Statistical Characteristics and Mapping of Near-Surface and Elevated Wind Resources in the Middle East

Statistical characteristics and mapping of near-surface and elevated wind resources in the Middle East

Dissertation by

Chak Man Andrew Yip

In Partial Fulﬁllment of the Requirements

For the Degree of

Doctor of Philosophy

King Abdullah University of Science and Technology

Thuwal, Kingdom of Saudi Arabia

November, 2018 2

EXAMINATION COMMITTEE PAGE

The dissertation of Chak Man Andrew Yip is approved by the examination committee

Committee Chairperson: Georgiy L. Stenchikov Committee Members: Marc G. Genton, Gerard T. Schuster, Kristopher B. Kar- nauskas 3

©November, 2018 Chak Man Andrew Yip

ABSTRACT

Statistical characteristics and mapping of near-surface and elevated wind resources in the Middle East Chak Man Andrew Yip

Wind energy is expected to contribute to alleviating the rise in energy demand in the Middle East that is driven by population growth and industrial development. However, variability and intermittency in the wind resource present significant challenges to grid integration of wind energy systems. The first chapter addresses the issues in current wind resource assessment in the Middle East due to sparse meteorological observations with varying record lengths. The wind field with consistent space-time resolution for over three decades at three hub heights over the whole Arabian Peninsula is constructed using the Modern Era Retrospective-Analysis for Research and Applications (MERRA) dataset. The wind resource is assessed at a higher spatial resolution with metrics of temporal variations in the wind than in prior studies. Previously unrecognized locations of interest with high wind abundance and low variability and intermittency have been identified in this study and confirmed by recent on-site observations. The second chapter explores high-altitude wind resources that may provide alternative energy resources in this fossil-fuel-dependent region. This study identifies areas favorable to the deployment of airborne wind energy (AWE) systems in the Middle East and computes the optimal heights at which such systems would best operate. AWE potential is estimated using realistic AWE system specifications and assumptions about deployment scenarios and is compared with the near-surface wind generation potential concerning diurnal and seasonal variability. The results show 5 the potential utility of AWE in areas in the Middle East where the energy demand is high. The third chapter investigates the potential for wind energy to provide a continuous energy supply in the region. We characterize the wind power variability at various time-scales of power operations to illustrate its effects across the Middle East via spectral analysis and clustering. Using a high-resolution dataset obtained from Weather Forecasting and Research (WRF) model simulations, this study showcases how aggregate variability may impact operation, and informs the planning of large- scale wind power integration in the Middle East in light of the scarcity of observational data. 6

ACKNOWLEDGEMENTS

The research reported in this dissertation was supported by King Abdullah Uni- versity of Science and Technology (KAUST) and Saudi Basic Industries Corporation (SABIC) under grant number RGC/3/1815-01. I want to thank my advisor Prof. Georgiy Stenchikov for his wisdom, kindness, and patience, and Dr. Udaya Bhaskar Gunturu for the many collaborative thoughts and actions throughout my research journey. I’d also thank Dr. Stoitchko Kalenderski for the historical simulations on WRF, Dr. Suleiman Mostamandi for his help in the setup and conﬁguration of the simulations, and Dr. Alexander Ukhov for his help in computing environment. I have my parents to thank for their unlimited support always and my dear for saving me in the many last-minute crises throughout my time at KAUST. 7

TABLE OF CONTENTS

Examination Committee Page 2

Abstract 4

Acknowledgements 6

List of Figures 9

List of Tables 14

1 Introduction 15 1.1 Near-surface wind resource assessment ...... 18 1.2 Elevated wind resource assessment ...... 19 1.3 High-resolution simulation of wind resource and spatial analysis . . . 20

2 Surface wind resource analysis 21 2.1 Overview ...... 22 2.2 Methods ...... 25 2.2.1 Data ...... 25 2.2.2 Metrics ...... 27 2.3 Results ...... 30 2.3.1 Abundance ...... 30 2.3.2 Variability ...... 40 2.3.3 Intermittency ...... 40 2.4 Discussions ...... 43

3 Airborne wind resource analysis 52 3.1 Overview ...... 52 3.2 Methods ...... 54 3.2.1 Wind Speed Maxima (WSM) ...... 56 3.2.2 Altitudes of WSM ...... 56 8 3.2.3 Capacity factor ...... 57 3.2.4 Deployment assumptions ...... 58 3.3 Results ...... 58 3.3.1 Average WSM ...... 59 3.3.2 Variability in WSM ...... 61 3.3.3 Altitude of WSM ...... 63 3.3.4 Variability in WSM altitudes ...... 63 3.3.5 Regional AWE potential ...... 64 3.3.6 Comparison with near-surface wind resources ...... 65 3.4 Discussions ...... 68

4 Variability mitigation with spatial clustering 70 4.1 Overview ...... 70 4.2 Methods ...... 74 4.2.1 Data ...... 74 4.2.2 Methods ...... 75 4.3 Results ...... 77 4.3.1 Dominant time scales of variability ...... 78 4.3.2 Clustering analysis ...... 80 4.3.3 Cluster variance ...... 80 4.3.4 Clustering characteristics ...... 82 4.3.5 In-cluster variability reduction ...... 82 4.4 Discussions ...... 84

5 Summary 87

References 90

Appendices 101 9

LIST OF FIGURES

2.1 Median WPD (W m−2) computed at 50 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 100 W m−2 and below 30 W m−2. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown...... 31 2.2 Categorized median WPD (fig. 2.1) at 50 m AGL into three regimes: regime I where median WPD is above 67 W m−2, regime II where median WPD is from 46 W m−2to67 W m−2, and regime III where median WPD is below 46 W m−2. Elevation contours at 400 m intervals are drawn using elevation data from MERRA. Selected locations of prior studies are shown...... 32 2.3 Average WPD (W m−2) computed at 50 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 200 W m−2 and below 60 W m−2. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown...... 33 2.4 Changes of median WPD (fig. 2.1) and average WPD (fig. 2.3) at 80 m and 140 m from those at 50 m AGL. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. The color scale is uniform except for values beyond 45%...... 37 2.5 Average wind speed (m s−1) computed at 100 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 6.6 m s−1 and below 2.4 m s−1. Selected locations of prior studies are shown...... 38 2.6 Differences in the average wind speed (m s−1) at 100 m AGL between our calculation and that from estimation of Vestas (Vestas - MERRA) are illustrated with a color scale that is uniform except for values beyond 2.5 m s−1 and below −2.5 m s−1. Selected locations of prior studies are shown. The MERRA dataset is regridded to 1 km resolution by nearest neighbor for illustration...... 39 10 2.7 Robust coefficient of variation computed at 50 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 1.45 and below 1.15. Selected locations of prior studies are shown...... 41 2.8 Availability of wind resource computed at 50 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 0.45 and below 0.1. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown...... 42 2.9 Changes of availability (fig. 2.8) at 80 m and 140 m from those at 50 m AGL. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. The color scale is uniform except for values beyond 40% and below 15%...... 43 2.10 Median length of wind episode (h) computed at 50 m, 80 m, and 140 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 10 h. Selected locations of prior studies are shown...... 44 2.11 Country map in the Middle East ...... 46 2.12 Illustration of a skewed distribution of WPD at Yanbu ...... 47 2.13 Average wind speed (m s−1) is computed at 50 m AGL using the wind fields reconstructed from the MERRA data. Selected locations of prior studies (table 2.1) are shown...... 48 2.14 Fractional change in WPD at 80 m and 140 m from 50 m over temporal median of roughness lengths ...... 49 2.15 Distribution of fractional changes in rCV at 80 m and 140 m from 50 m out of 1638 grid cells ...... 50 2.16 Distribution of differences in MEL at 80 m and 140 m from 50 m out of 1638 grid cells ...... 51

3.1 (a) Definition of wind speed maxima (WSM) and its associated altitude (H), where the size of the box indicates the magnitude of the wind speed (b) Power curve, power (kW) as a function of wind speed (m/s), of a lift-type groud-based AWES of 3 MW rating. The figure and all the text elements in (b) has been plotted using the software R version 3.3.2. 55 11 3.2 Average wind speed (m/s) of WSM during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. CV in the wind speed of WSM during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] Panels (a-d) share a uniform color scale from 5 to 13 m/s; values larger or smaller than the boundaries share the bounding colors. These boundaries are informed by the wind speed distributions (Fig. S1) and the wind speed maxima vertical profiles (profiles at selected locations are illustrated in Fig. S2)...... 60 3.3 Average altitude (meters above sea level, mASL) of WSM (H) during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. CV in the altitude of WSM (H) during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] . . 62 3.4 Map of AWE potential. The CF of the 3 MW AWE system (left) and annual AWE generation per capita (kWh) by country (right). The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] ...... 64 3.5 Difference in the average wind speed (m/s) between WSM and WS at 80 m during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. The difference in CV in wind speed between WSM and WS at 80 m during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] ...... 67

4.1 Transmission network map of the Middle East, note the lack of transmission across the border of Yemen and Oman (World Bank Group, 2017) ...... 77 4.2 Fraction of variance (color) accounted for in each temporal scale in wind power generation based on an N100 power curve at 100m above ground level ...... 78 12 4.3 Wind power generating locations are ﬁltered by capacity factor at 0.3 and clustered spatially using hierarchical clustering upon dissimilarities based on temporal correlations of the locations and their geographical distance at a mixing ratio of 0.5 towards geographical distance, resulting in 9 clusters (color) ...... 79 4.4 Fraction of variance (size, y-axis) accounted for in each temporal scale (x-axis) in aggregate wind power generation by cluster (color) based on an N100 power curve at 100m above ground level, where bubble size represents cluster annual generation potential, the grey bubble indicates variability by aggregating all generating clusters (aggregate generation at 40000 GWh a year) ...... 81 4.5 Fraction of variance in time-bands for each cluster where cluster-aggregated generation (agg) is compared to a cell with maximum fraction of variance (max) in each cluster...... 83

A.1 Spatial domain setup of the simulations: d01 (9 km x 9 km), d02 (3 km x 3 km), and d03 (1 km x 1 km) ...... 104 A.2 Average wind speed (m/s) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 105 A.3 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 106 A.4 Coefficient of variation in wind speed at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 107 A.5 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 108 A.6 Average wind speed (m/s) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 110 A.7 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 111 A.8 Coefficient of variation in wind speed at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 112 A.9 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 113 A.10 Average wind power generation (kW) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 114 13 A.11 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 115 A.12 Percentage difference in average power generation at 80m from 100m in WRF ...... 116 A.13 Percentage difference in average power generation at 140m from 100m in WRF ...... 117 A.14 Coefficient of variation in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution . . . . 119 A.15 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 120 A.16 Percentage difference in coefficient of variation in power generation at 80m from 100m in WRF ...... 121 A.17 Percentage difference in coefficient of variation in power generation at 140m from 100m in WRF ...... 122 A.18 Availability in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution ...... 124 A.19 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 125 A.20 Percentage difference in availability in power generation at 80m from 100m in WRF ...... 126 A.21 Percentage difference in availability in power generation at 140m from 100m in WRF ...... 127 A.22 Average episode length in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution . . . . 129 A.23 Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation ...... 130 A.24 Percentage difference in average episode length in power generation at 80m from 100m in WRF ...... 131 A.25 Percentage difference in average episode length in power generation at 140m from 100m in WRF ...... 132 14

LIST OF TABLES

2.1 Selected locations of prior studies in the region as indicated on the maps...... 46

3.1 Consumption (C) and potential AWE generation (G) for countries in the Middle East (per capita) with the maximum number of deployable generators (MNG), density of generators per 100 km2 (GD), and the population (P). C and G data are for 2010 from IEA Statistics OECD/IEA 2014...... 66

4.1 Time-scales and their corresponding physical processes and operations impacted ...... 72 4.2 Characteristics of the 9-clustering with the amount of turbines proposed, total power generated in a year in GWh, and fraction of total consumption fulﬁlled is also computed using per capita electricity consumption data from World Bank 2013...... 82 4.3 Coeﬃcient of variation (CV) for aggregated power generation time series (clusterCv) and the spatial statistical summaries (min, mean, max) of CV of individual cells in a cluster ...... 86 15

Chapter 1

Introduction

Saudi Arabia is on the cusp of energy transition [2]. The kingdom is endowed with the largest oil resources in the world [3]. But an imminent need to diversify the sources of energy emanates from several fronts.

• Power consumption increases as the lifestyle of the population improves and migration to urban regions increases.

• Increasing atmospheric temperatures exponentially raise the demand for power for air conditioning.

• The availability of water is tied to the availability of power as power-intensive desalination caters up to 50% of the water usage.

• As the economy is transitioning from an oil-exporting one to a manufacturing and service-based one, industrial and business demand for power is bound to increase.

• Substitution of oil as a fuel for power generation will free up oil for export.

• National commitment to lower greenhouse gas emissions to mitigate climate change.

This transition is an opportunity for the kingdom to seek solutions for sustainable development. 16 Wind and solar harvesting technologies constitute the dominant proportion of global renewable energy as both are spatially well distributed and are sustainable [4]. Between them, solar (Photovoltaic, PV) technology has low eﬃciency and the technology is still under active research and development [5]. But the wind turbines are technologically mature, performing near 80% of their theoretical maximum eﬃ- ciency, which is set by the Betzs limit (≈ 60%) in converting wind power density to electric power [6]. Further, technological advancements in wind turbines have made wind power attain near grid parity and compete with conventional generation [7]. Another merit of wind power compared to solar power is the high power density in unit cross-sectional area. Wind power, and renewable power, in general, have the following hurdles/disadvantages [8]:

• Variability. Conventional power generation is highly controllable. Hence, planning the energy system, generation adequacy and changing the generation proﬁle to match the variable demand proﬁle has been perfected over several decades.

• Wind power generation is intermittent, that is wind turbine does not generate any power due to low, or extremely high wind speed. In such situations, there is need for compensating the lack of wind generation from other generation technology.

• Wind resource is location dependent and expensive transmission needs to be built from the location of generation to the grid or to the location of consumption.

While the high performance and efficiency of wind turbines lowered the generation cost, these demerits increase the true cost of wind power. In fact, the countries and regions that have the increasing proportion of wind power have the highest consumer power tariff. 17 There are two main types of conventional ground-based turbines – those with a horizontal rotating axis (Horizontal Axis Wind Turbine, HAWT) and those with a vertical rotating axis (Vertical Axis Wind Turbine, VAWT)[9]. The HAWT are preferred in regions with high wind resource as they offer high cross-section to the wind and can generate large amounts of power. The VAWT usually can generate power even at low wind speeds and are preferred in low wind speed regions. But the topography and land-surface processes damp wind in the lower boundary layer of the atmosphere resulting in fluctuating and intermittent wind flow [6]. Since the turbines have high performance and availability (¿ 95%), the generated wind power is variable because of the variability of the resource. Also, because of the near- logarithmic vertical wind profile in the atmosphere, as the turbine height increases, the impact of the topography and land-surface processes diminishes and the abundance of the wind increases and the variability decreases. Therefore, the hub height of the HAWT has been increasing over the last decade and a half to catch the abundant and less variable wind at higher altitudes. As a result, the capital cost of wind turbine deployment of these bigger turbines is on the rise [10]. On the other hand, several research groups have been experimenting with harvesting devices floating in the air at different altitudes – kites and other devices. KiteGen, Makani (acquired by Google) are cases in point [11]. This dissertation aims to improve our understanding of the near-surface and elevated wind resources to facilitate large-scale deployment of wind turbines for harvesting wind. To this end, the wind resource in the region needs to be characterized in terms of abundance, variability, and intermittency.

• Although the industry uses wind speed as a measure of abundance, we also compute wind power density which is a more robust metric. We develop maps of abundance near the surface, in the boundary layer and in the free atmosphere.

• The variability of the wind resource is the most important metric of the resource 18 as it dictates the cost of wind power and also is a determining factor in how the non-wind part of the power generation has to be changed so that we have a steady power supply to meet the demand. The largest change in the wind is due to the diurnal cycle and the annual or seasonal cycle. Further, there are atmospheric forces that make wind vary on scales less than a season and year-to-year. Although grid operations and availability of power are impacted in the short term, the interannual variability has a large impact on the life cycle costs of wind power. Thus, the long-term viability of wind deployment is very strongly tied to interannual variability.

• We diﬀerentiate between the variability of wind and its intermittency. Inter- mittency is how the wind stops and blows. It is important to take this into consideration because a compensating power supply has to be switched on or ramped up which will increase the cost of wind power and also will impact the stability of the grid. We describe the intermittency in terms of two metrics: episode lengths (how long does wind blow at a stretch before it stops), and for what fraction of is there no wind at the location.

• Lastly, we know that as elevation from surface increases, the wind speed increases. This is generally true, but the proportion of increase in the resource may not justify the increase in deployment cost. So, we look at the above metrics at diﬀerent hub heights and assess the advantages and disadvantages.

This dissertation has three parts:

1.1 Near-surface wind resource assessment

The first part of the dissertation consists of analysis of data from different sources to assess the quality of wind resource in this region: an analysis of the MERRA reanalysis dataset for the surface resource. The MERRA reanalysis dataset is an 19 assimilation of observational data into the model to construct an optimum state of the atmosphere [12]. The data from surface stations, satellites, radiosondes and all other platforms is aggregated continuously and assimilated into a general circulation model to optimally estimate the state of the atmosphere. NASA generates this dataset. It stands for the Modern Era Retrospective-analysis for Research and Applications. It consists of hourly data from 1979 to now and has a spatial resolution of half a degree by two-thirds a degree. Using the MERRA reanalysis, we constructed a wind resource atlas for this region. Since the atlas is based on about 34 years of data, it will realistically represent the impact of the different atmospheric circulations like El-Nino on the wind resource in this region. If we have a shorter record, it will not be possible to assess the variability of the wind resource at interannual timescales. The constructed wind resource atlas was compared with the other available atlases of wind resource in this region. For instance, there is a Saudi wind atlas of 1986. Since there have been more robust estimates in the recent times, they were also being looked at. The study has been submitted for peer review and has been published in the journal Applied Energy [13]. It is represented in chapter 2.

1.2 Elevated wind resource assessment

The second part of the dissertation consists of assessing the airborne wind resource. The wind field data at different altitudes above the ground from MERRA dataset have been used to characterize the airborne wind resource in the MENA region. The characterization comprised of the identification of regions with abundant resources, seasonal variability of the resource, diurnal variability of the resource, and the altitudes at which the maximum resource occurs over each location in the region. As- suming a realistic airborne wind energy harvesting system, a system developed by KiteGen, the potential power that can be generated has been assessed. The results show that several regions in Saudi Arabia and Oman are endowed with potentially 20 harvestable airborne wind resource. The variability of this resource was also confirmed to be less than that of the surface wind resource in the region. The study has been submitted for peer review and has been published in Nature Scientific Reports. It is represented in chapter 3.

1.3 High-resolution simulation of wind resource and spatial analysis

The third part of the dissertation consists of the simulation of high-resolution surface and airborne wind resource and their assessment on the variability and intermittency as well as their potential mitigation through spatial aggregation. In previous experiments, the wind resource at diﬀerent altitudes, technologies, and resolutions have been assessed using the MERRA data that has a moderate resolution of 50 km × 67 km. Wind is generated, accelerated, and maintained by topographical features, such as mountains, sea coast, and land surface processes in the desert boundary layer and the surface. So, guided by the results in the previous experiments, surface-based and airborne wind resource has been simulated using the Weather Research and Forecasting model [14]. The simulation has been conducted for the years 2009 to 2014 at hourly time resolution and a spatial resolution of 5 km by 5 km, which is more than 100 times ﬁner than in the previous experiments. Using this dataset we study the wind resource near the surface and characterize the wind power generation variability at various timescales using the power curves of commercially available reference turbines. It illustrates how aggregate variability may impact operations and potentially inform the planning of large-scale wind power integration to the Middle Eastern energy mix. It is represented in chapter 4. 21

Chapter 2

Surface wind resource analysis

The potential adverse impacts of climate change and energy insecurity have encour- aged countries worldwide towards adopting renewable energy as an integral part of their future energy mix. Near-surface wind energy has the potential to power the world; it allows extracting energy at a rate of at least 400 TW [7]. It is suggested that following a moderate wind energy deployment plan by 2050 would delay the crossing of the 2 ◦C threshold for 1 to 6 years [15]. Wind energy provides a viable alternative energy source to energy intensive countries such as China, where it is estimated to be sufficient to replace 23% of the electricity generated from coal [16]. In the Middle East and North Africa (MENA), population growth has led to increases in demand for fuel and electricity for air-conditioning and desalination. Regional annual Total Primary Energy Supply (TPES) increased by 14.9% to 800 millions Mtoe (million tonnes of oil equivalent) in 2010 compared to the TPES of 2007 [17]. These steady increases in domestic consumption of energy drive the latest expansion of the renewable energy market [2]. Among net oil importers such as Jordan, energy insecurity and dependence on expensive oil imports have led to an expansion of the renewable energy program. Renewable energy has grown from 0.4 TW h in 2008 to 1.2 TW h in 2011 among net oil importers [17]. In the net oil exporting countries, renewable energy has grown from 0.8 TW h in 2008 to 1.6 TW h in 2011 [17]. This growth has been the result of rising opportunity cost of oil and gas accompanied by an increase in urbanization and a rapid rise in domestic demand for energy. These expansions are evident in the countries’ recent large-scale procurement of renewable energy systems to fulfill national renewable targets [2]. The surging interest in renewable energy calls for a better understanding of the spatial and temporal characteristics of the resource. This chapter focuses on the wind energy resource, the most variable and intermittent source of renewable energy in the Arabian Peninsula. There are two key challenges in assessing the wind resource in the Arabian Penin- 23 sula. Most of the observations available are sparse in space and inconsistent in time: spatially scattered observations with varying record lengths come from meteorological stations that are located mainly in clustered coastal and inland settlements. Hourly wind speeds were collected by 293 weather stations in the Peninsula during our period of study from 1979 to 2013. Among the stations, 42 collected data for at least half of the time. Only 17 stations have observations available for more than 80% of the record length [18]. Despite these challenges, Ansari et al. [19] constructed the Saudi Arabian Wind Energy Atlas in 1986 using hourly observations from 20 airport weather stations from 1970to1982. They described diurnal and seasonal variations of wind speed at measurement height at these locations and mapped prevailing wind directions. Rehman and Halawani [20] described diurnal, monthly, and inter-annual wind speed variations at 10 weather stations. Most of the studies focused on prominent sites of assessment that are mainly coastal. A similar tendency is observed for the MENA region [21, 22, 23], with the exception of Ohunakin et al. [24] where the focus was inland. These analyses concentrated on wind speed time series from meteorological stations with different record periods. The wind resource is frequently characterized by average wind speed or wind power density (WPD) that is at measurement height or is adjusted to hub height. Recent works have attempted to study the spatial variation of the wind resource. Jervase and Al-Lawati [25] performed an areal analysis of wind resource abundance in Oman using the NASA Surface Meteo-

rology and Solar Energy (SSE) Release 6.0 dataset with a spatial resolution of 1° × 1°. Al-Yahyai and Charabi assessed wind resource in Oman using a nested ensemble numerical weather prediction (NWP) approach, where two global models were used as boundary conditions to drive two local area models. The wind abundance has been assessed at the scale of a country [26] and a city [27]. Moreover, Charabi et al. [28] demonstrated that NWP models at 7 km are effective in resolving finer structures such as the sea breezes in this region. However, without well-formulated boundary 24 conditions based on a long-term and spatially and temporally consistent dataset, an NWP model would not capture the impact of large-scale circulations such as the El Nio. Since these circulations are of low frequency, they have higher spectral power and, therefore, have a significant impact on the wind resource. Existing meteorological observations contain missing data due to handling errors or malfunctioning of the instrument. These observations are spatially sparse, for instance, located mainly at airports and urban areas. These factors led to some potentially resource-rich regions being overlooked, which prevented prior studies in developing a comprehensive characterization of the wind resource for the entire Arabian Peninsula. Moreover, previous resource characterizations have focused on average wind abundance and annual energy production estimates. Wind power variability and intermittency present significant challenges to grid integration of wind energy systems, as identified by wind integration studies in the United States [29]. Variability and intermittency have been considered, most commonly using tower measurements where data are limited in the spatial and temporal dimensions [30, 31, 32]. Rehman and Halawani [20] provided a wind persistence measure via auto-correlation and auto-regression for ten weather stations. Rehman and Ahmad [33] presented a wind availability analysis for 5 coastal locations in Saudi Arabia in terms of frequency of wind speed within a specified interval. Wind speed time series from meteorological stations were fitted by Weibull distributions to investigate the monthly variation of wind speed and their changes with hub height in Saudi Arabia [34, 35] and Bahrain [36]. Ouarda et al. [37] fitted multiple distributions and assessed their goodness-of-fit with wind speed measurements in the United Arab Emirates (UAE). However, the variability and intermittency of the wind resource have not been studied in the entire Peninsular region. Recently, Chen et al. 2018 analyzed the wind power potential in Saudi Ara- bia using the Middle East North Africa Coordinated Regional Climate Downscaling Experiment (MENA CORDEX) model outputs [38]. 25 The primary goal of this study is to overcome the limitation of sparse station observations with varying record lengths by constructing the wind field using a gridded reanalysis dataset with a multi-decadal record period to arrive at a characterization of wind variability and intermittency. We characterize the wind resource using metrics proposed in Gunturu and Schlosser [8] (United States), Cosseron et al. [39] (Europe), Fant and Gunturu [40] (South Africa), and Hallgren et al. [41] (Australia). This work aims to answer the following questions:

• What methodology can be used to assess the wind energy resource in a region where observational data are sparse and non-concurrent? (section 2.2)

• Where are the areas with wind power potential that were not previously located due to lack of observations? (section 2.3.1)

• How do wind variability and intermittency diﬀer in spatial distributions from conventional metrics of resource abundance? (sections 2.3.2 and 2.3.3)

In the following sections, we explore regional wind resource and compare our results with results from prior studies. We ﬁrst describe wind resource abundance as characterized by the median WPD. Our wind ﬁeld reconstruction shows a qualitative agreement with previous studies. We then describe the regional wind resource using metrics of variability and intermittency.

2.2 Methods

2.2.1 Data

The spatial domain of interest spans the Arabian Peninsula bounded between 10°N and 35°N in latitude and 35°E and 60°E in longitude. This spatial extent allows investigation of the Red Sea, the Gulf of Aden, the Arabian Gulf, and part of the Arabian Sea along with inland areas. 26 The WPD ﬁeld is reconstructed using the Modern Era Retrospective-Analysis for Research and Applications (MERRA) dataset. MERRA is a reanalysis conducted by the Global Modeling and Assimilation Oﬃce (GMAO) at NASA using the God- dard Earth Observing System Version 5 (GEOS-5). GEOS-5 is a general circulation model (GCM) used within a data assimilation system where satellite and surface

observations are utilized [42]. The dataset has a spatial resolution of 0.5° (latitude) × 0.67° (longitude) and hourly output is available. For this study, a record period from January 1, 1979 midnight (UTC) to January 1, 2014 midnight (UTC) is chosen. This temporal range enables studying of wind variation over diﬀerent time scales from hours to decades. The spatial coverage of the dataset provides an opportunity to understand the regional wind patterns, those that coarse observations cannot address. Wind speed and WPD are two primary variables in wind resource assessments. Wind speed at the turbine hub height is calculated using the similarity theory for the surface layer in which the turbine submerges [6]. Wind speed is computed as

u z − d u(z) = ∗ ln − ψ, (2.1) κ z0

0 0 2 0 0 2 1/4 where z is the turbine hub height [6]. u∗ = [(u w ) + (v w ) ] is the friction velocity, defined with the surface kinematic momentum fluxes in x and y directions. κ = 0.41 is a standard accepted value of the von Krmn constant [43]. d is the displacement height that gives the vertical displacement of the entire flow regime over

areas densely covered with obstacles. Roughness length z0 defines the height that the wind speed is assumed to vanish near the ground. The parameter ψ depends on the stability of the boundary layer. A neutral boundary layer is assumed, thus ψ = 0. Neutral stratification assumes negligible effects of external heating on the vertical distribution of temperature. Although neutral stratification is widely assumed in previous studies of wind resource assessment to adjust wind speed to hub height from measurement height, the neutral boundary layer may not occur in the presence of 27 surface heating [44]. Intense surface heating causes warm air near the surface to rise and create turbulence as in the case of an unstable boundary layer. It implies a smaller change in wind speed with height than in a neutral case. WPD is the wind power available per unit swept area of a turbine defined as

1 WPD = ρu3, (2.2) 2

where ρ is the air density and u is the wind speed [6]. The fields for calculating wind speed and WPD at hub height have been extracted from the MERRA 2D surface turbulent flux diagnostics available at a single level at the top of the surface layer. Specifically, momentum roughness length (Z0M), friction velocity (USTAR), surface air density (RHOA), and displacement height (DISPH) are extracted from the dataset for the computation. Wind speed and WPD have been computed at all hourly time- steps at 50 m, 80 m, and 140 m corresponding to the typical hub heights of the three generations of wind turbines. Most prior studies in the region did not consider terrain and local conditions when adjusting wind speed to hub height (e.g.,[45]), where the effects of wind shear were demonstrated to be significant in power generation [46]. The WPD provides a proxy to wind power resource that is independent of a wind energy system’s specification. It combines the contributions of both wind speed and air density in illustrating the physical limit of wind power potential.

2.2.2 Metrics

Prior studies of the wind energy resource in the Arabian Peninsula were mostly con- cerned with the abundance of the wind resource in terms of average wind speed or annual energy production estimates. Variability and intermittency have not been investigated comprehensively for this region. The understanding of these temporal characteristics is essential for grid integration to pave the way for wind energy to become part of a larger generation network contributing to the generation of base- 28 load electricity. Therefore, we consider three essential metrics in evaluating the wind energy resource: abundance, variability, and intermittency. Abundance is the amount of wind energy available. It is conventionally measured by time-averaged wind speed (¯u) or time-averaged WPD, which takes the form of

1 Z T x¯ = x dt, (2.3) T 0

where x is the instantaneous wind speed or wind power density, and T is the length of the record period. Time-averaged WPD and wind speed are used to compare with results from prior studies in the region. For each grid cell in the domain, the hourly wind power density ﬁeld is used to compute the average WPD over the record period.

n 1 X p¯ = p , (2.4) n t t=1

where pt is the average WPD at each hour (t) and n is the total number of time-

steps in the record period. In addition, we compute the median WPD, pmed, which is the value lying at the midpoint of a frequency distribution of the time series at each grid point such that there is an equal probability to be above or below it. The median is a metric robust to extreme samples and hence is a better characterization of central tendency than the mean for skewed distributions such as those of wind speed

and WPD. The pmed is used to characterize the wind regimes in the further analysis:

−2 −2 −2 regime I where pmed ≥ 67 W m , regime II where 46 W m ≤ pmed < 67 W m , and

−2 regime III where pmed < 46 W m . The three regimes correspond to values separated by the 33rd and 67th percentiles of the median WPDs for all grid cells in the domain. This classification gives an indication of the relative spatial abundance of the wind resource in the region. Variability characterizes the fluctuations of the wind energy resource at a given location. It is measured by the robust coefficient of variation (rCV) of WPD [8], 29 computed as follows for each cell:

median(|p − p |) rCV = med . (2.5) pmed

The rCV provides a dimensionless measure of variability across the spatial domain. High rCV indicates highly variable wind resource, thus less desirable for operation. Intermittency is deﬁned by availability and persistence of the wind resource. Avail- ability is the fraction of time over a time series when the WPD exceeds a threshold of 200 W m−2. The threshold has been chosen keeping in mind the 300 W m−2 threshold (Wind Class 3) in ﬁltering sites for commercial scale power production in the Regional Energy Deployment System (ReEDS) model of the National Renewable Energy Lab- oratory (NREL) [47]. A lower threshold of 200 W m−2 (Wind Class 2) is chosen in light of recent developments in low-wind turbines which render conventionally low- wind areas viable for energy applications. Availability (A) is computed following [8],

n 1 X A = τ(p ), (2.6) n t t=1

 −2  1, pt > 200 W m τ(pt) =  0, otherwise

τ(pt) records the events when pt is greater than the threshold. Persistence reﬂects the steadiness over time in wind power generation above a given threshold. Episode length is deﬁned as the duration of consecutive time steps when pt is above the threshold. Persistence is measured by the median episode length (MEL). The median length of wind episodes is the median length of all recorded continuous periods when WPD is above the threshold [8], 30

MEL = median[τ(pt)]. (2.7)

It should be noted that intermittency concerns the statistics of threshold-crossing of the WPD while variability measures the fluctuation of the WPD in magnitude. This distinction was elaborated in Gunturu and Schlosser [8]. The present study aims at identifying regional features of wind resource while finer local characteristics and circulations could not be represented. Our analysis identifies regions for further down-scaling and micro-siting assessment where models with finer resolutions in space and time are appropriate. Dynamical downscaling in the mountainous regions of the Arabian Peninsula would benefit from the use of Kalman filters, as it has shown improvements in NWP wind speed forecast in complex terrain [48]. Sub-hourly data would be needed to characterize wind intermittency at scales relevant to grid integration of individual wind farms, as illustrated in the Wind Integration National Dataset (WIND) Toolkit by NREL [49]. Prevalence of intense surface heating in the region requires proper parameterization of the stability of the boundary layer to better estimate the variation of wind speed with height during daytime. It is also important to validate model results with current in situ measurements at different hub heights, such as the measurements taken by K.A.CARE in Saudi Arabia [50].

2.3 Results

2.3.1 Abundance

Near-surface wind resource over the Arabian Peninsula varies due to its diverse topographical features. We discuss the wind resource over this vast area by categorizing the spatial domain into three regimes based on the median WPD (ﬁg. 2.1). The three wind regimes are illustrated in ﬁg. 2.2. Selected locations of some prior studies can 31 199

Ar 100 30N Ma Aq R K 90

W DM 80 25N Do A Y 70

Dl S J 60 Latitude 20N 50

Jz Sa 40

15N 30

Ad 11.5 35E 40E 45E 50E 55E Longitude

Figure 2.1: Median WPD (W m−2) computed at 50 m AGL using the wind ﬁelds reconstructed from the MERRA data. The color scale is uniform except for values beyond 100 W m−2 and below 30 W m−2. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown. be found in table 2.1.

Regime I (relatively abundant)

The Arabian Peninsula shows a varied abundance of the near-surface wind resource spatially. Regime I shows the most abundant wind resource in the region with median WPD above 67 W m−2. Oﬀshore regions have the most abundant wind resource. Among the onshore locations, the Hejaz Mountains east of Jeddah in Saudi Arabia, the southern coast of Oman, and eastern Yemen show a relatively high abundance of wind resource. The mountainous coastline along the Gulf of Aqaba and central Jordan 32

Ar 30N Ma Aq R K

W DM 25N Do A Y III Dl S II J I Latitude 20N

Jz Sa

15N

35E 40E 45E 50E 55E Longitude

Figure 2.2: Categorized median WPD (fig. 2.1) at 50 m AGL into three regimes: regime I where median WPD is above 67 W m−2, regime II where median WPD is from 46 W m−2to67 W m−2, and regime III where median WPD is below 46 W m−2. Elevation contours at 400 m intervals are drawn using elevation data from MERRA. Selected locations of prior studies are shown. also have moderately abundant wind resource. These regions show a comparable spatial distribution of average and median WPD and the average WPD is about twice the median, illustrating a positively skewed distribution (fig. 2.12). One exception is along the coast of the Gulf of Aqaba, where the average and median WPD are close, leading to a less skewed distribution. The abundance of the wind resource is illustrated in average WPD (fig. 2.3). A map of average wind speed is also included for reference and comparison with prior studies where WPD is not available (fig. 2.13). The East of the Hejaz and Asir mountains in Saudi Arabia has average WPD of 143 W m−2. This is confirmed 33 532

Ar 200 30N Ma Aq R K 180

W DM 160 25N Do A Y 140

Dl S J 120 Latitude 20N 100

Jz Sa 80

15N 60

Ad 40.3 35E 40E 45E 50E 55E Longitude

Figure 2.3: Average WPD (W m−2) computed at 50 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 200 W m−2 and below 60 W m−2. Large extreme values beyond the 98- percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown. by recent meteorological measurements from 1998 to 2002 at Dhulum with average WPD of 186 W m−2 at 40 m AGL [51]. Before the use of reanalysis, wind resource assessments relied on data collected at meteorological stations and compared among known locations. The consensus from earlier studies has been that coastal areas possess more abundant wind resources (e.g., [52]). Our discovery of higher wind resources away from the coast points to the need for more in situ observations at locations indicated in this study. Bahrain and Qatar show average WPD (wind speed) of around 119 W m−2 (4.63 m s−1) and 139 W m−2 (4.71 m s−1) respectively. A prior study [36] shows an average wind speed of 8.65 m s−1 at 60 m AGL extrapolated from hourly observations at 10 m AGL 34 at Bahrain International Airport from 2003 to 2005. These observations point to the need for further investigations to better understand the disparity between the median and average WPD shown in Bahrain and Qatar where the ocean strongly influences circulation over the island and the peninsula. Along the Oman coast, our results show an average WPD of about 197 W m−2 for southern coast and 85.8 W m−2 near Sur. Meteorological observations reported at Thumrait, Sur, Masirah, and Marmul, show average WPD of 230, 194, 165, and 109 W m−2 at 10 m AGL [53]. Further investigations into the long-term time-series would help reveal climatological differences in the two sub-regions. In eastern Yemen around the Hadramaut Mountains, we find an average WPD (wind speed) of about 110 W m−2. The region surrounding Aden shows an average WPD of about 105 W m−2 (4.75 m s−1), where the wind speed decreases further east- ward. A prior study [54] using five years of meteorological observations from the airport at Aden shows an average wind speed of 4.5 m s−1 at 10 m AGL. The more abundant wind resource appears in the west of Aden, where future in situ observations would benefit from further studies into the local resource characteristics. Along the coast of the Gulf of Aqaba, the average WPD is around 105 W m−2. In central Jordan, average WPD is found to be 176 W m−2. This finding seems to deviate from prior observations where the wind resource is more abundant in the southwestern area. In particular, Fujaij shows an average WPD of around 91.6 W m−2 (4.3 m s−1), where the annual average wind speed of 6.88 m s−1 was reported [55].

Regime II (moderately abundant)

Regime II shows areas with a moderately rich wind resource of median WPD between 46 and 67 W m−2. It includes the northern coast of the Red Sea, the coast of the Arabian Gulf, the areas west of Riyadh and along the borders of Saudi Arabia with Yemen and Oman, and northwestern Kuwait. 35 The northern Arabian coast of the Red Sea has been popular for wind resource studies due to long-standing meteorological measurements at major settlements. Our results show that the surroundings of Yanbu have an average WPD around 106 W m−2, close to the observed 134 W m−2 [51]. Our results indicate that a more abundant wind resource is present within 100 km of the meteorological stations. Northwest of Yanbu shows a higher wind average WPD of around 138 W m−2. Comparable wind abundance can also be found in the mountains further north of Yanbu and along the coast of the Red Sea northwest of Yanbu. This analysis identifies coastal sites beyond those discussed and observed in prior studies (i.e., Rehman and Ahmad [33]). The coast of the Arabian Gulf illustrates a highly positively skewed distribution of WPD where the average value is about three times the median, indicating a potentially high variability in wind power previously not reported in the literature. The average WPD near Dhahran is shown to be around 136 W m−2, close to the observed 154 W m−2 [51]. Kuwait also shows a highly positively skewed distribution in WPD as illustrated by the difference between the median (fig. 2.1) and the average WPD (fig. 2.3). North- western Kuwait has an average WPD of around 164 W m−2, where southern Kuwait shows an average WPD of around 147 W m−2. A prior study shows an average WPD at 30 m AGL at Umm Omara (northwest) and Al-Wafra (south) to be 271 W m−2 and 273 W m−2 respectively. These discrepancies could arise from the choice of a constant exponent of 1/7 when adjusting the wind speed from 10 m AGL observations to 30 m [56]. The 1/7 constant exponent has been a popular choice in the wind resource assessment literature in extrapolating wind speed to hub height and deemed inappropriate for domains with complex surface characteristics [57]. This choice of the exponent in prior studies did not account for surface characteristics, leading to an overestimate of the wind resource. 36 Regime III (least abundant)

Regime III shows areas with the comparatively least abundant wind resource in the region where median WPD is below 46 W m−2. The area includes the southern Ara- bian coast of the Red Sea, the southern coastline of Yemen, the east coast of Oman, and Abu Dhabi in the UAE. The southern Arabian coast of the Red Sea has an average WPD (wind speed) of around 70.4 W m−2 (3.57 m s−1), which is conﬁrmed by observations at Jizan, where monthly average wind speed is between 3.8 m s−1 and 5.2 m s−1 [33]. The southern Yemen coast also shows comparable wind power density, with a similar mountainous terrain along the coastline. Northeastern Oman and the UAE show low average WPD of 80.7 W m−2, consistent with observations [53, 37].

Eﬀects of variation of hub height

The rate of change of average and median WPD with altitude depends on the roughness length (fig. 2.14). Roughness length defines the influence of surface characteristics on wind power resource. Figure 2.4 shows a general increase of WPD with altitude, much affected by the varying terrain. Offshore locations show lower increase than onshore sites. These observations are consistent across average and median WPD. With increased hub height, both the magnitude and the frequency of higher wind power resource increase and shift the distribution to higher WPD values, making it less skewed. The rate of increase is proportionally consistent across the average and median WPD at the two altitudes.

Comparison with Vestas’ wind map over Saudi Arabia at 100 m AGL

Figure 2.5 shows the annual average wind speed at 100 m AGL by our reconstruction and ﬁg. 2.6 shows its comparison to Vestas’ estimation. Vestas’ wind speed map was calculated using the output from the Weather Research and Forecasting (WRF) 37 48% average 80 m average 140 m

30N 45%

25N 40%

20N 35% 15N 30% median 80 m median 140 m

Latitude 25%

20%

15%

10% 35E 40E 45E 50E 55E Longitude

Figure 2.4: Changes of median WPD (ﬁg. 2.1) and average WPD (ﬁg. 2.3) at 80 m and 140 m from those at 50 m AGL. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. The color scale is uniform except for values beyond 45%.

model with a spatial resolution of 3 km × 3 km and hourly temporal resolutions from 2000 to 2013 [58]. The wind speed estimate is available at the Renewable Resource Atlas at King Abdullah City for Atomic and Renewable Energy (K.A.CARE) in Saudi Arabia [50]. The model was driven by the boundary conditions from the National Center for Environmental Prediction (NCEP) Global Forecast System Analysis of

1° × 1° spatial resolution and 6-hourly temporal resolution. Topography used in the Vestas model was obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) of 3000 × 3000 spatial resolution. Figure 2.5 indicates relatively high average wind speed at the tip of the Gulf of Aqaba, north of Duba, west of Yanbu, south of Jeddah, and to the east of the Hejaz and Asir Mountains. Figure 2.6 shows that the wind speed patterns of the two maps are qualitatively similar, with root-mean- 38 7.17

Ar 6.6 30N Ma R Aq K 6

5.4 W D M

25N Do 4.8 A Y 4.2 Latitude Dl S

J 3.6

20N 3

2.4

Jz Sa 0 35E 40E 45E 50E 55E Longitude

Figure 2.5: Average wind speed (m s−1) computed at 100 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 6.6 m s−1 and below 2.4 m s−1. Selected locations of prior studies are shown. square error (RMSE) of 1.15 m s−1. However, at these identified high wind locations the Vestas’ map holds higher values than from our reconstructions using MERRA, with the Vestas’ average wind speed being 0.424 m s−1 higher on average. The area in the north near Arar and Rafha, and in the east near the coast of the Arabian Gulf show similar spatial patterns in wind speed with the Vestas’ map, but higher values in general. This qualitative assessment reflects the higher spatial variability that is shown in the model with higher spatial resolution. A converse pattern is shown on the Vestas’ map where there is a higher wind speed east of Najran than in the Empty Quarter. There is a higher wind speed shown to the west of the Hejaz Mountains where fine topographical features exist in the Vestas’ map. It remains to investigate further these discrepancies and their relation to the models and assimilation schemes. 39 7.08

Ar 2.5 30N Ma R Aq K 2

1.5

1 W D M 0.5 25N Do A Y 0

Latitude -0.5 Dl S

J -1

-1.5 20N -2

-2.5 Jz Sa -2.9 35E 40E 45E 50E 55E Longitude

Figure 2.6: Diﬀerences in the average wind speed (m s−1) at 100 m AGL between our calculation and that from estimation of Vestas (Vestas - MERRA) are illustrated with a color scale that is uniform except for values beyond 2.5 m s−1 and below −2.5 m s−1. Selected locations of prior studies are shown. The MERRA dataset is regridded to 1 km resolution by nearest neighbor for illustration.

Comparison with 3TIER’s wind speed map at 80 m AGL

The average wind speed at 80 m AGL from our reconstruction using the MERRA dataset is compared with 3TIER’s average wind speed map. The 3TIER wind map is available with 5 km spatial resolution hosted at the Global Atlas for Renewable Energy of the International Renewable Energy Agency (IRENA) [59]. The 3TIER’s map uses over ten years of hourly data generated from statistical and dynamical downscaling where surface-atmosphere interactions are accounted for. The 3TIER wind speed estimates were validated with 229 NCEP-ADP stations in the Middle East and Africa, where only around 60 stations are in the Middle East [60]. The RMSE is 1.03 m s−1 between the annual average wind speed of station data and those from 40 the model output in Africa and the Middle East. However, the data for the 3TIER map are not retrievable. There is also a lack of publicly available documentation on the creation of the dataset. Hence, the following discussion is not illustrated. Our wind speed reconstruction and 3TIER’s estimates are qualitatively similar. On the coast of the Red Sea, the two maps agree on the lower wind speed between the southern Red Sea and the mountains. Prominent wind patterns agree over the north, the east, and the south of Saudi Arabia, the UAE, Qatar, Kuwait, the eastern and the southern coasts of Oman, and Yemen. Known locations at Yanbu, Jeddah, and Al Qahma appear to show higher wind speed in the 3TIER map than in the MERRA reconstruction.

2.3.2 Variability

The robust coefficient of variation (fig. 2.7) measures the median deviation as a fraction of median WPD. A higher rCV indicates relatively higher variability at a location in WPD. Low variability reduces the output volatility and is favorable for power generation fig. 2.7 indicates that the sea-ward side of elevated areas shows greater variability than other sites in the spatial domain. Within the Regime I area, the inland face of the Hejaz and Asir Mountains shows relatively low variability. Similarly, the Hadramaut Mountains in Yemen and the Wahiba Sands desert in eastern Oman show relatively low variability in WPD. Along the Gulf of Aqaba, the mountainous region also shows relatively low variability. Within Regime II, both the coastlines of the Red Sea and the Arabian Gulf show higher variability than inland locations. However, the surroundings of Yanbu show only moderate variability. Changes in hub height lead to insignificant changes in variability (fig. 2.15).

2.3.3 Intermittency

Intermittency is characterized by availability and persistence. 41 1.47

Ar 1.45 30N Ma Aq R K 1.4

W DM 25N Do 1.35 A Y

Dl S 1.3 J Latitude 20N 1.25

1.2 Jz Sa

15N 1.15

Ad 0.994 35E 40E 45E 50E 55E Longitude

Figure 2.7: Robust coeﬃcient of variation computed at 50 m AGL using the wind ﬁelds reconstructed from the MERRA data. The color scale is uniform except for values beyond 1.45 and below 1.15. Selected locations of prior studies are shown.

Availability

Availability (ﬁg. 2.8) measures the fraction of time when the wind power resource is above a given threshold (i.e., 200 W m−2). It represents the amount of time with meaningful power generation by a turbine. Most onshore locations have availability between 10% to 30%, except for the low wind abundance areas in Yemen, northeast Oman, and along the southern coast of the Red Sea. Areas with abundant wind resource appear to have high availability. The western mountains in Saudi Arabia, the southern coast of Oman, the tip of the Gulf of Aden, and Kuwait show availability of around 25%. Figure 2.9 shows that an increase in hub height contributes to increased availability onshore due to reduced eﬀect of boundary layer friction on the wind by the surface. 42 0.5

Ar 0.45 30N Ma Aq R K 0.4

W DM 0.35 25N Do A Y 0.3

Dl S J 0.25 Latitude 20N 0.2

Jz Sa 0.15

15N 0.1

Ad 0.024 35E 40E 45E 50E 55E Longitude

Figure 2.8: Availability of wind resource computed at 50 m AGL using the wind ﬁelds reconstructed from the MERRA data. The color scale is uniform except for values beyond 0.45 and below 0.1. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. Selected locations of prior studies are shown.

Increases in availability with hub height are most signiﬁcant in regions with low wind abundance, except southern coastal Yemen.

Persistence

Median episode length (fig. 2.10) measures the persistence of the wind resource, which indicates persistent up-time of energy production at a specific location. In Regime I, MEL is eight hours in the western mountains of Saudi Arabia, central Jordan, and the southern coast of Oman. It indicates that the wind resource at these locations is persistent for at least eight consecutive hours during half of the record period. At the tip of Yemen at the Gulf of Aden, the MEL is 14 hours. Moderate MEL above five hours by the coast of the Gulf of Aqaba is noticed. Areas with moderate 43 83% 80 m 140 m 60%

30N 50% 40% 25N 35% 30%

Latitude 20N 25% 20% 15N 15% 4% 35E 40E 45E 50E 55E Longitude

Figure 2.9: Changes of availability (fig. 2.8) at 80 m and 140 m from those at 50 m AGL. Large extreme values beyond the 98-percentile in the spatial domain are masked in grey. The color scale is uniform except for values beyond 40% and below 15%. wind speed near the Arabian Gulf such as Kuwait and eastern Saudi Arabia show a higher persistence of the wind with the MEL of greater than eight hours. Low wind abundance regions exhibit persistence with the MEL of below five hours. The MEL tends to increase with hub height with almost no exceptions (fig. 2.16).

2.4 Discussions

This study provides the first regional assessment of the abundance, variability, and intermittency of wind resource over the Arabian Peninsula. Employing the MERRA dataset, the wind field at different hub heights is reconstructed applying similarity theory using the roughness length, friction velocity, and displacement height. The reconstructed wind field spans over three decades with consistent spatial and temporal resolution. The wind power density field enables analysis of various aspects of the large-scale features of the wind energy resources in the Arabian Peninsula. The wind resource is also characterized using metrics of wind variability and persistence. This 44 15 50 m 80 m

Ar Ar Ma Ma 10 30N Aq R K Aq R K

W DM W DM Do Do 9 25N Y A Y A S S J J 20N 8

Jz Sa Jz Sa 15N 7 Ad Ad

140 m 6

Latitude Ar Ma Aq R K 5 W DM Do Y A S 4 J

Jz Sa 3

Ad 2 35E 40E 45E 50E 55E Longitude

Figure 2.10: Median length of wind episode (h) computed at 50 m, 80 m, and 140 m AGL using the wind fields reconstructed from the MERRA data. The color scale is uniform except for values beyond 10 h. Selected locations of prior studies are shown. work improves upon the earlier comprehensive studies in providing an areal overview of the wind resource with higher spatial resolution and metrics of temporal variations in the wind. Previously unrecognized locations of interest with high wind abundance and low variability and intermittency have been identified in this study and confirmed by recent on-site observations [51]. In particular, the western mountains of Saudi Arabia experience more abundant wind resource than most Red Sea coastal areas. The wind resource is more variable in coastal areas along the Arabian Gulf than their Red Sea counterparts at a similar latitude. More persistent wind is also found along the coast of the Arabian Gulf. This analysis points to the areas previously not recognized. Studies at finer resolutions for these identified areas are necessary to resolve spatial features and local 45 circulations relevant to wind power generation. Our reconstructed wind field will en- able investigations on the impact of large-scale circulations on regional wind resources. Effects of the El Nio Southern Oscillation (ENSO) [61] and the North Atlantic Os- cillation (NAO) [62] at various hub heights can be assessed using the reconstructed wind fields at different hub heights. The economic viability of wind energy applications at a regional scale can be conducted with higher spatial and temporal resolution using this dataset to provide greater consistency than in prior studies [63, 64, 65, 23]. Specifically, wind power production in this region can be modeled in a way similar to those performed in Northern Ireland [66], Great Britain [67], and Sweden [68]. In light of an estimated 25 years turbine lifetime [69], sustainable deployment of wind energy systems in the region requires assessments of the effects of climate change on the regional wind resource using developed approach [70]. A measure-correlate-predict (MCP) approach can be used to estimate the long-term wind resources at a target site using our reconstructed wind fields in conjunction with short-term wind measurement campaigns [71]. An investigation into the best pattern for wind power aggregation through region-wide interconnection to mitigate intermittency will be a timely and valuable next step towards an optimal integration of large-scale wind energy systems in the Arabian Peninsula [72]. 46

30 Countries Bahrain Jordan

25 Kuwait Oman Qatar Latitude 20 Saudi Arabia United Arab Emirates Yemen

35 40 45 50 55 60 Longitude

Figure 2.11: Country map in the Middle East

Mark Location Ma Maan Y Yanbu Ar Arar R Rafha D Dhahran Aq Aqaba Sa Salalah W Al Wajh Jz Jizan S Sur Do Doha M Manama A Abu Dhabi Ad Aden J Jeddah K Kuwait Dl Dhulum

Table 2.1: Selected locations of prior studies in the region as indicated on the maps. 47

90000

60000 count

30000

0 250 500 750 1000 WPD (W m−2)

Figure 2.12: Illustration of a skewed distribution of WPD at Yanbu 48

8.5

Ar 8 30N Ma Aq R K 7.5

W DM 6.5 25N Do A Y 6 Dl S J 5.5 Latitude 20N 5

4.5 Jz Sa 4 15N 3.5

Ad 3 35E 40E 45E 50E 55E Longitude

Figure 2.13: Average wind speed (m s−1) is computed at 50 m AGL using the wind ﬁelds reconstructed from the MERRA data. Selected locations of prior studies (table 2.1) are shown. 49

1.25

1.00

0.75 Hub height 140m AGL 80m AGL

0.50 Fractional change in WPD

0.25

0.001 0.100 Median roughness length (m in log10-scale)

Figure 2.14: Fractional change in WPD at 80 m and 140 m from 50 m over temporal median of roughness lengths 50

600

400 Hub height 140m AGL

count 80m AGL

200

-0.004 0.000 0.004 0.008 Fractional changes in rCV

Figure 2.15: Distribution of fractional changes in rCV at 80 m and 140 m from 50 m out of 1638 grid cells 51

900

600 Hub height 140m AGL

count 80m AGL

300

-2 -1 0 1 2 3 Changes in Median Episode Length (hour)

Figure 2.16: Distribution of diﬀerences in MEL at 80 m and 140 m from 50 m out of 1638 grid cells 52

Chapter 3

Airborne wind resource analysis

In the Middle East, near-surface wind resources are intermittent. However, high- altitude wind resources are abundant, persistent, and readily available and may provide alternative energy resources in this fossil-fuel-dependent region. Using wind field data from the Modern-Era Retrospective Analysis for Research and Applications Ver- sion 2 (MERRA-2), this study identifies areas favorable to the deployment of airborne wind energy (AWE) systems in the Middle East and computes the optimal heights at which such systems would best operate. AWE potential is estimated using realistic AWE system specifications and assumptions about deployment scenarios and is compared with the near-surface wind generation potential with respect to diurnal and seasonal variability. The results show the potential utility of AWE in areas in the Middle East where the energy demand is high. In particular, Oman and Saudi Arabia have a high level of the potential power generation with low annual variability.

3.1 Overview

In 2015, with a record 63 GW added to the total global renewable power capacity of about 433 GW in 2014, renewable energy became the largest contributor to new power-generating capacity in the United States and Europe and the second largest contributor in China [73]. Near-surface wind power is a mature technology and a fast-growing renewable power source that has the potential to contribute substan- tially to the reduction of greenhouse gas emissions [74]. In the Middle East, wind 53 power as a renewable energy resource has attracted interest primarily because of potential economic savings and energy resource diversiﬁcation [2]. Economic savings from wind power in the region would be primarily realized through the opportunity costs from saving fossil fuels from use in domestic power generation. With round-the- clock availability, wind power also provides a way to diversify the energy mix in the oil-exporting countries of the Middle East. Moreover, the renewable energy sector creates jobs that require technical expertise that is currently being cultivated in the younger generations of the local workforce, leading to a sustainable and knowledge- based economy [2]. These factors, coupled with the recent drop in oil prices, have incentivized wind power harvesting in the Middle East. A recent survey of near-surface wind resources showed that there are some areas with high and persistent winds in the Middle East [13]. However, near-surface wind resources in the Middle East are not as abundant or persistent as are wind resources in other regions of the world [8, 39]. Generation of airborne wind energy (AWE) is possible in areas with little near-surface wind [75]. Indeed, higher and steadier power generation from AWE can be attained due to the greater availability of persistent wind resources at higher altitudes [75]. AWE is generated using airborne devices that are connected to a ground station by a tether. Instead of land-based wind turbines that operate near the surface, AWE uses devices that extract kinetic energy from the wind available above the surface layer of the atmosphere and turn it into electricity [11]. There are currently two major types of AWE generators: drag type devices with generators on board with a tether that transmits electrical power, and lift type devices that transmit mechanical power in reeling the tether connected to a ground based generator. Active projects include KiteGen (lift) and Makani (drag), both are under active development and pilot studies. KiteGen is in the development of a 3 MW ground-based device and Makani is working on a 600 kW on-board generator prototype. Details of the technologies and their working principles have been described in 54 prior studies [76]. There have been interests in AWE in various geographical regions, most recently in Northern Ireland [77]. The potential of AWE in the Middle East has not previously been explored. This study contributes to the evaluation of AWE in the Middle East by identifying areas favorable for AWE system (AWES) deployment; analyzing the diurnal and seasonal variability of airborne wind resources; determining a range of optimal altitudes for AWES deployment; and estimating AWE generation potential compared with near-surface wind generation potential. The results reported in this chapter represent a survey of high-altitude wind resources at 3-hourly temporal resolution. Further downscaling studies and measurement campaigns, which are beyond the scope of this chapter should be conducted on site for siting purposes.

3.2 Methods

Wind speed vertical proﬁles were constructed using the three-hourly instantaneous output from MERRA-2, a project from the US National Aeronautics and Space Ad- ministration (NASA) on a 0.5◦ (latitude) × 0.625◦ (longitude) grid [78]. The horizontal wind vectors (U, V ) and altitudes (H) were retrieved at 13 pressure levels from sea-level to 700 hPa. The data spanned 35 years from 1 January 1980 to 31 √ December 2015. Wind speed (u) was computed using u = U 2 + V 2. The long record-length, even spatial and temporal coverage, state-of-the-art spatial and temporal resolutions as compared to other reanalyses products and prior validations in various studies make MERRA-2 a valuable dataset for wind resource assessment in the area of interest where reliable long-term measurements are rare. MERRA-2 assim- ilates conventional observations such as radiosondes and surface station observations from various sources, with over 1200 surface stations and 150 radiosonde locations in the region of interest [42, 79]. WSM was deﬁned as the maximum wind speed in the lowest 13 pressure levels 55

Figure 3.1: (a) Definition of wind speed maxima (WSM) and its associated altitude (H), where the size of the box indicates the magnitude of the wind speed (b) Power curve, power (kW) as a function of wind speed (m/s), of a lift-type groud-based AWES of 3 MW rating. The figure and all the text elements in (b) has been plotted using the software R version 3.3.2. from the sea surface to 700 hPa. The associated altitude at which WSM occurs is H (Fig. 3.1, a). Wind resource at high-altitude was characterized from WSM and H. The altitude of these pressure levels from the ground varied with the earth’s topography. Engineering constraints on the tethers due to increasing weight with length limited our focus to the first 3 km of the boundary layer [80]. The boundary layer in the Arabian Peninsula varies in thickness and could reach 5 km in height in the afternoon [81]. A similar definition of WSM was previously used with an imposed lower bound of 10 m/s [80, 82]. However, maximum wind speed in this study includes winds below 10 m/s because it has been technically demonstrated that current AWES could operate at nameplate capacity with wind speeds under 10 m/s [83]. For this study, the cut-in speed of the device was 2 m/s and the wind speed at which the nameplate capacity was reached was 6 m/s [83]. 56 3.2.1 Wind Speed Maxima (WSM)

For each grid cell in the domain, the average WSM (WSM) was computed using the maximum hourly wind speed ﬁeld across pressure levels over the data collection period, as follows:

n 1 X WSM = WSM , (3.1) n t t=1

where WSMt = max(ui), i ∈ [1, 13] at each hour (t) and n is the record length. Variability in WSM was characterized with the coeﬃcient of variation (CV) for each cell as follows:

σWSM CVWSM = , (3.2) WSM

where σWSM is the standard deviation over time for each cell. The CV provides a dimensionless measure of variability across the spatial domain. A high CV indicates highly variable wind resource, which would be less desirable for AWES operation.

3.2.2 Altitudes of WSM

The average altitude of WSM (H) for each cell was calculated as follows:

n 1 X H = H , (3.3) n t t=1

where Ht = Hj such that uj = max(ui), i ∈ [1, 13]. Variability in H was characterized with CV for each cell as follows:

σH CVH = , (3.4) H

where σH is the standard deviation over time for each cell. A high CV in H indicates the need for frequent ﬂight adjustments of the AWES, which would be less 57 desirable for AWES operation.

3.2.3 Capacity factor

The wind power potential of AWES operating at optimal altitudes is summarized using the capacity factor of a lift-type AWES based on its power curve (Fig. 3.1, b). The power curve is that of a prototype AWES at a test facility [83]. A lift-type AWES generates power from traction through cycles of reeling and unreeling of a tether connected between a kite and a base station. An automatically controlled airfoil in the form of a kite with flexible wings glides in the air-space above the base station. The CF was estimated based on the construction of a realistic power curve for the device. The power curve gives the power generation of a single AWES as a function of the ambient wind speed. The power curve of the simulated device has three power-generating regimes that correspond to three flight conditions of the AWES. When the wind speed is above the cut-in speed (2 m/s) but not high enough to exert the maximum force, a small amount of power is generated as the tether reels out. When the wind speed is sufficiently high to exert the maximum force (6 m/s), the tether reels out to maintain that maximum force exerted. The power generated in this second regime is proportional to the reel-out speed of the tether. When the wind speed is over the maximum possible for the device (16 m/s), the device will navigate such that the tether is at an angle from the prevailing wind direction, maintaining the maximum power generation. In this third regime, the power generation is constant at its maximum. Since the duration of one cycle of reeling for a lift-type AWES is in minutes, the device is assumed to adjust its altitude instantaneously to reach the wind speed maximum of the hour. The nameplate capacity of this prototype AWES is 3 MW. 58 3.2.4 Deployment assumptions

The annual potential for energy production at WSM by country in the region was computed. The annual AWE generation for each country was estimated based on assumptions about system availability, system density, and spatial exclusion:

• System availability: availability of an AWES when it is not under maintenance. The study assumes a 98 % system availability with current turbine technologies, with maintenance time distributed evenly under average generation conditions.

• System density: calculated based on how many AWE systems can be deployed in an area. Assuming that a fully extended tether to reach any altitude within the chosen portion of the boundary layer is 3 km, one unit of AWES is allowed in each 10 km x 10 km grid cell.

• Spatial exclusion: areas that are excluded for AWES deployment. Terrain slopes were computed with elevation data at 15 arcsecond spatial resolution with eight neighboring points. Regions with a slope greater than 0.2 were excluded. Inac- cessible terrain (wetlands, shrublands, and forests) and urban land-use were also excluded. It is further assumed that AWE systems are uniformly distributed on the available land after these exclusions.

3.3 Results

The analysis of AWE in the Middle East begins by studying the vertical characteristics of horizontal wind in the boundary layer relevant to AWE. Horizontal wind speed data at 13 pressure levels are gathered from the Modern-Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2) [78]. Wind speed maxima (WSM), the maximum wind speed at each time step across pressure levels at each grid cell, is used to measure the magnitude and altitude of winds at high-altitude. The average 59 magnitude and variability of WSM were computed to summarize the vertical proﬁle of WSM at every grid cell in the domain of interest by the time of day and season. Variability in the magnitudes of WSM and the altitudes at which they occur (H) is also characterized to gain an overview of the vertical range of operations and relative ease of deployment of AWE systems across the spatial domain where the length and the consequent weight of the tether create a deployment constraint. The following plots with discrete color scales are generated using the k-means algorithm, which clusters values around centroids that become the labels of each interval on the color scale.

3.3.1 Average WSM

The boundary layer over the Arabian Peninsula could reach up to 6 km above the surface in summer [84]. The wind speed in the lowest 3 km of the boundary layer does not necessarily increase monotonically with altitude due to the presence of low- level jets (LLJs) [75]. LLJs are streams of fast-moving air with super-geostrophic speeds located over 100 m above ground and formed by various mechanisms, such as inertial oscillations [85] and baroclinicity over the sloping terrain due to the horizontal temperature gradient [86]. Both mechanisms are discussed in the seminal book [87]. Figure 3.2 (a-d, with individual color scales due to diﬀerences in wind speed distribution) shows that there are areas with average WSM (WSM) over 12 m/s over eastern Saudi Arabia, Kuwait, Bahrain, and Qatar in winter. In summer, areas with persistent winds are located along the coast of Yemen and Oman, due to the presence of the Somali Jet associated with the Indian summer monsoon. A high AWE resource is located north of 22◦ N in winter. South of 22◦ N, high AWE resources appear to be in coastal areas. WSM is much higher over the Arabian Sea during the summer than during the winter. A slightly higher WSM is observed in summer in western Iran. This WSM 60

40 a b

22.1 30

17.3

12.5 20

11.4

10.4 40 c d 9.3

8.3 30

7.2

20 5.7

30 40 50 60 30 40 50 60 40 e f

0.55 30

0.49

0.43 20

0.4

0.36 40 g h 0.32

0.28 30

0.23

20 0.15

30 40 50 60 30 40 50 60

Figure 3.2: Average wind speed (m/s) of WSM during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. CV in the wind speed of WSM during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] Panels (a-d) share a uniform color scale from 5 to 13 m/s; values larger or smaller than the boundaries share the bounding colors. These boundaries are informed by the wind speed distributions (Fig. S1) and the wind speed maxima vertical profiles (profiles at selected locations are illustrated in Fig. S2). 61 is more prominent during the night than during the day. Higher WSM is observed during the winter in most areas in the region, particularly in central and northwestern Saudi Arabia, across the Gulf of Oman, and in western Turkey. WSM experiences little diurnal variation, except in July, when WSM is higher at night in western Saudi Arabia, Iran, and parts of Turkey.

3.3.2 Variability in WSM

Variability in WSM (CVWSM) is measured by the coeﬃcient of variation (CV), a dimensionless measure of spread that describes the variability relative to the mean.

Figure 3.2 (e-h) shows that CVWSM decreases by an average of 0.1 during the night in comparison with during the day in both January and July. Larger decreases above 0.1 are observed along the northern coast of the Red Sea and near the mountain ranges in Yemen. In July, a decrease of over 0.1 in variability during the night is also seen along the Mediterranean coast of North Africa and the Middle East and in Eastern Iran. The daytime WSM has higher variability in the southern coast of the Black Sea in January than in July.

CVWSM is lower by more than 0.1 in summer than in winter along the west coast of the Red Sea, the southern coast of the Arabian Peninsula, and the land area north of

◦ 20 N. Over Sudan, central and eastern Saudi Arabia, and the Arabian Gulf, CVWSM is lower by around 0.1 in winter than in summer.

In summer, CVWSM is higher during the night than during the day over most land areas except over Sudan and the Gulf of Aden. The spatial variation of CVWSM is

minimal except for areas surrounding the Gulf of Aden; for instance, the CVWSM in the daytime in winter is lower than that during the night. There is also slightly higher

variation in CVWSM along the Arabian coast of the southern Red Sea during the day. 62

40 a b

2832 30

2587

2371 20

2141

1915 40 c d 1683

1434 30

1158

20 820

30 40 50 60 30 40 50 60 40 e f

1.11 30

0.89

0.75 20

0.64

0.55 40 g h 0.46

0.37 30

0.27

20 0.15

30 40 50 60 30 40 50 60

Figure 3.3: Average altitude (meters above sea level, mASL) of WSM (H) during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. CV in the altitude of WSM (H) during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The ﬁgure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] 63 3.3.3 Altitude of WSM

The average altitude of WSM (H) (Fig. 3.3, a-d) follows similar diurnal patterns in January. In July, WSM is at a relatively low altitude over the Mediterranean coast of the Middle East at about 1 km. Also, a reduction in H (over 500 m) occurs over Egypt, northwestern Saudi Arabia, and northern Oman in July during the night when compared with the day. WSM occurs at a much lower altitude during the night in summer than during the night in winter, particularly along coastal areas where the reduction in H is more than 500 m. Egypt experiences a reduction of over 1 km in H at night in winter compared with at night in summer. Exceptions are found where the Red Sea joins the Gulf of Aden with an increase of 500 m in H.

3.3.4 Variability in WSM altitudes

Variability in WSM altitudes (CVH) is measured by CV. Low CVH (Fig. 3.3) is

associated with areas where H is high. High CVH is associated almost exclusively with the Mediterranean coast of the Middle East, the southern coast of Oman, and Egypt, where the average altitude is below 1 km.

Seasonal variation in CVH is consistent during the day and night. Areas above

◦ ◦ 20 N experience a higher CVH in summer than in winter. Areas below 20 N ex-

perience a higher CVH of more than 0.5 in winter than in summer. Central Saudi Arabia, northern Iran, western Turkey, Yemen, Sudan, and Ethiopia experience little

variation in CVH in summer and winter. In summer, Egypt, northern and eastern Saudi Arabia, Oman, the Mediterranean

coast of the Middle East, and Iraq experience slightly higher CVH of more than 0.2 during the night than during the day. In winter, the same is observed for eastern Sudan and Oman, especially along the coast. 64 3.3.5 Regional AWE potential

We explored regional AWE potential in three ways:

• Examining the average capacity factor (CF) of AWE generation in the region and identifying areas of interest, where CF is the ratio of power generation to the nameplate (maximum) capacity of the AWE system

• Computing annual energy production at the country level given certain assumptions about AWES deployment

• Comparing AWE potential with near-surface wind energy potential.

Average power generation at WSM a b

40 7500

30 5000 Egypt KSA Oman Turkey Jordan Kuwait Qatar UAE

20 production

2500

30 40 50 60

0.2 0.3 0.4 0.5 0.6 1980 1990 2000 2010 Year

Figure 3.4: Map of AWE potential. The CF of the 3 MW AWE system (left) and annual AWE generation per capita (kWh) by country (right). The ﬁgure including the map and all the text elements has been plotted using the software R version 3.3.2. [1]

A map of AWE feasibility (Fig. 3.4a) is constructed using the CF computed from the power curve of a reference 3 MW AWES [83]. Among the areas with CF greater than 0.5 are Kuwait, Syria, Jordan, and Egypt. The Aegean coast of Turkey 65 is another preferable location for deployment. The eastern border of Iran also has exceptional potential with a favorable terrain, where there are contiguous open areas. The southern coasts of Yemen and Oman have similar terrain and a high CF. AWE potential over Saudi Arabia is the most abundant in the northeast on the border with Kuwait and along a thin coastal strip along the northern Red Sea.

Annual per capita energy production at WSM

Figure 3.4b shows the projected average annual AWE generation per capita for each country. The average annual generation per capita in Oman and Saudi Arabia stands out for more than triple the per capita generation of other countries in the region mostly due to the high wind speed (Oman) and the vast area (Saudi Arabia). The coefficient of variation in annual generation per capita is highest (0.06) among Qatar, Bahrain, and Palestine, and lowest (0.02) among Yemen, Oman, and Saudi Arabia. The high average annual generation and low annual variability make Oman and Saudi Arabia promising candidates for large-scale AWE deployment. Table 3.1 presents the AWES potential for each country and the total potential power generation and energy consumption per capita for 2010. The number of deployable generators ranges from one to over fifteen thousand, reflecting the vast differences in the available land across the selected countries in the region. The spatial density of AWE systems varies from 0.18 to 0.96 per 100 km2, reflecting the different levels of urbanization and terrain features in the region. Oman, Saudi Arabia, Iraq, Egypt, and Yemen have power generation potential from AWE that could meet over 75 % of the energy consumption required for 2010 at the assumed deployment level.

3.3.6 Comparison with near-surface wind resources

In Figure 3.5 (a-d), the average wind speed at WSM is compared with the average wind speed (WS) at 80 m by season and time of day to understand the advantages 66

Country MNG GD (/100 km2) P (million) C (kWh) G (kWh) Oman 2836 0.91 3.42 5991 7841 Saudi Arabia 18392 0.96 28.69 8022 6463 Iran 12587 0.78 66.43 2634 1811 Iraq 3618 0.83 31.13 1187 1473 Jordan 780 0.88 6.34 2216 1462 Egypt 9644 0.96 83.08 1670 1437 Yemen 3491 0.77 23.82 259 1254 UAE 602 0.85 4.80 10746 1123 Syria 1497 0.81 20.18 1809 938 Qatar 63 0.57 0.83 15075 845 Turkey 6215 0.80 76.81 2498 812 Kuwait 145 0.84 2.69 16759 683 Cyprus 22 0.38 0.53 4622 488 Israel 184 0.83 7.23 6953 313 Lebanon 42 0.42 4.02 3475 122 Palestine 37 0.58 4.12 - 112 Bahrain 1 0.18 0.73 17578 22

Table 3.1: Consumption (C) and potential AWE generation (G) for countries in the Middle East (per capita) with the maximum number of deployable generators (MNG), density of generators per 100 km2 (GD), and the population (P). C and G data are for 2010 from IEA Statistics OECD/IEA 2014. 67 a 8 b 40 40

9 6 30 30 4 6

20 20 2 3

0 30 40 50 60 30 40 50 60 c d 40 12.5 40

10.0 9 30 30 7.5 6 5.0 20 20 2.5 3

0.0 30 40 50 60 30 40 50 60 e f 40 40 0.00 0.1

-0.25 30 0.0 30

-0.1 -0.50 20 20

-0.2 -0.75

30 40 50 60 30 40 50 60 g h 40 40 0.00 0.2

0.1 -0.25 30 30 0.0 -0.50

20 -0.1 20 -0.75 -0.2

30 40 50 60 30 40 50 60

Figure 3.5: Difference in the average wind speed (m/s) between WSM and WS at 80 m during: (a) day time in January, (b) night time in January, (c) day time in July, (d) night time in July. The difference in CV in wind speed between WSM and WS at 80 m during: (e) day time in January, (f) night time in January, (g) day time in July, (h) night time in July. The figure including the map and all the text elements has been plotted using the software R version 3.3.2. [1] 68 of AWE. The advantages are most obvious in winter when jet streams are active. In summer, AWE appears to be more advantageous near the coast of the Red Sea, the Arabian Gulf, and the Gulf of Aden. The gain of AWE over near-surface wind resources is minimal in regions with a more complex terrain, such as in western Saudi Arabia and Yemen, especially during the day. In summer daytime, surface heating overland causes a well-mixed and 5-km deep boundary layer. The strong turbulent mixing of momentum equalizes winds at different altitudes [84]. The difference in the CV (Fig. 3.5, e-h) shows that WSM is less variable than the near-surface wind in winter. In summer, the variability increases slightly over land during the day except in coastal areas. At night, coastal areas experience a large reduction in variability in WSM compared with that of near-surface wind, likely due to the prominence of low-level jets.

3.4 Discussions

In this study, areas favorable for AWES deployment in the Middle East are identified with the diurnal and seasonal characteristics of AWE potential characterized. The abundance of wind resources increases with altitude, except in the coastal areas in summer nights. Variability in wind resources increases with altitude over coastal areas in summer but not so much in winter. Optimal altitudes of AWES deployment were computed in light of current technological constraint on the length of the tether. The regional AWE potential was estimated using a realistic power curve specification with assumptions about deployment conditions and spatial distribution. Average WSM is higher over land in general in winter north of 20◦ N and over Sudan. Coastal WSM is more prominent in summer. There is a lower variability in coastal areas in summer. The average altitude of WSM is lower during the summer, especially in coastal areas where the variability is higher. The per capita annual energy generation demonstrates the potential of AWES in fulfilling electricity needs at current levels 69 for several countries in the Middle East. In particular, Oman and Saudi Arabia have a high level of the potential power generation with low annual variability. Our estimates also compare favorably to the near-surface wind power potentials from a previous study in which higher average wind speeds and less variability were observed [13]. This finding hints at the potential utility of AWE for areas with high energy demand in the region. In future studies, we will focus on characterizing the regional impact of AWES, where annual production estimates across countries in the region would offer insight into the potential contribution of AWE to respective electricity grids. The effects of spatial aggregation of AWES in reducing power variability would offer insights into the impacts of large-scale deployment of AWES on the grid. The untapped potential and technical viability of AWES suggest that studies on spatial optimization of wind power systems where conventional turbines and AWES may co-exist are warranted. AWES presents an excellent opportunity to champion the technological transfer and development of a maturing next-generation technology in a region with an increasingly knowledge-based and energy-intensive economy. 70

Chapter 4

Variability mitigation with spatial clustering

Providing a continuous energy supply is a challenge for wind power generation systems. Variability in power generation would affect the reliability and energy efficiency of the power grid. Power generation planning and operation scheduling span various time-scales. These range from day-of-economic-dispatch, where real-time pricing and emergency occur, day-ahead for bidding and scheduling, to months for operational planning and maintenance, and years for systems planning. To improve reliability, variability in power generation should be understood and better characterized. The conventional approach is to characterize a bulk wind power density variability using a coefficient of variability that is not time interval-specific. This study aims to characterize the wind power variability at various time-scales of power operations to illustrate its effects across the Middle East via spectral analysis and clustering. Us- ing a high-resolution dataset obtained from a local area model simulation, this study showcases how aggregate variability may impact operation, and informs the planning of large-scale wind power integration in the Middle East in light of the scarcity of observational data.

4.1 Overview

With an increasing awareness to sustainable development, countries in the Middle East have been planning to adopt renewable energy at a large scale. Along with solar power, wind power projects have been taking off both globally, exemplified by 71 major gains in wind power generation [88], and as the countries in the region work to diversify their energy mix. Unlike solar power, wind power could work around-the- clock and is complementary to other forms of renewable energy sources. However, wind power faces some unique challenges, specifically in its variability. Variability impacts wind power generation especially in situations of large-scale grid integration, where wind power behaves very differently from conventional power resources such as coal and gas. Wind resource at a location is inherently variable and intermittent due to the underlying atmospheric processes that span multiple time scales. At the interannual timescales, a large fraction of the variability in wind resource is attributed to the impact of the El Nio Southern Oscillation (ENSO) [89]. Monsoons affect local wind conditions at the month-to-season time scales. In the order of days, cyclones operate at the synoptic time scale. Sea breezes affect local winds in the order of hours on the diurnal time scale. Regional studies have explored the potential of wind power generation. Yip et al. 2015 assessed the near-surface wind resource at various hub heights over the whole Arabian Peninsula using a assimilated gridded dataset with long record-length and described the wind resource in terms of abundance, variability, and intermittency characteristics [13]. Specifically, variability and intermittency were characterized using global metrics for the whole time period. Langodan et al. 2016 studied the wind and wave energy potentials in the Red Sea with high-resolution regional atmospheric reanalysis data, where diurnal, monthly, and seasonal variability in wind power density are illustrated [90]. Recently, Yip et al. 2017 characterized the elevated wind resources in the Middle East, noting seasonal difference in wind behaviors when comparing to near-surface wind resources [91]. Independent of where the wind resource is harvested, near surface wind exhibits higher temporal variability spanning different time scales. As intermittent and variable renewable energy generation is integrated to the grid, 72 Physical processes Variability in time bands Operation impacted Sea Breezes Within-day Same-day response Weather fronts 2 - 7 days Operational planning & maintenance Inter-annual variability Year-beyond System planning Table 4.1: Time-scales and their corresponding physical processes and operations impacted it is important to understand how these renewable generations introduce temporal constraints into resource availability. Grid operations fall into different time scales. Most renewables integration studies explore the effects of high penetration scenarios on hour-ahead scheduling. Service restoration and day-ahead scheduling concern the temporal scale from day to the synoptic scale, beyond which maintenance scheduling is relevant. There are corresponding atmospheric processes at work that concern these time scales. For example, sea breezes affect coastal areas within the diurnal scale, while weather fronts would affect larger areas at the synoptic time scale. Wind power variability and their potential impact on large-scale integration has not been studied in the Middle East. Wind power varies due to fluctuation in wind patterns that happen in various time scales. Wan 2012 examined wind power data from different parts of the USA and suggested a relatively large inter-annual changes (up to 40% between peek and drought years) driven by the climate and regional weather pattern [92]. Coughlin 2011 characterized the changes in wind power output and load data over a range of time scales in several operating regions in the USA [93]. Gunturu and Hallgren 2017 showed negative co-variations in hydro and wind power generation driven by El Nino Southern Oscillation which has implications for higher generation system adequacy in Australia [94]. Studies have also shown that aggregating wind power generation from a wide area reduces fluctuation in generation. This is due to the decrease in correlation of wind speed time series as the distance among sites increases, resulting in a more smoothed aggregate generation profile. Holttinen 2011 showed an increase in predictability and 73 reduction in variability and near-zero output due to geographical spreading in the Nordics [95]. Archer and Jacobson 2007 suggested that interconnecting wind plants would lead to steady deliverable power to support the base load [96]. Most studies explored wind power variability in the temporal domain. Katzenstein et al. 2010 proposed a frequency-dependent analysis of the smoothing of variability in wind power generation as more wind power plants are interconnected [97]. It identifies significant inter-annual variability and informs about the severity of wind drought years for 20 wind plants studied in Texas. There are three areas that have not been amply explored in the Middle East: wind power variability, the effects of time scales on wind power variability, and the spatial aggregation of wind power generation through large-scale deployment. These are important issues to address as wind power is poised to be a growing integral part of the future energy mix in the region. Ramping and intermittency in wind power generation require further investigation that are tied to existing understanding of the working of regional weather patterns of different scales. Wind power generation needs to be assessed both at a local and aggregated level for large-scale deployment to be relevant. In this study we aim to answer the following questions for the Middle East:

• What are the dominant time scales in the variability of wind power at each location?

• How does the variability at each location aﬀect grid operations at diﬀerent time scales?

• How does spatial aggregation mitigate variability in power generation? 74 4.2 Methods

4.2.1 Data

To answer the questions posed above, we need data on the wind field beyond what is currently collected at weather stations, which are sparse, incomplete, and have varying period of coverage. We conducted a simulation using a meso-scale model to generate the wind field with hourly temporal resolution to suit the needs of this study and interpolated to turbine hub height to estimate wind power generation using a reference power curve. To determine the dominant time scales in wind power variability, spectral analysis is carried out on the wind power generation dataset which enables us to investigate time-scales from hours to years. Clustering is performed to explore the effects of spatial aggregation on wind power generation variability reduction at different time scales. For the clustering, the generation data is masked by a capacity threshold to subset most productive areas and by their distance to transmission network to ensure feasibility.

Wind Fields

The dataset comes from the hourly-outputs of the Weather Research and Forecasting (WRF) model developed by the National Center for Atmospheric Research (NCAR, Boulder, Colorado, USA) has been used. This data allows us to investigate time- scales from hours to years. The Weather Research and Forecasting (WRF) model is a numerical weather prediction (NWP) and atmospheric simulation system designed for both research and operational applications [14]. The simulations were performed with the conﬁguration in appendix A. 75 Wind data at hub height

Using the U and V at 10m above ground level, horizontal wind speed is calculated at 100m using the wind profile power law with a scaling coefficient of 1/7. The 6 years worth of hourly wind speed at 100m above ground level is fitted to the power curve of a Nordex N100/2500 turbine for the time series of wind power generation estimate at each grid cell. The N100/2500 is an IEC III/a (low wind) turbine, this generation profile has been used in previous studies of wind resource assessment in the region.

4.2.2 Methods

Spectral analysis

To analyze the time-series data in the frequency domain, the power spectrum density of the wind power output of a location is estimated through the Fourier transform. To remove seasonality in the time-series, diurnal and seasonal signals and their ﬁrst two harmonics are removed. The fraction of variance of a speciﬁc location represented by a temporal period is represented by the area under the power spectrum density over the total area under the curve.

Spatial clustering

To reduce the variability in the wind power output, we conduct a clustering analysis with the following requirements. First, the clusters should be geographically contiguous, which corresponds to a rough estimate of how operating regions would be divided. Second, the clusters should be of comparable size, where there are not too many clusters formed. There is also a need to avoid unreasonable amount of wind turbines to be populated within each cluster. With these constraints, multiple clustering conﬁgurations are evaluated, and further simpliﬁed by a sequence of operations. Assumptions: 76 • Filter locations by capacity factor, where low-wind areas are excluded.

• Exclude locations that are too distant from existing or planned transmission network.

• Perform clustering with the correlations of locations in terms of their wind power time series and their spatial distances.

• Determine an optimal clustering conﬁguration by maximizing in-cluster coherence and the diﬀerences among the clusters.

In terms of location filtering, the wind data is spatially masked by a capacity factor threshold and the locations distance to the transmission network (fig 4.1), a data set publicly available from the World Bank Group [98]. The data covers the network over the Middle East and North Africa, with specified transmission capacity on the edges, and cities that are connected as nodes in the network. This is a necessary pre-processing step to limit the vast area under investigation to locations that are most likely of interest to wind developers. Subsetting the data allows us to focus on investigating areas of highest deployment potential and the effects of aggregations amongst these areas. Hierarchical clustering is performed with geographical distance constraints incor- porated to preserve spatially contiguous clusters [99]. The algorithm performs a ward-like hierarchical clustering where the distance between any two clusters are the squared euclidean distance between all points. There are 2 dissimilarity matrices, one that is calculated from the distance arise from the correlations amongst the time series at all locations, and one that is computed for the geographical distance of all locations. The two dissimilarity matrices are scaled and summed with a mixing parameter to determine the relative importance of geographical distance as a constraint. The mixing parameter is between 0 and 1, and is determined by maximizing the homogeneity within clusters. 77

Figure 4.1: Transmission network map of the Middle East, note the lack of transmission across the border of Yemen and Oman (World Bank Group, 2017)

We further calculated cluster-integrated power generation (abundance), coefficient of variation and the variability accounted for in different time-scales, the fraction of consumption fulfilled, and the number of proposed turbines that achieved this configuration.

4.3 Results

The results follow our quest to answer the three questions raised in the introduction. We first examine the variability of wind power generation over the region and note the prominent regions displayed in each time scale. We then explore the aggregate variance and generation characteristics of a clustering configuration that preserves coherence of the wind behavior and geographical closeness of the sites. We end this session by looking into the reduction of variability in wind power generation resulted from this clustering configuration as compared to a generation scenario without spatial aggregation. 78

Figure 4.2: Fraction of variance (color) accounted for in each temporal scale in wind power generation based on an N100 power curve at 100m above ground level

4.3.1 Dominant time scales of variability

To examine the dominant time scales of variability in wind power generation, we investigate a time intervals ranging from within-day, day-ahead, within-week, within- 2-weeks, within-month, and within-year, and beyond-year. The fraction of variability accounted for in each of these time bands are computed for every location in the spatial domain. The physical processes associated to these time scales are discussed, and linked to their potential impact on grid operations. The fractional variance by temporal band is computed by first removing the first 2 harmonics of the diurnal and seasonal cycle in the data before aggregating the 6 respective bands for each grid-cell in the spatial domain. Figure 4.2 illustrates the distribution of variance accounted for during within a day for onshore locations, particularly near the coast of the Red Sea. Open waters see higher variations accounted for in a week to 2 weeks period, and in the case of the coast facing the Indian Ocean, the variance concentrates in beyond a year. A general understanding of the time-scales and their effects on every location in 79

Figure 4.3: Wind power generating locations are ﬁltered by capacity factor at 0.3 and clustered spatially using hierarchical clustering upon dissimilarities based on temporal correlations of the locations and their geographical distance at a mixing ratio of 0.5 towards geographical distance, resulting in 9 clusters (color) the spatial domain provides a glance of its impact. However, due to the various operational constraints of the grid, such as the high capital cost of connections, the necessities in geospatial compensation across operating regions, the natural aggregation of generating sites within a region, and the inherent coherence of the wind ﬁeld, a clustering analysis is performed to investigate the behavior of aggregated wind power in spatially coherent regions, where their impacts on interconnection and balancing can be studied. 80 4.3.2 Clustering analysis

Figure 4.3 illustrates a clustering based on their spatial proximity, electricity network reach, and similarity of the time-series in each cell. Low-wind areas are excluded. First, we see clear distinction of clustered wind behavior over the sea as opposed to the coast of the along the Red Sea as well as in the Gulf of Arabia. Western Jordan shows a coherent cluster of wind pattern that is diﬀerent from the gulf of Aqaba, which is more resembling to the pattern along the Red Sea Coast. It is of interest to note the mountain ranges along northern Red Sea exhibit mixed clusters, which points to the possibility of interconnection of areas where the winds behave diﬀerently. This is also the area that hosts the future sustainability city of NEOM. NEOM is a new city and economic zone to be constructed in Tabuk, Saudi Arabia close to the border of Egypt to promote sustainable development and future urban living through advanced technologies. Central Saudi Arabia exhibit wind patterns that are closer to the east coast than to the west, and that the Gulf of Aden exhibit wind patterns that are unlike those in inland Yemen.

4.3.3 Cluster variance

Figure 4.4 illustrates the fraction of variance accounted for in each of the 6 time-bands considered. The 2-day to 7-day band represents a major portion of the variance in each cluster, followed by the one-week to two-weeks band. The mountains along the Red Sea coast represents a more even distribution of variance across time-scales, as opposed to the cluster in Jordan, which has a relatively high concentration of variance in the 2-days to 7-day periodicity. It is also of interest to note that clusters over sea exhibit most concentration on the weekly band while exhibiting lowest variance at higher frequencies. They also show lower variation beyond a year compared to other clusters. Treating these clusters as proposed balancing areas, the following section investi- 81

Figure 4.4: Fraction of variance (size, y-axis) accounted for in each temporal scale (x- axis) in aggregate wind power generation by cluster (color) based on an N100 power curve at 100m above ground level, where bubble size represents cluster annual generation potential, the grey bubble indicates variability by aggregating all generating clusters (aggregate generation at 40000 GWh a year) 82 Cluster Turbines Generation Consumption Frac. 1 (Jordan & Lebanon) 747 6183 0.52 % 2 (North-eastern Arabia & Riyadh) 1202 9002 0.76 % 3 (Aqaba & Suez (NEOM)) 422 4962 0.42 % 4 (Arabian Gulf) 341 2680 0.23 % 5 (Northern Red Sea Coast) 601 4921 0.42 % 6 (Southern Red Sea Coast) 552 4471 0.38 % 7 (North-eastern Oman) 292 2600 0.22 % 8 (Middle Red Sea) 243 2347 0.20 % 9 (Gulf of Aden) 260 2924 0.25 % Table 4.2: Characteristics of the 9-clustering with the amount of turbines proposed, total power generated in a year in GWh, and fraction of total consumption fulﬁlled is also computed using per capita electricity consumption data from World Bank 2013. gates their interactions, how spatial diversity reduces risks of loss of load, and how interconnection should be prioritized in these proposed balancing areas.

4.3.4 Clustering characteristics

In table 4.2, the largest coherent generating area identiﬁed is north-eastern Arabian peninsula along the Arabian Gulf and in central Saudi Arabia, followed by the cluster that dominates Jordan and areas in the north-west. The gulfs of Aqaba and Suez and the northern Red Sea coast show to be a more productive area for wind power generation than its southern counterpart. Smaller clusters exit along the Arabian Gulf, middle Red Sea, and Gulf of Aden. A more spatially scattered cluster exist in north-eastern Oman.

4.3.5 In-cluster variability reduction

Figure 4.5 shows the eﬀect of variability in wind power generation due to cluster aggregation. Cluster-wide fraction of variance (agg) is computed using the aggregated generation time series for each cluster. Maximum fraction of variance of individual locations in each cluster (max) is calculated by taking the maximum among all values of the fraction of variance at each individual location (after spatial ﬁltering) in a 83

Figure 4.5: Fraction of variance in time-bands for each cluster where cluster- aggregated generation (agg) is compared to a cell with maximum fraction of variance (max) in each cluster. 84 cluster for each time-band. Aggregated generation exhibits an overall lower fraction of variance at high frequency (day and day-ahead), and shows a similar level of fraction of variance at synoptic time scales, where slightly lower fraction of variance is shown in most clusters at low frequencies.

4.4 Discussions

Variability reduction and intermittency mitigation are important tasks when considering large-scale integration of wind power into the energy mix. This study utilizes model simulations to study the impacts of different timescales have on wind power generation and examine how spatial aggregation reduces generation variability. Different atmospheric processes influence influence wind power generation at different time scales. Isolating these timescales allow us to investigate the impact of certain processes on wind power generation scenarios and help better characterize operations. Together with clustering analysis, it can help estimate area-wide impacts of variability and improve its predictability due to atmospheric processes. Using spectral analysis, this study shows that variability concentrates on the diurnal (24-hour) time scale predominantly for onshore locations (especially areas with complicated terrains) in the Middle East, where offshore locations experience a higher concentration of variability in the synoptic time scales from 1 to 2 weeks. These time scales affect the same-day response and operational maintenance of grid operations. Better predictions of atmospheric processes at these time scales could lead to better characterization of operations especially in areas where the variability is concentrated. Under a clustering scenario, filtered locations with high generating potential are aggregated into 9 clusters based on constraints in spatial proximity and similarity in wind power generation behavior. Fractions of variability concentrates on the synoptic time scales after clustering. There is also a large reduction in the fraction of variability 85 at the diurnal time scale, with moderate to minimal reductions at other time scales across all 9 clusters when comparing the cluster aggregate generating time series to the maximum within the cluster. Looking further into the clusters, we see that most of the clusters are onshore. This is due to the distance to transmission network constraints we impose on the generating locations. It is important to note that grid connection cost for offshore wind projects account for 15 - 30% versus 9 - 14% for onshore projects in developed countries. Given this magnitude in capital cost difference, we focus our analysis primarily on onshore locations for wind power generation. However, the many islands that embrace the coast of Red Sea pose opportunities of distributed power generations that could eventually be connected to take advantage of wider area balancing. Amongst the proposed clusters, the Aqaba & Suez cluster is of particular importance with the recent announcement of a city of the future project, NEOM. This particular cluster spots a relatively high aggregate generation potential. Its geographic proximity to nearby clusters proposes a potential for inter-cluster balancing for a further reduction of wind power variability in that location. Existing and planned transmission networks are also well-posed for taking advantage of inter-cluster balancing, particularly along the Red Sea coast, which spans 5 clusters, and the GCC interconnect along the Arabian Gulf, which spans 3 clusters.

Comparison to prior work

Table 4.3 uses an established metrics from prior work to illustrate variability reduction with clustering, coefficient of variation is computed for the aggregated power generation timeseries for each cluster. Coefficient of variation is also computed for each location in all cluster, where a set of cluster-wide spatial statistics (minimum, mean, maximum) is computed for comparison. Most clusters see a 50% reduction in coefficient of variation from the mean, except 86 cluster min mean max clusterCv 1 0.68 0.93 1.16 0.51 2 0.68 1 1.15 0.64 3 0.5 0.7 0.97 0.36 4 0.95 1.03 1.16 0.81 5 0.61 0.93 1.13 0.52 6 0.68 0.92 1.13 0.45 7 0.7 0.9 1.08 0.42 8 0.71 0.89 1.05 0.7 9 0.52 0.77 1.12 0.52

Table 4.3: Coefficient of variation (CV) for aggregated power generation time series (clusterCv) and the spatial statistical summaries (min, mean, max) of CV of individual cells in a cluster in cluster number 4, 8, and 9 which correspond to sea regions of the Arabian Gulf, Red Sea near Jeddah, and Red Sea near Aden. The aggregated CV is lower than the minimum CV in all clusters. Variability in power generation is an important issue concerning reliability and energy efficiency of a power grid. As planning and operation scheduling span different time-scales, variability in power generation at these time-scales should be understood and better characterized. This study characterizes the wind power variability at various time-scales of power operations to illustrate its effects across the Middle East via spectral analysis and clustering. Using a high-resolution dataset obtained from a local area model simulation, this study showcases how aggregate variability may impact operation. 87

Chapter 5

Summary

Background

Power generation in the MENA region is in transition, primarily in response to the increasing demand, rising opportunity costs of oil, and the increasing need for sustainability and lowering the carbon footprint of the region. As levelized cost of electricity generated from renewable resources is on a steady decline, there is a need for evaluating the potential of these resources for power generation. Among the renewable energy technologies (wind, solar, and biofuels), wind power is a mature technology and has high power density. But wind resource, like other renewable resources, suﬀers from variability and intermittency arising out of atmospheric motions and processes. So, it is imperative to deploy wind farms in regions/locations with high abundance and low variability. Further, as the technology of airborne wind harvesting devices matures, the potential airborne wind resource can be exploited.

Purpose

This dissertation aims to answer the following questions:

• What are the spatial-temporal attributes of wind abundance, variability, intermittency in the region?

• What are the temporal characteristics of the feasible set of simulated wind farms? 88 • How do these temporal characteristics diﬀer at diﬀerent time-scales?

• What are the eﬀects of diﬀerent spatial aggregations on the temporal characteristics?

Methodology

Modern Era Retrospective-analysis for Research and Analysis (MERRA) is a reconstruction of the state of the atmosphere using a General Circulation Model with assimilation of observations from satellites, surface stations, radiosondes, and many other sources. The time period of the dataset covers from 1979 to the present. The boundary layer fields from MERRA have been used to construct the wind power density at different hub heights for the MENA region for the surface wind resource. The wind fields at different heights from MERRA have been used to construct the airborne wind resource. The high resolution (5×5 km2) wind resource at the surface and at high altitude has been constructed using the Weather Research and Forecasting model developed and maintained by the National Center for Atmospheric Research (USA). Surface and upper air observations from the region have also been analyzed.

Key results

The Red Sea, the Arabian Gulf and the Hejaz mountainous region have high wind resource. There is persistent year-round wind resource in the northern Red Sea, its northern coast, and the Arabian Gulf region. The variability of wind power density is low in the northern Red Sea and its coast, and southern Arabian Gulf. The airborne wind resource is abundant over northern Saudi Arabia and Oman. The maximum wind speed are higher over land and in winter. An important observation has been that the regions of high wind resource (The Red Sea, the Hejaz mountainous region, and the Arabian Gulf) are also regions of much of the power consumption. So, 89 transmission of the generated wind power can be cost effective. The quantification of uncertainties of the results from the datasets used in the study would be necessary before informing any field activity and development.

Signiﬁcance

The generated high- and ultra-high resolution datasets can be used to develop scenarios for optimal deployment of wind farms, and to study the economics of wind power generation in the presence of diﬀerent energy policies. Further, economic and technical feasibility of individual locations in the region can be assessed using the high and ultra-high resolution datasets in conjunction with a computational ﬂuid dynamic model with complex terrain representation. 90

REFERENCES

[1] R Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, 2016. [Online]. Available: https://www.R-project.org/

[2] L. El-Katiri, B. Fattouh, and Oxford Institute for Energy Studies, A roadmap for renewable energy in the Middle East and North Africa, 2014.

[3] J. Krane, “Beyond 12.5: The implications of an increase in saudi crude oil production capacity,” Energy Policy, vol. 110, pp. 542–547, 2017.

[4] G. Luderer, R. C. Pietzcker, S. Carrara, H. S. de Boer, S. Fujimori, N. Johnson, S. Mima, and D. Arent, “Assessment of wind and solar power in global low- carbon energy scenarios: An introduction,” Energy Economics, vol. 64, pp. 542– 551, 2017.

[5] J. Ball, D. Reicher, X. Sun, and C. Pollock, “The new solar system: Chinas evolving solar industry and its implications for competitive solar power in the united states and the world,” Stanford Univ., CA (United States), Tech. Rep., 2017.

[6] S. Emeis, Wind Energy Meteorology Atmospheric Physics for Wind Power Gen- eration. Berlin, Heidelberg: Springer, 2013.

[7] K. Marvel, B. Kravitz, and K. Caldeira, “Geophysical limits to global wind power,” Nature Climate Change, vol. 3, no. 2, pp. 118–121, Feb. 2013.

[8] U. B. Gunturu and C. A. Schlosser, “Characterization of wind power resource in the united states,” Atmos. Chem. Phys., vol. 12, no. 20, pp. 9687–9702, Oct. 2012. 91 [9] Y. Zhang and S. Ula, “Comparison and evaluation of three main types of wind turbines,” in Transmission and Distribution Conference and Exposition, 2008. T&D. IEEE/PES. IEEE, 2008, pp. 1–6.

[10] M. Caduﬀ, M. A. Huijbregts, H.-J. Althaus, A. Koehler, and S. Hellweg, “Wind power electricity: the bigger the turbine, the greener the electricity?” Environ- mental science & technology, vol. 46, no. 9, pp. 4725–4733, 2012.

[11] U. Ahrens, Ed., Airborne Wind Energy, ser. Green energy and technology. Hei- delberg: Springer, 2014.

[12] M. M. Rienecker, M. J. Suarez, R. Gelaro, R. Todling, J. Bacmeister, E. Liu, M. G. Bosilovich, S. D. Schubert, L. Takacs, G.-K. Kim et al., “Merra: Nasas modern-era retrospective analysis for research and applications,” Journal of climate, vol. 24, no. 14, pp. 3624–3648, 2011.

[13] C. M. A. Yip, U. B. Gunturu, and G. L. Stenchikov, “Wind resource characterization in the Arabian Peninsula,” Applied Energy, vol. 164, pp. 826–836, Feb. 2016.

[14] W. C. Skamarock, J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, “A description of the advanced research wrf version 2,” Na- tional Center For Atmospheric Research Boulder Co Mesoscale and Microscale Meteorology Div, Tech. Rep., 2005.

[15] R. J. Barthelmie and S. C. Pryor, “Potential contribution of wind energy to climate change mitigation,” Nature Climate Change, vol. 4, no. 8, pp. 684–688, Aug. 2014.

[16] M. B. McElroy, X. Lu, C. P. Nielsen, and Y. Wang, “Potential for wind-generated electricity in china,” Science, vol. 325, no. 5946, pp. 1378–1380, Sep. 2009.

[17] J. Bryden, L. Riahi, and R. Zissler, “MENA renewables status report,” Tech. Rep., 2013. [Online]. Available: http://www.ren21.net/Portals/0/documents/ activities/Regional%20Reports/MENA 2013 highres.pdf 92 [18] A. Smith, N. Lott, and R. Vose, “The integrated surface database: Recent developments and partnerships,” Bulletin of the American Meteorological Society, vol. 92, no. 6, pp. 704–708, Jun. 2011.

[19] J. Ansari, I. K. Madni, and H. Bakhsh, “Saudi arabian wind energy atlas,” Riyadh, Saudi Arabia: KACST, pp. 1–27, 1986.

[20] S. Rehman and T. O. Halawani, “Statistical characteristics of wind in saudi arabia,” Renewable Energy, vol. 4, no. 8, pp. 949–956, Nov. 1994.

[21] A. Ucar and F. Balo, “Investigation of wind characteristics and assessment of wind-generation potentiality in uluda˘g-bursa,turkey,” Applied Energy, vol. 86, no. 3, pp. 333–339, Mar. 2009.

[22] I. Fyrippis, P. J. Axaopoulos, and G. Panayiotou, “Wind energy potential assessment in naxos island, greece,” Applied Energy, vol. 87, no. 2, pp. 577–586, Feb. 2010.

[23] S. A. Akda˘g,H. S. Bagiorgas, and G. Mihalakakou, “Use of two-component weibull mixtures in the analysis of wind speed in the eastern mediterranean,” Applied Energy, vol. 87, no. 8, pp. 2566–2573, Aug. 2010.

[24] O. Ohunakin, M. Adaramola, and O. Oyewola, “Wind energy evaluation for electricity generation using WECS in seven selected locations in nigeria,” Applied Energy, vol. 88, no. 9, pp. 3197–3206, Sep. 2011.

[25] J. A. Jervase and A. M. Al-Lawati, “Wind energy potential assessment for the sultanate of oman,” Renewable and Sustainable Energy Reviews, vol. 16, no. 3, pp. 1496–1507, Apr. 2012.

[26] S. Al-Yahyai, Y. Charabi, A. Al-Badi, and A. Gastli, “Nested ensemble NWP approach for wind energy assessment,” Renewable Energy, vol. 37, no. 1, pp. 150–160, Jan. 2012.

[27] S. Al-Yahyai and Y. Charabi, “Assessment of large-scale wind energy potential in the emerging city of duqm (oman),” Renewable and Sustainable Energy Reviews, vol. 47, pp. 438–447, Jul. 2015. 93 [28] Y. Charabi, S. Al-Yahyai, and A. Gastli, “Evaluation of NWP performance for wind energy resource assessment in oman,” Renewable and Sustainable Energy Reviews, vol. 15, no. 3, pp. 1545–1555, Apr. 2011.

[29] B. Kirby, E. Ela, and M. Milligan, “Chapter 7 - analyzing the impact of variable energy resources on power system reserves,” in Renewable Energy Integration, L. E. Jones, Ed. Boston: Academic Press, 2014, pp. 83–99.

[30] S.-D. Kwon, “Uncertainty analysis of wind energy potential assessment,” Applied Energy, vol. 87, no. 3, pp. 856–865, Mar. 2010.

[31] J. Lin, Y.-z. Sun, L. Cheng, and W.-z. Gao, “Assessment of the power reduction of wind farms under extreme wind condition by a high resolution simulation model,” Applied Energy, vol. 96, pp. 21–32, Aug. 2012.

[32] S. Jung, O. Arda Vanli, and S.-D. Kwon, “Wind energy potential assessment considering the uncertainties due to limited data,” Applied Energy, vol. 102, pp. 1492–1503, Feb. 2013.

[33] S. Rehman and A. Ahmad, “Assessment of wind energy potential for coastal locations of the kingdom of saudi arabia,” Energy, vol. 29, no. 8, pp. 1105–1115, Jun. 2004.

[34] S. Rehman, T. O. Halawani, and T. Husain, “Weibull parameters for wind speed distribution in saudi arabia,” Solar Energy, vol. 53, no. 6, pp. 473–479, Dec. 1994.

[35] S. Rehman, A. M. Mahbub Alam, J. P. Meyer, and L. M. Al-Hadhrami, “Wind speed characteristics and resource assessment using weibull parameters,” Inter- national Journal of Green Energy, vol. 9, no. 8, pp. 800–814, Mar. 2012.

[36] F. A. L. Jowder, “Wind power analysis and site matching of wind turbine generators in kingdom of bahrain,” Applied Energy, vol. 86, no. 4, pp. 538–545, Apr. 2009.

[37] T. B. M. J. Ouarda, C. Charron, J. Y. Shin, P. R. Marpu, A. H. Al-Mandoos, M. H. Al-Tamimi, H. Ghedira, and T. N. Al Hosary, “Probability distributions 94 of wind speed in the UAE,” Energy Conversion and Management, vol. 93, pp. 414–434, Mar. 2015.

[38] W. Chen, S. Castruccio, M. G. Genton, and P. Crippa, “Current and Future Estimates of Wind Energy Potential Over Saudi Arabia,” Journal of Geophysical Research: Atmospheres, vol. 123, no. 12, pp. 6443–6459, Jun. 2018.

[39] A. Cosseron, U. B. Gunturu, and C. A. Schlosser, “Characterization of the wind power resource in europe and its intermittency,” Energy Procedia, vol. 40, pp. 58–66, 2013.

[40] C. Fant, B. Gunturu, and A. Schlosser, “Characterizing wind power resource reliability in southern africa,” Applied Energy.

[41] W. Hallgren, U. B. Gunturu, and A. Schlosser, “The potential wind power resource in australia: A new perspective,” PLoS ONE, vol. 9, no. 7, p. e99608, Jul. 2014.

[42] M. M. Rienecker, M. J. Suarez, R. Gelaro, R. Todling, J. Bacmeister, E. Liu, M. G. Bosilovich, S. D. Schubert, L. Takacs, G.-K. Kim, S. Bloom, J. Chen, D. Collins, A. Conaty, A. da Silva, W. Gu, J. Joiner, R. D. Koster, R. Lucchesi, A. Molod, T. Owens, S. Pawson, P. Pegion, C. R. Redder, R. Reichle, F. R. Robertson, A. G. Ruddick, M. Sienkiewicz, and J. Woollen, “MERRA: NASA’s modern-era retrospective analysis for research and applications,” Journal of Cli- mate, vol. 24, no. 14, pp. 3624–3648, Jul. 2011.

[43] A. J. Dyer and B. B. Hicks, “Flux-gradient relationships in the constant ﬂux layer,” Quarterly Journal of the Royal Meteorological Society, vol. 96, no. 410, pp. 715–721, Oct. 1970.

[44] C. L. Archer, “Spatial and temporal distributions of u.s. winds and wind power at 80 m derived from measurements,” Journal of Geophysical Research, vol. 108, no. D9, 2003.

[45] S. Rehman, “Wind energy resources assessment for yanbo, saudi arabia,” Energy Conversion and Management, vol. 45, no. 13–14, pp. 2019–2032, Aug. 2004. 95 [46] E. Fırtın, O.¨ Güler,and S. A. Akda˘g,“Investigation of wind shear coefficients and their effect on electrical energy generation,” Applied Energy, vol. 88, no. 11, pp. 4097–4105, Nov. 2011.

[47] W. Short and P. Sullivan, “Regional energy deployment system (ReEDS),” Na- tional Renewable Energy Laboratory, Tech. Rep. NREL/TP-6A20-46534, Dec. 2011.

[48] F. Cassola and M. Burlando, “Wind speed and wind energy forecast through kalman ﬁltering of numerical weather prediction model output,” Applied Energy, vol. 99, no. 0, pp. 154–166, Nov. 2012.

[49] C. Draxl, A. Clifton, B.-M. Hodge, and J. McCaa, “The wind integration national dataset (WIND) toolkit,” Applied Energy, vol. 151, pp. 355–366, Aug. 2015.

[50] King Abdullah City for Atomic and Renewable Energy (K.A.CARE), “Renewable resource atlas,” 2015. [Online]. Available: https://rratlas.kacare. gov.sa

[51] N. M. Al-Abbadi, “Wind energy resource assessment for ﬁve locations in saudi arabia,” Renewable Energy, vol. 30, no. 10, pp. 1489–1499, Aug. 2005.

[52] S. Rehman, “Prospects of wind farm development in saudi arabia,” Renewable Energy, vol. 30, no. 3, pp. 447–463, Mar. 2005.

[53] S. AL-Yahyai, Y. Charabi, A. Gastli, and S. Al-Alawi, “Assessment of wind energy potential locations in oman using data from existing weather stations,” Renewable and Sustainable Energy Reviews, vol. 14, no. 5, pp. 1428–1436, Jun. 2010.

[54] A. H. Algifri, “Wind energy potential in aden-yemen,” Renewable Energy, vol. 13, no. 2, pp. 255–260, Feb. 1998.

[55] K. M. Bataineh and D. Dalalah, “Assessment of wind energy potential for selected areas in jordan,” Renewable Energy, vol. 59, pp. 75–81, Nov. 2013. 96 [56] W. Al-Nassar, S. Alhajraf, A. Al-Enizi, and L. Al-Awadhi, “Potential wind power generation in the state of kuwait,” Renewable Energy, vol. 30, no. 14, pp. 2149– 2161, Nov. 2005.

[57] M. N. Schwartz and D. L. Elliott, “Wind shear characteristics at central plains tall towers,” National Renewable Energy Laboratory, Tech. Rep., 2006. [Online]. Available: http://www.mapcruzin.com/wind-power-publications/wind-issues/ 40019.pdf

[58] Vestas, “Mesoscale model description,” Tech. Rep., 2013. [On- line]. Available: http://rratlas.kacare.gov.sa/RRMMDataPortal/Content/pdf/ 121019mazag MesoscaleModelDescription.pdf

[59] International Renewable Energy Agency, “Global atlas for renewable energy 2.0,” 2015. [Online]. Available: http://irena.masdar.ac.ae

[60] 3TIER Inc., “3tier global wind dataset annual mean validation,” 3TIER Inc., Tech. Rep., Jan. 2009. [Online]. Available: http://www.3tier.com/static/ttcms/us/documents/publications/ validations/3TIER WP FL gbl wind validation.pdf

[61] K. Klink, “Atmospheric circulation eﬀects on wind speed variability at turbine height,” Journal of Applied Meteorology and Climatology, vol. 46, no. 4, pp. 445–456, Apr. 2007.

[62] H. M. Cullen, A. Kaplan, P. A. Arkin, and P. B. deMenocal, “Impact of the north atlantic oscillation on middle eastern climate and streamﬂow,” Climatic Change, vol. 55, no. 3, pp. 315–338, Nov. 2002.

[63] M. J. Shawon, L. El Chaar, and L. A. Lamont, “Overview of wind energy and its cost in the middle east,” Sustainable Energy Technologies and Assessments, vol. 2, pp. 1–11, Jun. 2013.

[64] S. Rehman, T. O. Halawani, and M. Mohandes, “Wind power cost assessment at twenty locations in the kingdom of saudi arabia,” Renewable Energy, vol. 28, no. 4, pp. 573–583, Apr. 2003. 97 [65] A.-H. Maraﬁa and H. A. Ashour, “Economics of oﬀ-shore/on-shore wind energy systems in qatar,” Renewable Energy, vol. 28, no. 12, pp. 1953–1963, Oct. 2003.

[66] M. L. Kubik, D. J. Brayshaw, P. J. Coker, and J. F. Barlow, “Exploring the role of reanalysis data in simulating regional wind generation variability over northern ireland,” Renewable Energy, vol. 57, pp. 558–561, Sep. 2013.

[67] D. J. Cannon, D. J. Brayshaw, J. Methven, P. J. Coker, and D. Lenaghan, “Using reanalysis data to quantify extreme wind power generation statistics: A 33 year case study in great britain,” Renewable Energy, vol. 75, pp. 767–778, Mar. 2015.

[68] J. Olauson and M. Bergkvist, “Modelling the swedish wind power production using MERRA reanalysis data,” Renewable Energy, vol. 76, pp. 717–725, Apr. 2015.

[69] I. Staﬀell and R. Green, “How does wind farm performance decline with age?” Renewable Energy, vol. 66, pp. 775–786, Jun. 2014.

[70] C. Fant, C. Adam Schlosser, and K. Strzepek, “The impact of climate change on wind and solar resources in southern africa,” Applied Energy, 2015.

[71] J. A. Carta, P. Cabrera, J. M. Mat´ıas,and F. Castellano, “Comparison of fea- ture selection methods using ANNs in MCP-wind speed methods. a case study,” Applied Energy, vol. 158, pp. 490–507, Nov. 2015.

[72] U. B. Gunturu and C. A. Schlosser, “Behavior of the aggregate wind resource in the ISO regions in the united states,” Applied Energy, vol. 144, pp. 175–181, Apr. 2015.

[73] REN21, “Renewables 2016 Global Status Report,” REN21 Secretariat, Paris, Tech. Rep., 2016.

[74] M. Z. Jacobson and M. A. Delucchi, “Providing all global energy with wind, water, and solar power, Part I: Technologies, energy resources, quantities and areas of infrastructure, and materials,” Energy Policy, vol. 39, no. 3, pp. 1154– 1169, Mar. 2011. 98 [75] C. L. Archer and K. Caldeira, “Global Assessment of High-Altitude Wind Power,” vol. 2, no. 2, pp. 307–319, May 2009.

[76] A. Cherubini, A. Papini, R. Vertechy, and M. Fontana, “Airborne Wind En- ergy Systems: A review of the technologies,” Renewable and Sustainable Energy Reviews, vol. 51, pp. 1461–1476, Nov. 2015.

[77] E. Lunney, M. Ban, N. Duic, and A. Foley, “A state-of-the-art review and feasibility analysis of high altitude wind power in northern ireland,” Renewable and Sustainable Energy Reviews, vol. 68, pp. 899–911, 2017.

[78] S. P. Global Modeling and Assimilation Oﬃce, “MERRA2 inst3 3d asm Np: 3d,3-Hourly,Instantaneous,Pressure-Level,Assimilation,Assimilated Meteorolog- ical Fields V5.12.4,” 2015.

[79] R. Gelaro, W. McCarty, M. J. Surez, R. Todling, A. Molod, L. Takacs, C. A. Randles, A. Darmenov, M. G. Bosilovich, R. Reichle, K. Wargan, L. Coy, R. Cullather, C. Draper, S. Akella, V. Buchard, A. Conaty, A. M. da Silva, W. Gu, G.-K. Kim, R. Koster, R. Lucchesi, D. Merkova, J. E. Nielsen, G. Partyka, S. Pawson, W. Putman, M. Rienecker, S. D. Schubert, M. Sienkiewicz, and B. Zhao, “The modern-era retrospective analysis for research and applications, version 2 (merra-2),” Journal of Climate, vol. 30, no. 14, pp. 5419–5454, 2017. [Online]. Available: https://doi.org/10.1175/JCLI-D-16-0758.1

[80] C. L. Archer, L. Delle Monache, and D. L. Rife, “Airborne wind energy: Optimal locations and variability,” Renewable Energy, vol. 64, pp. 180–186, Apr. 2014.

[81] P. Jish Prakash, G. Stenchikov, S. Kalenderski, S. Osipov, and H. Bangalath, “The impact of dust storms on the Arabian Peninsula and the Red Sea,” Atmo- spheric Chemistry and Physics, vol. 15, no. 1, pp. 199–222, Jan. 2015.

[82] D. L. Rife, J. O. Pinto, A. J. Monaghan, C. A. Davis, and J. R. Hannan, “Global Distribution and Characteristics of Diurnally Varying Low-Level Jets,” Journal of Climate, vol. 23, no. 19, pp. 5041–5064, May 2010. 99 [83] M. Canale, L. Fagiano, and M. Milanese, “KiteGen: A revolution in wind energy generation,” vol. 34, no. 3, pp. 355–361, Mar. 2009.

[84] M. Gamo, “Thickness of the dry convection and large-scale subsidence above deserts,” Boundary-Layer Meteorology, vol. 79, no. 3, pp. 265–278, May 1996. [Online]. Available: http://dx.doi.org/10.1007/BF00119441

[85] A. K. Blackadar, “Boundary layer wind maxima and their signiﬁcance for growth of nocturnal inversions,” Bull. Amer. Meteor. Soc., vol. 38, pp. 283–290, 1957.

[86] J. R. Holton, “The diurnal boundary layer wind oscillation above sloping terrain,” Tellus, vol. 19, no. 2, pp. 200–205, 1967.

[87] R. B. Stull, An Introduction to Boundary Layer Meteorology. Springer Science & Business Media, Jul. 1988.

[88] “Clean Energy Gains Ground,” https://eos.org/articles/clean-energy-gains- ground.

[89] C. Turney, “El Ni˜noand the Southern Oscillation. Multiscale variability and global and regional impacts, H. F. Diaz and V. Markgraf (eds). Publisher Cam- bridge University Press, Cambridge 2000 (496 pp) ISBN 0-521-62138-0,” Journal of Quaternary Science, vol. 18, no. 5, pp. 467–468, Jul. 2003.

[90] S. Langodan, Y. Viswanadhapalli, H. P. Dasari, O. Knio, and I. Hoteit, “A high-resolution assessment of wind and wave energy potentials in the Red Sea,” Applied Energy, vol. 181, pp. 244–255, Nov. 2016.

[91] C. M. A. Yip, U. B. Gunturu, and G. L. Stenchikov, “High-altitude wind resources in the Middle East,” Scientiﬁc Reports, vol. 7, no. 1, p. 9885, Aug. 2017.

[92] Y. H. Wan, “Long-Term Wind Power Variability,” Renewable Energy, p. 39, 2012.

[93] K. Coughlin, “Analysis of Wind Power and Load Data at Multiple Time Scales,” Jan. 2011. 100 [94] U. B. Gunturu and W. Hallgren, “Asynchrony of wind and hydropower resources in Australia,” Scientiﬁc Reports, vol. 7, Aug. 2017.

[95] H. Holttinen, P. Meibom, A. Orths, B. Lange, M. O’Malley, J. O. Tande, A. Es- tanqueiro, E. Gomez, L. S¨oder,G. Strbac, J. C. Smith, and F. van Hulle, “Im- pacts of large amounts of wind power on design and operation of power systems, results of IEA collaboration,” Wind Energy, vol. 14, no. 2, pp. 179–192, Mar. 2011.

[96] C. L. Archer and M. Z. Jacobson, “Supplying Baseload Power and Reducing Transmission Requirements by Interconnecting Wind Farms,” Journal of Applied Meteorology and Climatology, vol. 46, no. 11, pp. 1701–1717, Nov. 2007.

[97] W. Katzenstein, E. Fertig, and J. Apt, “The variability of interconnected wind plants,” Energy Policy, vol. 38, no. 8, pp. 4400–4410, Aug. 2010.

[98] “ENERGYDATA.INFO,” https://energydata.info/dataset/mena-electricity- transmission-network-2017.

[99] M. Chavent, V. Kuentz-Simonet, A. Labenne, and J. Saracco, “ClustGeo: An R package for hierarchical clustering with spatial constraints,” Computational Statistics, vol. 33, no. 4, pp. 1799–1822, Dec. 2018. 101

APPENDICES

A High resolution simulations: Analysis, Data Description, and Processing

High resolution simulations and analysis

Chapters 2 and 3 utilized atmospheric state at a coarse resolution of 50 × 67 km2 to construct the wind resource at that spatial scale. Particularly, the MERRA (Reanal- ysis) dataset developed by NASA was utilized. It has the following advantages:

• Observational data has been assimilated in the model run at each time-step to derive a dynamically consistent atmospheric state. Thus, a high ﬁdelity to observed state is maintained.

• The long record length ( 37 years) is able to capture the long-term variability of the wind climatology in the region. Further, the long record length allows to assess the impact of large-scale circulations on the wind resource in the region.

But the wind that transfers momentum to the harvesting devices (turbines or airborne devices) is strongly impacted by the features of the boundary layer. Most importantly, it is impacted by:

Land surface processes The intense thermals generated during the day (particularly during hot summers) enhance the turbulence in the atmospheric boundary layer and the wind speeds are impacted. 102 Topography Topographic features like hills, mountains, and escarpments are locations where wind is usually accelerated. But at the same time, these features of complex terrain generate turbulence and turbulent wakes.

Land-sea breezes Coastal regions are impacted by land and sea breezed in response to the diﬀerential heating of sea and land during the day and night.

Convective events and clouds Convective events like clouds and dry convective motions create strong updrafts and downdrafts create strong turbulence in the boundary layer.

Mountain passes & barrier jets Mountain passes are the gaps between topographic features, and wind is accelerated in these regions following Bernoulli principle. Long mountain ranges like that on the west coast of Saudi Arabia create barrier jets that ﬂow along the mountains.

Nocturnal jets The heating and cooling of the land surface, during the day and the night respectively, creates accelerated wind ﬂow close to the surface (100-300 m AGL) at night, peaking in the early morning. This phenomenon is the dominant mechanism for abundant wind resource in diﬀerent regions of the world. The central plains of the United States and northeast China are cases in point.

While the coarse resolution assessment in the previous two chapters gives a ﬁrst approximation of the wind resource in a region, it is imperative to resolve these atmospheric and land-surface characteristics of a region to get accurate characterization of wind resource in a region. Resolution of these features and processes mandates the use of an atmospheric model at high resolution. To assess the wind resource at high resolution, the Weather Research and Fore- casting (WRF) model developed by the National Center for Atmospheric Research (NCAR, Boulder, Colorado, USA) has been used. The speciﬁc version of the WRF 103 model used is WRF-ARW version 3.9.1. WRF is a computationally expensive model for long-term simulations. So, to better capture the inter-annual variations of the wind resource while maintaining the high spatial resolution, two sets of simulations have been performed:

HRS 1 A simulation at 5x5km2 for six years (2009-2014), and

HRS 2 Another simulation at 1x1km2 for one year (2014).

Domain of the simulations A meticulous of the model domain is imperative such that the large-scale and synoptic-scale circulations over the region are appropriately accounted for.

Domain for HRS 1 The spatial setup of this simulation consists of the geographical region between x1 and x2 latitudes and y1 and y2 longitudes.

Domain for HRS 2 The spatial setup of the simulation consists of the geographical region described (see ﬁg. A.1). This larger domain further consists of the following sub-domains.

d01 The larger domain with a grid size of 9×9 km2, called the mother domain. This larger domain captures the large-scale dynamics of the atmosphere.

d02 A sub-domain of d01 consists of the smaller domain drawn in white color in the ﬁgure. This sub-domain has a resolution of 3 × 3 km2.

d03 The smallest sub-domain of the model setup of HRS 2 consists of the region drawn in the ﬁgure in red color. The spatial resolution of this nested sub- domain is 1 × 1 km2.

Initialization of the model and the boundary conditions Both the simulations – HRS 1 and HRS 2 – were initialized with the state of the atmosphere from 104

Figure A.1: Spatial domain setup of the simulations: d01 (9 km x 9 km), d02 (3 km x 3 km), and d03 (1 km x 1 km) 105

Figure A.2: Average wind speed (m/s) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution the European Center for Medium-range Weather Forecasting (ECMWF)/Integrated Forecasting System (IFS) analysis dataset. This dataset has a spatial resolution of 15 × 15 km2 and a temporal resolution of 6 hours. The same dataset was used as the boundary condition for the two simulations. Model results follow.

Mean wind speed

The regions that are known to have higher wind speeds, such as the Bab-el-Mandeb, the Gulf of Aqaba, and the Gulf of Suez show further higher wind speeds in the 106

Figure A.3: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 107

Figure A.4: Coeﬃcient of variation in wind speed at 100m spatial distribution by 10 classes from the WRF model at 5km resolution 108

Figure A.5: Normalized differences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 109 high resolution simulation HRS 1. Much of the increase in the wind speeds is due to the topographical features around these regions. The west coast of the Arabian Peninsula along the Red Sea has lower wind speeds while the mountainous regions along the coast have higher wind speeds. The Tokar gap in the mountainous region along the east coast of Africa along the Red Sea has the highest wind speeds, which is corroborated by several reports that studied the impact of the accelerated wind jet through this gap on the oceanography of the Red Sea. The southern mountainous region in Yemen, also has high wind resource, primarily driven by the topography and the monsoonal jet that flows along this mountain range. The coast of the Gulf of Oman around Muscat also has modest wind resource. The most important inference that can be drawn from these results is that the topographical features and the two seas – the Red Sea and the Arabian Sea – drive much of the wind resource in this region. It is instructive to compare these results with those from chapter 2 (henceforth referred to as SWMERRA). Largely, the regions identified to have appreciable wind resource in SWMERRA have increased resource in HRS-1 and the regions that have moderate to low resource in SWMERRA have decreased resource from SWMERRA in HRS-1, as shown in fig. A.7.

Variability of wind speed

The variability (unsteadiness of wind speed) is measured in terms of the coeﬃcient of variation in this study. Generally, the regions inﬂuenced by the Red Sea such as the Gulf of Aqaba, the Gulf of Aden, and the Gulf of Suez, have high resource as discussed above, but have low variability. The southern mountainous region of Yemen also has high resource and low variability. There are pockets of regions with wind resource with low variability along the Arabian coast of the Red Sea. The mountainous region 110

Figure A.6: Average wind speed (m/s) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution 111

Figure A.7: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 112

Figure A.8: Coeﬃcient of variation in wind speed at 100m spatial distribution by 10 classes from the WRF model at 5km resolution 113

Figure A.9: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 114

Figure A.10: Average wind power generation (kW) at 100m spatial distribution by 10 classes from the WRF model at 5km resolution

(the Hejaz range) along the Red Sea (west coast) also has high wind resource but the variability is high. The regions with moderate and low resource usually have high variability. For instance, the region between Medina and Riyadh have moderate resource but high variability. Comparison with SWMERRA shows that (ﬁg. A.9) MERRA dataset tends to overestimate the variability in areas with complex terrain. 115

Figure A.11: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 116

Figure A.12: Percentage diﬀerence in average power generation at 80m from 100m in WRF 117

Figure A.13: Percentage diﬀerence in average power generation at 140m from 100m in WRF 118 Mean power generation

The generated power is cubic in wind speed. Therefore, the linear changes in wind speed unintuitively map to nonlinear differences in power generation. Qualitatively, regions with high wind speed correspond to regions with high generation. Therefore, the Gulfs of Aden, Suez, and Aqaba have persistently high potential generation of wind power. The mountainous region in the west coast of the Arabian Peninsula has high potential mean generation, but the adjoining coast has slightly lower generation. The Tokar gap has one of the highest potential generation. The southern topographic region of Yemen and the Oman coast also have high potential generation. In comparison, the regions with moderate wind speeds have low potential generation. Regions surrounded by topographical features like mountains themselves have lower potential generation, but the surrounding topography features moderate to high potential generation. This fact is exemplified by the Dead Sea. Compared with SWMERRA, the east coast of the Arabian Peninsula has little difference in potential generation due to lack of topographic features. In other regions that show considerable difference in wind speeds from those in SWMERRA, there is an appreciable increase in the potential power generation.

Influence of the turbine hub height It is generally accepted that the wind speed and hence the potential power generation increases with hub height, as the influence of the surface features is minimized. But this is generally not true everywhere. For instance, the regions that have low surface roughness, like the sea surfaces, coastal regions are likely exceptions to this general rule. Figures A.12 and A.13 show the impact of raising or lowering the hub height. The effect is most significant along the Omani coast of the Gulf of Oman, and along the southern Red Sea. Interestingly, the potential power generation increases when the hub height is decreased from 100m to 80m in the Gulfs of the Red Sea and the Tokar 119

Figure A.14: Coeﬃcient of variation in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution gap. This is due to the presence of low level jets in those regions. Thus, increasing the hub height may not justify the investment in these regions.

Variability in power generation

Potential wind power generation in regions/locations with low roughness length usually has lower variability. In the MENA region, the Gulfs of Aqaba, Suez and Aden have very low variability. High generation variability is usually a result of periodic 120

Figure A.15: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 121

Figure A.16: Percentage diﬀerence in coeﬃcient of variation in power generation at 80m from 100m in WRF 122

Figure A.17: Percentage difference in coefficient of variation in power generation at 140m from 100m in WRF 123 change of wind like in land and sea breezes, and intermittent windy events like storms. Thus, the coastal regions have high variability. But in the Omani coast of the Gulf of Oman, the high resource is generated by topography and hence is less variable. SWMERRA underestimates the variability in generation in regions with low roughness length and overestimates in the coasts along the Arabian Gulf. The variabiity in HRS 1 is significantly higher for the region between Medina and Riyadh where the roughness length is low. The potential generation is also low in this region. Similarly, SWMERRA underestimates the variability in potential generation off the coast of Salalah in Oman.

Impact of the hub height on variability In conformance with the atmospheric boundary layer theory, the variability of potential generation decreases as the hub height decreases from 100m to 80m. A signiﬁcant diﬀerence is seen in the Gulfs of Aqaba and Suez where the variability decreases, and it remains unchanged in the Gulf of Aden. As the hub height increases from 100m to 140m, in general, the variability decreases. But it increases in the three gulfs – Aqaba, Suez, and Aden, as well as in the Tokar gap.

Availability of generated power

In the context of this study, availability is deﬁned as the percentage of time during which the wind power generation is above 10% of the nameplate capacity. Open seas are characterized by very high availability. In waters close to the coast and in the gulfs (Aqaba, Suez, and Aden), and in several regions along the coasts of the Red Sea, the availability is high. In the mountainous regions of southern Yemen and along the coast of the Gulf of Oman, the availability is high. The area between Medina and Riyadh also shows moderate availability. 124

Figure A.18: Availability in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution 125

Figure A.19: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 126

Figure A.20: Percentage diﬀerence in availability in power generation at 80m from 100m in WRF 127

Figure A.21: Percentage diﬀerence in availability in power generation at 140m from 100m in WRF 128 SWMERRA tends to overestimate the availability over the sea compared to HRS 1 and underestimates in regions with topography along the Red Sea, along the coast of the Gulf of Oman, and the mountainous areas in southern Yemen.

Impact of the turbine hub height Lowering the hub height from 100m to 80m results in lower availability in general as the wind speeds decrease at the lower altitude. But over the Red Sea, there is an increase in availability. Further, the magnitude of decrease is also lower in the mountainous regions in the Red Sea coast and in Oman. The inland regions have a signiﬁcant decrease in availability with lowered hub height. When the hub height increases from 100m to 140m, the eﬀects are opposite to the above.

Mean episode length

Episode length is the time period between the time the generated power crosses 10% of the nameplate capacity while increasing and that while decreasing. Lower mean episodes are found usually in open seas. Inland, long generation episode lengths are found in the southern mountains in Yemen, coast of Gulf of Oman, and in Kuwait. The Hejaz mountains along the Red Sea coast alos have long mean episode lengths, but the adjoining coastal plains has low values. The region between Medina and Riyadh has long episode lengths on the windward side of the mountains and shorter ones on the lee side. SWMERRA slightly overestimates the generation episode lengths in open seas. HRS-1 and SWMERRA agree over near coastal waters, generally. There are diﬀer- ences over the three gulfs of the Red Sea (Aqaba, Suez, and Aden) an SWMERRA underestimates in these regions. Generally, there is disagreement when there is topography. 129

Figure A.22: Average episode length in wind power generation at 100m spatial distribution by 10 classes from the WRF model at 5km resolution 130

Figure A.23: Normalized diﬀerences between the WRF model (5km) and MERRA (50km), where the red indicates MERRA underestimation 131

Figure A.24: Percentage diﬀerence in average episode length in power generation at 80m from 100m in WRF 132

Figure A.25: Percentage diﬀerence in average episode length in power generation at 140m from 100m in WRF 133 Impact of the hub height As the hub height is lowered from 100m to 80m, the mean generation episode length decreases over land. Over waters close to the coast, the decrease is slow. The episode length decreases with height over the gulf regions of the Red Sea and also in the Tokar gap. Raising the hub height impacts oppositely. 134 Description of Datasets

The dissertation utilized data from diﬀerent sources. The following paragraphs describe the data used.

MERRA near-surface data

This data has been downloaded from the NASA website. The features of this dataset are:

• Surface variables to reconstruct wind power density within the boundary layer at 3 altitudes: 80 m, 100 m, and 140 m.

• Spatial resolution: 50 km × 67 km

• Temporal resolution: hourly

• Stored in a directory with raw data (daily ﬁles), and time-merged ﬁle, which was used for this dissertation.

• Format: NetCDF

MERRA-2 upper air data

This dataset has been used to create the airborne wind resource. The details of this dataset are:

• Variables at 13 pressure levers from the ground to approximately 3 km above the ground.

• Spatial resolution: 50 km × 67 km

• Temporal resolution: hourly

• Stored in a directory with raw data (daily ﬁles), and time-merged ﬁle, which was used for this dissertation. 135 • Format: NetCDF

High resolution simulation datasets

HRS 1 dataset This dataset has been used to study the high resolution wind resource for this dissertation. Its main features are:

• Fields at 35 vertical levels

• Spatial resolution: 5 km × 5 km

• Temporal resolution: hourly

• Each year of simulation is stored in one ﬁle

• Format: NetCDF

HRS 2 dataset This dataset is being provided to further understand the wind resource at very high resolution. Its main features are:

• The ﬁelds are available at 40 vertical levels

Domain 1– Spatial resolution: 9 km × 9 km

– Temporal resolution: daily output in daily ﬁles

Domain 2– Spatial resolution: 3 km × 3 km

– Temporal resolution: 30 minute output in daily ﬁles

Domain 3– Spatial resolution: 1 km × 1 km

– Temporal resolution: 10 minute output in daily ﬁles

• All this data is stored in multiple directories, each directory containing a month worth of data for the three domains in NetCDF 136 Other data

Other data that has been used in this dissertation include the following.

• Weather observations from surface weather stations (hourly measurements)

• Radiosonde observations of upper air (twice daily)

• ECMWF operational analysis dataset used for initializing the WRF model to generate the HRS 1 and HRS 2 datasets. 137 WRF Model Conﬁguration

Advanced Research Weather Research and Forecasting Model (ARW) v3.5.1 Simulation period: 2009 - 2014 Domain size: 770 x 660 Domain center: 20N, 47.5E Horizontal resolution: 5km x 5km Mapping: cylindrical equidistant Vertical resolution: 35 model levels Boundary conditions: ECMWF operational analysis Output frequency: hourly Physics options Microphysics: Lin et al. scheme (2) Longwave radiation: RRTMG (4) Shortwave radiation: RRTMG (4) Surface layer: Monin-Obukhov (Janjic Eta) (2) Boundary layer: Mellor-Yamada-Janjic (Eta) TKE (2) Cumulus: Kain-Fritsch (new Eta) (1) Land-surface model: Uniﬁed Noah LSM (2) Sea-surface temperature updates: true (1) 138 Data Processing Scripts

This appendix shows the steps to be followed to generate the two following ﬁelds:

PowerGeneratedTurbine Power generated at the surface using a turbine.

PowerGeneratedKite Airborne power generated using a kite.

The convention of dimensions followed derives from the MERRA dataset. The dimensional information is shown below. double lon(lon) ; lon:standard_name = "longitude" ; lon:long_name = "longitude" ; lon:units = "degrees_east" ; lon:axis = "X" ; double lat(lat) ; lat:standard_name = "latitude" ; lat:long_name = "latitude" ; lat:units = "degrees_north" ; lat:axis = "Y" ; double time(time) ; time:standard_name = "time" ; time:long_name = "time" ; time:units = "day as %Y%m%d.%f" ; time:calendar = "standard" ; time:axis = "T" ;

For processing MERRA ﬁles: 139 cdo -f nc -a monmean -expr,"_u=sqrt(U^2+V^2);p = _u < 2 || _u > 30 ? 0

,→ : -312.181372549021+ 96.3736240110083*_u+ 13.1364809081527*_u^2+

,→ -0.519908840729273*_u^3; p = _u >= 18 && _u <= 30 ? 2675 : p"

,→ merra_uv.nc4 merra_monmean_p.nc

From already computed u at vertical levels cdo -f nc -a monmean -expr,"p = u < 2 || u > 30 ? 0 :

,→ -312.181372549021+ 96.3736240110083*u+ 13.1364809081527*u^2+

,→ -0.519908840729273*u^3; p = u >= 18 && u <= 30 ? 2675 : p"

,→ -vertmax u.nc4 merra_monmean_p.nc

For processing WRF ﬁles: cdo -a setgridtype,lonlat -monmean -expr,"_u=sqrt(U^2+V^2);p = _u < 2

,→ || _u > 30 ? 0 : -312.181372549021+ 96.3736240110083*_u^1+

,→ 13.1364809081527*_u^2+ -0.519908840729273*_u^3; p = _u >= 18 && _u

,→ <= 30 ? 2675 : p" wrf_uv.nc wrf_monmean_p.nc

To remove times bnds artifacts using NCO and CDO: ncks -C -x -v time_bnds merra_monmean_pmax.nc kite_int.nc cdo selname,p kite_int.nc merra_monmean_p_clean.nc

To rename variable and dimension using NCO: ncatted -O -a units,p,c,c,"kW" merra_monmean_p_clean.nc ncatted -O -a long_name,p,c,c,"Power at Wind Speed Maxima produced by

,→ a 3MW kite" merra_monmean_p_clean.nc ncatted -O -a units,pmax,c,c,"kW" test_monmean_kite_pmax_a_2014.nc ncatted -O -a long_name,pmax,c,c,"Power at Wind Speed Maxima produced

,→ by a 3MW kite" test_monmean_kite_pmax_a_2014.nc 140 where ”-a” is followed by ”attribute name, variable name, mode (append, create, delete, modify, overwrite), attribute variable type (ﬂoat, character, ...), attribute value” ncrename -d Times,time -v pmax,PowerGeneratedKite -v Times,time

,→ test_monmean_kite_pmax_a_2014.nc

,→ monmean_PowerGeneratedKite_2014.nc cdo setgridtype,lonlat monmean_PowerGeneratedKite_2014.nc

,→ monmean_PowerGeneratedKite_lonlat_2014.nc To compute turbine power, monthly mean power generated computation in nc with absolute time axis using CDO: cdo -f nc -a monmean

,→ -expr,"_u100=(USTAR/0.41)*log((100-DISPH)/Z0M);p100 = _u100 < 3 ||

,→ _u100 > 20 ? 0 : 1529.43356643357*_u100^0+

,→ -926.653846153847*_u100^1+ 163.502331002331*_u100^2+

,→ -6.64568764568765*_u100^3; p100 = _u100 >= 13 && _u100 <= 20 ?

,→ 2500 : p100" tavg1_cat.nc4 merra_monmean_p100.nc cdo -a setgridtype,lonlat -monmean

,→ -expr,"_u100=sqrt(U10^2+V10^2)*(10^(1/7));p100 = _u100 < 3 ||

,→ _u100 > 20 ? 0 : 1529.43356643357 -926.653846153847*_u100 +

,→ 163.502331002331*_u100^2+ -6.64568764568765*_u100^3; p100 = _u100

,→ >= 13 && _u100 <= 20 ? 2500 : p100" wrf_u10v10.nc

,→ wrf_monmean_p100.nc To remove time bnds artifacts: ncks -C -x -v time_bnds merra_monmean_p100.nc turbine_int.nc cdo selname,p100 turbine_int.nc merra_monmean_p100_clean.nc 141 To rename variable using NCO: ncatted -O -a units,p100,c,c,"kW" merra_monmean_p100_clean.nc ncatted -O -a long_name,p100,c,c,"Power at hub height 100m produced by

,→ a 2.5MW turbine" merra_monmean_p100_clean.nc ncatted -O -a units,p100,c,c,"kW" test_monmean_turbine_p100_a_2014.nc ncatted -O -a long_name,p100,c,c,"Power at hub height 100m produced by

,→ a 2.5MW turbine" test_monmean_turbine_p100_a_2014.nc

where ”-a” is followed by ”attribute name, variable name, mode (append, create, delete, modify, overwrite), attribute variable type (ﬂoat, character, ...), attribute value” ncrename -d Times,time -v p100,PowerGeneratedTurbine -v Times,time

,→ test_monmean_turbine_p100_a_2014.nc

,→ monmean_PowerGeneratedTurbine_2014.nc cdo setgridtype,lonlat monmean_PowerGeneratedTurbine_2014.nc

,→ monmean_PowerGeneratedTurbine_lonlat_2014.nc