emission factors. Ensure that the applied emission factors account for country specifics such as system boundaries of non-energy use, feedstock composition, and plant efficiencies. (iv) Aim at reducing uncertainties of emission estimates by applying more detailed approaches according to Tier 2 and Tier 3 methodology (IPCC, 1997).

Acknowledgements This research was funded by the Federal Environmental Agency (Umweltbundesamt) of Germany (Environmental Research Plan 2003, Project FKZ 203 412 53/02) and is closely linked to the activities of the ‘International Network on Non-Energy Use and CO2 Emissions’ (NEU-CO2, Phase III, funded by the European Commission). The authors are grateful to Michael Strogies and his colleagues from the Federal Environmental Agency of Germany for the fruitful cooperation. The authors would furthermore like to thank Alexandra Newman from the Colorado School of Mines (Golden, USA) and two anonymous reviewers for their valuable comments on earlier drafts of this chapter.

Appendix Table A4.1: Chemical conversion processes and process-specific emission factors as implemented in NEAT Production of Feedstock consumption Generation of by- Chemical conversion losses in Feedstock By-products in t/t product products in t/t product t CO2 equivalents/t product Cumene 1.35 Acetone 0.61 - Phenol Toluene 1.20 Benzene 0.01 1.17 Dichloropropane 0.11 Propylene 0.88 0.33 Dichloroethylether 0.03 Propylene 0.90 misc. Acetone 0.25 - Propylene oxide 87 Isobutene 2.35 Butyl alcohol 2.45 0.97 Propylene 0.74 0.06 Styrene 2.29 Ethylbenzene 2.52 0.62 Cyclohexane 1.03 - - 0.88 Caprolactam Phenol 0.92 - - 0.25 o-Xylol 0.92 Maleic acid 0.05 0.60 emissionsinGermany Non-energy useandrelated Phthalic anhydride Naphthalene 0.92 - - 0.78 Acrylonitrile 1.13 - - 0.54 Adipic acid 1.48 - - 0.40 Adiponitrile Butadiene 0.63 - - 0.76 Hydrogen cyanide 0.60 - - - Acrylonitrile Propylene 1.06 Hydrogen cyanide 0.08 0.71 Adipic acid Cyclohexane 0.75 - - 0.55 Ethylene oxide Ethylene 0.78 - - 0.45 Toluene 0.67 - - - Toluene diisocyanate Carbon monoxide 0.43 Heavy products 0.08 0.44 Bisphenol-A Phenol 0.88 - - - Acetone 0.29 - - 0.37 Toluene 0.04 - - -

88 Chapter 4 Table A4.1 (cont.): Chemical conversion processes and process-specific emission factors as implemented in NEAT Production of Feedstock consumption Generation of by- Chemical conversion losses in Feedstock By-products in t/t product products in t/t product t CO2 equivalents/t product p-Xylene 0.63 - - 0.27 Dimethyl terephthalate Methanol 0.38 - - 0.07 Polyamide-6 Caprolactam 1.11 - - 0.26 Isopropanol Propylene 0.78 - - 0.25 Bisphenol-A 0.90 - - - Polycarbonate Carbon monoxide 0.23 - - 0.19 2-Ethylhexanol 0.73 - - - Dioctyl phthalate Phthalic anhydride 0.38 - - 0.17 p-Xylene 0.66 - - 0.07 Terephthalic acid Acetic acid 0.05 - - 0.07 Aniline 0.76 - - - Methelene di-phenylene- Formaldehyde 0.14 - - 0.07 isocyanate Carbon monoxide 0.26 - - 0.07 Formaldehyde Methanol 1.15 - - 0.12 Acetaldehyde Ethylene 0.67 - - 0.11 Acetaldehyde 0.76 - - 0.06 Acetic acid Methanol 0.54 - - 0.05 Carbon monoxide 0.53 - - 0.05 Propylene 0.66 i-Butyraldehyd 0.09 0.07 n-Butanol Carbon monoxide 0.44 - - 0.02 Vinyl chloride Ethylene 0.47 - - 0.07 Aniline Benzene 1.35 - - 0.06 Diethylene glycol 0.1 Ethylene glycol Ethylene oxide 0.83 0.05 Triethylene glycol 0.01

Table A4.1 (cont.): Chemical conversion processes and process-specific emission factors as implemented in NEAT Production of: Feedstock consumption Generation of by- Chemical conversion losses in Feedstock By-products in t/t product products in t/t product t CO2 equivalent/t product Benzene 0.01 Styrene Ethylbenzene 1.07 0.05 Toluene 0.02 Polyvinyl chloride Vinyl chloride 1.03 - - 0.04 Ethylene glycol 0.33 - - 0.03 PET Terephthalic acid 0.87 - - - Propylene 0.35 - - 0.02 Cumene Benzene 0.66 - - 0.01 Polystyrene Styrene 1.01 - - 0.03 Acetone Isopropanol 1.05 - - 0.03 Polyethylene Ethylene 1.01 - - 0.03 Glycerol 0.03 - - -

Polyether polyols emissionsinGermany Non-energy useandrelated Propylene oxide 1 - - 0.02 Ammonia 0.57 - - - Urea Carbon dioxide 0.75 - - 0.02 Cyclohexane Benzene 0.93 - - 0.02 Polypropylene Propylene 1.01 - - 0.02 Benzene 0.74 - - - Ethylbenzene Ethylene 0.27 - - 0.01

5 Market diffusion, technological learning, and cost-benefit dynamics of condensing gas boilers in the Netherlands

Martin Weiss, Lars Dittmar, Martin Junginger, Martin K. Patel, and Kornelis Blok Published in: Energy Policy 37 (2009), pp. 2962-2976.

Abstract High costs often prevent the market diffusion of novel and efficient energy technologies. Monitoring cost and price decline for such technologies is thus important in order to establish effective energy policy. Here, we present experience curves and cost-benefit analyses for one efficient energy demand technology, i.e., condensing gas boilers produced and sold in the Netherlands between 1981 and 2006. For condensing gas combi boilers, which are the most dominant boiler type on the Dutch market, we identify learning rates of 14 ± 1% for the average price and 16 ± 8% for the additional price relative to non- condensing devices. Economies of scale, competitive sourcing of boiler components, and improvements in boiler assembly are among the main drivers behind the observed price decline. The net present value of condensing gas combi boilers shows an overall increasing trend. Purchasing in 2006 a gas boiler of this type instead of a non-condensing device generates a net present value of 970 EUR (Euro) and realizes CO2 (carbon dioxide) emission savings at negative costs of -120 EUR per tonne CO2. We attribute two thirds of the improvements in the cost-benefit performance of condensing gas combi boilers to technological learning and one third to a combination of external effects and governmental policies.

91 Chapter 5

5.1 Introduction Introducing renewable energy technologies on a large scale and improving the efficiency of energy demand and supply technologies are the two main trajectories towards a sustainable energy system (IEA, 2008c). Energy efficiency improvements are thereby often regarded as the most cost-effective and readily available means for reducing non- renewable energy consumption and greenhouse gas GHG emissions (EZ, 2005; IEA, 2008d, e). Despite substantial improvements in past decades, considerable energy efficiency potentials still exist across both countries and economic sectors (IEA, 2008d). Whether these potentials can be realized will depend on the market diffusion of novel and efficient energy demand and supply technologies. However, high initial investment costs often present a key barrier for the market success of these technologies. Novel technologies are relatively expensive at the point of market introduction but eventually become cheaper due to mechanisms such as learning-by-doing, technological innovation, and economies of scale. The combined effect of these mechanisms on production costs is generally referred to as technological learning. Aspects of technological learning can be captured by the so-called experience curve approach1.

The experience curve approach is an empirical concept hypothesizing that production costs of a technology decline at a constant rate with each doubling of cumulative production (BCG, 1972). In the 1970s and 1980s, the experience curve approach was primarily applied as a management tool in manufacturing industries (Argote and Epple, 1990). In the 1990s, it gained importance as an instrument to forecast costs and diffusion rates of energy technologies and for building energy and CO2 emission scenarios (Wene et al., 2000; IEA, 2000, 2008c; van Vuuren et al., 2006). The experience curve approach has been extensively applied to and redefined for renewable and non-renewable energy supply technologies (Junginger et al., 2004, 2005, 2006, 2008; McDonald and Schrattenholzer, 2001; Neij, 1999). Its application to efficient energy demand technologies is, however, still scarce (e.g., Iwafune, 2000; Laitner and Sanstad, 2004).

Here, we address this knowledge gap by constructing experience curves for one efficient energy demand technology, i.e., condensing gas boilers. In addition, we analyze the importance of technological learning for costs and benefits of condensing gas boilers from both a consumer and a governmental perspective. We focus our analysis on condensing gas boilers in the Netherlands. Choosing this technology is justified because condensing gas boilers offer substantial energy savings in the residential and commercial building sector of the European Union (EU). Weber et al. (2002) estimate that condensing gas boilers save about 5% of primary energy and 4% of CO2 emissions related to residential space heating in the EU. National as well as European policies targeted at energy efficiency improvements address condensing gas boilers as an important energy- efficiency technology (EU, 2005, 2006a). We focus on the Netherlands because this country has been an international front-runner in the manufacturing and market diffusion of condensing gas boilers. This technology was introduced to the Dutch boiler market already in 1981. Nowadays, condensing gas boilers are a mature and fully developed

1 We use here the term experience curve instead of learning curve because the latter is typically used for approaches that quantify the decline in labor costs only (Junginger et al., 2008).

92 Technological learning of condensing gas boilers in the Netherlands technology. Condensing gas boilers are by far the most dominant boiler technology in the Netherlands, reaching market shares of more than 90% in 2006. This allows us to study an efficient energy demand technology that has completed all stages of its technology life cycle, i.e., invention, market introduction, market diffusion, and market saturation (see, e.g., Grübler et al., 1999).

In the next section of this chapter, we give a brief overview of the history of condensing gas boilers in the Netherlands. In Section 5.3, we explain our methodology for (i) constructing experience curves and (ii) analyzing costs and benefits of this technology. We present our results in Section 5.4, and we discuss methodology as well as implications of our findings for policy makers in Section 5.5. In Section 5.6, we draw conclusions.

5.2 A short history of condensing gas boilers in the Netherlands In the 1960s, conventional non-condensing gas boilers with relatively low efficiencies of around 80%2 became standard technology for space heating in the Netherlands (van Overbeeke, 2001; Mooi, 2004). Triggered by increasing energy prices after the first oil crisis in 1973 and by expectations in a growing replacement market for technically obsolete gas boilers, the natural gas supplier Gasunie3 began researching options to improve the energy efficiency of conventional non-condensing gas boilers. In 1978, Gasunie applied for patents for the first prototype of a so-called condensing gas boiler, which was able to utilize the latent heat of evaporation contained in the water vapour of flue gases. Based on this first prototype, Dutch boiler manufacturers started to develop own models of condensing gas boilers (Gasunie, 1982; Aptroot and Meijnen, 1993; Weber et al., 2002). In 1981, condensing gas boilers from six Dutch manufacturers were successfully introduced to the Dutch boiler market. The first generation of these boilers reached efficiencies of around 101–103% (Gasunie, 1982). Although receiving subsidies until 1985, condensing gas boiler sales remained low due to the following reasons (Gasunie, 1982; Brezet, 1994; Gasterra, 2007; Sijbring, 2007; Nefit, 2008): (i) Installers lacked training and experience with the new technology and invested only insufficiently in marketing activities. (ii) Condensing gas boilers put additional requirements on household infrastructure (e.g., installation of non-corrosive flues, condensate drainage to the sewage system, air supply). (iii) Condensing gas boilers were less reliable and far more expensive than conventional non-condensing gas boilers. (iv) In 1981, boiler producers also introduced conventional non-condensing boilers with improved efficiencies (approximately 88%) to the market. These boilers were cheaper and easier to install than condensing gas boilers.

2 Throughout this chapter, we express all efficiencies based on the lower heating value (LHV) of fuels, e.g., natural gas. The thermodynamic maximum efficiency of condensing gas boilers is 111%. This efficiency assumes that the energy content of natural gas (including the energy contained in the water vapor of the flue gas) is fully converted into useful heat. 3 N.V. Nederlandse Gasunie is a leading gas company in the Netherlands.

93 Chapter 5

The first three obstacles were addressed by the various stakeholders in consecutive years. Condensing gas boiler sales remained nevertheless lower than expected, reaching a market share of only 8% by 1987. Aptroot and Meijnen (1993) argue that high consumer investment costs and falling natural gas prices between 1985 and 1989 caused unattractively high payback times for condensing gas boilers in the mid to late 1980s. However, from 1990 onwards the Dutch gas boiler market experienced a rapid shift from non-condensing to condensing gas boilers. This development lead to a displacement of mainly non-condensing gas space heating boilers by so-called condensing gas combi boilers, which provide both space heating and hot tap water from one single boiler unit (Figure 5.1).

500 Condensing gas space heating boilers 450 Condensing gas combi boilers Non-condensing gas boilers 400 (space heating and combi) Non-condensing gas space heating boilers 350 Non-condensing gas combi boilers

300

250

200 in 1000 units

150

100 Gas boiler sales inthe Netherlands 50

0 1970 1980 1990 2000 Year Figure 5.1: Sales of condensing and non-condensing gas boilers in the Netherlands4 (data sources: CBS, 2007; Aptroot and Meijnen, 1993; Remeha, 2007; Sijbring, 2007)

By 1996, condensing gas boilers were the dominant boiler technology in the Netherlands. Since 2000, the market share of condensing gas boilers exceeds 80%. The rapid market diffusion in the 1990s was mainly caused by the following factors (Gasunie, 1982; Brezet, 1994; Gasterra, 2007; Sijbring, 2007; Nefit, 2008): (i) the reintroduction of a subsidy scheme, which had been absent in 1988 and 1989

4 In years prior to 1992, available data do not allow differentiating between non-condensing gas space heating and combi boilers.

94 Technological learning of condensing gas boilers in the Netherlands

(ii) technological developments in conventional non-condensing gas boiler manufacturing that led to a switch from inexpensive open to more expensive closed boiler systems5, thereby making condensing gas boilers increasingly attractive for the replacement market, (iii) continuous cost decline in production, installation, and maintenance of condensing gas boilers

In 2006, condensing gas combi boilers reached sales of 387,000 units thereby accounting for 85% of total gas boiler sales and even 93% of condensing gas boiler sales in the Netherlands. Throughout past decades, the Dutch government did not follow a consistent subsidy policy for condensing gas boilers. A first subsidy program was set up for the period between 1981 and 1985, granting subsidies of around 250 Dutch Guilders (NLG), i.e., 113 Euro (EUR) per condensing gas boiler6. Between 1985 and 1987, subsidies covered 33–40% of additional investment and installation costs for condensing gas boilers (Consumentenbond, 1983–2006). A new subsidy program started in 1990, granting 350 NLG (159 EUR) per condensing gas boiler. The government stopped this program in 1993 for budgetary reasons (Weber et al., 2002). In 1994, a final subsidy program was initiated, granting 100 NLG (45 Euro) per condensing gas boiler. This program ended in 2002. From this year onwards, no subsidies have been provided for condensing gas boilers in the Netherlands. Based on Dougle and Oosterheert (1999), Eiff et al. (2001), Joosen et al. (2004), and Oude Lohuis (2004), we estimate that in the Netherlands total nominal subsidies of roughly 70 ± 10 million EUR were spent in direct support of condensing gas boilers.

Having described the market diffusion of condensing gas boilers in the Netherlands, we set the stage for our empirical analysis. We now explain in detail the methodology of both experience curve approach and cost-benefit analysis.

5.3 Methodology and data sources In this section, we first present our methodology for constructing experience curves. Afterward, we explain in detail approach and data sources used for our cost-benefit analysis.

5 Open boilers receive air from the buildings’ interior. Closed boilers, by contrast, receive air via a pipe from outside of the building. Whereas condensing gas boilers have generally always been produced as closed boilers, this has not been the case for conventional non-condensing boilers. Non-condensing boilers were in the majority open boilers until the early 1990s (Gasterra, 2007; Sijbring, 2007; Nefit, 2008). The shift from open to closed boilers caused a temporary increase in the price of non-condensing gas boilers. The introduction of closed non-condensing gas boilers was, however, necessary to reduce heat losses due to convection. 6 We uniformly apply an exchange rate of 2.20371:1 to convert NLG into EUR. Data expressed in monetary units are given in this chapter either in nominal terms [EUR] or in real terms, deflated to the base year of 2006 [EUR2006]. For deflation, we use consumer price indices as given by CBS (2007).

95 Chapter 5

5.3.1 Constructing experience curves The experience curve approach models costs of a technology as a power-law function of cumulative production:

= ⋅ bi Ccumi C0,i (Pcumi ) (5.1) where Ccumi [EUR2006/kWth] represents here the price at Pcumi, C0,i [EUR2006/kWth] the price of the first unit produced, Pcumi [MWth] the cumulative experience (i.e., cumulative production), and bi the product-specific experience index of boiler type i. By applying log10 transformation to Equation (5.1), we can plot a linear experience curve with bi as slope parameter and log C0,i as the price-axis intercept. Based on this methodology, we calculate product-specific progress ratio (PRi) [%] and learning rate (LRi) [%] as:

= bi PRi 2 (5.2)

= − = − bi LRi 1 PRi 1 2 (5.3)

We estimate the error interval of PRi and LRi as the implicit error of the regression analysis, i.e., the 95% confidence interval of the slope parameter of the experience curve. We uniformly use market prices as proxy for actual production costs and we use Dutch market sales as a proxy for actual condensing gas boiler production. The first proxy is a simplification that is, strictly speaking, only valid for competitive markets where prices closely follow production costs (BCG, 1972). This is generally the case for condensing gas boilers in the Netherlands (see also the discussion in Section 5.5.1). Moreover, the use of prices is widely accepted practice in experience curve analysis because data on actual production costs are generally kept confidential by producers (see, e.g., Junginger et al., 2008). We justify the use of Dutch sales data as a proxy for condensing gas boiler production based on two considerations. First, Dutch boiler producers claim that condensing gas boilers have been invented (late 1970s), developed (early 1980s), as well as produced (until the early 1990s) without substantial technology spillover from other countries and regions (Nefit, 2008; Remeha, 2007). It is hence reasonable to assume a national learning system. Second, Dutch sales data proved to be more reliable and more readily available than Dutch production data. Furthermore, net trade of condensing gas boilers is negligible until the early 1990s and only minor relative to domestic production in the years afterwards (Remeha, 2007; Nefit, 2008). The use of Dutch sales data thus allows us to construct reliable experience curves for relatively long time periods. However, our approximation of cumulative condensing gas boiler production by Dutch sales data disregards knowledge spillover to and from the Netherlands since the early 1990s. To quantify the resulting uncertainties, we conduct a sensitivity analysis based on cumulative sales of condensing gas boilers within the fifteen Western European member countries of the European Union (EU-15; see discussion in Section 5.5.1).

In the first instance, we construct experience curves separately for condensing gas combi and space heating boilers. Subsequently, we extend the conventional experience curve approach by constructing experience curves for the additional price of the two

96 Technological learning of condensing gas boilers in the Netherlands condensing gas boiler types relative to non-condensing improved-efficiency gas boilers. We regard such analysis as useful because it reveals insight into the dynamics of costs that result from upgrading non-condensing gas boilers to condensing devices (e.g., costs related to the installation of an additional heat exchanger and the adjustment of internal boiler settings).

We obtain price data for condensing and non-condensing gas combi and space heating boilers from various sources, i.e., AGPO (2007), AWB (2007), Consumentenbond (1983–2006), Itho (2007), Nefit, (2001, 2007), Remeha (2007), Vaillant (2007), and Warmteservice (2007). We chose to include data from all these sources into our experience curve analysis because this allows us to estimate average prices for a higher number of individual years. We include in our analysis only gas boilers with a capacity less than or equal to 30 kWth because these boilers are typically used for central heating and hot tap water production in Dutch households (CBS, 2007; Remeha, 2007). Based on available price data, we calculate price averages and related standard deviations for individual years.

We estimate cumulative condensing gas boiler sales in the Netherlands based on data as provided by Aptroot and Meijnen (1993), Remeha (2007), and Sijbring (2007). For our sensitivity analysis, we estimate condensing gas boiler sales in the EU-157 using available sales data for France, Germany, the UK, and the Netherlands. These four countries account together for roughly 90% of the total condensing gas boiler sales in the EU-15 (Remeha, 2007; Weber et al., 2002).

5.3.2 Cost-benefit analysis We perform cost-benefit analysis for condensing gas combi and space heating boilers of 8 25 kWth capacity from both a consumer and a governmental perspective. We first take a consumers perspective and calculate net present value9, internal rate of return, and simple payback time10 of condensing gas combi and space heating boilers relative to non- condensing improved-efficiency gas boilers. The internal rate of return (IRRij) thereby represents the consumer discount rate (rij) at which the net present value (NPVij) of

7 Despite substantial gas boiler sales in countries like Korea and Japan, we limit our sensitivity analysis to the EU-15 because complete time-series data that would allow an estimation of cumulative global condensing gas boiler sales or production are not available. Our approach is justified because knowledge spillover is more likely to occur between European countries than with countries outside of Europe due to close spatial proximity of European countries and a de-regulated domestic European market. 8 The average capacity of condensing and non-condensing gas combi and space heating boilers varies between 16 and 26 kWth but shows an overall increasing trend in the period between 1983 and 2006. To assure data consistency, we recalculate boiler prices uniformly for a boiler of 25 kWth capacity. We assume a linear relationship between price and boiler capacity in the specified capacity range based on Consumentenbond (1983–2006). 9 We assume that parameters affecting costs and benefits (e.g., natural gas prices, household natural gas consumption) remain constant at the level of the year of investment. 10 Simple payback time is a widely applied criterion for evaluating the profitability of investments. The simple payback time is a simplified parameter for approximating actual payback time that disregards (i) interest rate, i.e., the implicit discount rate of consumer investments and (ii) changes in costs and benefits that might become effective during the use phase of condensing gas boilers.

97 Chapter 5 investments into a condensing gas boiler of type i in year j becomes zero. The internal rate of return in individual years fulfils the following criterion:

= ; = IRRij rij NPV (rij ) 0 (5.4)

To obtain more detailed insight into the drivers of the observed cost-benefit dynamics, we disaggregate simple payback times for condensing gas combi boilers, which is the most dominant boiler type on the Dutch market. We distinguish three principle factors, (i) technological learning (i.e., changes in installation and maintenance costs, boiler prices, and boiler efficiencies), (ii) external effects (i.e., price and consumption of natural gas), and (iii) governmental policy (i.e., taxes and subsidies).

In the second part of our cost-benefit analysis, we take a governmental perspective and calculate specific costs for realized CO2 emission savings as:

α × I + C − B C = ij i i (5.5) CO2 ,ij ΔM CO2 ,ij whereC [EUR /t CO ) stands for the costs of realized CO emission savings, for CO2 ,ij 2006 2 2 the capital recovery factor, Iij for the initial investment, Bij for the annual benefits

[EUR ], Cij for the annual costs (excluding capital costs) [EUR ], and ΔM 2006 2006 CO2 ,ij

[t CO2] for CO2 emissions saved by a condensing gas boiler of type i bought in year j instead of a non-condensing improved-efficiency gas boiler.

Our choice to compare condensing gas boilers with improved-efficiency non- condensing gas boilers is justified because consumers who installed a condensing gas boiler would have chosen a so-called improved-efficiency non-condensing gas boiler (instead of a less efficient non-condensing standard-efficiency gas boiler), if condensing gas boilers were not available. We perform cost-benefit analysis for condensing gas combi boilers only for the period from 1988 to 2006 because data on both boiler prices and actual natural gas consumption for hot tap water production in Dutch households are not available to us for earlier years.

We base our cost-benefit analysis on the same data sources as used for our experience curve analysis. Additional information is provided by CBS (2007), EnergieNed (1981–2006, 2006), and Visser (2007) (see Tables A5.1-A5.3 in the Appendix of this chapter). For Dutch natural gas, we assume a lower heating value of 31.67 MJ/m3 and a 3 specific CO2 emission factor of 1.78 kg CO2/m (Gasunie, 1988; IPCC, 1995).

98 Technological learning of condensing gas boilers in the Netherlands

5.4 Results We begin by presenting and explaining the results of our experience curve analysis in Section 5.4.1. Afterward, we analyze costs and benefits of condensing gas boilers relative to non-condensing gas boilers from both a consumer and a governmental perspective (Section 5.4.2).

5.4.1 Experience curve analysis We first provide an overview of price data used for our experience curve analysis (Figure 5.2). Condensing gas combi boilers show the highest price decline in real terms, i.e., 50% between the years 1988 and 2006 or on average 4% per year.

100 th /kW

2006 80

40 Average boiler price in EUR

20 1985 1990 1995 2000 2005 Year

Non-condensing improved-efficiency gas combi boilers Non-condensing improved-efficiency gas space heating boilers Non-condensing standard-efficiency gas combi boilers Non-condensing standard-efficiency gas space heating boilers Condensing gas combi boilers Condensing gas space heating boilers

Figure 5.2: Average prices for condensing and non-condensing gas boilers in the Netherlands between 1981 and 2006

The price of condensing gas space heating boilers declines by 39% (1983–2006), whereas the price decline of non-condensing gas boilers ranges from 40% (1988–2006) for improved-efficiency gas combi boilers to 29% (1983–2006) for standard-efficiency gas combi boilers (Figure 5.2). Based on the price data presented in Figure 5.2, we now construct two sets of experience curves (i) for the price of condensing gas combi and space heating boilers and (ii) for the additional price of condensing gas combi and space heating boilers relative to non-condensing improved-efficiency gas boilers.

99 Chapter 5

The price of condensing gas combi and space heating boilers declines at learning rates of 14 ± 1% and 6 ± 1%, respectively (Figure 5.3 and Figure 5.4).

1990 (2)

100

90 1988 (4) 1998 (13) 80 th 2002 (35)

/kW 70 1994 (7) 2006 60 2004 (3) = ± (−0.22±0.01) in EUR y (615 67)x 50 2001 (8) R2 = 0.98 LR = 14 ± 1% 2005 (8) 40

Price of condensing gas combi boilers boilers combi gas of condensing Price PR = 86 ± 1% 2006 (51)

104 105 Cumulative sales of condensing gas boilers in the Netherlands in MW th Figure 5.3: Experience curve for condensing gas combi boilers in the Netherlands for the period from 1988 to 2006; in parentheses, number of price data included in our analysis; error bars indicate the standard deviation of price data

100 90

80 1983 (6) 70 1990 (5) th 1988 (4) 60 /kW 1994 (6) 2006 50 1997 (13)

in EUR 1998(13) 40 y = (163 ± 16 )x ( −0 .10± 0.01) 2 R = 0.92 2006 (16) 30 LR = 6 ± 1% PR = 94 ± 1% 2001 (7) 2002 (22) Price of condensing gas space heating boilers

103 104 105 Cumulative sales of condensing gas boilers in the Netherlands in MW th Figure 5.4: Experience curve for condensing gas space heating boilers in the Netherlands for the period from 1983 to 2006; in parentheses, number of price data included in our analysis; error bars indicate the standard deviation of price data;

100 Technological learning of condensing gas boilers in the Netherlands

Our results exceed the findings of Martinus et al. (2005), who identified a learning rate of 4% for condensing gas boilers based on price data and cumulative installed capacities referring to the Netherlands. By recalculating the results of Martinus et al. (2005), we can entirely explain the observed differences. Martinus et al. (2005) cover with their analysis only the period between 1983 and 1997 and they do not differentiate between non-condensing gas combi and space heating boilers. If we consider the same time period as Martinus et al. (2005), we identify a learning rate of 4 ± 5% for condensing gas boilers as a whole. If we distinguish between the two types of condensing gas boilers, we find learning rates of 8 ± 3% for condensing gas combi boilers and 2 ± 3% for condensing gas space heating boilers, respectively.

The following factors characterize product improvements and explain the observed price decline for both condensing gas combi and space heating boilers (Gasterra, 2007; Remeha, 2007; Sijbring, 2007; Nefit, 2008): (i) economies of scale and increased automation in both boiler assembly and component manufacturing during the entire period since market introduction in 1981 (ii) reduction of boiler size, thereby reducing material costs for, e.g., heat exchangers by roughly 50% in the entire period since market introduction in 1981 (iii) improvements in quality and reliability of boiler components (i.e., shift from non-modulating to modulating pre-mix burners11 in the early 1990s, introduction of internal boiler diagnosis systems, and modulating ventilators) (iv) price reduction and performance improvement of control electronics since the early 1990s (v) increasing competition among component manufacturers with rising shares of components being imported from low-wage countries like China (since the 1990s) (vi) standardization of boiler components and competitive outsourcing of component production to specialized companies (especially since the end of the 1990s) (vii) further streamlining of boiler assembly lines, decreasing assembly times, custom-made just-in-time manufacturing, and reduction of on-site stocks of components, semi-finished and finished boilers in recent years

Major changes regarding individual components refer especially to heat exchangers and control electronics. In the early 1980s, heat exchangers were the single largest cost component in condensing gas boiler manufacturing, accounting for 30% of total boiler production costs. This share has been reduced drastically in the past 25 years. Nowadays, control electronics are the most important cost component (Gasterra, 2007; Remeha, 2007; Sijbring, 2007). Looking at the whole time period between 1981 and 2006, technological developments allowed for a reduction in both volume and weight of

11 Modulating burners allow for a dynamic adjustment of burner capacity depending on actual heat demand. Non-modulating burners function based on a simple on-off mechanism that does not allow for dynamic capacity adjustments.

101 Chapter 5 condensing gas boilers by a factor of 2–3. Especially competitive outsourcing of component production to specialized companies since the end of the 1990s has been a major driver for decreasing production costs of condensing gas boilers (Gasterra, 2007; Remeha, 2007; Sijbring, 2007). The recent merging of boiler producers offers further potential for economies of scale, thus declining costs for raw materials, components, marketing, and research and development.

We explain the difference in the learning rates for condensing gas combi and space heating boilers with: (i) the effect of additional technological learning in adapting internal boiler settings for providing combined space heating and hot tap water (ii) scale effects of the market, i.e., the market dominance of condensing gas combi boilers enabled producers to realize additional economies of scale, thus reducing production costs of condensing gas combi boilers to a larger extent than it is the case for space heating devices (Nefit, 2008)12

In the second part of our experience curve analysis, we now focus on the price difference between condensing and non-condensing gas boilers. This part of the analysis is especially relevant for policy makers because the price difference between both gas boiler types is one important indicator of the effort that is required to stimulate market up-take of condensing gas boilers13.

In line with price trends depicted in Figure 5.2, we identify a general trend towards declining price differences between condensing and non-condensing gas boilers (Figure 5.5). For the additional price of condensing gas combi boilers, which is nowadays the dominant boiler type on the Dutch market, we identify a learning rate of 16 ± 8%. Our results also indicate that the price difference between combi boilers (condensing versus non-condensing improved-efficiency combi boilers) declines roughly twice as fast as the price difference between the various types of space heating boilers. This finding is explained to some extent by additional economies of scale in the manufacturing of condensing gas combi boilers.

We also find that differences in the length of the analyzed time periods (i.e., 1988– 2006 for combi boilers versus 1983-2006 for space heating boilers) have an impact on our results. The shorter time period covered by the data for combi boilers leads to a lower number of doublings of cumulative sales. By recalculating our estimates for space heating boilers, we identify for the period 1988-2006 learning rates of 11 ± 3% and 19 ± 8% for the additional price of condensing gas boilers compared to non-condensing improved- efficiency and standard-efficiency gas boilers, respectively. This result shows that

12 Note that both increasing cumulative production and rising levels of actual production potentially reduce production costs. High volumes of actual production enable economies of scale that often help to reduce production costs beyond levels caused by increased cumulative production alone. 13 The price difference between condensing and non-condensing gas boilers is only one important indicator of the efforts required to assure market diffusion. The realized savings of energy and related costs is another one. In the next section, we analyze total life cycle costs of condensing gas boilers compared to non- condensing improved-efficiency gas boilers, thereby providing more comprehensive insights regarding, e.g., subsidy requirements.

102 Technological learning of condensing gas boilers in the Netherlands deviations regarding the analyzed time periods contribute 37% and 42% to the observed differences between the learning rates for the additional price of condensing gas combi and space heating boilers, respectively.

60 1988 50 1983

th 40 /kW 30 2006 2006

15 Price differencePrice in EUR 10 9 8 103 104 105 Cumulative sales of condensing gas boilers

in the Netherlands in MWth

Space heating boilers - condensing versus non-condensing standard-efficiency: R2 = 0.54; LR = 14 ± 9%; PR = 86 ± 9% Space heating boilers - condensing versus non-condensing improved-efficiency: R2 = 0.64; LR = 8 ± 4%; PR = 92 ± 4% Combi boilers - condensing versus non-condensing standard-efficiency: R2 = 0.99; LR = 26 ± 3%; PR = 74 ± 3% Combi boilers - condensing versus non-condensing improved-efficiency: R2 = 0.72; LR = 16 ± 8%; PR = 84 ± 8%

Figure 5.5: Experience curves for the additional price of condensing gas combi and space heating boilers relative to conventional, non-condensing gas boilers in the Netherlands; numbers in the diagram indicate base year and final year of our analysis

5.4.2 Cost-benefit analysis We now present our analysis of costs and benefits associated with condensing gas boilers in the Netherlands. First, we take a consumer perspective and calculate the net present value per condensing gas combi and space heating boiler relative to a non-condensing improved-efficiency gas boiler in the periods between 1988 and 2006 and between 1981

103 Chapter 5 and 2006, respectively14. The net present value of both condensing gas boiler types shows large fluctuations but an overall increasing trend in the time periods analyzed (Figure 5.6).

1000 Condensing gas space heating boilers (excluding subsidies) Condensing gas space heating boilers (including subsidies) Condensing gas combi boilers (excluding subsidies) 800 Condensing gas combi boilers (including subsidies) 2006 600

400

200

0 Net present value in EUR -200

-400 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year Figure 5.6: Net present value of condensing gas boilers relative to non-condensing improved- efficiency gas boilers; assuming 25 kWth boiler capacity and a discount rate of 7%

With the exception of the periods 1988-1989 and 1997-1999, we find a positive net present value for condensing gas combi boilers. By contrast, the net present value of condensing gas space heating boilers (excluding subsidies) is largely negative until 1996. In the years after 1999, both condensing gas boiler types show a positive and steadily increasing net present value relative to improved-efficiency gas boilers. Figure 5.6 also shows that subsidies increased the net present value of condensing gas boilers considerably. The net present value of condensing gas combi boilers is on average higher than the one of condensing gas space heating boilers. We explain this finding by the higher natural gas savings of combi boilers that over-compensate for additional consumer costs relative to condensing gas space heating boilers.

14 For condensing and non-condensing gas space heating boilers, we cover in our cost-benefit analysis the entire period between 1981 and 2006. To do so, we have to estimate average boiler capacities for the years 1981 and 1982 by extrapolation. For this reason, we do not cover these years in our experience curve analysis. Note again that we assume here costs and benefits (e.g., natural gas prices, household natural gas consumption, and maintenance costs) to remain constant at the level of the year in which the investment into a 25 kWth gas boiler is made.

104 Technological learning of condensing gas boilers in the Netherlands

The main drivers for the observed net present value dynamics are (see also below for a detailed analysis of simple payback time): (i) the declining price difference between condensing and non-condensing gas boilers (ii) increased efficiency of condensing gas boilers relative to non-condensing improved-efficiency devices (iii) declining natural gas consumption in households (iv) on average increasing natural gas prices

In particular, the rise of the net present value in the period between 2000 and 2006 strongly follows natural gas price dynamics. By 2006, condensing gas combi and space heating boilers generate a net present value of 970 EUR and 550 EUR per boiler, respectively. Reaching a positive net present value at a consumer discount rate of 7% does not necessarily imply that consumers chose a condensing gas boiler. In fact, consumers often appear to apply much higher implicit discount rates than 7% when making their purchasing decisions (Blok, 2007; Meier and Whittier, 1983; Train, 1985). Therefore, we also estimate the internal rate of return, i.e., the discount rate at which the net present value per condensing gas boiler relative to an improved-efficiency device becomes zero (Figure 5.7).

14 1

12 3

5 10

8 10

6 20 Simple time payback in a

4 Internal rate of in return % 30 2 50

0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year

Condensing gas space heating boilers (excluding subsidies) Condensing gas space heating boilers (including subsidies) Condensing gas combi boilers (excluding subsidies) Condensing gas combi boilers (including subsidies)

Figure 5.7: Simple payback time and internal rate of return for condensing gas boilers relative to a non-condensing improved-efficiency gas boilers; assuming a boiler capacity of 25 kWth

105 Chapter 5

Condensing gas combi boilers show zero net present value at an internal rate of return below 5% in 1988, 1989, and 1998. However, as early as in 1990 subsidies increase the profitability of consumer investments in condensing gas combi boilers to such an extent that the net present value becomes positive even at an internal rate of return of 100% (Figure 5.7). Overall, subsidies considerably increased the internal rate of return for condensing gas combi and space heating boilers. After 1999, both condensing gas combi and space heating boilers reach zero net present value at an internal rate of return of at least 10%. In 2006, the internal rate of return for condensing gas combi and space heating boilers reaches 49% and 25%, respectively. This result indicates the high attractiveness of condensing gas boilers for consumers in recent years, even when applying much higher implicit discount rates than 7%.

Next to net present value and internal rate of return, the simple payback time provides an easily applicable and widely used indicator for the profitability of consumer investments. The simple payback time for purchasing a condensing gas boiler instead of a non-condensing improved-efficiency gas boiler follows the inverse dynamics of net present value (Figure 5.7).

In 1988, 1989, and 1993 condensing gas space heating boilers cannot regain additional consumer investments within their life time without receiving subsidies. After the year 2000, consumer investments in condensing gas combi boilers can be recovered within less than 5 years. The simple payback time for investments in condensing gas combi boilers in the year 1993 specifically highlights the effect of subsidies. In this year, no governmental subsidies were granted for condensing gas boilers. Compared to the years 1992 and 1994, the simple payback time for a condensing gas combi boiler increases from around 5–6 years to 8.5 years, making consumer investments far less attractive. This example points to the importance of subsidies for improving the attractiveness of condensing gas boilers in years when gas prices were low and costs for purchase and installation were high.

To obtain quantitative insight into the drivers of the observed dynamics, we now perform a more detailed analysis for condensing gas combi boilers. In the first step, we analyze the entire time period from 1988 to 2006. We attribute two thirds of the decline in simple payback time to technological learning (i.e., declining prices for purchase and installation as well as efficiency improvements of condensing gas combi boilers)15 and one third to external effects (e.g., changes in natural gas price) and governmental policies (e.g., subsidies and taxes). However, depending on the year of analysis, these shares might vary. To illustrate the variability in the contribution of individual parameters, we differentiate two distinct time periods: (i) 1988–1995 in which natural gas prices remain relatively constant (ii) 1996–2006 in which natural gas prices as well as fees and taxes on energy show a steady increase

15 Note that technological learning is also influenced indirectly by governmental policies through direct and indirect policy effects on cumulative production.

106 Technological learning of condensing gas boilers in the Netherlands

In the period between 1988 and 1996, payback times for condensing gas combi boilers dropped by 50% from 16 years to 8 years. We identify improved efficiency of condensing gas combi boilers to be the single major driver for the observed decline in simple payback time (Figure 5.8)16. In the same period, external effects increase and governmental policies decrease simple payback time, whereas the opposite is true for the period from 1996 to 2006.

16 -7% -8% 14 1

-35% +29% -17% 12 3 4) -21% 5 10 1) +24%

1) -14%

8 3)

1) 10 2) 6

Simple paybacktime 20 4

30 % in rate of Internal return 5) in the period from 1988 to 1996 in to 1996 in a from the 1988 in period 2 50 Parameters 1988 of at the frozen year the level + chang. household natural gas cons. + chang. household + chang. eff. of non-cond. natural gas price changing + + changing installation costs changing installation + maintenance costs + changing + changing boiler prices boiler + changing + subsidies + + chang. eff. of cond. 0 + taxes Technological learning External effects Governmental policy Figure 5.8: Drivers for the dynamics of simple payback time in the period from 1988 to 1996; percentages indicate the contribution to the change of simple payback time (1) aggregated effect of changing prices, installation costs, and maintenance costs for condensing and non-condensing improved-efficiency gas boilers; 2) changing efficiency of condensing gas boilers; 3) changing efficiency of non-condensing improved-efficiency gas boilers; 4) changing household natural gas consumption; 5) including levy for the Environmental Action Plan)

In the latter period, declining prices for condensing gas combi boilers and rising prices for natural gas provide major contributions to the decline of simple payback time (Figure 5.9). Between 1996 and 2006, technological learning contributes two thirds to the reduction of simple payback time from 7 to 2 years, whereas the combination of external effects and governmental policies together contribute around one third (Figure 5.9). Our results demonstrate that technological learning (i.e., declining prices for purchase and installation as well as improvements in the efficiency of condensing gas combi boilers) contributed in both time periods substantially to the cost-effectiveness of condensing gas boilers in the Netherlands. Declining natural gas consumption in Dutch households, primarily due to improvements in wall and roof insulation as well as in window glazing,

16 Note that (i) that scales of y-axes in Figure 5.8 and Figure 5.9 differ from each other and (ii) that the effect of individual parameters on the simple payback time depends on the order of disaggregation. We try to correct for this order-dependency by stating percentage changes in simple payback time always relative to the previous level of disaggregation. Thus, the sum of percentage changes does not equal 100%.

107 Chapter 5 has a substantial adverse effect on the payback time for condensing gas combi boilers in the entire period from 1988 to 2006.

8 10

-37%

-13% 20 -25%

4 4) +28% 1) 1) +6% 1) -56% 30 3) 2) -28% Simple payback time payback Simple

2 +48% 50 Internal rate of return in % 5) in the period from 1996 to 2006 in a in 2006 to 1996 from period the in Parameters frozen at the level of the year 1996 + changing inst. costs inst. changing + + changing maint.costs eff. n.-cond. of ch. + cons. gas n. house. ch. + + ch. eff. of cond. + changing boiler prices + subsidies + taxes 0 + gas price Technological learning External effects Governmental policy Figure 5.9: Drivers for the dynamics of simple payback time in the period from 1996 to 2006; percentages indicate the contribution to the changes of simple payback time (1) aggregated effect of changing prices, installation costs, and maintenance costs for condensing and non-condensing improved-efficiency gas boilers; 2) changing efficiency of condensing gas boilers; 3) changing efficiency of non-condensing improved-efficiency gas boilers; 4) changing household natural gas consumption; 5) including levy for the Environmental Action Plan)

For the second part of our cost-benefit analysis, we now take a governmental perspective. In the period between 1981 and 2006, condensing gas combi and space heating boilers saved in the Netherlands 8.4 Gm3 of natural gas, being equivalent to: 2.1 billion EUR in energy costs (excluding taxes), approximately 270 PJ in primary energy, and 15 Mt in CO2 emissions. In 2006 only, condensing gas boilers saved 2.0 Mt CO2 emissions, which account for roughly 1.2% of total energy-related CO2 emissions of the Netherlands (UNFCCC, 2008). These energy savings were associated with additional consumer costs for purchase and installation of condensing gas boilers. The costs of realized CO2 emission savings (excluding direct governmental taxes and subsidies on natural gas and the purchase, installation, and maintenance of gas boilers) follow the trend of net present value. Both condensing gas combi and space heating boilers save CO2 emissions at steadily declining costs (Figure 5.10). Condensing gas combi boilers are in general more cost-efficient than condensing gas space heating boilers and realize in the year 2006 emission savings at negative costs of -116 EUR/t CO2.

108 Technological learning of condensing gas boilers in the Netherlands

2 75

/t CO /t 50 2006

-25 emissions in EUR 2 -50

-75

Condensing gas space heating boilers -100 Condensing gas combi boilers Costs forCosts avoidedCO 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year Figure 5.10: Costs for CO2 emission savings generated by condensing gas boilers in the Netherlands; excluding direct subsidies and taxes on natural gas as well as on purchase, installation, and maintenance of gas boilers

5.5 Discussion In the first part of this section, we discuss the strengths and weaknesses of our methodology. Afterwards, we discuss the implications of our results for energy and CO2 emission mitigation policy.

5.5.1 Discussion of methodology Both experience curve analysis and cost-benefit calculations are subject to uncertainty. We begin with a qualitative discussion of uncertainties and caveats related to our experience curve analysis.

Our approach of using cumulative Dutch boiler sales as an indicator for cumulative experience in condensing gas boiler manufacturing is justified because condensing gas boilers were developed and produced in the Netherlands without considerable exogenous technology spillover until the early 1990s (Remeha, 2007; Nefit, 2008). However, in subsequent years, condensing gas boilers became more frequently traded. The mergers of boiler producers additionally contributed to international knowledge transfer. One might argue that using data on Dutch cumulative boiler sales introduces a bias into our experience curve analysis. Therefore, we perform a sensitivity analysis of our results, plotting Dutch condensing gas boiler prices as a function of cumulative condensing gas boiler sales in the EU-15. We estimate learning rates of 14 ± 1% and 6 ± 2% for condensing gas combi and space heating boilers, respectively. The results of our

109 Chapter 5 sensitivity analysis are within the uncertainty margins of our estimates, thus indicating that the uncertainties related to the use of data on cumulative Dutch boiler sales are small.

The use of market prices as a proxy for actual production potentially introduces uncertainty into our results, if profit margins of producers vary considerably. We regard this source of uncertainty as minor because the Dutch boiler market has been highly competitive, leaving only relatively small but nevertheless declining profit margins for producers (Nefit, 2008). Furthermore, quality and availability of price data are other sources of uncertainty. We analyzed the sensitivity of our experience curve results with respect to the inclusion and exclusion of the various sources of price data (see Section 5.3.1). We find that the differences in the learning rates are small, i.e., within the uncertainty range of our results.

In more general, uncertainty might be introduced into our results by substantial and sudden changes in the price of production factors used for the manufacturing of gas boilers, i.e., the price of capital, labor, energy, and materials. To understand this source of uncertainty, we reiterate the basic assumption of the experience curve approach: production costs decline at a constant rate with each doubling of cumulative production due to technological learning in manufacturing. One might argue that technological learning causes foremost a reduction in the quantity of production factors used for manufacturing but not a reduction in the price of production factors. Changes in the price of production factors are generally triggered from outside of the learning system and might lead to a temporary or permanent change in the costs of manufacturing. A prominent example in the case of condensing gas boiler manufacturing is the reduction of labor costs at the end of the 1990 by competitive outsourcing of component manufacturing to China. The realized cost reduction can be largely attributed to low wages in China, i.e., a decline in the price for the production factor labor, but not to a decline in the quantity of labor needed to manufacture condensing gas boilers. Similarly, a substantial increase in oil and energy prices as observed in 2007 might have a substantial adverse effect on manufacturing costs, despite continuous technological learning of manufacturers17.

Furthermore, meaningful experience curve analysis requires that the technology studied remains homogenous with regard to its components and the consumer services provided. Condensing gas boilers generally meet this requirement. However, the switch from open to closed non-condensing gas boilers in the early 1990s lead to a temporary price increase, and thus introduces uncertainty into the learning rates identified for the additional price of condensing gas boilers relative to non-condensing devices. The temporary price increase for non-condensing gas boilers between 1988 and 1994 (see

17 One might argue that technological learning is more complex than discussed here: manufacturers not only aim at minimizing production costs by reducing the quantities of production factors but also by substituting production factors (e.g., energy for labor), if substantial price changes occur. This is indeed the case in reality where factor substitution is a main driver behind cost decline in manufacturing. Regardless, it is unlikely that prices of production factors always follow an experience curve pattern because these depend not only on technological learning in other sectors (e.g., growing experience in energy production, in the manufacturing of raw materials and semi-finished components, as well as in information technology) but also on additional factors such as resource scarcity, availability of skilled labor, or the dynamics of profit margins.

110 Technological learning of condensing gas boilers in the Netherlands

Figure 5.2) suggests that the applicability of the experience curve approach to this boiler type is limited, given the level of detail of available price data. This finding also indicates that the time period chosen for experience curve analysis strongly influences the learning rate in cases where technologies undergo substantial changes during their life cycle (see data comparison with Martinus et al. (2005) in Section 5.4.1). Despite the sources of uncertainty discussed so far, we regard the results of our experience curve analysis as valid.

We now discuss uncertainties of our cost-benefit analysis. For our calculations, we use average gas boiler prices, efficiencies, as well as natural gas consumption in centrally heated dwellings of the Netherlands. We exclude a detailed sensitivity analysis for these parameters, because we primarily analyze trends regarding costs and benefits of condensing gas boilers. The level of detail of our analysis is, therefore, insufficient to reflect the situation in individual cases, e.g., for households with natural gas consumption below or above average and for cases where specific boiler prices deviate considerably from the averages used here. These limitations can, however, be addressed to some extent by more detailed analysis with the Dutch residential energy model (DREM) (Dittmar et al., 2007).

Moreover, the results of our cost-benefit analysis reflect the situation of the year in which the investment decision is made. We thereby assume that conditions (e.g., natural gas consumption, natural gas price) remain unchanged during the life time of condensing gas boilers. We chose this approach because it probably best reflects the perceived consumer costs and benefits at the time of purchase. However, given the substantial increase in natural gas prices in recent years, our results potentially underestimate the real consumer benefits of condensing gas boilers. Uncertainties of our cost-benefit analysis also refer to data quality. We estimate natural gas consumption for hot tap water production based on (EnergieNed, 1981-2006) and (EnergieNed, 2006). Data for years prior to 1996 are estimated based on data extrapolation and are, therefore, uncertain.

5.5.2 Discussion of results The results of our analysis indicate a trend towards declining prices and rising benefits for consumers related to condensing gas boilers in the Netherlands. The identified learning rates of 14 ± 1% and 16 ± 8% for the absolute and additional price of condensing gas combi boilers are in line with the findings of Weiss et al. (2009d), who identified in an overview study learning rates of on average 18 ± 9% for energy demand technologies.

Our research identifies (i) economies of scale, (ii) increased specialization and automation of production processes, and (iii) outsourcing of production to low-wage regions as main drivers for price reductions of condensing gas boilers (Gasterra, 2007; Remeha, 2007; Nefit, 2008). It is not possible for us to quantify the contributions of these individual factors to the observed overall price decline. Such analysis would require far more disaggregated cost data that are generally not available due to confidentiality reasons.

111 Chapter 5

The cost-benefit analysis shows that purchasing of a condensing gas boiler was not profitable for consumers in several years after market introduction of this technology. Condensing gas boilers only gained market shares slowly, requiring more than 10 years to achieve a breakthrough in the market (Figure 5.1). However, our cost-benefit analysis also demonstrates that the cost-effectiveness of condensing gas boilers in saving both non- renewable energy resources and CO2 emissions increased rapidly after 1999. Our results indicate that technological learning (i.e., declining prices for purchase and installation as well as improvements in the efficiency of condensing gas boilers) is indeed a main driver for this development. However, especially in the period between 1999 and 2006, also rising natural gas prices greatly improved the cost-effectiveness of condensing gas combi boilers (Figure 5.9).

Our analysis also provides a rationale for establishing subsidy schemes for novel but not yet cost-effective technologies (e.g., heat pumps, innovative lighting technologies, and renewable energy supply technologies). At the example of condensing gas boilers, we have shown that subsidies can indeed substantially improve the cost-performance of novel and efficient energy technologies in early years when technology costs are high and market volumes small. Depending on the dynamics of market diffusion and energy prices, subsidies might be, however, necessary for long time periods of a decade or more.

The Dutch experience with condensing gas boilers in the early years after market introduction (i.e., 1981-1985) also indicates that cost-effectiveness is not a guarantee for market success, if (i) experience with the new technology is limited and (ii) the technology does not comply with legal regulations, household infrastructure characteristics, as well as non-cost-related consumer preferences (Brezet, 1994).

5.6 Conclusions Experience curve analyses for efficient energy demand technologies are still scarce to date. Here, we address this knowledge gap by applying the experience curve approach along with cost-benefit analysis to condensing gas boilers in the Netherlands. We regard the experience curve approach as applicable and useful for analyzing price-dynamics of condensing gas boilers. For the past two decades, we identify a trend towards declining prices as well as rising consumer and governmental benefits. The dynamics of net present value and simple payback time are driven by (i) technological learning, (ii) external effects, and (iii) governmental policies. Technological learning (i.e., declining prices for purchase and installation as well as improvements in the efficiency of condensing gas boilers) explains two thirds of the observed dynamics, whereas the two latter factors together explain one third. Our results highlight the importance of both technological learning and non-technology-related factors such as energy prices for realizing cost- effective emission savings. The example of condensing gas boilers in the Netherlands shows how product innovation can improve energy efficiency in the residential sector. Limited subsidy support of roughly 70 ± 10 million EUR contributed to savings of 270 PJ primary energy and 15 Mt CO2 emissions in the period from 1981 to 2006. Condensing

112 Technological learning of condensing gas boilers in the Netherlands gas boilers in the Netherlands, however, also show that energy policy aiming at improving energy efficiency might need perseverance over several years to decades.

We conclude that our analysis provides an important component of ex-post technology analysis, which adds valuable insight into market diffusion as well as price- and cost-dynamics of one efficient energy demand technology. Our research thereby assists policy makers in designing effective governmental support for other novel and efficient energy demand technologies as well.

Acknowledgements This research was funded by the Dutch Ministry of Economic Affairs. We thank Klaas-Jan Koops from the Dutch Ministry of Economic Affairs for the fruitful cooperation during this research project. We thank Alexandra Newman (Colorado School of Mines, Golden, USA) for her valuable comments on an earlier draft of this chapter.

113 114

Chapter 5 Appendix Table A5.1: Background data for our cost-benefit analysis of condensing gas boilers (data sources: CBS, 2007; Consumentenbond, 1983-2006; EnergieNed, 1981-2006; Visser, 2007) Natural gas Natural gas Average Improved-efficiency non- Condensing Nominal Condensing gas boilers consumption consumption efficiency of gas condensing gas boiler gas boiler natural Year for space for hot tap boilers installed sales gas price Average Market share Average Market heating in ICH water in ICH in ICH dwellings 3 [1000 units] 3 [EUR/m ] efficiency[%] [%] Efficiency [%] share [%] dwellings [m3] dwellings [m ] [m3] 1981 13 3110 280 75 0.19 101 7 88 1982 15 2980 270 76 0.22 101 7 88 10 1983 16 2850 260 76 0.23 102 9 88 21 1984 19 2720 245 76 0.25 102 10 88 29 1985 20 2590 245 77 0.26 102 9 89 32 1986 21 2460 260 77 0.25 102 9 89 37 1987 23 2330 330 78 0.18 103 8 89 42 1988 29 2200 335 79 0.18 103 11 89 45 1989 34 2070 355 79 0.17 103 12 89 50 1990 47 1940 360 80 0.20 103 16 89 55 1991 94 1810 360 81 0.22 104 30 90 56 1992 127 1680 360 83 0.21 104 38 90 57 1993 135 1640 370 84 0.20 104 41 90 55 1994 145 1720 370 86 0.21 105 40 91 52 1995 160 1690 370 87 0.21 105 44 91 50 1996 190 1785 375 89 0.28 106 51 91 45 1997 233 1640 380 91 0.25 107 58 92 37 1998 267 1545 375 92 0.25 107 68 92 29 1999 280 1565 375 94 0.24 107 75 92 24 2000 307 1580 375 95 0.31 107 81 92 18 2001 322 1614 375 96 0.38 107 83 92 17 2002 323 1559 375 97 0.40 107 86 92 13 2003 340 1514 377 99 0.43 107 86 92 13 2004 358 1494 380 99 0.44 107 87 92 12 2005 384 1432 380 100 0.50 107 91 92 8 2006 416 1432 380 101 0.55 107 91 92 7

Table A5.2: Background data for our cost-benefit analysis for condensing gas boilers (data source: Consumentenbond, 1983-2006; Warmteservice, 2007; Visser, 2007) Nominal average boiler price [EUR]a Nominal Improved- Nominal Nominal subsidy Nominal Improved- installation Condensing efficiency subsidy per per improved- installation Condensing efficiency non- costs per Sales Consumer price Year gas space non- condensing efficiency non- costs per non- gas combi condensing gas condensing tax [%] index [%] heating condensing gas boiler condensing gas condensing gas boilers space heating gas boiler boilers gas combi [EUR] boiler [EUR] boiler [EUR] boilers [EUR] Technological learningofcondensinggasboilersintheNetherlands boilers 1981 - 1224 - 794 113 - 343 250 18.0 43 1982 - 1230 - 800 113 - 343 255 18.0 50 1983 - 1236 - 806 113 - 343 265 18.0 52 1984 - 1245 - 831 113 - 353 275 18.0 54 1985 - 1255 - 855 154b - 353 285 18.0 56 1986 - 1265 - 880 151b - 363 290 18.0 52 1987 - 1275 - 905 175c - 363 295 18.0 41 1988 1737 1285 1294 930 - - 372 309 18.0 40 1989 1696 1300 1417 987 - - 372 309 18.0 38 1990 1655 1316 1539 1044 159 - 379 327 18.0 43 1991 1628 1325 1475 1042 159 91 393 340 18.0 46 1992 1602 1334 1411 1041 159 91 393 340 17.5 45 1993 1575 1343 1347 1039 - - 417 372 17.5 42 1994 1548 1352 1283 1036 91 - 417 372 17.5 44 1995 1579 1337 1244 1042 91 - 417 386 17.5 44 1996 1610 1322 1205 1049 91 - 424 397 17.5 47 1997 1641 1308 1166 1056 - - 424 420 17.5 51 1998 1672 1316 1127 1037 - - 488 465 17.5 52 1999 1602 1321 1124 1013 45 - 545 511 17.5 52 2000 1532 1327 1121 988 45 - 556 556 17.5 59 2001 1463 1332 1115 963 45 - 635 635 19.0 68 2002 1575 1428 1177 1027 45 - 630 630 19.0 72 2003 1522 1431 1192 1029 - - 635 635 19.0 77 2004 1470 1434 1207 1030 - - 655 655 19.0 80 115 2005 1548 1437 1223 1032 - - 655 655 19.0 91 2006 1520 1440 1238 1034 - - 680 680 19.0 100 a b, c Assuming average boiler capacity of 25 kWth, including sales taxes Subsidy covers 33% (b) and 40% (c) of additional costs for purchase and installation

Chapter 5

Table A5.3: Assumptions for our cost-benefit analysis of condensing gas boilers Electricity savings of condensing gas boilers in kWh/a (Vaillant, 2007)a 0 Difference in nominal installation costs between condensing and non-condensing gas 20 boilers in EURb Interest rate in % 7 Life time of boilers in years 15 a Based on information from Vaillant (2007), we assume average power requirements of 140 Wel for both condensing and non-condensing gas boilers. This assumption contrasts estimates from Consumentenbond (1983-2006) according to which condensing gas boiler save 200-300 kWhel per year compared to non- condensing gas boilers. We explain the discrepancy with the fact that Consumentenbond (1983-2006) compares condensing gas boilers with outdated non-condensing gas boilers of the existing boiler stock but not with most recent non-condensing gas boilers competing in the market with condensing devices for purchase. b including sales tax

116

6 Analyzing price and efficiency dynamics of large appliances with the experience curve approach

Martin Weiss, Martin K. Patel, Martin Junginger, and Kornelis Blok Accepted for publication in: Energy Policy

Abstract Large appliances are major power consumers in households of industrialized countries. Although the energy efficiency of large appliances has been increasing considerably in past decades, substantial additional energy efficiency potentials still exist. Energy policy that aims at realizing these potentials faces, however, growing concerns about possible adverse effects on commodity prices. Here, we address these concerns by analyzing long- term price and energy efficiency trends of three large wet appliances (washing machines, laundry dryers, and dishwashers) and two large cold appliances (refrigerators and freezers) with the experience curve approach. We identify a robust long-term decline in both the specific price and specific energy consumption of large appliances. Specific prices of wet appliances decline at average learning rates (LR) of 29 ± 8% and thereby much faster than those of cold appliances (LR of 9 ± 4%). Our results demonstrate that technological learning leads to substantial price decline and thus indicate that the introduction of novel and initially expensive energy efficiency technologies does not necessarily imply adverse price effects in the long term. By extending the conventional experience curve approach, we find a steady decline in the specific energy consumption of wet appliances (LR of 18 ± 3% to 35 ± 3%) and cold appliances (LR of 13 ± 3% to 17 ± 2%). Our analysis suggests that energy policy might be able to bend down the slope of energy experience curves.

117 Chapter 6

6.1 Introduction Appliances currently consume 6% (6 EJ, exajoules) of the economy-wide final energy supply in IEA-19 countries1 (IEA, 2008c). After space heating and cooling, appliances are the second largest energy function in households, accounting for 21% of household final energy demand. The energy consumption of appliances continues to grow rapidly, albeit with differentiation for large and small appliances. Large appliances (also referred to as large household appliances or white goods) such as washing machines and refrigerators account for half of the appliance-related final energy demand (IEA, 2008c). Their share is, however, falling for two reasons. Firstly, ownership of small appliances (e.g., juicers, cellular phones, audio and video devices) shows an over-proportional increase in recent years. Secondly, the specific energy consumption of large household appliances has been falling considerably in the past 30 years, partially due to effective energy efficiency policies (Bertoldi and Atanasiu, 2007; Ellis et al., 2007; IEA, 2008c,d,e; Dale et al., 2009).

Despite this development, still substantial and untapped potentials exist to further increase the energy efficiency of large appliances. Policy initiatives that aim at realizing these potentials face, however, growing concerns of manufacturers, consumers, and policy makers about possible adverse effects on the price of large appliances (EU, 2008a). These concerns were previously fuelled by ex-ante engineering analysis that suggested a direct relationship between improved energy efficiency on the one hand and rising commodity prices on the other (Greening et al., 1996; Ellis et al., 2007; Dale et al., 2009).

In reality, however, both prices and energy consumption of large appliances have been falling simultaneously for several decades (Schiellerup, 2002; Ellis et al., 2007; Bertoldi and Atanasiu, 2007; Dale et al., 2009). Hence, conventional ex-ante engineering analysis fails to provide reliable price and cost projections because it disregards potentials for cost reductions by assuming constant additional costs of energy efficiency improvements. This assumption neglects that efficiency measures are introduced as superposition in a dynamic rather than a static product system. The entire product system (including the newly implemented energy efficiency technologies) continuously undergoes technological changes and offers substantial potentials for price and cost decline due to technological learning (e.g., growing experience of manufacturers, economies of scale, and technological innovation).

In this chapter, we aim at supplementing ex-ante engineering analysis by studying price and efficiency trends of large appliances over long time periods.In particular, we want to obtain a better understanding of the extent to which technological learning influences both price and energy efficiency of large appliances. One tool that allows for quantifying technological learning from this perspective is the experience curve approach2. Typically, the experience curve approach models production costs of a

1 We refer here to the 19 member countries of the IEA (International Energy Agency), i.e., Austria, Belgium, Canada, Denmark, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, The Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, and the USA. 2 The experience curve approach has so far mainly been used to analyze price and cost trends in manufacturing (see, e.g., Argote and Epple (1990) and Junginger et al. (2008)). Exceptions refer to its application for analyzing the specific energy consumption of ammonia and urea production

118 Price and efficiency dynamics of large appliances technology as a power-law function of cumulative production (BCG, 1972). The experience curve approach gained importance as management tool in manufacturing industries (Argote and Epple, 1990) and as instrument for technology forecasting in energy and CO2 emission scenarios (e.g., IEA, 2000; Wene et al., 2000; IEA, 2008c; van Vuuren et al., 2006).

The experience curve approach has been extensively applied to and redefined for renewable energy supply technologies (e.g., Neij, 1999; McDonald and Schrattenholzer, 2001; Junginger et al., 2004, 2005, 2006, 2008; Neij, 2008). Its application to energy demand technologies and in particular to large appliances is, however, still scarce. Bass (1980) analyzed technological learning of refrigerators (period from 1922 to 1940), dishwashers (periods from 1947 to 1960 and 1947 to 1974), and laundry dryers (periods from 1950 to 1961 and 1950 to 1974) based on appliance sales and price data for the USA. Laitner and Sanstad (2004) quantify technological learning of washing machines, dishwashers, laundry dryers, refrigerators, and freezers for the period between 1980 and 1998 based on production and price data for the USA. These studies cover either past time periods more than three decades ago (Bass, 1980) or only relatively short time series (Laitner and Sanstad, 2004).

Here, we apply the experience curve approach to three wet appliances (i.e., washing machines, laundry dryers, dishwashers) and two cold appliances (i.e., refrigerators, and freezers). With our analysis, we cover very long time periods of three to four decades until most recent years. This assures as far as possible that our results correctly represent the actual price and efficiency dynamics of large appliances. First, we identify the rate at which prices, i.e., consumer investment costs for large appliances decline. To provide a more comprehensive picture of technological learning, we extend the conventional experience curve approach by analyzing also the dynamics of specific energy consumption of large appliances as function of cumulative production. Such a methodological extension is new and allows for quantifying technological learning of large appliances from a broader perspective. We justify our methodological extension by two considerations: (i) At the system level, energy efficiency improvements might follow autonomous technological innovation and often result from the quest of producers to decrease production costs. For example, improved wall insulation of cold appliances might allow for smaller, thus cheaper compressors; innovative rubber door gaskets might be cheaper, while having longer life-times and offering better thermal insulation. By means of such inter-linkages, energy efficiency improvements and declining production costs may go hand in hand, thereby contributing to increasing energy efficiency alongside technological learning that is directed at improved and cheaper appliances.

(Ramirez and Worrell, 2006) as well as ethanol production from corn (Hettinga et al., 2009). Our extension of the experience curve approach to analyze the specific energy consumption of large appliances is hence new and will be justified below.

119 Chapter 6

(ii) Over the past decade, energy efficiency became a product feature that is decisive for the market success of large appliances. Thus, producers are nowadays forced to increase the efficiency of their products as well as to decrease production costs, if they want to remain competitive on the market (Ecowet, 2007a).

Following the introduction, we explain in Section 6.2 methodology and data sources used for constructing cost and energy experience curves. We present the results of our analysis in Section 6.3. In Section 6.4, we discuss our findings, thereby paying special attention to the discussion of a conceptual framework for devising energy experience curves. We draw conclusions in Section 6.5.

6.2 Methodology and data sources In this section, we first present our methodology for constructing experience curves. Afterward, we explain in detail data sources used and data adaptations made to construct price and energy experience curves for large appliances.

6.2.1 The experience curve approach With the experience curve approach, we trace price and efficiency dynamics of five large household appliances, i.e., washing machines, laundry dryers, dishwashers, refrigerators, and freezers by modelling specific price and specific energy consumption as a power-law function of cumulative production:

= ⋅ bi Ccumi C0,i (Pcumi ) (6.1a)

= ⋅ bi Ecumi E0,i (Pcumi ) (6.1b) where Ccumi [EUR2006 (real Euros deflated to the base year of 2006) per functional unit] represents the specific price and Ecumi [kWhel (kilowatt hour electricity) per functional 3,4 unit, EEI (Energy Efficiency Index)] the specific energy consumption at Pcumi. C0,i [EUR2006 per functional unit] stands for the specific price and E0,i [kWhel (kilowatt hour electricity) per functional unit, EEI] for the specific energy consumption of the first unit produced. Pcumi represents cumulative experience (i.e., the cumulative worldwide production), and bi the product-specific experience index of the large household appliance i5. By applying the logarithmic function to Equation 6.1, we plot a linear experience curve

3 We analyze price and efficiency dynamics of wet appliances based on the following functional units: kg (kilogramme) laundry capacity for washing machines and laundry dryers, as well as standard place setting for dishwashers. In the case of cold appliances, we analyze price dynamics based on 100 l (liters) of volume and efficiency dynamics based on the dimensionless EEI (see Section 6.2.2). 4 Dishwashers typically have a capacity of 12 standard place settings. One standard place setting consists of a dinner plate, a soup plate, a dessert plate, a glass tumbler, a tea cup and saucer, as well as a set of knife, fork, soup spoon, dessert spoon, and teaspoon. 5 Energy experience curves should generally include a constant term that accounts for the thermodynamic minimum energy requirements of technologies and processes (e.g., heat effects of chemical reactions). For the five large household appliances, the thermodynamic minimum energy requirements are virtually zero. One exception from this is the first-time cooling of food products. We, however, neglect energy

120 Price and efficiency dynamics of large appliances

with bi as slope parameter and log C0,i as intercept with the y-axis. Based on the experience index bi, we calculate appliance-specific learning rates (LRi) [%] and progress ratios (PRi) [%] as rates, at which both, specific prices and specific energy consumption decline with each doubling of cumulative production:

= − = − bi LRi 1 PRi 1 2 (6.2)

We estimate the error interval of LRi and PRi based on the implicit error of the regression analysis, i.e., the 95% confidence interval for the slope parameter of the experience curve. We devise experience curves based on the average specific price and energy consumption of large appliances in individual years.

6.2.2 Data sources and data adaptations To construct experience curves for large household appliances, we use price and energy consumption data that refer to the Netherlands. This choice is justified by two reasons: (i) Data availability for this particular country allows for analyzing long and consistent time series (i.e., three to four decades). (ii) Despite considerable price variation between countries (GfK, 2003, 2004; Ecocold, 2007b), price and efficiency trends are generally similar throughout the world (Ellis et al., 2007).

As single source for data on specific price and energy consumption, we use the Dutch consumer organization Consumentenbond. Typically, Consumentenbond (1964- 2008), provides data on prices and energy consumption along with other supplementary product information when publishing product evaluations. Consumentenbond (1964-2008) tests large appliances frequently, i.e., depending on the type of appliance in intervals between twice per year to once in two years. We generally regard the data as presented by Consumentenbond (1964-2008) as reliable proxy for the actual price and energy consumption of large appliances produced in respective years. However, communication with Consumentenbond (2009) revealed that in some cases, data might not be representative because individual tests can focus on either cheap or expensive appliances, which are most successful at the Dutch market, or on appliances of a specific energy label category. Prior to our experience curve analysis, we therefore corrected parts of the bias in the datasets as provided by Consumentenbond (1964-2008). We are now going to explain our data adaptation in more detail. We will revert to the problem of data bias and the related uncertainties in our discussion (see Section 6.4.1).

requirements for this service because they are in reality negligible compared to the actual energy consumption of cold appliances.

121 Chapter 6

To assure consistency of price data, we first exclude built-in models of dishwashers, refrigerators, and freezers from our analysis. Built-in appliances have a considerably higher specific price than standard models. Without corrections, differences regarding the frequency of data for these models in our datasets would introduce substantial bias into our experience curve analysis.

Secondly, we deduct sales tax from the price data and we deflate prices to the base year of 2006 by using consumer price indices for the Netherlands as obtained from CBS (2007). In third instance, we use data on absolute prices and supplementary product information to calculate specific prices expressed in EUR2006/kg laundry capacity for washing machines and laundry dryers, EUR2006/standard place setting for dishwashers, and EUR2006/100 l for refrigerators and freezers.

In our price experience curve analysis for freezers, we differentiate between upright and chest freezers because both freezer types show considerable and systematic differences regarding their specific price (Consumentenbond, 1964-2008; Waide, 2001). Under the category of refrigerators, we uniformly include data for refrigerators without freezer compartments, refrigerator with freezer compartments, as well as two-door refrigerator-freezer combinations. We do not apply price corrections here because data obtained from Consumentenbond (1964-2008) indicate no systematic price differences between these three types of cold appliances.

Next to adaptation of price data, we also make adjustments for data on specific energy consumption as given by Consumentenbond (1964-2008) for the various types of appliances. In this way, we assure consistent reproduction of the historic development in the specific energy consumption of large appliances. In the case of washing machines, we correct for temperature differences of test cycles. We consistently refer the specific energy o consumption of washing machines to an average 60 C (degrees centigrade) cotton washing cycle. We recalculate the energy consumption as given by Consumentenbond (1964-2008) for 90 oC washing cycles by making three assumptions: (i) 90% of the energy use during a 60 oC cotton washing cycle is consumed for water heating and 10% for laundry spinning and water pumping, (ii) the inlet water temperature of washing machines is 15 oC, and (iii) the energy consumption for water heating is directly proportional to the difference between inlet temperature and washing temperature (i.e., 75 oC in the case of a 90 oC washing cycle and 45 oC in the case of a 60 oC washing cycle). These assumptions yield a correction factor of 62.5% for converting the energy consumption of a 90 oC cotton washing cycle into the energy consumption of a 60 oC cotton washing cycle. Our correction factor is in line with estimates presented by GEA (1995) and Ecowet (2007a).

Accounting for the specific energy consumption of laundry dryers is complicated because data availability is limited. Consumentenbond (1964-2008) states with a few exceptions (e.g., the years 1990, 1988, 1984) only energy ratings or energy labelling categories. We therefore estimate the actual energy consumption of laundry dryers based on supplementary information provided by producers and retailers. For both washing machines and laundry dryers, we calculate the specific energy consumption [kWhel/kg] by

122 Price and efficiency dynamics of large appliances

dividing absolute energy consumption [kWhel] (as given by Consumentenbond (1964- 2008)) by actual laundry capacity [kg].

In the case of dishwashers, refrigerators, and freezers, we express energy consumption as energy efficiency index (EEI). For calculating the EEI, we follow the official methodology used for determining the efficiency category of these appliances within the European energy labelling scheme (EU, 1994, 1997, 2003b)6. The energy consumption given by Consumentenbond (1964-2008) for refrigerators and freezers refers in individual years to ambient air temperatures of 18, 20, and 25 oC (degrees centigrade). In a first step, we uniformly recalculate energy consumption for an ambient air temperature of 25 oC by assuming that energy consumption at 18 oC and 20 oC is 35% and 25% lower than at 25 oC 7. Afterward, we recalculate the corrected energy consumption data into an EEI. This way, we assure consistency with the standard methodology that is used to evaluate the energy efficiency of dishwashers and cold appliances within the European energy labelling scheme (EU, 1997, 2003b). Based on the adapted data, we calculate averages of specific prices and specific energy consumption as well as the related standard deviations for individual years.

We estimate cumulative experience in the manufacturing of large appliances based on global production data. This choice is justified because for more than two decades, major producers have been simultaneously operating on each of the three major global appliance markets, i.e., Europe, America, and Asia (Ecocold, 2007c; Dahlman, 2007). Although producers adjust their products for specific consumer preferences on individual markets, it is plausible to assume a global learning system for the manufacturing of large appliances and components thereof. We estimate cumulative production such that it represents the sum of production since the appliance under consideration was introduced to the market based on data from UN (2008), Eurostat (2008), and Ecocold (2007a). The data provided by these sources do, however, not allow for constructing complete time series that cover global yearly production of appliances, starting from the point of market commercialisation almost a century ago up until most recent years. We therefore supplement available data with information provided by various sources, including Bowden and Offer (1994), Waide (2001), Laitner and Sanstad (2004), AM (2007), the statistical offices of Canada, China, Germany, Japan, and the USA, as well as the manufacturers association AHAM (2004a, b). We apply data interpolation and extrapolation to close remaining gaps in our datasets. The availability of production data for dishwashers and freezers is particularly limited. We estimate cumulative global freezer production based on three sources of information: (i) sales data for the USA, (ii) the fraction of refrigerator to freezer production in the EU between 1995 and 2005, and (iii) the fraction of refrigerator to freezer production and sales as provided by Ecocold (2007a)

6 For refrigerators and freezers, we base our calculation of the EEI consistently on the usable volume of cold appliances as determined through standardized product testing by Consumentenbond (1964-2008). These volumes are in general smaller than the ones stated by appliance manufacturers. The EEIs presented in our experience curve analyses are hence higher than the ones used for energy labeling. 7 This approach is in line with the method used for temperature corrections by EU (2003b). We account here only for the impact of the temperature difference between interior and ambient air on the overall energy consumption of refrigerators, thereby neglecting the effect of temperature differences on the coefficient of performance of the refrigerator’s heat pump.

123 Chapter 6 for several non-European countries. We estimate cumulative global dishwasher production based on Destatis (1990-2003a), Laitner and Sanstad (2004), and CECED (2007). We conduct a sensitivity analysis for data on dishwasher and freezer production to identify the uncertainties of our approach. Based on data availability, we cover the following time periods by our experience curve analysis: (i) 1965-2008 for washing machines, (ii) 1969- 2003 for laundry dryers, (iii) 1968-2007 for dishwashers, (iv) 1964-2008 for refrigerators, and (v) 1970-2003 for freezers.

6.3 Results We first provide an overview of average yearly changes in the specific price and energy consumption of large household appliances (Table 6.1). We distinguish between (i) the whole time period for which data are available to us and (ii) a shorter time period since 1990, suitable to identify more recent price and efficiency trends.

Table 6.1: Average yearly change in the specific price and energy consumption of large appliances in the Netherlands (data source: Consumentenbond, 1964-2008) Average yearly change in Average yearly change in Appliance specific pricea [%] specific energy consumption [ %] entire period truncated periodb entire period truncated periodb Washing machines -2.4 -2.9 -2.5 -2.4 Laundry dryers -2.1 -2.3 -1.5 -2.5 Dishwashers -3.8 -3.3 -2.3 -2.3 Refrigerators -1.2 -1.3 -2.3 -3.0 Upright freezers -1.5 0.0 -1.9 -1.6 Chest freezers -1.1 -0.9 a prices in real terms b covering the years 1990 and afterward

Both wet appliances (washing machines, laundry dryers, dishwashers) and cold appliances (refrigerators and freezers) show a trend towards a decline in specific price and specific energy consumption. Considering first the long time periods, we find specific energy consumption of wet and cold appliances to decline at similar rates of roughly 2% per year. However, specific prices of wet appliances decline at 2-4% per year and thereby considerably faster than the ones of cold appliances. Our results for short time periods confirm these findings (Table 6.1). The decline in specific energy consumption of washing machines and dishwashers is in both time periods accompanied by an over-proportional decline in water consumption of 2-6% per year (data not shown in Table 6.1).

Using the same data as above, we now construct two sets of experiences curves, i.e., for the specific prices and for the specific energy consumption of wet and cold appliances (Figures 6.1 and 6.2). In line with the results presented in Table 6.1, our experience curve analysis indicates a clear trend towards a decline in both specific prices and specific energy consumption with increasing cumulative production of large household appliances. Wet appliances show relatively high price learning rates of 33 ± 9% for washing machines, 28 ± 7% for laundry dryers, and 27 ± 7% for dishwashers.

124 Price and efficiency dynamics of large appliances

By contrast, cold appliances show price learning rates, which are a factor three lower, i.e., 9 ± 4% for refrigerators as well as 10 ± 5% and 8 ± 2% for upright and chest freezers.

600 Energy consumption 0.6 100 Water consumption Specific water consumption 400 80 0.4 in l

300 0.3 60 /kg laundry capacity 0.2 200 40 R2 = 0.92 150 LR = 35 ± 3% 30 0.1 PR = 65 ± 3% /kg laundry capacity /kg laundry 0.08 20

100 capacity laundry /kg Specific price price Specific el 0.06 2 2006 80 Washing machines R = 0.57 15 Time period: 1965-2008 0.04 LR = 28 ± 8% 60 R2 = 0.56 PR = 72 ± 8% in kWh 0.03 Washing machines 10 Specific energy consumption consumption energy Specific in EUR LR = 33 ± 9%; PR = 67 ± 9% Time period: 1965-2008 8 40 0.02 300 400 500 600 8001000 1500 2000 300 400 500 600 8001000 1500 2000 Cumulative global production in million washing machines Cumulative global production in million washing machines

1.5 200

150 1 100 0.9 90 80 0.8 /kg laundry capacity /kg laundry 70 capacity laundry /kg el Specific price price Specific

2006 0.7 60 Laundry dryers Laundry dryers Time period: 1969-2003 Time period: 1969-2003 50 2 0.6 2 R = 0.80 in kWh Specific energy consumption consumption energy Specific R = 0.84 in EUR in 40 LR = 28 ± 7%; PR = 72 ± 7% LR = 20 ± 6%; PR = 80 ± 6% 0.5 50 60 80100 150 200 250 60 80100 150 200 250 Cumulative global production in million laundry dryers Cumulative global production in million laundry dryers

300 50 Energy consumption

2 Specific water consumption 200 Water consumption 30 in l/standard place setting 20 150 2 1 R = 0.89 LR = 18 ± 3% 10 100 0.7 PR = 82 ± 3% 7 80 0.5 5 60 2

/standard place setting place /standard 50 R = 0.84

Specific price price Specific 0.3 3 LR = 31 ± 5% 2006 40 Dishwashers 2 Time period: 1968-2007 0.2 PR = 69 ± 5%

30 2 index efficiency Energy R = 0.82 Dishwashers 1 in EUR LR = 27 ± 7%; PR = 73 ± 7% Time period: 1968-2007 20 0.1 0.7 20 30 40 60100 150 200 300 400 600 20 30 40 60100 150 200 300 400 600 Cumulative global production in million dishwashers Cumulative global production in million dishwashers Figure 6.1: Experience curves for specific prices (left column) and specific energy consumption (right column) of wet appliances; error bars indicate the standard deviation of data; error intervals represent the 95% confidence intervals of learning rates and progress ratios based on the implicit error of the regression analysis8

8 For washing machines and dishwashers, we also include water consumption in our experience curve analysis. This choice is justified because water consumption is (next to energy consumption) the second important parameter determining environmental impacts and use phase costs during the life cycle of wet appliances.

125 Chapter 6

300 500

400 200

/100 volume l 300 150 2006

100 200 80 Refrigerators Refrigerators

150 index efficiency Energy Time period: 1964-2008 60 Time period: 1964-2008 2 2 R = 0.43 50 R = 0.87 LR = 9 ± 4%; PR = 91 ± 4% LR = 17 ± 2%; PR = 83 ± 2%

Specific price in EUR 100 40 40 60100 200 400 600 1000 2000 40 60100 200 400 600 1000 2000 Cumulative global production in million refrigerators Cumulative global production in million refrigerators

800 Upright freezers 700 Upright freezers Chest freezers 150 600 Time period: 1970-2003 R2 = 0.59; LR = 10 ± 5%; PR = 90 ± 5% 500 /100 l /100 volume 400

2006 100 90 300 80 70 200 Refrigerators

Chest freezers efficiencyEnergy index 60 Time period: 1964-2008 2 150 Time period: 1970-1998 R = 0.79 R2 = 0.87; LR = 8 ± 2%; PR = 92 ± 2% 50 LR = 13 ± 3%; PR = 87 ± 3% Specific price EUR in 20 30 40 60 80100 150 200 300 400 600 800 20 30 40 60 80100 150 200 300 400 600 800 Cumulative global production in million freezers Cumulative global production in million freezers Figure 6.2: Experience curves for specific prices (left column) and specific energy consumption (right column) of cold appliances; error bars indicate the standard deviation of data; error intervals represent the 95% confidence intervals of learning rates and progress ratios based on the implicit error of the regression analysis

The learning rates identified for specific energy consumption clearly indicate a trend towards increasing energy efficiency albeit with differentiation for wet and cold appliances. For wet appliances, learning rates range from 35 ± 3% for washing machines to 20 ± 6% for laundry dryers and 18 ± 3% for dishwashers. Cold appliances show slightly lower learning rates of 17 ± 2% and 13 ± 3% for refrigerators and (upright and chest) freezers, respectively.

We now provide some explanations for the decline in specific prices. Afterward, we try to explain improvements in the energy efficiency of large household appliances. The decline in specific prices is to a large extent caused by an overall decline of production costs but also by declining markups in the wholesale and retail sector, represented by the difference between producer prices and consumer prices).

Addressing the first point, production costs for large appliances declined mainly due to the following factors (Ecocold, 2007a; Siderius, 2008; Dale et al., 2009; Kemna, 2009): (i) technological learning and economies of scale in component manufacturing and appliance assembling in past decades, partly realized by mergers of producers

126 Price and efficiency dynamics of large appliances

(ii) increasing substitution of capital for labor, lowering labor requirements and increasing automation and overall productivity of appliance manufacturing in the period of 1970-1990 (iii) standardization and competitive outsourcing of components and sub assemblies production to specialized companies in low-wage regions like China (since the 1990s) (iv) streamlining of assembly lines, decreasing assembly times, just-in-time manufacturing, mass customization, and reduction of on-site stocks of components, semi-finished and finished products in recent years (v) shifting of Western European assembly lines to low-income Eastern European countries such as Hungary, Poland, or Turkey (vi) standardization and homogenization9 of components as well as simplification of product design leading to a decrease in the number of materials and components (vii) technological improvements in other areas of the economy (e.g., information technology, material sciences, mechanical engineering)

During past decades, components of large appliances have been continuously improved (e.g., introduction of direct drive motors in washing machines and variable speed-drive compressors in cold appliances). Furthermore, wet appliances experienced the introduction of additional product functions (e.g., centrifuge drying, large variety of washing programs). This has been only to a minor extent the case for cold appliances (e.g., introduction of automatic defrosting function, ice cube production, or water dis- pensing). Such functionality improvements potentially increase production costs of large household appliances (see discussion in Section 6.4.2).

Our results show that specific prices of wet appliances decline faster than the ones of cold appliances both as a function of time and cumulative production. Changes in product functionality can therefore not explain the relatively high learning rates for wet appliances in comparison to the relatively low learning rates for cold appliances. The differences in the learning rates for wet and cold appliances might, however, be explained by three factors (Siderius, 2008; Kemna, 2009): (i) Wet appliances are typically composed of more individual components than cold appliances. This enables higher cost reduction potentials through outsourcing of component and sub-assembly manufacturing to low wage regions like China. (ii) Due to the large number of components, assembling of wet appliances and their components was relatively labour intensive and time intensive in the 1960s and 1970s. Prices of wet appliances thus profited over-proportionally from increased automation.

9 The number of cold appliance categories (i.e., spanning from simple refrigerators to multi-use cabinets) sold on the market in noticeable quantities decreased from ten to four, with refrigerator-freezer combinations constituting to date market shares of more than 60% in Europe (Ecocold, 2007c). This development was partly facilitated by the EU energy label because producers can achieve stringent energy standards for some categories of cold appliances more easily than for others.

127 Chapter 6

(iii) The EU energy label was updated in 2003 for cold appliances but not for wet appliances. More stringent energy efficiency labeling might have incurred adverse short-term price effects for cold appliances thus leading to a slower price decline in most recent years (see Figure 6.2).

Declining markups in the wholesale and retail sector have been realized by (Ecocold, 2007a, b; Dale et al., 2009): (i) cost reductions due to increases in productivity, economies of scale, decrease of on-site stocks and thus storage costs (ii) increasing market shares of retail chains at the expense of small local retailers10, thus (iii) enhanced market competition that lead to declining profit margins

Increasing competition in recent years supported, however, trends towards market concentration in all price segments, which might induce adverse price effects in the future (Ecowet, 2007a).

We now focus on the decline of specific energy consumption. Technological changes that lead to an increase of energy efficiency in the past include (Ecocold, 2007e; Ecowet, 2007b; Siderius, 2008; Kemna, 2009): (i) improving insulation by increased wall thickness and the introduction new insulation materials (e.g., replacing polystyrene by polyurethane foams), improving compressor technology, increasing the size of condensers, improving in refrigerants, heat exchangers, control electronics, as well as internal temperature distribution by fans leading to an optimization of the cooling system in the case of cold appliances (ii) reducing water consumption (e.g., by introducing the jet-system around 1985 for laundry spraying instead of bathing, centrifuge drying between washing cycles, and improving tub shape in washing machines), internal heat recovery (dishwashers) as well as progress in other areas of manufacturing (i.e., improvement of detergent quality) in the case of wet appliances11 (iii) improving heat exchangers, water vapour condensation, ventilation, and tumbling of laundry in laundry dryers

So far, we have been focusing on the dynamics of specific energy consumption of large appliances in the past three to four decades. However, if we now compare the last data points in Figures 6.1 and 6.2 with data for earlier years, we might observe an accelerated decline of specific energy consumption for refrigerators and partly also for dishwashers and freezers in recent years. This observation might be attributed to the combined effect of in principle three energy policy measures, i.e., the implementation of

10 In Europe, the market share of the five largest retailers rose from 12% in 1990 to 30% in 2005 (Ecocold, 2007a). 11 The relatively high learning rates for specific water consumption (Figure 6.1) indicate an over- proportional decline in the specific consumption of non-heated rinsing water. These water savings were, to some extent, achieved by technological innovations as explained above but also to some extent by simply reducing the number of cold-water rinsing cycles within the entire washing cycle.

128 Price and efficiency dynamics of large appliances the European energy labeling (EU, 1992), the European minimum energy performance standards for cold appliances (EU, 1996), and the Dutch energy premium regulation (SenterNovem, 2000). To obtain a more detailed insight into the dynamics of energy efficiency, we devise two separate sets of energy experience curves: one covering the period before the introduction of Energy policies in the Netherlands and one covering the period afterward (Figure 6.3; Table 6.2)12.

2 300 R = 0.86 LR = 16 ± 2% 250 PR = 84 ± 2% 200 Time period: 1964-1994

150

1995 100

80 R2 = 0.71 60 LR = 49 ± 17% Energy efficiency index (EEI) index efficiency Energy 50 PR = 51 ± 17% 2002 Time period: 1995-2008 2008 40 40 60100 200 400 600 1000 2000 Cumulative global production of refrigerators in million

Decline of EEI in the period of 1964-1994 Hypothetical autonomous decline of EEI without energy policy Decline of EEI including the effects of energy policy

Figure 6.3: Potential effect of energy policy on the decline of EEI of refrigerators; numbers on error bars indicate the year of analysis13

Our results indicate at first sight that learning rates for specific energy consumption of all wet and cold appliances are higher in the period of enforced energy policy than in the period before. However, the error intervals indicate that differences are only significant in the case of dishwashers, refrigerators and freezers. The findings in Figure

12 We use here the year in which energy labeling was introduced in the Netherlands to discriminate the time periods with and without energy policy. We assume that energy labels were introduced in the Netherlands in 1995 for refrigerators and freezers, in 1996 for washing machines and laundry dryers, and in 1999 for dishwashers (Luttmer, 2006). We justify this choice because it probably captures best the effect of policy measures on the energy efficiency of large appliances in the Netherlands. Other relevant policy measures were introduced around the same time or later: the introduction of minimum energy performance standards for cold appliances dates back to 1996 and the introduction of the Dutch energy premium regulation followed in 2000. 13 The relatively low energy efficiency index of refrigerators in the year 2002 suggests bias in the dataset presented by Consumentenbond (1964-2008) for this particular year. Both Consumentenbond (2009) and a comparison of data from Consumentenbond (1964-2008) with results of a market analysis conducted by Ecocold (2007d) indicate that efficient label-A refrigerators are indeed slightly over-represented in the data set of Consumentenbond (1964-2008) for this particular year.

129 Chapter 6

6.3 nevertheless suggest that energy policy was to some extent able to accelerate energy efficiency improvements by bending down the slope of energy experience curves.

Table 6.2: Energy learning rates for large household appliances before and after the introduction of energy policy; in brackets coefficient of determination (R2) Time period prior to Time period after Appliance the introduction of LR in % (R2) the introduction of LR in % (R2) the energy policy energy policy Washing machines 1964-1995 36 ± 5 (0.87) 1996-2008 39 ± 11 (0.79) Laundry dryers 1969-1995 18 ± 10 (0.68) 1996-2003 36 ± 37 (0.66) Dishwashers 1968-1998 15 ± 4 (0.85) 1999-2007 33 ± 12 (0.82) Refrigerators 1964-1994 17 ± 2 (0.86) 1995-2008 49 ± 17 (0.71) Freezers 1970-1994 11 ± 4 (0.72) 1995-2003 47 ± 3 (1.00)

In the case of dishwashers and refrigerators, our data indicate another interesting phenomenon. Between the years 2003 and 2004 (dishwashers, see last three data points in Figure 6.1) as well as between 1999 and 2000 (refrigerators, see Figure 6.3), we find a relatively substantial decrease of specific energy consumption. In the years afterward, however, the decline of specific energy consumption has been far less pronounced. This finding might be explained by a rapid shift of the appliance market towards efficient label- A washing machines, dishwashers, and refrigerators shortly after energy policies (e.g., energy labeling, energy premium regulation) became effective (Luttmer, 2006; Ecocold, 2007b, d). As soon as manufacturers reached compliance with labeling standards, the rate at which additional efficiency improvements were realized seems to have declined again14. Caution is, however, necessary because the number of data points in our analysis is limited. It is therefore too early to draw firm conclusions on these issues.

6.4 Discussion We begin with discussing the strengths and weaknesses of our methodology. Afterward, we discuss justification, opportunities, and limits for the applicability of the experience curve approach to specific energy consumption. In the last part of this section, we compare our findings to results from literature and we focus on implications of our results for effective energy policy.

6.4.1 Discussion of methodology As for any empirical analysis, the reliability of our results depends on the applied methodology and on the quality of underlying empirical data. We regard both as reliable, although they are subject to several uncertainties. The strength of the experience curve approach compared to simple time-series analysis refers to its ability to relate the dynamics of specific prices and specific energy consumption directly to the cumulative

14 In line with this reasoning, we find that the specific energy consumption of refrigerators continues to decline while the specific energy consumption of washing machines and dishwashers remains relatively constant after label-A products reached market saturation. This observation might be attributed to the introduction of additional and more stringent label-A+ and label-A++ categories for refrigerators in 2003 (EU, 2003b).

130 Price and efficiency dynamics of large appliances experience in manufacturing. With our experience curve analysis, we extend existing analyses (i) by covering long time periods until most recent years15 and (ii) by providing uncertainty intervals of empirical data and estimated learning rates. Our analysis thus contributes to a more reliable and transparent quantification of price and efficiency dynamics of large appliances and can thus be used to improve the quality of energy and CO2 (carbon dioxide) emission scenarios.

Uncertainties refer in first instance to the reliability of the experience curve approach for quantifying price and efficiency dynamics of technologies. Price experience curves provide trustworthy results, (i) if prices reliably approximate actual production costs and (ii) if the observed price dynamics are predominantly driven by growing experience in manufacturing, i.e., a decline in the quantities of production factors used for manufacturing.

Addressing the first point, we argue that the approximation of actual production costs by market prices is accepted practice in experience curve analyses because data on actual production costs are generally kept confidential by producers (see, e.g., Junginger et al., 2008). However, such approximation is only valid for competitive markets, where sales prices closely follow production costs (BCG, 1972). This is in general the case for large appliances, for which markets are traditionally characterized by low yet declining profit margins (Ecocold, 2007a)16.

Addressing the second point, the experience curve phenomenon applies strictly speaking only to the cost of value added (Sallenave, 1985); abd more specifically to costs related to changes in the quantity of production factors used for manufacturing. Parts of the observed price dynamics are, however, not attributed to changes in the quantity of production factors used in manufacturing but to changes in the price of production factors. Examples are declining labour costs due to outsourcing of production to low wage regions, or changes in the price of energy and materials. Changes in the price of production factors are often exogenous of the learning system and can hardly be influenced by technological learning of manufacturers. Variability in the price of production factors might hence lead to a singular change of production costs that might not be repeatable or could even be reversed in the future. Uncertainties relate also to changes in the functionality of large appliances. In principle, we do not analyze with the experience curve approach a product or a technology but a service, provided by large household appliances to consumers (i.e., the cleaning of 1kg laundry or the preserving of fresh food for a defined time period). In past decades, especially the functionality of wet appliances has been improving substantially. Adding functionality such as the introduction of centrifuge drying and various washing programs potentially increases in first instance the specific prices of appliances. Correcting for improved functionality would hence very likely lead to even higher learning rates for especially wet appliances than the ones found in Figure 6.1.

15 The variability of learning rates identified for different time periods (e.g., Bass, 1980) indicates that such extensions are important to improve the reliability of product-specific learning rates. 16 Ecocold (2007c) finds price differences of up to 41% for refrigerators sold in Western and Eastern European countries. This observation indicates that pricing policy of producers might indeed present a source of uncertainty for the results of our experience curve analysis.

131 Chapter 6

However, measuring and quantifying increasing functionality as a single parameter is rather difficult. Related uncertainties result from covering by our analysis only parts of the time periods in which large appliances are offered at the market. This approach is necessary due to limited data availability but it inherits the risk of neglecting radical technology innovations before the period of analysis (thus potentially overestimating cumulative production and technology-specific learning rates in the period of analysis (Wene, 2008a)). However, we regard it unlikely that technology innovations occurred in any of the analyzed appliances to such an extent that would make it necessary to reset cumulative production17.

Uncertainties also result from the data used for our experience curve analysis. Consumentenbond (1964-2008) pre-selects large household appliances for their tests (i.e., according to price, efficiency, functionality, or other more subjective criteria). Their datasets are therefore not always representative for the whole Dutch appliance market. We nevertheless argue that using data from Consumentenbond (1964-2008) only introduces a random error into our analysis, which is unlikely to affect the robust average price and efficiency trends identified here for relatively long time periods. The situation is however different for our analysis presented in Table 6.2. Here, pre-selection of data with regard to appliances of specific price or efficiency categories might affect our experience curve results for the relatively short time period after the introduction of energy policies and introduces therefore uncertainty into this part of our analysis. Including additional data from other European countries could potentially ameliorate this problem; such additional data collection was, however, outside the resources for this research.

Our estimates for cumulative global appliance production are subject to uncertainties because we had to use interpolation and extrapolation to fill data gaps. This introduces an explicit error into our analysis that is excluded from the error intervals displayed in Figures 6.1 and 6.2. In the case of dishwashers and freezers, for which data availability was relatively poor, we perform a sensitivity analysis. This sensitivity analysis indicates that our estimates on cumulative global dishwasher and freezer production introduce an additional uncertainty of 3-5% percent points into the error intervals quantified for these two appliances in Figures 6.1 and 6.2.

6.4.2 Justification for the applicability of the experience curve approach to specific energy consumption Extending the conventional experience curve approach to the specific energy consumption of large household appliances is new and was not attempted in this form before. We think that our method is generally valid and that it offers new insights into the dynamics of energy efficiency improvements in energy demand technologies. In this section, we want to provide a broader justification for our approach. We start out with Ramírez and Worrell (2006), who modeled the specific energy consumption for ammonia and urea production

17 One example in which the resetting of cumulative production is important for estimating reliable learning rates is lighting. Lighting technologies such as conventional incandescent light bulbs, compact fluorescent light bulbs, or light emitting diodes differ in the physical principle that they use for the conversion of electricity into light. Hence, separate experience curves must be developed for each of the individual lighting technologies.

132 Price and efficiency dynamics of large appliances with the experience curve approach. They argue that the experience curve approach is applicable to their case because (i) energy-related costs account for a large part (i.e., more than 70%) of total production costs and (ii) production costs decline at a constant rate with each doubling of cumulative production. Reducing the specific energy consumption of chemical processes is thus a constant point of attention for chemicals manufacturers. It is however less obvious, why improving the energy efficiency of large household appliances should be a constant point of attention of appliance manufacturers, too. One explanation might be that energy efficiency improvements follow autonomously technological innovation. In addition to these autonomous energy efficiency improvements, further energy efficiency potentials might be realized by producers in their quest for decreasing production costs. Improving insulation of cold appliances, for example, might allow for smaller compressors; improving the tub shape of washing machines might reduce water consumption and thus material demand for pipes and pumps. As a consequence, energy efficiency improvements and declining production costs may go hand in hand at the system level. Caution is however required because autonomous energy efficiency improvements might be partially reversed, if product functionality requires so18.

Next to these autonomous and cost-driven energy efficiency improvements, appliance manufacturers might have experienced only little incentives in the past to improve the energy efficiency of their products. Consumers are likely to have chosen large appliances in the 1960s until the early 1990s based on information on primary product functions such as cleaning capacity, storage volume of cold appliances, design, formats, and price because information on these parameters were more openly available than information on energy consumption and use-phase costs. Without adequate product labeling, it is reasonable to assume that consumer awareness for energy consumption and energy-related costs was relatively low for many decades19.

The introduction of minimum energy performance standards for cold appliances (EU, 1996) but more importantly energy labeling of household appliances (EU, 1992) made energy efficiency a product feature. This development was successfully supported in the Netherlands by the introduction of the energy premium regulation (EPR) in the year 2000, which granted subsidies to consumers who bought efficient household appliances. The combined effect of these policy measures increased the market elasticity with regard to the energy efficiency of large household appliances and provided additional incentives for producers to increase the energy efficiency of their products (Ecowet, 2007a). The accelerated decline in the specific energy consumption of cold appliances and dishwashers during the period of enforced energy policy (see, e.g., Figures 6.2 and 6.3 as well as Table 6.2) indicates the potential policy impact on the energy efficiency of large household appliances. Today, specific energy consumption of wet and cold appliances is an important criterion for consumers when purchasing large household appliances (Ecowet,

18 One example is the recent improvement of light chromaticity at the expenses of bulb efficacy in the case of compact fluorescent light bulbs (PL, 2007). 19 Consumers buy large appliances far less frequently than other goods. They hence often lack knowledge and base their decisions upon advice provided by sales personnel or product labels. Forsa (2004) found that 70% of German consumers receive and follow advice from sales personal prior to purchasing a new appliance.

133 Chapter 6

2007a). Producers are hence forced to improve the energy efficiency of their products in a similar way that they aim at reducing production costs.

Based on these considerations, we argue that it is valid to model specific energy consumption with the experience curve approach for recent and also future years. The results of the experience curve analysis are, however, subject to greater uncertainties in the more distant past (e.g., before the enforcement of energy policy in the early 1990s). We argue that modeling energy efficiency dynamics of energy demand technologies with the experience curve approach is useful for future scenario projections because this approach traces specific energy consumption more closely to actual production than simple time- series analyses.

6.4.3 Discussion of results Our findings indicate a systematic decline in both specific prices and specific energy consumption of large appliances. Literature data on yearly price and efficiency changes confirm this trend for many countries and regions. Yearly rates of decline in prices and specific energy consumption show, however, relatively large variation (e.g., Dale et al., 2002; Bertoldi and Atanasiu, 2007; Ellis et al., 2007; see Table A6.1 in the Appendix of this chapter). Unlike our findings, data from literature do not indicate systematic differences between the rates of yearly price decline for wet and cold appliances. The data variation observed in Table A6.1 might be to some extent explained by the following factors: (i) The average yearly changes in price and energy consumption are often calculated in literature based on data for the base year and in the final year of analysis rather than based on regression analysis. Due to relatively high price variability in individual years, the former approach might especially lead to extreme results, if the analyzed time periods are short (e.g., see Table A1 in the Appendix; price decline of refrigerators in Japan at yearly rates of 15% in the period of 2001-2005). (ii) The use of absolute prices and energy consumption that disregard changes in the size and capacity of large appliances (we corrected for these changes by analyzing specific prices and specific energy consumption, whereas most did not). (iii) International differences in market characteristics, governmental taxation policies, or product pricing of producers and retailers.

The experience curve analyses on household appliances as presented by Bass (1980) and Laitner and Sanstad (2004) confirm a general trend towards declining prices of large appliances (Figure 6.4). However, with the exception of freezers, our learning rates generally exceed the ones published in literature (Figure 6.4).

134 Price and efficiency dynamics of large appliances

this study 40 Laitner and Sanstad (2004) Bass (1980)

30 1 1 2 Period: 1980-1998 1 20 Period: 1980-1998 Period: 1970-2003 2 Period: 1964-2008 Learning in % rate Learning Period: 1980-1998 Period: 1950-1961 Period: 1980-1998 Period: 1980-1998 Period: 1947-1974 Period: 1980-1998 Period: 1970-1998 10 Period: 1947-1968 Period: 1922-1940 Period: 1950-1974 Period: 1969-2003 Period: 1965-2008 0 Period: 1968-2007 Washing Laundry dryers Dishwashers Refriger- Freezers machines ators Figure 6.4: Comparison of experience curve results (1 electric laundry dryers, 2 gas laundry dryers)

Due to lack of detailed insight, we can only semi-quantitatively explain parts of the deviations observed in Figure 6.4. The systematic differences between our estimates and the ones of Laitner and Sanstad (2004) for wet appliances are to a large extent caused by the incomplete accounting of cumulative production in the latter publication. Laitner and Sanstad (2004) use the year 1980 as base year of their analysis but do not account for production of appliances in earlier years. This leads firstly to a substantial underestimation of cumulative production in the base year of their analysis and thus secondly to an overestimation of doublings of cumulative production in the time period analyzed. Laitner and Sanstad (2004) therefore substantially underestimate price learning rates. Recalculating the learning rates as identified by Laitner and Sanstad (2004) by using our estimates for cumulative production yields substantially higher learning rates of 46% for washing machines, 31% for refrigerators, and 37% for freezers. Applying data corrections decreases differences between the learning rates for wet appliances, whereas differences for cold appliances become even larger. Inconsistencies similar to the ones identified for Laitner and Sanstad (2004) might also explain deviations between our estimates and the findings of Bass (1980).

The remaining deviations might be explained by differences regarding the time period analyzed. By recalculating our estimates for the same time period as analyzed by Laitner and Sanstad (2004), i.e., 1980-1998, we identify learning rates, which are generally lower and which are subject to substantially higher uncertainty intervals than our estimates presented in Figures 6.1 and 6.2. Further explanations for differences might include the calculation of learning rates by Laitner and Sanstad (2004) based on price differences between base year and final year of analysis (rather than based on regression

135 Chapter 6 analysis) and in the case of freezers the combining of upright and chest freezers into one freezer category20.

Despite considerable energy efficiency improvements in the past decade, our data do not indicate adverse effects on the price of large household appliances. This finding is in line with worldwide trends for large household appliances as found in literature (Ellis et al., 2007; Bertoldi and Atanasiu, 2007; Dale et al., 2009). Not only did absolute and specific prices of large household appliances decline, Dale et al. (2009) also found evidence that the average incremental price for energy efficiency improvements in refrigerators and air conditioners declined as well. This finding indicates that, in general, still considerable and low-cost potentials for future energy efficiency improvements exist.

For utilizing these potentials, effective energy policy will be crucial. Our results demonstrate that energy policy might be able to bend down the slope of energy experience curves thereby accelerating energy efficiency improvements of large appliances. Such a finding is remarkable because literature on price and cost experience curves provides so far no indication that governmental policy can bend down the slope of experience curves (Junginger et. al., 2008). We therefore argue that governmental policy might have larger potentials to accelerate energy efficiency improvements than to accelerate the decline of production costs or retail prices of energy demand technologies. However, caution is required because so far policies with an explicit focus on cost reduction have been rare.

The effectiveness of policy instruments like energy labels depends in particular on their ability to sufficiently differentiate products of low and high energy efficiency. To satisfy this requirement, energy labeling schemes need to be continuously monitored and updated (Ellis et al., 2007; EU, 2003). Updates of energy labels were introduced for cold appliances (EU, 2003) but not for wet appliances for which the market nowadays consists almost entirely of most efficient label-A appliances. Currently the EU is, however, updating energy labels and efficiency standards for large appliances. In this respect, it is important to note that in the case of washing machines and dishwashers, additional efficiency potentials at current test conditions might be relatively limited. However, substantial energy savings can be realized, e.g., due to novel enzyme detergents or ozone treatment of the wash liquor to allow for a decrease in washing temperatures (Ecowet, 2007b).

The extent to which novel and energy-efficient appliances enter the market depends often on their price. Our results show that technological learning offers substantial potentials for cost decline of large appliances. It is hence likely that novel and initially expensive components of large appliances become substantially cheaper after short time periods. Policy makers can thus expect that support of promising but initially expensive energy technologies might result in both declining consumer costs and improved energy efficiency. The case of heat pump laundry dryers illustrates these

20 Such an approach is particularly problematic because upright freezers are considerably more expensive that chest freezers. Changes in the contribution of price data for upright and chest freezers to the average freezer price can have a substantial effect on the calculated learning rate.

136 Price and efficiency dynamics of large appliances dynamics: By 2005, heat pump laundry dryers were newly introduced to the market at prices roughly 650 EUR higher than the ones for conventional condensing laundry dryers (Barthel et al., 2005). Assuming now that heat pump laundry dryers would have been forced into the market and the additional costs for the heat pump and its integration into the dryer system decline at similar learning rates than the prices of laundry dryers (i.e., 28 ± 7%, see Figure 6.1), the price difference between the two competing dryer technologies would have decreased to 30-130 EUR by 200921. On the Dutch market, heat pump laundry dryers are currently 100-300 EUR more expensive than conventional devices (Kieskeurig, 2009). The example of heat pump laundry dryers shows that novel and efficient technologies offer substantial potentials for cost reduction because low initial market sales enable substantial growth of cumulative production within short time periods. Here, governmental policies can support the market uptake of novel and efficient appliances, thereby contributing to a rapid and substantial decline of consumer investment costs. However, policy makers need to be cautious because next to market prices other factors such as product features, consumer convenience, as well as education and awareness of consumers and sales personnel are as well critical factors for the market success of novel and efficient energy demand technologies (see, e.g., Brezet (1994)).

6.5 Conclusions In this chapter, we construct experience curves for the specific price and the specific energy consumption of large wet and cold appliances. We regard the experience curve approach as applicable and as useful for analyzing long-term price and efficiency dynamics of large appliances. For the past three to four decades, we identify a trend towards a continuous decline in specific prices, albeit with differentiation for wet and cold appliances. Our analysis suggests that technological learning is a powerful mechanism that enables a substantial decline in the prices and production costs of large appliances. We hence argue that introducing novel and initially expensive technologies to increase the energy efficiency of large appliances does not need to cause substantial and permanent adverse price effects. The example of heat pump laundry dryers shows, how experience curve analysis can supplement ex-ante bottom-up engineering analyses in providing more reliable forecasts on future technology costs. Applying the experience curve analyses to the specific energy consumption of large appliances is new and reveals useful insights into the dynamics of energy efficiency improvements. In analogy to the rates of price decline, we find that wet appliances show a higher learning rate than cold appliances with respect to specific energy consumption. Our results suggest that energy policy might be able to bend down the slope of energy experience curves, thereby accelerating the decline in the specific energy consumption of large appliances. This finding highlights the importance of energy policy for energy efficiency improvements of large household appliances as well as of other energy demand technologies.

21 For heat pump laundry dryers, we assume here market shares of 0.3% in 2005 and 5% in 2010 (based on European market shares and conservative estimates for future market potentials; see Bush and Nipkow (2006)). If the market for heat pump laundry dryers grows faster than assumed, the additional price will decline even more rapidly.

137 Chapter 6

Acknowledgements This research was funded by the Dutch Ministry of Economic Affairs. We thank Klaas-Jan Koops from the Dutch Ministry of Economic Affairs (The Hague, the Netherlands) for the fruitful and very pleasant cooperation. We are grateful to Faruk Dervis and Mauricio Solano for assisting data collection and to John A. ‘Skip’ Laitner (American Council for an Energy-Efficient Economy, ACEEE), René Kemna (Van Holsteijn and Kemna B.V.), and Hans-Paul Siderius (SenterNovem) for providing valuable background information.

138 Price and efficiency dynamics of large appliances

Appendix Table A6.1: Literature overview: yearly changes in price and specific energy consumption of large appliances Yearly change in % Appliance Source Countrya Time period Price SEC b This study NL 1965-2008 -2.4 -2.5 Bertoldi and Atanasiu (2007) EU-15 1996-2004 - -3.3 CECED (2003)c EU 1994-2002 - -4.5 Washing machines Dale et al. (2002) USA 1983-2001 -2.4 -0.9 EES (2006) AUS 1993-2005 -2.6 -1.3 Laitner and Sanstad (2004) USA 1980-1998 -3.4 - Waide (2001)d EU-15 1996-1998 - -2.5 This study NL 1969-2003 -2.1 -1.5 Bass (1980)e USA 1950-1961 -2.3 - Bass (1980)e USA 1950-1974 -2.2 - Laundry dryers EES (2006) AUS 1993-2005 -1.1 -0.7 Laitner and Sanstad (2004)e USA 1980-1998 -3.2 - Laitner and Sanstad (2004)f USA 1980-1998 -2.9 - This study NL 1968-2007 -3.8 -2.3 Bass (1980) USA 1947-1968 -2.0 - Dishwashers Bass (1980) USA 1947-1974 -2.0 - Ennen (2006)g, c EU 1998-2004 - -5.1 Ennen (2006)h, c EU 1998-2004 - -6.0 This study NL 1964-2008 -1.2 -2.4 Bass (1980) USA 1922-1940 -2.6 Bertoldi and Atanasiu (2007)d EU-15 1993-2005 - -4.3 Bertoldi and Atanasiu (2007)c EU 1993-2004 - -4.5 CECED (2004)c, i EU 1999-2003 - -3.5 Dahlman (2007) AUS 1993-2005 - -3.9 Refrigerators Dale et al. (2002) USA 1980-2001 -2.5 -4.6 EES (2006) AUS 1993-2005 -1.7 -4.6 ECCJ (2006) JPN 2001-2005 -15.1 -5.1 Laitner and Sanstad (2004) USA 1980-1998 -3.2 - Schiellerup (2002) UK 1992-1999 -6.3 -3.9 Schiellerup (2002)j UK 1992-2000 - -3.4 Waide (2001)c, k EU-15 1994-1998 - -2.3 This studyl NL 1970-2003 -1.5 -1.9 This studym NL 1970-2003 -1.1 EES (2006) AUS 1993-2005 -2.5 -3.3 Freezers Laitner and Sanstad (2004) USA 1980-1998 -5.3 - Schiellerup (2002)l UK 1992-1999 -5.1 -3.1 Schiellerup (2002)m UK 1992-1999 -5.0 -5.6 a Abbreviations: AUS – Australia, EU – Europe, EU-15 – 15 member countries of the European Union, JPN – Japan, NL – the Netherlands, UK – United Kingdom, USA – United States of America b SEC – specific energy consumption c including member countries of CECED (European Committee for Domestic Equipment Manufacturers) d sales weighted averages e electric laundry dryers f gas laundry dryers g referring to dishwashers with a capacity of 12 standard place settings h referring to dishwashers with a capacity of 9 standard place settings i total of cold appliances j refrigerator-freezer combinations k covering the total of cold appliances l upright freezers m chest freezers

139

7 A review of experience curve analyses for energy demand technologies

Martin Weiss, Martin Junginger, Martin K. Patel, and Kornelis Blok Accepted for publication in: Technological Forecasting and Social Change

Abstract Transitioning towards sustainable energy systems requires a large-scale introduction of novel energy demand and supply technologies. Such novel technologies are often expensive at the point of their market introduction but eventually become cheaper due to technological learning. In order to quantify potentials for price and cost decline, the experience curve approach has been extensively applied to renewable and non-renewable energy supply technologies. However, its application to energy demand technologies is far less frequent. Here, we provide the first comprehensive review of experience curve analyses for energy demand technologies. We find a widespread trend towards declining prices and costs at an average learning rate of 18 ± 9%. This finding is consistent with the results for energy supply technologies and for manufacturing in general. Learning rates for individual energy demand technologies are symmetrically distributed around the arithmetic mean of the data sample. Absolute variation of learning rates within individual technology clusters of 7 ± 4% and between technology clusters of 7 ± 5% both contribute to the overall variability of learning rates. Our results show that technological learning is as important for energy demand technologies as it is for energy supply technologies. Applying the experience curve approach to forecast technology costs involves, however, unresolved uncertainties, as we demonstrate in a case study for the micro-cogeneration technology.

141 Chapter 7

7.1 Introduction Introducing renewable energy technologies and improving the economy-wide energy efficiency are key strategies for a sustainable global energy system (IEA, 2008c). The quest for sustainable energy supply and demand is, however, complex and extremely challenging for various reasons. Given the magnitude of environmental and social problems associated with the current energy system, novel technologies must be deployed globally at large scale. Furthermore, investments into novel and efficient energy technologies are often unprofitable because current energy markets are distorted in favor of incumbent technologies (Ferioli et al., 2009). Novel energy supply and demand technologies in particular face the problem that they are relatively expensive at the point of their market commercialization. However, these technologies might become cheaper due to technological learning, i.e., the combination of various mechanisms such as learning-by-doing, economies of scale, technological innovation, or factor substitution in manufacturing. Berglund and Söderholm (2006) argue that technological learning as driver of cost decline for energy supply and demand technologies is likely to be the single most important factor for shaping our future global energy system. For manufacturers and policy makers alike, it is therefore of critical importance to obtain quantitative insight into the prospects for cost decline of novel energy technologies.

One widely applied tool for quantifying the cost dynamics of technologies is the experience curve approach that models production costs of technologies as a power-law function of cumulative experience, i.e., cumulative production. Cost dynamics are thereby quantified by learning rates, which indicate the rate of cost decline with each doubling of cumulative production. For several decades, experience curves have been used for strategic planning in manufacturing (Dutton and Thomas, 1984; Argote and Epple, 1990). Since the 1990s, experience curves are increasingly applied to establish efficient energy technology policies and to forecast technology diffusion and technological change by energy and greenhouse gas (GHG) emission models (Wene et al., 2000; IEA, 2000; Kahouli-Brahmi, 2008).

In the context of sustainable energy supply, the experience curve approach has been widely applied and redefined for renewable and conventional energy supply technologies (Neij, 1997; Claeson Colpier and Cornland, 2002; Junginger et al., 2004, 2006). Neij (2008) finds that the results of experience curve analyses for energy supply technologies are generally reliable, i.e., consistent with the outcomes of bottom-up technology assessments. Using the extensive body of literature, comprehensive overview studies were prepared in the past with the aim of devising average technology-specific learning rates for energy supply technologies (IEA, 2000; McDonald and Schrattenholzer, 2001; Neij et al., 2006; Neij, 2008; Kahouli-Brahmi, 2008). However, similar efforts for energy demand technologies are largely missing.

Experience curves for energy demand technologies are scarce (e.g., Laitner and Sanstad, 2004; Jakob and Madlener, 2004; Weiss et al., 2008c, 2009b, c) and a comprehensive literature review is still unavailable to date. The absence of average learning rates for individual energy demand technologies and technology clusters makes it

142 A review of experience curve analyses for energy demand technologies still necessary to devise experience curves for each technology in order to evaluate prospects of cost decline. For the large and very heterogeneous group of energy demand technologies, this is a cumbersome and time consuming task.

In this chapter, we investigate whether it is possible to devise generic ex-ante learning rates for individual energy demand technologies and technology clusters. To address existing knowledge gaps, we first present a comprehensive literature review of experience curve studies on energy demand technologies. Based on this review, we quantify technology-specific learning rates and compare our results with findings in literature for energy supply technologies. We furthermore identify potentials and limi-tations of the experience curve approach to generate reliable ex-ante cost estimates for energy demand technologies. Such analysis can provide additional insights into the usefulness of the experience curve approach for energy policy and energy modeling.

In the next section, we briefly explain the experience curve approach and our research methodology. We present our results in Section 7.3 and we provide a discussion of our findings in Section 7.4. We draw conclusions in Section 7.5.

7.2 Methodology The experience curve approach dates back to the 1930s, when Wright (1936) found that unit labor costs in airframe manufacturing decline at a constant rate with each doubling of cumulative production. Wright (1936) noted the particular relevance of his findings for the investigation of future cost developments in manufacturing. The graphical representation of Wright’s discovery is nowadays referred to as learning curve, which applies to the effects of learning-by-doing, i.e., the decline in labor costs due to decreasing working time requirements for manufacturing. Arrow (1962) introduced the notion that declining labor costs are the result of growing experience. The Boston Consulting Group extended the concept of technological learning by analyzing the dynamics of total production costs as a function of cumulative production (BCG, 1972). This (black-box) modeling of total production costs as function of cumulative production is generally referred to as experience curve approach. Dutton and Thomas (1984) differentiate experience curves and progress curves; the first represent average production costs of multiple manufacturers, whereas the second represent production costs at the level of individual firms. Here, we are interested in the price and cost dynamics of energy demand technologies. Technology- specific data in the literature generally refer to price and cost averages of various manufacturers in a specific country, region, or even on a global level. Hence, we uniformly use the term experience curves throughout this chapter when referring to the graphical representation of prices and cost as a function of cumulative production.

143 Chapter 7

The experience curve approach expresses production costs (and under specific preconditions also prices)1 of technologies as a power-law function of cumulative production:

= ⋅ bi Ccumi C0,i (Pcumi ) (7.1) where Ccumi represents the price or costs at Pcumi, C0,i stands for the price or costs of the first unit produced, Pcumi for the cumulative production, and bi for the technology- specific experience index of technology i. By applying the logarithmic function to Equation 7.1, a linear experience curve can be plotted with bi as the slope parameter and log C0,i as the price or cost axis intercept (Figure 7.1). A technology-specific progress ratio (PRi) [%] and learning rate (LRi) [%] can be calculated as rate at which the price or costs of a technology decrease with each doubling of cumulative production:

= bi PRi 2 (7.2)

= − = − bi LRi 1 PRi 1 2 (7.3)

100 1988 90

th 80 1983 //kW

2006 70

in EUR 60 Condensing gas combi boilers: R2 = 0.98; PR = 86 ± 1%; LR = 14 ± 1% 50 Condensing gas space heating boilers:

Price ofcondensing gas combiboilers 2 R = 0.92; PR = 94 ± 1%; LR = 6 ± 1% 2006

103 104 105

Cumulative sales of condensing gas boilers in the Netherlands in MWth

Condensing gas combi boilers (providing combined space heating and hot tap water) Condensing gas space heating boilers

Figure 7.1: Experience curves for two types of condensing gas boilers in the Netherlands; numbers in the diagram indicate the year of analysis (data source: Weiss et al., 2009b)

1 In practical applications of the experience curve approach, production costs are often approximated by market prices because data availability is limited. This approximation introduces uncertainty into experience curve analysis and is, strictly speaking, only valid, if profit margins of producers remain constant in the period of study (see discussion in Section 7.4.1).

144 A review of experience curve analyses for energy demand technologies

Experience curves have been constructed for a wide range of products, technologies, and processes. Empirical results almost always indicate declining prices and costs with increasing cumulative production (Nemet, 2009). However, the rates at which production costs decline are technology dependent and even vary for identical technologies depending on the time period and geographical system boundary (i.e., the country or region) chosen for the analysis (Dutton and Thomas, 1984; McDonald and Schrattenholzer, 2001; Kahouli-Brahmi, 2008, Nemet, 2009). In the past, several attempts were made to conceptually extend the experience curve approach. These include: (i) the modeling of specific energy consumption instead of costs as the dependent variable, thereby modeling energy efficiency as a power-law function of cumulative production (ii) introducing additional explanatory variables next to cumulative production into the experience curve approach, thereby modeling cost decline as a multiple factor process

Ramírez and Worrell (2006), for example, analyzed specific energy consumption of ammonia and urea production as function of cumulative production. Hettinga et al. (2009) applied the experience curve approach to the specific energy consumption of ethanol production from corn in the United States. Weiss et al. (2009c) modeled the dynamics of specific energy consumption of large appliances as function of cumulative appliance production. All studies found a relatively good fit of empirical data (coefficient of determination R2 > 0.7) to the hypothesized experience curve pattern. Klaassen et al. (2005), Jamasb (2007), and Söderholm and Klaassen (2007) developed two-factor experience curves to identify the effect of cumulative production, cumulative spending for research and development (R&D), as well as cumulative R&D-based knowledge stock on cost decline of energy supply technologies. These studies broadly support the experience curve hypothesis, indicating that learning rates for production costs are generally higher as a function of cumulative R&D spending than as a function of cumulative production.

In this chapter, we focus on conventional one-factor price and cost experience curves because these represent the vast majority of published studies. We compile in the first instance an overview table of experience curve analyses on energy demand technologies. Based on this overview table, we plot a frequency histogram of learning rates for energy demand technologies. We identify average learning rates and their associated uncertainty intervals (i.e., the mean and standard deviation of data) for energy demand technologies in aggregate and for individual technologies and technology clusters.

We differentiate individual technology clusters based on function and characteristics of technologies as well as type of the energy conversion process performed. Such an analysis might reveal insight into pattern of learning rates for the individual clusters of energy demand technologies, thereby potentially contributing to a more detailed and accurate forecasting of future technology costs. Again, we estimate uncertainty intervals of un-weighted averages by the standard deviation of data for individual technology clusters. We apply a simple t-test to identify whether the differences in the learning rates between individual technologies and technology clusters are significant.

145 Chapter 7

To test whether learning rates for energy demand technologies are symmetrically distributed around the mean, we calculate the coefficient of skewness of the entire data sample as:

n B (x − x)3 g = i=1 i (7.4) nσ 3 where g represents the sample coefficient of skewness, n stands for the sample size, for the sample standard deviation, for the mean learning rate of the sample, and xi for the learning rate of energy demand technology i. Furthermore, we apply one-way ANOVA (Analysis of variance) to test whether skewed distributions (i.e., Weibull distribution and Lognormal distribution) provide a better fit to our data than the Normal distribution. We compare our findings with results for energy supply technologies and for manufacturing in general. This comparison might reveal, in particular, differences regarding average learning rates as well as the distribution of learning rates of energy demand and supply technologies, allowing for more detailed insights into technological learning and future technology costs. For this comparison, we update existing inventories of experience curve studies as prepared by Dutton and Thomas (1984), McDonald and Schrattenholzer (2001), Kahouli-Brahmi (2008), and Junginger et al. (2008).

7.3 Results We start out by presenting a comprehensive overview of experience curve studies on energy demand technologies (see Table A7.1 in the Appendix of this chapter)2. The experience curve approach has been applied frequently to compact fluorescent light bulbs (CFLs), large household appliances, gas boilers, and various consumer electronics. Technology-specific learning rates span a relatively wide range from 4% (condensing gas boilers; Martinus et al. (2005)) to 41% (CFLs and ballasts thereof; Iwafune (2000)). Average learning rates for individual technologies and technology clusters range from 9 ± 3% for refrigerators to 26 ± 3% for consumer electronics (Figure 7.2). We identify substantial variation of learning rates within individual technology clusters (in the following referred to as within-technology variation) and between the various technology clusters (in the following referred to as between-technology variation).

Based on the standard deviation for each sample of technology-specific learning rates, we find that absolute within-technology variation of learning rates ranges between 2% for Ford’s model T to 14% for washing machines. Ford’s model T was a clearly defined and technically homogeneous product, which did not change considerably over time; it shows thus only a small span of learning rates. By contrast, washing machines show large differences regarding product design and the services they provide. Variation

2 We include here and in the following analyses, to the best of our knowledge, an overview of learning rates for energy demand technologies as identified in research reports and peer reviewed articles. We do not claim absolute completeness with regard to all learning rates published. Our overview aims at (i) illustrating the spectrum of learning rates for energy demand technologies and (i) giving an indication about data averages and data distribution.

146 A review of experience curve analyses for energy demand technologies might be introduced into learning rates by, e.g., analyzing different time periods and by uncertainties related to the correct accounting of cumulative production of washing machines (which is a technology that has been produced commercially for almost a century). Overall, absolute within-technology variation of learning rates averages at 7 ± 4%. This result indicates that averages of technology-specific learning rates for energy demand technologies are subject to considerable uncertainties. It is hence not surprising that we find average absolute between-technology variation of learning rates (7 ± 5%) to be in the same value range as the within-technology variation.

Automotive; Ford, model T (3) Building insulation and glazing (5) Residential heat pumps (2) Other residential heating technologies (6) Air conditioners (6) Washing machines (2) Laundry dryers (5) Dishwashers (4) Refrigerators (3) Freezers (3) Compact fluorescent light bulbs (8) Lamp ballasts (6) Television sets (4) Other consumer electronics (4) Electronic components (14)

02 4 6 8 10203040 12 14 16 18 22 24 26 28 32 34 36 38 Learning rates in % Figure 7.2: Average learning rates and associated uncertainty intervals (i.e., standard deviations of data samples) for individual energy demand technologies and technology clusters; values in parentheses indicate the number of studies included in our analysis3

Both within-technology variation and between-technology variation of learning rates are to some extent caused by differences in performance measures applied by individual studies, i.e., the use of either absolute prices, specific prices, or unit costs as the dependent variable as well as the use of either cumulative capacity or cumulative production referring to individual countries, regions, or the world as a whole as the independent variable. Nevertheless, the substantial between-technology variation of learning rates (Figure 7.2) suggests that it might be possible to differentiate technology- specific learning rates. For this reason, we apply the t-test to identify whether between- technology differences of learning rates are statistically significant.

Our results indicate that between-technology deviations of learning rates are often insignificant at a 95% confidence level (Table A7.2 in the Appendix of this chapter). However, in several cases, we find significant differences between individual technologies and technology clusters. For example, residential heat pumps as well as consumer

3 For an explanation of technologies comprised in individual technology clusters we refer the reader to Table A1 in the Appendix.

147 Chapter 7 electronics and components thereof show significantly higher learning rates than most other technologies (e.g., building insulation and window glazing, other residential heating technologies, or refrigerators). Significance testing with the t-test reveals further that the combined cluster of consumer electronics and components thereof shows a higher average learning rate (23 ± 4%) than the combined group of the remaining energy demand technologies (16 ± 9%). Consumer electronics and components thereof consist of novel semiconductor materials and components4, which seem to offer a large potential for cost decline along the manufacturing chain. One explanation for this finding might be that these materials and components attain fast doublings of their cumulative production. By contrast, the majority of other energy demand technologies consists to a larger extent of materials and components which have already been manufactured for many decades and which are widely used as components of many other technologies as well. Furthermore, several of the other energy demand technologies require a comparatively low level of integration in manufacturing (e.g., building insulation and glazing, cold appliances). Weiss et al. (2009c) argue along these lines when attributing deviations in the learning rates of cold appliances and wet appliances to differences regarding technical complexity as well as potentials for enhanced automation and innovation in component manufacturing and final assembly. Following this argument, we explain the relatively high learning rates for residential heat pumps mainly with substantial technological learning and scale effects in the assembly of entire heat pumps systems.

Caution is however required when interpreting our empirical findings, because sample sizes for individual technologies are too small to draw definite conclusions. Our literature overview shows that experience curve studies are still absent for several important energy demand technologies such as electric motors, entire motor systems, and many small appliances (e.g., fans, coffee makers, computers, novel plasma and liquid- crystal-display televisions, as well as other portable and non-portable consumer electronics). This finding points to an important knowledge gap because the mentioned technologies together consume a substantial amount of energy and show a rapid and disproportional increase in absolute energy consumption (IEA, 2008c). At the same time they offer substantial potentials for energy efficiency improvements and therefore deserve special attention in future energy demand modeling5.

Based on our literature review, we now plot the overall frequency distribution of learning rates for energy demand technologies (Figure 7.3). We find an average (mean) learning rate of 18 ± 9%. The frequency distribution of learning rates for energy demand technologies is slightly asymmetric with a coefficient of skewness of 0.51. This finding indicates that a small majority of learning rates shown in Table A1 is located right of the mean. We attribute this asymmetry largely to the extremely high learning rates of 41%,

4 We define here novel components as components which reach doublings of their cumulative production quickly; thus showing a substantial cost decline in a distinct time period. By contrast, mature components reach doublings of their cumulative production only slowly, thereby showing only marginal cost decline in the time period under consideration. 5 Motor systems account for 70% of total electricity consumption in industry and roughly 30% of electricity use in the service sector of the European Union. To date, they still offer energy efficiency potentials of 20- 30% (de Almeida et al., 2008).

148 A review of experience curve analyses for energy demand technologies which Iwafune (2000) identified for modular-magnetic CFLs and CFL ballasts6. Fitting continuous probability distributions to our data, we find that positively skewed distributions such as the Weibull distribution or the Lognormal distribution do not provide a significantly better fit to our data than the symmetric Normal distribution7.

9 Automotive; Ford, model T 8 Building insulation and glazing Residential heat pumps 7 Other residential heating technologies Air conditioners 6 Washing machines Laundry dryers y Dishwashers 5 Refrigerators Freezers

requenc 4 Compact fluorescent light bulbs F Lamp ballasts 3 Television sets Other consumer electronics 2 Electronics components Normal distribution (R2 = 0.84) 1 Weibull distribution (R2 = 0.86) Lognormal distribution (R2 = 0.82) 0 1-2 3-4 5-6 7-8 9-10 (-1)-0 11-12 13-14 15-16 17-18 19-20 21-22 23-24 25-26 27-28 29-30 31-32 33-34 35-36 37-38 39-40 41-42 43-44 45-46 47-48 49-50 (-5)-(-4) (-3)-(-1) Learning rate in % Figure 7.3: Frequency histogram and fitted distributions of learning rates for energy demand technologies; n = 75

A comparison of our findings with learning rates for energy supply technologies indicates relatively high conformity. Learning rates for energy supply technologies span from -3% for electricity produced from wind energy in Germany during the period from 1991 to 1999 (Ibenholt, 2002) to 47% for photovoltaic modules in the period from 1984 to 1987 (IEA, 2000)8. The average learning rates for individual technologies and technology clusters range from 9 ± 8% for nuclear power plants and 9 ± 5% for coal and lignite power plants and coal boilers to 22 ± 8% and 24 ± 14% for photovoltaics and biomass production, respectively9 (Figure 7.4). Our results for energy supply technologies are in line with average learning rates as proposed by Neij (1997). However, technology-specific learning rates often span relatively wide ranges. Differences in performance parameters again provide one explanation for data variability; technology-specifics such as differences regarding the novelty of components and the inclusion and exclusion of efficiency improvements provide another.

6 Excluding the learning rates of 41% from the data sample reduces the coefficient of skewness to 0.2, which indicates roughly a symmetrical distribution of learning rates for energy demand technologies. 7 Based on one-way ANOVA testing of ordinary residuals at a 95% confidence level (F = 1.22; p = 0.30). 8 We include in the overview, to the best of our knowledge, learning rates for energy supply technologies as identified for capacity costs and energy costs in research reports and peer reviewed articles. We do not claim absolute completeness with regard to learning rates published for energy supply technologies in the various time periods. Our overview aims at illustrating the spectrum of learning rates, giving an indication about data averages and data distribution. 9 The span of averages presented here excludes technologies for which only one learning rate could be indentified in literature (i.e., hydro power plants and high voltage direct current converter stations).

149 Chapter 7

Energy extraction (3)1 Energy transport (3)2 Electricity conversion (1)3 Accessories of energy production (5)4 Miscellaneous energy production (8)5 Fossil fuels - coal and lignite (9)6 Fossil fuels - natural gas (11)7 Nuclear energy (4)8 Hydro energy (1)9 Wind energy (41)10 Photovoltaics (27)11 Energy and fuels from biomass (16)12 Biomass production (5)13

02 4 6 8 10203040 12 14 16 18 22 24 26 28 32 34 36 38 Learning rates in % Figure 7.4: Average learning rates and associated uncertainty intervals (i.e. standard deviations of data samples) for individual energy supply technologies and technology clusters; principal data sources: McDonald and Schrattenholzer (2001), Kahouli-Brahmi (2008), Junginger et al. (2008); values in parentheses indicate the number of learning rates included in our analysis (1including oil extraction, coal production for electric utilities, and crude oil production at well; 2including submarine HVDC (high-voltage direct current) cables, on-shore and off-shore pipelines; 3including HVDC converter stations; 4including flue gas desulfurization and selective catalytic reduction; 5including retail gasoline processing, electric power production, LNG (liquefied natural gas) production, fluid petroleum cracking, bitumen production from non-conventional oil and oil sands; 6including coal, supercritical coal, pulverized coal and lignite power plants, as well as pulverized coal boilers; 7including gas turbines and gas turbine combined cycle (GTCC) power plants; 8including nuclear power plants; 9including hydro power plants; 10including wind power plants and components thereof; 11including solar modules, panels, and entire photovoltaic systems; 12including bio-ethanol, biogas, bio-diesel, and electricity from biomass; 13including the production of corn, sugar cane, rapeseed, as well as logistic chains for forest wood chips)

Overall, the costs of energy supply technologies decline on average at rates of 16 ± 9% with each doubling of cumulative production (Figure 7.5). Similar to energy demand technologies, learning rates for energy supply technologies are approximately normally distributed; showing only a small positive asymmetry in their frequency distribution (coefficient of skewness of 0.73). However, by contrast to the relatively homogenous distribution of learning rates for energy demand technologies, Figure 7.5 indicates two peaks in the distribution of learning rates for energy supply technologies. Such a two-peak-distribution of learning rates might be explained with a top-down framework based on cybernetic theory as suggested by Wene (2008a). This framework identifies a learning rate of 20%, if there is no systematic disturbance of an operationally closed learning system. Systematic deviations can result, e.g., in learning rates of 7% and

150 A review of experience curve analyses for energy demand technologies below, if the learning system is subject to perturbations such as ill-defined system boundaries, technology spillover, or governmental interventions10.

16 Energy extraction Energy transport 14 Electricity conversion Accessories of energy production 12 Miscellaneous energy production Fossil fuels - coal and lignite 10 Fossil fuels - natural gas Nuclear energy 8 Hydro energy Wind energy Frequency Photovoltaics 6 Energy and fuels from biomass Biomass production 4 Normal distribution (R2 = 0.78) Weibull distribution (R2 = 0.79) 2 Lognormal distribution (R2 = 0.75)

0 1-2 3-4 5-6 7-8 9-10 (-1)-0 45-46 47-48 49-50 11-12 13-14 15-16 17-18 19-20 21-22 23-24 25-26 27-28 29-30 31-32 33-34 35-36 37-38 39-40 41-42 43-44 (-5)-(-4) (-3)-(-2) Learning rate in % Figure 7.5: Frequency histogram and fitted distributions of learning rates for energy supply technologies; n = 132; including experience curve studies for resource extraction, energy conversion, energy transport, as well as fossil, nuclear, and renewable energy supply technologies (principal data sources: McDonald and Schrattenholzer, 2001; Kahouli-Brahmi, 2008; Junginger et al., 2008)

We now take a broader perspective and compare our findings for energy demand technologies with learning rates as identified for a wide range of manufacturing industries. We base our comparison on an early literature review presented by Dutton and Thomas (1984) that comprises of 108 learning rates for individual manufacturers of electronics and components thereof, machine tools, paper, aircraft, steel, apparel, and automobiles11. We extend these data with learning rates identified by Junginger et al. (2008) for the manufacturing of chemical and petrochemical products. Again, learning rates are approximately normally distributed around the mean of 19 ± 8% (Figure 7.6).

10 The theoretical framework developed by Wene (2008b) can be applied only to a limited extent to energy demand technologies because it requires that all technical characteristics of a technology remain constant. This criterion is generally fulfilled by energy supply technologies but only to a far less extent by energy demand technologies, which experience a frequent addition of new technical functions and features. 11 Dutton and Thomas (1984) exclude from their overview industry-level experience curve analyses that analyze average economy-wide price decline of technologies. Here, we combine the results of their study with industry-level learning rates for chemicals.

151 Chapter 7

10 Cellophane 8 Polymers Non-specified chemicals Manufacturing (Dutton and Thomas, 1984) Frequency 6 Normal distribution (R2 = 0.85) Weibull distribution (R2 = 0.91) 2 4 Lognormal distribution (R = 0.90)

0 1-0 3-2 5-4 7-6 9-8 11-10 13-12 15-14 17-16 19-18 21-20 23-22 25-24 27-26 29-28 31-30 33-32 35-34 37-36 39-38 41-40 43-42 45-44 47-46 49-48 51-50 (-5)-(-6) (-3)-(-4) (-1)-(-2) (-7)-(-8) Learning rate in % Figure 7.6: Frequency histogram and fitted distributions of learning rates in the manufacturing industry; n = 124 (data sources: Dutton and Thomas, 1984; Junginger et al., 2008)

Overall, our literature overview indicates a widespread and robust development towards declining prices and production costs for energy demand and supply technologies as well as for manufacturing in general. The statistical properties of the distribution of learning rates for energy demand and supply technologies as well as for other manufacturing are remarkably similar. This finding strongly supports McDonald and Schrattenholzer (2001), who argued that learning rates identified for non-energy-related technologies provide a useful starting point for energy modelers until more detailed technology-specific data are available. However, our findings also indicate substantial within-technology and between-technology variation of learning rates for both energy demand and supply technologies. Within-technology variation often makes it difficult to differentiate learning rates for individual energy demand technologies. In the next section, we discuss in greater detail the strength and weaknesses of the experience curve approach and we explore the possibility of using learning rates to generate reliable technology- specific cost projections.

7.4 Discussion 7.4.1 Discussion of uncertainties In the absence of other reliable and easy-to-use methodologies, the experience curve approach has been widely applied to forecast technology costs. Experience curves provide a very useful first-order approximation of cost decline based on a simple linear regression analysis. Alberth (2008) showed that experience curves often allow tracing costs more precisely than simple time-series analysis. In comparison to conventional bottom-up engineering analysis, experience curves permit more realistic cost projections because they account for technological learning rather than statically assuming constant technology costs (Dale et al., 2009). However, the experience curve approach has several caveats that limit its ability to generate reliable cost projections for technologies. In practice, the usefulness of experience curve analysis is determined by two factors: (i) the reliability of

152 A review of experience curve analyses for energy demand technologies the approach itself and (ii) the quality of model input data. In this section, we address both points, thereby clarifying major limitations of the experience curve approach.

The approach of modeling production costs as a power-law function of cumulative production has been widely discussed in the literature (e.g., Alchian, 1963; Dutton and Thomas, 1984; Sallenave, 1985; McDonald and Schrattenholzer, 2001; Alberth, 2008; Kahouli-Brahmi, 2008; Junginger et al., 2008; Ferioli et al., 2009). The phenomenon that production costs decline at a constant rate with each doubling of cumulative production is an empirical observation but not a natural law (Dutton and Thomas, 1984). The classical case of airplane manufacturing discussed by Argote and Epple (1990) and the analysis of nuclear power in the USA (Hultman and Koomey, 2007) both indicate that neither do technology costs have to decline with increasing production nor does the rate of cost decline per se needs to remain constant. Although several studies demonstrate a relatively strong time-dependency of learning rates (e.g., IEA (2000), Claeson Colpier and Cornland (2002), Alberth (2008)) cost decline is neither fixed nor automatic but mainly the result of deliberate decision making. Hence, cost decline must be regarded as the outcome of complex multi-level interactions of mechanisms, which are both endogenous and exogenous of the manufacturing process (Figure 7.7).

At the company level, costs tend to decline due to mechanisms such as learning- by-doing, research and development (R&D), or economies of scale. Factors like organizational forgetting, employee turnover, or transfer of knowledge might counterbalance the effect of these mechanisms. The observed experience curve pattern of production costs refers thereby strictly only to the costs of value added (Sallenave, 1985). Manufacturers can aim at minimizing the quantities of production factors that are employed to generate value added and they can substitute production factors (e.g., energy and capital for labor). However, they have generally little if no influence on the price of their production factors. In reality, prices of raw materials and intermediate goods, services, capital, energy, and labor are volatile. Changing prices of production factors introduce time dependency into experience curve analysis and limit the reliability of experience curve results, if these are used to forecast technology costs (Ferioli et al., 2009).

153 Chapter 7

Exogenous learning in manufacturing: • general scientific and technological progress • capital depreciation and periodic replacement of equipment • learning of capital goods suppliers • learning elsewhere

Endogenous learning in manufacturing: • learning-by-doing • learning-by-using • learning-by-interacting • investments into improved capital goods • private and public investments into research and development • copying technological innovations of competitors • increased tooling and automation • process innovation in manufacturing • factor substitution

Changes in the price of production Organizational learning: factors: • employee turnover • changing costs for resource extraction • management decisions • changing profit margins of suppliers • knowledge transfer • others

Multilevel interactions

Market competition Mark-ups of Profit margins wholesalers and retailers Taxation

Production costs Producer prices Market prices

Figure 7.7: Production costs and market prices of technologies as determined by mechanisms, which are exogenous and endogenous of the learning system

Modeling cost decline as a function of cumulative production only raises problems related to omitted variables (Kahouli-Brahmi, 2008) because in reality costs may also decline as a function of time, cumulative investment, research and development spending, autonomous technological change, scale effects, or volatility of input prices (e.g., Arrow (1962), Newell et al. (1999), Kahouli-Brahmi (2008), Alberth (2008)). Although in practice cumulative production may be the best single proxy for the various drivers of cost decline, the omission of other explanatory variables introduces bias into any experience curve analysis. Related to the problem of omitted variable bias is also the role of technology spillover: Production costs decline not only due to technological learning in the manufacturing of a specific technology but also by building knowledge stock in other sectors of the economy. Future analyses should address this point by providing a more detailed account of the effect of technology spillover on technological learning (Söderholm and Klaassen, 2007).

Alberth (2008) points to an econometric caveat of the experience curve approach that results from the convexity of the logarithmic function. The results of cost forecasts based on experience curves are found to be symmetric and unbiased in log format. However, due to the convexity of the logarithmic function, cost estimates might become

154 A review of experience curve analyses for energy demand technologies increasingly asymmetric and biased towards cost decline, if costs projections are made far into the future. Focusing on the practical application of the experience curve approach, van Sark (2007) and Nemet (2009) criticize experience curve analyses that indicate the goodness of fit of empirical data by providing the coefficient of determination (R2) but not the implicit error of the regression analysis (i.e., the error of the learning rate PR). Providing the error of estimated learning rates is, however, essential because such a practice reveals quantitative information about the implicit uncertainty of experience curve analysis to modelers and policy makers.

The methodological limitations of the experience curve approach are to some extent linked to the second source of uncertainty in experience curve analysis, i.e., the quality of model input data. Ideally, data on the costs of value added should be analyzed in a regression analysis as a function of multiple drivers as discussed above. In reality, however, data availability is often limited to such and extent that makes it necessary to approximate total production costs and their various drivers by market prices and cumulative production only. The approach to approximate actual production costs by market prices is common practice but only valid, if profit margins of producers as well as markups of wholesalers and retailers remain constant. In reality, this is almost never the case, making prices an imperfect measure of production costs (McDonald and Schrattenholzer, 2001). According to IEA (2000), market prices initially decline at lower rates than production costs. In the following phase of enhanced market competition, price decline tends to exceed cost decline. Only after products mature, market prices parallel production costs. However, even then, prices may temporarily increase in times of rising demand. Also, induced demand by policy support measures or policy targets can (at least temporarily) trigger slower price decline or even price increase. The use of price data thus introduces time dependency into experience curve analyses. This conclusion is substantiated by Bass (1980) who finds that learning rates can vary by more than 50% for identical technologies, if different time periods are analyzed.

In practice, limited data availability often causes erroneous experience curve analyses, which suffer from partially invalid definitions of geographic and technological system boundaries. Empirical studies often remain vague on whether they estimate cumulative experience correctly by including total production of a technology or whether they only consider cumulative production in the time period and region of analysis. In particular, the incomplete accounting of cumulative production (i.e., the exclusion of production prior to the time periods covered by price or cost data) leads to a systematic overestimation of doublings of cumulative production and thus to an underestimation of actual learning rates.

In the case of energy demand technologies, additional peculiarities limit the applicability of the experience curve approach: Unlike the relatively small and well confined group of energy supply technologies, which serve the purpose of producing energy (e.g., to supply 1 kWh of power), energy demand technologies provide services to consumers (i.e., cleaning laundry or preserving food). Energy use is here only a means to an end. This leads to product differentiation and market segmentation to a far greater extent than is the case for energy supply technologies. The heterogeneous group of energy

155 Chapter 7 demand technologies shows therefore a large variability in prices, use-phase costs, and product functionality. This introduces additional uncertainty into price and cost data. The complex interplay of changes in technology components (e.g., changes from non- modulating burners and ventilators to modulating burners and ventilators in gas boilers) and improvements in product functionality (e.g., introduction of centrifuge-drying into washing machines) suggests that the decline in absolute production costs of energy demand technologies might be even higher than suggested by the learning rates in Figure 7.2 (Weiss et al., 2009b, c). It is hence somewhat surprising to find similar results for energy demand and supply technologies. Apparently, improved functionality does not introduce a substantial systematic error into the overall average learning rate as calculated for energy demand technologies. This finding might indicate that also additional measures for improving the energy efficiency of energy demand technologies do not need to cause substantial adverse effects on production costs and consumer prices in the medium and long term.

Apart from these more principle uncertainties, our approach of estimating average learning rates for energy demand and supply technologies based on learning rates of individual technologies might include a bias. In our analysis, we include several experience curve studies for the same type of technologies to calculate average learning rates of energy demand technologies. In particular, the average learning rate of energy demand technologies is based on a relatively large number of relatively high learning rates for electronics components, where as other technologies are not included in our analysis at all (e.g., several small appliances or electric motors) or are underrepresented (e.g., cold appliances). To identify whether such an approach introduces substantial bias into our results, we conduct a sensitivity analysis. For this, we calculate average learning rates of energy demand and supply technologies based on averages for individual technologies and technology clusters. This way, we correct for differences in the amount of studies that analyze individual technologies. We find, that average learning rates calculated in our sensitivity analysis for energy demand technologies (17 ± 7%) and energy supply technologies (16 ± 8%) deviate only marginally from our original estimates (i.e., LR of 18 ± 9% for energy demand technologies and LR of 16 ± 9% in the case of energy supply technologies. This finding indicates that the potential overrepresentation of individual technologies does not introduce a substantial systematic error into the average learning rates identified for both energy demand and supply technologies.

7.4.2 Discussion of methodological extensions The discussion so far has shown that the experience curve approach is subject to caveats when used to make precise and accurate cost projections for technologies. These caveats open areas for further research. Multiple factor experience curves that account for additional independent variables (such as R&D spending or cumulative knowledge stock) in addition to cumulative production present one way to expand our understanding of technological learning. The two-factor learning curves as prepared by Klaassen et al. (2005), Jamasb (2007), and Söderholm and Klaassen (2007) resent important steps in this

156 A review of experience curve analyses for energy demand technologies direction12. These studies also point to two distinct aspects, which are worth being explored by future research. First, in addition to multi-factor experience curves, the problem of omitted variables should be addressed by developing a system of regression equations in which each equation addresses a specific aspect of the technology development process. Second, the studies mentioned above focus next to cumulative production on public (rather than private) R&D spending, mainly due to data availability reasons. Future research efforts addressing also the role of private R&D should be encouraged to allow for more complete insights into the role of R&D as driver for cost decline.

Ferioli et al. (2009) propose to regard technologies or processes as aggregates of their components. Overall technology-specific costs can then be modeled as function of the sum of costs for individual components:

n = ⋅ bi = ⋅ b1 + ⋅ b2 + + ⋅ bn Ccumi BC0,i (Pcumi ) C0,1 (Pcum1 ) C0,2 (Pcum2 ) ... C0,n (Pcumn ) (7.5) i=1 where i represents the individual cost components of a given technology. Such a disaggregate analysis allows for differentiating experience indices and cumulative production for individual components and provides modelers in theory ample possibilities to introduce additional terms for cost components, which do not learn at all or which show strong time dependency. However, in practice, such analysis is frequently hampered by the lack of publicly available data as well as by difficulties in obtaining consistent time-series data (e.g., because the mix of components may alter over time due to technological innovation).

Analogous to Ferioli et al. (2009), cost decline of technologies might be differentiated into two major categories: (i) cost decline for individual components and (ii) cost decline for the assembly of components into the final product. Novel components such as semiconductors, which require new materials and innovative engineering, can thereby reach faster doublings of their cumulative production than components, which are mature and produced on a large scale for many decades. Opportunities for cost decline multiply in the former case due to potentials for technological progress and economies of scale (Ferioli et al., 2009). In the case of technologies, which consist of such novel and innovative components, both component-learning and learning in assembly contribute to declining production costs. For technologies, which consist of mature components, component-learning contributes little to the overall price and cost decline, while technological learning in the assembly of components is of major importance for decreasing overall production costs. For technology-specific cost forecasting, the assumption of different learning rates for individual cost components (costs for technical components or assembly) implies the possibility of declining learning rates with

12 Incorporating two-factor learning curves in bottom-up energy systems models allows, e.g., to project the optimal allocation of R&D budgets across a set of competing energy technologies, thus enabling a more adequate representation of the energy innovation process (see, e.g., Barreto and Kypreos (2004)).

157 Chapter 7 progressing maturity of technologies. Although such dynamics were already proposed by Argote (1999) and Kouvaritakis et al. (2000), empirical evidence remains weak.

Whether or not more detailed approaches such as multiple-factor experience curves and component learning can be applied in reality depends on the availability of data. Based on our own experience, we argue here that data at a sufficient level of detail and disaggregation are often non-existent for long enough time periods or are confidential and thus unavailable for public research. More detailed methods almost inevitably require additional assumptions on, e.g., cumulative production of components, which might add additional uncertainty into experience curve estimates. Although data problems restrict the application of more detailed experience curve approaches for public research, these tools might gain importance for internal production planning of manufacturers. Furthermore, we argue here that methodological extensions of the experience curve approach might be powerful in explaining historic cost dynamics of technologies. In view of practical data limitations and volatility of cost components, it remains, however, doubtful whether these extensions will yield more reliable cost projections.

7.4.3 Discussion of results Energy demand technologies show learning rates of 18 ± 9% that are approximately normally distribution around the mean. This finding is in line with results for energy supply technologies as well as for manufacturing in general. Our results thereby substantiate the finding of Ferioli et al. (2009), who identified a Normal distribution of learning rates based on the data presented by Dutton and Thomas (1984) but stated that more studies should be conducted before drawing rigorous conclusions about the distribution of learning rates.

We find that between-technology differences are often insignificant given the substantial within-technology variation of learning rates. As a rough approximation, policy makers and researchers might hence assume a learning rate of 18 ± 9% to derive cost projections for energy demand technologies. However, for many cases, such an assumption is too rough. Assuming a standard deviation of 9%-points could result in substantial under- or overestimation of production costs, if projections exceed three or four doublings of cumulative production. Here, our results provide some scope for identifying more accurately estimates of learning rates for individual technologies and technology clusters. These estimates can be used for technologies, which are novel and not yet commercially available at the market or for which detailed experience curve analyses are not feasible for other reasons.

We demonstrate now at the example of micro-cogeneration of heat and power (micro-CHP) in the Netherlands to what extent the experience curve approach can be useful to derive cost projections. Micro-CHP technology is currently being tested in several pilot projects and might be introduced into the Dutch heating market within the next couple of years. Micro-CHP technology could potentially replace conventional condensing gas combi boilers in the Netherlands on a large scale within the next two decades. SMP (2007) anticipates that micro-CHP systems will be 4,500 EUR more expensive than conventional condensing gas combi boilers (for which we assume a market

158 A review of experience curve analyses for energy demand technologies price of 1,250 EUR in 2009) at the point of market commercialization. We develop two sets of experience curves: (i) based on an aggregate learning rate of 18 ± 9% for the entire micro-CHP system and (ii) based on component learning by assuming a learning rate of 14 ± 1% for condensing gas boilers (Weiss et al., 2009b) and 18 ± 9% for the sterling engine and its integration into the boiler system (Figure 7.8)13.

7500 2010 5000

2500 2000

1500

1000

750 Accebtable price to reach projected 500 market diffusion LR = 18 ± 9%

Market price of micro-CHP in EUR Component learning: Condensing gas combi boiler: LR = 14 ± 1% 250 Sterling engine: LR = 18 ± 9% 2030

104 105 106 Cumulative micro-CHP sales in the Netherlands Figure 7.8: Projected price decline for micro-CHP systems in the Netherlands; numbers in the diagram indicate the year of analysis

The area between the experience curve and the line for the acceptable price represents learning investments and can be regarded as an estimate for subsidy requirements. The component learning approach allows us to narrow down uncertainty intervals of learning rates. However, both the component learning approach and the conventional experience curve approach yield substantial uncertainty intervals, if used to estimate break-even sales and subsidy requirements for micro-CHP systems (Table 7.1)14.

These results highlight the general problem with projections based on experience curve results: already small deviations in the learning rate lead to huge differences in estimated learning investments and subsidy requirements for novel technologies (see also: McDonald and Schrattenholzer (2001) and van der Zwaan and Seebregts (2004)).

13 The micro-CHP technology in the Netherlands can be regarded simplistically as a condensing gas combi boiler with an integrated sterling engine. For more details on the Dutch micro-CHP technology, we refer the reader to Weiss et al. (2008c) and Faber et al. (2009). 14 In the case of slow learning, subsidy requirements are extremely high. This result justifies more in-depth analysis. Note that our estimates are sensitive to changes in the price of natural gas and electricity as well as to changes in the heat demand of households.

159 Chapter 7

Table 7.1: Break-even sales and subsidy requirements for micro-CHP systems in the Netherlands Assumed learning Break-even sales Cumulative subsidy Decline of market price rate in % in 1000 units requirements in MEUR Learning of energy demand technologies: LR = 18 ± 9% Fast 27 15 13 Average 18 41 20 Slow 9 1,357 527 Component learning: LR = 14 ± 1% for condensing gas combi boiler; LR = 18 ± 9% for sterling engine Fast 15 and 27 79 27 Average 14 and 18 318 36 Slow 13 and 9 7,118 2,217

The time period in which subsidies need to be paid depends on both learning rates and market diffusion of the micro-CHP technology and might span periods from three years up to several decades. However, measuring the decline of real production costs over the first few doublings of cumulative production may yield new empirical data (while subsidy requirements are still rather modest), and may improve the accuracy of further projections.

Our example shows that reliable estimates for learning rates are one necessity for trustworthy technology forecasting and efficient energy policy. For this purpose, experience curve estimates can be improved by: (i) providing error margins to indicate uncertainty intervals of results (van Sark, 2007; Nemet, 2009) (ii) extending the empirical data basis (Söderholm and Sundquist, 2007) (iii) supplementing experience curve studies with bottom-up engineering analysis, technology assessment, cost-foresight methods, and stakeholder interviews to explain and cross-check the observed cost and price dynamics with qualitative information (Neij, 2008; Weiss et al., 2009b, c) (iv) conducting in-depth analysis on component learning and multiple-factor experience curves as well as developing more sophisticated models for reproducing the process of technological learning (e.g., Jamasb, 2007; Ferioli et al., 2009) (v) correcting price and cost estimates for changes in factor prices based on disaggregated cost indices15

Whereas the first point can be easily addressed by any thorough experience curve analysis, achieving the other objectives is difficult because necessary data might be non- existent or confidential. Here, close cooperation between science, policy, and industry might allow extending the empirical basis of experience curve analyses. Regardless, improving data availability and correcting for factor prices can only reduce uncertainty of experience curve forecasts to some extent. Future cost dynamics still might be highly uncertain as products mature and volatility in the price of production factor might become increasingly important for production costs. Correcting cost data for changes in factor

15 Hamilton et al. (2009) provide one example by stating indices for, e.g., capital costs for power supply, steel prices, and chemical plant costs in their cost analysis of pulverized coal power plants.

160 A review of experience curve analyses for energy demand technologies prices can help to assure comparability of technologies in cases where cost structures are similar. Furthermore, additional in-depth bottom-up analysis is necessary to better explain learning rates of individual technologies as well as to interpret the overall distributions of learning rates. Such explanation and interpretation can provide important insights for estimating more reliable ex-ante learning rates of novel technologies.

Already to date many efficient energy demand technologies offer relatively inexpensive opportunities for energy and greenhouse gas emission savings (IEA, 2008c, d). Hence, required investments to reach the price and cost break-even point with incumbent technologies are comparatively low. Purchasing decisions for energy demand technologies depend, however, to a greater extent on subjective consumer preferences than it is the case for energy supply technologies. This requires caution, in particular, from policy makers because non-monetary product features such as product design, functionality, and compliance with household infrastructure often present important purchasing criteria next to technology-related costs.

Our analysis has shown that technological learning occurs for a wide range of energy demand technologies. This finding has ambivalent implications for energy consumption in general. On the one hand, novel and efficient technologies become cheaper and can potentially replace inefficient and outdated equipment. On the other hand, also production costs and market prices of new products, which offer additional services and enhanced consumer satisfaction decline (e.g., consumer electronics and small appliances). Such a development might support the market diffusion of new products, thereby inducing additional energy demand, which potentially compensates for energy efficiency improvements elsewhere.

7.5 Conclusions In this chapter, we provide a comprehensive review of experience curve studies for energy demand technologies. We find a widespread development towards price and cost decline. Energy demand technologies become cheaper at an average learning rate of 18 ± 9%, thereby showing rates of cost decline similar to energy supply technologies. The learning rates for energy demand technologies are approximately normally distributed; the arithmetic mean provides therefore a good representation of the data sample. Within- technology deviations of learning rates are often as large as between-technology deviations. We nevertheless identify significantly higher learning rates for, e.g., consumer electronics and components thereof compared to several household appliances or relative to building insulation and window glazing. Our results provide scope to devise generic ex- ante learning rates for individual energy demand technologies and technology clusters. However, learning rates tend to show time dependency and variability depending on the system boundary chosen for analysis. This limits the applicability of the experience curve approach for modeling technological change in energy and emission scenarios. A major challenge of experience curve analysis is to isolate relevant parameters for quantifying their effects on the observed overall cost decline. Such detailed analysis remains, in particular, difficult because data are often non-existent or only available via closer

161 Chapter 7 industry-science cooperation. We argue that more detailed experience curve studies can provide additional insight into the drivers of cost decline and the potentials for policy intervention. The contribution of more detailed experience curve analysis to more reliable cost forecasts is nevertheless limited due to volatility of factor prices.

Acknowledgements This research was funded by the European Commission under the 7th framework program on ‘Environment’; ENV.2008.3.3.2.1: PROSUITE - Sustainability Assessment of Technologies, grant agreement number 227078. We thank Alexandra Newman (Colorado School of Mines, Golden, USA) and the two anonymous reviewers for their valuable comments on earlier drafts of this chapter.

162

Appendix Table A7.1: Overview of experience curve studies for energy demand technologies; subscript numbers indicate the base year of currency deflation Technology Time LR Errora Reference Technology Dependent variable Independent variable Country R2 Nb cluster period [%] [%] Cunningham (1980) Ford, model T price cum. production [units] USA 1910-1926 14 n. s. n. s. 10 c Automotive Lipman and Sperling (1980) Ford, model T price [USD1958/car] cum. production [units] USA 1909-1918 15 n. s. 0.96 ~9 d Laitner and Sanstad (2004) Ford, model T costs [USD1978/car] cum. production [units] USA 1909-1923 12 1 0.97 9.0

163 Laitner and Sanstad [(2004)d Selective window coatings prod. costs cum. production [m2] n. s. 1992-2000 20 4 0.93 5.0 f Building Jakob and Madlener (2004) Building façades insulation costs [CHF/kWhconserved) cum. energy conserved [GWh] CH 1975-2001 15 n. s. n. s. ~3.3 g insulation Jakob and Madlener (2004) Building façades insulation costs [CHF/kWhconserved) cum. energy conserved [GWh] CH 1975-2001 18 n. s. n. s. ~3.2 and glazing f 2 Jakob and Madlener (2004) Building façades insulation costs [CHF/kWhconserved) cum. façade area [m ] CH 1975-2001 17 n. s. n. s. ~2.5 g 2 Jakob and Madlener (2004) Building façades insulation costs [CHF/kWhconserved) cum. façade area [m ] CH 1975-2001 21 n. s. n. s. ~2.5

Residential Martinus et al. (2005) Heat pumps investment costs [EUR2000/kWth] inst. German capacity [MWth] NL 1980-2002 30 n. s. n. s. ~1.4

heat pumps Weiss et al. (2008c) Heat pumps price [EUR2006/kWth] cum. Swiss sales [MWth] CH 1980-2004 35 1 0.99 3.5

Martinus et al. (2005) Condensing gas boilers inv. costs [EUR2000/kWth] cum. German capacity [MWth] GER 1992-1999 4 n. s. n. s. ~3.6 Martinus et al. (2005) Condensing gas boilers inv. costs [EUR /kW ] cum. Dutch sales [MW ] NL 1983-1997 4 n. s. n. s. ~5 Other 2000 th th residential Weiss et al. (2009b) Condensing gas space heating boilers price [EUR2006/kWth] cum. Dutch sales [MWth] NL 1983-2006 6 1 0.92 6.8

heating Weiss et al. (2009b) Condensing gas combi boilers price [EUR2006/kWth] cum. Dutch sales [MWth] NL 1988-2006 14 1 0.98 5.0 technologies Newell (2000) Gas water heaters price cum. US shipments [units] USA 1962-1993 25 2 0.87 n. s. d Laitner (2002) Residential electrical water heaters costs [USD1995/unit] cum. US shipments [units] USA 1982-1995 5 2 0.91 4.1 Bass (1980) Room air conditioners price cum. US industry sales [units] USA 1946-1961 8 1 0.89 n. s. Bass (1980) Room air conditioners price cum. US industry sales [units] USA 1946-1974 12 2 0.87 n. s. Air Akisawa (2000) Air conditioners sales price [Yen/unit] cum. sales [units] JPN 1972-1997 10 n. s. 0.82 n. s. conditioners Newell (2000) Room air conditioners price cum. US shipments [units] USA 1958-1993 23 1 0.95 n. s. Newell (2000) Central air conditioners price cum. US shipments [units] USA 1967-1988 24 2 0.83 n. s. Laitner and Sanstad (2004) Room air conditioners unit costs cum. US shipments [units] USA 1980-1998 15e n. s. n. a. 4.7 e Washing Laitner and Sanstad (2004) Washing machines unit costs cum. US shipments [units] USA 1980-1998 13 n. s. n. a. 4.6

machines Weiss et al. (2009c) Washing machines price [EUR2006/kg l. c.] cum. global production [units] NL 1965-2008 33 9 0.56 2.5

164

Weiss et al. (2009c) Laundry dryers price [EUR2006/kg l. c.] cum. global production [units] NL 1969-2003 28 7 0.80 2.3 Bass (1980) Dishwashers price cum. US industry sales [units] USA 1947-1968 10 2 0.75 n. s. Bass (1980) Dishwashers price cum. US industry sales [units] USA 1947-1974 11 2 0.85 n. s. Dishwashers Laitner and Sanstad (2004) Dishwashers unit costs cum. US shipments [units] USA 1980-1998 16e n. s. n. a. 4.7

Weiss et al. (2009c) Dishwashers price [EUR2006/SPS] cum. global production [units] NL 1968-2007 27 7 0.82 4.7 Bass (1980) Refrigerators price cum. US industry sales [units] USA 1922-1940 7 1 0.83 n. s. Refrigerators Laitner and Sanstad (2004) Refrigerators unit costs cum. US shipments [units] USA 1980-1998 12 e n. s. n. a. 4.6

Weiss et al. (2009c) Refrigerators price [EUR2006/hl] cum. global production [units] NL 1964-2008 9 4 0.43 5.7 Laitner and Sanstad (2004) Freezers unit costs cum. US shipments [units] USA 1980-1998 22 e n. s. n. a. 3.9

Freezers Weiss et al. (2009c) Upright Freezers price [EUR2006/hl] cum. global production [units] NL 1970-2003 10 5 0.59 4.8

Weiss et al. (2009c) Chest Freezers price [EUR2006/hl] cum. global production [units] NL 1970-1998 8 2 0.87 4.4

Iwafune (2000) Modular-electronic CFLs price [USD1995/klm] cum. global production [units] USA 1992-1998 20 n. s. 0.56 ~2.3

Iwafune (2000) Integral-electronic CLFs price [USD1995/km] cum. global production [units] USA 1992-1998 16 n. s. 0.66 ~2.8

Iwafune (2000) Modular-magnetic CFLs price [USD1995/klm] cum. global production [units] USA 1992-1998 41 n. s. 0.90 ~2.3 Compact h Iwafune (2000) CFLs price [USD1995/km] cum. global production [units] USA 1992-1998 21 ~6 0.90 ~2.3 fluorescent light bulbs Ellis (2007) CFLs price [USD2004/unit] cum. global sales [units] USA 1990-2004 10 n. s. n. s. ~4.2

Weiss et al. (2008c) CFLs price [EUR2006/klm] cum. global sales [Glm] GER, NL 1988-2006 19 4 0.91 6.2

Weiss et al. (2008c) CFLs Price [EUR2006/We] cum. global sales [MWe] GER, NL 1988-2006 19 5 0.89 5.9

Weiss et al. (2008c) CFLs Price [EUR2007] cum. global sales [units] INT 1985-2007 21 5 0.71 8.5

Iwafune (2000) Magnetic ballasts for CFLs price [USD1995/unit] cum. US production [units] USA 1981-1988 16 n. s. 0.80 ~3.9

Iwafune (2000) Magnetic ballasts for CFLs price [USD1995/unit] cum. US production [units] USA 1990-1993 41 n. s. 0.97 ~0.3 d Lamp Laitner and Sanstad (2004) Magnetic ballasts for FLs prod. costs [USD1993/unit] cum. shipments [units] USA 1977-1993 3 3 0.26 4.1

ballasts Duke and Kammen (1999) Electronic ballasts for FLs price [USD1997/unit] cum. prodcion [units] USA 1986-1997 11 2 n. s. ~8.9

Iwafune (2000) Electronic ballasts for CFLs price [USD1995/unit] cum. US production [units] USA 1986-1998 13 n. s. 0.98 ~7.5 d Laitner and Sanstad (2004) Electronic ballasts for FLs prod. costs [USD1996/unit] cum. shipments [units] USA 1986-2001 11 2 0.91 9.7

Table A7.1 (cont.): Overview of experience curve studies for energy demand technologies; subscript numbers indicate the base year of currency deflation Technology Time LR Errora Reference Technology Dependent variable Independent variable Country R2 Nb cluster period [%] [%] Television Bass (1980) Black-and-white TV price cum. US industry sales [units] USA 1948-1960 13 4 0.78 n. s. sets (cathode Bass (1980) Black-and-white TV price cum. US industry sales [units] USA 1948-1974 22 5 0.73 n. s. ray tube) Bass (1980) Color TV price cum. US industry sales [units] USA 1961-1971 5 1 0.88 n. s. Bass (1980) Color TV price Cum. US industry sales[units] USA 1961-1974 7 2 0.78 n. s. Maycock and Wakefield (1975) 4-function pocket calculators price [USD/unit] cum. production [units] USA early 1970s 30 n. s. n. s. n. s. Other Cunningham (1980) Digital watches av. factory price [const. USD/unit] cum. production [units] USA 1975-1978 26 n. s. n. s. 4 consumer electronics Cunningham (1980) Hand-held calculators av. factory price [const. USD/unit] cum. production [units] USA 1975-1978 26 n. s. n. s. 2 c c

165 Lipman and Sperling (2000) Sony laser diodes prod. costs [Yen/unit] cum. production [units] prod. by Sony 1982-1994 23 n. s. 0.95 ~17 BCG (1972) Integrated circuits av. price [const. USD/unit] cum. industry experience [units] USA 1964-1972 25 n. s. n. s. ~10 Cunningham (1980) Integrated circuits av. price [const. USD/unit] cum. production [units] USA 1964-1972 28 n. s. n. s. 10 d Laitner and Sanstad (2004) Integrated circuits prod. costs [USD1993/unit] cum. production [USD] USA 1962-1968 26 5 0.95 7.7 Cunningham (1980) MOS/LSI av. price [const. USD/unit] cum. production [units] USA 1970-1976 20 n. s. n. s. 10 Cunningham (1980) MOS dynamic RAM av. factory price [const. USD/unit] cum. number [bits] USA 1973-1978 32 n. s. n. s. 6 Cunningham (1980) Disk memory drives av. price [const. USD/bit] cum. number [bits] USA 1975-1978 24 n. s. n. s. 3 Irwin and Klenow (1994) 4 kB - DRAM price [USD/MB] cum. global shipments [units] E, JPN, USA ~1974-1986 20 2 0.91 n. s. Electronics Irwin and Klenow (1994) 16 kB - DRAM price [USD/MB] cum. global shipments [units] E, JPN, USA ~1977-1922 24 2 0.92 n. s. components Irwin and Klenow (1994) 16 kB-5 - DRAM price [USD/MB] cum. global shipments [units] E, JPN, USA ~1977-1992 18 1 0.95 n. s. E, JPN, ROK, Irwin and Klenow (1994) 64 kB - DRAM price [USD/MB] cum. global shipments [units] ~1978-1992 23 1 0.97 n. s. USA E, JPN, ROK, Irwin and Klenow (1994) 256 kB - DRAM price [USD/MB] cum. global shipments [units] ~1982-1992 21 2 0.93 n. s. USA E, JPN, ROK, Irwin and Klenow (1994) 1 MB - DRAM price [USD/MB] cum. global shipments [units] ~1985-1992 16 2 0.86 n. s. USA E, JPN, ROK, Irwin and Klenow (1994) 4 MB - DRAM price [USD/MB] cum. global shipments [units] ~1988-1992 20 2 0.97 n. s. USA JPN, ROK, Irwin and Klenow (1994) 16 MB - DRAM price [USD/MB] cum. global shipments [units] n. s. 16 2 0.98 n. s. USA a representing the 95% confidence interval of learning rates b N - number of doublings of cumulative production c information obtained from McDonald and Schrattenholzer (2001) d experience parameters recalculated based on original data of Laitner and Sanstad (2004) e referring to the cost difference between base year and final year of analysis f building facades insulation of 1.0 W/m2K g building facades insulation of 1.25 W/m2K h weighted average of modular and integral CFLs

Chapter 7 166 Table A7.2: Probability that learning rates for individual energy demand technologies are identical; based on two-tailed t-test, significant differences at a 95% confidence level are marked bold numbers

Building insulation and window glazing Residential heat pumps Other residential technologies heating Air conditioners machinesWashing Laundry dryers Dishwashers Refrigerators Freezers Compact fluorescent light bulbs Lamp ballasts and (magnetic electronic) sets Television (cathodetube) ray Other consumer electronics components Electronics Automotive Automotive (Ford, model T) Automotive (Ford, model T) 1.00 Building insulation and window glazing 0.03 1.00 Residential heat pumps 0.00 0.00 1.00 Other residential heating technologies 0.30 0.06 0.00 1.00 Air conditioners 0.59 0.37 0.01 0.23 1.00 Washing machines 0.52 0.72 0.51 0.39 0.58 1.00 Laundry dryers 0.97 0.31 0.01 0.44 0.75 0.52 1.00 Dishwashers 0.55 0.65 0.02 0.11 0.85 0.62 0.66 1.00 Refrigerators 0.08 0.01 0.02 0.93 0.10 0.40 0.32 0.17 1.00 Freezers 0.95 0.38 0.03 0.54 0.72 0.50 0.94 0.64 0.46 1.00 CFLs 0.06 0.44 0.03 0.03 0.21 0.87 0.18 0.38 0.01 0.23 1.00 Lamp ballasts (magnetic and electronic) 0.70 0.68 0.03 0.36 0.94 0.60 0.76 0.95 0.29 0.73 0.44 1.00 Television sets (cathode ray tube) 0.66 0.19 0.01 0.70 0.48 0.45 0.71 0.44 0.59 0.80 0.11 0.55 1.00 Other consumer electronics 0.00 0.00 0.18 0.00 0.01 0.80 0.03 0.07 0.00 0.09 0.15 0.11 0.03 1.00 Electronic components 0.00 0.02 0.10 0.01 0.05 0.96 0.08 0.21 0.00 0.17 0.67 0.28 0.06 0.11 1.00

8 Discussion

8.1 Scope of this thesis The manufacturing industry plays a dual role in the context of energy use and GHG emissions, namely (i) as consumer of energy and (ii) as producer of energy consuming equipment and technologies. Focusing on the manufacturing industry as energy consumer, still knowledge gaps exist concerning the accuracy of energy statistics and GHG emission inventories. In particular, uncertainties refer to the accounting of fossil fuels, which are used for non-energy purposes (Olivier and Peters, 2002; UBA, 2003; Patel et al., 2005). In the first part of this thesis, we address these uncertainties by improving the accuracy and completeness of estimates on non-energy use and related CO2 emissions at national, regional, and global scale.

In the second part of this thesis, we focus on the manufacturing industry as producer of energy consuming technologies. Here, prevailing knowledge gaps exist regarding technological learning and the rate at which cost of energy demand technologies decline. To address existing knowledge gaps, we study technological learning of condensing gas boilers and large appliances, and we provide a review of technological learning for a wider range of energy demand technologies.

In this chapter, we discuss our principle findings, we examine the strengths and limitations of our research, and we derive final conclusions and recommendations. We first focus on the non-energy use of fossil fuels; afterwards we discuss technological learning of energy demand technologies.

8.2 Non-energy use of fossil fuels 8.2.1 Principal findings To monitor energy efficiency and GHG emissions, reliable energy statistics are of key importance. In Chapter 2, we recalculate non-energy use as given by international energy statistics for the world as a whole and for the 50 countries with the highest consumption of fossil fuels for non-energy purposes. We quantify worldwide non-energy use in the year 2000 to be 20 ± 2 EJ, thereby accounting for 5% of the global total primary energy supply. In the case of 12 countries, our estimates, assuming a 95% confidence interval, are in line with data from international energy statistics (IEA, 2005a, b). Our estimates are lower than data from international energy statistics in the case of 26 countries and for the world as a whole. For another 12 countries, our estimates exceed official data of international energy statistics. We find that the inclusion of feedstock shares that are used for energy as well as errors in energy statistics provide one explanation for deviations between our estimates and IEA (2005a, b) data.

167 Chapter 8

For several countries (e.g., the USA, China, France, Indonesia, and Poland), non- energy use data in international energy statistics most likely include fuel use of feedstock, which indicates a lack of compliance with the accounting guidelines given by the International Energy Agency (IEA, 2006). For these countries, the system boundaries of non-energy use data should urgently be clarified. Furthermore, the Ukraine, Venezuela, Kuwait, and Colombia are likely to report incomplete values with regard to feedstock use in the chemical industry.

In Chapter 3, we estimate that non-energy use in 2000 leads to 700 ± 90 Mt CO2 emissions on a global scale. Annex I countries account for 51% (360 ± 50 Mt CO2) and non-Annex I countries for 49% (340 ± 70 Mt CO2) of this total. Around 75% of global non-energy use emissions are related to industrial processes; the remainder is attributed to product use, agriculture, and waste management.

The case study in Chapter 4 substantiates our findings from Chapter 2 by showing that non-energy use is not defined consistently in German energy statistics. These include the fuel use of feedstock in the non-energy use of coal, lignite, and oil products, whereas they exclude fuel use of feedstock in the case of natural gas. Yearly non-energy use emissions in Germany increased from 22.0 ± 1.8 Mt CO2 in 1990 to 27.7 ± 2.3 Mt CO2 in 2003. Our analysis allowed closing major gaps in the German GHG inventory with regard to emissions from methanol and carbon black production as well as from chemical conversion losses. Our case study for Germany points to remaining data gaps and uncertainties within the German GHG inventory. Emissions from steam cracking as well as parts of emissions from non-ferrous metals production and product use are still not included in the German GHG inventory. The comparison between NEAT and the German IPCC-SA (UNFCCC, 2007a) furthermore reveals that the applied IPCC default emission factor for ammonia production is incorrect given feedstock distribution, system boundaries of non-energy use, and CO2 sequestration for urea production.

8.2.2 Strengths and limitations of our research Strength of our research The proper accounting of non-energy use and related emissions is complicated by the complex inter-linkages of material and energy flows within the chemical industry and requires detailed and country-specific insights into production routes and use patterns of products. Two issues which are particularly problematic refer to (i) the allocation of feedstock, which is used for fuel, to either energy or non-energy use and (ii) the complete accounting of non-energy use emissions along all life cycle stages of products, including production, product use, and waste disposal. Our research improves the accounting of non- energy use and related emissions by developing and applying bottom-up models, which function independently from data of energy statistics.

168 Discussion

The simple bottom-up model presented in Chapter 2 is capable of reproducing non- energy use without the need of conducting in-depth analyses. The model results are useful for identifying major errors as well as inconsistencies in the system boundaries of non- energy use as stated in energy statistics. Experience shows that such analysis can serve also as starting point for correcting data on fuel use in energy statistics (Neelis and Pouwelse, 2008).

The extended bottom-up model presented in Chapter 3 is useful (i) for providing first estimates of non-energy use emissions in non-Annex I countries, which have not yet established complete GHG inventories and (ii) for crosschecking non-energy use emission data in GHG inventories of Annex I countries. Our model calculations can thereby contribute important building blocks for establishing reliable GHG emission inventories in both non-Annex I and Annex I countries.

In Chapter 4, we conduct an in-depth analysis of non-energy use and related emissions by applying the detailed Non-energy use Emission Accounting Tables (NEAT) model in a case study for Germany. With NEAT, we estimate non-energy use emissions in more detail than required by the IPCC (2006) guidelines, thus providing a more complete accounting of relevant source categories. Our bottom-up analysis reveals gaps and inconsistencies in the German GHG inventory. The Environmental Agency of Germany (Umweltbundesamt, UBA) used parts of our emission estimates to improve the completeness of the German GHG inventory.

A general point of uncertainty in GHG inventories are estimates on product use emissions. We make an important contribution to the correct accounting of product use emissions by providing disaggregate estimates for the most important source categories (i.e., the use of solvents, lubricants, waxes, and paraffins) for Germany as well as for major developing countries (see Chapters 3 and 4). This way, our research enables a more accurate and complete reporting of non-energy use emissions. Nevertheless, our model approaches are subject to uncertainties, which we discuss next.

Limitations of our research Limitations of our approaches and results refer in particular to the accounting of non- energy use of refinery products in Chapters 2 and 3 and the application of regression analysis and default emission factors for estimating parts of non-energy use emissions in Chapters 3 and 4. In the following, we discuss these principal sources of uncertainty in greater detail. (i) Due to limited data availability, we approximate in Chapters 2 and 3 the consumption of refinery products by domestic production capacities. This simplification is mainly problematic for small countries, where net trade of refinery products can be substantial relative to domestic production (e.g., in the case of Algeria, Austria, and Kuwait). However, for most countries the net trade of refinery products is small. Moreover, since feedstock use typically represents the far more important component of non-energy use, the use of refinery capacities as proxy for domestic consumption of refinery products results overall only in minor uncertainties.

169 Chapter 8

(ii) In Chapters 2 and 3, we estimate the consumption of electrodes and other solid carbon based on regression analysis as a function of aluminium production and GDP. In addition, we estimate emissions from the use of solvents, waxes, and paraffins, as well as from the oxidation of surfactants during wastewater treatment by regression analysis as function of GDP. Overall, we estimate approximately 20% of non-energy use emissions in Chapter 3 based on regression analysis. These estimates are subject to uncertainties in the order of around 25%; hence adding an overall uncertainty of 5% to our non-energy use emission estimates. (iii) The bottom-up model presented in Chapter 3 does not allow for estimating emissions that result from carbon losses in chemical conversions processes. This shortcoming leads to a systematic underestimation of non-energy use emissions of around 10% for countries with a large chemical industry. (iv) In our in-depth analysis with NEAT (Chapter 4), we estimate emissions from carbon losses in chemical conversion processes based on process-specific default emission factors as given by Neelis et al. (2007). More analysis is necessary to assure accuracy of these emission factors, in particular to avoid potential double counting of emissions in cases where carbon losses occur in the form of low value fuel grade by-products. (v) Our NEAT estimates for emissions from the use of lubricants, waxes, and paraffins are based on default emission factors. Company surveys and detailed bottom-up studies on the use and fate of these products could substantially reduce the uncertainty intervals of our estimates. The use of lubricants, waxes, and paraffins accounts approximately for 8% of non- energy use emissions and is subject to uncertainties of approximately 40%; thereby adding 3% uncertainty to our total non-energy use emission estimates for Germany. (vi) A final and more general source of uncertainty refers to the quality of official production statistics. In our case study for Germany, we identify in several cases errors in chemicals production data as given by the German statistics office (Destatis, 1990-2003a). This source of uncertainty can be reduced to some extent by crosschecking official data with information provided from other sources.

We conclude that although our estimates are subject to uncertainties, these affect only limited parts of non-energy use and related emissions. Our research thus allows for more detailed, complete, and accurate accounting of non-energy use emissions in comparison to existing approaches.

170 Discussion

8.2.3 Conclusions and recommendations Our research (i) identifies inconsistencies and potential errors in international and national energy statistics, (ii) provides first bottom-up estimates of non-energy use emissions in major developing countries, and (iii) improves the accuracy and completeness of non-energy use emission estimates in the German GHG inventory. Our results indicate several areas where the accounting of non-energy use and related emissions can be substantially improved. To address these, non-energy use, its conversion, and fate along the entire life cycle of products needs to receive more widespread attention.

Immediate action is required to ensure that countries comply with the reporting instructions of the International Energy Agency (IEA, 2006) and uniformly exclude fuel use of feedstock from non-energy use in their energy statistics. Such an approach would (i) allow international comparisons of non-energy use data as well as (ii) simplify the calculation of non-energy use emissions for the various relevant source categories. Second, to assure that countries follow the reporting instructions of the IEA requires that IEA questionnaires provide more detailed guidance on how to handle the complex feedstock and energy conversions within the chemical industry. In particular, it is important to specify, how energy statisticians should differentiate feedstock used for fuel and for non-energy purposes in the most important chemical processes, i.e., steam cracking and the production of ammonia, methanol, and carbon black. The IEA should further clarify the position of chemical grade refinery aromatics in energy statistics. This requires in particular specifying whether to regard these items as other petroleum products or as chemicals and whether to include them under feedstock use or under non-energy use of refinery products.

In this thesis, we have pinpointed countries for which non-energy use data as reported in international energy statistics should be urgently rechecked. However, experience in the past 10 years has shown that the harmonization and correction of national and international energy statistics is a very slow and tedious process. To improve the quality of non-energy use estimates, more attention and expertise is necessary.

Next to energy statistics, the accuracy of official production statistics requires improvement to allow accurate monitoring of GHG emissions and energy efficiency improvements. Current GHG emission inventories are particularly uncertain with regard to product use emissions, for which the improved IPCC (2006) guidelines request more detailed calculations. In the case of Germany, we can recommend to complement current emission estimates for solvent use by our emission estimates for other important sources of product use emissions, i.e., the consumption of lubricants, waxes, and paraffins. We furthermore suggest using our estimates on emissions from wastewater treatment as a benchmark for further and more detailed analysis.

We summarize our discussion on the accounting of non-energy use and related emissions with the following recommendations for policy makers, energy statisticians, and inventory experts: (i) The quality of energy statistics and production statistics needs to be substantially improved to warrant accurate emissions accounting in

171 Chapter 8

international and national GHG emission inventories. Considerably more attention, staffing, and expertise are required in order to achieve this aim. (ii) National and international energy statistics should assure a correct and internationally harmonized accounting of non-energy use to enable, in particular, international data comparability. Achieving this aim requires commitment and better guidance in energy questionnaires. (iii) The processes of preparing energy statistics and GHG inventories should be linked more closely together to assure that energy statistics fulfil requirements for accurate emissions accounting. (iv) Even though the IPCC (2006) guidelines exclude non-energy use emissions from the IPCC-RA, we recommend continuing to use the IPCC-RA according to the 1996 IPCC guidelines as macro-check for non-energy use emissions, which are estimated according to the IPCC-SA. (v) Inventory experts should pay particular attention to the complete and correct accounting of emissions from (i) conversion losses in chemical processes and (ii) product use because estimates for these source categories are subject to major uncertainties. (vi) To assist in emissions accounting, our bottom-up models provide means to crosscheck, add, and improve estimates of non-energy use and related emissions at a very detailed level. We recommend using our models and results to inventory experts around the world.

8.3 Technological learning in energy demand technologies 8.3.1 Principal findings Our research provides new insights into technological leaning of energy demand technologies. We find (i) that the experience curve approach is in general applicable to study long-term technological learning of condensing gas boilers and large appliances and (ii) that a large variety of energy demand technologies shows a robust trend towards price and cost decline. We quantify this price and cost decline as learning rate, which represents the rate of price and cost decline with each doubling of cumulative production.

In Chapter 5, we estimate that the price of condensing gas combi boilers declines at a learning rate of 14 ± 1%. Similarly, the additional price of condensing gas combi boilers, relative to the incumbent non-condensing gas boiler technology, declines at a learning rate of 16 ± 8%.

Likewise, we find a robust price decline for large appliances in Chapter 6. The specific price of wet appliances (washing machines, laundry dryers, and dish washers) declines at learning rates of on average 29 ± 8% and thereby much faster than the prices of cold appliances (refrigerators and freezers), which show an average learning rate of only 9 ± 4%.

172 Discussion

In Chapter 7, we review experience curve studies, which address 15 different groups of energy demand technologies. We find a widespread trend towards declining prices and costs at an average learning rate of 18 ± 9%. Learning rates for energy demand technologies show a wide and approximately symmetric distribution around this average. Data variation within individual technology clusters is often as large as data variation between technology clusters. These findings are consistent with results for energy supply technologies and for the manufacturing industry in general, indicating that technological learning is as important for energy demand technologies as it is for energy supply technologies.

We present an extension of the conventional experience curve approach in our analysis of large appliances. Here, we model specific energy consumption as function of cumulative production. We refer to the resulting experience curves as energy experience curves. We find a decline in the specific energy consumption of wet appliances (LR of 18 ± 3% to 35 ± 3%) and cold appliances (LR of 13 ± 3% to 17 ± 2%). Our results suggest that energy policy might be able to bend down the slope of energy experience curves, thereby accelerating energy efficiency improvements of large appliances. We justify the application of the experience curve approach to specific energy consumption of large appliances by arguing that specific energy consumption (and its inverse energy efficiency) might follow autonomous technological innovation and, furthermore, result from the quest of producers to decrease production costs. As an example, advances in insulation of cold appliances, might allow for smaller compressors; improving the tub shape of washing machines might reduce water consumption and thus material demand for pipes and pumps. Consequently, energy efficiency improvements and declining production costs may go hand in hand at the system level. Apart from these cost- driven energy efficiency improvements, manufacturers experienced over long periods only small incentives to improve the energy efficiency of their products. However, the enforcement of energy policy since the 1990s (e.g., via subsidies and energy labeling) made energy efficiency increasingly a product feature and introduced an additional incentive for producers to improve the efficiency of their products. Our analysis suggests that this effect of energy policy is to some extent visible by a steeper slope in energy experience curves of several large appliances. This finding indicates that energy experience curves can reveal insights into the effectiveness of energy policy.

Our experience curve analysis demonstrates the importance of technological learning for the decline of prices and costs of energy demand technologies. Due to this mechanism, novel and efficient but initially expensive technologies do not necessarily incur higher costs on consumers in the long-term. In a sample case, we analyze the effect of technological learning on investment and use phase costs of condensing gas boilers, which were initially expensive, but became cheaper during pervasive market diffusion. For this efficient energy demand technology, the net present value was negative at -330 EUR2006 (real Euro, deflated to the base year of 2006) in 1988 but increased over the years and reached 970 EUR in 2006. We attribute two thirds of the improvements in the cost-benefit performance of condensing gas combi boilers to technological learning and one third to a combination of external effects (e.g., heat demand in households) and governmental taxation policy. From a social perspective, we find that costs for emission

173 Chapter 8

savings achieved by condensing gas combi boilers decreased from 60 EUR2006/t CO2 in 1988 to -120 EUR2006/t CO2 in 2006. Our findings show how limited subsidy support of 70 ± 10 million EUR contributed to (i) a substantially improved cost-benefit performance of one particular energy-efficient technology, (ii) consumer cost savings of a multiple of the provided subsidies, and (iii) savings of approximately 270 PJ primary energy and 15 Mt CO2 emissions in the residential sector of the Netherlands between 1981 and 2006.

8.3.2 Strengths and limitations of our research Strengths of our research We demonstrate that technological learning is an important and widespread phenomenon in the manufacturing of energy demand technologies. For these technologies, we provide detailed learning rates, which we explain qualitatively based on bottom-up technology analysis. For the identified learning rates, we provide uncertainty intervals, which quantify the implicit error of the applied experience curve approach. Such practice has been largely neglected in previous experience curve analyses (van Sark, 2007) but presents an important step towards a more transparent quantification of technological learning.

Analyzing specific energy consumption of large appliances with the experience curve approach allows for quantifying technological learning from a broader perspective and reveals new insights into the dynamics of energy efficiency improvements. Our cost- benefit analysis for condensing gas boilers in the Netherlands demonstrates in a case study the importance of technological learning for improving the life cycle cost performance of one efficient energy demand technology.

Overall, our analysis contributes to a more reliable and detailed quantification of technological learning in energy demand technologies. However, our research is also subject to several uncertainties, which we discuss next.

Limitations of our research The experience curve approach inherits several problems related to methodology and data availability. First, we address methodological problems. The phenomenon that production costs decline at a constant rate with each doubling of cumulative production is an empirical observation but not a natural law (Dutton and Thomas, 1984). Although experience curves have demonstrated that cost decline is a general power-law function of cumulative production, cost decline is driven by complex and multi-level interactions of mechanisms, which are both endogenous and exogenous of the manufacturing process. A general problem of the experience curve approach refers to its inability to differentiate technological learning from singular effects such as changing prices for raw materials and energy and its failure to explain drivers for the observed cost dynamics on a sub-system level. In the case of wet and cold appliances, possible explanations for the observed deviation in learning rates might include differences regarding technical complexity and chances for outsourcing as well as disproportionate potentials for automation and innovation throughout the manufacturing chain. Nevertheless, we cannot explain the deviation in the learning rates of these two technology clusters in a straightforward and compelling manner. Such shortcomings introduce uncertainty into our experience curve results, which is particularly relevant if learning rates are used for long-term technology

174 Discussion forecasting. Furthermore, the shortcomings discussed so far limit possibilities to generate ex-ante technology-specific learning rates for novel technologies.

The approximation of technological learning by cumulative production may be in practice the best single proxy for the various drivers of cost decline, but the exclusion of other explanatory variables (such as scale effects or research and development spending) creates bias related to omitted variables.

The methodological shortcomings of the experience curve approach are closely linked to the limited availability of detailed empirical data, which represents the second major source of uncertainty. We follow in our research the common practice of approximating actual production costs by market prices. Prices are generally an imperfect measure of production costs because they deviate from the latter depending on profit margins of producers and markups in the wholesale and retail sector. These in turn depend on the life cycle stage of products, demand-supply driven price dynamics, or the degree of which markets are competitive (IEA, 2000; Junginger, 2005; Junginger et al., 2008). Markets for condensing gas boilers and large appliances have been highly competitive in the Netherlands, leaving only small profit margins to producers (Nefit, 2008). Therefore, we argue that uncertainties introduced into our analysis by using prices as proxy for actual production costs are limited.

Data availability determines also our choice to model prices of condensing gas boilers based on cumulative production in the Netherlands only. This choice introduces limited uncertainty until the early 1990s because condensing gas boilers were developed and produced in the Netherlands without considerable exogenous technology spillover from other countries (Remeha, 2007; Nefit, 2008). However, in subsequent years, condensing gas boilers were increasingly manufactured abroad, while mergers of boiler producers contributed to additional international knowledge transfer.

Finally, the problem of changing product functionality is a particular point of uncertainty for experience curve analyses on energy demand technologies. Purchasing decisions of consumers are typically based on the ability of products to provide multiple services (rather than simply producing energy in the least costly manner as it is the case for energy supply technologies). Producers have therefore strong incentives to achieve functionality improvements (e.g., introducing centrifuge drying into washing machines) and product differentiation, which potentially introduces non-learning related price changes and thus uncertainty into experience curve analyses. These uncertainties are to some extent reflected by the low coefficients of determination (i.e., R2 < 0.7) in our experience curve analysis of large appliances.

8.3.3 Conclusions and recommendations The experience curve approach is typically applied to three areas, i.e., for strategic planning in manufacturing (Dutton and Thomas, 1984; Argote and Epple, 1990), in energy-environment-economy models (see, e.g., Kahouli-Brahmi, 2008), and for energy policy making (Wene et al., 2000; Junginger et al., 2008). In these applications, experience curves are used to explore potentials for cost decline in the mid and long term as well as to

175 Chapter 8 estimate necessary investments or subsidy requirements to reach cost break-even in comparison with incumbent technologies. Our research demonstrates that energy demand technologies show substantial technological learning, which has wider implications for science and policy.

We summarize our discussion on technological learning of energy demand technologies with the following recommendations for scientists and policy makers: (i) It is important to report error margins of learning rates. This allows for sensitivity analysis in long-term model projections that are susceptible to small variations in learning rates. (ii) Despite a robust cost decline in past decades, caution is required when projecting future technology costs because prices for materials and energy might increase and potentially compensate for learning effects. (iii) Our results suggest that static approaches for estimating least life cycle costs of novel technologies overlook future potentials for the decline in up-front technology costs and specific energy consumption. While designing new energy policy to improve the efficiency of energy demand technologies (such as energy labeling programs), policy makers need to be aware of potentials for cost decline. (iv) Modelling energy consumption with energy experience curves reveals new insights into the dynamics of energy efficiency and should be analyzed further. We recommend identifying the potentials of such energy experience curves throughout all areas of energy demand in the manufacturing industry as well as in transport and the residential and commercial buildings sector. (v) Technological learning of energy demand technologies should be accounted for in energy modelling because this mechanism contributes to a substantial decline of investment and use phase costs for these technologies in the mid and long term.

Finally, we conclude that energy efficiency improvements are generally regarded as important mechanism towards a sustainable energy system. Technological learning helps realizing efficiency potentials in a less costly manner. However, to achieve an economy-wide decline of absolute energy consumption requires substantially more efforts and commitment than in the past and makes it in particular necessary to exploit also initially costly and economically less attractive energy efficiency potentials.

176

9 References

AGE (2007): Vollständige Energiebilanzen 1990–2003. AGE - Arbeitsgemeinschaft Energiebilanzen (Working Group Energy Balances). Berlin, Germany. Source: http://www.ag-energiebilanzen.de/viewpage.php?idpage=1. Accessed: 17 June 2009. AGPO (2007): Adviesprijzen leveringsprogramma 2005, 2006, 2007. AGPO Ferroli B.V. Breda, the Netherlands. AHAM (2004a): The story of the refrigerator. AHAM - Association of Home Appliance Manufacturers. Source: http://www.aham.org/consumer/ht/action/GetDocument Action/id/1409. Accessed: 12 June 2007 AHAM (2004b): Clothes washers: the end of blue Monday. AHAM - Association of Home Appliance Manufacturers. Source: http://www.aham.org/ht/a/Get DocumentAction/i/1407. Accessed: 12 June 2007. Akisawa, A. (2000): Technological development and market substitution of heat pumps – a case of room air-conditioners in Japan. Unpublished data, cited from McDonald and Schrattenholzer (2001). Alberth, S. (2008): Forecasting technology costs via the experience curve – myth or magic? Technological Forecasting and Social Change 75, pp. 952-983. Alchian, A. (1963): Reliability of progress curves in airframe production. Econometrica 31, pp. 679-693. AM (2007): Various information on appliances. AM – Appliance Magazine. Canon communications LLC. Los Angeles, USA. Source: www. Appliancemagazine.com. Accessed: 10 May 2007. Aptroot, R., Meijnen, A.J. (1993): Ontwikkeling in de c.v.-ketel techniek. Het milieu en de consument gaan erop vooruit. Verwarming en Ventilatie 1, pp. 19–28. Argote, L., Epple, D. (1990): Learning curves in manufacturing, Science 247, pp. 920– 924. Argote, L. (1999): Organizational learning: creating, retaining and transferring knowledge. Kluwer. Dordrecht, the Netherlands. Arrow, K.J. (1962): The economic implications of learning by doing. Review of Economic Studies 29, pp. 155-173. AWB (2007): Adviesprijzen AWB producten. AWB B.V. Helmond, the Netherlands. Ayres, R.U., Ayres, L.W., Warr, B. (2003): Exergy, power and work in the US economy, 1900-1998. Energy 28, pp. 219-273. Ayres, R.U., van den Bergh, J.C.J.M. (2005): A theory of economic growth with material/energy resources and dematerialization: Interaction of three growth mechanisms. Ecological Economics 55, pp. 96-118. Baker, J. (2008): New technology and possible advances in energy storage. Energy Policy 36, pp. 4368-4373. Barker, T., Bashmakov, I., Alharthi, A., Amann, M., Cifuentes, L., Drexhage, J., Duan, M., Edenhofer, O., Flannery, B., Grubb, M., Hoogwijk, M., Ibitoye, F.I., Jepma,

177 References

C.J., Pizer, W.A., Yamaji, K. (2007): Mitigation from a cross-sectoral perspective. In: Climate change 2007: Mitigation. Contribution of working group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Metz, B., Davidson, O.R., Bosch, P.R., Dave, R., Mayer, L.A. (eds.)], Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Barreto, L., Kypreos, S. (2004): Endogenzing R&D and market experience in the “bottom- up” energy-systems ERIS model. Technovation 24, pp. 615-629. Barthel, C., Irrek, W., Thomas, S., Hohmeyer, O. (2005): Energieeffiziente Wäschetrockner. Wuppertal Institute für Klima, Umwelt, und Energie mbH. Wuppertal, Germany. Source: http://www.wupperinst.org/uploads/tx wibeitrag/programm9.pdf. Accessed: 30 January 2009. Bass, F. (1980): The relationship between diffusion rates, experience curves, and demand elasticities for consumer durable technological innovations. The Journal of Business - Interfaces between Marketing and Economics 53, pp. 51-67. BCG (1972): Perspectives on experience. BCG - Boston Consulting Group Inc., third edition. Boston, USA. Berglund, C., Söderholm, P. (2006): Modeling technical change in energy system analysis: analyzing the introduction of learning-by-doing in bottom-up energy models. Energy Policy 34, pp. 1344-1356. Bernstein, L., Roy, J., Delhotal, K.C., Harnisch, J., Matsuhashi, R., Price, L., Tanaka, K., Worrell, E., Yamba, F., Fengqi, Z. (2007): Industry. In: Climate change 2007: Mitigation. Contribution of working group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Metz, B., Davidson, O.R., Bosch, P.R., Dave, R., Meyer, L.A. (eds)], Cambridge University Press. Cambridge, United Kingdom and New York, USA. Bertoldi, P., Atanasiu, B. (2007): Electricity consumption and efficiency trends in the enlarged European Union. Status report 2006 - EUR 22753 EN. Joint Research Centre, Institute for Environment and Sustainability. Ispra, Italy. Blok, K, de Visser, E. (2005): Energiebesparing: de onbegrensde mogelijkheden. Ecofys. ECS05066. Utrecht, the Netherlands. Blok, K. (2007): Introduction to energy analysis. First edition. Techne Press, Amsterdam, the Netherlands. Bowden, S., Offer, A. (1994): Household appliances and the use of time: The United States and Britain since the 1920s. The Economic History Review, New Series 47, pp. 725-748. Brezet, H. (1994): Van prototype tot standaard. De diffusie van energiebesparende technologie. Een onderzoek naar het diffusieprocess van hoog-rendement CV- ketels in de particuliere huishoudelijke sector in Nederland over de periode 1981– 1992. Dissertation. Erasmus University Rotterdam. Rotterdam, the Netherlands. Bush, E., Nipkow, J. (2006): Tests and promotion of energy efficient heat pump dryers. Swiss Agency for Efficient Energy Use. Source: http://mail.mtprog.com/ CD_Layout/Day_1_21.06.06/1615-1800/ID26-74_ Bush _final.pdf. Accessed: 30 January 2009. Caldeira, K., Wickett, M.E. (2003): Anthropogenic carbon and ocean pH. Nature 425, p. 365.

178 References

CBS (2006): Industrial production in the Netherlands. Confidential data for the period of 1990–2003. CBS - Centraal Bureau voor de Statistiek (Statistics Netherlands). Voorburg, the Netherlands. CBS (2007): Statline. CBS - Centraal Bureau voor de Statistiek (Statistics Netherlands). Source: www.cbs.nl. Accessed: 10 February 2007. CECED (2003): CECED voluntary commitment II on reducing energy consumption of household washing machines. First annual report to the Commission of the European Communities. CECED – European Committee of Domestic Equipment Manufacturers. Brussels, Belgium. CECED (2004): CECED unilateral commitment on reducing energy consumption of household refrigerators and freezers. 1st Annual report for 2003 to the Commission of the European Communities. CECED – European Committee of Domestic Equipment Manufacturers. Brussels, Belgium. CECED (2007): Personal communication Mr. Axt. CECED - European Committee of Domestic Equipment Manufacturers. Brussels, Belgium. Claeson Colpier, U., Cornland, D. (2002): The economics of the combined cycle gas turbine – an experience curve analysis. Energy Policy 30, pp. 309-316. Consultic (1990–2003): Plastics production, consumption, and recycling data for Germany. Consultic Marketing und Industrieberatung GmbH. Study prepared for Plastics Europe Germany; Verband der Kunststofferzeuger, VKE (German association of plastics manufacturers). Frankfurt am Main, Germany. Consumentenbond (1983–2006): Consumentengids. Maandblad van de Nederlandse Consumentenbond. Consumentenbond. The Hague, the Netherlands. Consumentenbond (1964-2008): Consumentengids. Maandblad van de Nederlandse Consumentenbond. Consumentenbond. The Hague, the Netherlands. Consumentenbond (2009): Personal communication with Mrs. Jongeling. Consumentenbond. The Hague, the Netherlands. Cunningham, J.A. (1980): Using the learning curve as a management tool. IEEE Spectrum. IEEE - Institute of Electrical and Electronics Engineers. New York, USA. CW (2000): Product focus - ammonia. CW – Chemical Week. New York, USA. CW (2001): Product focus - methanol. CW – Chemical Week. New York, USA. CW (2005): Product focus - carbon black. CW – Chemical Week. New York, USA. Dahlman, T. (2007): Market trends for energy efficient major appliances. IEA workshop on technology learning and development. IEA – International Energy Agency. 8-9 October 2007. Paris, France. Dale, L., Antinori, C., McNeil, M., McMahon, J. (2002): Retrospective evaluation of declining price for energy efficient appliances. Lawrence Berkeley national Laboratory, paper 9.55 presented at ECEEE 2002. ECEEE – European Council for an Energy Efficient Economy. Stockholm, Sweden. Dale, L., Antinori, C., McNeil, M., McMahon, M.E., Fujita, K.S. (2009): Retrospective evaluation of appliance price developments. Energy Policy 37, pp. 597-605. de Almeida, A.T., Ferreira, F.J.T.E., Fong, J., Fonseca, P. (2008): EUP Lot 11 Motors. Eco Motors. Source: http://www.ecomotors.org/documents.htm#1. Accessed: 25 May 2009.

179 References

Destatis (1990–2003a): Produzierendes Gewerbe. Fachserie 4, Reihe 3.1. Produktion im Produzierenden Gewerbe. Destatis - Statistisches Bundesamt (Federal Statistical Office of Germany). Wiesbaden, Germany. Destatis (1990–2003b): Aussenhandel. Fachserie 7, Reihe 2 Aussenhandel nach Waren und Ländern (Spezialhandel). Destatis - Statistisches Bundesamt (Federal Statistical Office of Germany). Wiesbaden, Germany. de Vries, B. (2006): Setting the stage: matter and money, long-term prospects for world development. In: Lysen, E.H. (ed.) (2006): Climate change scientific assessment and policy analysis – Assessment of the interaction between economic and physical growth. Report 500102005. PBL - Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency). Bilthoven, the Netherlands. Dittmar, L., Faaij, A.P.C., Turkenburg, W.C. (2007): DREM - The Dutch residential energy model. Utrecht University. Utrecht, the Netherlands. DIW (2005): Personal communication with Mr. Wittke. DIW - Deutsches Institut für Wirtschaftsforschung (German Institute for Economic Research). Berlin, Germany. Dougle, P.G., Oosterheert, R.J. (1999): Case studies on energy conservation and employment in the Netherlands: subsidy on condensing boilers, subsidy on energy management systems and introduction of an energy performance standard (EPN). Report ECN-C-99-060. ECN - Energy Research Center of the Netherlands. Petten, the Netherlands. Duke, R., Kammen, D. (1999): The economics of energy market transformation programs. The Energy Journal 20, pp. 15-64. Dutton, J.M., Thomas, A. (1984): Treating progress functions as a managerial opportunity. The Academy of Management Review 9, pp. 235-247. ECCJ (2006): Survey and analysis of sales and prices for energy saving products. ECCJ - Energy Conservation Center Japan. Cited from Ellis et al. (2006). Ecocold (2007a): Part I – present situation. A portrait of the household appliance industry and market worldwide. Lot 13: Domestic refrigerators & freezers. Lot 14: Domestic dishwashers & washing machines. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40-2005). Ecocold. Source: http://www.ecocold-domestic.org/index.php. Accessed: 8 December 2008. Ecocold (2007b): Part I – present situation. A portrait of the household appliance industry and market in Europe. Lot 13: Domestic refrigerators & freezers. Lot 14: Domestic dishwashers and washing machines. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40-2005). Ecocold. Source: http://www.ecocold-domestic.org/index.php. Accessed: 8 December 2008. Ecocold (2007c): Part I – present situation. Task 1: Definitions. A portrait of the household appliance industry and market worldwide. Lot 13: Domestic refrigerators and freezers. Rev. 4.0. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40-2005). Ecocold. Source: http://www.ecocold-domestic.org/index.php. Accessed: 8 December 2008. Ecocold (2007d): Part I – present situation. Task 2: Economic and market analysis. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40- 2005). Ecocold. Source: http://www.ecocold-domestic.org/ index.php. Accessed: 8 December 2008.

180 References

Ecocold (2007e): Part II – Improvement potential. Task 6: Technical analysis, Rev. 4.0. Lot 13: Refrigerators and freezers. Rev. 4.0. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40-2005). Ecocold. Source: http://www.ecocold-domestic.org. Accessed: 16 January 2009. Ecowet (2007a): Part I – Resent situation. Task 4: Product system analysis. Lot 14: Domestic dishwashers and washing machines. Preparatory studies for Eco-design requirements of EuPs. Ecowet. Source: http://www.ecowet- domestic.org/index.php?option=com_docman&task=cat_view&gid=27&Itemid=4 0. Accessed: 15 December 2008. Ecowet (2007b): Part II – Improvement potential. Task 6: Technical analysis. Lot 14: Domestic washing machines and dishwashers. Rev. 3.0. Preparatory studies for Eco-design requirements of EuPs (Tender TREN/D1/40-2005). Ecowet. Source: http://www.ecowet-domestic.org. Accessed: 16 January 2009. EES (2006): Greening white goods: a report into the energy efficiency trends of major household appliances in Australia from 1993-2005. EES - Energy Efficiency Strategies. Report 2006/06 for Equipment Energy Efficiency (E3) Committee. Cited from Ellis (2007). Eiff, V.L., Sons, E.P., Mol, A.D., ter Hedde, M.M., Pen, M. (2001): Evaluatieonderzoek Milieu Actie Plan 1991–2000. Eindrapport. Berenschot Groep B.V. Utrecht, the Netherlands. Ellis, M. (2007): Experience with energy efficiency regulations for electrical equipment. IEA - International Energy Agency. Paris, France. Ellis, M., Jollands, N., Harrington, L., Meier, A. (2007): Do energy efficient appliances cost more? ECEEE 2007 Summer Study. ECEEE – European Council for an Energy Efficient Economy. Source: http://www.eceee.org/ conference_proceedings /eceee/ 2007/Panel_6/6. 025 /. Accessed: 12 March 2008. EnergieNed (1981-2006): Basisonderzoek aardgasverbruik kleinverbruikers. EnergieNed. Arnhem, the Netherlands. EnergieNed (2006): Energy in the Netherlands. EnergieNed. Arnhem, the Netherlands. Ennen, G. (2006): Household dishwashers’ energy consumption reduction – a success story. CECED – European Commitee of Domestic Equipment Manufacturers. Brussels, Belgium. Source: http://mail.mtprog.com/CD_Layout/Day3_23. 06.06/ 0900-1045/ID19 5_Ennen_final.pdf. Accessed: 12 December 2008. EU (1992): Council Directive 92/75/EEC of 22 September 1992 on the indication by labeling and standard product information of the consumption of energy and other resources by household appliances. EU - European Union. Official Journal L 297, 13 October 1992, pp. 16 - 19. EU (1994): Commission Directive 94/2/EC of 21 January 1994 implementing Council Directive 92/75/EEC with regard to energy labelling of household electric refrigerators, freezers and their combinations. EU - European Union. Official Journal L 45, 17 February 1994, pp. 1–22. EU (1996): Directive 96/57/EC of the European parliament and of the council of 3 September 1996 on energy efficiency requirements for household electric refrigerators, freezers and combinations thereof. EU - European Union. Official Journal L 236, pp. 36-43.

181 References

EU (1997): Commission Directive 97/17/EC of 16 April 1997 implementing Council Directive 92/75/EEC with regard to energy labelling of household dishwashers. EU - European Union. Official Journal L 118, pp. 1-26. EU (1999): Council Directive 1999/31/EC of 26 April 1999 on the landfill of waste. EU – European Union. Official Journal L 182, 16 July 1999, pp. 1-19. EU (2003a): Directive 2002/91/EC of the European Parliament and of the Council of 16 December 2002 on the energy performance of buildings. EU – European Union. Official Journal L 1, 4 January 2003, pp. 65–71. EU (2003b): Commission Directive 2003/66/EC of 3 July 2003 amending Directive 94/2/EC implementing Council Directive 92/75/EEC with regard to energy labelling of household electric refrigerators, freezers and their combinations. EU- European Union. Official Journal L170, pp. 10-14. EU (2005): Directive 32/2005/EC of the European Parliament and of the Council of 6 July, 2005 - establishing a framework for the setting of ecodesign requirements for energy-using products and amending Council Directive 92/42/EEC and Directives 96/57/EC and 2000/55/EC of the European Parliament and of the Council. EU - European Union. Official Journal L 191, 22 July, 2005, pp. 29 - 58. EU (2006a): Communication from the commission - action plan for energy efficiency: realizing the potential. EU – European Union. Source: http://ec.europa.eu/ energy/action_plan_energy_efficiency/doc/com_2006_0545_en.pdf. Accessed: 18 August 2009. EU (2006b): Green paper – a European strategy for sustainable, competitive, and secure energy. EU – European Union. Source: http://www.energy.eu/ directives/2006_03_08_gp_document_en.pdf. Accessed: 18 August 2009. EU (2006c): Directive 2006/32/EC of the European Parliament and of the Council of 5 April, 2006 on energy end-use efficiency and energy services and repealing Council Directive 93/76/EEC. EU - European Union. Official Journal L 114, 27 April, 2006, pp. 64 - 85. EU (2008a): Personal communication with Mr. Jürgens. EU – European Union, European Commission, Enterprise and Industry Directorate-General. Brussels, Belgium. EU (2008b): Questions and answers on the Commission's proposal to revise the EU Emissions Trading System. Memo 08/35. EU – European Union. Source: http://europa.eu/rapid/pressReleasesAction.do?reference=MEMO/08/35&format= HTML&aged=0&language=EN&guiLanguage=en. Accessed: 3 July 2009. EU (2009a): Directive 2009/…/EC of the European Parliament and of the Council on the promotion of the use of energy from renewable sources amending and subsequently repealing Directives 2001/77/EC and 2003/30/EC. EU – European Union. Source: http://register.consilium.europa.eu/pdf/en/08/st03/ st03736.en08.pdf. Accessed: 18 August 2009. EU (2009b): Directive 2009/…/EC of the European Parliament and of the Council of amending Directive 2003/87/EC so as to improve and extend the greenhouse gas emission allowance trading scheme of the community. EU – European Union. Source: http://register.consilium.europa.eu/pdf/en/08/st03/st03737.en08.pdf. Accessed: 18 August 2009. EU (2009c): Decision No .../2009/EC of the European Parliament and of the Council on the effort of member states to reduce their greenhouse gas emissions to meet the

182 References

community's greenhouse gas emission reduction commitments up to 2020. EU – European Union. Source: http://register.consilium.europa.eu/pdf/en/08/st03/ st03738.en08.pdf. Accessed: 18 August 2009. EU (2009d): Directive 2009/.../EC of the European Parliament and of the Council of amending Directive 98/70/EC as regards the specification of petrol, diesel and gas- oil and introducing a mechanism to monitor and reduce greenhouse gas emissions and amending Council Directive 1999/32/EC as regards the specification of fuel used by inland waterway vessels and repealing Directive 93/12/EEC. EU – European Union. Source: http://register.consilium.europa.eu/pdf/en/08/st03/ st03740-re01.en08.pdf. Accessed: 18 August 2009. Eurostat (2008): Statistics on the production of manufactured goods. Source: http://epp.eurostat.ec.europa.eu. Accessed: 7 November 2008. EZ (2005): Nu voor later - Energierapport 2005. EZ - Ministerie van Economische Zaken (Ministry of Economic Affairs). The Hague, the Netherlands. Source: http://appz.ez.nl/publicaties/pdfs/05D GE01.pdf. Accessed: 4 June 2008. Faber, A., Valente, M., Janssen, P. (2009): Exploring domestic micro-cogeneration in the Netherlands: an agent-based demand model for technology diffusion. Unpublished document, forthcoming in 2009. Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency). Bilthoven, the Netherlands. FAO (2006): Pesticides consumption. FAO - Food and Agriculture Organization of the United Nations. Source: http://faostat.fao.org/site/424/default.aspx. Accessed: 5 January 2007. Ferioli, F., Schoots, K., van der Zwaan, B.C.C. (2009): Use and limitations of learning curves for energy technology policy: a component-learning hypothesis. Energy Policy 37, pp. 2525-2535. Fischedick, M., Lechtenböhmer, S., Hanke, T., Barthel, C., Jungbluth, C., Assmann, D., von der Brüggen, T., Nitsch, J., Trieb, F., Nast, M., Langniss, O., Brischke, L.-A. (2002): Langfristszenarien für eine nachhaltige Energienutzung in Deutschland. UFOPLAN FKZ 200 97 104. Wuppertal Institut für Klima, Umwelt und Energie GmbH, Deutsches Zentrum für Luft- und Raumfahrt e.V. Wuppertal and Stuttgart, Germany. Forsa (2004): Evaluierung der Effizienzkampagne der Initiative EnergieEffizienz. Abschlussbericht. Forsa - Gesellschaft für Sozialforschung und Statistische Analysen mbH. Berlin, Germany Source: http://www.stromeffizienz.de/ fileadmin/InitiativeEnergieEffizienz/stromeffizienz/downloads/sonstigeDownloads /Abschlussbericht_IEE.pdf. Accessed: 16 January 2009. Freed, R., Mintz, C., Lanza, R., Hockstadt, L. (2005): Analytic framework for analyzing non-energy uses of fossil fuels as petrochemical feedstock in the USA. Resources, Conservation and Recycling 45, pp. 275–294. Gasterra (2007): Personal communication with Mr. Overdiep. GasTerra B.V. Groningen, the Netherlands. Gasunie (1982): Speciale uitgave hr-ketel. Gas+Architectuur - Een international tijdschrift over energiebewust bouwen. Nederlandse Gasunie. Groningen, the Netherlands. Gasunie (1988): Physical properties of natural gases. Nederlandse Gasunie. Groningen, the Netherlands.

183 References

GDA (2007): Aluminium in Zahlen. GDA - Gesamtverband der Aluminium Industrie e.V. (German association of aluminium manufacturers). Düsseldorf, Germany. Source: http://www.aluinfo.de. Accessed: 22 June 2009. GEA (1995): Washing machines, driers and dishwashers. Final report. GEA – Group for Efficient Appliances. Working Group - European Energy Network. ISBN 97-7844- 014-9. Danish Energy Agency. Copenhagen, Denmark. GfK (2003): Evolution of sales of domestic appliances in Western Europe. GfK - Gesellschaft für Konsumforschung. Presentation by M. Soregaroli, 3rd intenational conference on energy efficiency in domestic appliances and lighting - EEDAL, 1-3 October 2003. Torino, Italy. GfK (2004): Electricity end-use efficiency in new member states and candidate countries. GfK - Gesellschaft für Konsumforschung. Presentation in Brussels, 9-10 December 2004. Brussels, Belgium. Gielen, D.J., Yagita, H. (2002): Carbon accounting for Japanese petrochemicals. Journal of Mater Cycles and Waste Management 4, pp. 51–61. Greening, L. Sanstad, A., McMahon, J., Wenzel, T., Pickle, S. (1996): Retrospective analysis of national energy efficiency standards for refrigerators. Lawrence Berkeley National Laboratory. 1996 ACEEE Summer study on energy efficiency in buildings. ACEEE - American Council for an Energy Efficient Economy. Washington DC, USA. Groenenberg, H. (2002): Development and convergence - a bottom-up analysis for the differentiation of future commitments under the climate convention. Dissertation. Utrecht University. Utrecht, the Netherlands. Grübler, A., Naki4enovi4, N., Victor, D.G. (1999): Dynamics of energy technologies and global change. Energy Policy 27, pp. 247–280. Hamilton, M.R., Herzog, H.J., Parson, J.E. (2009): Cost and US public policy for new coal power plants with carbon capture and sequestration. Energy Procedia 1, pp. 4487- 4494. Hettinga, W.G., Junginger, H.M., Dekker, S.C., Hoogwijk, M., McAloon, A.J., Hicks, K.B. (2009): Understanding the reductions in US corn ethanol production costs: an experience curve approach. Energy Policy 37, pp. 190-203. Holdren, J.P. (2008): Science and technology for sustainable well-being. Science 219, pp. 424-434. Hultman, N.E., Koomey, J.G. (2007): The risk of surprise in energy technology costs, Environmental Research Letters 2, pp. 1-6. Ibenholt, K. (2002): Explaining learning curves for wind power. Energy Policy 30, pp. 1181-1189. IEA (2000): Experience curves for energy technology policy. IEA - International Energy Agency. Paris, France. IEA (2005a): Energy balances of non-OECD countries 1971–2003. CD-ROM. IEA - International Energy Agency. Paris, France. IEA (2005b): Energy balances of OECD countries 1960–2003. CD-ROM. IEA - International Energy Agency. Paris, France. IEA (2005c): Energy statistics manual. IEA – International Energy Agency. Paris, France. Source: http://www.iea.org/Textbase/nppdf/free/2004/statistics _manual. pdf. Accessed: 10 June 2009.

184 References

IEA (2006): Annual questionnaires. IEA - International Energy Agency. Paris, France. Source: http://www.iea.org/Textbase/stats/questionnaire/quest.asp. Accessed: 16 June 2008. IEA (2007a): Personal communication with Mrs. Reece. IEA – International Energy Agency. Paris, France. IEA (2007b): Tracking industrial energy efficiency and CO2 emissions - In support of the G8 plan of action. IEA - International Energy Agency. Paris, France. IEA (2008a): Key world energy statistics. IEA – International Energy Agency. Paris, France. IEA (2008b): Energy balances of non-OECD countries. 2008 edition. IEA – International Energy Agency. Paris, France. IEA (2008c): Energy technology perspectives 2008. IEA - International Energy Agency. Paris, France. IEA (2008d): Worldwide trends in energy use and efficiency - key insights from IEA indicator analysis. IEA - International Energy Agency. Paris, France. IEA (2008e): Promoting energy efficiency investments - case studies in the residential sector. IEA - International Energy Agency. Paris, France. IEA (2008f): CO2 emissions from fuel combustion. 2008 edition. IEA - International Energy Agency. Paris, France. IEA (2009): Energy technology transitions for industry – strategies for the next industrial revolution. IEA – International Energy Agency. Paris, France. IFA (2004): IFADATA-statistics, from 1973–1973/1974 to 2002–2002/2003. CD-ROM. IFA - International Fertilizer Industry Association. Paris, France. IPCC (1995): Greenhouse gas inventory reference manual. IPCC guidelines for national greenhouse gas inventories, vol. 3. IPCC - Intergovernmental Panel on Climate Change. IPCC WG I-Technical Support Unit. Bracknell, UK. IPCC (1997): Greenhouse gas inventory reporting instructions. Revised 1996 IPCC guidelines for national greenhouse gas inventories, volumes 1–3. IPCC - Intergovernmental Panel on Climate Change. WG I - Technical Support Unit. Bracknell, United Kingdom. IPCC (2006): 2006 IPCC guidelines for national greenhouse gas inventories. IPCC - Intergovernmental Panel on Climate Change. Source: http://www.ipcc-nggip .iges.or.jp/public/2006gl/index.html. Accessed: 23 May 2007. IPCC (2007a): Climate change 2007: synthesis report. Contribution of working groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Pachauri, R.K. and Reisinger, A. (eds.). IPCC - Intergovernmental Panel on Climate Change. Geneva, Switzerland. IPCC (2007b): Climate change 2007: Summary for policymakers, the physical science basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, USA. IPTS (2001): Reference document on best available technologies in the non-ferrous metal industries. IPTS - Institute for Prospective Technological Studies. Seville, Spain. Source: http://eippcb.jrc.es/pages/FActivities.htm. Accessed: 10 August 2004.

185 References

IPTS (2003): Integrated pollution prevention control (IPPC) - reference document on best available technologies in the large volume organic chemical industry. IPTS - Institute for Prospective Technology Studies. Seville, Spain. Source: http://eippcb.jrc.es/pages/FActivities.htm. Accessed: 20 May 2007. IPTS (2004): Reference document on best available technologies in the large volume inorganic chemicals, ammonia, acids, and fertilizers industries, draft March 2004. IPTS - Institute for Prospective Technological Studies. Seville, Spain. Source: http://eippcb.jrc.es. Accessed 10 August 2004. Irwin, D.A., Klenow, P.J. (1994): Learning-by-doing spillovers in the semiconductor industry. The Journal of Political Economy 102, pp. 1200-1227. Itho (2007): Personal communication with Mr. Groenendal. Itho B.V. Schiedam, the Netherlands. Iwafune, Y. (2000): Technology progress dynamics of compact fluorescent lamps. Interim Report IR-00-009. IIASA - International Institute for Applied Systems Analysis. Laxenburg, Austria. Jakob, M., Madlener, R. (2004): Riding down the experience curve for energy-efficient building envelops: the Swiss case for 1970-2000. International Journal of Energy Technology and Policy 2, pp. 153-178. Jamasb, T. (2007): Technical change theory and learning curves: patterns of progress in electricity generation technologies. The Energy Journal 28, pp. 51-71. Jepsen, D., Ipsen, A., Theloke, J., Zangel, S. (2004): VOC-Emissionskataster. Emissionsdaten für flüchtige organische Verbindungen aus der Lösemittelanwendung-Methodenevaluierung, Datenerhebung und Prognosen. UFOPLAN Vorhaben. FKZ 20143306. ÖKOPOL-Institut für Ökologie und Politik GmbH, Hamburg; Institut für Energiewirtschaft und Rationelle Energieanwendung (IER), Universität Stuttgart. Commissioned by Umweltbundesamt (Federal Environmental Agency of Germany). Berlin, Germany. Joosen, S., Harmelink, M., Blok, K., (2004): Evaluatie van het klimaatbeleid in de gebouwde omgeving 1995–2002. Rapport in opdracht van het Ministerie van Volkshuisvesting, Ruimtelijke Ordening en Milieubeheer. Ecofys B.V. Utrecht, the Netherlands. Junginger, H.M., Faaij, A., Turkenburg, W.C. (2004): Cost reduction prospects for offshore wind farms. Wind Engineering 28, pp. 97–118. Junginger, H.M. (2005): Learning in renewable energy technology development. Dissertation, Utrecht University. Utrecht, the Netherlands. Junginger, H.M., Faaij, A., Turkenburg, W.C. (2005): Global experience curves for wind farms, Energy Policy 33, pp. 133–150. Junginger, H.M., de Visser, E., Hjort-Gregersen, K., Koornneef, J., Raven, R., Faaij, A., Turkenburg, W.C. (2006): Technological learning in bioenergy systems. Energy Policy 34, pp. 4024–4041. Junginger, H.M., Lako, P., Lensink, S., van Sark, W., Weiss, M. (2008): Technological learning in the energy sector. Report NWS-E-2008-14, Utrecht University. Utrecht, the Netherlands. Kahouli-Brahmi, S. (2008): Technological learning in energy-environment-economy modeling: a survey. Energy Policy 36, pp. 138-162.

186 References

Kaiser, J. (2005): Mounting evidence indicts fine-particle pollution. Science 307, pp. 1858-1861. Kamp, L. (2002): Learning in wind turbine development. A comparison between the Netherlands and Denmark. Dissertation, Utrecht University. Utrecht, the Netherlands. Kemna, R. (2009): Personal communication with Mr. Kemna. van Holsteijn en Kemna B.V. (VHK). Delft, the Netherlands. Klaassen, G. Miketa, A., Larsen, K., Sundquist, T. (2005): The impact of R&D on innovation for wind energy in Denmark, Germany and the United Kingdom. Ecological Economics 54, pp. 227-240. Kieskeurig (2009): Wasdroger. Kieskeurig. Culembourg, the Netherlands. Source: http://www.kieskeurig.nl/wasdroger. Accessed: 26 March 2009. Kouvaritakis, N., Soria, A., Isoard, S. (2000): Modeling energy technology dynamics: methodology for adaptive expectations models with learning by doing and learning by searching. International Journal of Global Energy Issues 14, pp. 1-4. Kuramochi, T. (2006): Differentiation of greenhouse gas emissions reduction commitments based on a bottom-up approach: focus on industrial efficiency benchmarking and future industrial activity indicators. Masters thesis. Utrecht University. Utrecht, the Netherlands. Laitner, J.A. (2002): Unpublished data on experience curves, made available via personal communication. Laitner, J.A., Sanstad, A.H. (2004): Learning-by-doing on both the demand and the supply sides: implications for electric utility investments in a heuristic model. International Journal of Energy Technology and Policy 2, pp. 142–152. La Motta, S., Ancona, P, Weiss, M. (2005): CO2 emissions accounting for the non-energy use of fossil fuels in Italy: a comparison between NEAT model and the IPCC approaches. Resources, Conservation and Recycling 45, pp. 310–330. Lintona, J.D., Walsh, S.T. (2008): A theory of innovation for process-based innovations such as nanotechnology. Technological Forecasting and Social Change 75, pp. 583-594. Lipman, T.E., Sperling, D. (2000): Forecasting the costs of automotive PEM fuel cell systems – using bounded manufacturing progress functions, in: C.-O. Wene, A. Voß, T. Fried (eds.), Experience curves for policy making – the case of energy technologies, Proceedings of the IEA International Workshop 10-11 May 1999. Band 67. Institut für Energiewirtschaft und Rationelle Energieanwendung, Universität Stuttgart. Stuttgart, Germany. Lundvall, B.A. (1988): Innovation as an interactive process: from user-producer interaction to the national system of innovation. In: Dosi, G., Freeman, C., Nelson, R., Soete, L. (eds.): Technical Change and Economic Theory. Pinter, pp. 349-369. Luttmer, M. (2006): Evaluation of labeling of appliances in the Netherlands. Case study executed within the framework of the AID-EE project, contract number EIE-2003- 114. Marland, G., Boden, T.A., Andres, R.J. (2006): Global, regional, and national CO2 emissions. In: Trends: a compendium of data on global change. Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy. Oak Ridge, USA.

187 References

Martinus, G.H., Blesl, M., Smekens, K.E.L., Lako, P., Ohl, M. (2005): Technical and economic characterization of selected energy technologies. Contributions to the EU SAPIENTIA project. Report ECN-C-05-056. ECN - Energy Research Center of the Netherlands. Petten, the Netherlands. Maycock, P.D., Wakefield, G.F. (1975): Business analysis of solar photovoltaic energy conversion. Cited from: McDonald and Schrattenholzer (2001). McDonald, A., Schrattenholzer, L. (2001): Learning rates for energy technologies. Energy Policy 29, pp. 255–261. Meier, A.K., Whittier, J. (1983): Consumer discount rates implied by purchases of energy- efficient refrigerators. Energy 8, pp. 957–962. Mooi, R. (2004): HR-ketel bestaat 25 jaar, Deel I: 1978–1985. GAWALO 10/2004. MWV (2005): Personal communication with Mr. Lorenz. MWV - Mineralölwirtschaftsverband (German Petroleum Association). Hamburg, Germany. Neelis, M.L., Patel, M.K., de Feber, M. (2003): Improvement of CO2 emission estimates from the non-energy use of fossil fuels in the Netherlands. Report NWS-E-2003- 10. Utrecht University. Utrecht, the Netherlands. Neelis, M.L., Patel M.K., Blok, K. (2005a): CO2 emissions and carbon storage resulting from the non-energy use of fossil fuels in the Netherlands, NEAT results for 1993- 1999. Resources, Conservation and Recycling 45, pp. 251-274. Neelis, M.L., Patel, M.K., Gielen, D.J. Blok, K. (2005b): Modelling CO2 emissions from non-energy use with the non-energy use emission accounting tables (NEAT) model. Resources, Conservation and Recycling 45, pp. 226–250. Neelis, M.L. (2006): De bruikbaarheid van de Nederlandse Prodcom- en energiestatistieken voor de berekening van CO2-emissies en energiebesparing in de chemische industrie. Centraal Bureau voor de Statistiek (Statistics Netherlands). Voorburg/Herlen, the Netherlands. Neelis, M.L., Patel, M.K., Blok, K., Haije, W., Bach, P. (2007): Approximation of theoretical energy-saving potentials for the petrochemical industry using energy balances for 68 key processes, Energy 32, pp. 1104–1123. Neelis, M.L. (2008): Monitoring industrial energy and carbon flows. Dissertation. Utrecht University, Utrecht, the Netherlands. Neelis, M.L., Pouwelse, H. (2008): Towards consistent and reliable energy statistics for the chemical industry - a Dutch experience with international lessons. Energy Policy 36, pp. 2719–2733. Nefit (2001): Adviesprijzen Nefit cv-ketels. Nefit B.V. Deventer, the Netherlands. Nefit (2007): Adviesprijzen Nefit cv-ketels. Nefit B.V. Deventer, the Netherlands. Nefit (2008): Personal communication with Mr. Vloon. Nefit B.V. Deventer, the Netherlands. Neij, L. (1997): Use of experience curves to analyze the prospects for diffusion and adoption of renewable energy technology. Energy Policy 23, pp. 1099-1107. Neij, L. (1999): Dynamics of energy systems - methods of analyzing technology change. Dissertation. Lund University. Lund, Sweden. Neij, L, Borup, M., Blesl, M., Mayer-Spohn, O. (2006): Cost development – an analysis based on experience curves, NEEDS - New energy externalities development for sustainability. Project no. 502687. Lund University. Lund, Sweden.

188 References

Neij, L. (2008): Cost development of future technologies for power generation - a study based on experience curves and complementary bottom-up assessments. Energy Policy 36, pp. 2200-2211. Nemet, G.F. (2009): Interim monitoring of cost dynamics for publicly supported energy technologies. Energy Policy 37, pp. 825-835. Newell, R.G., Jaffe, A., Stavins, R. (1999): The induced innovation hypothesis and energy saving technological change. Quarterly Journal of Economics 114, pp. 941-975. Newell, R.G. (2000): Incorporation of technological learning into NEMS building modules. Washington DC, USA. NOAA (2009): Trends in atmospheric carbon dioxide – Mauna Loa. NOAA – National Oceanic and Atmospheric Administration. Earth System Research Laboratory. Boulder, USA. OGJ (1999): Worldwide refining/worldwide production. OGJ – Oil and Gas Journal. PennWell Petroleum Group. Tulsa, USA. OGJ (2001): Ethylene report. OGJ – Oil and Gas Journal. PennWell Petroleum Group. Tulsa, USA. Olivier, J.G.J., Peters, J.A.H.W. (2002): Uncertainties in global, regional and national emission inventories. In: Van Ham, J., Baede, A.P.M., Guicherit, R., Williams- Jacobse, J.F.G.M. (eds.): Non-CO2 greenhouse gases: scientific understanding, control options and policy aspects. Proceedings of the Third International Symposium, vol 21–23. Maastricht, the Netherlands, pp 525–540. Olivier, J.G.J. (2005a): Part III: greenhouse gas emissions: 1. Shares and trends in greenhouse gas emissions; 2. Sources and methods - greenhouse gas emissions for 1990 and 1995, pp III-1-III-31. In: IEA (2005): CO2 emissions from fuel combustion 1971-2003, 2005 edition. IEA - International Energy Agency. Paris, France. Olivier, J.G.J. (2005b): EDGAR 32FT2000. Greenhouse gas emissions. http://www.mnp.nl/edgar/model/v32ft2000edgar/edgv32ft-ghg/edgv32ftco2. jsp. Accessed: 3 July 2007. Olivier, J.G.J., Brandes, L.J., te Molder, R.A.B. (2009): Uncertainty in the Netherlands’ greenhouse gas emissions inventory – Estimation of the level and trend uncertainty using the IPCC Tier 1 approach. PBL – Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency). Bilthoven, the Netherlands. Oude Lohuis, J. (2004): Gamma-onderzoek: meer dan alleen acceptatie van nieuwe technieken. Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency). Bilthoven, the Netherlands. Park, H. (2005): Fossil fuel use and CO2 emissions in Korea: NEAT approach. Resources, Conservation and Recycling 45, pp. 295–309. Patel, M.K. (1999): Closing carbon cycles - Carbon use for materials in the context of resource efficiency and climate change. Dissertation, Utrecht University. Utrecht, the Netherlands. Patel, M.K., Jochem, E., Marscheider-Weidemann, F., Radgen, P. (1999): C-STRÖME: Abschätzung der Material-, Energie- und CO2-Ströme für Modellsysteme im Zusammenhang mit dem nicht-energetischen Verbrauch, orientiert am Lebensweg - Stand und Szenarienbetrachtung - Band I: Abschätzungen für das Gesamtsystem).

189 References

Fraunhofer Institut für System- und Innovationsforschung (Institute for Systems and Innovation Research). Karlsruhe, Germany. Patel, M.K. (2005): International network non-energy use and CO2 emissions (NEU-CO2). Source: http://www.chem.uu.nl/nws/www/nenergy/. Accessed: 8 September 2009. Patel, M.K., Neelis, M.L., Gielen, D.J., Olivier, J., Simmons, T., Theunis, J. (2005): Carbon dioxide emissions from non-energy use of fossil fuels: summary of key issues and conclusions from the country analyses. Resources, Conservation and Recycling 45, pp. 195-209. PBL (2009): Emission database for global atmospheric research (EDGAR), release version 4.0. PBL - Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency); European Commission, Joint Research Centre. Source: http://edgar.jrc.ec.europe.eu. Accessed: 4 September 2009. PE (2007): Business data and charts 2006. Plastics Europe Market Research Group, PE – Plastics Europe. Brussels, Belgium. Phylipsen, D. (2000): International comparisons and national commitments - analysing energy and technology differences in the climate debate. Dissertation, Utrecht University. Utrecht, the Netherlands. PL (2007): Personal communication with Mr. Taelman. PL - Philips Lighting. Roosendaal, the Netherlands. Prognos (2000): Prognos. Anwendung des IPCC-Referenzverfahrens zur Ermittlung der verbrennungsbedingten CO2-Emissionen in Deutschland. Report commissioned by Umweltbundesamt (Federal Environmental Agency of Germany), FKZ 29744850. Berlin, Germany. Ramírez, C.A., Worrell, E. (2006): Feeding fossil fuels to the soil – an analysis of energy embedded and technological learning in the fertilizer industry. Resources, Conservation and Recycling 46, pp. 75-93. Remeha (2007): Personal communication with Mr. van Maaren. Remeha B.V. Apeldoorn, the Netherlands. Rosenberg, N. (1982): Inside the black box: technology and economics. Cambridge University Press, pp. 304. Sallenave, J.-P. (1985): The uses and abuses of experience curves. Long Range Planning 18, pp. 64-72. Schiellerup, P. (2002): An examination of the effectiveness of the EU minimum standard on cold appliances: the British case. Energy Policy 30, pp. 327-333. Schmidt-Stejskal, H., Steinlechner, S., Windsperger, A. et al. (2004): Studie zur Anpassung der Zeitreihe der Lösemittelemissionen der österreichischen Luftschadstoffinventur 1980–2002. Report commissioned by Umweltbundesamt Österreich (Federal Environmental Agency of Austria). Vienna, Austria. Schoots, K., Ferioli F., Kramer G.J., van der Zwaan, B.C.C. (2008): Learning curves for hydrogen production technology: an assessment of observed cost reductions. International Journal of Hydrogen Energy 33, pp. 2630-2645. SenterNovem (2000): Energiepremieregeling (EPR). SenterNovem. Source: http://duurzaambouwen.senternovem.nl. Accessed: 29 January 2009. Sijbring (2007): Personal communication with Mr. Sijbring. Sijbring B.V. Tilburg, the Netherlands.

190 References

Siderius, P.J.S. (2008): Personal communication with Mr. Siderius. SenterNovem. Utrecht, the Netherlands. Sjardin, M. (2003): CO2 emission factors for non-energy use in the non-ferrous metals, ferroalloys, and inorganics industry. Report NWS-I-2003-24. Science, Technology, and Society, Utrecht University. Utrecht, the Netherlands. SMP (2007): Grootschalige toepassing van micro-WKK (HRE-ketel). SMP – Smart Power Foundation. Document provided via personal communication with Mr. Overdiep, GasTerra. Groningen, the Netherlands. Söderholm, P., Klaassen G. (2007): Wind power in Europe: a simultaneous innovation- diffusion model. Environmental and Resource Economics 36, pp. 163-190. Söderholm, P., Sundquist, T. (2007): Empirical challenges in the use of learning curves for assessing the economic prospects of renewable energy technologies. Renewable Energy 32, pp. 2559-2578. Sonnenberg, A., Sietz, M. (2007): Pestizide in der Umwelt. Fachhochschule Lippe und Höxter (University of Applied Sciences). Lemgo, Germany. Theloke, J., Obermeier, A., Friedrich, R. (2000): Ermittlung der Lösemittelemissionen 1994 in Deutschland und Methoden zur Fortschreibung. Insitut für Energiewirtschaft und Rationelle Energieanwendung (IER), Universität Stuttgart. Report No. 295 42 628 commissioned by Umweltbundesamt (Federal Environmental Agency of Germany). Berlin, Germany. Torcellini, P., Pless, S., Deru, M., Crawley, D. (2006): Zero energy buildings: a critical look at the definition. National Renewable Energy Laboratory, U.S. Department of Energy. Source: http://www.nrel.gov/docs/fy06osti/39833.pdf. Accessed: 25 August 2009. Train, K. (1985): Discount rates in consumers’ energy-related decisions: a review of the literature. Energy 10, pp. 1243–1253. Trischler (1997): Ermittlung von Altölvermeidungs-potenzialen. Trischler und Partner GmbH. Report No. 103 60 11 UBA-FB 97-034, commissioned by UBA - Umweltbundesamt (Federal Environmental Agency of Germany). Berlin, Germany UBA (2003): Leistungsskizze zum Forschungs- und Entwicklungsvorhaben 203 412 53/02 des UFOPLAN 2003. Methodenaktualisierung für die Emissionsberechnung 2003. Teilvorhaben 02: NEU-CO2 Energie-Nicht-energetischer Verbrauch. UBA – Umweltbundesamt (Federal Environmental Agency of Germany). Berlin, Germany. UBA (2004): Workshop Nationales System Emissionsinventare. Themenpapier: Emissionen von Rest-und Sekundärrohstoffen. UBA - Umweltbundesamt (Federal Environmental Agency of Germany). 8–9 November 2004. Berlin, Germany. UBA (2006): Personal communication with Mr. Strogies. UBA - Umweltbundesamt (Federal Environmental Agency of Germany). Dessau-Rosslau, Germany. UBA (2007): Umwelt Daten Online. UBA - Umweltbundesamt (Federal Environmental Agency of Germany). Berlin, Germany. Source: http://www.umweltbundesamt- umwelt-deutschland.de. Accessed: 22 June 2009. Ullmann, F. (1997): Ullmann’s encyclopedia of industrial chemistry, 5th edition on CD- ROM. John Wiley and Sons Ltd. Ullmann, F. (2007): Ullmann's encyclopedia of industrial chemistry – the Ultimate reference release. 7th edition. Wiley-VCH. Weinheim, Germany.

191 References

UN (1992): United Nations Framework Convention on Climate Change. UN – United Nations. Source: http://unfccc.int/resource/docs/convkp/conveng.pdf. Accessed: 3 July 2009. UN (2000): Industrial commodity production statistics 1950–2000. CD-ROM. UN - United Nations. New York, USA. UN (2008): Industry commodity production data 1950-2005. CD-ROM. UN – United Nations, Statistics division. New York, USA. UNEP (2007): Global environmental outlook 4. UNEP – United Nations Environmental Programme. Source: http://www.unep.org/geo/geo4/report/ GEO-4_Report_Full_ en.pdf. Accessed: 18 August 2009. UNFCCC (2004): Common reporting format tables for Germany (submitted 30 April 2004). UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/national_reports/annex_i_ghg_inventories/national_ inventories_submissions/items/3473.php. Accessed: 22 June 2009. UNFCCC (2005a): Common Reporting Format Tables for Germany (submitted 15 April 2005). UNFCCC - United Nations Framework Convention on Climate Change Source: http://unfccc.int/national_reports/annex_i_ghg_inventories/national_ inventories_submissions/items/2761.php. Accessed: 12 August 2006. UNFCCC (2005b): Germany - Report of the Individual Review of the Greenhouse Gas Inventory. Submitted in the year 2004. Executive Summary. UNFCCC - United Nations Framework Convention on Climate. Source: http://unfccc.int /national_reports/annex_i_ghg_inventories/national_inventories_submissions /items/2761.php. Accessed: 27 April 2005. UNFCCC (2005c): National Inventory Report Germany. UNFCCC - United Nations Framework Convention on Climate Change Source: http://unfccc.int/national _reports/annex_i_ghg_inventories/national_inventories_submissions/items/2761.p hp. Accessed: 21 September 2007. UNFCCC (2006a): National inventory submissions 2006, GHG inventories of Annex I countries. UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/national_reports/annex_i_ghg_inventories/national_ inventories_submissions/items/3734.php. Accessed: 12 August 2006. UNFCCC (2006b): Common Reporting Format Tables for Germany (submitted 31 January 2007). UNFCCC - United Nations Framework Convention on Climate Change Source: http://unfccc.int/national_reports/annex_i_ghg_inventories/ national_inventories_submissions/items/3734.php. Accessed: 12 August 2006. UNFCCC (2006c): Inventory review reports. UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/national_reports/ annex_i_ghg_inventories/inventory_review_reports/items/4030.php. Accessed: 19 August 2009. UNFCCC (2007a): Common reporting format tables for Germany (submitted 2 May 2007). UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/national_reports/annex_i_ghg_inventories/national_ inventories_submissions/items/3929.php. Accessed: 10 May 2007. UNFCCC (2007b): National Inventory Report Germany. UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/

192 References

national_reports/annex_i_ghg_inventories/national_inventories_submissions/items/ 3929.php. Accessed: 10 September 2007. UNFCCC (2008): National inventory submissions 2008, GHG Inventories of Annex I countries. UNFCCC - United Nations Framework Convention on Climate Change Source: http://unfccc.int/national_reports/annex_i_ghg _inventories/ national _ inventories _submissions/items/4303.php. Accessed: 18 June 2008. UNFCCC (2009): Kyoto protocol - status of ratification. UNFCCC - United Nations Framework Convention on Climate Change. Source: http://unfccc.int/files/ kyoto_protocol/status_of_ratification/application/pdf/kp_ratification.pdf. Accessed: 3 July 2009. Vaillant (2007): Prijslijst januari 2007. Vaillant B.V. Amsterdam, the Netherlands. van der Zwaan, B.C.C., Seebregts, A. (2004): Endogenous learning in climate-energy- economic models – an inventory of key uncertainties. International Journal of Energy Technology Policy 2, pp. 130-141. van Overbeeke, P. (2001): Kachels, geisers en fornuizen. Keuzeprocessen en energieverbruik in Nederlands huishoudens 1920–1975. Dissertation. Technical University Eindhoven. Eindhoven, the Netherlands. van Sark, W.G.J.H.M. (2007): Introducing errors in progress ratios determined from experience curves. Technological Forecasting and Social Change 75, pp. 405-415. van Vuuren, D.P., van Ruijven, B., Hoogwijk, M., Isaac, M., de Vries, B. (2006): TIMER 2.0: model description and application. In: Bouwman, L., Kram, T, Klein- Goldwijk, K. (eds.), IMAGE 2.4: An overview. Planbureau voor de Leefomgeving (Netherlands Environmental Assessment Agency). Bilthoven, the Netherlands. VCI (2004a): Primärchemikalienerfassung. Confidential data made available via personal communication Mr. Rothermel. VCI - Verband der Chemischen Industrie e.V. (Association of German Chemicals Manufacturers). Frankfurt am Main, Germany. VCI (2004b): Personal communication with Mr. Rothermel. VCI - Verband der Chemischen Industrie e.V. (Association of German Chemicals Manufacturers). Frankfurt am Main, Germany. Visser (2007): Personal communication with Mr. Visser. Installatiebedrijf Visser. Twisk, the Netherlands. Waide, P.A. (2001): Monitoring of energy efficiency trends for refrigerators, freezers, washing machines, washer-dryers and household lamps sold in the EU. Final Report. PW Consulting. Manchester, United Kingdom. Warmteservice (2007): Verwarming. Source: www.warmteservice.nl. Accessed: 2 February 2007. WB (2004): World development indicators 2004. WB – World Bank. Washington DC, USA. Source: www.worldbank.org. Accessed: 20 November 2006. WBMS (2002): Metal statistics (1991–2002), 89th edition. WBMS - World Bureau of Metal Statistics. Ware, England. Weber, C., Gebhards, B., Fahl, U. (2002): Market transformation for energy efficient technologies - success factors and empirical evidence for gas condensing boilers. Energy 27, pp. 287–315. Weiss, M., Neelis, M.L., Patel, M.K. (2007): Estimating CO2 emissions from the non- energy use of fossil fuels in Germany. Report NWS-E-2007-21, prepared for the

193 References

Federal Environmental Agency of Germany. Utrecht University. Utrecht, the Netherlands. Weiss, M., Neelis, M.L., Zuidberg, M.C., Patel, M.K. (2008a): Applying bottom-up analysis to identify the system boundaries of non-energy use data in international energy statistics. Energy 33, pp. 1609-1622. Weiss, M., Neelis M.L., Blok K., Patel M.K. (2008b): Non-energy use and related carbon dioxide emissions in Germany: a carbon flow analysis with the NEAT model for the period of 1990-2003. Resources, Conservation and Recycling 52, pp. 1252-1265. Weiss, M., Junginger, M.H., Patel, M.K. (2008c): Learning energy efficiency - experience curves for household appliances and space heating, cooling, and lighting technologies. Report NWS-E-2008-26. Utrecht University. Utrecht, the Netherlands. Weiss, M., Neelis, M.L., Patel, M.K., Blok, K. (2009a): Non-energy use of fossil fuels and resulting CO2 emissions: bottom-up estimates for the world as a whole and for major developing countries. Climatic Change 95, pp. 369-394. Weiss, M., Dittmar, L., Junginger, H.M., Patel, M.K., Blok, K. (2009b): Market diffusion, technological learning, and cost-benefit dynamics of condensing gas boilers in the Netherlands. Energy Policy 37, pp. 2962-2976. Weiss, M., Junginger, H.M., Patel, M.K., Blok, K. (2009c): Analyzing price and efficiency dynamics of large appliances with the experience curve approach. Accepted for publication in Energy Policy. Weiss, M., Junginger, H.M., Patel, M.K., Blok, K. (2009d): A review of experience curve analyses for energy demand technologies. Accepted for publication in Technological Forecasting and Social Change. Weissermel, K., Arpe, H.-J. (2003): Industrial organic chemistry. 4th edition, Wiley-VCH, Weinheim, Germany. Weizsäcker, E.U., Lovins, A.B., Lovings, L.H. (1997): Faktor vier. Doppelter Wohlstand – halbierter Naturverbrauch. 10th edition, Droemer Knaur. Munich, Germany. Wene, C.-O., Voß, A., Fried, T. (2000): Experience curves for policy making – the case of energy technologies, Proceedings of the IEA International Workshop 10-11 May 1999, Band 67. Institut für Energiewirtschaft und Rationelle Energieanwendung (IER), Stuttgart University. Stuttgart, Germany. Wene, C.-O. (2008a): A cybernetic perspective on technological learning. In: Foxon, T., et al. (Eds.): Innovations for a low carbon economy: economic, institutional and management approaches. Edward Elgar. London, UK. Wene, C.-O. (2008b): Energy technology learning through deployment in competitive markets. The Engineering Economist 53, pp. 340-364. Worrell, E. (1994): Potential for improved use of industrial energy and materials. Dissertation. Utrecht University. Utrecht, the Netherlands. Worrell, E., Bernstein, L., Roy, J., Price, L., Harnisch, J. (2009): Industrial energy efficiency and climate change mitigation. Energy Efficiency 2, pp. 109-123. Wright, T.P. (1936): Factors affecting the cost of airplanes. Journal of the Aeronautical Sciences 3, pp. 122-128. Yara (2008): Personal communication with Mr. Wodsack. Yara GmbH & Co. KG. Dülmen, Germany.

194

Summary

Fossil energy is still a main driver of the global economy. However, in recent decades concerns have been growing about the tremendous environmental impacts as well as the long-term availability and affordability of fossil fuels. In particular, fossil fuel-related greenhouse gas (GHG) emissions as cause of global climate change have triggered policy action on a national and international scale. Various policy initiatives directly or indirectly address the manufacturing industry, which plays a dual role in the context of energy use and GHG emissions, namely (i) as a consumer of energy and (ii) as a producer of energy- consuming technologies. Industrial manufacturing itself uses one third of the global primary energy supply; the other two thirds are consumed by energy conversion and end- use technologies, which are produced by the manufacturing industry but which are utilized in other sectors of the economy.

The dual function of the manufacturing industry has been addressed by a large body of scientific work. However, we still lack a detailed quantitative understanding of several important elements of industrial manufacturing in the context of energy use and GHG emissions. Existing knowledge gaps regarding the manufacturing industry as an energy consumer are related to the accuracy of energy statistics and GHG emission inventories. Specifically, we still lack a detailed and accurate accounting of fossil fuels that are used for non-energy purposes (e.g., as feedstock in the chemical industry) and the associated carbon dioxide (CO2) emissions. With respect to the manufacturing industry as a producer of energy consuming technologies, knowledge gaps exist regarding technological learning, i.e., the rate at which costs decline for novel and efficient energy demand technologies. This recognition forms the starting point of our research, in which we address two distinct research questions related to the industry-energy complex: (i) How can we improve accuracy and completeness in the accounting of emissions caused by the non-energy use of fossil fuels? (ii) At which rates does technological learning occur for energy demand technologies?

To address the first research question, we develop in Chapters 2-4 bottom-up models that quantify non-energy use of fossil fuels and related emissions at various levels of detail. To address the second research question, we analyze in Chapters 5-7 technological learning by applying the experience curve approach to condensing gas boilers and large household appliances, and by reviewing the results of experience curve analyses for a wide range of energy demand technologies.

195 Summary

We quantify in Chapter 2 the worldwide non-energy use of fossil fuels in the year 2000 to be 20 ± 2 EJ (exajoules), thereby accounting for 5% of the global total primary energy supply. Our estimates, i.e., their 95% confidence intervals are in line with data from international energy statistics in the case of 12 countries, whereas they are lower than data from international energy statistics in the case of 26 countries and for the world as a whole. For another 12 countries, our estimates exceed the data of international energy statistics. We find that inconsistent definitions of system boundaries (i.e., the inclusion of the fuel use of feedstock for process energy requirements) as well as errors in official non- energy use data provide two important explanations for deviations between our estimates and data given by international energy statistics.

In Chapter 3, we estimate that global non-energy use emissions in the year 2000 amount to 700 ± 90 Mt (megatonnes) CO2. The forty industrialized countries that submit a GHG emission inventory to the UNFCCC (i.e., the so-called Annex I countries) account for 51% (360 ± 50 Mt CO2) of this total. Developing countries that do not submit a GHG emission inventory to the UNFCCC (i.e., the so-called non-Annex I countries) account for 49% (340 ± 70 Mt CO2). Around 75% of global non-energy use emissions are related to industrial processes; the remainder is attributed to product use, agriculture, and waste management.

Chapter 4 shows that yearly non-energy use emissions in Germany increased on average by 1.8% per year from 22.0 ± 1.8 Mt CO2 in 1990 to 27.7 ± 2.3 Mt CO2 in 2003. Out of these totals, 77% of the emissions are related to industrial processes, 17% to solvent and other product use, 2% to the use of urea fertilizers and pesticides in agriculture, and 4% to wastewater treatment. Our results close major gaps in the German GHG inventory by providing emission estimates for methanol and carbon black production as well as for carbon losses in chemical conversion processes. However, our analysis also shows that the German GHG inventory does not yet fully include emissions from steam cracking, non-ferrous metals production, and product use.

By identifying inconsistencies and errors in the non-energy use estimates of international energy statistics, our research emphasizes that substantially more attention is necessary to account for non-energy use in a correct manner. Our research provides the first detailed estimates for non-energy use emissions of major non-Annex I countries, which do not yet have established complete GHG inventories. Although our bottom-up approaches are subject to uncertainties, they are capable of providing insights into non- energy use and they allow for a more detailed, complete, and accurate accounting of non- energy use emissions in comparison to existing methodologies. Our methods are, in particular, useful for providing crosschecks of emission estimates in GHG inventories, which are prepared according to the 2006 Intergovernmental Panel on Climate Change (IPCC) guidelines and for which internal quality checks with the IPCC-Reference Approach (IPCC-RA) are no longer foreseen.

196 Summary

We conclude our research on non-energy use and resulting emissions with the following recommendations for policy makers, energy statisticians, and inventory experts: (i) Substantially more attention, staffing, and expertise are required in national and international statistical offices in order to improve the quality of energy and production statistics to such an extent that makes them suitable for GHG mitigation purposes. (ii) Inconsistencies regarding the system boundaries of non-energy use data in national and international energy statistics have to be resolved based on international agreements and more detailed guidance as provided by the questionnaires of the International Energy Agency. National energy statistics must strictly apply international standards to assure comparability of data. (iii) The processes of preparing energy statistics and GHG inventories should be linked more closely together to assure that energy statistics fulfil the requirements for accurate emissions accounting. (iv) Exact knowledge of the system boundaries of data in energy statistics is critical for GHG inventory makers in order to correctly quantify emissions in the relevant source categories as outlined by the IPCC-Sectoral Approach (IPCC-SA) (i.e., energy, industrial processes, solvent and other product use, agriculture, and waste). (v) Whereas the 2006 IPCC guidelines exclude non-energy use emissions from the IPCC-RA, we recommend continuing to use the IPCC-RA as a crosscheck for the total of non-energy use emissions, which are estimated for individual activities according to the IPCC-SA. (vi) To assist in reliable emissions accounting, we recommend that inventory experts around the world use our bottom-up models as well as the results presented in this thesis.

To address the second research question, we quantify technological learning of energy demand technologies by applying the experience curve approach. Experience curves model production costs or prices of a technology as a power-law function of cumulative production. The change in production costs or prices is quantified by so-called learning rates (LR), which represent the percentage of cost or price decline with each doubling of cumulative production. We find (i) that the experience curve approach is generally applicable to energy demand technologies and (ii) that energy demand technologies show a widespread trend towards price and cost decline.

In Chapter 5, we estimate that prices of condensing gas combi boilers decline at a learning rate of 14 ± 1%. For large appliances, we find in Chapter 6 a robust long-term price decline for wet and cold appliances at learning rates of, on average, 29 ± 8% and 9 ± 4%, respectively. Possible explanations for the observed deviation between the learning rates of the two clusters of appliances might include differences regarding technical complexity and potentials for enhanced automation and innovation in component manufacturing and final assembly. However, a thorough explanation of deviations cannot be provided in a straightforward manner because this would require more detailed analysis. The results for condensing gas boilers and large appliances are in line with the findings of our literature review (Chapter 7), where we identify that learning rates of

197 Summary energy demand technologies are approximately normally distributed around a mean of 18 ± 9%. Variation of learning rates within individual technology clusters is often as large as variation between technology clusters. This limits possibilities to derive technology- specific ex-ante learning rates for energy demand technologies.

By modelling the specific energy consumption of large appliances as a function of cumulative production, we extend the conventional experience curve approach to quantify technological learning from a broader perspective. The resulting energy experience curves indicate a robust decline in the specific energy consumption of wet appliances (LR of 18 ± 3% to 35 ± 3%) and cold appliances (LR of 13 ± 3% to 17 ± 2%).

With our experience curve analysis, we show that technological learning leads to a substantial decline in the prices of energy demand technologies. Therefore, our results indicate that novel and efficient but initially expensive energy demand technologies do not necessarily have to impose higher costs on consumers in the long term. This conclusion is supported by our cost-benefit analysis for condensing gas boilers, which demonstrates the importance of technological learning for improvements in the cost competitiveness of novel and efficient energy demand technologies. The net present value of condensing gas combi boilers was negative at -330 EUR2006 (real Euro, deflated to the base year of 2006) in 1988 but increased to 970 EUR in 2006. We attribute two thirds of this increase to technological learning and one third to a combination of external effects (e.g., heat demand in households) and governmental policies. From a social perspective, costs for emission savings achieved by condensing gas combi boilers decreased from 60 EUR2006/t CO2 in 1988 to -120 EUR2006/t CO2 in 2006. Our findings show how limited subsidy support of 70 ± 10 million EUR substantially improved the cost-benefit performance of condensing gas boilers, contributed to consumer cost savings of a multiple of the provided subsidies, and enabled energy and emission savings in the residential sector of the Netherlands.

Although experience curves possess advantages over simple time-series analysis, they are also subject to methodological and empirical uncertainties. In particular, experience curve analysis should be complemented by bottom-up engineering analysis to allow a more detailed understanding of mechanisms that are responsible for the cost decline at the system level. Furthermore, sensitivity analysis of experience curve results is important because already small variations in learning rates can substantially change long- term cost projections.

We summarize our research on technological learning with the following recommendations for scientists and policy makers: (i) Technological learning of energy demand technologies should be accounted for in energy modelling and energy policy because this mechanism contributes to a substantial decline of investment and use-phase costs in the mid and long term. (ii) Providing error margins of learning rates is essential for the sensitivity analysis of long-term cost projections because small variations in learning

198 Summary

rates can lead to substantial deviations in long-term technology forecasts in energy models. (iii) Modelling energy consumption with experience curves reveals new insights into the dynamics of energy efficiency. The potentials of such energy experience curves for analyzing energy efficiency dynamics should be explored further throughout all areas of energy demand in the manufacturing industry as well as in the transport and buildings sector.

Energy efficiency improvements are regarded as an important mechanism for realizing a sustainable energy system. We conclude that technological learning helps to achieve a more efficient use of energy at declining costs. However, enabling an economy- wide decline of absolute energy consumption requires additional efforts and commitment in comparison to the current situation and makes it necessary to also adopt initially costly, economically less attractive, but highly efficient energy demand technologies.

199

Samenvatting

De mondiale energievraag wordt op dit moment nog steeds gedomineerd door fossiele brandstoffen. Het grootschalig gebruik van fossiele brandstoffen heeft ernstige gevolgen voor het milieu. Daarnaast dreigt de beschikbaarheid van fossiele brandstoffen op de lange termijn in gevaar te komen. De uitstoot van broeikasgassen ten gevolge van het gebruik van fossiele brandstoffen heeft geleid tot nationale en internationale beleidsmaatregelen. Een aantal van deze maatregelen zijn gericht op de industrie. In relatie tot energie- en klimaatbeleid heeft de industrie een dubbele functie. Enerzijds is de industrie gebruiker van energie, anderzijds produceert de industrie goederen die energie gebruiken in de gebruiksfase.

Ongeveer een derde van de wereldwijde primaire fossiele energie wordt gebruikt door de industrie. De overige tweederde wordt gebruikt door technologieën die geproduceerd zijn in de industrie, maar die gebruikt worden in andere sectoren van de economie. De industrie en industrieel energiegebruik zijn onderwerp van uitgebreid wetenschappelijk onderzoek. Maar tot nu toe ontbreekt een gedetailleerd kwantitatief inzicht van een aantal belangrijke aspecten. Ten eerste is de kennis van het gebruik van fossiele brandstoffen voor niet-energetische doeleinden en de daarmee samenhangende broeikasgasemissies onvoldoende. Een voorbeeld voor het niet-energetisch gebruik is het gebruik van ruwe olie als grondstof in de chemische industrie. Ten tweede is de kennis van technologisch leren op het gebied van energieverbruikende technologieën beperkt. Een voorbeeld is de snelheid waarmee de kosten en het specifieke energiegebruik van technologieën daalt. Deze aspecten vormen het uitgangspunt van dit onderzoek, waarin de volgende twee onderzoeksvragen worden beantwoord: (i) Hoe kunnen we de emissies ten gevolge van het niet-energetisch gebruik van fossiele brandstoffen nauwkeuriger berekenen? (ii) Wat is de snelheid van technologisch leren voor nieuwe technologieën voor het energiegebruik?

Voor het beantwoorden van de eerste vraag ontwikkelen wij bottom-up modellen. Deze modellen maken het mogelijk om op verschillende detailniveaus het niet-energetisch gebruik van fossiele brandstoffen en de daarmee samenhangende emissies te kwantificeren. In hoofdstuk 2 ramen wij het niet-energetisch gebruik van fossiele brandstoffen in het jaar 2000 op 20 ± 2 EJ. Dit komt overeen met 5% van de mondiale primaire energieproductie. Onze raming, met inbegrip van het 95%-betrouw- baarheidsinterval, is voor 12 landen in overeenstemming met de gegevens uit de internationale energiestatistieken. Voor 26 landen en voor de wereld als geheel zijn onze resultaten lager dan de gegevens in de internationale energiestatistieken. Voor 12 landen zijn onze schattingen hoger dan de gegevens uit de internationale energiestatistieken. Deze verschillen zijn het gevolg van inconsistente definities van systeemgrenzen die betrekking hebben op het gebruik van grondstoffen voor het opwekken van energie.

201 Samenvatting

In hoofdstuk 3 wordt de mondiale uitstoot van broeikasgassen door het niet- energetisch gebruik van fossiele brandstoffen geschat op 700 ± 90 Mt CO2. Hiervan is 51% (360 ± 50 Mt CO2) afkomstig uit geïndustrialiseerde (Annex I) landen en 49% (340 ± 70 Mt CO2) uit ontwikkelings- (Non-Annex I) landen. Ongeveer 75% van de niet- energetische emissies zijn gerelateerd aan industriële processen. Overige bronnen zijn productgebruik, landbouw en afvalverwerking.

Hoofdstuk 4 laat zien dat niet-energetische emissies in Duitsland zijn gestegen met een gemiddeld 1.8% per jaar van 22.0 ± 1.8 Mt CO2 in 1990 tot 27.7 ± 2.3 Mt CO2 in 2003. Hiervan is 77% gerelateerd aan industriële processen, 17% aan oplosmiddel- en ander productgebruik, 2% aan het gebruik van N-kunstmest en pesticiden in de agrarische sector en 4% aan afvalwaterzuivering. Onze resultaten hebben substantieel bijgedragen aan het dichten van belangrijke gaten in het broeikasgas emissie-inventarisatie van Duitsland door het beschikbaar stellen van emissieramingen voor de productie van methanol en zwartsel en koolstofverliezen in chemische conversieprocessen. Onze analyse laat ook zien dat de officiële broeikasgas emissie-inventarisatie van Duitsland nog niet volledig is voor emissies van stoomkraken, de productie van non-ferrometalen en het gebruik van smeermiddelen en parafijnen.

Met het identificeren van fouten en inconsistenties in de ramingen van het niet- energetisch gebruik in energiestatistieken, toont ons onderzoek aan dat er nog substantieel werk verricht moet worden om het niet-energetisch gebruik geharmoniseerd en correct op te nemen in de nationale en internationale energiestatistieken. Ons onderzoek geeft de eerste gedetailleerde ramingen van niet-energetische emissies voor de belangrijkste Non- Annex I landen die nog geen complete broeikasgas emissie-inventarisaties hebben opgezet. Onze bottom-up modellen bieden de mogelijkheid om het niet–energetisch gebruik inzichtelijker te maken en daarmee het detailniveau, de compleetheid en nauwkeurigheid van niet-energetische emissieramingen beter te berekenen in vergelijking tot bestaande methoden. Deze modellen zijn voornamelijk geschikt voor het controleren van emissieramingen in broeikasgas emissie-inventarisaties die gemaakt zijn volgens de 2006 richtlijnen van het Intergovernmental Panel on Climate Change (IPCC). Omdat interne kwaliteitscontrole voor niet-energetisch gebruik met de ‘IPCC-Reference Approach’ (IPCC-RA) niet meer mogelijk is. We eindigen ons onderzoek van niet- energetisch energiegebruik en gerelateerde emissies met de volgende aanbevelingen voor beleidsmakers, energiestatistici en inventarisatie experts: (i) Er is substantieel meer aandacht, personeel en expertise nodig in om de kwaliteit van energie en productiestatistieken te verbeteren en daarmee bruikbaar te maken voor broeikasgasmitigatie doeleinden. (ii) Inconsistenties in systeemgrenzen van niet-energetisch energiegebruikdata in nationale en internationale statistieken moeten worden voorkomen door: het maken van internationale afspraken en gedetailleerdere vragenlijsten voor energiestatistieken van het Internationale Energie Agentschap (IEA). De nationale energiestatistieken moeten strikt voldoen aan internationale standaarden om de vergelijkbaarheid van data te garanderen.

202 Samenvatting

(iii) Een verdere integratie van het voorbereidingsproces van energiestatistieken en van broeikasgas emissie-inventarisaties is nodig zodat de energiestatistieken voldoen aan de eisen voor emissieberekeningen. (iv) Exacte kennis van systeemgrenzen in energiestatistieken is cruciaal voor de makers van broeikasgas emissie-inventarisaties om de emissies in de relevante broncategorieën correct te kwantificeren (d.w.z., energie, indust- riële processen, oplosmiddel- en ander productgebruik, landbouw en afval). (v) Hoewel de 2006 IPCC richtlijnen emissies van niet-energetisch gebruik uitsluiten van de IPCC-Reference Approach (IPCC-RA), raden wij aan om de IPCC-RA te gebruiken als macrocontrole voor het totaal van alle niet- energetische emissies zoals geraamd voor de individuele activiteiten volgens de IPCC-Sectoral Approach (IPCC-SA) methode. (vi) Voor het berekenen van emissieramingen, raden wij inventarisatie experts aan om onze bottom-up modellen toe te passen en de resultaten, zoals gepresenteerd in dit proefschrift, te gebruiken voor de betreffende landen.

Voor het beantwoorden van de tweede onderzoeksvraag kwantificeren we het leerpotentieel van energiegebruik technologieën met de zogenaamde leercurve methode. Met deze leercurves, is het mogelijk om het zogenaamde leerpercentage te berekenen. Deze leerwaarden geven het percentage aan waarmee kosten dalen bij elke verdubbeling van het cumulatieve productievolume. De uitkomsten voor deze onderzoeksvraag zijn dat ten eerste, de leercurve benadering algemeen toepasbaar is op energiegebruik technologieën en ten tweede, dat al deze technologieën een trend in kostenreductie vertonen.

In hoofdstuk 5 schatten we dat de prijzen voor hoogrendement (HR) combi-ketels verminderen met een leerpercentage van 14 ± 1%. In hoofdstuk 6, vinden we voor zogenaamde witgoedapparaten een prijsverlaging die robuust is op de lange termijn met gemiddelde leerpercentages van 29 ± 8% voor wasmachines, -drogers en afwasmachines en gemiddelde leerpercentages van 9 ± 4% voor koel- en vrieskasten en voor koel/vries combinaties. Er is geen directe verklaring voor de verschillen tussen de leerpercentages van deze verschillende technologieën. Mogelijkheden zijn verschillen in technische complexiteit en verschillen in het potentieel voor automatisering en innovatie in de productie van de apparaten of componenten hiervan.

De resultaten voor HR-combi ketels en witgoedapparaten zijn in overeenstemming met de bevindingen uit de literatuurstudie (hoofdstuk 7). In dit hoofdstuk vinden wij leerpercentages voor energiegebruik technologieën met een Gauss-verdeling van gemiddeld 18 ± 9%. De variatie in leerpercentages van dezelfde technologieën is vaak even groot als de variatie in leerpercentages tussen verschillende technologieën. Dit beperkt de mogelijkheden om technologiespecifieke ex-ante leerpercentages voor energiegebruik technologieën te ontwikkelen. We breiden de conventionele leercurve methode uit door het modelleren van het specifieke energiegebruik van witgoedapparaten als een functie van het cumulatieve historische productievolume. De resulterende energie leercurves tonen een robuuste vermindering in het specifiek energiegebruik van

203 Samenvatting wasmachines, -drogers en afwasmachines (leerpercentages van 18 ± 3% tot 35 ± 3%) en koel- en vrieskasten en combinaties (leerpercentages van 13 ± 3% tot 17 ± 2%).

Met de hier gepresenteerde leercurve analyse tonen we aan dat technologisch leren leidt tot een substantiële vermindering in productiekosten, prijzen en specifiek energiegebruik van energiegebruik technologieën. Onze resultaten geven aan dat nieuwe, aanvankelijk dure en efficiënte energiegebruik technologieën, op de lange termijn niet noodzakelijk leiden tot hogere kosten voor consumenten. Deze conclusie wordt ondersteund door onze kosten-batenanalyse van HR-combi ketels. De geactualiseerde totale waarde van de investering (inclusief alle verdisconteerde kosten en baten tijdens de levensduur van HR-combi ketels was in 1988 negatief, nl. X -330, maar nam toe tot X 970 in 2006. Tweederde van deze verbetering kan worden toegeschreven aan technologisch leren, een derde aan een combinatie van externe effecten (zoals de warmtevraag van huishoudens) en belastingheffing op energie. Vanuit maatschappelijke perspectief blijkt dat de kosten voor emissie-reducties met HR-combi ketels zijn afgenomen van 60 X/t CO2 in 1988 tot -120 X/t CO2 in 2006. De hier gepresenteerde resultaten tonen aan dat een beperkte subsidie van X 70 ± 10 miljoen een substantiële verbetering bracht in de kosten- baten balans van HR-ketels. Dit droeg bij aan kostenbesparingen voor consumenten die een veelvoud waren van de verstrekte subsidie en zorgden voor emissie-reducties in de Nederlandse huishoudens.

Ondanks dat leercurves veel voordelen hebben in vergelijking tot eenvoudige tijdreeksanalyses, heeft deze methode ook onzekerheden. Om precies te zijn, de leercurve benadering zou moeten worden gecombineerd met een bottom-up technologische analyse om inzicht te krijgen in de componenten en mechanismen die verantwoordelijk zijn voor kostendalingen op het systeemniveau.

De hier gepresenteerde studie naar technologisch leren besluiten we met de volgende aanbevelingen voor onderzoekers en beleidsmakers: (i) In energiemodellen en energiebeleid moet rekening worden gehouden met het proces van technologisch leren voor energiegebruik technologieën, omdat dit mechanisme bijdraagt aan een substantiële vermindering van investerings- en gebruikskosten van deze technologieën op de middellange en lange termijn. (ii) Het geven van foutenmarges voor leerpercentages is essentieel voor een gevoeligheidsanalyse van kostenprojecties, omdat kleine variaties in leerpercentages al kunnen leiden tot substantiële afwijkingen in lange termijn projecties. (iii) Het modelleren van het specifieke energiegebruik met behulp van leercurves, geeft nieuwe inzichten in de dynamiek van energie-efficiëntie veranderingen. Het potentieel van zulke energie leercurves moet verder worden verkend voor alle aspecten van het energiegebruik in de industrie, het transport en de bebouwde omgeving.

204 Samenvatting

Verbeteringen van energie-efficiëntie worden beschouwd als een belangrijk mechanisme voor het realiseren van een duurzaam energiesysteem. We concluderen dat technologisch leren bijdraagt aan bereiken van het bestaande energie-efficiënte potentiëlen op een minder kostbare manier. Echter, voor het bereiken van een substantiële vermindering in absoluut energiegebruik, moet ook worden geïnvesteerd in economisch minder aantrekkelijk, maar hoog efficiënte technologieën voor energiegebruik. Om dit te bereiken is meer politieke invloed noodzakelijk.

205

Zusammenfassung

Ein Großteil des weltweiten Energiebedarfs wird durch fossile Energieträger gedeckt. Im Hinblick auf den Verbrauch fossiler Energie sowie die daraus resultierenden Kohlendioxid-Emissionen kommt der Industrie eine besondere Bedeutung zu. Sie ist zunächst Abnehmer von etwa einem Drittel des weltweiten Primärenergieangebots. Darüber hinaus beansprucht die Nutzung von industriell hergestellten Anlagen, Maschinen und Konsumgütern in anderen Bereichen der Wirtschaft die restlichen zwei Drittel des globalen Aufkommens an Primärenergie.

Vor dem Hintergrund der gegenwärtigen Bemühungen sowohl Energie-verbrauch als auch Treibhausgas-Emissionen zu senken, kommt sowohl der statistischen Erfassung des Verbrauchs fossiler Energieträger als auch der Steigerung von Energieeffizienz eine wesentliche Bedeutung zu. In diesem Zusammenhang ist festzustellen, dass Rohstoffeinsatz und Emissionen aus Industrieprozessen noch immer statistisch unzureichend erfasst werden. Dies trifft in besonderem Maße auf fossile Brennstoffe zu, die für nicht-energetische Zwecke genutzt werden. Darüber hinaus besteht ein unzureichender Kenntnisstand im Hinblick auf technologisches Lernen, d.h. hinsichtlich der Lernraten mit denen sich Produktionskosten, Verbraucherpreise und spezifischer Energieverbrauch von effizienten Energieverbrauchstechnologien verringern. Vorliegende Dissertation will einen Beitrag zur Erweiterung des allgemeinen Kenntnisstands im Hinblick auf die genannten Wissensdefizite leisten. Dazu sollen folgende Fragen beantwortet werden: (i) Wie können die Emissionen aus dem nicht-energetischen Verbrauch fossiler Energieträger exakter und vollständiger erfasst werden? (ii) Welche Bedeutung hat technologisches Lernen für die Reduzierung von Produktionskosten, Verbraucherpreisen sowie des spezifischen Energieverbrauchs von Technologien des Endenergieverbrauchs?

Im ersten Teil dieser Dissertation werden bottom-up Modelle entwickelt, die den nicht-energetischen Verbrauch sowie die daraus resultierenden Kohlendioxid-Emissionen mit unterschiedlichem Detaillierungsgrad erfassen. In Kapitel 2 ermitteln wir, dass im Jahr 2000 weltweit 20 ± 2 EJ (Exajoule) fossile Energieträger für nicht-energetische Zwecke (zum Beispiel zur Herstellung von Chemikalien, Bitumen, oder Schmiermitteln) genutzt wurden. Damit hat der nicht-energetische Verbrauch einen Anteil von etwa 5% am globalen Primärenergieverbrauch. Für 12 Länder stimmen die in dieser Arbeit durchgeführten Berechnungen mit den Daten aus der internationalen Energiebilanz überein. Bei 26 Ländern sowie hinsichtlich des gesamten weltweiten nicht-energetischen Verbrauches, sind die vorliegenden Ergebnisse niedriger als die Daten aus der internationalen Energiebilanz. Für weitere 12 Länder liegen unsere Resultate über den Werten aus der internationalen Energiebilanz.

207 Zusammenfassung

Wir erklären die gefundenen Abweichungen mit Fehlern in der internationalen Energiebilanz sowie mit Inkonsistenzen in der Definition des nicht-energetischen Verbrauchs. Die Internationale Energieagentur (IEA – International Energy Agency) fordert bei ihren Erhebungen, fossile Energieträger, die in chemischen Prozessen für energetische Zwecke genutzt werden, als energetischen Verbrauch auszuweisen. Im Gegensatz dazu, deuten unsere Analysen darauf hin, dass ein Großteil der Länder dieser Forderung nicht folgt und die Nutzung fossiler Energieträger zur Erzeugung von Prozessenergie als nicht-energetischen Verbrauch deklariert.

Kapitel 3 zeigt, dass der nicht-energetische Verbrauch im Jahr 2000 weltweit zu Emissionen von 700 ± 90 Mt CO2 geführt hat. Industrieländer, die im Anhang I der Klima- Rahmenkonvention der Vereinten Nationen aufgeführt werden, sind für 51% (360 ± 50 Mt CO2) dieser Emissionen verantwortlich. Schwellen- und Entwicklungs- länder, haben einen Anteil von 49% (340 ± 70 Mt CO2). Etwa drei Viertel der globalen Emissionen aus dem nicht-energetischen Verbrauch entstehen bei Industrieprozessen, der restliche Teil wird durch die Nutzung von Produkten, durch Pestizid- und Düngemitteleinsatz in der Landwirtschaft, sowie bei der Abwasserreinigung freigesetzt.

In Kapitel 4 wird das detaillierte und erweiterte NEAT (Non-energy use Emission Accounting Tables) Model angewandt, um den nicht-energetischen Verbrauch sowie die daraus resultierenden Emissionen für Deutschland in der Periode von 1990-2003 zu berechnen. Die Emissionen aus dem nicht-energetischen Verbrauch fossiler Energieträger steigen im Durchschnitt mit etwa 1.8% pro Jahr (von 22.0 ± 1.8 Mt CO2 im Jahr 1990 bis 27.7 ± 2.3 Mt CO2 im Jahr 2003). Etwa 77% dieser Emissionen entstehen in Industrieprozessen, 17% bei der Nutzung von Lösemitteln und anderen Produkten, 2% durch die Verwendung von Pestiziden und Harnstoffdüngern in der Landwirtschaft sowie 4% bei der Abwasserbehandlung. Unsere Abschätzungen zu den Emissionen aus der Herstellung von Methanol, Ruß sowie aus Umwandlungsprozessen in der chemischen Industrie wurden vom Umweltbundesamt genutzt, um Datenlücken im deutschen Treibhausgasinventar zu schließen. Darüber hinaus konnten durch unsere Arbeiten Lücken im deutschen Treibhausgasinventar im Hinblick auf Emissionen aus Crack-Anlagen, aus der Herstellung verschiedener Nichteisen-Metalle so wie aus der Verwendung von Schmierstoffen und Wachsen nachgewiesen werden.

Die vorliegende Arbeit zeigt, dass der Erfassung des nicht-energetischen Verbrauchs sowie der daraus resultierenden Kohlendioxid-Emissionen international wesentlich mehr Aufmerksamkeit gewidmet werden muss. Zugleich wird erstmals eine detaillierte Abschätzung der nicht-energetischen Emissionen in Schwellen- und Entwicklungsländern vorgenommen. Unsere bottom-up Modelle ermöglichen ein tiefer gehendes Verständnis des nicht-energetischen Verbrauchs gegenüber bisherigen Methoden und sind daher geeignet, Daten in internationalen und nationalen Energiebilanzen und Treibhausgasinventaren zu überprüfen und gegebenenfalls zu korrigieren.

208 Zusammenfassung

Folgende Empfehlungen für Energiestatistiker und Inventarexperten werden abgeleitet: (i) Um den nicht-energetischen Verbrauch korrekt und international harmonisiert in den Energiebilanzen zu erfassen, sind verbindliche Vorgaben erforderlich. Insgesamt muss die Qualität von Energiebilanzen und offiziellen Produktionsstatistiken dringend verbessert werden, um den gestiegenen Anforderungen an die Erfassung von Treibhausgas-Emissionen zu entsprechen. (ii) Um Inkonsistenzen in der Definition des nicht-energetischen Verbrauchs zu beheben, muss die Internationale Energieagentur detailliertere Richtlinien erstellen, welche auch in den nationalen Energiebilanzen strickt befolgt werden müssen. Nur so kann eine Vergleichbarkeit von internationalen Daten gewährleistet werden. (iii) Die Erstellung von Energiebilanzen und Treibhausgasinventaren sollte enger verzahnt werden, um sicherzustellen, dass Emissionen aus der Nutzung fossiler Energieträger vollständig und richtig erfasst werden. (iv) Inventarexperten benötigen eine detaillierte Kenntnis der Definition des nicht-energetischen Verbrauchs, um Emissionen korrekt und vollständig in den jeweils relevanten Quellkategorien (zum Beispiel Industrie-prozesse, Lösemittel und sonstige Produktnutzung, Landwirtschaft, und Abfall) zu ermitteln. (v) Obwohl die Richtlinien zur Emissionsberichterstattung aus dem Jahr 2006 den IPCC Referenz Ansatz (IPCC-RA) nicht mehr zur Erfassung von Emissionen aus dem nicht-energetischen Verbrauch vorsehen, wird empfohlen, diesen weiterhin zur Überprüfung von Emissions-abschätzungen zu verwenden. Dadurch kann weitestgehend sichergestellt werden, dass Emissionen aus dem nichtenergetischen Verbrauch vollständig und exakt erfasst werden. (vi) Die Anwendung der in dieser Arbeit genutzten Modelle zur Abschätzung des nicht-energetischen Verbrauches sowie der sich daraus ergebenden Emissionen wird generell empfohlen.

Im zweiten Teil dieser Arbeit wird das technologische Lernen bei der Herstellung von Energieverbrauchstechnologien mit Hilfe so genannter Lernkurven analysiert. Lernkurven quantifizieren diesen Prozess mit Hilfe so genannter Lernraten, welche angeben, um welchen Prozentsatz sich Produktionskosten oder Preise einer Technologie verringern, nachdem eine Verdopplung der kumulierten Produktion erreicht wurde. Die durchgeführten Analysen zeigen, (i) dass Lernkurven prinzipiell geeignet sind, technologisches Lernen von Technologien des Endenergieverbrauchs zu quantifizieren sowie (ii) dass diese Technologien einem generellen Trend zu niedrigeren Produktionskosten und Preisen unterliegen.

In Kapitel 5 berechnen wir, dass sich die Preise von Kondensationsgas-Boilern, die in den Niederlanden zur kombinierten Warmwasserproduktion und Raumheizung eingesetzt werden, mit einer Lernrate von 14 ± 1% im Zeitraum von 1988-2006 verringert haben. Auch bei Haushaltsgeräten finden wir einen robusten Langzeittrend zu niedrigeren

209 Zusammenfassung

Preisen (siehe Kapitel 6). Waschmaschinen, Wäschetrockner und Geschirrspülmaschinen weisen dabei im Durchschnitt mit 29 ± 8% wesentlich höhere Lernraten auf, als Kühl- und Gefrierschränke mit 9 ± 4%. Eine mögliche Erklärung für diese Abweichung könnte in der unterschiedlichen technischen Komplexität der beiden Gerätegruppen liegen und darüber hinaus in unterschiedlichen Potenzialen für die Automatisierung bei der Herstellung von Komponenten und Endprodukten begründet sein. Die Ergebnisse für Kondensationsgas-Boiler und Haushaltsgeräte liegen im Wertebereich von Lernraten, wie sie für eine größere Gruppe von Technologien des Endenergieverbrauchs ermittelt wurden. Unsere Literaturstudie in Kapitel 7 ergibt, dass Lernraten für Energieverbrauchs- technologien um einen Mittelwert von 18 ± 9% normalverteilt sind. Abweichungen in den Lernraten, wie sie für identische Technologien ermittelt wurden, sind oft genauso hoch, wie Abweichungen für unterschiedliche Technologien. Es ist daher nur sehr begrenzt möglich, exakte Lernraten für neue Technologien zu ermitteln, ohne vorher detaillierte Lernkurven für diese Technologien zu erstellen.

Für Haushaltsgeräte wird der herkömmliche Lernkurvenansatz erweitert, in dem wir den spezifischen Energieverbrauch als Potenzfunktion der kumulierten Produktion darstellen. Die dabei resultierenden Energie-Lernkurven zeigen, dass sich die Effizienz von Waschmaschinen, Wäschetrocknern und Geschirrspülmaschinen mit Lernraten von 18 ± 3% bis 35 ± 3% steigert, während die Lernraten für Kühl- und Gefrierschränke etwas niedriger bei 13 ± 3% und 17 ± 2% liegen.

Technologisches Lernen führt zu substanziellen Preisabnahmen und Effizienzsteigerungen von Technologien des Endenergieverbrauchs. Dieses Ergebnis bedeutet, dass neue, und anfänglich teure aber effiziente Energieverbrauchstechnologien mittel- bis langfristig nicht notwendigerweise zu höheren Kosten für Konsumenten führen müssen. Gestützt wird diese Schlussfolgerung von den Ergebnissen unserer Kosten- Nutzenanalyse für Kondensationsboiler. Der jährlich realisierte Wert für Investitionen in Kondensationsgas-Kombiboilern war im Jahr 1988 mit -330 EUR2006 negativ, stieg aber bis 2006 auf 970 EUR2006. Zwei Drittel dieses Anstiegs resultieren aus technologischem Lernen während das verbleibende Drittel mit einer Kombination von externen Effekten (zum Beispiel der Abnahme des Wärmebedarfs in Haushalten) sowie steigenden Energiesteuern erklärt werden kann. Aus sozialer Perspektive haben sich die Kosten zur Emissionsvermeidung durch Kondensationsgas-Kombiboiler von 60 EUR2006/t CO2 im Jahr 1988 auf -120 EUR/t CO2 im Jahr 2006 verringert. Unsere Ergebnisse zeigen, dass begrenzte Subventionen von etwa 70 ± 10 Million Euro zur Verbesserung der Kosten- Nutzen Bilanz von Kondensationsgasboilers beigetragen haben. Subventionen haben darüber hinaus ein Mehrfaches an Ersparnissen von Energiekosten bei den Konsumenten ermöglicht und wesentlich zur Verminderung von Emissionen in holländischen Haushalten beigetragen.

210 Zusammenfassung

Obwohl die vorgenommenen Lernkurvenanalysen wesentliche Vorteile gegenüber einfachen Zeitreihenanalysen bieten, so unterliegen unsere Ergebnisse doch methodischen und empirischen Unsicherheiten. Insbesondere erfordern Lernkurvenanalysen zusätzliche technologische Detailanalysen, um Mechanismen und Komponenten, die zur Preis- und Kostenreduktion auf Technologieebene führen, besser erklären zu können. Die Untersuchungen zum technologischen Lernen von Technologien des Endenergie- verbrauchs lassen sich mit den folgenden Empfehlungen für Wissenschaftler und Entscheidungsträger in Politik und Wirtschaft zusammenfassen: (i) Technologisches Lernen von Energieverbrauchstechnologien sollte von Energiemodellen und Energiepolitik berücksichtigt werden, weil dieser Prozess mittel- und langfristig wesentlich zur Abnahme von Produktionskosten und Verbraucherpreisen beiträgt. (ii) Lernraten sind immer zusammen mit ihrem Fehlerintervall anzugeben, da schon geringe Unsicherheiten zu wesentlichen Abweichungen in mittel- und langfristigen Kostenprojektionen führen. (iii) Das Modellieren von Energieeffizienz mit Energie-Lernkurven erlaubt neue Einsichten in die Effizienzdynamik von Energieverbrauchs-technologien. Das Potenzial von Energie-Lernkurven sollte auf allen Gebieten des Energieverbrauchs in der Industrie, im Transportsektor sowie in privaten Haushalten weiter erkundet werden.

Abschließend bleibt festzuhalten, dass die Erhöhung der Energieeffizienz eine wesentliche Komponente auf dem Weg zu einem nachhaltigen globalen Energiesystem darstellt. Technologisches Lernen kann dabei helfen, ökonomisch begründete Effizientpotenziale zu tendenziell abnehmenden Kosten zu realisieren. Um allerdings eine Reduktion des absoluten Energieverbrauchs zu erzielen, ist die verstärkte Nutzung von anfänglich teureren und ökonomisch wenig attraktiven Effizienztechnologien notwendig. Um dies zu gewährleisten sind nachhaltige Entscheidungen in Politik und Wirtschaft, zielgerichtete Subventionsmaßnahmen und eine ökologische Bewusstseinsbildung bei den Konsumenten erforderlich.

211

List of abbreviations and units

a - year av. - average oC - degree centigrade CFL(s) - compact fluorescent light bulb(s) CH - Switzerland CO2 - carbon dioxide const. - constant cont. - continued cum. - cumulative DRAM - dynamic random access memory EJ - exajoule EPROM - erasable programmable read only memory E - Europe EAF - electric arc furnace eds. - editors EU-15 - European Union, consisting of 15 member states EUR - Euro EUR2006 - real Euro deflated to the base year 2006 excl. - excluding FL - fluorescent light bulbs GDP - gross domestic product GER - Germany GHG - greenhouse gas GJ - Gigajoule Glm - gigalumen Gm3 - giga cubic meter GWh - gigawatt hour hl - hectoliter ICH - individually and centrally heated IEA - International Energy Agency inst. - installed INT - international data inv. - investment IPCC - Intergovernmental Panel on Climate Change IPCC-RA - IPCC-Reference Approach IPCC-SA - IPCC-Sectoral Approach JPN - Japan K - kelvin kB - kilobyte kg - kilogram klm - kilolumen

213 Abbreviations and units kt - kilotonne ktoe - kilotonne oil equivalent kWel - kilowatt electric kWth - kilowatt thermal kWhel - kilowatt hour electric kWhth - kilowatt hour thermal l.c. - laundry capacity LR - learning rate LSI - large-scale integration m2 - square meter m3 - cubic meter MB - megabyte MEUR - million Euro MJ - megajoule MOS - metal oxide semiconductor Mt - megatonne MWe - megawatt electric n - sample size NEU-CO2 - non-energy use and CO2 emissions NEAT - non-energy use emission accounting tables NL - the Netherlands NLG - Dutch guilders NMVOC - non-methane organic compound n.s. - not specified PE - Plastics Europe PJ - petajoule ppm - parts per million PR - progress ratio prod. - production RAM - random access memory R&D - research and development ROK - Republic of Korea (South Korea) SPS - standard place setting t - tonne TFEC - total final energy consumption TPES - total primary energy supply TV - television UNEP - United Nations Environmental Programme UNFCCC - United Nations Framework Convention on Climate USD - United States dollars vs. - versus W - watt Wel - watt electric WMO - World Meteorological Organization

214

Acknowledgements

Running on the trail of science towards this dissertation has been a true challenge for me. To reach the finish line, I received assistance and motivation from many people. First, I would like to thank my supervisor, Kornelis Blok, who taught me how to write a scientific article. Our meetings sharpened my thinking and writing. I am also grateful to my co- supervisors, Martin K. Patel and Martin Junginger, who introduced me to the research presented in this thesis and who guided my work in several joint research projects.

I thank Klaas-Jan Koops (now working at the Ministry of Housing, Spatial Planning and the Environment, The Hague, the Netherlands) and Micheal Strogies (Federal Environmental Agency, Dessau, Germany) for the fruitful cooperation in the course of two independent research projects. I am grateful to experts, whose ideas and comments provided important contributions to this thesis, especially to René Kemna (Van Holsteijn and Kemna B.V.), Ton van Maaren (Remeha B.V.), Jos Olivier (Netherlands Environmental Assessment Agency), Hans Overdiep (GasTerra B.V), Henk Seijbring (H. Seijbring B.V.), Hans-Paul Siderius (SenterNovem), Willy Taelman (Philips Lighting), and Paul Vloon (Nefit B.V.).

During the past five years, I shared my office with Maarten Neelis and Deer Saygn, who facilitated my work in multiple ways: Maarten, thanks for explaining non- energy use and the NEAT model to me; Deer, thanks for proofreading my articles, for introducing me to the advanced internet, to MS-Office, to the Turkish cuisine, and foremost for being a friend. I would like to thank Aisha, Edward, Jules, and Ric for translating the summary of this thesis into Dutch, and Birka, Ju, Berthold, Deer, Stevie, and Takeshi for proofreading parts of the text.

Thanks to my colleagues Andrea, Barbara, Birka, Floor, Janske, Jenssie, Jinke, Li, Machteld, Pita, Roos, André, Arie, Arjan, Bert, Bothwell, Edward, Erik Alsema, Erik Lysen, Ernst, Evert, Hans, Jeroen, Joris, Jos, Lex, Loek, Marc, Martijn, Michiel, Nils, Oscar, Ric, Richard, Rob, Sander, Takeshi, Tao, Wilfried, and Wim at Science, Technology and Society for the pleasant working atmosphere. Special credits go to the ladies at the secretariat office (Aisha, Cosy, Mirjam, Siham, Petra, and Sylvia) for their organizational support. I am grateful to Ernst Ulrich von Weizsäcker und Roland Geyer for enabling my research visit at the Donald Bren School of Environmental Science and Management, University of California at Santa Barbara (USA). Thanks to Maggie, Brandon, and Francesco for making my stay a pleasant Californian experience.

I would like to thank all my friends who shared with me non-academic joys in Utrecht. Special thanks go to the runners (Barbara, Birka, Li, Hans, Martin Junginger, Takeshi) and to the bikers (Barbara, Floor, Edward, Roald, Sjoerd) for joining the anti- aging program. Thanks also to Emilia, Barbara, Bogdana, Juliane, Raluca, Wiebke, Xin, Deer, Edward, Ric, Roald, Takeshi, Angeles and Watse, Birka and Joseph, Li and Peter, Nidhi and Partesh, Pelin and Edwin, Sonia and Ryan, all former housemates at

215 Acknowledgements

Schutstraat 84, as well as to the close friends Sara and Anand for making Utrecht a home. I thank the meanwhile old friends Deborah, Denise and Thomas, Patricia and Christian, and Mirko for staying in touch after all these years.

Alex, thanks for proofreading my articles and even more for joining in the miles of trials and trials of miles. I know that the ultimate interval is still ahead of us! I thank my parents, Elke and Berthold, my grandparents Ursel and Manfred, and my uncle Franz for making their homes a safe haven. Stevie, thanks for great vacations and good discussions in all those years. Many thanks and much love to Ju; your enthusiasm and affection inspire me.

Finally, I would like to especially thank Martin K. Patel. From my days at the Wuppertal Institute to the most recent evening hours, you are always willing to listen, to help, or to provide comments, regardless of how high work piles up on your desk. Your friendliness and reliability are characteristics, which others sacrifice first in these competitive and often opportunistic times. I value the experience of working with you beyond everything that I have learned in the past five years.

216

Curriculum vitae

I was born on December 10, 1976 in Suhl, Germany. From 1982 to 1995, I received primary and secondary education in Maputo (Mozambique) as well as in Suhl and Oberhof (Germany). In 1997, I began to study environmental sciences. I received a M.Sc. degree in environmental science and engineering from the Colorado School of Mines (USA) in 2001 and a M.Sc. degree in geo-ecology from the Technische Universität Bergakademie Freiberg (Germany) in 2004. Between 1997 and 2004, I worked several months as a student assistant and intern at the ENVIAm Energy AG (Markleeberg, Germany), the Federal Ministry for Nature Protection and Nuclear Safety (Berlin, Germany), and the Wuppertal Institute for Climate, Environment, and Energy (Wuppertal, Germany). Since August 2004, I have been a junior researcher at Utrecht University (the Netherlands). Under daily supervision of Dr. Martin Patel, I conduct research on (i) carbon emissions resulting from the non-energy use of fossil fuels, (ii) decoupling of physical and economic activity in the bulk materials industry, and (iii) technological learning of energy demand technologies. In early 2009, I worked as a visiting researcher at the Donald Bren School for Environmental Science and Management, University of California at Santa Barbara (USA). In my free time, I enjoy running, biking, and reading.

217 Coal, Reducing agents for the

Lignite Coke oven production of non- Learning accounting in carbon and energy efficiency ferrous metals, ferroalloys,Learning and in Tars, Pitch inorganic chemicalscarbon accounting Production of non-ferrous Electrodes metals, ferroalloys,and and energy efficiency inorganic chemicals Martin Weiss Aromatics Petroleum large appliances industrial process emissions condensing gas boilers carboncoke dioxides solvent and other product use experience curvesPyga refineriess production costs steam crackers energy demand technologies sustainable energy system greenhouse gas inventories non-energy use technological learning chemical industry electricity learning rates surfactants energy experienceence curves polymerspo y e households energyener statisticsstat ammonia production developing countries energy policy naphtha experience ImportImport EthaneEthane Organic Ethylene chemicals Crude NaphthNaphtha htha SteamS production oil Refineryery Propylene BitumenBitumen crackea r

Other C4 ExportExport LubricantsLubricants InternalIntern fuel use

Liquid petroleumpetroleumt gas (LPG)(LLPG) Martin Martin Weiss HeavyHeavy oils

Gasoline, KeKerosene,rosenerosene,e,, Dieselies Natural gas Energy production