The Dirty Footprint of the Broken Grid The Impacts of Fossil Fuel Back-up Generators in Developing Countries September 2019 © International Finance Corporation 2019. All rights reserved. 2121 Pennsylvania Avenue, N.W. Washington, D.C. 20433 Internet: www.ifc.org The material in this work is copyrighted. Copying and/or transmitting portions or all of this work without permis- sion may be a violation of applicable law. IFC does not guarantee the accuracy, reliability or completeness of the content included in this work, or for the conclusions or judgments described herein, and accepts no responsibility or liability for any omissions or errors (including, without limitation, typographical errors and technical errors) in the content whatsoever or for reliance thereon. Table of Contents FORWARD AND ACKNOWLEDGEMENTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv EXECUTIVE SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Major Findings .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Next Steps. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii GLOSSARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 BACKGROUND AND RESEARCH METHODS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 COPING WITH BROKEN GRIDS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 PRIMER ON BACKUP GENERATORS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Generator Types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Many Costs of Generators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 RESEARCH METHODS OVERVIEW.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 NIGERIA: A UNIQUE AND LARGE-SCALE BACKUP GENERATOR MARKET.. . . . . . . . . . . . . . . . . . . . . . . . . . 7 RESULTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 THE GLOBAL FLEET OF BACKUP GENERATORS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Fleet Size & Composition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Installed Fleet Capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Energy Generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Fuel Consumption.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 THE ECONOMIC COSTS OF BACKUP GENERATORS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Capital investment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Fuel Related Costs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Consumption Subsidies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 POLLUTANT EMISSIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 High Priority Opportunity for Pollution Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 BUGS as Significant Source of Pollution .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Implications of Data Gaps on Pollutant Emissions and Impact Estimates.. . . . . . . . . . . . . . . . . . . . . . . . . . 28 COUNTRY-LEVEL ACCURACY AND UNCERTAINTY.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 CONCLUSION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 APPENDIX 1: METHODOLOGICAL DETAILS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 APPENDIX 2: OPPORTUNITIES TO REDUCE UNCERTAINTY IN ESTIMATES. . . . . . . . . . . . . . . . . . . . . . . . . 49 FORWARD AND ACKNOWLEDGEMENTS The following research models the global fleet of back-up fossil fuel generators. It is part of IFC’s emerging work to support solar and energy storage solutions that can provide reliable, sustainable, affordable energy to people and businesses relying on fossil fuel generators. The research findings include estimates of fleet size, composition, energy service, fuel con- sumption, and resulting financial costs and pollutant output (pollutant emissions) as an indicator for health and climate impacts. Our modeling focused on understanding global and regional trends to help clarify the overall footprint and related opportunity for alter- native solutions. It applied a broad geographic scope including 167 developing countries (excluding China). We limited our view to this scope and did not account for non-fuel maintenance costs, nor estimate the value of lost productivity from generator downtime and management, or costs passed onto customers from enterprises reliant on generators for day to day operations. We only present the part of the picture that we felt we could reasonably estimate with avail- able data from multiple sources. We rely on official import/export data, and therefore do not account for generators imported unofficially or produced locally. The available data for generator performance typically comes from laboratory testing, which would likely under- estimate fuel use and emissions for generators in use on the ground. Overall, the estimates presented in this summary are conservative, we believe significantly so. This is the foundation piece of an open source resource that we hope becomes a broader collaborative effort at producing and sharing data. Because of our global focus and stan- dardized approach to modeling, the specific results should be treated as a starting point for further research, rather than a final result. Focused work in national and local markets will be crucial to follow through on this first effort. This is the impressionistic painting. We hope it leads to a more detailed and fuller picture. We would like to acknowledge and thank our research partner, the Schatz Energy Research Center at Humboldt State University. This research and IFC’s engagement in this area will be further developed in partnership with the IKEA Foundation, Netherlands Ministry of Foreign Affairs and the Italian Ministry of Environment, Land and Sea. The authors of the study include Nicholas L. Lam, Eli Wallach and Chih-Wei Hsu, Arne Jacobson , and Peter Alstone from the Schatz Energy Research Center (SERC); Pallav Purohit and Zbigniew Klimont from the International Institute for Applied Systems Analysis (IIASA). The contributing editors are Russell Sturm, Daniel Tomlinson, Bill Gallery, and Rwaida Gharib from the World Bank Group’s International Finance Corporation (IFC). Executive Summary About 1.5 billion people around the world live day-to-day with “broken” electricity grids and experience blackouts for hundreds and sometimes thousands of hours a year. For this population, reliance on distrib- uted diesel and gasoline backup generators, or BUGS, is a common stopgap measure. These generators are deployed across the globe on a large scale both on- and off-grid, at homes, businesses, and industrial sites. They support access to energy but come with significant costs. The goal of this research project is to estimate the scale and impacts of generators serving energy access needs within developing regions of the world. With a broad geographic scope, including 167 develop- ing countries (excluding China), the coverage represents 94 percent of the population living in low- and middle-income regions of the world.. We develop and use a modeling framework using the best available data for each country to estimate the size and composition of the fleet of generators, operational time, fuel consumption, and financial, health, and climate impacts. The estimates are designed to help clarify the opportunity in developing countries for clean technologies such as solar and storage (solar + storage) to replace generators, and to avoid these costs and impacts. Major Findings The fleet of generators in the developing countries modeled serves 20 to 30 million sites with an installed capacity of 350 to 500 gigawatts (GW), equivalent to 700 to 1000 large coal power stations. The fleet has a replacement value of $70 billion and about $7 billion in annual equipment investment. Over 75 percent of the sites where generators are deployed are “grid-connected.” The map in Figure 1.1 illustrates the volume of diesel and gasoline fuel burned annually across modeled countries. FIGURE 1.1: TOTAL DIESEL AND GASOLINE CONSUMED IN 2016 ACROSS ALL MODELED COUNTRIES. vi Backup generators are a major source of electricity access maintenance costs for generators could add an additional in some developing regions, providing 9 percent of the 10 percent to 20 percent to fuel service costs.2 electricity consumed in Sub-Saharan Africa, and 2 percent in South Asia. In western Africa, generators account for Backup generators are a significant source of air pol- over 40 percent of the electricity consumed annually. This lutants that negatively impacts health and the environ- requires considerable quantities of fossil fuel; 20 percent ment. As a pollution source, generators are often hidden of the gasoline and diesel consumed in Sub-Saharan Africa from policymakers since their fuel consumption may be is burned for electricity generation. In regions where gen- lumped in with the transport sector in official statistics. erators are a predominant source of energy access, spend- Generators consume the same fuels and also emit the ing on fuel can be equivalent to or higher than the total same pollutants as cars and trucks, except they are used in national spending on the grid. Figure 1.2 shows how the closer proximity to people’s homes and businesses. Often, spending is notably similar in size to the overall utility emission limits for generators are also less stringent than electricity sector in some regions of Africa. Western Africa for vehicles. As a result, the pollutants emitted from gen- is a particularly significant market for backup generators, erators may represent meaningful but largely unaccounted owing largely to Nigeria, with its large economy, popula- or misclassified impacts on population health and the tion, and low-reliability power sector that together drive environment. . Generators emit the same pollutants as many homes and businesses to rely on backup generators. cars and trucks, except they are used in closer proxim- ity to people’s homes and businesses, and emission limits Electricity from backup generators is expensive, with are often less stringent than for vehicles. In Sub-Saharan $28 billion to $50 billion spent by generator users on fuel Africa, we estimate that generators account for the major- each year. This corresponds to an average service cost of ity of power sector emissions of nitrogen oxides (NOx) $0.30/kWh for the fuel alone (ranging from $0.20/kWh and fine particulate matter (PM2.5), with their contribu- to $0.60/kWh depending on generator size and fuel type), tion to PM2.5 being equivalent to 35 percent of the emis- usually much higher than the cost of grid-based energy sions from the entire transportation sector. BUGS are a ($0.10–0.30 / kWh) and on par with current estimates modest contributor to CO2, accounting for roughly 1 per- of the levelized cost of solar + storage.1 Operations and cent of annual emissions across modeled countries. FIGURE 1.2: ANNUAL EXPENDITURE ON GRID-BASED ELECTRICITY VS. FUEL FOR BACKUP GENERATORS BY REGION, AND THE TOTAL INSTALLED FLEET CAPACITY, IN AFRICA Grid vs. Generator Fuel Expenditures Backup Generator Fleet Capacity Map Total Expenditure on Electricity (Billion USD /yr) Installed Capacity (MW) 0 5 10 15 10 100 1000 10000 40000 Total Expenditure on Electricity Northern Africa (Billion USD /yr) 0 5 10 15 Southern Africa Spending Northern Africa Category Western Africa Generator Fuel Southern Africa Utility Grid Spending Eastern Africa Category Western Africa Generator Fuel Middle Africa Utility Grid Eastern Africa Middle Africa vii FIGURE 1.3: A CLUSTER OF SMALL GASOLINE GENERATORS LEAKING FUEL AND LUBRICATING OIL INTO A STORMWATER TRENCH IN A MARKET IN ABUJA, NIGERIA Photo: A. Jacobson Our modeling focused on understanding global and methodology for estimating fleet characteristics across regional characteristics to help clarify the overall oppor- countries with comparable data sources whenever pos- tunity. It is important to emphasize the need for focused sible. We chose not to include “expert based” estimates for work in national and local markets to follow through and sectors or countries with missing data. The estimates we solidify the market intelligence groundwork. Because of make are benchmarked against national and regional fos- our global focus and standardized approach to model- sil fuel inventories as an additional verification step. Based ing, the specific results for every one of the 167 countries on these decision factors and known data gaps, our central we included in the modelling effort should be treated as a estimates of fleet characteristics are likely conservatively starting point for further insight, rather than a final result. low and could be treated as a reasonable lower bound. Despite the negative impacts that unreliable electricity supply has on populations and economies, there remains Reported fleet sizes and impacts could be underestimated limited data on power systems and the operational charac- for the following reasons (among others):  teristics of BUGS fleets in specific developing country con- • Gray market or untracked imports of generators. Genera- texts, and significant discrepancies exist in coverage and tors that are missed by formal tracking are not counted in reporting which make comparison across what few data the import / export data that we used as the basis for fleet sets exist difficult. In addition, there are gaps in our ability size in most countries. to estimate the scale of unregulated sales of generators and a weak understanding of the true cost of operations and • Locally assembled generators. Most generators are as- maintenance, including lost opportunities for productivity. sembled in industrial centers in Asia, but domestically assembled units may be missing from our data. When interpreting our analysis, it is important to keep • Longer generator lifetimes. It is possible that generators these tradeoffs and assumptions in mind. During the in some areas are maintained to run beyond the assump- model development our priority was to use a consistent viii tions we make about lifetime. This would lead to our eliminate them. The uncertainty decomposition technique estimates of active fleet size to be conservatively low. we used in our model reveals where additional research could contribute most to improving understanding of the • Very poorly performing/high pollution generators. Based fleet, operations, and impacts of generators. We found that on the available data, we apply performance values from for gasoline generators, about 60 percent of uncertainty generators measured in developed countries, typically is related to the number of sites using generators, due to under controlled laboratory settings. We expect this to poor understanding of the service life of these relatively lead to conservatively low estimates of fuel demand and small and inexpensive generators. Targeted research and related impacts compared to poorly performing genera- better survey coverage of homes and businesses, including tors that may be in use. more detailed data on service quality using instruments Regardless, the results are still significant and large—and like the World Bank Multi-Tier Framework surveys, could the reality that could be uncovered with more detailed significantly improve certainty in the estimates related to understanding of local markets could be even larger. gasoline generators. For diesel generators, about 60 per- cent of the uncertainty is related to the sizing of the fleet Next Steps of diesel generators and the loads they serve. For these, a detailed survey of sites, including monitoring of loading Overall, our results indicate a significant opportunity and fuel consumption, could help address this uncertainty. to reduce costs and negative health and environmental For all classes of generators, data on the frequency, dura- externalities by replacing diesel and gasoline generators. tion, and patterns of blackouts contributes to 10 percent To follow through, it is important to develop both the to 20 percent of the uncertainty in estimates. Grid status technology and business model solutions needed and to data could limit this uncertainty and also help inform the improve the understanding of generator impacts in local design of clean technologies such as solar and storage that contexts. The local realities of the solar industry, grid reli- would serve needs of customers facing particular reliability ability, fossil fuel competitiveness, and the utility and regu- realities. latory approach to distributed generation are among the important factors. While our modeling approach was not There are also remaining areas of missing fundamental designed to reveal special insight on how to deploy such data related to the emissions from backup generators and clean technologies, there are some clear next steps that their impacts on community health and air quality. There could be taken. is a scarcity of data on the performance of generators used in developing countries. This has led to a reliance on per- First, development and private sector actors should work formance data from well-maintained generators that are to accelerate and support emerging clean energy technol- very likely to be better performing than the units deployed ogy deployment and markets to better serve the needs in countries modeled in our study. Furthermore, the expo- of people who now rely on generators. In parallel with sure contribution to people is not well mapped or under- market transformation, improving the fidelity of data and stood, nor are the resulting health impacts. If generators knowledge on generators could help focus and target these follow similar trends to other energy service technologies, efforts. our results likely lead to highly conservative estimates of emission impacts. Making measurements of emissions Our initial results suggest that a large opportunity exists, from generators operating in practice is a high priority to but that there is still significant uncertainty in many facets better understand the health and environmental benefits of our estimates that could affect local decision making. from relegating or replacing fuel-based generators. Improving understanding of backup generators could help ix Glossary TERM (SYNONYM / ABBREVIATION) DEFINITION BC Black carbon BUGS Backup fossil-fueled generator Capacity factor Fraction of rated capacity that the generator operates at CIA United States Central Intelligence Agency CO2 Carbon dioxide EID Experienced interruption duration ER Emission rate; the quantity of pollutant released to the atmosphere per unit of time EF Emission factor; the quantity of pollutant released to the atmosphere per unit of activity associated with that release GAINS Greenhouse Gas - Air Pollution Interaction and Synergies Model GBD Global burden of disease GDP Gross domestic product GEE Generalized estimating equation : diesel large Generator­ Diesel-fueled generator with a rated capacity greater than 300 kW Generator: diesel small Diesel-fueled generator with a rated capacity of less than 60 kW Generator: petrol or gasoline Petrol-fueled generator (any rated capacity) Generator: diesel medium Diesel-fueled generator with a rated capacity of between 60 and 300 kW GW Gigawatt IEA International Energy Agency IFC International Finance Corporation IIASA International Institute of Applied Systems Analysis IMF International Monetary Fund kt Kiloton kVa Kilo-volt-ampere kW Kilowatt x kWh Kilowatt hour LMICs Low- and middle-income countries (World Bank, 2018) MJ Megajoule Mt Megaton MW Megawatt NMVOC Non-methane volatile organic compounds NOX Nitrogen oxides O&M Operation and maintenance O3 Ozone OC Organic carbon PM2.5 Particulate matter with a diameter of 2.5 micrometer or less Pollutant concentration Mass of pollutant contained per unit volume of media PPP Purchasing power parity PV Photovoltaics—A type of solar electricity technology. The typical technology used for “solar panels” that are installed on buildings and in utility-scale generation. Rated capacity/ nameplate capacity Intended full-load sustained output of a generator (nameplate capacity) Runtime Duration of time a generator is running over a specified time period SAIDI System average interruption duration index SE4ALL Sustainable Energy for All SLCFs Short lived climate forcers SO2 Sulfur dioxid solar+storage An energy system combining distributed solar electricity generation with battery energy storage, often with the capability to operate and serve on-site loads without the grid. TWh Terawatt hours (10^12 watts) UI Uncertainty interval UN United Nations USD United States dollar UV Ultraviolet VOC Volatile organic compounds Introduction Living with an unreliable electricity connection is a day-to-day reality for billions of people in develop- ing countries. Blackouts can be regular or unex- pected, stretching to hours or days. To better meet their energy needs, tens of millions of people purchase and operate distributed genera- tion to supplement their unreliable grid connection at households and businesses, or for off-grid power. For decades the only viable option has been fos- sil fuel “backup” generator sets (BUGS) like the one pictured here.3 These generators are usually designed for intermittent service but are used for thousands of hours a year in places with the worst grid reliability or in off-grid locations. Continued Photo: A. Jacobson reliance on them brings financial, environmental, and health hardships. Reducing reliance on BUGS through replacement with integrated solar and energy storage systems presents an opportunity to reduce these hardships. However, understanding the scale of this opportunity requires an understanding of the extent of their use and the impacts of their operation. Because of the distributed and untracked nature of BUGS, however, there has been limited or incomplete information available around the current impacts of BUGS and the level of energy service they provide. This study contributes to addressing this knowledge gap by performing the most detailed characterization to date of backup generator fleets, the cost of their operation, and their contribution to health and climate damaging pollutant emissions. We use existing data to model the fleets and operations of BUGS in 167 developing countries,4 addressing several questions: • How many generators are installed and at what size range? • What are the patterns of grid (un)reliability that drive generator use? • How much energy service do generators provide? • How much fuel is burned and at what welfare and environmental cost? • What are the major knowledge gaps affecting our understanding of generator operations and impacts? This report describes our approach and results, which address the questions above. The results reveal the vast scale of reliance on BUGS. 2 Background and Research Methods COPING WITH BROKEN GRIDS People use fossil fuel BUGS primarily because of an inability to access reliable electricity service from an area electric power system (i.e., the grid). This access gap can stem from an inability to make physical connection to the grid or from intermittent grid service. For many, grid outages are a part of everyday life. The duration and frequency of such outages varies widely across countries and time of year depend- ing on demand and the availability of energy sources needed to generate electricity. The reliability of power systems also varies, from highly stable and reliable grids to power systems with frequent rolling or unplanned blackouts that can stretch for hours or days. Surveys conducted by the World Bank indicate that the duration of outages (often measured in terms of the System Average Interruption Duration Index, or SAIDI) ranges from hundreds to thousands of hours annually in coun- tries with weak grids.5 Based on published SAIDI estimates, we estimate that more than 2 billion people live with blackouts more than 100 hours a year and 1 billion with more than 1,000 hours. PRIMER ON BACKUP GENERATORS In response to uncertain grid conditions, backup generators—while only a stop-gap measure—have the potential to impose significant monetary and non-monetary costs on users, communities, businesses, and the environment. This background section briefly describes some background information on generators and their operation, followed by an overview of the methods we used to estimate them. In Appendix 1 we provide more depth and details on the background and methods. Generator Types There is a vast range in generator scales serving sites across the world, from less than a kilowatt to sev- eral megawatts, powering sites ranging from small households to industrial facilities. Understanding the size of these segments is important for evaluating the scale of the opportunity to replace generators. For smaller systems, a more standardized approach may be appropriate, while for larger generators there could be a business case for more customized design. Generators are typically installed so that they run as standalone alternatives to the grid or operate as an alternative power source during grid outages. Figure 3.1 illustrates a typical arrangement for grid- connected sites that use a transfer switch (often automated to switch on during blackouts) to connect the loads at a home or business to the grid or to a generator. Some sites do not use automatic transfer switches, instead relying on more manual, less intrinsically safe methods for powering loads in parallel with the electricity grid. We distinguish generators by the fuel they run on (diesel vs. gasoline) and the amount of power they can generate (watts). Both factors affect the efficiency of electricity generation and the size of applications. It 4 FIGURE 3.1: OUTLINE OF A SAFELY INSTALLED BACKUP GENERATOR INSTALLATION USING A TRANSFER SWITCH (NOT TO SCALE) TO ISOLATE THE GENERATOR AND HOME OR BUSINESS BEING SERVED FROM THE REGIONAL GRID is important to note that direct drive generating units for on core income-generating activities. For households, this agricultural and industrial applications are not considered means less time to focus on family, leisure, and producing in our fleet or impact estimates. a household income. These additional costs of operation are not included in our estimates due to a lack of support- The Many Costs of Generators ing data and knowledge beyond anecdote. They present additional opportunities to provide value to people who The continued reliance and operation of BUGS impose a replace generators with less burdensome pathways to elec- variety of costs on users, communities, governments, and tricity access. the environment; we distinguish and examine some of these costs as impacts within our modeling framework. Subsidies and public costs The use of BUGS to meet energy service needs is often The costs to users include: incentivized and enabled through government subsidies • Capital costs to purchase and install a generator (estimat- on fossil fuels. Despite the well-intentioned goals of many ed based on import value and retail markup) subsidy schemes, they are often inefficient and incur direct and indirect costs to users, governments, and the environ- • Fuel costs to operate the generator (based on expected ment. These subsidies make alternative pathways to elec- runtime due to grid outage) tricity services less competitive by creating artificially low • Operation and Maintenance (O&M) costs are not service costs for BUGS. included as a cost in our model estimates but can be considerable in some BUGS applications—conservatively Reducing reliance on generators could ease the subsidy on the order of 10 percent to 20 percent of the fuel costs burden on government budgets, while removing or reduc- in most situations.6 ing subsidies could better signal the cost of backup genera- tion to customers who may have other options. Indirect costs In addition to direct costs related to fuel, replacement Air Pollution parts, and technician labor, the effort spent to operate, BUGS are a potentially significant pollutant source, espe- maintain, and cope with generators imposes an oppor- cially at a local level. In areas where they are deployed, tunity cost on users. Depending on the frequency of use, BUGS contribute to the emissions of health and climate purchasing fuel and refueling the generator can be a daily damaging pollution. The emissions from BUGS contrib- or more frequent chore that exposes people to harmful ute directly or indirectly to nearly all pollutants found on fumes and spilled fuel, and may require considerable travel major priority (criteria) pollutant lists developed for the and transportation costs to refill containers. The time protection of human health. BUGS also contribute to cli- spent managing a generator is lost to other valuable activi- mate change through their emissions of carbon dioxide and ties. For business operators, this means less time to focus numerous short lived climate forcing pollutants (SLCFs). 5 Community Disruption classified by fuel type (i.e., diesel, gasoline) and size Noise pollution and accidental injuries are important (maximum power output) categories. In most countries impacts of BUGS, especially at the local level, but were (except India and Nigeria) we did not attempt to account not examined in detail as part of this study. Exposure to for domestically produced generators, which is a known excessive noise contributes to the local burden of disease source of conservative bias in our approach. through increased risk of heart disease, cognitive impair- 2. The total duration of power outages (i.e., system aver- ment in children, and loss of sleep, among others. BUGS age interruption duration index, or SAIDI) was the basis are also disruptive to social and business activities and are for the hours of BUGS operation (runtime); this was a frequently mentioned nuisance in accounts from people combined with manufacturer data about their efficiency who live with them. and assumptions about loading factor of generators to RESEARCH METHODS OVERVIEW estimate energy generation and fuel consumption. 3. Fuel consumption results were used to update a widely This study characterizes backup generator operations in used fuel and emissions inventory in order to estimate the 167 countries, representing 94 percent of the population contribution of BUGS to fossil fuel demand and emissions living in low- and middle-income regions of the world, of health and climate damaging pollutants.7 Fuel esti- excluding China. For most countries we applied a stan- mates were compared to IEA statistics for the power and dardized approach for modeling the backup generator sec- commercial sector and adjusted so that the overall energy tor based on globally available data sets. For Nigeria and use is consistent with IEA. India (the top two markets in terms of total load served by generators) a more customized approach was taken to 4. T hese fuel consumption quantities are used to estimate improve user segmentation and improve the model fidelity. fuel-related costs and pollutant emissions: a. Fuel cost based on consumption and estimated retail Figure 3.2 shows the workflow and types of data sources prices used to support our estimates, including. b. Cost of subsidizing fuel for BUGS based on estimated 1. Global import/export trade data on generators and consumption subsidies national surveys were used to estimate the number of c. Pollutant emissions from available data for generator generators used in 167 developing countries (fleet size), performance. FIGURE 3.2: OVERVIEW OF MAJOR PROJECT COMPONENTS, MODEL FLOW-DOWN, AND KEY DATA SOURCES Major Project Components Key Sources 1. Generator Fleet Size National trade records; Deployed units, installed capacity, fleet segmentation household and business surveys 2. Energy Generation SAIDI, stakeholder Grid reliability (SAIDI), generator runtimes, capacity interviews, factors, energy generation 3. Fuel Consumption Generator performance Fuel consumption curves, fuel consumption, global characteristics, IEA statistics, energy inventory GAINS 4a. Direct Monetary Costs Fuel prices, grid revenue, fuel subsidies, capital Fuel price records, national investment, others 4b. Health & Environmental Costs (Emissions) Emission factors, global emissions inventory, national Emission factors, GAINS and regional pollutant emission rates 6 We consider the uncertainty of input data in our report- outcomes. In the results we use error bars and ranges that ing of results. Our modeling approach uses Monte Carlo contain 90 percent of the possible cases we estimated. We simulations to randomly vary uncertain parameters (like also performed a more in-depth uncertainty decomposi- the number of hours of blackout or generator capacity) tion to identify the biggest sources of error in our model, within reasoned boundaries to estimate a range of possible with details described in Appendix 2. Nigeria: a unique and large-scale backup generator market Nigeria is a notoriously large market for backup generators. While it has the largest population (200 million people) and economy ($1.1 trillion GDP PPP adjusted) in Africa,8 there are only 5.3 GW of large- scale power stations reliably connected to the regional grid,9 which is 10 percent of the capacity of South Africa (with 55 million people and $0.767 trillion GDP PPP adjusted). This installed power capacity amounts to about 30 Watts per person, a similar installed capacity per capita to Ethiopia, Afghanistan, and the Democratic Republic of the Congo (DRC). As a point of reference, the global average is about 900 Watts per person. Figure 3.3 below shows that, compared to other large countries in the world, Nigeria is among the lowest per capita for generation capacity on the grid. However, Nigeria also has nearly the highest level of economic output in terms of GDP per installed watt of grid-scale generation, at over $100/Watt. The grid in Nigeria is not sufficient to serve the needs of the country, and the massive population and economy of Nigeria is instead largely powered with electricity from small-scale generators. FIGURE 3.3: GRID GENERATION CAPACITY IN THE CONTEXT OF COUNTRY POPULATION AND ECONOMY SIZE ●●● ● ● ● ● ● ●● ● ●● ● ● ●●●● ● ●● ●● ● ●● ●● ● ● ● ●● ● ●●●● ● ● ●● United States ●● ●● ●● ● ● Indonesia Nigeria ●● ●●● ● ● ● ● ●● ●● ● ● Cumulative total global population ● ●● ●● ● ● ●● ● ●● ●● ● ● ●● ● Pakistan ●● ●● ● ●●● ● ●● ●●● ● ● ● Cumulative total global GDP ● ●● ● ● 6 China ● ● ● ● ● 100 India ●● ● ● ● ●● ● (Trillion USD) ●● ●● ●● ● ● ● ● ● Brazil (Billions) ● ● ● ● ● ● ● ● ●● ● ●● ● 4 ●● ● ● ● ● ● Brazil ● ● ● ● ● ● ● ●● ● ● ● ●● ● India United States 50 ● 2 ● ●● ● ● ● ● ● Indonesia ● ● ● ● ● ● ● ● ● Pakistan China ●● ● ● ● ● ● Nigeria ● ● ● ● 0 ● ● 0 ● ● ● ●● ● ●● ● ●● ● 0 1000 2000 3000 4000 5000 0 50 100 150 Grid capacity per capita GDP per grid capacity [A] (Watts / person) [B] (USD / Watt) Labels are included for the seven countries with 200 million people or more. Panel [A] shows generators per capita. Panel [B] shows economic production in terms of GDP per installed watt of grid generator capacity. The source data are from CIA World Factbook,10 with a modification of Nigeria generation capacity data based on an SE4All prospectus.11 8 FIGURE 3.4: DIESEL GENERATORS TYPICAL OF THOSE THAT POWER LARGE HOUSING, COMMERCIAL, AND INSTITUTIONAL BUILDINGS Photo: A. Jacobson In the background is a solar street lamp and a presumably low-reliability electric distribution circuit. The “backup” generators deployed in Nigeria include by the government in 2015 over concerns about local air both diesel units and smaller gasoline-powered generators. pollution.13 In spite of the ban, these units remain widely Large diesel generators power offices, industry, and large available in retail markets. Two images below illustrate the homes and businesses (as is common in many parts of the ubiquity of these generators. Both show how merchants world with poor or no electricity access). The cost to oper- and small businesses in the market rely on generators for ate these large generators is significant. A recent estimate power in Abuja, Nigeria. by the Nigeria Labor Congress shows that “as much as N3.5 tn” (approximately $17 billion USD) is spent each The preponderance of generators in Nigeria is both an year by industrial generator users.12 The generators are economic and health burden. In our modeling study we also used at institutional, commercial, and large housing are able to estimate capital expenses, fuel costs, and air sites, like the one pictured in Figure 3.4. pollution quantities, but the effect of generator operation on quality of life is best understood through testimonials There are many large diesel generators in Nigeria, but the from people who live with them. country is also well known for widespread use of small gasoline generators. These inexpensive units have become On economic burdens: “Without electricity, no nation newly available and emerged in recent years as a fast- can go ahead. Without electricity there is nothing that is growing segment. Many are two-stroke generators that happening in the country. So we need power. … Three or burn a mixture of gasoline and lubricating oil (as opposed four days there will be a power supply. The next day off, to quieter and typically less polluting four-stroke engines the next three days on. That is the challenge we are hav- like those used in cars). In popular culture, this category of ing now.”14 The generators in use also impose a significant generator is known as, “I better pass my neighbor.” Units burden of effort and cost for operations and maintenance. have proliferated across households and small businesses As one shopkeeper described, “I have three generators. and were later banned from import in large quantities 9 FIGURE 3.5: SMALL GASOLINE GENERATORS POWERING SHOPS IN AN ABUJA MARKET Photo: A. Jacobson FIGURE 3.6: GENERATORS LINE THE STREET IN A MARKET IN ABUJA Photo: A. Jacobson 10 Sometimes when one is spoiled I take it to the mechanic. The marketplace has begun to respond to the emerging When the second one spoils I take it to the mechanic.” opportunity presented by reliable, economic, safer, and quieter solar and storage options. An investment prospec- Air pollution and the cacophony of ambient noise from tus for Nigeria’s Solar Energy for All (SE4ALL) efforts generators is top of mind for people who live with them described a pipeline of over 20 projects incorporating clean as well. One shop owner interviewed in Abuja explained energy. The description for one of them crystallizes the that, “Everything about generators is not good. Because opportunity to replace burdensome generators with solar16: number one, noise! … You cannot hear well anywhere. … The smoke causes a lot of sickness in the body. It is not good for human beings.”15 “Over reliance on gasoline generators and its attendant high cost of maintenance leads to the failure of many small scale enterprise (SSE) start-ups in Nigeria. It also leads to low return on investment for those with forbear- ance to survive among these enterprises. It also has negative impacts on the work environment in terms of noise and pollution, contributing to climate change due to CO2 emissions. This is despite the fact that their quantum [of] energy demand can be met by an alternative low cost source of energy— Solar PV as the most feasible.” —Project Description from SE4ALL Prospectus Results THE GLOBAL FLEET OF BACKUP GENERATORS Fleet Size & Composition The global fleet of BUGS is substantial and underscores the potential burden resulting from poor service quality. We estimate that 25 million generators (90 percent UI: 10 to 40 million units) were deployed in 2016 within developing countries (Figure 4.1).17 Nineteen million units, or 75 percent of the global fleet, are operated at sites with grid connection, reflecting the fact that the need for generators often results from weak or broken grids rather than a lack of grid connection. The global backup generator fleet is dominated in numbers by small gasoline and diesel generating units that provide service for loads less than 60 kW. Nearly 20 million small gasoline generators are currently deployed across modeled countries, accounting for over three quarters of the global fleet. Five million small diesel generators (< 60 kW) are currently deployed, accounting for 20 percent of the global fleet and the majority of diesel backup units. Medium (60 to 300 kW) and Large (> 300 kW) sized diesel gen- erators together account for around 2 percent of the global fleet and 10 percent of diesel generating sets (0.5 million units). The largest regional fleets exist in South Asia (3.4 million), Sub-Saharan Africa (6.5 million), and the Middle East and North Africa (5.3 million), with generator compositions similar to that of the fleet across all modeled countries (Figure 4.2). FIGURE 4.1: BACKUP GENERATOR FLEET COUNT ESTIMATES FOR 2016 ACROSS ALL MODELED COUNTRIES 12 FIGURE 4.2: BACKUP GENERATOR SIZES BY REGION AND SIZE CLASSIFICATION Small Petrol Medium Diesel Generator Classification Small Diesel Large Diesel 10.0 25 Fleet size of BUGS (Millions) 20 7.5 15 5.0 10 2.5 5 0 0.0 Northern Africa Middle Africa Western Africa Micronesia Central America Southern Africa South America Western Asia Melanesia Caribbean Central Asia Eastern Africa Southern Asia Polynesia South−eastern Asia Eastern Asia All Modeled Countries Error bars correspond to the 90 percent uncertainty interval Figure 4.3 shows the number of generators per 100 people (90 percent UI: 550-1300 GW) power plants. Considering in the ten low- and middle-income countries (LMICs) only LMICs, the total fleet capacity is 350 GW (90 percent with the largest total fleets. Within this group, there is UI: 220 to 530 GW), a 22 percent reduction. This change is one generator for every 165 people (30 households). In largely attributed to the exclusion of seven countries in the Nigeria, which has one of the largest fleets at three million Middle East with particularly large fleets. deployed units, there is one generator for every 60 people (12 households). It is important to note that while fleet A small number of countries in Africa and Asia account size is an important component for assessing the result- for most of the installed capacity of backup generators. ing impacts of generator operation, it does not necessar- The twelve countries19 with the largest fleet capacities ily reflect populations’ reliance on them or their resulting account for 40 to 60 percent of all backup generating impacts in that area. capacity across modeled countries; the top thirty-two (20 percent) modeled countries account between 60 and Installed Fleet Capacity 90 percent of the total backup capacity. Based on central estimates, Sub-Saharan Africa accounts for roughly We estimate that BUGS account for 450 GW (90 percent 20 percent of the population living in countries modeled, UI: 275 to 650 GW) of installed generating capacity across but 25 percent of total installed capacity—the largest frac- modeled countries (Figures 4.4 and 4.5). By comparison, tion of any single region. East and South Asia combined the capacity of a typical coal-fired power plant is 0.5 GW (excluding China) account for 50 percent of the popula- (500 MW),18 making the capacity of generator fleets cur- tion living in countries modeled, but 36 percent of the rently deployed in developing countries equivalent to 900 total installed capacity. Among LMICs with the largest 13 FIGURE 4.3: PREVALENCE RATE OF BUGS IN THE TEN LOW- OR MIDDLE-INCOME COUNTRIES WITH THE LARGEST FLEETS, EXPRESSED AS GENERATOR UNITS PER 100 PEOPLE East Asia & Pacific Middle East & North Africa Sub−Saharan Africa Region Latin America & Caribbean South Asia 4 Generators per 100 people 3 2 1 0 South Africa Angola Venezuela Nigeria Iraq Viet Nam Indonesia Egypt Philippines India Error bars correspond to the 90 percent uncertainty interval. Note that direct drive generators for agricultural and industrial applications are not included as part of generator fleet size estimates. fleet capacities, India, Angola, Indonesia, Nigeria, and the Energy Generation Philippines account for 10 percent, 9 percent, 8 percent, 5 percent, and 4 percent of capacity across all modeled BUGS provide 130 terawatt hours (TWh) (90 percent countries, respectively. UI: 68 to 260 TWh) of energy service per annum across modeled countries (Figures 4.6 and 4.7). By comparison, Accounting for installed capacity of generators on the a typical coal-fired power plant generates 3 TWh21 in a power grid indicates that BUGS make up a significant typical year, making the service provided by BUGS equiva- fraction of electricity generating capacity in developing lent to that of 43 (90 percent UI: 23 to 87) power plants. countries. Across all modeled countries, backup generator These energy services are distributed across a range of capacity is equivalent to 27 percent (90 percent UI: countries and regions, not just in areas with the poorest 18 percent, 40 percent) of the capacity of power plants on grid reliability or largest populations and generator fleets. the grid, and accounts for 22 percent (90 percent UI: Considering only LMICs, annual generation is 120 TWh 15 percent, 29 percent) of total generating capacity— (90 percent UI: 63 to 234 TWh). This modest 8 percent grid and backup capacities combined.20 In Sub-Saharan change to total generation relative to the larger 22 percent Africa, the backup capacity is roughly equal to that of change observed for installed capacity is indicative of the power plants on the grid; excluding South Africa, installed low utilization rates (reliable grids) of several high-income backup capacity is twice that of the grid. countries with substantial backup fleets, primarily in the Middle East. 14 FIGURE 4.4: INSTALLED CAPACITY OF BUGS ACROSS ALL MODELED COUNTRIES FIGURE 4.5: INSTALLED CAPACITY OF BUGS ACROSS ALL MODELED COUNTRIES BY REGION AND GENERATOR SIZE CLASSIFICATION Small Petrol Medium Diesel Generator Classification Small Diesel Large Diesel 125 400 100 Installed Capacity of BUGS (GW) 300 75 200 50 100 25 0 0 Western Asia Southern Asia South−eastern Asia South America Middle Africa Northern Africa Western Africa Eastern Africa Caribbean Central America Southern Africa Central Asia Eastern Asia Melanesia Polynesia Micronesia All Modeled Countries Error bars correspond to a 90 percent uncertainty interval. 15 FIGURE 4.6: ENERGY GENERATION BY BUGS ACROSS ALL MODELED COUNTRIES FIGURE 4.7: ENERGY GENERATION FROM BUGS ACROSS ALL MODELED COUNTRIES BY REGION AND GENERATOR SIZE CLASSIFICATION Small Petrol Medium Diesel Generator Classification Small Diesel Large Diesel Generated Energy from BUGS (TWh/Yr) 60 100 40 50 20 0 0 All Modeled Countries Southern Asia Western Africa Western Asia South America Eastern Africa South−eastern Asia Middle Africa Northern Africa Caribbean Central America Southern Africa Central Asia Melanesia Eastern Asia Polynesia Micronesia Error bars correspond to a 90 percent uncertainty interval 16 FIGURE 4.8: GENERATION FROM BUGS AS A PORTION OF GRID GENERATION (RATIO) BY REGION 0.4 Portion of Grid Energy Generation Generated Energy from BUGS as a Bars scaled horizontally by population 0.3 0.2 0.1 0.0 Western Africa Middle Africa Caribbean Eastern Africa Southern Asia Western Asia Northern Africa South America South−eastern Asia Southern Africa Central America Central Asia Eastern Asia The horizontal width of bars is scaled based on regional populations. Energy services (generated energy) from backup genera- of the degree to which various populations are dependent tion are heavily concentrated within several countries in on backup sources of electricity (Figure 4.8). The impact Africa and to a lesser extent Asia. The five countries with of poor grid reliability is particularly pronounced across the most generation account for 50 to 60 percent of all Sub-Saharan Africa, where the energy service provided backup generator service; fifteen (9 percent) of the 167 from BUGS is equal to 11 percent (90 percent UI: 6 to 21 modeled countries account for 70 to 80 percent of the percent) of that from the grid.22 Western Africa is among total service provided by BUGS. Using central estimates, the most affected, where the energy generated each year Sub-Saharan Africa alone accounts for 30 percent of total from backup generator sets is equivalent to 40 percent backup generation, the most of any single region, but that of the grid. roughly 20 percent of the population living in countries modeled. East and South Asia, excluding China, together Fuel Consumption account for roughly the same fraction of total backup gen- 55 billion liters (90 percent UI: 25 to 110 liters) of eration as Sub-Saharan Africa, but 50 percent of the popu- diesel and gasoline are consumed annually by BUGS lation living in countries modeled. Among the LMICs, (Figure 4.9). Diesel accounts for the majority of total Nigeria, India, Iraq, Pakistan, and Venezuela account for consumption, at 38 billion liters per year (70 percent). 16 percent, 15 percent, 11 percent, 9 percent, and 4 per- Gasoline consumption is slightly less than half that of die- cent of all backup energy service from generators. sel at 17 billion liters per year (30 percent). Although gas- A comparison of the amount of generated energy provided oline generating units outnumber diesel units in the fleet from BUGS to service from the grid provides a better sense by approximately three to one, the maximum capacity of a 17 FIGURE 4.9: ANNUAL GASOLINE AND DIESEL FUEL USED IN BUGS BY REGION AND GENERATOR SIZE CATEGORY Small Petrol Medium Diesel Generator Classification Small Diesel Large Diesel 40 20 Fuel Usage (Billion L/Yr) 20 10 0 0 All Modeled Countries Southern Asia Western Africa Western Asia South America South−eastern Asia Eastern Africa Middle Africa Northern Africa Central America Caribbean Southern Africa Central Asia Melanesia Eastern Asia Polynesia Micronesia Error bars correspond to a 90 percent uncertainty interval. gasoline generator is considerably lower than the average for nearly 15 percent of total gasoline consumption and diesel generator, reducing fuel requirements. The distribu- 22 percent of total diesel consumption. Several countries tion of fuel consumption across regions closely mirrors in South Asia also have meaningful fractions of diesel and that of generation, with Southern Asia and Western Africa gasoline fuel being used in generators, including Pakistan accounting for the greatest portions, at 26 percent (14 bil- (20 percent), Bangladesh (22 percent), Nepal (9 percent), lion liters) and 22 percent (12 billion liters), respectively. and India (4 percent). Small sized diesel and gasoline BUGS account for roughly Figure 4.11 shows generator fuel consumption as a per- two thirds of all diesel or gasoline fuel (35 billion liters) centage of transportation sector demand, the single larg- consumed for backup generation across modeled coun- est consuming sector in all countries and regions. In the tries. In Western Africa, where the fleet and operation time absence of detailed accounting of fossil fuel use, as is the of gasoline generators is especially high, gasoline accounts case in many LMICs, it is often assumed that nearly all for half of all fossil fuel consumed for backup electricity fossil fuel is used for transportation. Our results reveal, generation—nearly five times the fraction of other regions however, that in areas with weak and failing grids, demand in Sub-Saharan Africa (excluding Southern Africa) and for BUGS are comparable to that of leading sectors with more than three times that of South Asia. respect to fossil fuel demand. In many locations, including Sub-Saharan Africa and several countries in South Asia, Powering BUGS accounts for a significant portion of the quantity of fuel required for generators is upwards of total fossil fuel demand in several regions and countries 20 percent of the amount of diesel used for transportation, (Figure 4.10). In Sub-Saharan Africa, generators account and upwards of 10 percent of the amount of gasoline. 18 FIGURE 4.10: DIESEL AND GASOLINE CONSUMPTION FOR POWERING BUGS AS A PERCENTAGE OF TOTAL FUEL CONSUMPTION Percent of Fuel Usage Atributed to Generators 0 5 10 15 20 Sub−Saharan Africa Pakistan Korea (North) Bangladesh Myanmar Afghanistan (Islamic Republic of) Middle East Nepal Chile Caribbean Venezuela Cambodia Central America Mongolia Vietnam India Republic of South Africa Philippines Georgia Indonesia North Africa Other Former USSR Kyrgyzstan Argentina Azerbaijan Egypt Kazakhstan Colombia Uruguay Paraguay Israel Cyprus Sri Lanka Turkey Laos Bolivia Mexico Malaysia Ecuador Brunei Armenia Taiwan Bhutan Iran Petrol Saudi Arabia Peru Diesel Brazil Fuel estimates were compared to IEA statistics so that the overall energy use is consistent with IEA. Regional and country classifications are based on those used in the GAINS model. 19 FIGURE 4.11: BUGS FUEL CONSUMPTION AS A PERCENTAGE OF FUEL CONSUMED IN THE TRANSPORTATION SECTOR Bugs as a Percentage of Transportation Fuel Usage 0 10 20 30 40 Sub−Saharan Africa Bangladesh Pakistan Korea (North) Myanmar Venezuela Middle East Caribbean Afghanistan (Islamic Republic of) Chile Nepal Mongolia Central America Cambodia India Kazakhstan Philippines Kyrgyzstan Vietnam Other Former USSR Georgia Republic of South Africa Indonesia North Africa Argentina Azerbaijan Uruguay Egypt Colombia Cyprus Laos Paraguay Israel Sri Lanka Bolivia Mexico Turkey Ecuador Bhutan Malaysia Brunei Taiwan Saudi Arabia Armenia Petrol Iran Peru Diesel Brazil Fuel estimates were compared to IEA statistics so that the overall energy use is consistent with IEA. Regional and country classifications are based on those used in the GAINS model. 20 THE ECONOMIC COSTS OF with a replacement cost of $19 billion. Medium sized BACKUP GENERATORS diesel units comprise the smallest fraction, at $4.9 billion. It is important to note that although small gasoline and Capital investment small diesel are valued similarly, small gasoline generators are typically less robust and require replacement more fre- Over 1.2 million generators were transferred to develop- quently than diesel units. ing countries through international trade in 2016, with a total value of $5.3 billion. From 2011 to 2016, import Fuel Related Costs values totaled $45 billion, averaging $9 billion per year over this time period.23 Diesel generating units accounted Expenditures on fuel for BUGS is estimated at $40 bil- for only 25 percent of total units sold, but 80 percent of lion per year, or eight times the annual investment in the total import value in 2016. We estimate the replacement generators themselves in 2016. Figure 4.13 shows how value of the generator fleet across all modeled countries to there is a vast range in the marginal fuel cost of backup be approximately $70 billion.24 These estimates are before generator operation, from $0.20 to $0.50 per kWh. These accounting for local taxes, duties, and distribution costs. differences are mainly due to differences in the retail cost of gasoline and diesel, but also include variations in the Figure 4.12 shows the estimated value of backup genera- makeup of generator fleets (e.g., generator types, capac- tor fleets across modeled regions, assuming average 2016 ity) and assumptions about the part-load efficiency and unit costs. Small gasoline ($25 billion) and small diesel capacity factors of generators during operation. It is ($22 billion) dominate globally and across all regions with important to note that marginal costs reported here are for the largest fleets. Large diesel units are not far behind, the cost of the fuel alone, and do not consider capital or FIGURE 4.12: REPLACEMENT COST OF BACKUP GENERATOR FLEETS Small Petrol Medium Diesel Generator Classification Small Diesel Large Diesel 20 Fleet Replacement Cost ($ Billions) 60 15 40 10 20 5 0 0 Western Asia Southern Asia South−eastern Asia South America Northern Africa Western Africa Middle Africa Eastern Africa Central America Caribbean Southern Africa Central Asia Eastern Asia Melanesia Polynesia Micronesia All Modeled Countries Error bars correspond to the 90 percent uncertainty interval of totals. 21 FIGURE 4.13: ESTIMATED SERVICE COSTS FOR BUGS BASED ON FUEL PRICES ALONE, WITH COMPARISONS TO THE AVERAGE COST OF ELECTRICITY FROM UTILITY GRIDS Marginal Retail Cost of Electricity (USD/kWh) 0.0 0.1 0.2 0.3 0.4 0.5 Eastern Asia Southern Africa Middle Africa Eastern Africa South−eastern Asia Service Caribbean Type Southern Asia BUGS Operation Western Africa Central America Utility Grid Western Asia Central Asia South America Northern Africa The midpoint estimate of backup generator marginal cost is shown. maintenance costs, or the external costs of pollutant emis- of strategies that support electricity service in Africa are sions and other impacts on welfare. just as much or more a story of reducing the reliance on BUGS with distributed systems as it is one of providing In every region the grid is lower cost than BUGS. These clean energy through grid-serving renewables. marginal costs of service provide a benchmark against which solar + storage and other strategies could compete. Consumption Subsidies During blackouts, service from solar + storage, for exam- We estimate that the cost of subsidizing fuel used in BUGS ple, would avoid the marginal fuel cost of backup genera- was $1.6 billion (90 percent UI: $0.8 to $3.2 billion) in tor use. During normal operation, the generation from 2016. Like the fleet characteristics discussed in previous onsite solar could also offset retail electricity consumption. sections, much of the subsidy cost is concentrated in a few countries with large unit subsidies. While modest in com- Another view on the cost of fuel for BUGS versus grid parison to other costs of backup generation, it is impor- service is to compare overall spending on each category of tant to consider that the consumption subsidies reported service, showing the overall scale of each electricity access here are before adding production subsidies and external pathway. Figure 4.14 shows how these two energy sources costs of pollutant emissions on health and climate, which compare across regions. In much of Asia and the Americas, can be considerable. A recent valuation of global fossil there are large and heavily relied upon utility grids that fuel subsidies conducted by the International Monetary provide the vast majority of energy service. Thus, spending Fund (IMF) found pollutant impacts on climate and air on grid-based power is dominant in these regions, albeit quality to account for over half of the total cost.25 Given with significant spending on BUGS as well, between the highly concentrated nature of generator fleet deploy- $1 billion and $10 billion per year. In Africa, however, the ments, it is reasonable that external costs would have a scale of spending on BUGS is similar to the grid. Western similarly large contribution if valued. In effect, the true Africa spends approximately the same amount on genera- cost of fossil fuel use for BUGS could be roughly twice the tor fuels as it does for grid electricity, and in specific coun- $40 billion mentioned above, if we account for the pollut- tries (such as Nigeria) there is more spending on generator ant impacts discussed in the next section. fuel than on the grid. The implication is that deployment 22 FIGURE 4.14: TOTAL SPENDING BY RETAIL CUSTOMERS ON FUEL FOR BUGS AND UTILITY GRID SERVICE (Billion USD / year) 0 100 200 300 Spending Southern Asia South America Category South−eastern Asia BUGS Fuel Eastern Asia Western Asia Utility Grid Central America Northern Africa Southern Africa Africa Details 0 5 10 15 Western Africa Northern Africa Eastern Africa Southern Africa Central Asia Western Africa Caribbean Eastern Africa Middle Africa Middle Africa The total spending on utility grid service is shown for comparison to BUGS. The spending on fuel for generators includes a 90 percent uncertainty interval error bar. POLLUTANT EMISSIONS Our results reveal that BUGS are a significant source of pollutant emissions in many countries and regions. Like the emissions from the engines of cars and motorcy- Measuring generator performance and impacts in areas cles, the “tailpipe” emissions of BUGS contain thousands with frequently operated fleets could reveal they are an of chemicals, including many that impact human health even more significant local source of air pollution, and and the environment. A key objective of this study was mitigation opportunity, than indicated here. One implica- to establish the most comprehensive coverage to date on tion of our work is an increased recognition of BUGS as a the current (baseline) emissions from BUGS in developing source of pollutant emissions in most developing countries countries based upon the characteristics of their fleets and and regions of the world. the energy service they provide. 23 High Priority Opportunity for Pollution Reduction Air pollution is a leading cause of premature studies examining various parts of the African death and disease in many countries. This is continent have reported that BUGS are a sig- especially true in developing countries, where nificant and growing source of NOx emissions, exposure to particulate matter (PM) air pollution and an important contributor to ozone-forming was responsible for 2.5 million premature deaths pollutants.30 An accounting of BUGS emissions in 2016, with an additional 400 thousand prema- based on existing country and regional estimates ture deaths resulting from exposure to ground- of fuel consumption found BUGS to be a modest level ozone.26 Many of the same pollutants that contributor to pollutant emissions globally, but a harm health also contribute to climate change potentially important source of local black carbon and can have adverse effects on ecosystems. A (BC) and NOx emissions, especially in develop- critical step toward mitigating these pollutant ing countries.31 Two previous reports from the impacts is identifying and controlling important World Bank found BUGS to be a modest contribu- pollutant sources. Despite the pervasive use tor to black carbon (BC) emissions in Nigeria and of BUGS across developing countries, a limited the Kathmandu Valley of Nepal, but noted that understanding of their contribution to local and limited data were available on the size and char- regional pollutant emissions persists and hampers acteristics of generators in the fleet. Nearly all the ability to assess the benefits of strategies that existing studies on BUGS impacts have focused reduce their operations. As a source of pollution on diesel generators only, and estimated genera- that has been poorly understood to date, the tor operations by assuming power plants on the global and local burdens resulting from generator grid represent total electricity demand, or do not emissions represent unaccounted costs of opera- explicitly connect the energy services of BUGS to tion, and eliminating them provides extended their emission impacts. value from programs that mitigate generator use, beyond monetary savings from avoided fuel and Several of the pollutants in generator emissions other expenses. are of particular importance given the robust evidence of their effects on health and the envi- Existing evidence suggests that BUGS can be a ronment. The emissions from BUGS contribute, potentially important source of local and regional either directly or indirectly, to all pollutants found air pollution in developing countries. Compared on major priority pollutant lists. The World Health to power plants on the grid, BUGS can emit sev- Organization (WHO) recognizes four pollutants eral times more pollution from each unit of fuel relevant to outdoor air pollution: particulate burned and unit of electricity delivered.27 When matter, ozone (O3), nitrogen dioxide (NO2), and deployed at scale, as they often are in weak-grid sulfur dioxide (SO2), all of which are directly emit- areas, BUGS have been found to be an impor- ted or formed from pollutants found in generator 32 tant source of local and regional air pollutants. A exhaust fumes. Table 4.1 provides a brief sum- recent assessment of sources of pollution in 20 mary of several important pollutants associated cities across India indicate that BUGS account with backup generator operation. for 2 to 6 percent of total ambient PM2.5,28 while a separate study of Indian cities found BUGS to The emissions from BUGS contribute, either account for between 8 and 28 percent of PM2.5 directly or indirectly, to all pollutants found on in the residential areas they examined.29 Several major priority pollutant lists. 24 TABLE 4.1: A SUMMARY OF HEALTH AND CLIMATE RELEVANT POLLUTANTS ASSOCIATED WITH THE EMISSIONS OF BACKUP DIESEL AND GASOLINE GENERATORS Major Impact Estimated in Pollutant Areas This Study Description Carbon Dioxide Environment Yes CO2 is the single most important contributor to climate change. [CO2] Particulate Health, Yes PM2.5 is perhaps the best pollutant indicator for health risk; combustion Matter, Environment of fossil fuels, like diesel, is a major source globally, especially in urban Black Carbon, populations. Once in the atmosphere, PM2.5 goes on to affect air quality, Organic Carbon while black (BC) and organic carbon (OC), components of PM2.5, contribute [PM2.5, BC, OC] to climate impacts. The importance of BUGS as a source of PM2.5 health risk is dependent on other sources of PM2.5 nearby. In cities, for example, vehicle emissions are a dominant source. Black carbon (BC) and organic carbon (OC), components of particulate matter, in addition to health risks, absorb and reflect solar radiation leading to climate impacts. Nitrogen Health, Yes Most NOX emissions come from the combustion of fossil fuels and are Oxides Environment typically associated with the vehicle and energy generation sources. Once [NOX] emitted, NOX form pollutants that damage health (i.e., ozone, particles) and the ecosystem (i.e., acid rain, ozone). Exposure to NOX has been associated with numerous respiratory illnesses. High levels of nitrogen dioxide are also harmful to vegetation—damaging foliage, decreasing growth and reducing crop yields. NOX are ozone precursors, reacting with other pollutants in the air to form potentially harmful ground level ozone. Sulfur Dioxide Health, Yes SO2 is a pollutant emitted from burning fuels that contain sulfur, such as coal, [SO2] Environment diesel, and kerosene. Inhaling SO2 can exacerbate respiratory diseases and can also form small particles, which contribute to PM exposure. In the atmosphere, SO2 can contribute to acid rain and reduce visibility. Carbon Health No CO is the leading cause of accidental poisonings globally. Carbon monoxide Monoxide poisoning is a significant threat when generators are used inside or too close [CO] to occupied buildings.33 This is especially true for smaller two-stroke generators often used by homes and small businesses.34,35 CO is an ozone precursor, reacting with other pollutants in the air to form potentially harmful ground level ozone. This occurs close to the site of emission. It does not have any significant environmental effects at a global level. Non-Methane Health, No NMVOCs are a large group of chemical compounds that easily evaporate Volatile Environment into the surrounding air. Exposure to some NMVOCs such as benzene, Organic formaldehyde, and acetone can pose direct health risks. NMVOCs are also Compounds ozone precursors, reacting with other pollutants in the air to form potentially [NMVOC] harmful ground level ozone. The emissions of NMVOCs from generators are not reported here or well documented, but remain important, especially in areas with large numbers of gasoline-fueled generators. Formation of ozone in the lower atmosphere (ground-level ozone) occurs from reactions between NOx (a component of NOx ), carbon monoxide (CO), and volatile organic compounds (VOCs) in the presence of ultraviolet light (UV). Unlike the ozone in the upper atmosphere, which protects from harmful Ozone Health, No UV radiation, ozone exposure in the air we breathe can lead to increased [O3] Environment risk of various respiratory diseases, such as asthma, and cause abnormal lung development in children. We do not model the contribution of BUGS to ozone formation, but it has been identified as a potentially important source in Africa and particularly Nigeria.36 25 BUGS as Significant Source of Pollution the type of energy services they provide (i.e., transporta- tion, power). Figure 4.17 presents estimates of absolute Backup generator fleets contribute significantly to some pollutant emissions across sectors and regions. pollutant emissions, but their importance for reducing impacts on population health and climate varies by loca- The large contribution of NOx emissions by BUGS stands tion. Figure 4.15 presents BUGS emissions as a percent- out among other pollutants examined here. Across all age of all emissions in three large global regions. In each modeled countries, 1500 kilotons (kt) of NOx are emitted region, total emissions account for the contributions of as a result of backup generation each year. In Africa, this many sources of pollution, including cars, trucks, power accounts for 7 percent of total NOx emissions annually, plants, manufacturing plants, wildfires, stoves used for but significantly less in South Asia where vehicle fleets are cooking, heating, and fuel-based lighting, to name just a much larger. In Sub-Saharan Africa, generators account few. As a single source, generators account for a significant for 15 percent of total NOx emissions—equivalent to 35 fraction of some pollutants but differences in the charac- percent of the NOx from the entire transportation sector. It teristics of other pollutant sources affecting a region leads also accounts for 65 percent of NOx emitted from power to variation in BUGS contributions in the same region, and generation in Africa, and more than 10 percent in Asia across different regions. Overall, the contribution of BUGS and the Americas. to pollutant emissions is greatest on the African continent. Regional emissions of PM2.5 and other aerosol species from Comparing the emissions from BUGS to those of other BUGS are modest in comparison to that of several domi- sources and energy sectors can be useful for determining nant sectors, but may still be an important source of local their importance relative to other mitigation opportuni- pollution. Across all modeled countries, the annual PM2.5 ties. Figure 4.16 presents BUGS emissions as a portion of emission rate for BUGS is estimated to be 1,000 kt/year, that from major energy sectors, by pollutant. Sectors are and 400 and 300 kt/year for BC and OC, respectively. aggregates of many sources of pollution, often grouped by Across regions shown in Figure 4.16, BUGS contribute 20 FIGURE 4.15: CONTRIBUTION OF BUGS TO REGIONAL EMISSIONS Africa Americas Asia Percentage of Total Emissions from BUGS 6 4 2 0 PMOC PMOC PMOC PM2.5 PMBC PM2.5 PMBC PM2.5 PMBC NOx NOx NOx SO2 SO2 SO2 CO2 CO2 CO2 Emission Species 26 to 75 kt of PM2.5 emissions per year—equivalent to 5 to results suggest. Several studies of pollutant source contri- 15 percent of transportation emissions. As a major source butions in Indian cities found generators to account for as of energy generation, BUGS account for 10 to 75 percent much as 28 percent of local PM2.5 pollution.37 of PM2.5 from the power sector. In urban areas, where the use of solid fuels in homes is typically much lower than The fact that emissions from BUGS, an individual source indicated by national averages—and where generators are within the Power Sector, can be compared to the emissions most prevalent—BUGS likely account for a larger fraction of entire sectors is indicative of their likely importance as a of local particulate emissions and air pollution than our pollutant source in some countries. Also, while examining FIGURE 4.16: EMISSIONS FROM BUGS EXPRESSED AS A FRACTION OF TOTAL SECTORAL EMISSIONS IN 2016 Africa Americas Asia 100 75 Power 50 25 0 100 75 Industry 50 BUGS Emissions as a Percentage of Major Sectors 25 0 100 75 Transport 50 25 0 100 Commercial 75 Residential 50 25 0 PM2.5 PM2.5 PM2.5 NOx NOx NOx SO2 SO2 SO2 CO2 CO2 CO2 Emission Species Note that BUGS are an emitting source within the Power Sector, so percentages are interpreted as the fraction of total Power Sector emissions attributable to BUGS. 27 a pollutant source at coarse geographic scales is useful as gasoline-fueled generators, notably in Nigeria and other a first-approximation of its importance, it can dilute the parts of West Africa. Measurements of two stroke engines source’s contribution to local burdens, especially when its and generators have been reported to emit as much as use is concentrated in small areas. Our results suggest that 40 percent38 of their fuel as unburned vapor (VOCs) and BUGS are one such source, given that fleets are predomi- can generate acutely dangerous concentrations of carbon nantly deployed in urban areas with grid access. monoxide.39 These high emissions of both NMVOCs, in combination with their NOx emissions, make generators Likely important but not reported here are the emissions a potentially potent source for promoting ground-level of non-methane volatile organic compounds (NMVOCs) ozone formation,40 a pollutant associated with numerous and carbon monoxide. This is particularly relevant for respiratory diseases. populations employing large numbers of two-stroke FIGURE 4.17: EMISSIONS (MEGATONS (MT)/YEAR) FROM BUGS COMPARED TO SELECTED EMITTING SECTORS, FOR COMPARISON Africa Americas Asia 3000 2000 CO2 1000 0 10.0 7.5 SO2 5.0 2.5 0.0 12 9 NOX 6 Emissions (Mt) 3 0 2 PM2.5 1 0 0.4 0.3 PMBC 0.2 0.1 0.0 1.00 0.75 PMOC 0.50 0.25 0.00 BUGS Power Industry Transport Other BUGS Power Industry Transport Other BUGS Power Industry Transport Other Note that because BUGS are a source within the Power Sector, their contribution has been subtracted out of the sectoral total to avoid double counting. The residential sector is not shown for presentation purposes but is the dominant source in all regions for particulate matter species. 28 Table 4.20 summarizes the dimensions of environmental technologies based on lab tests, especially those affected and health risks we identified from generator emissions. by duty cycles and sensitive to poor maintenance, often Not all of these impact dimensions were directly modeled yield better performance indicators than those based on as part of our research effort, but they are mentioned as in-field measurements under typical usage conditions. they remain important issues. There are risks associated This has been true for several energy service technolo- with the pollutants presented, with some at a higher level gies (and major emissions sources) relevant to develop- of certainty than others due to a lack of good quality data ing countries, including cookstoves, fuel-based lighting, on BUGS emission characteristics. and automobiles. For some pollutants such as CO2 and NOX we expect our results are less sensitive to the lack of Implications of Data Gaps on Pollutant context-specific performance data given their formation Emissions and Impact Estimates mechanisms. Other pollutants, including PM2.5, BC, SO2, are likely more affected and so are probably conservatively Our work revealed that major gaps exist in the under- low. Poor fuel quality can also negatively affect generator standing of backup generator performance in developing performance, and while anecdotal accounts of fuel adul- countries, requiring us to apply assumptions that likely teration are common, we are unable to account for the bias at least some of our pollutant estimates low. In the effect of this on our estimates. absence of data from generators in developing country fleets, we often relied upon lab-based performance data The spatial resolution of our estimates may understate the for new generators tested in industrialized countries. These importance of BUGS as a local source of air pollution. In devices are likely better performing than the typical unit an effort to provide coverage and consistency across as used by residents in developing countries to power their many countries as possible, we focused our analysis on homes and businesses. Moreover, performance of energy providing national and regional emission estimates. Our TABLE 4.20: SUMMARY OF BUGS EMISSIONS FOR POLLUTANTS MODELED IN THIS STUDY Potential Scale of Impact Pollutant from BUGS Data Quality Impact Summary Carbon Dioxide Modest Good We estimate that 100 (Mt) of CO2 are emitted each year from [CO2 ] contributor at generators in modeled countries. In Sub-Saharan Africa, the CO2 national and emitted by generators is equivalent to 20 percent of the CO2 regional scales emissions from vehicles in the region. Particulate Matter, Modest Low In Sub-Saharan Africa, emissions of PM2.5 from BUGS is equivalent to Black Carbon, contributor at 35 percent of the PM 2.5 emitted from vehicles. It also contributes the Organic Carbon national and Limited data majority of PM2.5 , BC, and OC from the Power Sector in Sub-Saharan [PM2.5 , BC, OC] regional scales. on emission Africa. Many BUGS are used near where people live and work and in Potentially characteristics of densely populated (urban) areas, meaning that a larger fraction of significant source generators used in what BUGS emit is likely to be inhaled by people. at local scale key regions. Nitrogen Oxides Potentially a Good BUGS are a potentially significant source of NOx in some countries [NO X ] significant source and regions. We estimate that BUGS account for around 5 percent of at national and Limited data all NOx emissions in developing countries, 7 percent in Africa, and 15 local scales. on emission percent in Sub-Saharan Africa. characteristics of generators used in key regions. Sulfur Dioxide Minor source Low Overall emissions of SO2 from BUGS are minor at national and [SO2 ] at national and regional scales. Emissions from generators are equivalent to 50 regional scales; Limited data on percent of emissions from transportation in Sub-Saharan Africa, potentially actual fuel quality. but account for less than 0.5 percent of total emissions in Africa, important source Currently assumes Asia, and the Americas. BUGS may be an important local driver of at local scale. local fuel quality exposure, given that major emitting sources tend to exist further standards. from densely populated areas. 29 fleet analysis, however, indicated that the use of BUGS is discrepancies, survey data were not always available and highly localized, even within a country, being predomi- not all sectors are represented in surveys. It is expected nantly used in urban areas with grid connections. Other that in cases with significant untracked trade or domestic major sources of pollution are also localized, but not nec- production, these adjustments are unlikely to fully capture essarily in the same direction as BUGS. In most developing the true volume and probably result in low fleet counts. countries, for example, household air pollution accounts for a major and often dominant fraction of PM2.5. The use For example, in Nigeria, where we highlighted a particu- of solid fuel, however, is more prevalent in rural and off- larly well-known generator market in the introductory grid communities, and typically less so in areas where gen- section to this report, there is reason to believe that not erators are widely deployed. In these instances, generators all generator trade is captured in the global trade data. would likely account for a greater proportion of air pol- This would lead to undercounting and we attempted to lutants than indicated by national or regional aggregates. adjust fleet counts using analyses of nationally represen- Finally, generators are often installed in densely populated tative household and business surveys that are publicly areas, and in close proximity to homes and businesses,41 available. With these adjustments, we estimate that there increasing the likelihood that the pollution they emit is are 2.8 million residential and 210 thousand commercial eventually inhaled by people. These finer-level assessments sites with actively used generators, totaling 13 GW overall were beyond the scope of this work, but our results under- (two times the installed capacity of power plants on the score the importance of context specific examinations into grid), with $5.4 billion in annual spending on fuel alone. the operational characteristics and impacts of BUGS at a Other estimates previously reported in the press (but not local (i.e., sub-national, city-level) scale. from well-described or in some cases any cited sources) report higher numbers of sites (e.g., 12 million active COUNTRY-LEVEL ACCURACY sites across the country42) and spending of “as much as” AND UNCERTAINTY $17 billion in the industrial sector alone.43 This may be a case where even the adjusted model approach does not The modeling framework we developed was designed to capture the full market, and/or the result of exaggerated accommodate widely available data on trade and house- or high-side estimates reported in press. The model did, hold and business surveys and may not be accurate at the however, identify that Nigeria is clearly a country with a country level where nuances of a local market and use significantly large fleet and higher-than-normal spending cases are not captured. This is the reason we chose and fuel consumption compared to many surrounding and to focus on regional averages in most of the results pre- other countries. Regardless, the Nigeria case highlights sented above. that country-level results from our study can be taken as indicative, and additional work to understand and engage It is possible that in some countries our estimates are in local contexts is important in advance of investments lower than reality, particularly in places with significant or engagement to address the market. For similar reasons, domestic generator manufacturing and/or untracked we expect that our estimates for India, Lebanon, Thailand, imported generators. While our combined approach of Brazil, and Malaysia are also conservatively low. estimating fleet sizes using import records and sectoral cross-sectional surveys should address some of these 30 Conclusion It is early days for replacing BUGS with clean energy technologies such as solar and storage, and this report is an attempt to identify and clarify the welfare and environmental opportunities that could accompany such transitions. Our characterization of BUGS fleet compositions, operations, and pollutant emissions enables greater understanding of the impacts of weak grids. Importantly, it reveals the level of avoidable environmental burdens that can begin to be addressed through actions that lead to a reduced reliance on BUGS. The global capacity of BUGS is immense—equivalent to 700 to 1000 large power plants, nearly double the capacity of generators powering the entire grid in India. In countries where the grids are especially poor, the energy service provided by BUGS rivals, and sometimes exceeds, power plants on national grids. With annual spending on fuel alone in excess of $40 billion, the heavy reliance on BUGS imposes significant costs on families, businesses, and governments. In Western Africa, there is more spent on the fuel for BUGS than on electricity from the grid each year. Much of the financial cost of BUGS opera- tions is driven by the staggering quantity of fossil fuel they consume, which in some regions of Africa is more than half that used by the entire transportation sector. BUGS contribute to emissions of health and climate damaging pollutants, sometimes significantly, even at regional scales. BUGS are predominantly deployed in grid-connected urban centers, and so likely impose greater impact to local air quality than indicated by our results, given the spatial scale of our analysis. Efforts to refine estimates of operational characteristics, performance, and impacts of BUGS through measurements in areas where they are widely deployed may reveal a significant opportunity to improve public health through replacement. As the cost of clean technologies continue to fall and the understanding of welfare impacts of BUGS improves, there is an emerging and significant value proposition to replace BUGS. The cost of generated electricity from diesel and petrol generators is not likely to fall dramatically due to any current or near- future technology improvements. As a result, the cost of clean technologies may have already reached a level of parity in some markets—or is not far off. Could distributed clean energy systems replace BUGS and even support the local distribution circuits or regional grids where they are installed? The answers to these and other critically important technology and policy questions will help accelerate clean transitions, but also needs to be informed by a better understanding of local conditions. Replacing BUGS belongs in the global conversation along with efforts to decarbonize electricity grids, transportation systems, and other sectors of the economy. The sector is far-reaching, and in some coun- tries with particularly poor grids accounts for significant financial burdens. Avoiding the emissions, health impacts, and operational efforts imposed by BUGS represents a potentially significant opportunity to improve the welfare of people who rely on them. 32 Appendix 1: Methodological Details GENERATOR BACKGROUND DETAILS Generator Power Ratings The power rating of generators is defined in terms of “apparent power” with units of “kilo-volt-amps” (kVA). These ratings are similar to the familiar “real” electrical power rating in “kilowatts” (kW), but with the important distinction that the kVA rating also accounts for the voltage-stabilizing “reactive power” that needs to be provided with generators in standalone operation. The idea behind reactive power support is that some loads (such as motors, power supplies, and others) have electrical characteristics where they do not just need real power (kW) but also require voltage sta- bilization to help balance the circuit. This additional work is called “reactive power,” and uses up some capacity of the generator. The combination of real and reactive power is what needs to be supported by a generator overall to serve loads with stable voltage; this is given in terms of kVA. For most buildings, the level of kVA required from a generator is between 1 and 1.5 times the sum total kW rating of the loads being served. Fuel Types Diesel fuel is typically used in generators designed to service larger (greater than 3-5 kW) loads. Diesel fueled generators are generally more efficient than gasoline generators for the same output level. They are sometimes called “compression ignition” generators because of the design of the engines, which take advantage of a property of diesel fuel where the fuel auto-ignites at a sufficiently high pressure. We distin- guish three sizes of diesel generators: small (< 75 kVA/60 kW), medium (75–375 kVA, 60–300 kW) and large (> 375 kVA/ > 300 kW). Previously reported estimates of the levelized cost of electricity (i.e., the average cost including buying the generator, fuel, and maintenance) is between 0.20 and 0.30 $/kWh for large diesel generators44 and $0.20 to $0.50 for smaller units.45 Gasoline (petrol) generators service small (less than 3–5 kW) loads, often providing just enough energy to run lights and basic appliances. They are generally less efficient and robust than diesel units, so rated capacities greater than 3 to 5 kW are not common. Gasoline units are powered by either two or four- stroke motors, with two stroke motors being far more affordable but poorer performing. A typical gasoline generator used throughout Nigeria, referred to locally as “I better pass my neighbor,” runs on a two-stroke engine with rated capacity of around 0.5 kW. Given the narrow range of capacities avail- able for gasoline generators, we do not distinguish size categories. Because of lower efficiency than diesel, gasoline generators have a higher levelized cost of electricity, around 0.60 $/kWh.46 Note on BUGS costs above vs. Solar: As of 2018, the levelized cost of rooftop PV has fallen to 0.20 $/ kWh, with the cost of delivered energy from PV+ storage at 0.40 to 0.70 $/kWh.47 The projections for future costs of PV and storage suggest continued progress toward a transition point where PV and storage 34 could effectively foreclose on the market for distributed Fuel and O&M Costs energy currently served by backup generators, depending on the marginal costs of their operations. BUGS provide energy services by burning fossil fuels to generate electricity. Expenditures on fuel are typically User Segments the dominant cost associated with their operation. The The loads and utilization characteristics of BUGS may regular maintenance and servicing of generators, espe- vary across user segments, affecting decisions around the cially for larger capacity units, also have associated costs. type and size of generator deployed. These application We combine estimates of fleet size and composition with characteristics may also be important in affecting market- estimates of runtime (from country-level SAIDI values) driving factors such as affordability and payback duration. and generator performance curves to estimate fuel con- Our modeling framework maintains flexibility to distin- sumption. Historic pump prices of fossil fuels are used to guish user segments in order to consider these factors. convert volumetric consumption of diesel or gasoline to costs. Operation and Maintenance (O&M) costs are not In all modeled countries, generator fleets are classified into included as a cost in our estimates, but can be considerable residential or commercial sectors, urban or rural, and on- in some BUGS applications—conservatively on the order or off-grid. National surveys often report ownership status of 10 to 20 percent of the fuel costs in most situations.50 of generators, urban/rural designation, and grid connec- tion status, but rarely collect characteristics of the genera- EXPENDITURE ON GRID VS. BUGS tor units needed to inform fleet disaggregation by type and size. In the absence of these data, a rough apportioning of When the grid is reliable, the main cost of BUGS is related the fleet to residential or commercial sectors is performed to capital investment and routine maintenance. In the using input from experts in major generator markets and regions we studied, the grid is very weak. Frequent opera- review of backup generator literature. User segments tions of generators mean that fuel costs can become a accounting for large portions of fleet deployments have significant or dominant expense, thus comparing expen- been identified. These include the telecom sector and off- ditures on BUGS versus the grid can provide a valuable shore diesel generators used on barges, for example. We point of context and is included in some presentations of do not explicitly examine these sectors in-depth here, but the results. A recent study of several countries in Africa our methodological approach does account for these units reported that the reliance on BUGS for electricity genera- in the electric generator fleet in each country/region. tion increases fossil fuel consumption (and associated soci- etal costs for fuel) by a factor of 1.5 to 1000 depending on COSTS OF BACKUP GENERATORS assumptions about local conditions and the capability of existing grid capacity to satisfy electricity demand.51 Our Capital Investments approach has a different method for estimating generator costs than that study, using import/export data and more Capital costs represent the cost of purchasing BUGS. Fleet granular information on fleets. Our estimates for expen- size estimates and import rates are used to examine capital ditures on the grid are based on the cost of electricity and costs of BUGS purchased each year and for valuing the the total energy generated, adjusted for transmission and replacement cost of the fleet. To estimate the size of fleets, distribution losses. we examine country-level import and export records of generators and combine this information with data on grid Subsidies reliability to approximate generator runtimes and corre- sponding lifetimes. Results on the prevalence of generator The use of BUGS to meet energy service needs is often ownership from an analysis of over 70 nationally represen- incentivized and enabled through government subsidies tative household48 and business surveys49 is used to adjust on fossil fuels. Despite the well-intentioned goals of many country-level fleet sizes, with regional adjustment factors subsidy schemes, they are often inefficient and incur direct applied when country-level surveys are not available. and indirect costs to users, governments, and the environ- ment. These subsidies make alternative pathways to elec- tricity services less competitive by creating artificially low 35 service costs for BUGS. Reducing reliance on generators pollution and accidental injuries. These impacts are likely can ease the burden of subsidies on government budgets, important, especially at a local scale, but were not exam- and removing or reducing subsidies could better signal ined as part of this study. Exposure to excessive noise con- the cost of backup generation to customers who may have tributes to the local burden of disease through increased other options. risk of heart disease, cognitive impairment in children, and loss of sleep, to name a few. A recent study by the World We apply the widely used “price-gap” approach52 to esti- Health Organization estimated that at least one million mate country-level consumption subsidies and the corre- life years are lost annually due to exposure to traffic noise sponding cost of subsidizing the fossil fuel used in BUGS. pollution in Western Europe.53 Anecdotal accounts of the When valuing fossil fuel subsidies, external costs from the noise pollution generated by BUGS is widely documented impact of combustion products on climate and air qual- in the gray literature, but no study that we are aware of ity add additional (and sometimes significant) cost. We do has examined the potential health implications on local or not report on the external cost of subsidies here, but our national populations. results provide the information necessary to perform these additional cost calculations. STUDY SCOPE AND METHODOLOGY Health & Environmental Hazards This study uses the best available global data sets on BUGS and the drivers of generator use to estimate backup BUGS are a potentially significant pollutant source, espe- fleet operations and pollutant emissions for 167 coun- cially at a local scale. In areas where they are deployed, tries. This work aims to establish the most comprehen- BUGS contribute to the emissions of health and climate sive understanding of the scale of backup generator fleet damaging pollution. However, it is often difficult to dif- deployment and operations in developing countries and ferentiate their contribution from that of cars, trucks, and their contribution to national and regional emissions. Our other technologies that burn the same fuels. As a result, approach attempts to reflect the mechanism by which grid the extent to which mechanisms that reduce their opera- quality affects reliance on BUGS, and in turn the impact of tions could contribute to achieving health and climate BUGS on economies and pollutant emissions. goals remains unclear. Geographic Scope To begin to address this impact gap, we estimate the con- This study characterized backup generator operations in tribution of BUGS to emissions of health and climate rel- 167 countries, representing 94 percent of the population evant pollutants. Our results are used to update a global living in low- and middle-income regions of the world, emissions inventory and comparisons to other pollutant excluding China. Figure 6.1 provides an overview of the sources and sectors performed at a national and regional countries modeled as part of this work and their regional scale. While this study does not explicitly examine con- categorizations, and Table 6.1 summarizes the regional tributions to outdoor air pollution and exposure, disease, populations. For most countries we applied a standard- or radiative forcing, it does establish the groundwork and ized approach for modeling the backup generator sector provides the necessary inputs for such assessments for based on globally available data sets. For Nigeria and most developing countries in the world. There is limited India (the top two markets in terms of total load served information on the emission characteristics of generators by generators) a more customized approach was taken used in developing countries, leading us to make assump- to improve user segmentation and improve the model tions that likely result in conservatively low emission fidelity. We exclude several developing countries from estimates for several pollutants of importance to health our analysis due to limited data from which to perform a and the environment. These data gaps and their implica- country-specific analysis when our standardized approach tions for our results are discussed in the main report and could not be applied. The most notable of these countries in Appendix 2. is China, which is a major producer of electric generating sets globally. There are also non-pollutant hazards that arise from the operation of BUGS, such as their contribution to noise 36 Country-level results are aggregated over several global provided by BUGS and the reliability of the grids they region classifications. A sub-classification of World Bank compensate for. The first-order modeling approach empha- regions is used whenever possible in order to provide sizes impacts that directly relate to the consumption of more granular representation of Sub-Saharan Africa and and expenditure on fuels and generating units, as deter- several parts of Asia. We use 2018 World Bank income mined by utilization characteristics. classifications to differentiate between all modeled coun- tries and low- and middle-income countries (LMICS) Global import/export trade data on generators and when presenting results for several backup generator fleet national surveys were used to estimate the number of characteristics. Fuel consumption and pollutant emissions generators used in 167 developing countries (fleet size), from generators are aggregated using country and regional classified by fuel type (i.e., diesel, gasoline) and their size classifications employed by IIASA’s Greenhouse Gas-Air (maximum power output) categories. In most countries Pollution Interaction and Synergies (GAINS) model.54 (except India and Nigeria) we did not attempt to account for domestically produced generators, which is a known Technological Scope source of conservative bias in our approach. The total Our analysis estimates fleet characteristics for diesel and duration of power outages (i.e., system average interrup- gasoline-fueled electric generating sets. We distinguish tion duration index, or SAIDI) were the basis for the hours three capacity (size) categories for diesel (compression of BUGS operation (runtime); this was combined with ignition) generators: small (< 75 kVA, < 60 kW), medium manufacturer data about their efficiency and assumptions (75–375 kVA, 60–300 kW) and large (> 375 kVA, > 300 about loading factor of generators to estimate energy gen- kW). All gasoline (spark ignition) generators are aggre- eration and fuel consumption. Fuel consumption results gated into single group. Our analysis is exclusive to elec- were used to update a widely used fuel and emissions tric generating sets and excludes direct drive generators for inventory55 in order to estimate the contribution of BUGS agricultural and industrial applications. to fossil fuel demand and emissions of health and climate damaging pollutants. Fuel estimates were compared to Methodological Overview IEA statistics for the power and commercial sectors and The framework we employed uses existing data on genera- adjusted so that the overall energy use is consistent with tor sales, ownership prevalence, grid reliability, and gen- IEA. We examine the factors affecting uncertainty in our erator performance characteristics to estimate various fleet results and discuss the implications of these results on characteristics and their impacts of operation. Resulting future efforts to address major knowledge gaps. impact estimates are directly linked to the level of service TABLE 6.1: SUMMARY OF COUNTRIES MODELED, AND THEIR POPULATIONS AGGREGATED ACROSS WORLD BANK REGIONS Number of Countries Modeled Population Of Modeled Countries World Bank Region (LMICs Only) (Millions) East Asia & Pacific 29 (22) 607 Europe & Central Asia 10 (9) 167 Latin America & Caribbean 39 (24) 634 Middle East (“Western Asia”) & North Africa 20 (13) 436 South Asia 8 (8) 1766 Sub-Saharan Africa 47 (46) 1030 Other 14 (0) 0.86 Total 167 (122) 4642 37 FIGURE 6.1: GEOGRAPHIC SCOPE OF THIS STUDY AND DETAILED REGIONAL CLASSIFICATIONS ANALYSIS METHODS The fuel consumption, A, of generators using fuel type, k, in country i can be described as: Overview Ai,k=∑ gNi,k,gPi,k,gTiCFk,gBRk,g          (1) Our first-order modeling approach emphasized impacts that directly relate to the consumption and expenditure Where Ni,k,g represents the number of generators in coun- on fuels and generating units, as determined by utiliza- try i using kth fuel of the gth size (capacity) category. We tion characteristics. It is based on a variety of existing distinguish three size categories for diesel generators and and assembled data sets, including global trade records, one size category for gasoline generators. Pg,k is the aver- national surveys, reported grid reliability (SAIDI), and age rated power output of a generator in size category g, generator performance characteristics. Fuel estimates were informed by review of literature and discussion with gen- used to update inventories within IIASA’s Greenhouse Gas- erator distributors. Tk is the average runtime of a genera- Air Pollution Interaction and Synergies (GAINS) model to tor based on SAIDI values calculated at the national level estimate emissions and facilitate comparison across other or based on regional averages in the minority of instances sources and sectors. where country-level estimates were not available. CFk,g is the fraction of the rated capacity that is utilized when To the extent possible, it was important that our approach operated (capacity factor). The product of parameters up reflect the connection between energy services from BUGS to this point yields an estimate of energy generation of and grid performance with their downstream impacts. To generators (i.e. kWh). BRk,g is the average fuel consump- provide relevant national level insights, we rely upon as tion rate, calculated from fuel consumption curves to many country-level data sets as possible to inform fleet account for differences and dependencies on generator sizes, sectoral allocations, and generator runtimes while type, size categories, and output levels. still maintaining a consistent procedure across all coun- tries. Our bottom-up approach differs from many previous For modeled simulations, parameters in Eq.1 were allowed efforts to size backup generator service and impacts in that to vary around distribution parameters informed by results we do not begin with the assumption that power plants on from underlying analyses, literature review, and consul- the grid, even if reliable, reflect electricity demand. tations with generator industry experts. Note that our framework also maintains flexibility to account for differ- ences across user segment, grid access, and urban status. 38 In nearly all instances, however, there was inadequate data indicator set includes country level estimates of SAIDI as to account for differences in performance and operational well as other electrical grid reliability metrics. For some parameters at such a granular segmentation scale. Thus, countries, reliability metrics are reported for two cities, generator operation characteristics are assumed to be the and in these cases the average of the two cities was calcu- same across these classifications within a country. lated and used to represent the nation. Doing Business sur- veys are representative of the country’s largest economic Several impacts are calculated from fuel consumption center and are therefore not nationally representative. To (activity) estimates. For example, annual fuel consumption account for this, we adjusted SAIDI estimates from Doing multiplied by an emission factor (EF) for nitrogen dioxide, Business Surveys with a scaling factor based on a compari- yields the nitrogen dioxide emission rate; multiplying die- son of SAIDI for countries sampled in both Doing Business sel consumption by the local pump price yields an estimate and Enterprise Surveys. of annual expenditure on fuel for BUGS. SAIDI Estimation The framework for estimating emissions from generators Due to differences in data availability for each of the can be generally expressed as: countries for which SAIDI is estimated, multiple estima- tion methodologies have been implemented. Each of the Ei,p=∑k∑mAi,k,EFi,k,m,pXi,k,m,p          (2) possible methodologies for generating an estimate of SAIDI in each country is described below. where i, k, m, p respectively represents the country, fuel type, abatement measure, and pollutant. Ei,p is the emis- Average EID from Enterprise Surveys sions of pollutant p in country i, and Ai,k the activity level If a country has Enterprise Survey data for the year 2016, of fuel type k estimated in Eq 1. EFi,k,m,p is the pollutant the average EID from firms surveyed is used as the SAIDI emission factor of pollutant p, of fuel type k, in country i, estimate. For the purposes of representing the uncertainty after application of control measure M. Xi,k,m,p is the share in this estimate the standard error is also calculated. The of total activity of type k in country i which control mea- World Bank Enterprise Surveys are firm-level surveys sure m for pollutant p is applied. For calculating baseline conducted through interviews with business owners and emissions, we apply default emission factors and control managers. Results from the enterprise survey were used measures from GAINS described in Klimont et al. (2017).56 to calculate an Experienced Interruption Duration (EID) for each firm. The EID is defined as the number of hours Generator Runtimes of outage experienced by the firm in the survey year and Generator runtimes were based on SAIDI values calculated when averaged, interpreted as the SAIDI value. from an analysis of the World Bank Enterprise Surveys Country Level GEE model and the World Bank Doing Business Surveys. SAIDI values If a country had Enterprise Survey data for at least two reflect the hours of grid outage per year experienced by the years but neither were from 2016, a 2016 SAIDI value average customer. For consistency, we apply 2016 values was estimated using a Generalized Estimating Equation of SAIDI, based on data from that survey year, or based on (GEE) model. The GEE model estimated SAIDI as a func- modeled trends. SAIDI was assigned at the country-level, tion of year using the EID for each firm in the country as based on country-specific data or regional trends if coun- an input. All observations from the same location listed in try data were not available. the Enterprise Survey (usually cities) were treated as inde- Data Sources pendent. This model was then used to estimate SAIDI in Data sources used for each country are described in the country for the year 2016. The standard error was also Table 6.2 calculated from the GEE model to represent the uncer- tainty in the SAIDI estimate. Doing Business Report Estimates from the World Bank Doing Business Report, Scaled Doing Business Survey specifically the Getting Electricity indicator set, were This approach was applied if a country had one or fewer also used to estimate grid outage. Starting in 2015 the years of Enterprise Survey data available (not 2016) but 39 a SAIDI estimate available from the World Bank Doing as a function of time for a region, then used to estimate Business Report, specifically the Getting Electricity indi- the 2016 EID for the country. All observations from the cator set. Starting in 2015 the indicator set included same location listed in the Enterprise Survey (usually cit- country-level estimates of SAIDI as well as other electri- ies) were treated as independent. The regions used are cal grid reliability metrics based on interviews with utility the UN regions with the exception of Oceania; the three companies. Doing Business surveys are representative of UN Regions (Polynesia, Melanesia and Micronesia) are the country’s largest economic centers and are therefore combined and treated as one region due to limited data not nationally representative and reflect a different sam- availability. This trend in SAIDI was used to extrapolate pling frame than the Enterprise Surveys. As a result, Doing from the SAIDI value calculated from the one year that Business SAIDI values were adjusted using results from a Enterprise Data was available for the country. To represent regression model of SAIDI values from Doing Business and the uncertainty in this estimate, the standard error was Enterprise Surveys, where there was country overlap. In calculated based on the regional SAIDI trend. all instances, this adjustment increased SAIDI. Three data points were thrown out as outliers due to high estimates of Regional Level GEE model SAIDI from Getting Electricity (South Sudan, Honduras, If no country data on SAIDI were available, a regional- and eSwatini). level average was applied based on results from a GEE model for that region. The defined regions are consis- Country Enterprise Survey Scaled With Regional tent with the UN regions with the exception of Oceania SAIDI Trend which is a combination of three UN regions (Polynesia, This approach was used if a country had one year of data Melanesia and Micronesia). All observations from the from the Enterprise Surveys that was not in 2016 and no same location listed in the Enterprise survey (usually cit- available data from Doing Business. A regional level GEE ies) are treated as a dependent. model was used to estimate the average change in SAIDI 40 TABLE 6.2: COUNTRIES MODELED AND CORRESPONDING DATA SOURCES USED TO INFORM ESTIMATES OF FLEET SIZE AND COMPOSITION NOTES AND REFERENCES C United Nations Statistical Division COMTRADE 2005–2016; Atlas of Economic Complexity57 E World Bank Enterprise Surveys58 D USAID Demographic and Health Surveys59 L World Bank Living Standards Measurement Study Household Survey60 I Telecom Base Transceiver Stations count61 NT GSMA “Powering Telecoms: West Africa Market Analysis” (2013) 62 NO World Bank “Diesel Power Generation Inventories and Black Carbon Emissions in Nigeria” (2004) 63 Country ISO Country Name Region Reference AFG Afghanistan Southern Asia C, E, D DZA Algeria Northern Africa C ASM American Samoa Polynesia C AGO Angola Middle Africa C, E, D AIA Anguilla Caribbean C ATG Antigua and Barbuda Caribbean C, E ARG Argentina South America C, E ARM Armenia Western Asia C, E ABW Aruba Caribbean C AZE Azerbaijan Western Asia C, E BHS Bahamas Caribbean C, E BHR Bahrain Western Asia C BGD Bangladesh Southern Asia C, E, D BRB Barbados Caribbean C, E BLZ Belize Central America C, E BEN Benin Western Africa C, E, D BTN Bhutan Southern Asia C, E BOL Bolivia South America C, E BWA Botswana Southern Africa C, E BRA Brazil South America C, E VGB British Virgin Islands Caribbean C BRN Brunei Darussalam South-Eastern Asia C BFA Burkina Faso Western Africa C, E BDI Burundi Eastern Africa C, E KHM Cambodia South-Eastern Asia C, E CMR Cameroon Middle Africa C, E, D CPV Cape Verde Western Africa C, E CYM Cayman Islands Caribbean C CAF Central African Republic Middle Africa C, E TCD Chad Middle Africa C, E CHL Chile South America C, E COL Colombia South America C, E COM Comoros Eastern Africa C COK Cook Islands Polynesia C CRI Costa Rica Central America C, E 41 Country ISO Country Name Region Reference CIV Côte d’Ivoire Western Africa C, E CUB Cuba Caribbean C CUW Curaçao Caribbean C CYP Cyprus Western Asia C PRK Democratic People’s Republic of Korea Eastern Asia C COD Democratic Republic of the Congo Middle Africa C, E, D DJI Djibouti Eastern Africa C, E DMA Dominica Caribbean C, E DOM Dominican Republic Caribbean C, E, D ECU Ecuador South America C, E EGY Egypt Northern Africa C, E SLV El Salvador Central America C, E GNQ Equatorial Guinea Middle Africa C ERI Eritrea Eastern Africa C, E ETH Ethiopia Eastern Africa C, E, L FLK Falkland Islands South America C FSM Federated States of Micronesia Micronesia C, E FJI Fiji Melanesia C, E PYF French Polynesia Polynesia C ATF French Southern and Antarctic Lands Seven seas (open ocean) C GAB Gabon Middle Africa C, E, D GEO Georgia Western Asia C, E GHA Ghana Western Africa C, E, D GRD Grenada Caribbean C, E GUM Guam Micronesia C GTM Guatemala Central America C, E GIN Guinea Western Africa C, E GNB Guinea-Bissau Western Africa C, E GUY Guyana South America C, E, D HTI Haiti Caribbean C HND Honduras Central America C, E IND India Southern Asia E, D, I IDN Indonesia South-Eastern Asia C, E IRN Iran Southern Asia C IRQ Iraq Western Asia C, E, L ISR Israel Western Asia C, E JAM Jamaica Caribbean C, E JOR Jordan Western Asia C, E KAZ Kazakhstan Central Asia C, E KEN Kenya Eastern Africa C, E KIR Kiribati Micronesia C KWT Kuwait Western Asia C KGZ Kyrgyzstan Central Asia C, E LAO Lao People’s Democratic Republic South-Eastern Asia C, E LBN Lebanon Western Asia C, E LSO Lesotho Southern Africa C, E, D 42 Country ISO Country Name Region Reference LBR Liberia Western Africa C, E, D LBY Libya Northern Africa C MDG Madagascar Eastern Africa C, E MWI Malawi Eastern Africa C, E, L MYS Malaysia South-Eastern Asia C, E MDV Maldives Southern Asia C MLI Mali Western Africa C, E MHL Marshall Islands Micronesia C MRT Mauritania Western Africa C, E MUS Mauritius Eastern Africa C, E MEX Mexico Central America C, E MNG Mongolia Eastern Asia C, E MSR Montserrat Caribbean C MAR Morocco Northern Africa C, E MOZ Mozambique Eastern Africa C, E MMR Myanmar South-Eastern Asia C, E BES Bonaire, Sint Eustatius and Saba Caribbean C MYT Mayotte Eastern Africa C TKL Tokelau Polynesia C TUV Tuvalu Polynesia C NRU Nauru Micronesia C NPL Nepal Southern Asia C, E NCL New Caledonia Melanesia C NIC Nicaragua Central America C, E NER Niger Western Africa C, E, L NGA Nigeria Western Africa C, E, L, NO, NT NIU Niue Polynesia C MNP Northern Mariana Islands Micronesia C OMN Oman Western Asia C PAK Pakistan Southern Asia C, E PLW Palau Micronesia C PSE Palestine Western Asia C PAN Panama Central America C, E PNG Papua New Guinea Melanesia C, E PRY Paraguay South America C, E PER Peru South America C, E, D PHL Philippines South-Eastern Asia C, E QAT Qatar Western Asia C COG Republic of Congo Middle Africa C, E RWA Rwanda Eastern Africa C, E SHN Saint Helena Western Africa C KNA Saint Kitts and Nevis Caribbean C, E LCA Saint Lucia Caribbean C, E VCT Saint Vincent and the Grenadines Caribbean C, E BLM Saint-Barthélemy Caribbean C 43 Country ISO Country Name Region Reference WSM Samoa Polynesia C, E STP São Tomé and Principe Middle Africa C SAU Saudi Arabia Western Asia C SEN Senegal Western Africa C, E SYC Seychelles Eastern Africa C SLE Sierra Leone Western Africa C, E, D SXM Sint Maarten Caribbean C SLB Solomon Islands Melanesia C, E SOM Somalia Eastern Africa C ZAF South Africa Southern Africa C, E SGS South Georgia and South Sandwich Islands Seven seas (open ocean) C SSD South Sudan Eastern Africa C, E LKA Sri Lanka Southern Asia C, E SDN Sudan Northern Africa C, E SUR Suriname South America C, E SWZ Swaziland Southern Africa C, E SYR Syria Western Asia C TWN Taiwan, China Eastern Asia C TJK Tajikistan Central Asia C, E, L TZA Tanzania Eastern Africa C, E, D GMB The Gambia Western Africa C, E TLS Timor-Leste South-Eastern Asia C, E, L TGO Togo Western Africa C, E TON Tonga Polynesia C, E TTO Trinidad and Tobago Caribbean C, E TUN Tunisia Northern Africa C, E TUR Turkey Western Asia C, E TKM Turkmenistan Central Asia C TCA Turks and Caicos Islands Caribbean C UGA Uganda Eastern Africa C, E, L ARE United Arab Emirates Western Asia C URY Uruguay South America C, E UZB Uzbekistan Central Asia C, E VUT Vanuatu Melanesia C, E VEN Venezuela South America C, E VNM Vietnam South-Eastern Asia C, E WLF Wallis and Futuna Islands Polynesia C ESH Western Sahara Northern Africa C YEM Yemen Western Asia C, E, D ZMB Zambia Eastern Africa C, E ZWE Zimbabwe Eastern Africa C, E, D 44 Capacity Factor the average electrical demand for each building. Finally, we divided the average electrical demand by the inferred We use the term capacity factor to mean the average nameplate capacity of the backup generator to arrive at an power output of a generator during operation divided by estimated capacity factor for each building. its rated power output. Another interpretation is that it is the portion of the maximum power output of a genera- Figure 6.2 illustrates components from the building load tor that is provided, on average. A capacity factor of 0.5 curve used to estimate the average capacity factor of a indicates that the generator would, on average, provide generator. The figure depicts the hourly average, max, and half of its maximum rated power output. Within the BUGS min load of a building for each hour of the day. The dot- workflow, the capacity factor is used in the estimates of ted line (Overall Maximum Load) is taken as the rated energy generation, and thus affects fuel consumption and capacity of the backup generator and the solid black line all resulting impacts downstream of this (for more detail (Overall Average Load) is taken as the average power out- on derivation of fuel consumption rates, see the “Fuel put of the generator. Using these values, capacity factor is Consumption Rates” document). calculated as Overall Average Load / Overall Maximum Load. If, in reality, the generator is drastically oversized To inform our estimate of capacity factor we used a data so that the maximum load is significantly smaller than the set containing smart meter data from over 60,000 com- actual rated capacity, this approach will yield an overesti- mercial buildings in California. We assumed that the peak mate of the capacity factor. power demand at each building was a proxy for nameplate capacity of the backup generator. Next, we calculated FIGURE 6.2: ESTIMATING GENERATOR CAPACITY FACTOR FROM BUILDING LOAD PROFILES Hourly Average Load Hourly Maximum Load Hourly Base Load Overall Maximum Load 50 Electrical Load (kW) 40 30 Overall Average Load 20 0 4 8 12 16 20 24 Time of Day 45 The average capacity factor for California commercial fuel consumption rates (liters/hour) on generator power buildings was around 0.3 (30 percent). For the purposes output (kW). A database containing hourly consumption of the BUGS model this value served as the mode of a rates and corresponding generator power outputs was triangular distribution of capacity factors used in Monte assembled from a review of 73 manufacturer specification Carlo simulations. The lower and upper bounds of the sheets of currently manufactured units. A separate linear distribution were assumed to be 0.2 and 0.8, respectively. regression was performed for each of the four generator The average value drawn from this distribution was categories. The generator fuel curve slope is in units of approximately 0.45 across all model runs. A right skewed liters per kWh (liters/kWh). distribution was used to account for the possibility that users would purchase a generator that would only be able Figure 6.3. shows generator fuel consumption curves to support base loads (i.e., the generator may be sized applied in the BUGS modeling framework. Each point such that load shedding is necessary during grid outages), represents one operating point for a generator, meaning making the necessary generator capacity much lower and that one generator model may be represented by multiple increasing the capacity factor. points (up to 4) on the graph. Individual data points are taken from performance specification sheets of currently FUEL CONSUMPTION CURVES manufactured generators. Solid vertical lines represent the median output of a generator in each category based on We estimated fuel consumption rates (liters/kWh) from simulated runs that vary the average generating capacity fuel curves generated from a linear regression of hourly of a generator and its operating capacity factor. Ninety FIGURE 6.3: GENERATOR FUEL CONSUMPTION CURVES THE FOUR GENERATOR CATEGORIES CONSIDERED IN THE BUGS MODELING FRAMEWORK Petrol Small Diesel 6 20 15 4 10 2 5 Fuel Usage (L/Hr) 0 0 0.0 2.5 5.0 7.5 0 20 40 60 Medium Diesel Large Diesel 100 600 75 400 50 200 25 0 0 0 100 200 300 0 500 1000 1500 2000 Generator Operating Point (kW) 46 FIGURE 6.4: IMPLIED EFFICIENCY CURVES FOR THE FOUR GENERATOR CATEGORIES USED IN THIS STUDY Petrol Small Diesel 40 40 30 30 20 20 10 10 0 0 % E ciency 0.0 2.5 5.0 7.5 0 20 40 60 Medium Diesel Large Diesel 40 40 30 30 20 20 10 10 0 0 0 100 200 300 0 500 1000 1500 2000 Generator Operating Point (kW) percent of modeled estimates fall within the dashed lines modeled relationships shown in Figure 6.4 combined with (90 percent confidence interval). heating values for respective fuels. Implied Efficiency Curves Figure 6.4. shows the implied efficiency curves for the four generator categories in the BUGS modeling frame- In general, the efficiency of a generator changes depend- work. Solid vertical lines represent the median output of ing on the electrical load relative to its maximum output. a generator in each category based on simulated runs that Over the operational range of a generator, its efficiency vary the average generating capacity of a generator and its may vary by as much as 35 percent, being lowest near the operating capacity factor. Ninety percent of modeled esti- bottom end of its operating range. However, while the mates fall within the dashed lines (90 percent confidence effect of changing capacity factor on operating point does interval). To calculate efficiency, hourly fuel consumption result in changes in efficiency, it does not decrease overall rates are converted to power assuming a heating value for fuel usage. This is because the dominating factor affecting gasoline (32 MJ/liter) and diesel (36 MJ/liter). fuel usage is energy generated—given the same runtime, a higher capacity factor always leads to more energy gen- UNCERTAINTY ANALYSIS eration and fuel usage. If a generator is running a small load (low capacity factor) it delivers a relatively small An in-depth uncertainty analysis was performed using amount of energy at a lower efficiency. The same genera- Sobol uncertainty decomposition with Mauntz estima- tor running a larger load (high capacity factor) delivers tors (Pujol et al. 2017). Our approach applies procedures much more energy at a slightly improved efficiency. Figure outlined in Sobol and Saltelli (Saltelli et al. 2010; Sobol 6.4 presents implied efficiency curves estimated from the et al. 2007) and has been applied widely in the systems 47 modeling literature (Saltelli et al. 2008). The specific com- subsidies. This implied subsidy is estimated as the differ- putational algorithm was selected for its ability to accu- ence between the domestic consumer (pump) price and the rately calculate small first and total order indices (Pujol et international spot price, adjusting for transportation, dis- al. 2017; Sobol et al. 2007). Another benefit of this algo- tribution, and retailing costs. Informed by previous appli- rithm is that it can simultaneously calculate both first and cations of this approach, this adjustment is assumed to total order Sobol indices.64 be $0.20 per liter for oil importing/net zero countries and zero for oil exporting countries (Davis 2014, IMF 2013).65 Sobol first-order indices represent the reduction in out- put variance which would occur if the variable were to We used historic consumer pump prices for diesel and be fixed to a single value. They represent the amount of gasoline freely available through World Bank data banks output variance which would be present if all variables (World Bank, 2018). Oil market status was determined were fixed except the variable in question (Saltelli et al. using crude oil imports and exports from UN Comtrade 2010). For this reason, first-order indices sum to one, as if International Trade Statistics Database (Center for all variables were fixed to single values there would be no International Development at Harvard University). output variance (100 percent reduction). International spot prices were taken from EIA databases (EIA, 2018). Due to the large number of variables used within our model, Sobol indices were calculated for variable cat- Total subsidy cost for fuel used in BUGS is calculated at a egories. This reduces computation time and accuracy national level using the estimated united subsidy per liter of indices due to the reduction in dimensionality. These estimated above, and the estimated fuel usage for the same groupings also allow each of our inputs to be indepen- country from our model. For the purpose of this analysis dently sampled, which is an assumption requirement of we only consider consumer subsidies. Several countries the procedure. Variables are grouped into three catego- modeled by BUGS did not have subsidies calculated (27 ries: Fleet Characteristics, Generator Characteristics, and percent) as domestic consumer price data were not avail- Runtime. Each category represents the combined influence able. However, these countries represent a small portion of of up to 32 individual input parameters and is applied at total BUGS fuel consumption and would likely have little the country and generator fuel type levels. impact on the total subsidy value. SUBSIDIES Implied unit subsidies for both gasoline and diesel fuels are estimated using the price gap approach, a widely implemented method of determining post-tax consumer 48 Appendix 2: Opportunities to Reduce Uncertainty in Estimates As with many distributed energy systems, significant gaps in the understanding of backup generator use and performance characteristics exist, affecting the precision and accuracy of final impact estimates.It was important that this work consider, to the extent possible, how these gaps contributed to the uncer- tainty of final results, and use this insight to provide data-driven recommendations for informing future research and market intelligence efforts. We identify that there is both uncertainty resulting from an attempt to apply a consistent modeling approach across countries, and also uncertainty in the parameters of our model arising from data gaps. Overall, these assumptions have likely resulted in conservatively low estimates in most countries and regions. This Appendix describes the implications of our results on strategies for improving understanding and reducing the uncertainty based on generator type (gasoline or diesel) and location. As a result, the strate- gies and measurements to address areas of greatest need are differentiated depending on the types of gen- erators deployed and the population in question. MODEL UNCERTAINTY DECOMPOSITION Sources of uncertainty arising from various model inputs were grouped into three knowledge categories: 1. Fleet Size Characteristics: Assumptions affecting the size of fleets 2. Generator Characteristics: Assumptions affecting the size, performance, and operation of generators in the fleet. 3. Runtime Characteristics: Assumptions affecting the utilization rate of generators in the fleet. Figure 7.1 shows the portion of diesel and gasoline consumption uncertainty attributed to each knowl- edge category. For gasoline, factors affecting Fleet Size dominate, largely as a result of discrepancies between trade records and survey-based measures of fleet size in the residential and commercial sectors. Uncertainty in diesel consumption is dominated by Generator Characteristics, particularly factors influ- encing how units in the largest (> 300 kW) size category are operated. Although these units account for a small fraction of units in the fleet (by number), they have the potential to account for a large fraction of generation and consume large quantities of fuel in a short period of runtime. Gasoline generators have maximum output capacities that are roughly 100 times less than the largest diesel generator classes, mak- ing Generator Characteristic less influential on total gasoline consumption estimates. Efforts that address key knowledge gaps in several regions can provide large reductions to overall uncer- tainty of fuel consumption estimates in developing regions of the world. The large gasoline generator fleets in Western Africa account for most of the total uncertainty in gasoline consumption in this region (Figure 7.2). For diesel, Southern Asia dominates, followed by Western Africa. Notability, the relative importance of addressing specific knowledge categories for improving diesel consumption estimates changes by region, suggesting that there may be value in tailoring monitoring strategies accordingly. 50 FIGURE 7.1: CONTRIBUTION TO UNCERTAINTY IN TOTAL DIESEL AND GASOLINE CONSUMPTION ESTIMATES FOR ALL MODELED COUNTRIES BY KNOWLEDGE CATEGORY Petrol Diesel Fleet Characteristics Generator Characteristics Runtime 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 Fraction of Uncertainty Conversely, the relative importance of knowledge cat- power outages) and its relationship with BUGS utilization egories in contributing to gasoline uncertainty remain for various user groups would not only improve accuracy, relatively similar across regions, suggesting that a single but would also provide critical knowledge for understand- strategy may be adequate, at least at the regional scale. ing the viability of generator alternatives. Despite the negative impacts that unreliable electricity supply has on Addressing areas contributing to the uncertainty in diesel populations and economies, there remains limited data on consumption appears most important for developing coun- global power systems, the operational runtimes of genera- tries given that it accounts for the majority of fuel con- tors, and significant discrepancies in coverage and report- sumed in most regions for backup generation. Our results ing, making comparison across what few data sets exist reveal, however, that small gasoline generators remain difficult.66 important, accounting for a meaningful fraction of total fossil fuel demand for backup power, and are dominant There are some areas of impact that we report on for in some countries, notably Nigeria. How these generator which our treatment of uncertainty was not applied but classes reflect distinctions between user segments is also are still important for establishing baseline impacts and important; for example, an emphasis on gasoline genera- mitigation potential. tors and small diesel will likely target domestic and small business users, while larger diesel categories are likely to Wherever possible, we applied consistent estimation pro- emphasize commercial and industrial applications. Policy cedures across all countries examined as part of our study. mechanisms and control strategies for affecting change For some countries, particularly those that manufacture may also vary by sector and user segment, justifying a or export large numbers of BUGS, an alternative approach more granular analysis. was needed. In India, for example, a bottom-up (sector-by- sector) estimation approach was performed that did not Model assumptions affecting generator Runtime rely on global trade data. More detailed approaches were Characteristics had similar importance for gasoline not possible for all countries for which a standardized and diesel, accounting for 12 percent and 16 percent of approach was deemed inappropriate, however. China and the uncertainty in total fuel consumption, respectively. Namibia for example, were excluded from our analysis Although Runtime is identified as lowest priority in terms but are likely important for generator markets and pos- of improving the precision of our model results, 12 to 16 sibly impacts. percent of the total model uncertainty is still large, and improving estimates of SAIDI (the number of hours of 51 FIGURE 7.2: FRACTION OF UNCERTAINTY IN TOTAL GASOLINE (LEFT) AND DIESEL CONSUMPTION (RIGHT) ESTIMATES, APPORTIONED BY REGIONS AND KNOWLEDGE CATEGORY Petrol Consumption Diesel Consumption Other Modeled Other Modeled Regions Regions Middle Africa Western Asia South America Western Asia Southern Asia Western Africa Western Africa Southern Asia 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Fraction of Uncertainty Attributed to Model Inputs Runtime Generator Characteristics Fleet Characteristics Another potentially important area not explicitly exam- characteristics of new generators, based primarily on ined in our uncertainty analysis were gaps in the under- laboratory measurements conducted in industrialized standing of emission characteristics of generators. There countries. Such measurements do not reflect the effects of is extremely limited data on the emissions from genera- poor maintenance, age, or fuel quality, for example, on the tors used in developing countries under typical opera- emission strength of generators. tion. In the absence of these data, we relied on emission 52 ENDNOTES 1 Marginal grid cost: Trimble, Christopher Philip, Masami Kojima, Ines Perez Arroyo, and Farah Mohammadzadeh. 2016. “Financial Viability of Electricity Sectors in Sub-Saharan Africa: Quasi-Fiscal Deficits and Hidden Costs.” WPS7788. The World Bank. http://documents.worldbank.org/curated/en/182071470748085038/Financial-viability-of-electricity-sectors-in-Sub-Saharan-Africa-quasi-fiscal-deficits-and-hidden-costs. 2 Solar+storage levelized cost: Lazard. 2018. “Levelized Cost of Energy and Levelized Cost of Storage 2018.” 2018. http://www.lazard.com/perspective/levelized-cost-of-energy-and-levelized-cost-of-storage-2018/. 3 Photo Credit: Bhushan Tuladhar, from Diesel Power Generation: Inventories and Black Carbon Emissions in Kathmandu Valley, Nepal. 4 Study scope includes Latin America, South America, Africa, the Middle East, Pacific Islands, and most of Asia (excluding China) 5 Doing Business, The World Bank (http://www.doingbusiness.org) 6 Based on an analysis of business expenditure surveys conducted by IFC in Nigeria in 2018–2019 (unpublished). 7 Greenhouse Gas­ — Air Pollution Interaction and Synergies (GAINS) model maintained by the International Institute of Applied Systems Analysis (IIASA). Amann, M. et al. Cost- effective control of air quality and greenhouse gases in Europe: modeling and policy applications. Environ. Model. Softw. 26, 1489–1501 (2011). 8 The World Factbook. Washington, DC: Central Intelligence Agency. 2019. https://www.cia.gov/library/publications/the-world-factbook/index.html. 9 SE4All projects in progress in Nigeria. 2017. http://se4all.ecreee.org/sites/default/files/Nigeria_IP.pdf. 10 CIA World Factbook. 11 SE4All. 2017. 12 https://www.premiumtimesng.com/business/business-news/185668-nigerian-manufacturers-spend-n3-5trn-yearly-on-generators-nlc.html 13 https://guardian.ng/sunday-magazine/when-powerless-government-banned-powerful-generator/ 14 Access to Energy Institute (2019) “Solar Killed the Generator Star” (Video Production) https://vimeo.com/341730105. 15 Access to Energy Institute (2019) “Solar Killed the Generator Star” (Video Production) https://vimeo.com/341730105. 16 http://se4all.ecreee.org/sites/default/files/Nigeria_IP.pdf. 17 The uncertainty of variables used to estimate generator fleet characteristics and impacts are considered in our modeling framework. Each input variable is given a range of possible values The 90 percent uncertainty interval (UI) indicates that 90 percent of model runs fell within the specified interval. 18 Koomey, Jonathan, et al. 2010. “Defining a standard metric for electricity savings.” Environmental Research Letters 5.1 (2010): 014017. 19 India, Angola, Indonesia, Argentina, Saudi Arabia, Nigeria, Philippines, Venezuela, Bangladesh, Chile, Algeria, Iraq 20 Among the subset of 111 countries that were modeled and total grid capacity estimates were available (98 percent of the modeled population). 21 Koomey, Jonathan, et al. 2010. “Defining a standard metric for electricity savings.” Environmental Research Letters 5.1 (2010): 014017. 22 Estimated from 35 of 48 countries in Sub-Saharan Africa for which data on grid generation was available, adjusted for transmission and distribution losses. 23 Values based on import records and do not account for generators assembled or manufactured in country, or units sold via illegal markets. 24 Values based on fleet size estimates and the average unit prices across generator classes from trade records. 25 IMF. 2019. Global Fossil Fuel Subsidies Remain Large: An Update Based on Country-Level Estimates. https://www.imf.org/en/Publications/WP/Issues/2019/05/02/Global-Fossil-Fuel-Subsidies-Remain-Large-An-Update-Based-on-Country-Level-Estimates-46509 26 Global Burden of Disease Study 2017 (GBD 2017) Results. Seattle, United States: Institute for Health Metrics and Evaluation (IHME). 27 Farquharson, DeVynne, Paulina Jaramillo, and Constantine Samaras. 2018. “Sustainability implications of electricity outages in sub-Saharan Africa.” Nature Sustainability 1.10 (2018): 589. 28 Guttikunda, Sarath K., K. A. Nishadh, and Puja Jawahar. 2019. “Air pollution knowledge assessments (APnA) for 20 Indian cities.” Urban Climate 27 (2019): 124-141. 29 http://www.indiaenvironmentportal.org.in/files/Rpt-air-monitoring-17-01-2011.pdf p. 97. 30 Marais, Eloise A., and Christine Wiedinmyer. 2016. “Air quality impact of diffuse and inefficient combustion emissions in Africa (DICE-Africa).” Environmental science & tech- nology 50.19 (2016): 10739–10745. 31 Klimont, Zbigniew, et al. 2017. “Global anthropogenic emissions of particulate matter including black carbon.” Atmospheric Chemistry and Physics 17.14 (2017): 8681–8723. 32 https://www.who.int/en/news-room/fact-sheets/detail/ambient-(outdoor)-air-quality-and-health 33 Wang, Liangzhu, Steven J. Emmerich, and Andrew K. Persily. 2010. “In situ experimental study of carbon monoxide generation by gasoline-powered electric generator in an enclosed space.” Journal of the Air & Waste Management Association 60.12 (2010): 1443–1451. 34 Afolayan, J. M., N. P. Edomwonyi, and S. E. Esangbedo. 2014. “Carbon monoxide poisoning in a Nigerian home.” The Nigerian postgraduate medical journal 21.2 (2014): 199–202. 35 Seleye-Fubara, D., E. N. Etebu, and B. Athanasius. 2011. “Pathology of deaths from carbon monoxide poisoning in Port Harcourt: an autopsy study of 75 cases.” Nigerian journal of medicine: journal of the National Association of Resident Doctors of Nigeria 20.3 (2011): 337–340. 36 Marais, Eloise A., and Christine Wiedinmyer. 2016. “Air quality impact of diffuse and inefficient combustion emissions in Africa (DICE-Africa).” Environmental science & tech- nology 50.19 (2016): 10739–10745. 37 Guttikunda, Sarath K., K. A. Nishadh, and Puja Jawahar. 2019. “Air pollution knowledge assessments (APnA) for 20 Indian cities.” Urban Climate 27 (2019): 124–141. 38 Nuti, Marco. 1998. Emissions from two-stroke engines. SAE. 39 Wang, Liangzhu, Steven J. Emmerich, and Andrew K. Persily. 2010. “In situ experimental study of carbon monoxide generation by gasoline-powered electric generator in an enclosed space.” Journal of the Air & Waste Management Association 60.12 (2010): 1443–1451. 40 Marais, Eloise A., and Christine Wiedinmyer. 2016. “Air quality impact of diffuse and inefficient combustion emissions in Africa (DICE-Africa).” Environmental science & tech- nology 50.19 (2016): 10739–10745. 41 Akin, Akindele O. 2016. “False adaptive resilience: The environmental brutality of electric power generation use in Ogbomoso, Nigeria.” World Environment 6.3 (2016): 71–78. 42 https://cleantechnica.com/2018/08/02/the-nigerian-entrepreneur-who-wants-his-country-to-be-generator-free/ 43 https://www.premiumtimesng.com/business/business-news/185668-nigerian-manufacturers-spend-n3-5trn-yearly-on-enerators-nlc.html 44 Lazard. 2017. “Levelized Cost of Energy 2017.” https://www.lazard.com/media/450337/lazard-levelized-cost-of-energy-version-110.pdf. 45 Oviroh, Peter, and Tien-Chien Jen. 2018. “The energy cost analysis of hybrid systems and diesel generators in powering selected base transceiver station locations in Nigeria.” Energies 11.3 (2018): 687. 53 46 Oviroh, Peter, and Tien-Chien Jen. 2018. 47 Solar+storage levelized cost: Lazard. 2018. “Levelized Cost of Energy and Levelized Cost of Storage 2018.” 2018. http://www.lazard.com/perspective/levelized-cost-of-energy-and-levelized-cost-of-storage-2018/. 48 This included country-level analyses of Living Standards Measurement Surveys (World Bank), Demographic and Health Surveys (USAID), and surveys conducted by National Statistics Bureaus. 49 Enterprise Surveys (World Bank), Doing Business Surveys (World Bank). 50 Based on an analysis of business expenditure surveys conducted by IFC in Nigeria in 2018–2019 (unpublished). 51 Farquharson, DeVynne, Paulina Jaramillo, and Constantine Samaras. 2018. “Sustainability implications of electricity outages in sub-Saharan Africa.” Nature Sustainability 1.10 (2018): 589. 52 Davis, Lucas W. 2014. “The economic cost of global fuel subsidies.” American Economic Review 104.5 (2014): 581–585. Clements, Mr Benedict J., et al. Energy subsidy reform: lessons and implications. International Monetary Fund, 2013. 53 https://www.who.int/quantifying_ehimpacts/publications/e94888.pdf?ua=1 54 While GAINS regions are less granular in Africa, aggregation to these levels provides a more consistent basis from which to compare backup generator fuel use and emissions to other relevant sectors. With this approach, fossil fuel demand for backup generation is balanced to regional and national totals. 55 Greenhouse Gas - Air Pollution Interaction and Synergies (GAINS) model maintained by the International Institute of Applied Systems Analysis (IIASA). 56 Klimont, Zbigniew, et al. 2017. “Global anthropogenic emissions of particulate matter including black carbon.” Atmospheric Chemistry and Physics 17.14 (2017): 8681–8723. 57 “The Atlas of Economic Complexity.” Center for International Development at Harvard University, http://www.atlas.cid.harvard.edu 58 https://www.enterprisesurveys.org/ 59 https://www.dhsprogram.com/ 60 http://surveys.worldbank.org/lsms 61 https://community.data.gov.in/base-transceiver-stations-btss-installed-at-mobile-towers-as-on-29-02-2016/ 62 https://www.gsma.com/mobilefordevelopment/resources/powering-telecoms-west-africa-market-analysis/ 63 https://openknowledge.worldbank.org/handle/10986/28419 64 Pujol, A. G., Iooss, B., Janon, A., Veiga, D., Delage, T., Fruth, J., Gilquin, L., Guil-, J., Gratiet, L. Le, Lemaitre, P., Nelson, B. L., Oomen, R., Ramos, B., Roustant, O., Staum, J., Touati, T., Weber, F., and Iooss, M. B. 2017. “Package ‘sensitivity.’” Saltelli, Andrea, et al. 2010. “Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index.” Computer Physics Communications 181.2 (2010): 259–270. Saltelli, Andrea, et al. 2008. Global sensitivity analysis: the primer. John Wiley & Sons, 2008. Tarantola, S., et al. 2007. “Estimating the approximation error when fixing unessential factors in global sensitivity analysis.” Reliability Engineering & System Safety 92.7 (2007): 957–960. 65 Davis, Lucas W. 2014. “The economic cost of global fuel subsidies.” American Economic Review 104.5 (2014): 581–585 Clements, Mr Benedict J., et al. 2013. Energy subsidy reform: lessons and implications. International Monetary Fund. 66 Taneja, Jay. 2018. Measuring Electricity Reliability in Kenya. Working paper, available at http://blogs.umass.edu/jtaneja/files/2017/05/outages.pdf, accessed in 2018. September 2019 2121 Pennsylvania Ave. NW Washington, DC 20433 USA T: 202.473.1000 www.ifc.org