Preliminary Greenhouse Gas (GHG) Emissions Analysis of Four Gates Sanitation Systems Emissions during Steady-State Operation

By John T. Trimmer, Diana M. Byrne, Hannah A.C. Lohman, & Jeremy S. Guest

University of Illinois at Urbana-Champaign

Prepared for: Duke Internal Subaward #: TO 283-1325 Brian Stoner, Ph.D. Center for WaSH-AID [email protected]

Primary Author Contact Information: Date Submitted: 12/20/2018 Jeremy S. Guest, Ph.D. Assistant Professor Department of Civil and Environmental Engineering University of Illinois at Urbana-Champaign 3221 Newmark Civil Engineering Laboratory, MC-250 205 North Mathews Avenue Urbana, IL 61801-2352 [email protected]

BMGF OPP: OPP1173370 Project #: 13C ______This work is, in part, supported by a grant, OPP1173370, from the Bill & Melinda Gates Foundation through Duke University’s Center for WaSH-AID. All opinions, findings, and conclusions or recommendations expressed in this work are those of the author(s) and do not necessarily reflect the views of the Foundation, Duke, or the Center. Statement of Confidentiality This report and supporting materials contain confidential and proprietary information. These materials may be printed or photocopied for use in evaluating the project, but are not to be shared with other parties without permission. Jeremy S. Guest, Ph.D. 2 Assistant Professor Department of Civil and Environmental Engineering University of Illinois at Urbana-Champaign 3221 Newmark Civil Engineering Laboratory, MC-250 205 North Mathews Avenue Urbana, IL 61801-2352 Preliminary Greenhouse Gas (GHG) Emissions Analysis of Four Gates Sanitation Systems Emissions during Steady-State Operation John T. Trimmer, Diana M. Byrne, Hannah A.C. Lohman, Jeremy S. Guest

Executive Summary This analysis estimates greenhouse gas (GHG) emissions associated with steady-state operation of four technologies in the Bill & Melinda Gates Foundation’s sanitation portfolio: the HTClean system (Helbling), the Empower Sanitation Platform (Duke Centre for WaSH-AID), the Electrochemical Reinvented Toilet (Eco-San), and the Janicki Omni-Processor (Sedron Technologies). Specifically, we estimated emissions from (i) the degradation of bodily waste during containment, treatment, and recovery; (ii) electricity and materials consumed during operation; and (iii) transportation of waste to its treatment and/or disposal site. We also estimated GHG offsets from recovery of fertilizer nutrients (nitrogen, phosphorus, potassium). This steady- state analysis does not include the emissions associated with the construction, maintenance, and end-of-life of these technologies. All estimates are expressed as equivalent kilograms of carbon -1 -1 dioxide per year, normalized to the estimated population served (i.e., kg CO2 eq·cap ·year ). With the exception of the HTClean system, direct emissions from the current portfolio technologies have the potential to compare favorably against pit latrines. Our results suggest that electricity demand tends to drive emissions trends across three of the four systems (excluding the Omni-Processor). HTClean is associated with the highest total GHG emissions per person -1 -1 (780-1,800 kg CO2 eq·cap ·yr ), with 90% of emissions coming from its large electricity demand. As the Omni-Processor is reported to require no outside electricity, it is associated with the lowest -1 -1 emissions (33-64 kg CO2 eq·cap ·yr for the full system including latrine containment and passive pretreatment). Latrine containment and pretreatment contribute most of the emissions related to -1 -1 the Omni-Processor system. The Empower (100-180 kg CO2 eq·cap ·yr ) and Eco-San (250- -1 -1 470 kg CO2 eq·cap ·yr ) systems produce intermediate levels of emissions (compared with the HTClean and Omni-Processor systems), with the Eco-San total being larger due to its higher electricity demand. The Empower system offers two alternatives for final processing of dried solids (combustion or land application), but total emissions are similar for both options (as land-applied solids may continue to degrade). The Empower estimates are similar to preliminary values associated with degradation of bodily waste in pit latrines (as calculated by the University of Leeds CACTUS team in ongoing work; final values may differ slightly). In all full systems, recovered nutrients contribute offsets that are relatively small compared with total system emissions. Broadly, these preliminary results are associated with large uncertainties as the analysis required numerous assumptions. Results tend to be most sensitive to the emissions factor for electricity -1 (i.e., kg CO2 eq·kWh ) and multiple parameters related to direct emissions from bodily waste (e.g., carbon and nitrogen excretion, emissions during combustion). Electricity emissions will vary depending on the local source of electricity (e.g., coal, hydroelectric, solar), resulting in significant variability across implementation locations. This variability can be reduced if a power source is built into the system itself, and we show that integrating renewable energy sources (e.g., photovoltaics) may provide the greatest opportunity for reductions in total GHG emissions. Direct emissions vary based on storage and treatment conditions of bodily waste and will also depend upon implementation location, as local diets will affect excretion of carbon and nitrogen into sanitation systems. Moving forward, we hope to expand these analyses in collaboration with the design and modeling teams and develop a full life cycle assessment (LCA) of each system.

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Introduction This report presents the preliminary findings from an analysis of greenhouse gas (GHG) emissions associated with four technologies in the Bill & Melinda Gates Foundation’s sanitation portfolio. The four technologies include the HTClean system (Helbling), the Empower Sanitation Platform (Duke Centre for WaSH-AID), the Electrochemical Reinvented Toilet (Eco-San), and the Janicki Omni-Processor S250 (Sedron Technologies), providing an array of systems ranging from a decentralized, single-household toilet up to a centralized, community-scale treatment facility. For this preliminary report, we present estimates of steady-state, operation phase emissions for each system. Three general categories of emissions are defined: • Direct emissions from degradation of bodily waste (biogenic methane and N2O given off during containment, treatment, and recovery; biogenic carbon dioxide is not included); • Technology operation (emissions associated with electricity and consumables needed for system functioning; replacement parts are not included); and • Transportation (truck conveyance of latrine sludge to centralized treatment facilities or recovered products to locations for land application). We also estimated offsets associated with nutrients (nitrogen, phosphorus, potassium) recovered in the products from each system. These products could replace conventional fertilizers and the emissions associated with their production. All emissions and offset estimates were converted to equivalent kilograms of carbon dioxide per year and were normalized according to the population -1 -1 served by each system (kg CO2 eq·cap ·yr ). In the future, we hope these analyses can be expanded to further inform decision-making through a full life cycle assessment of each system.

Results and Discussion Estimates of greenhouse gas emissions during steady-state operation. To account for the uncertainties associated with our analysis, we present expected emissions estimates (calculated using single likely values of all assumed parameters) along with the results of an uncertainty analysis (in which we varied parameter values across 10,000 simulations; see Methods for more detail). All reported ranges represent the 5th and 95th percentile output values from this analysis. Generally, electricity demand tends to drive comparative performance concerning GHG emissions per person per year (Figure 1). The HTClean system is associated with the highest -1 -1 total emissions (780-1,800 kg CO2 eq·cap ·yr ), and 90% of emissions come from its large electricity demand. The Helbling Technik team (via Dr. Christian Seiler) estimates this system currently requires an average of 650 watts of continuous power. The average power draw of the next generation model may drop below 400 watts, representing a potential 40% reduction in energy-related emissions, but its total emissions would still be much larger than those of the other three systems. In the HTClean system, direct emissions during the high temperature/pressure treatment process are particularly uncertain, due to a lack of experimental data (we used estimates related to dehydration-stage pyrolysis1 at <200°C for methane and gaseous losses of 2 protein in hydrothermal liquefaction for N2O, placing large uncertainty ranges of ±100% around these values). However, the dominance of electricity emissions in this system reduces the importance of these tentative direct emissions assumptions. In contrast, the Omni-Processor is reported to require no outside electricity and is associated with the lowest emissions. We investigated three cases for the Omni-Processor: (i) emissions from the processor itself; (ii) emissions from the processor and a passive pretreatment approach that dewaters input sludge using unplanted drying beds and treats the resulting liquid effluent using anaerobic and facultative lagoons; and (iii) emissions from the processor, passive pretreatment, and latrine containment (including direct emissions due to degradation in latrines prior to collection and transport emissions to move collected sludge to a centralized treatment facility). Based on

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

data provided by the Gates 2000 Foundation (via Dr. John Duffy), the Direct emissions processor itself (i.e., on its own) is Technology operation estimated to generate low emissions. Transportation Most (>90%) of its total comes from 1000 direct emissions released during drying and combustion, while the ) contribution from consumables -1 400 yr needed for system startup (diesel, -1 natural gas) and emissions control 300

(hydrated lime) is minimal. Beyond eq cap the processor, infrastructure and 2 operational needs for sludge pretreatment (to raise solids content CO (kg 200

of input sludge and treat the resulting Greenhouse gas emissions liquid) may generate substantial Pit latrines emissions, but only direct fugitive 100 emissions (CH4 and N2O) from pretreatment were included at this time (based on data availability). 0 Without pretreatment to dry the incoming sludge, the Omni-Processor HTClean Eco-San would need to use auxiliary fuel sources, which would increase technology operation emissions. EmpowerEmpower (with combustion) (land application) Omni-Processor (processor only) Within the fully integrated system Omni-Processor (with pretreatment) (latrines, pretreatment, Omni- Omni-Processor (with pretreatment & latrines) Processor), emissions from latrine Figure 1. Estimated annual greenhouse gas (GHG) emissions per storage contributed most (58% in the person served during steady-state operation of each system. Each expected case) of the total GHG bar represents the expected emissions for a given system, -1 -1 calculated using likely values of all input parameters. Bars are emissions (33-64 kg CO2 eq·cap ·yr colored to show the portion of the total coming from each emissions in total). However, we assumed that category (direct emissions, technology operation, transportation). latrines were well-managed and Transportation emissions are too small to be visible, as are frequently emptied. Under less emissions from consumables (related to technology operation) used favorable assumptions3, latrine in the Omni-Processor. Fertilizer offsets are not considered in this emissions become even more figure and tend to be minor relative to total emissions. Error bars represent 5th and 95th percentile values from the uncertainty important. In the Omni-Processor analysis. This analysis provides a probable range of total system itself, thermal drying of pretreated emissions, generated by a set of 10,000 simulations pulling input sludge results in minor direct values from each uncertain parameter’s distribution. The graph’s 4 gray shaded area (49-128 kg CO eq·cap-1·yr-1) represents the emissions , and while N2O may be 2 5 range of preliminary emissions estimates from pit latrines (as released during combustion , calculated by the University of Leeds CACTUS team in ongoing methane emissions are expected to work; final values may differ slightly). Note the break and scale -1 -1 be minimal in this process. change in the y-axis at 400 kg CO2 eq·cap ·yr . The remaining two systems are estimated to produce relatively intermediate levels of emissions, -1 -1 with the Eco-San system’s total (250-470 kg CO2 eq·cap ·yr ) being the larger of the two because of its higher electricity demand (data provided by Chris Buckley and Siphesihle Khumalo of the University of KwaZulu-Natal). We expect that Eco-San’s more intensive energy needs are due to this system’s use of electrochemical (EC) oxidation to degrade organic matter from a mixed waste stream. In contrast, Empower’s EC cell focuses only on disinfection of a separated liquid stream (requiring less energy), with most organics retained in the solid stream. While EC oxidation of

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

organic matter is expected to remove organic carbon without emitting methane (i.e., carbon is removed as CO2), the Eco-San system’s septic tank and lack of solids separation lead to predominantly anaerobic conditions where a large quantity of methane may be produced. -1 -1 Total emissions for the Empower Sanitation Platform (100-180 kg CO2 eq·cap ·yr ) include results from two alternative solids management approaches, in which dried solids are either incinerated or land applied (information provided by Dr. Brian Hawkins of Duke University). Both options produce similar emissions, as land-applied solids may continue to degrade on fields (balancing the N2O emissions of combustion). Generally, emissions from separated solids are relatively small, as solids processing steps are similar to those in the Omni-Processor (thermal drying and possibly combustion). More direct emissions originate from the liquid processing stream, where anaerobic conditions are likely to predominate. As most of the liquid is recycled as flush water, we assumed most of the carbon and nitrogen in the liquid stream remain in the system and degrade over time. As a point of comparison, total emissions from the Empower system and the fully integrated Omni- Processor system appear to be similar to preliminary values associated with pit latrines (as calculated by the University of Leeds CACTUS team in ongoing work led by Prof. Barbara Evans; final values may differ slightly). Generally, with the exception of the HTClean system, direct emissions from the portfolio technologies have the potential to compare favorably, with modest improvements bringing them to the lower end of the range estimated for pit latrines. The HTClean system’s direct emissions are estimated to be larger, but a high degree of uncertainty surrounds these estimates (as stated previously). Across all systems, transport contributed only a minor fraction of total emissions. In most cases, treatment processes generate relatively small masses of products such as dried solids or ash. Transport was most important for the fully integrated Omni-Processor system, as wet latrine sludge must be hauled to a centralized facility. Even in this case, however, transport emissions account for only 1% of the overall total, and drastically increasing our assumption of average transport distance to 100 km (from 2-10 km) only brings transport’s contribution up to 10%. Fertilizer offsets were also relatively inconsequential, representing 0-3% of total emissions across all full systems. While nutrient recovery can generate potentially valuable and marketable resources for local users or regional farmers, it is unlikely to substantially reduce a system’s net GHG emissions. The Omni-Processor, when considered in isolation (i.e., without emissions from pretreatment or latrine containment), is one exception to this pattern. Given our current estimates, nutrients in the processor’s ash may offset most of its direct emissions. However, evaluating the Omni-Processor by itself neglects potentially considerable emissions from pretreatment and latrine containment (both of which are needed for the processor to function and should be treated as integrated components of the sanitation value chain). Finally, we note that total emissions estimates are highest in the most decentralized system (HTClean) and lowest in the most centralized alternative (Omni-Processor). However, the energy requirements of each system’s treatment approach may be more responsible for this trend than system scale (i.e., centralized vs. decentralized). Treating bodily waste under high temperature/pressure conditions (HTClean) is inherently more energy-intensive than using energy embedded in the waste itself to fuel combustion (Omni-Processor, Empower); of course, low energy dewatering in these latter systems underpins this assertion. Key sources of uncertainty. Given the uncertainty associated with these estimates, it is valuable to explore which input parameters are most highly correlated with the uncertainty of the outputs. Our analysis included over 70 uncertain parameters (each characterized based on information from the design and modeling teams, relevant literature, or our own experience). For each uncertain parameter, we calculated Spearman’s rank correlation coefficients with the total

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

emissions from each system as a form of sensitivity analysis. Spearman’s coefficients estimate the degree to which a system output is correlated with one parameter’s value (a coefficient with a larger absolute value indicates a stronger monotonic correlation), enabling us to identify parameters that are most critical to the uncertainty surrounding results from individual systems or across multiple systems (Figure 2). The strongest correlations may point toward key areas where more information is needed, or it may suggest how variations in certain contextually-dependent factors (e.g., unit GHG emissions from local grid electricity) can impact system performance.

Spearman’s rank correlation coefficient 0.1 0.5 1.0

-1 Electricity emissions (kg CO2eq·kWh ) HTClean: Required electricity (W) HTClean: Users per day Empower: Required electricity (kWh·cap-1·d-1) Flush frequency (flushes·cap-1·d-1) COD (g·cap-1·d-1) Particulate COD (% of total) N (g·cap-1·d-1) N in urine (%) -1 Maximum direct methane emission (g CH4·g COD ) MCF: Land application MCF: Well-managed pit latrine

Combustion: N2O emissions (% of N) Time to full degradation (years) Latrine sludge: N leached from latrine (%) Latrine sludge: Emptying period (years) Reduction representing full degradation (%)

Figure 2. The relative importance of key parameters to the uncertainty associated with each system’s total emissions. The size of each bubble is proportional to the absolute value of the Spearman’s rank correlation coefficient associated with each parameter in each system. Spearman’s coefficients estimate the degree to which variations in a system output (in this case, total emissions) have a monotonic relationship with variations in a specific parameter’s value. Each parameter shown in this figure had a coefficient with an absolute value of at least 0.20 in at least one system. Parameters are divided into two groups: those related to electricity emissions (top, orange shading) and those related to direct emissions from the degradation of waste (bottom, blue shading). Intersections with an “X” signify parameters that do not apply to the given system. Broadly, our estimates tend to be most sensitive to the electricity emissions factor and multiple parameters related to direct emissions from the degradation of bodily waste. For all systems that -1 require electricity, the electricity emissions factor (i.e., kg CO2 eq·kWh ) was particularly impactful. Essentially, this factor is based on the mix of energy sources used to produce local electricity. If more polluting sources (e.g., fossil fuels) make up a larger portion of the overall mix, the emissions factor will be higher. As such, the electricity mix in a given local setting could have a considerable impact on a system’s actual GHG emissions. The effects of variations in electricity mix from location to location are examined with a scenario analysis in the following section. Estimates are also sensitive to a number of parameters related to direct emissions from bodily waste, with the most critical factors varying from system to system. In the HTClean and Eco-San systems, however, the impacts from these parameters are relatively minor, because electricity emissions considerably outweigh direct emissions. In the Empower system, excretion of chemical oxygen demand (COD, an aggregate measure of organic content proportional to total organic

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

carbon in a given waste stream6) and the factor representing the maximum potential methane emission from the degradation of a given mass of COD were most critical to direct emissions. COD excretion is related to caloric intake and will vary depending upon typical diets in a given setting. Country-specific COD excretion estimates (based on calorie consumption)7 can be employed to generate more precise estimates for a given implementation context. Emissions from the Omni-Processor alone are most sensitive to the fraction of nitrogen released as N2O during combustion, as combustion is the primary source of emissions from the processor itself. When latrine containment and pretreatment are integrated into this system, a number of other parameters, such as COD and nitrogen excretion and maximum potential methane emissions, become much more important. The impacts of variations in COD and nitrogen excretion across countries are evaluated in the following section’s scenario analysis. Additionally, assumptions about environmental conditions within latrines (represented by the methane correction factor, MCF; see Methods for more detail) and how often latrines are emptied will affect the degree to which methane and N2O are released before latrine sludge reaches the treatment facility. Scenario analysis. Based on our uncertainty and sensitivity analyses, a key parameter for systems using electricity is the electricity emission factor. Essentially, this factor relates to the mix of energy sources used to produce electricity in a given context and is higher when fossil fuels account for a larger fraction. The emission factor can vary considerably from place to place. Similarly, excretion rates of COD and nutrients (particularly nitrogen) vary based on calorie and protein intake and affect the quantities of methane and N2O that may be released from bodily waste. Therefore, we performed a scenario analysis to assess how context-specific diet and electricity characteristics may impact emissions in various locations of interest. We included three country-specific scenarios (Uganda, South Africa, India), in which we used per capita COD and nutrient excretion rates estimated based on calorie and protein availability in each country7, as well as statistics on each country’s electricity mix8 to calculate the electricity emissions factor. A fourth scenario focused solely on electricity source and examined the emissions reductions that could result from using solar panels to provide all required electricity. For all scenarios, any altered assumptions were assigned context-specific uncertainty distributions, and we ran the uncertainty analysis in each case to estimate likely ranges for scenario-specific results. If we assume the three electricity-using technologies (HTClean, Empower, Eco-San) draw power from the grid, then their total emissions are highly sensitive to the local mix of electricity sources (Figure 3; note the panels have differing y-axis scales). Total emissions for systems installed in South Africa and India are somewhat similar to the ranges estimated in our general analysis (which employed a generalized electricity emissions factor for Africa, with a wide uncertainty distribution to account for other contexts). These countries tend to be largely reliant on fossil fuels (>80% in India, >90% in South Africa). Although a larger fraction of South Africa’s electricity mix comes from fossil fuels, India’s electricity emissions per kilowatt-hour are larger overall because fossil fuel use for electricity production is estimated to be more polluting in India9. In contrast, much of Uganda’s electricity generation (>90%) comes from hydroelectric power. While technologies more reliant on electricity may be associated with substantially lower emissions in Uganda, toilets such as the HTClean system would still require a substantial amount of energy, potentially limiting feasibility in a country where only 27% of the population in 2016 was estimated to have access to electricity10. Country-specific differences in COD and nutrient excretion influence direct emissions associated with all four systems, although changes are most noticeable in the Omni-Processor (since variations associated with electricity mix do not impact this system). As typical diets in Uganda, South Africa, and India include less caloric and protein content than the global average (in our general analysis, we used estimated average global excretion values, consistent with the current

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

methodology of the ongoing University of Leeds CACTUS project), people in those countries excrete less COD and nitrogen. These smaller excretion rates contribute to lower estimates of direct methane and N2O emissions. For the three technologies that rely on electricity, use of renewable energy sources may provide the greatest opportunities for emissions reductions. Incorporating solar panels into the system itself could lessen the variability in emissions from diverse mixes of electricity from country to country and drastically reduce total emissions. For example, integrating solar panels into the HTClean system could reduce total emissions by >80% compared with expected values in the general analysis.

Figure 3. Findings from a scenario analysis examining country context and electricity source. For each of the four systems, total emissions estimates from the expected case and uncertainty analysis (5th and 95th percentile values) are shown under different assumptions regarding diet and electricity supply. In each plot, that system’s results from the general analysis (Figure 1) are shown first, followed by country-specific results for Uganda, South Africa, and India. Note each plot uses a different y-axis scale. In each scenario, per capita COD and nutrient excretion rates (affecting how much methane and N2O may be released during degradation of bodily waste) were varied depending on typical 7 -1 calorie and protein intake within the specified country , while the electricity emissions factor (i.e., kg CO2 eq·kWh ) was varied based on the country’s reported mix of electricity sources8. The final scenario assumes that all electricity is provided using solar panels, while assumptions regarding dietary intake revert to those used in the general analysis. Gray boxes signify the range of preliminary pit latrine estimates for each scenario, calculated using current University of Leeds CACTUS project methodology with country-specific COD and nitrogen excretion estimates as inputs.

Conclusions This analysis estimated GHG emissions associated with steady-state operation of four technologies in the Bill & Melinda Gates Foundation’s sanitation portfolio. Given the specified scope of the analysis, we did not include the emissions associated with the construction and end- of-life of these technologies, nor did we include replacement parts. Broadly, results suggest that energy demand tends to drive emissions differences across the four systems. The HTClean system was associated with the highest total emissions per person served (780-1,800 kg CO2 eq·cap-1·yr-1), due to its large electricity demand relative to the other systems. As the Omni- Processor is reported to require no outside electricity, it is associated with the lowest emissions

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

-1 -1 (33-64 kg CO2 eq·cap ·yr for the full system). For this system, we included latrine containment and estimates of passive (i.e., no energy) pretreatment before sludge enters the processor, as these are integrated components of the sanitation value chain and contribute large fractions of -1 -1 the system’s total emissions. The Empower (100-180 kg CO2 eq·cap ·yr ) and Eco-San (250- -1 -1 470 kg CO2 eq·cap ·yr ) systems produce intermediate levels of emissions, with the Eco-San system requiring more electricity, likely due to its use of electrochemical oxidation to reduce organic matter (whereas the Empower system uses this process only for disinfection). As a point of comparison, the Empower and Omni-Processor systems’ total emissions appear to be comparable with the range of preliminary estimates for direct emissions from pit latrines (as calculated by the University of Leeds CACTUS team in ongoing work; final values may differ slightly). In all full systems, while recovered nutrients may represent valuable products, the GHG offsets they contribute are relatively small compared with total system emissions. Generally, with the exception of the HTClean system, direct emissions from the current portfolio technologies have the potential to compare favorably against pit latrines. Large uncertainties are associated with these preliminary results, and results are most sensitive to emissions factors for electricity and multiple factors related to direct emissions from bodily waste. Emissions will vary depending on local sources of electricity (e.g., coal, hydroelectric, solar) and local diets (which impact excretion of carbon and nitrogen, and thus direct emissions of methane and N2O). As such, implementing these systems in different locations may cause considerable disparities in GHG emissions from place to place, unless a power source is built into the system itself (e.g., solar panels may provide the greatest opportunity for emissions reductions in systems requiring electricity). In the future, we hope to expand this work in collaboration with the design and modeling teams, moving toward a full life cycle assessment of each system.

References (including additional references from Methods and Appendices) 1. Shao, J. et al. Pyrolysis Characteristics and Kinetics of Sewage Sludge by Thermogravimetry Fourier Transform Infrared Analysis. Energy Fuels 22, 38–45 (2008). 2. Li, Y. et al. Quantitative multiphase model for hydrothermal liquefaction of algal biomass. Green Chem. 19, 1163–1174 (2017). 3. IPCC. 2006 IPCC Guidelines for National Greenhouse Gas Inventories. (Prepared by the National Greenhouse Gas Inventories Programme, Eggleston H.S., Buendia L., Miwa K., Ngara T. and Tanabe K. (eds.), 2006). 4. Deng, W.-Y. et al. Emission characteristics of volatile compounds during sludges drying process. J. Hazard. Mater. 162, 186–192 (2009). 5. Barton, P. K. & Atwater, J. W. Nitrous Oxide Emissions and the Anthropogenic Nitrogen in Wastewater and Solid Waste. J. Environ. Eng. 128, 137–150 (2002). 6. Tchobanoglous, G. et al. Wastewater Engineering: Treatment and Resource Recovery. (Metcalf & Eddy, Inc., AECOM, McGraw-Hill, 2014). 7. Trimmer, J. T., Cusick, R. D. & Guest, J. S. Amplifying Progress toward Multiple Development Goals through Resource Recovery from Sanitation. Environ. Sci. Technol. 51, 10765–10776 (2017). 8. EIA. International energy statistics. U.S. Energy Information Administration (2018). Available at: https://www.eia.gov/beta/international/. (Accessed: 8th December 2018) 9. ecoinvent 3.2 database. (2016). 10. World Bank. DataBank. The World Bank Group (2018). Available at: http://databank.worldbank.org/data/home.aspx. (Accessed: 24th February 2018) 11. Bare, J. TRACI 2.0: the tool for the reduction and assessment of chemical and other environmental impacts 2.0. Clean Technol. Environ. Policy 13, 687–696 (2011).

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

12. Myhre, G. et al. Anthropogenic and Natural Radiative Forcing. in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Stocker, T. F. et al.) (Cambridge University Press, 2013). 13. Rose, C., Parker, A., Jefferson, B. & Cartmell, E. The Characterization of Feces and Urine: A Review of the Literature to Inform Advanced Treatment Technology. Crit. Rev. Environ. Sci. Technol. 45, 1827–1879 (2015). 14. Noyan, K., Allı, B., Taş, D. O., Sözen, S. & Orhon, D. Relationship between COD particle size distribution, COD fractionation and biodegradation characteristics in domestic sewage. J. Chem. Technol. Biotechnol. 92, 2142–2149 (2017). 15. Friedler, E., Butler, D. & Alfiya, Y. Wastewater composition. in Source Separation and Decentralization for Wastewater Management (eds. Larsen, T. A., Udert, K. M. & Lienert, J.) 241–258 (IWA Publishing, 2013). 16. Muff, J. Chapter 3 - Electrochemical Oxidation – A Versatile Technique for Aqueous Organic Contaminant Degradation. in Chemistry of Advanced Environmental Purification Processes of Water (ed. Søgaard, E. G.) 75–134 (Elsevier, 2014). doi:10.1016/B978-0-444-53178- 0.00003-1 17. Samolada, M. C. & Zabaniotou, A. A. Comparative assessment of municipal sewage sludge incineration, gasification and pyrolysis for a sustainable sludge-to-energy management in Greece. Waste Manag. 34, 411–420 (2014). 18. McKay, M. D., Beckman, R. J. & Conover, W. J. A Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of Output from a Computer Code. Technometrics 21, 239–245 (1979). 19. Hauke, J. & Kossowski, T. Comparison of values of Pearson’s and Spearman’s correlation coefficient on the same sets of data. Quaest. Geogr. 30, (2011). 20. Rogers, T. W. et al. A granular activated carbon/electrochemical hybrid system for onsite treatment and reuse of blackwater. Water Res. 144, 553–560 (2018). 21. Hawkins, B. T. et al. Remediation of suspended solids and turbidity by improved settling tank design in a small-scale, free-standing toilet system using recycled blackwater. Water Environ. J. 0, (2018). 22. Orner, K. D. & Mihelcic, J. R. A review of sanitation technologies to achieve multiple sustainable development goals that promote resource recovery. Environ. Sci. Water Res. Technol. 4, 16–32 (2018). 23. Tilley, E., Ulrich, L., Luthi, C., Reymond, P. & Zurbrugg, C. Compendium of Sanitation Systems and Technologies. (Swiss Federal Institute of Aquatic Science and Technology (Eawag), 2014). 24. Mayer, P. W. et al. Residential End Uses of Water. (AWWA Research Foundation and American Water Works Association, 1999). 25. Strande, L. et al. Methods to reliably estimate faecal sludge quantities and qualities for the design of treatment technologies and management solutions. J. Environ. Manage. 223, 898–907 (2018). 26. Jacks, G., Sefe, F., Carling, M., Hammar, M. & Letsamao, P. Tentative nitrogen budget for pit latrines – eastern Botswana. Environ. Geol. 38, 199–203 (1999). 27. Lagerstedt, E., Jacks, G. & Sefe, F. Nitrate in groundwater and N circulation in eastern Botswana. Environ. Geol. 23, 60–64 (1994). 28. Nyenje, P. M., Foppen, J. W., Kulabako, R., Muwanga, A. & Uhlenbrook, S. Nutrient pollution in shallow aquifers underlying pit latrines and domestic solid waste dumps in urban slums. J. Environ. Manage. 122, 15–24 (2013). 29. Phillips, I. & Burton, E. Nutrient Leaching in Undisturbed Cores of an Acidic Sandy Podosol Following Simultaneous Potassium Chloride and Di-Ammonium Phosphate Application. Nutr. Cycl. Agroecosystems 73, 1–14 (2005).

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

30. Mihelcic, J. R., Fry, L. M., Myre, E. A., Phillips, L. D. & Barkdoll, B. D. Field Guide to Environmental Engineering for Development Workers: Water, Sanitation, and Indoor Air. (American Society of Civil Engineers, 2009).

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Methods The objective of this preliminary analysis was to estimate steady-state GHG emissions during normal operation of four technologies in the Bill & Melinda Gates Foundation sanitation portfolio. The four sanitation technologies include the HTClean system (Helbling), the Empower Sanitation Platform (Duke Centre for WaSH-AID), the Electrochemical Reinvented Toilet (Eco-San), and the Janicki Omni-Processor (Sedron Technologies). The following sections describe the processes and assumptions used to estimate emissions from (i) degradation of bodily waste during containment, treatment, and recovery; (ii) electricity and materials consumed during operation; and (iii) transportation. GHG offsets from recovery of fertilizer nutrients (nitrogen, phosphorus, potassium) were also considered.

Data collection. This analysis and the assumptions it employed were based primarily on a combination of scientific literature, life cycle assessment software (the ecoinvent v3.2 database9 and the U.S. EPA’s Tool for the Reduction and Assessment of Chemicals and Other Environmental Impacts [TRACI 2.1 v1.03]11, accessed and implemented within SimaPro v8.5.2.0), a preliminary model of direct emissions from latrines (from the University of Leeds’ ongoing CACTUS project), and documentation and estimates from the technology design teams. Our primary source of information about the characteristics and treatment processes of each sanitation system was a series of two-page cut sheets (one for each system), and additional information was obtained through direct correspondence with the design teams. Transforming this understanding of each system into estimates of GHG emissions required numerous assumptions, which were based on a review of relevant literature and emissions factors from models related to life cycle environmental impacts and direct emissions from latrines. Specific assumptions related to each category of emissions (direct emissions from degradation of bodily waste, operational electricity and consumables, transportation) and fertilizer offsets will be described in the following paragraphs. To account for uncertainty and variability associated with assumed parameters, we performed uncertainty and sensitivity analyses, which are described in a subsequent section.

Greenhouse gas emissions estimates. The following subsections detail the procedures and assumptions used to estimate emissions within each of the three categories (direct emissions, technology operation, transportation) and offsets from nutrient recovery. Direct emissions. Direct emissions include biogenic (i.e., produced from biological organisms) methane and N2O released from the degradation of bodily waste during containment, treatment, or recovery processes (biogenic carbon dioxide emissions are considered to not contribute to climate change). By mass, methane from biogenic sources is estimated to contribute 34 times the climate change impact of non-biogenic carbon dioxide (34 kg CO2 eq per kg CH4), while N2O has 12 an impact 298 times that of carbon dioxide (298 kg CO2 eq per kg N2O) . Total CH4 and N2O emissions can be multiplied by these factors to represent all direct emissions as an equivalent mass of carbon dioxide. The factors are consistent with those currently used in the University of Leeds CACTUS project and include climate-carbon feedbacks, which reflect future changes in carbon storage due to changes in climate. While these feedbacks are likely to occur, their magnitude remains highly uncertain12. If we were to use factors that do not include climate-carbon feedbacks (28 kg CO2 eq per kg CH4; 265 kg CO2 eq per kg N2O), direct emissions estimates would be slightly reduced.

Methane and N2O released from bodily waste in sanitation systems depend on the environmental conditions present and the treatment processes being employed, and they are directly related to the quantities of carbon (assumed to be proportional to COD6) and nitrogen excreted in bodily waste. In the general analysis, we used recommended values of COD and nutrient excretion from the literature13, consistent with the values currently used by the University of Leeds team (Table

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

1; phosphorus and potassium excretion are also included to estimate fertilizer offsets, as described in a subsequent section). Subsequently, the scenario analysis (investigating specific country contexts) used excretion estimates from those countries based on our previous work7. Other characteristics of excreted waste, and partitioning between urine and feces, were based on literature data6,13–15. Table 1. Characteristics of excreted bodily waste employed in the general analysis, based on literature data6,13–15. Parameter Expected Value Minimum Maximum Distribution COD (g·cap-1·d-1) 71 46 96 Triangular Nitrogen (g·cap-1·d-1) 12.8 4.3 20.2 Triangular Phosphorus (g·cap-1·d-1) 2.06 0.75 3.37 Triangular Potassium (g·cap-1·d-1) 3.2 0.98 5.39 Triangular COD in feces (%) 81 69 90 Triangular Nitrogen in urine (%) 88 74 93 Triangular Phosphorus in urine (%) 61 33 75 Triangular Potassium in urine (%) 74 53 93 Triangular Total fecal mass (g·cap-1·d-1) 250 75 520 Triangular Feces moisture content (%) 85 76 88 Triangular Total urine volume (L·cap-1·d-1) 1.4 0.8 2.5 Triangular Carbon:COD ratio (g C·g COD-1) 0.35 0.25 0.4 Triangular Particulate COD (% of total) 66 60 72 Uniform

For any instances of passive storage (i.e., similar to latrine containment) within the four systems being studied, we made assumptions about likely storage conditions (e.g., mostly anaerobic, mostly aerobic, localized aerobic and anoxic zones, etc.) and matched these conditions with preliminary emissions factors from the University of Leeds CACTUS project and the Intergovernmental Panel on Climate Change (IPCC)3. To account for imperfect agreement between conditions in the Gates technologies and emissions factors from the CACTUS project, we chose multiple CACTUS factors judged to represent conditions similar to those in each Gates technology, defining an uncertainty distribution for each factor (Table 2). Consistent with the current CACTUS methodology, we assumed only 70% of COD entering passive storage was easily biodegradable. As such, we set aside 30% of initial COD and assumed it did not degrade during passive storage, while the remaining 70% can fully degrade over time. For other treatment processes, we assumed all COD could degrade and produce emissions (as literature on these processes often report emissions as a percentage of total influent carbon). For passive storage where the waste remains indefinitely, we followed the IPCC methodology3. In the case of methane, a given quantity of influent COD is associated with a maximum potential methane production (0.175-0.325 kg CH4 per kg COD). Based on environmental conditions, an appropriate methane correction factor (MCF, Table 2) is then applied to estimate actual methane emissions. For example, a MCF of 0.7 signifies conditions likely to produce 70% of the maximum potential methane emissions. For N2O, the emission factor is multiplied by the total influent nitrogen to estimate how much nitrogen leaves the system as N2O (e.g., a factor of 0.005 kg N2O- N per kg N suggests that 0.5% of nitrogen will be emitted as N2O).

Table 2. Methane and N2O emissions factors for various passive storage conditions (defined using preliminary factors from the University of Leeds CACTUS project and the IPCC guidelines3). Factor Expected Value Minimum Maximum Distribution Methane correction factors (fraction of maximum methane production capacity, 0.25 kg CH4 per kg COD) Mostly anaerobic conditions 0.7 0.6 1.0 Triangular Anaerobic deep lagoon/digester 0.8 0.8 1.0 Triangular Shallow (facultative) lagoon 0.2 0.0 0.3 Triangular Well-managed pit latrine 0.25 0.1 0.4 Triangular Land application 0.1 0.05 0.3 Triangular N2O emission factors (kg N2O-N per kg N) Mostly anaerobic conditions 0.005 0.0005 0.0065 Triangular Anaerobic deep lagoon/digester 0.005 0.0005 0.006 Triangular Well-managed pit latrine/shallow lagoon 0.008 0.0065 0.01 Triangular Land application 0.009 0.008 0.01 Triangular

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

If waste contained in a given passive storage environment does not remain indefinitely (e.g., latrine sludge is pumped out and taken for centralized treatment before all COD and nitrogen have degraded), we made several assumptions to estimate what fraction of complete degradation occurs during the given storage period. We assumed that each new waste deposit in a given storage location forms a layer that does not mix with previous deposits. Easily biodegradable -kt COD and nitrogen in each layer degrade according to first-order reaction kinetics (C = Coe ), assuming that full degradation occurs after 1-3 years and that “full” degradation can be represented by 99-99.99% degradation (as the first-order exponential decay function does not allow for complete 100% degradation). The first-order rate constant (k) was calculated based on these assumptions. When the collected waste is evacuated from storage, all layers are removed together, with each layer at a different point along the first-order decay curve. We assumed the COD and nitrogen concentrations in the evacuated waste can be represented as the average of the concentrations across all layers, calculated using the mean value theorem for integrals (i.e., integrating the first-order decay function from the start time to the ending time, and dividing by the total elapsed time). The differences between the initial concentrations and the averages calculated after storage were then used with the IPCC methodology (summarized in the preceding paragraph) to calculate methane and N2O emissions during the storage period. Beyond passive storage or containment processes, emissions from waste during specific treatment processes (e.g., thermal drying, combustion, high temperature/pressure treatment, electrochemical oxidation) were based on those reported in or calculated from relevant literature (Table 3). While carbon dioxide emissions continue to contribute no climate change impacts (as they are biogenic), we incorporate them to track all carbon losses throughout each system. Thermal drying processes are expected to release minimal quantities of methane and ammonia 4,5 (a fraction of which may transform to N2O in the atmosphere) . Combustion of bodily waste is assumed to release all remaining carbon as carbon dioxide (such that no methane is emitted), 5 but N2O emissions may represent up to approximately 3% of total nitrogen . Emissions from high temperature/pressure treatment are highly uncertain, due to a lack of relevant literature. As an approximate estimate, we used reported carbon emissions from the dehydration stage (<200°C) 1 2 of pyrolysis and gaseous protein losses in hydrothermal liquefaction for N2O, placing a large uncertainty distribution around these values (±100%). Finally, any carbon removed through electrochemical oxidation is assumed to be released as carbon dioxide16, while nitrogen removal is assumed to be negligible. Table 3. Technology-specific emissions estimates and other characteristics, based on relevant literature1,2,4,5,16,17. Parameter Expected Value Minimum Maximum Distribution Thermal drying Carbon dioxide emissions (% of total C) 1.6 0 3.2 Uniform Methane emissions (% of total C) 0.03 0 0.06 Uniform Ammonia emissions (% of total N) 0.26 0 0.52 Uniform Transformation of NH3 to N2O (% of NH3) 1 0 2 Uniform Final moisture content (%) 10 5 10 Uniform Combustion Methane emissions (% of total C) 0 - - (Constant assumption) N2O emissions (% of total N) 1.5 0.07 2.95 Uniform Ash weight remaining (% of initial solids) 30 24 30 Uniform Nitrogen remaining in ash (% of initial N) 0 - - (Constant assumption) High temperature/pressure treatment Carbon dioxide emissions (% of total C) 13 0 26 Uniform Methane emissions (% of total C) 11 0 22 Uniform N2O emissions (% of total N) 3.5 0 7 Uniform Electrochemical oxidation Carbon dioxide emissions (% of degraded C) 100 - - (Constant assumption) Methane emissions (% of degraded C) 0 - - (Constant assumption)

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

While the previous paragraphs account for many of the conditions under which direct emissions may occur, each sanitation system considered in this report is distinct and required additional assumptions related to its operation and performance. These assumptions, along with diagrams showing all components of each technology included within the system boundaries, are provided in Appendices I-IV (Figures A.1-A.4; Tables A.1-A.4). To briefly summarize here, emissions from the HTClean system include those occurring while mixed waste (urine, feces, flush water) is stored in the holding tank, while it undergoes high temperature/pressure treatment, and after it has been filtered. Filtered solids are assumed to be land applied (during which remaining carbon and nitrogen emissions may occur), while liquid is recycled as flush water (with the exception of excess water that is discharged to balance out urine additions into the system). Any easily biodegradable COD and nitrogen remaining in the portion of liquid that is recycled are assumed to degrade over time under mostly anaerobic conditions. The Empower Sanitation Platform is the only system of the four that separates liquid (urine, flush water) and solid (feces) streams prior to treatment. The solids are stored in an extruder/macerator, where conditions are assumed to be mostly anaerobic (as the material eventually becomes densely packed). This assumption is relatively inconsequential, as solids remain in this portion of the system for a short time (<1 day). Solids then undergo thermal drying, and the dried material may either be collected for centralized combustion or land application. The liquid stream moves through a series of baffle tanks, where some solids and COD are removed through settling. It is then treated with granular activated carbon (GAC) and electrochemical oxidation, and treated liquid is recycled as flush water (with excess discharged to balance urine additions). In this case, electrochemical treatment is assumed to be used for disinfection but not COD reduction. Any easily biodegradable COD removed through settling or GAC filtration is assumed to remain in the liquid system and degrade over time under mostly anaerobic conditions. Minimal nutrient removal is assumed to occur in this system, although most nitrogen in the liquid system does remain in the system as recycled water, such that it degrades over time as well. The Eco-San system directs mixed waste to a septic tank, where a fraction of COD and solids (including nutrients contained in the solids) settle and degrade anaerobically until the tank is emptied after 6-12 months. These collected solids are assumed to be land applied, during which remaining easily biodegradable COD and nitrogen can release emissions. Effluent from the septic tank is treated through electrochemical oxidation and membrane filtration. In the Eco-San system, electrochemical treatment removes 50-60% of COD, with the carbon released as carbon dioxide. Following treatment, the liquid is recycled as flush water. Any easily biodegradable COD and nitrogen that are filtered or recycled are assumed to remain in the system and degrade over time. Finally, the Omni-Processor includes drying and combustion processes to treat sludge. However, the processor represents only a part of the full sanitation value chain. User interfaces, containment, conveyance, and pretreatment are also needed, and these processes will entail additional emissions. To account for these steps, we assumed the population served by the Omni- Processor uses well-managed pit latrines. These latrines are emptied regularly by tanker trucks that transport collected sludge to a centralized facility. Before the sludge enters the Omni- Processor, its solids content must be raised to 20%, so we incorporated a passive dewatering process (unplanted drying beds) into the full system. Finally, the liquid draining out of the drying beds also requires treatment, so we included a series of anaerobic and facultative lagoons, each of which removes a fraction of influent COD and emits methane and N2O. Technology operation. Electricity required for each technology was determined using provided cut sheets or communication with design teams. An inventory of emissions (including uncertainty) associated with various electricity types was acquired using the ecoinvent v3.2 database9 accessed within SimaPro v8.5.2.0, a software for implementing life cycle assessment. Emissions were then converted to climate change impacts (reported as kg of CO2 equivalent per kWh of

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

electricity) using the U.S. EPA’s Tool for the Reduction and Assessment of Chemicals and Other Environmental Impacts (TRACI; 2.1 v1.03, implemented within SimaPro v8.5.2.0)11. In the scenario analysis (described in a subsequent section), country-specific data for India, South Africa, and Uganda were used whenever available, otherwise global estimates were applied. These electricity type-specific factors for climate change impacts (per kWh of electricity) were then used to develop country-specific factors for India, South Africa, and Uganda, using a weighted average calculated based on 2016 electricity mix data from the U.S. Energy Information Administration8 (Tables 4-5). These final country-specific factors for climate change impacts (as kg CO2 eq per kWh) were then multiplied by the electricity consumption of each technology to determine climate change impacts due to electricity. Table 4. Country-specific electricity mixes. Electricity Type Unit India South Africa Uganda Nuclear billion kWh 35.01 15.21 0 Hydroelectricity billion kWh 121.28 0.69 3.21 Geothermal billion kWh 0 0 0 Wind billion kWh 44.86 3.70 0 Tide and Wave billion kWh 0 0 0 Solar billion kWh 14.13 3.10 0.038 Biomass and Waste billion kWh 43.64 0.26 0 Fossil Fuels billion kWh 1127.52 213.07 0.22 Coal % 58.6 74.8 0 Oil % 33.7 22.0 100.0 Natural Gas % 77 3.2 0 Total Generation billion kWh 1386.44 234.51 3.46 Source: U.S. Energy Information Administration8

Table 5. Country-specific climate change impacts from electricity. Expected Value Minimum Maximum Assumed Country -1 -1 -1 (kg CO2 eq·kWh ) (kg CO2 eq·kWh ) (kg CO2 eq·kWh ) Distribution India 1.298 1.291 1.306 Uniform South Africa 1.032 1.028 1.036 Uniform Uganda 0.113 0.106 0.121 Uniform

Although the Omni-Processor is not reported to require outside electricity, it does consume certain materials during operation, and these materials were also incorporated into the analysis. The design team reported that hydrated lime is used for emissions control, while diesel fuel and natural gas are used during system startup. The team estimated startup occurs approximately twelve times per year. We combined estimated annual quantities of each of these materials with emissions factors extracted from the ecoinvent database9 accessed through SimaPro (Table 6) to estimate the equivalent carbon dioxide emissions associated with the use of these consumable materials. Country-specific data were not available for emission factors associated with consumables and therefore global estimates were used.

Table 6. Climate change impacts from consumables (kg CO2 eq per MJ or per kg). Expected Value Minimum Maximum Assumed Inventory Item Unit -1 -1 -1 (kg CO2 eq·unit ) (kg CO2 eq·unit ) (kg CO2 eq·unit ) Distribution Diesel fuel MJ 0.175 0.088 0.261 Uniform Butane Kg 0.382 - - (Constant) Hydrated lime Kg 0.923 0.905 0.942 Uniform

Transportation. Transportation emissions are primarily derived from the transport of solids to (only Janicki Omni-Processor) and from (all four) the analyzed technologies. The following equation (Eq. 1) was used to determine the GHG emissions resulting from transportation.

��� = � ∗ � ∗ � (Eq. 1)

Where GHGtransport is the total CO2 equivalent emissions resulting from transportation of solids (kg CO2 eq per person per year); msolids is the mass of solids to be transported (metric tonnes per

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

person per year); dtransport is the transport distance of solids (km); and etruck is the CO2 emission factor of a truck (kg CO2 eq per tonne·kilometer). The mass of solids that needs to be transported was estimated in the direct emissions calculations. The transport distance was assumed to lie between 2-10 km, and the emission factors of various trucks were found in the ecoinvent database (Tables 7-8). Emission factors varied by truck type and hauling capacity, and the full range were used to characterize its uncertainty distribution. Country-specific data were not available, and general global values were used. Table 7. Climate change impacts from solids transport. Truck Capacity Emissions Inventory Item -1 (metric ton) (kg CO2 eq·tkm ) Transport, freight, lorry >32 metric ton > 32 0.0576 Transport, freight, lorry 16-32 metric ton 16-32 0.168 Transport, freight, lorry 7.5-16 metric ton 7.5-16 0.220 Transport, freight, lorry 3.5-7.5 metric ton 3.5-7.5 0.526

Table 8. Uncertain parameter values for transportation. Assumed Parameter Expected Value Minimum Value Maximum Value Distribution Transport distance (km) 5 2 10 Uniform Truck Transport emissions -1 0.194 0.0576 0.526 Uniform (kg CO2 eq·tkm )

Fertilizer Offsets. Based on our estimates of nutrient excretion7, along with emissions and losses that occur during processing (see the section on direct emissions), we estimated the quantities of nutrients (nitrogen, phosphorus, potassium) embedded in the solid recovery products generated by each system. We then combined these recovered nutrient quantities with emissions factors associated with the production of single-nutrient fertilizers (from ecoinvent within SimaPro): • nitrogen: urea ammonium nitrate, calcium ammonium nitrate, urea, ammonium nitrate, calcium nitrate, ammonium sulfate • phosphorus: single superphosphate, triple superphosphate • potassium: potassium chloride, potassium sulfate For each nutrient, all relevant single-nutrient fertilizer emissions factors were used to characterize an uncertainty distribution (Table 9). The expected value was the average of the factors for individual fertilizers, and the minimum and maximum values were defined by the smallest and largest individual factors. In each system, the masses of recovered nitrogen, phosphorus, and potassium were multiplied by the relevant factor to estimate GHG offsets associated with reduced fertilizer production (Eq. 2):

��� = � ∗ � (Eq. 2)

Where GHGfertilizert is the total CO2 equivalent emissions offset by recovery of each nutrient (kg CO2 eq per person per year); mnutrient is the mass of nitrogen, phosphorus, or potassium recovered (kg per person per year); and efertilizer is the CO2 emission factor associated with the given nutrient (kg CO2 eq per kg N, P, or K). After estimating offsets from each individual nutrient, the total offset was calculated as the sum from the three nutrients. Table 9. Uncertain parameter values for fertilizer offsets. Assumed Parameter Expected Value Minimum Value Maximum Value Distribution Climate change impacts of N fertilizer 5.4 1.8 8.9 Triangular production (kg CO2 eq per kg N) Climate change impacts of P fertilizer 4.9 4.3 5.4 Triangular production (kg CO2 eq per kg P) Climate change impacts of K fertilizer 1.5 1.1 2.0 Triangular production (kg CO2 eq per kg K)

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Uncertainty and sensitivity analyses. For a variety of reasons (e.g., contextual differences, experimental uncertainty), many individual parameters cannot be specified exactly, and a key aspect of our general design philosophy involves the incorporation of uncertainty and sensitivity analyses. While we first analyzed a “base case” in which we assumed a single value for each uncertain parameter that we believed to be most likely, the uncertainty analysis enabled us to go beyond this one scenario. We defined distributions that are likely to contain most or all possible values of uncertain parameters (Tables 1-9, A.1-A.4), and then we ran numerous simulations in which random values for all uncertain parameters are pulled from the distributions we have defined. For this project, we ran 10,000 simulations with parameter values generated through Latin hypercube sampling18. This process produced a distribution of emissions defining a likely range for each system. When presenting results, ranges indicate the 5th and 95th percentile output values from the uncertainty analysis. The input and output distributions from the uncertainty analysis can also be used to calculate Spearman’s rank correlation coefficients as a measure of results’ sensitivity to individual parameters. Sensitivity refers to the degree to which an output (e.g., total GHG emissions) is related to a single input parameter. Spearman’s coefficients are calculated by ranking the values in each input and output distribution (e.g., the lowest value is assigned a rank of 1, the next lowest value is assigned 2, etc.) and determining the correlation between these ranks. The coefficient value indicates the degree to which an arbitrary monotonic function can describe the relationship between the input parameter and the output value (without requiring assumptions about the frequency distribution of each variable or the relationship’s linearity)19. Coefficient values range from -1 to 1, and coefficients with larger absolute values signify a stronger correlation. Values close to 1 indicate a positive correlation (the output is likely to increase with an increase in the input), while values close to -1 indicate a negative correlation (the output is likely to decrease with an increase in the input). Scenario analysis. Beyond the general uncertainty and sensitivity analyses, we also considered several scenarios designed to evaluate the impacts of key context-specific parameters related to diet and electricity characteristics. Diet (specifically, calorie and protein intake) affects excretion rates of organic carbon (in our model, we represent aggregate organic content using COD) and nitrogen, which determine the total quantities of methane and N2O that may be released during degradation of bodily waste under a given set of environmental conditions. Regarding electricity, the mix of sources used (e.g., coal, hydroelectric, solar) determines the GHG emissions associated with generation and delivery of a certain quantity of electrical energy (e.g., kg CO2 eq per kWh). Locations where electricity generation is heavily reliant on fossil fuels will be associated with higher emissions. In our scenario analysis, we investigate locations of interest through three country-specific scenarios (Uganda, South Africa, India). For each country, we used per capita COD and nutrient excretion rates based on calorie and protein availability in that country7, and we used the country’s reported electricity mix8 to calculate the electricity emissions factor. In contrast, our general analysis used recommended global values of COD and nutrient excretion13 to be consistent with preliminary estimates from the ongoing University of Leeds CACTUS project. The general analysis also employed a generalized electricity emissions factor for Africa from the ecoinvent database, with a wide uncertainty distribution encompassing other contexts. We also considered a fourth scenario, in which we focused on the use of solar panels to provide all electricity required by each system. In this case, diet assumptions were the same as those in the general analysis we had already completed. For all scenarios, any altered assumptions were given context-specific uncertainty distributions, and we repeated the uncertainty analysis using these altered distributions to estimate likely ranges for scenario-specific results.

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Appendix I. HTClean system boundary and assumptions

Pretreatment / Post-treatment / Influent Treatment containment recovery

CH4 CO2eq N2O

CH4 CH4 N O N2O Transport 2 Filter cakes Land applied Feces Holding tank High temp/pressure 1-2 day retention >160ºC, <25 bar Filtration Urine

Flush water Electricity Water Recycled for flushing

CO2eq Excess CH 4 water N2O discharged Figure A.1. All components of the HTClean system included within the system boundary (represented by the dashed box). System parameters and assumptions are based on relevant literature1–3,6,7,12–14.

Table A.1. Additional parameters and assumptions for the HTClean system. Expected Parameter values Min Max Distribution Reference value General HTClean: Users per day 5 4 6 Uniform cutsheet cutsheet (0.2-0.9 L per flush; expected HTClean: Flush water 1.5 0.8 3.6 Triangular value assumes 0.2 L for urine 3 times per (L·cap-1·d-1) day , 0.9 L for feces 1 time per day) Pretreatment/containment Holding tank HTClean: Holding tank Helbling team (assume plug flow reactor, 1.5 1 2 Uniform retention time (days) pumped out all at once) Operating electricity HTClean: Required 650 400 650 Uniform Helbling team (next gen may be < 400 W) electricity (W)

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Appendix II. Empower system boundary and assumptions

Pretreatment / Post-treatment / Influent Treatment containment recovery

CO2eq N2O

CH4 CH4 Transport N2O Combustion N2O

CH4 80-90% Drying N O Feces Extruder or 2 <10% moisture content

Electricity Soil amendment S/L separator Land applied

CO2eq Water Urine Baffle tanks GAC and EC Recycled for flushing

CH4 CH4 Excess Flush water CH4 N2O discharged Figure A.2. All components of the Empower system included within the system boundary (represented by the dashed box). System parameters and assumptions are based on relevant literature3–7,12,13,20,21.

Table A.2. Additional parameters and assumptions for the Empower system. Parameter values Expected value Min Max Distribution Reference General Empower: Users per day 30 10 50 Uniform cutsheet Flush water (L·cap-1·d-1) 15 - - (Constant) design team Pretreatment/containment Extruder Empower: Solid/liquid separation (%) 85 80 90 Uniform design team Empower: Extruder solids retention time (hr) 8 4 12 Uniform assumption Baffle tanks Empower: Baffle tank COD removal (%) 38 30 45 Uniform 6,21 Empower: Baffle tank N removal (%) 0 - - (Constant) assumption Treatment GAC Empower: GAC COD removal (%) 60 45 80 Triangular 20 Empower: GAC N removal (%) 0 - - (Constant) assumption EC Empower: EC COD removal (%) 0 - - (Constant) assume COD deg. is negligible Empower: EC N removal (%) 0 - - (Constant) assumption Operating electricity Empower: Required electricity (kWh·cap-1·d-1) 0.2 0.17 0.22 Uniform Design team

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Appendix III. Eco-San system boundary and assumptions

Pretreatment / Post-treatment / Influent Treatment containment recovery Excess CH CH4 4 water CH4 N O N2O 2 discharged N2O Feces EC treatment Water Septic tank With membrane polishing Recycled for flushing Urine Electricity CO2eq Flush water Transport Undegraded settled CO2eq Soil amendment sludge Land applied

CH4 N2O Figure A.3. All components of the Eco-San system included within the system boundary (represented by the dashed box). System parameters and assumptions are based on relevant literature3,6,7,12,13,22–24.

Table A.3. Additional parameters and assumptions for the Eco-San system. Expected Parameter values Min Max Distribution Reference value General Eco-San: Flushes per day 400 50 800 Uniform cutsheet Eco-San: Flush water (L·flush-1) 7 - - (Constant) Eco-San team Flush frequency (flushes·cap-1·d-1) 5 4 6 Uniform 24 Pretreatment/containment Septic tank Eco-San: Septic tank solids retention time (yr) 0.5 0.5 1 Triangular cutsheet Eco-San: Septic tank solids settling (%) 50 40 60 Triangular 3,23 Eco-San: Septic tank COD settling (%) 35 30 40 Uniform 22,23 6 (assume similar to lagoon Eco-San: Total solids of settled sludge (%) 25 25 30 Uniform solids) Treatment EC oxidation Eco-San: EC COD removal (%) 55 50 60 Uniform Eco-San team Eco-San: EC N removal (%) 0 - - (Constant) assumption Operating electricity Eco-San: Required electricity (kWh·m-3) 18 - - (Constant) Eco-San team

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest

Appendix IV. Omni-Processor system boundary and assumptions.

Pretreatment / Post-treatment / Influent Treatment containment recovery

CO eq CO eq CH 2 2 CH4 CH4 4 N O N2O N2O N2O 2 Consumables Transport Feces Sludge Boiler combustion Ash Pit latrine Dryer dewatering To ash Land applied Urine Omni-Processor only Transport Anaerobic Facultative lagoon lagoon

CO2eq

CH4 CH4 N2O N2O Omni-Processor with pretreatment

Omni-Processor with pretreatment and latrines Discharged water Figure A.4. All components of the Omni-Processor system included within the system boundary (represented by the dashed boxes). In this case, three boundaries are shown: the Omni-Processor itself, the processor with pretreatment, and the processor with pretreatment and latrine containment. System parameters and assumptions are based on relevant literature3,6,7,12,13,22,23,25–30.

Table A.4. Additional parameters and assumptions for the Omni-Processor system. Parameter values Expected value Min Max Distribution Reference General Omni-processor: Users per day 400,000 300,000 500,000 Uniform cutsheet Pretreatment/containment Collected fecal sludge (single-household latrines) Latrine sludge: Total volume (L·cap-1·yr-1) 270.0 - - (Constant) 25 Latrine sludge: Total solids (g·L-1) 25.0 15 48 Triangular 25 Latrine sludge: Emptying period (years) 0.5 0.2 1 Triangular 25 Latrine sludge: N leached from latrine (%) 13.0 1 50 Uniform 22,26–29 Latrine sludge: P leached from latrine (%) 18.0 0 37 Uniform 22,28,29 Latrine sludge: K leached from latrine (%) 21.0 11 31 Uniform 29 Dewatering O-P Drying bed: Final solids content (%) 20.0 - - (Constant) cutsheet O-P Drying bed: Drying time (days) 12.5 10 15 Uniform 23 Anaerobic lagoons for liquid from dewatering Anaerobic lagoons: COD removal (%) 65 60 70 Uniform 30 Anaerobic lagoons: HRT (days) 35 20 50 Uniform 30 Facultative lagoons for liquid from dewatering (after anaerobic lagoons) Facultative lagoons: COD removal (%) 80 70 90 Uniform 30 Facultative lagoons: HRT (days) 100 20 180 Uniform 30 Consumables Diesel fuel (L·yr-1) 1,440 - - (Constant) Design team Natural gas (m3·yr-1) 18,000 - - (Constant) Design team Hydrated lime (kg·yr-1) 35,762 - - (Constant) Design team (9 lb·hr-1)

12/20/2018 BMGF OPP1173370, Project 13C (Duke TO 283-1325) Guest