Mainstreaming Disruptive Technologies in Energy (P166854) FINAL REPORT Kwawu Mensan Gaba, World Bank Brian Min, University of Michigan Olaf Veerman, Development Seed Kimberly Baugh, National Oceanic and Atmospheric Administration Acknowledgments The report was prepared by a team led by Kwawu Mensan Gaba, Global Lead – Power Systems in the Energy and Extractive Industries Global Practice (EEXGP), under the guidance of the Senior Director of the EEXGP, Riccardo Puliti. The work took place under the Infrastructure Vice President Makhtar Diop and his predecessor, the Sustainable Development Vice President, Laura Tuck. The report is the result of a collaboration between the World Bank, the University of Michigan (UM), the National Oceanic and Atmospheric Administration’s National Center for Environmental Information (NOAA-NCEI) (formerly National Geophysical Data Center), and Development Seed (software developer). Members of the core Bank team included Trevor Monroe and Bruno Sanchez Andrade Nuno. The UM team led by Brian Min included Zachary O’Keeffe, Htet Thiha Zaw, and Paul Atwell. The NOAA-NCEI team led by Kimberly Baugh included Chris Elvidge (NOAA-NCEI), Mikhail Zhizhin, Feng-Chi Hsu and Tilottama Ghosh from Cooperative Institute for Research in Environmental Sciences (CIRES), University of Colorado. The Development Seed team led by Derek Liu and Olaf Veerman included Vitor George, Alireza Jazayeri, Laura Gillen, Kim Murphy and Ian Schuler. The team is grateful for the guidance received from the peer reviewers, Martin Heger (Economist, GEN05), Tigran Parvanyan (Energy Specialist, ESMAP), Jun Erik Rentschler (Young Professional, GFDRR), Benjamin Stewart (Geographer, GTKM1) and Borja Garcia Senna (Consultant, GEE04). Other colleagues, Dana Rysankova (Global Lead, Energy Access), Rhonda Lenai Jordan-Antoine (Energy Specialist, GEE01), Zubair K.M. Sadeque (Senior Energy Specialist, GEE08) and Varun Nangia (Consultant, GEE07), also provided inputs during the course of this activity and their contribution is greatly appreciated. The methodology and outputs of this analytical work were discussed and disseminated at several events over the past year: (i) Nightlights Panel at the 2018 SatSummit (Washington DC, September 2018), (ii) Launch of the Global Nights Platform on November 15, 2018 (Washington DC), (iii) Launch of the Kenya National Electrification Strategy on December 6, 2018 (Nairobi, Kenya), (iv) World Bank Group Data Day on February 13, 2019 (event sponsored by the WBG Development Data Council), and (v) BBL: Looking in the Dark: Kenya Geospatial and Nightlight Assessment on February 21, 2019. The team is honored that the Interim WBG President & Chief Executive Officer, Kristalina Georgieva, referred to this work as “a very cool platform” in her opening remarks during the WBG Data Day. The team is also very grateful for the various positive feedbacks garnered and support received during these events. Finally, we are grateful for the support of the EEXGP Management Team, who ultimately has the responsibility to create the conducive environment for mainstreaming this innovative approach into energy operations. 2 Table of Contents Acknowledgments......................................................................................................................................... 2 Figures and Tables .................................................................................................................................... 4 Introduction ................................................................................................................................................... 5 What has changed since the Detection of Rural Electrification in Africa using DMSP-OLS Night Lights Imagery? ....................................................................................................................................................... 8 Box 1: At broad-scale VIIRS & DMSP data look quite similar but there is no saturation with VIIRS . 11 Box 2: How do the DMSP and VIIRS products differ? .......................................................................... 12 How Did We Improve the Analytical Framework? .................................................................................... 19 Method One: Threshold-Based Classification ........................................................................................ 23 Method Two: Likelihood Estimates of Electricity Access ..................................................................... 28 Validation................................................................................................................................................ 34 Detecting changes over time ................................................................................................................... 38 What are the New Features in the Global Night Lights Platform? ............................................................. 43 Improving and Iterating .......................................................................................................................... 43 Delivering Data at Scale ......................................................................................................................... 44 Data Service ............................................................................................................................................ 46 Improved Data Model ......................................................................................................................... 46 Flexible Data Ingest ............................................................................................................................ 46 Scalable API........................................................................................................................................ 47 Extracting High-Resolution Settlement Data .......................................................................................... 47 Constantly Improving ............................................................................................................................. 48 What Possibilities Do We Now Have? ....................................................................................................... 49 Conclusion .................................................................................................................................................. 53 Annex 1: Classification of Electrified Settlements in 33 countries (Method One) ..................................... 55 Sub-Saharan Africa ................................................................................................................................. 55 Middle East and North Africa ................................................................................................................. 66 South Asia ............................................................................................................................................... 67 East Asia and Pacific .............................................................................................................................. 68 Latin America and the Caribbean ........................................................................................................... 70 Annex 2: Data Appendix ............................................................................................................................ 73 Ghana ...................................................................................................................................................... 74 Cote D’Ivoire .......................................................................................................................................... 79 The Philippines ....................................................................................................................................... 81 Rwanda ................................................................................................................................................... 84 Sri Lanka ................................................................................................................................................. 86 Tanzania .................................................................................................................................................. 95 3 Figures and Tables Figure 1: Structure of Global Night Lights program. ................................................................................... 7 Figure 2: VIIRS and DMSP-OLS footprints compared. ............................................................................. 12 Table 1: Comparison between DMSP-OLS and VIIRS DNB .................................................................... 12 Figure 3: Sample VIIRS nighttime lights image. Source: NOAA. ............................................................. 13 Figure 4: Processing of VIIRS Satellite Data by EOG NOAA for Electrification Access Estimates ........ 15 Table 2: Countries with processed data as part of the Global Night Lights program. ................................ 20 Table 3: Data Sources for Subnational Census Data on Electricity Access................................................ 21 Figure 5: Methods used to determine electrification................................................................................... 22 Figure 6: Average Light Output. ................................................................................................................. 23 Figure 7: Average Light Output overlaid with human settlements. ............................................................ 24 Figure 8: Method One (Threshold-Based Classification) resulting from analysis...................................... 24 Table 4: Optimal VIIRS Thresholds to Match Official Electrification Rates, 2015 ................................... 27 Figure 9: Background Noise Levels in Unsettled, Unelectrified Areas Across Different Landcover Types ............................................................................................................................................................ 28 Figure 10: Process Flow for Method Two for computation of Likelihood Electricity Access Estimates... 30 Figure 11: Likelihood a settlement is electrified, Kenya 2017. .................................................................. 33 Figure 12: Likelihood a settlement is electrified, Ghana 2017. .................................................................. 34 Figure 13: Comparison of Official Vs. Night Lights Satellite Estimates (Method 2) Electrification Rates for 33 Countries (2012–2016). ............................................................................................................ 36 Figure 14: Comparison of Census vs. Satellite-Based Subnational Electrification Rates (Method Two) .. 37 Figure 15: Comparison of Ghana Electrification Rates by District (2010 Census vs. 2012 Satellite- Derived Method Two Estimates) ........................................................................................................ 38 Figure 16: Annual estimated electricity access rates for Ghana (2012-17) ................................................ 39 Figure 17: Annual estimated electricity access rates for Ghana (2012-17), by district .............................. 39 Figure 18:Changes in electrified settlements, Ghana (2012-17) ................................................................. 40 Table 5: Top 10 Districts with Largest Gains in Population Electrified, Ghana 2012–17.......................... 41 Table 6: Top 10 Districts with Largest Gains in Proportion Electrified, Ghana 2012–17 .......................... 41 Figure 19: Screenshot of Global Night Lights website. .............................................................................. 44 Figure 20: Architecture diagram for Global Night Lights platform............................................................ 45 Figure 21: Sample GeoJSON query response. ............................................................................................ 47 Figure 22: Kenya – High Priority Areas by Concentration of Unlit Settlements ....................................... 49 Figure 23: Comparison between Kenya and other African Countries ........................................................ 50 Figure 24: Comparison between Rwanda, Tanzania, Uganda, South Africa, Philippines and Sri Lanka .. 50 Figure 25: Comparison between Haiti, Burkina Faso, Algeria, Cambodia, Guatemala and Indonesia ...... 51 Figure 27: Electric infrastructure and night lights ...................................................................................... 52 4 Introduction Time is running out to reach the goal of universal electricity access by 2030, according to the latest Tracking SDG 7 Report.1 The development community needs innovative ways to collect systematic and geographically precise data about electrified regions to identify focus areas where collective actions from policymakers and other stakeholders can make a real and lasting impact on electricity access. Our previous work has demonstrated convincingly that nighttime satellite data could help detect the availability and use of electricity in rural areas.2 With the advent of new satellite sensors and the development of computational techniques, we can now push the frontier beyond what was not even deemed possible five years ago. This activity, for the first time, uses new data collected from a recently launched powerful satellite to evaluate electricity access down to settlement levels, and seeks to mainstream disruptive technologies into energy operations. According to the 2019 Tracking SDG 7 Report, about 840 million people – or about 11 percent of the world’s population – still live without electricity. The number of people having newly gained access to electricity has been accelerating since 2015 and it is estimated that more than 920 million have moved from darkness to light between 2010 and 2017. However, progress has been uneven across regions, and needs to become more widespread if the SDG 7 goal of universal access to electricity is to be met by 2030. While electrification efforts have been mostly successful in several parts of Asia, Latin America and the Caribbean, Sub-Saharan Africa remains the region with the largest deficit: according to the same report, 573 million people – more than one in two – lack access to electricity. The global urban-rural chasm in access also remains wide, with almost 87 percent of the world’s population without electricity living in rural areas. This is particularly acute in countries affected by Fragility, Conflict and Violence (FCV) where there are many difficult to reach communities and households. However, off-grid solar solutions ranging from solar home systems to solar mini-grids are emerging as an important driver of rural energy access, complementing grid electrification in some countries. Under a business-as-usual scenario, if 1 IEA, IRENA, UNSD, WB, WHO (2019), Tracking SDG 7: The Energy Progress Report 2019, Washington DC 2 Gaba, K.M., Min, B., Elvidge, C. and Thakker, A., (2016), “World: Using Satellite Imagery to Monitor Progress of Rural Electrification”, World Bank Report ACS19333. 5 current policies and population trends continue, as many as 674 million people will continue to live without electricity in 2030, mostly in Sub-Saharan Africa.3 Blimpo and Cosgrove-Davies4 argue that access to reliable electricity is a prerequisite for the economic transformation of African economies, especially in a digital age. The low rate of electricity access rate in Sub-Saharan African countries, coupled with the poor reliability and high costs of service, imposes substantial constraints on economic activities, provision of public services, adoption of new technologies, and quality of life. They also noted that the solution lies in understanding that the overarching reasons for the unrealized potential involve tightly intertwined technical, financial, political, and geographic factors. While policymakers, development institutions, utilities, private sector and other stakeholders are working to address this conundrum, much remains to be done in terms of more fundamental issues: measuring access, identifying target areas, selecting interventions and monitoring their sustainability as well as impact over time. Measuring electrification is rather complex given the differences in methodologies adopted by leading organizations such as the World Bank (WB) and the International Energy Agency (IEA). According to the SDG 7 Energy progress report (2018)5, the WB and IEA each maintains a country-by-country database of global electricity access rates. The former, included in the Global Tracking Framework (GTF), derives estimates from a suite of standardized household surveys that are conducted in most countries every two to three years, with a multilevel nonparametric model used to extrapolate for missing years. The IEA Energy Access Database sources data where possible from government-reported values for household electrification (usually based on utility connections), supplemented with a new measurement of off-grid access. While each of these traditional approaches gives different and important quantifications of electrification, they do not represent geographical coverage or evolution across time adequately. A third way, complementary to the two traditional approaches, has been emerged thanks to the use of nighttime satellite imagery and development of data processing technologies and is the primary subject of this report. 3 IEA, IRENA, UNSD, WB, WHO (2018), Tracking SDG 7: The Energy Progress Report 2018, Washington DC 4 Blimpo, Moussa P., and Malcolm Cosgrove-Davies. 2019. Electricity Access in Sub-Saharan Africa: Uptake, Reliability, and Complementary Factors for Economic Impact. Africa Development Forum series. Washington, DC: World Bank. 5 IEA, IRENA, UNSD, WB, WHO (2018), Tracking SDG 7: The Energy Progress Report 2018, Washington DC 6 The following diagram describes the working arrangements underlying the multi-year World Bank-led collaboration between the University of Michigan (UM), the National Oceanic and Atmospheric Administration (NOAA), and Development Seed (DevSeed), which has made breakthroughs in the detection of electricity access using nighttime satellite imagery, statistically rigorous data methods to construct consistent and reliable indicators on electrification, and advanced computational techniques for fast data processing, analysis and visualization (see Figure 1). Figure 1: Structure of Global Night Lights program. The activity is fully aligned with the Bank’s new approach to disruptive technologies, as endorsed by the 2018 Development Committee. By making these tools and data broadly accessible to practitioners and policymakers alike, this activity operationalizes the Build-Boost-Broker value proposition of the Bank. By building the data infrastructure, and boosting the capacity of institutions, communities, firms, and individuals to leverage and adapt disruptive technologies for electricity access monitoring, the activity positions the Bank to broker further partnerships across sectors to more quickly solve the electricity access challenge. The Bank thus consolidates its role of “partner of choice” for governments, donors, rural electrification agencies, power utilities, technology firms, private sector players and other stakeholders to ensure that disruptive technologies are harnessed to accelerate progress toward SDG 7 and the twin goals. 7 What has changed since the Detection of Rural Electrification in Africa using DMSP-OLS Night Lights Imagery? The launch of new satellites with more powerful sensors as well as advances in computational processing power and algorithmic sophistication have helped us make great strides in our efforts to monitor electricity access globally. Our prior successful efforts using satellite data from the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) to track rural electrification opened the way for a new approach of monitoring rural electrification given the geographical and temporal limitations of the traditional methods of surveys, technical visits, utility reports, and third party audits. We have demonstrated that DMSP-OLS can distinguish between electrified and unelectrified villages, even in developing countries with low rates of electricity access and consumption.6 While the DMSP-OLS data are unmatched in their historical coverage (collected data digitalized since 1992), weaknesses in the data quality of the raw data stream are well known. For example, the lack of on-board calibration and changes in gain settings mean that accurate changes in brightness cannot be readily inferred by comparing two images captured at different time points. Our team has developed methods to statistically control for changes in gain settings and to correct other sources of bias in the raw data stream. These methods have enabled us to generate more accurate data on trends and changes in daily brightness for every village in India over two decades.7 These results also provided insights on the performance and impact of the rural electrification program. Prior studies have demonstrated that nighttime satellites can detect light output originating from cities, fires, gas flares, and heavily lit fishing boats.8 Several studies have also shown that nighttime light output 6 Min, B. and K.M. Gaba. 2014. Tracking Electrification in Vietnam Using Nighttime Lights. Remote Sensing 6(10): 9511–9529; Min, B., K.M. Gaba, O.F. Sarr, and A. Agalassou. 2013. Detection of Rural Electrification in Africa using DMSP-OLS Night Lights Imagery. International Journal of Remote Sensing 34(22): 8118–8141. 7 Gaba, K.M., Min, B., Elvidge, C. and Thakker, A., (2016), “World: Using Satellite Imagery to Monitor Progress of Rural Electrification”, World Bank Report ACS19333. 8 Croft, T.A., 1978, "Night-time Images of the Earth From Space", Scientific American, 239, 68-79. Elvidge, C.D., K.E. Baugh, E.A. Kihn., H.W. Kroehl, E.R. Davis, and C. Davis. 1997. Relation between satellite observed visible- near infrared emissions, population, economic activity, and power consumption. International Journal of Remote Sensing 18(6): 1373–1379. 8 strongly correlates with electricity generating capacity and economic activity at the regional and national levels.9 Some recent research has identified the ways in which the delivery of electrical power can be politically motivated.10 While most of these studies have been conducted in the industrialized world or in urban environments, our recent research has demonstrated that nighttime light data can be used to detect lower levels of electricity use and outdoor lighting in the developing world.11 However, the resolution and saturation of the DMSP-OLS satellite data did not allow a granular analysis across the globe. A radical change in satellite data quality occurred in 2011 when NASA and NOAA launched the Suomi National Polar Partnership (SNPP) satellite carrying the first Visible Infrared Imaging Radiometer Suite (VIIRS) instrument. In contrast to DMSP-OLS, the VIIRS was designed to collect high quality radiometric data for digital analysis and input into numerical models. The VIIRS instrument includes a day / night band (DNB) which collects standard panchromatic image data by day and low light imaging data at night. The DNB low light imaging is based on a time delay and integration (TDI) charge coupled device (CCD). The VIIRS instrument offers improvement in each of these shortcomings, except multispectral low light imaging. The NASA & NOAA instrument began collecting low light imaging day/night band (DNB) data in 2012. From the raw satellite data, NOAA generates several product types: • Sub-orbits: The data are acquired as strips that circle the earth from pole to pole. Approximately half of each orbit is nighttime data. It is possible to search for the orbits covering a specific region and deliver the nighttime data as sub-orbits. 9 De Souza Filho, C.R., Zullo, J. Jr., Elvidge, C., 2004, "Brazil's 2001 energy crisis monitored from space", International Journal of Remote Sensing, 25(12), 2475-2482; Doll C.N.H., Muller J.-P., Morley J.G., 2006, "Mapping regional economic activity from night-time light satellite imagery", Ecological Economics, 57, 75-92; Sutton, P. C., C. D. Elvidge and T. Ghosh, 2007, "Estimation of gross domestic product at sub-national scales using nighttime satellite imagery", International Journal of Ecological Economics and Statistics, 8 (SO7), 5 – 21; Ghosh, T., Powell, R., Elvidge, C. D., Baugh, K. E., Sutton, P. C., & Anderson, S., 2010, "Shedding light on the global distribution of economic activity", The Open Geography Journal , 3, 148-161; Henderson, J. Vernon & Adam Storeygard & David N. Weil, 2012. "Measuring Economic Growth from Outer Space," American Economic Review, American Economic Association, vol. 102(2), pages 994-1028. 10 Min, B. 2015. Power and the Vote: Elections and Electricity in the Developing World . New York: Cambridge University Press; Min, B. and M. Golden. 2014. Electoral cycles in electricity losses in India. Energy Policy 65: 619–625; Baskaran, T., Min, B., & Uppal, Y. (2015). Election cycles and electricity provision: Evidence from a quasi-experiment with Indian special elections. Journal of Public Economics, 126, 64-73. 11 Min and Gaba, 2014, “Tracking Electrification in Vietnam Using Nighttime Lights”, Remote Sensing. 6(10):9511–9529. 2014; Min, Gaba, Sarr, and Agalassou, 2013, “Detection of Rural Electrification in Africa using DMSP-OLS Night Lights Imagery”, International Journal of Remote Sensing 34(22):8118–8141. 2013. 9 • Geolocated Sub-orbits: The data in sub-orbits can be geolocated to form grids in a specific map projection. The standard output is in a straight latitude-longitude grid, known as a Plate Carrere projection. The standard for DMSP is a 30 arc second grid, with grid cells close to 1 km on a side at the equator. For VIIRS it is possible to use a 15 or 30 arc second grid. Possible ancillary files include: cloud layer indicating presence / absence of clouds based on long wave infrared analysis, lunar illuminance, a stray light layer to indicate zones with some level of solar contamination. These files can be provided as Georeferenced Tagged Image File Format files (GeoTIFFs). • Nightly mosaics: It is possible add geolocated data from multiple orbits to cover extended geographic regions from a single nightly image product. • Monthly cloud-free composites: This style of product is typically made with data collected with zero to near zero lunar illuminance. Cloudy data are excluded. Fires and flares are included in the data. Separate image files tally the number of coverages, number of cloud-free coverages, average visible band signal, minimum visible band value, maximum visible band value, standard deviation of the visible band data. May also include the percent frequency of detection based on results from a light detection algorithm. A separate layer indicating locations of gas flares can be provided. • Annual cloud-free composites: These are similar to the monthly cloud-free composites, but cover a full twelve months, typically a calendar year. Input data are filtered to remove sunlit and moonlit data, plus lightning features. A separate layer indicating locations of gas flares can be provided. • Annual stable lights: This product is filtered to exclude extraneous features not related to manmade lighting. This includes removal of background noise (where no lighting was detected), lightning, fires, and aurora. Separate image files tally the number of coverages, number of cloud- free coverages, average visible band signal, minimum visible band value, maximum visible band value, standard deviation of the visible band data, and percent frequency of detection. A separate layer indicating locations of gas flares can be provided. 10 Box 1: At broad-scale VIIRS & DMSP data look quite similar but there is no saturation with VIIRS VIIRS DMSP Cloud Free Composite 2014 Annual stable lights 2013 Allahabad (India) VIIRS Allahabad (India) DMSP Annual Cloud Free Composite Annual Cloud Free Composite 11 Box 2: How do the DMSP and VIIRS products differ? A major distinction is that the VIIRS DNB pixel footprint is 45 times smaller than the DMSP footprint (Figure 2) in the ground instantaneous field of view (GIFOV). The VIIRS DNB data are collected with a constant 742 m x 742 m pixel footprint from nadir out to the edge of scan. In contrast, the DMSP-OLS nighttime visible band starts at nadir with a 5 km x 5 km footprint (after on-board averaging) and the footprint expands toward the edge of scan. Thus, the DNB pixel footprint is at Figure 2: VIIRS and DMSP- least 45 times smaller than the OLS pixel footprint. OLS footprints compared. In addition, the newer VIIRS instrument offers significant improvements in other areas as well: • Quantization: VIIRS 14 bit versus 6 bit for DMSP. • Dynamic Range: Due to limited dynamic range, DMSP data saturate on bright lights in operational data collections. • Lower Detection Limits: VIIRS can detect dimmer lighting than DMSP. • Quantitative: VIIRS is well calibrated, the DMSP visible band has no in-flight calibration. • Multispectral: VIIRS has additional spectral bands to discriminate combustion sources from lights and to characterize the optical thickness of clouds. Table 1 below summarizes the dramatic improvements in spatial, spectral and radiometric resolution between DMSP-OLS and VIIRS DNB. Table 1: Comparison between DMSP-OLS and VIIRS DNB Attribute DMSP-OLS VIIRS DNB Orbit Sun-synchronous, ~850km Sun-synchronous, 827km Nighttime Nodal Overpass Time ~1930 UTC ~0130 UTC Swath Width 3000km 3000km Spectral Response (full-width half-maximum) Panchromatic 500-900nm Panchromatic 500-900nm Ground-projected Instantaneous Field of View 5km (nadir) / ~7km (edge) 0.740 ± 0.043km (scan) 0.755 ± 0.022km (track) Spatial Resolution (Ground Sample Distance) 2.7km; “smooth data” < 0.820km (scan) < 0.750km (track) Radiometric Quantization 6-bit 14-bit Accompanying Spectral Bands 1 (10-12µm) 11 night / 21 day Radiometric Calibration None On-Board Solar Diffuser Saturation In Urban Cores None 12 VIIRS has greatly improved the ability to measure and detect economic activity and electric power consumption for smaller geographic units than previously possible.12 With VIIRS capabilities, it is possible to better distinguish the different sources of nighttime lights (from cities and human settlements, aurora, lightning, boats, gas flares, industrial sites, fires) and filter images to remove the unwanted information. NOAA produces three global product lines using Nighttime VIIRS data: (a) VIIRS Boat Detections (VBD): Offshore detections of lights used by fishery agencies; (b) VIIRS Nightfire (VNF): Multispectral detections of fires, flares and other infrared (IR) emitters. Used for annual surveys of gas flare locations and flared gas volumes. (c) VIIRS nighttime lights (VNL): Global monthly and annual average radiances filtered to remove sunlit, moonlit, and cloudy observations. Additional filtering is also performed to remove lightning, fires, aurora, and background noise. An example is shown in Figure 3. Figure 3: Sample VIIRS nighttime lights image. Source: NOAA. 12 Beyer, R. C., Chhabra, E., Galdo, V., & Rama, M. (2018). Measuring districts' monthly economic activity from outer space. The World Bank; Bruederle, A., & Hodler, R. (2018). Nighttime lights as a proxy for human development at the local level. PloS one, 13(9), e0202231; Keola, S., Andersson, M., & Hall, O. (2015). Monitoring economic development from space: using nighttime light and land cover data to measure economic growth. World Development, 66, 322-334; Shi, K., et al. (2014). Evaluating the ability of NPP-VIIRS nighttime light data to estimate the gross domestic product and the electric power consumption of China at multiple scales: A comparison with DMSP-OLS data. Remote Sensing, 6(2), 1705-1724. 13 As described above, typical NOAA products are composites, either monthly or annual, which average out brightness values collected during cloud-free nights. To estimate electrification rates, the Earth Observation Group (EOG) of NOAA delivers to the team through University of Michigan all the daily images using the process described below and summarized in Figure 4. VIIRS Nighttime data was acquired from the NOAA Comprehensive Large Array-data Stewardship System (CLASS) archive, screened for nighttime, run through EOG’s nighttime algorithms to create ancillary information layers, orthorectified to 15 arc-second grids, and then delivered as GeoTIFF files. VIIRS data are available from the NOAA CLASS archive as “aggregate” HDF5 format files containing ~6minutes of acquired satellite imagery. For each nighttime aggregate, EOG pulled the following 10 Sensor Data Records (SDRs): DNB, M7, M8, M10, M11, M12, M13, M14, M15, and M16, the corresponding DNB and M-band geolocation files, ellipsoid geolocation (GDNBO) and terrain-corrected geolocation (GMTCO), and the VIIRS Cloud-mask Environmental Data Record (EDR). The VIIRS Day/Night Band (DNB) is the primary “nighttime lights” band, recording radiances from a broad spectral band covering the visible to near-infrared boundary (0.5-0.9µm). To assist in quantitative use of the DNB radiances for nighttime lights work, EOG provided ancillary information layers to provide context for the DNB radiance images. The full suite of orthorectified files provided is shown by file extension. File extension Description .rade9 DNB radiance image, radiance units nanowatts/cm2/sr multiplied by 10E9. .vflag companion information VIIRS flag file, containing bitfields described below .li lunar illuminance image, units lux .samples DNB along-scan sample position, unitless .lines DNB along-track line number, unitless 14 Figure 4: Processing of VIIRS Satellite Data by EOG NOAA for Electrification Access Estimates 15 The vflag file contains the results of a series of EOG algorithms packaged as a 32-bit integer into the following bit-fields, which are described further below. Flag Name, Start Bit, Number of Bits VIIRS_FLAG_CLOUD2, 2, 3 VIIRS_FLAG_ZERO_LUNAR_ILLUM, 5, 1 VIIRS_FLAG_DAY_NIGHT_TERM, 6, 2 VIIRS_FLAG_FIRE_DETECT, 8, 6 VIIRS_FLAG_STRAY_LIGHT, 14, 2 VIIRS_FLAG_CLOUD2_REJ, 18, 1 VIIRS_FLAG_DNB_LIGHTNING, 22, 2 VIIRS_FLAG_DNB_HEP, 24, 1 VIIRS_FLAG_NO_DATA, 31, 1 VIIRS_FLAG_CLOUD2 (A.K.A. VIIRS_CLOUD_MASK, VCM): The first bit is a quality flag indicating quality of the VIIRS Cloud Mask (VCM) and the second two bits are the VCM. This field can be used, for example, when it is necessary to filter out cloud-impacted data. 000 :0 Clear GOOD 100 :4 Confident Cloudy GOOD 001 :1 Clear POOR 101 :5 Confident Cloudy POOR 010 :2 Probably Cloudy GOOD 110 :6 Spare 011 :3 Probably Cloudy POOR 111 :7 NO_DATA VIIRS_FLAG_ZERO_LUNAR_ILLUM: Lunar illumination impacts brightness of the DNB and has other impacts on the data quality. This field can be used to filter out data impacted by lunar illumination. 0 :0 Lunar Illum > threshold 1 :1 Dark VIIRS_FLAG_DAY_NIGHT_TERM: For each orbit, polar orbiting satellites such as the Suomi- NPP have a daytime section, a nighttime section, and a section in transition from day to night, or night to day, called the terminator zone. This field can be used to exclude data from, for example, the day and terminator zones if high quality nighttime data is required. 00 :0 Day Zone 10 :2 Night Zone 01 :1 Terminator Zone 11 :3 NO_DATA 16 VIIRS_FLAG_FIRE_DETECT: This field can be used to understand where fires/flares are present in the DNB data if excluding fires from an analysis is required. 000000 :0 Background 0*1*** :- M12/M13 Local max 0****1 :- M10 Detection 01**** :- Valid Temperature 0***1* :- M12/M13 Detection 111111 :63 NO_DATA 0**1** :- M10 Local max VIIRS_FLAG_STRAY_LIGHT: Stray light happens when the satellite is sunlit, but the earth is not, and a small amount of “stray” light enters sensor and impacts the DNB data. These bits can be queried to understand when the DNB imagery is possibly impacted, and if so, has a correction been applied. 00 :0 No stray Light *1 :- Stray Light Impact possible 1* :- Stray Light Corrected VIIRS_FLAG_CLOUD2_REJ: The VCM incorrectly labels gas flares as clouds in the nighttime algorithm. This field can be queried to return some of those “clouds” to a “clear” state. This field is useful if an analysis requires working with gas flare locations. 0 :0 Accept as a cloud 1 :1 Reject as a cloud VIIRS_FLAG_DNB_LIGHTNING: Lightning shows up in the DNB data as bright streaks, one scan (16 lines) wide. This field can be queried for locations of lightning signatures in the DNB. 00 :0 No Lightning 10 :3 Spare 01 :1 Lightning Signature 11 :4 NO_DATA VIIRS_FLAG_DNB_HEP: This field flags locations of High Energy Particle hits recorded in the DNB data. They are most prevalent in the South Atlantic Anomaly region centered over Brazil. This field should be queried and these pixels excluded in any analyses over that region. 0 :0 clear 1 :1 HEP hit The final step in the process was to package and deliver these data to the University of Michigan’s nighttime lights data repository. The suite of 5 files, for each ~6minute aggregate, were packaged into gzip-compressed tar files and a corresponding sha1 checksum was generated. The entire VIIRS archive from April 1, 2012 through December 31, 2017 for the Suomi NPP satellite was processed and delivered 17 in satellite-year batches via the Globus file transfer application. This new Global Night Lights archive of VIIRS image files and associated metadata is massive, comprising some 250 terabytes (TB) of data across hundreds of thousands of files, describing visible band brightness, cloud cover, and associated metadata from sub-orbits capturing every corner of the globe on each night. 18 How Did We Improve the Analytical Framework? Besides the VIIRS data, the team leveraged new population and settlement estimates from CIESIN/Facebook, which identify human-built structures through computer vision techniques. The combination of high-resolution nighttime light and settlement data enables estimates of electricity access at a higher spatial resolution than ever before. Beginning with a subset of developing countries for which settlement data and ground-based census data exist for small subnational units, this report presents new methods to generate High Resolution Energy Access (HREA) estimates, describes our results, and presents validation analysis from comparisons against official administrative data on electricity access. First, we estimate electricity access by studying images of nighttime light output. Drawing upon the Global Nightlights Archive, we extract brightness levels from GeoTIFFs of VIIRS’ DNB, which measures radiance from the visible/reflective band of wavelengths between .5–.9 μm.13 The VIIRS DNB sensor has dramatically improved spatial resolution and dynamic range over its predecessor, the DMSP- OLS series. The detection limit of the VIIRS sensor is also much improved, now ~2E-11 Watts/cm2/sr as opposed to ~5E-10 Watts/cm2/sr for DMSP-OLS. Because the radiance values are so small, NOAA multiplies them by 109 in the data stream (rade9 units). Data from the VIIRS DNB sensor has a nominal spatial resolution of 750m across the full scan, which NOAA projects onto a 15 arcsecond grid, with a pixel size of roughly 430m2.14 One important feature of the VIIRS sensor is that its overpass time is now after midnight (typically within an hour or so of 01:30 local time), as compared to several hours earlier for DMSP-OLS. Thus, while VIIRS has significant technical advantages over DMSP, the overpass time is less ideal for detection of access and typical electricity use. However, where electricity is available and can be used for outdoor electrical lighting, it is plausible to assume that some lighting is left on even after midnight.15 Indeed, across the VIIRS satellite imagery, outdoor lighting is easily detectable, even in the early morning hours in many remote and sparsely populated areas. 13 For technical details, see: http://rammb.cira.colostate.edu/projects/npp/VIIRS_bands_and_bandwidths.pdf, https://www.star.nesdis.noaa.gov/smcd/spb/nsun/snpp/VIIRS/VIIRS_SDR_Users_guide.pdf 14 Raw VIIRS DNB GeoTIFFs are stored using unprojected WGS84 geodetic coordinates. The rasters are projected to relevant UTM zones and converted to polygons in order to determine the area of cells in square meters and to intersect the cells with the settlement polygons. 15 Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh (2017) VIIRS night- time lights, International Journal of Remote Sensing, 38:21, 5860-5879, DOI: 10.1080/01431161.2017.1342050 19 Our estimates and validation focus on individual access and therefore require data on the location and size of human settlements. However, in most countries, the availability, accuracy, and reliability of population data is limited. Even in cases where high quality administrative data on population is available, the data are typically reported only for relatively large spatial units like the census precinct or tract. We sidestep these issues by relying instead on new computer-generated data on built-up areas. These efforts rely on machine learning and computer vision techniques applied to very high resolution daytime imagery to identify and trace the outline of every building or cluster of buildings within a country. The building outlines are then georeferenced and, in some cases, linked with census tract population estimates to yield high resolution settlement maps that can be directly compared against other georeferenced data.16 We link VIIRS data to high resolution settlement maps along a constant spatial grid. This provides a spatially consistent reference on which to map the nightly values based on cell centroid coordinates, and with which to intersect the settlement polygons. In other words, we identify the nightly radiance values that correspond to the point coordinates in the middle of each cell and identify the settlement shapes that fit within the quadrilaterals of each cell. Relying upon currently available data from the High Resolution Settlement Layers project, we are able to run our analysis for 33 countries. Many of these are also priority countries identified by the Sustainable Energy For All (SE4ALL) Global Tracking Framework (GTF) and some of them are included in the 2019 Tracking SDG 7 report.17 The countries that have been processed to date include: Table 2: Countries with processed data as part of the Global Night Lights program. Algeria Ghana Liberia Nigeria*+^ Tanzania+^ Argentina Guatemala Madagascar+^ Philippines*+ Thailand* Benin Guinea Malawi+^ Puerto Rico Tunisia Botswana Haiti Mali^ Rwanda Uganda+^ Burkina Faso+^ Indonesia*+ Mauritania Sierra Leone Zambia Cambodia Ivory Coast Mexico* South Africa* Central African Kenya+^ Mozambique+^ Sri Lanka Republic * Fast-moving countries (among top 20 countries with greatest annual increase in electricity access, 1990-2010, SE4All Global Tracking Framework 2013, Fig O.5) 16 Prominent efforts using the settlement detection approach are being led by the Gates Foundation and the Oak Ridge National Laboratory; and Columbia’s Earth Institute and the Connectivity Lab at Facebook. 17 IEA, IRENA, UNSD, WB, WHO (2019), Tracking SDG 7: The Energy Progress Report 2019, Washington DC. 20 + High-impact countries (among top 20 countries with highest electricity access deficit 2010, SE4All Global Tracking Framework 2013, Fig O.4A) ^ The 20 countries with the Largest Electricity Access Deficit over the 2010-2017 Tracking Period (IEA, IRENA, UNSD, WB, WHO (2019), Tracking SDG 7: The Energy Progress Report 2019, Washington DC, Fig ES3) We focused our initial efforts and algorithm development on a subset of these countries for which reliable “ground-truth” data exist on electricity access, namely recent census data at the sub-national level on household electricity rates. For these countries, we acquired census data and spatially linked these to administrative regions using publicly available shapefiles.18 Table 3: Data Sources for Subnational Census Data on Electricity Access 19 Country Census Question Lowest No. of Data Source Shapefile Year (Census) Subnat'l Obs. Source Unit Cote d'Ivoire 2014 Q51 Region 58 Cote d'Ivoire Data Portal GADM Ghana 2010 H08 District 170 Ghana Statistical Service IPUMSI Philippines 2015 H1 Province 1,634 Philippine Statistical HDX Authority Rwanda 2012 H12 District 30 Rwanda Data Portal HDX Sri Lanka 2012 H4 Municipality 13,985 Lanka Stat Map HDX Tanzania 2012 G49 District 151 Census Info Tanzania HDX Radiance levels from VIIRS are recorded on all nights from April 2012 to December 2017. Individual values are subsequently dropped if they meet any of the following criteria for being considered low- quality by NOAA: a) obstructed by clouds; b) sunlit and outside the nighttime cutoff zone (i.e., above the solar zenith angle 101°); c) moonlit, with lunar illumination above .0005 lux; d) high energy particles were detected; e) obstructed by stray light (solar zenith angle at nadir between 90–118.5°); f) surface lightning was detected; or, g) gas flares were detected (temperature > 1200 K and frequency > 1%).20 If more than one high-quality value was observed on the same single night, the value recorded earliest in the night was kept. Because radiance values are heavily right-skewed (i.e., most pixels are relatively dim, but the brightest areas have extremely large positive values), and some are slightly negative (the technical 18 Of these, we omit Tunisia from our analysis and validation as the country is nearly fully electrified and has almost no sub-national variation. 19 These data are described in greater detail in the accompanying appendix (Annex 2) 20 Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh (2017) VIIRS night- time lights, International Journal of Remote Sensing, 38:21, 5860-5879. 21 minimum is -1.5), we add 2.5 to the remaining values and apply the natural logarithm, following NOAA. To generate annual estimates of average brightness, we calculate the mean radiance values across all quality nightly values for each pixel in each calendar year. Computational processing and analysis were performed on Flux, the University of Michigan’s high- performance computing cluster, using tools including R and Python for data extraction and processing; QGIS for spatial operations; and R and Stata for data analysis and visualization. Method One: Threshold-Based Classification Method Two: Nightly Comparative Analytics Classify settlements to match SE4ALL Independently classify settlements based on Electricity Access rates daily comparisons of brightness values for all settlements. • Extract nighttime light levels for all • For each night, compare brightness of each settlements on all nights. settlement against background light in • Key assumption: areas brighter than a isolated, unpopulated cells with same land critical threshold are electrified. type. • Identify optimal light threshold resulting in • Controls for nightly variation in the VIIRS classification of access matching the DNB signal due to e.g., atmospheric SE4ALL access rate in each year. conditions, lunar illumination (and interaction with land type). • For each year and each settlement, aggregate the nightly data to estimate: • Electrified settlement (0 or 1): binary • Likelihood lit (0 to 1): overall confidence a classification of each settlement as settlement is brighter than background noise electrified or not for each year. • Frequency lit (0 to 1): proportion nights a settlement is significantly brighter than background noise. Figure 5: Methods used to determine electrification. We developed two methodological approaches to generate electricity access estimates at the level of individual settlements (see Figure 5). Method One is a simple approach that identifies an optimal light output threshold to classify settlements in order to match the official electrification rate of each country in each year. The result is a binary partition of settlements that is efficient to compute but is anchored on often problematic official electrification rates. Method Two is a more sophisticated approach that identifies settlements as electrified if they are statistically determined to be brighter than similar non- settlement areas across each individual night of imagery. By aggregating the statistical comparisons across all nights in a calendar year, we generate probabilistic estimates of electricity access for every settlement while accounting for uncertainty and measurement error. In sum, the approach generates an independent partition of electrified and unelectrified settlements in every year that does not rely on prior 22 estimates or the availability and accuracy of official electrification rates. We describe the two methods more fully below. Method One: Threshold-Based Classification Our first approach relies on an estimation procedure relying on country and year specific thresholds of light output, above which a settlement has a high likelihood of having electricity access. To identify the optimal country-year thresholds, we calculate the radiance level that results in an imputed national electricity access rate that corresponds most closely to the official national electrification rate from the World Bank’s Sustainable Energy for All (SE4ALL) database. We then classify every settlement as electrified or not, based on whether its radiance level is above or below the threshold. We run the process separately for every country in each year to account for country-level differences and variation in calibration settings on the VIIRS sensor which are adjusted over time. The following three figures summarize the processing steps under Method One for Kenya and for the year 2017: Figure 6: Average Light Output. 23 Figure 7: Average Light Output overlaid with human settlements. Figure 8: Method One (Threshold-Based Classification) resulting from analysis. 24 For illustration purposes, Figure 8 above only shows the area around Mwingi, in the south eastern part of Kenya where it is easier to distinguish between electrified and un-electrified settlements. Given variations in VIIRS calibration settings and differences in atmospheric conditions and other sources of noise across geography, our analysis reveals that there is no single radiance level that consistently distinguishes anthropogenic light from background noise across time and countries. As a result, this method requires calculating different thresholds for each country-year. This approach remains advantageous for its computational efficiency and resultant classifications that yield overall access rates matching official electrification rate estimates. 25 Table 4 below shows the optimal thresholds generated by Method One for 2015. The threshold is the mean “rade9” (radiance x 109) values across all good quality VIIRS observations from the year. As is evident in the table, the range of optimal thresholds varies widely across countries. The variance indicates there is no global threshold that can be used to classify settlements in a way that would match official rates consistently across cases. In addition, the results also suggest there may be variations in electricity use patterns post-midnight. Neighboring Kenya and Tanzania, for example, have very different rade9 thresholds of .232 and 1.88. If the lower threshold of .232 were used for Tanzania, many more settlements would be classified as electrified, yielding an estimated electrification rate far higher than the official rate of 18.5 percent. The highest threshold is in Malawi, which requires a rade9 threshold of 4.79 to generate a classification that matches the official 10.8 percent electrification rate in 2015. The threshold is substantially higher than any other country in the sample. One reason is that the electrification rate in Malawi is very low and almost all electrified homes are in just a couple of large cities. The optimal threshold thus needs to be high enough that only a tenth of the populated settlements would be classified as electrified. The high threshold captures settlements in the brightest cells in the country, mostly found in the center of urban areas. A lower threshold would result in large swaths of peri-urban areas being classified as electrified, leading to too many electrified settlements compared to the official rate. The sensitivity of the classifications is partly a result of the assumption of the approach that all homes located within brightly lit cells have access to electricity. While Method One results in consistency with official estimates and is efficient to compute, it has important limitations. 26 Table 4: Optimal VIIRS Thresholds to Match Official Electrification Rates, 2015 Country Official Electrification Optimal VIIRS Rade9 Rate (Percent, SE4ALL) threshold Algeria 99.34 0.10892 Benin 40.03 0.44189 Botswana 58.53 0.23685 Burkina Faso 18.47 1.41752 Cambodia 47.57 0.11507 Central African Republic 13.38 1.38376 Ghana 75.72 0.24860 Guatemala 90.51 0.12155 Guinea 30.96 0.38492 Haiti 38.22 0.51699 Indonesia 97.54 0.06044 Ivory Coast 64.09 0.20826 Kenya 41.60 0.23156 Liberia 13.84 1.55659 Madagascar 19.04 0.06833 Malawi 10.80 4.78686 Mali 37.60 0.14164 Mauritania 39.50 0.45161 Mozambique 24.00 1.35434 Nigeria 52.50 0.28737 The Philippines 89.08 0.05060 Puerto Rico 100.00 -0.00905 Rwanda 22.80 0.23866 Sierra Leone 16.47 1.19183 South Africa 85.50 0.54098 Sri Lanka 93.89 0.21650 Tanzania 18.50 1.88039 Thailand 99.60 0.07544 Tunisia 100.00 0.02813 Uganda 18.50 0.30317 Zambia 31.10 1.55487 27 Method Two: Likelihood Estimates of Electricity Access We introduce a new method, Method Two, to generate probability estimates of electricity access for all areas with human settlements within a country by using computational methods to identify brightness values that are plausibly associated with electricity use, and statistically unlikely to be due to exogenous, non-anthropogenic factors. By repeating the process across all nights within each calendar year, we establish statistically robust estimates and generate new high-resolution data that identifies variation in the level and reliability of electricity use, even across areas with low-levels of electricity access or assumed to be electrified but lacking usable or reliable service. Every night, the VIIRS DNB sensor collects data on the observed brightness over all locations within a country, capturing both electrified and unelectrified areas as well as populated and unpopulated areas. Our objective is to classify populated areas as electrified or not using all the brightness data over a country. But the challenge is that light output can be due to multiple sources unrelated to electricity use. Notably, the VIIRS sensor is so sensitive that it picks up light from overglow, atmospheric interactions, moonlight, and due to variations in surface reflectance which varies across types of land cover. We refer collectively to these exogenous sources as background noise, which must be accounted for to classify whether an area is brighter than expected on any given night. Figure 9: Background Noise Levels in Unsettled, Unelectrified Areas Across Different Landcover Types We use data on light output detected over areas with no settlements or buildings to train a statistical model of background noise. Figure 9 shows that background noise levels vary across time. These are 28 mean brightness levels for geographically isolated, unpopulated pixels with no settlements that are unlikely to have human presence at night. Each line represents a different land type, revealing that surface reflectance varies significantly across different types of terrain like forests, savannas, or wetlands.21 The model can be used to generate an expected brightness value on every given night for any given location. We then compare the observed brightness over on each night against the expected baseline brightness value. Areas with human settlements with brighter light output than expected are assumed to have access to electricity on that night. We classify all settlements on all nights and then aggregate the estimates to generate a “likelihood electrified” estimate for each calendar year for all settlement areas. Areas that are much brighter than would be expected on most nights have the highest probability of being electrified. Areas that are as dim as areas with no settlements have the lowest probability of being electrified. Areas that are marginally brighter on some nights have middle levels of probability. Another way of looking at this is classifying settlements based on the frequency of nights when significant light output is detected in a given location. Meaning a location could be classified as lit on all nights of the year or some smaller portion of the year. The advantage of this process is that it fully uses all available nightly data from the VIIRS data stream while considering sources of known noise and variability. The process also generates probability estimates that allow for the identification of areas where the likelihood of electricity access and use is more uncertain. This is significant given that traditional binary measures of access do not account for variations in levels of use or reliability of power supply, and generally assume uniform access across areas that are nominally electrified. This data may therefore be helpful in identifying baseline variations in access and reliability within countries, consistent with the objectives of the Multi-tier Framework for measuring energy access (ESMAP22 2015). The Multi-tier Framework provides comprehensive profiles of electricity use, access, and reliability, but can do so only in settings where surveys have been fielded. Bolstering these efforts, the satellite-based approach can provide estimates of access and reliability using autonomous methods that observe light output during a limited overpass time window and at a fixed 21 Landcover types are from the Global Land Cover Facility (http://glcf.umd.edu/data/lc/). The lines are natural cubic splines plotted through the mean daily observed radiance (rade9) values colored by land type, for a sample of isolated non-settlements (max 500 sampled cells per land type, outliers > 4 sd above mean removed by land type and day). 22 The Energy Sector Management Assistance Program, a multi-donor trust fund housed at the World Bank. 29 spatial resolution but with complete geographic coverage, regular and repeated nightly data collection, and straightforward replicability. The process is summarized in Figure 10 below and explained in more detail below. Figure 10: Process Flow for Method Two for computation of Likelihood Electricity Access Estimates 30 1) Select random sample of locations with no settlements to measure background noise. We select a stratified random sample of non-settlement points drawn from the 1 arcsecond pixel grid. To maximize the N per pixel, we use stratified random sampling based on the interaction of 1) land type and 2) proportion of cells around the cell that are lit. In Ghana, this results in a sample of about 22,000 non-settlement 1 arc-second pixels. 2) Select observations. Following NOAA guidelines and their data quality flags, we drop bad quality data, including those with heavy cloud cover and excessive sensor noise. NOAA also drops many nights with high lunar illumination but we relax this threshold slightly and keep observations with modest lunar illumination to preserve more data. Furthermore, on nights with multiple overpasses, we use data with the earliest local timestamp for settlement points but allow multiple observations for non-settlement points. 3) Remove outliers. To generate a reliable estimate of background noise, we need to exclude outliers. Presumably, an unusually high brightness values in an unsettled area is not due to background noise but rather due to external, non-systematic phenomena. We do this first by stratifying by land type and date and remove observations if they are above 3 times the standard deviation above the mean. Then, for the whole sample of brightness values over non-settlement areas, we remove values that are above the 99.9th percentile. 4) Create statistical model of background noise. For each calendar year, we run a linear mixed effects model on light output for all pixels in areas with no settlements. The aim is to understand the exogenous factors that explain variation in light level for areas where there are no human settlements, and presumably no electricity. The model includes observations from all non- settlement pixels from all good quality nights, and includes controls for date, land type, lunar illumination, local time, and the proportion of surrounding cells that have a settlement. Notably, the regression diagnostics are excellent with strong linearity, few outliers, and limited heteroskedasticity. Using these statistical parameters learned from data on non-settlement areas, we then calculate the expected level of light output for all areas with settlements. These predicted values represent a counterfactual estimate of how much light would be expected on that specific day on that type of land, if the only sources of light were from background noise and other exogenous factors. Areas 31 with higher observed light output than expected light output are assumed to have electricity access. 5) Identify electrified settlement areas on each night. We compare the actual observed level of light output against the expected light output level from the model above for every settlement pixel on every night. This difference in the observed versus expected light output is our measure of anthropogenic light generation on each night. We standardize these values by dividing by the standard deviation to generate z-scores for each pixel on each night. Higher z-scores imply higher light output than expected by exogenous factors (e.g. land type, lunar illumination, etc.) alone. The key assumption is that higher scores indicate higher likelihood that a settlement is using electricity on that specific night. 6) Aggregate nightly estimates to generate “Probability Electrified” values for all settlement areas for each year. We aggregate all nightly estimates for each settlement to generate probability of electrification values for all 15 arcsecond pixels with settlements across the country. We repeat the process for all years of VIIRS data. 7) Determine a threshold above which probabilities of electrification correspond to electrified settlements For purposes of comparison against standard electrification access rate data, we can generate a binary classification of electrified settlements by setting thresholds. In the comparisons against census data below, we code settlement pixels as electrified when the z-scores are above 1.68 (corresponding to the 90% confidence interval that the settlement is brighter than background noise). The results of the process are robust and have good statistical properties. The figures that follow depict output for Kenya (Figure 11) and Ghana (Figure 12), and readily show that many urban areas have high probability scores, while areas with lower scores are often in remote and rural areas. The data also identify many settlements where electricity access is less certain, indicating that light output over settlements is only marginally higher than matched background light levels. 32 The overall process relies on the assumption that electrified areas are brighter at night than non-electrified areas. A growing body of research speaks to the plausibility of this conjecture.23 But there are nonetheless situations that violate the premise. For example, some areas without human settlements are bright on some or many nights. This includes areas with fires accompanying slash and burn agricultural practices in some rural areas. There is also light on roads from street lighting and vehicular traffic. Bright lights are also generated in some industrial areas with no or few buildings, such as around open-pit mining sites. Figure 11: Likelihood a settlement is electrified, Kenya 2017. Our electricity access estimates ignore light from these unpopulated areas, focusing only on light output over settlement areas to generate our estimates of electricity access. 23 Wang, Z., Román, M. O., Sun, Q., Molthan, A. L., Schultz, L. A., & Kalb, V. L. (2018). Monitoring disaster-related power outages using NASA black marble nighttime light product. ISPRS Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci, 1853-1856; Mann, M., Melaas, E., & Malik, A. (2016). Using VIIRS day/night band to measure electricity supply reliability: Preliminary results from Maharashtra, India. Remote Sensing, 8(9), 711; Min and Gaba 2014; Min, Gaba, Sarr, and Agalassou 2013; 33 It is also plausible that not all electrified settlements generate nighttime lighting. Outdoor lighting at night might be limited due to cost, cultural norms, or the absence or failure of lighting infrastructure. These limitations and considerations speak to the importance of understanding local context and situational factors. But overall, because of the great value humans place on illumination, electrified areas are brighter at night than unelectrified areas, all else equal. Building off this premise, this process results in new data on the location of electrified settlements in settings where problems of access and reliability are difficult to track using conventional methods. Figure 12: Likelihood a settlement is electrified, Ghana 2017. Validation In order to investigate the reliability and validity of our new HREA settlement classification methods to estimate electricity access, we compare our estimates against official data on electricity access at both the 34 national and subnational levels. Except where specified, all of the analysis below focuses on the likelihood estimates generated by Method Two. We compare HREA estimates of electrification against national-level data from the SE4ALL database. The SE4ALL data provide annual estimates of electricity access for all countries in the world. The two datasets overlap from 2012 to 2016, providing 5 years of comparison for the 33 countries in our HREA sample. Overall, the correlation between the SE4ALL data and these new satellite-derived estimates (Method Two) is very high, with an overall correlation of .95 across the 165 observations as shown in Figure 13. The results suggest that the satellite-based methods yield results that are highly comparable to official estimates. In comparison to official estimates, which are laboriously collected relying on household surveys, government estimates, and statistical imputation, the satellite-based methods rely only on autonomously collected and coded data which can be efficiently calculated as new data arrives. Since the satellite-based methods rely on high resolution data, we can also generate access estimates at lower subnational levels which are not reported in many countries. To validate these lower level estimates, we were able to collect subnational census data on household electrification rates for six countries: Côte d’Ivoire, Ghana, the Philippines, Rwanda, Sri Lanka, and Tanzania. In Figure 14, we present scatterplots showing the correlation of our estimates against subnational census data (from questions regarding the main source of household lighting) for each of the countries, typically at the district level or second-level administrative division. The plots show comparisons of satellite-based estimates of electricity access in the year closest to each country’s most recent census. Overall, the correlations are strong. The correlations are lowest in the Philippines and Sri Lanka, countries with high levels of access according to official records. While there is general agreement in the direction and rank- order of access estimates across both sources, the scatterplots indicate that many areas have higher census-reported electrification rates than what is detected by satellite. The data annex, Annex 2, presents the underlying figures used to prepare the scatterplots. Overall, the close match between the data is notable given that census estimates have their own limitations. In particular, census data is typically collected only once every decade, and so we compare satellite data either from that year or the closest year available if the last census was prior to 2012 (for example, as in Figure 15. Moreover, many census estimates of electrification rely upon surveys using national sampling schemes that may not always yield accurate estimates in smaller areas with characteristics that differ from the sampling scheme. Certain countries, such as Côte d’Ivoire, have 35 experienced political problems that led to population displacement and difficulties to access certain areas. As a result, there may be a range of measurement errors in the official data. Figure 13: Comparison of Official Vs. Night Lights Satellite Estimates (Method 2) Electrification Rates for 33 Countries (2012–2016). 36 Figure 14: Comparison of Census vs. Satellite-Based Subnational Electrification Rates (Method Two) 37 Figure 15: Comparison of Ghana Electrification Rates by District (2010 Census vs. 2012 Satellite-Derived Method Two Estimates) Detecting changes over time Having examined the relationship of satellite-based electrification access estimates against census data, we can also examine temporal trends made possible by studying changes in spatial access over the period for which we have satellite data. Figure 16 shows the electricity access rate for Ghana as a whole increased from about 70 percent to 80 percent over the 2012-17 period, while Figure 17 shows estimated electrification rates for individual districts in Ghana annually from 2012 to 2017. 38 Figure 16: Annual estimated electricity access rates for Ghana (2012-17) Figure 17: Annual estimated electricity access rates for Ghana (2012-17), by district 39 By comparing settlement classification maps across years, it is also possible to identify changes in electrification status for individual settlements over time. The map below (Figure 18) shows change in electrification status for settlements in northern Ghana: • green represents settlements that were already electrified in 2012 • blue are newly electrified settlements between 2012-17 • red are settlements that are still unelectrified in 2017 Figure 18:Changes in electrified settlements, Ghana (2012-17) We also examined additional data on average population density in settlements in different areas to make the following estimates. According to the satellite-based estimates, from 2012 to 2017, 2.5 million people gained electricity in Ghana. Much of this growth was concentrated in Upper East, Upper West, and Northern regions, where access rates were generally lowest in 2012. In Bawku Municipal district, for example, the electrification rate more than doubled from 32 to 67 percent over this timespan. Table 5 lists the 10 districts with the largest absolute gains from 2012 to 2017 in population residing in electrified areas and Table 6 lists the largest proportional gains in electrification over the same period. 40 Table 5: Top 10 Districts with Largest Gains in Population Electrified, Ghana 2012 –17 Region District Newly # HRSL Population Pct Pct Change Rank Electrified settlement Electrified Electrified Pct Pop pixels 2017 2012 Electrified (2012-17) (2012-17) 1 Upper Bawku 71,813 37,110 206,721 66.9 32.2 34.7 East Municipal 2 Northern Mamprusi 68,276 26,651 194,035 64.5 29.3 35.2 West 3 Upper Talensi 49,901 17,464 98,190 76.6 25.7 50.8 East Nabdam 4 Northern Savelugu 49,007 19,038 167,954 83.7 54.5 29.2 Nanton 5 Upper Bongo 48,170 15,908 88,125 68.3 13.6 54.7 East 6 Eastern Suhum- 48,118 19,123 162,192 91.1 61.5 29.7 Kraboa Coaltar 7 Volta North 44,933 25,923 152,824 82.9 53.5 29.4 Tongu 8 Northern Yendi 44,928 25,834 238,416 55.7 36.9 18.8 9 Brong Pru 44,595 14,676 142,734 66.6 35.3 31.2 Ahafo 10 Upper Jirapa 40,170 22,826 97,067 59.8 18.5 41.4 West Table 6: Top 10 Districts with Largest Gains in Proportion Electrified, Ghana 2012–17 Region District Newly # HRSL Population Pct Pct Change Rank Electrified settlement Electrified Electrified Pct Pop pixels 2017 2012 Electrified (2012-17) (2012-17) 1 Upper Bongo 71,813 15,908 88,125 68.3 13.6 54.7 East 2 Upper Talensi 68,276 17,464 98,190 76.6 25.7 50.8 East Nabdam 3 Upper Jirapa 49,901 22,826 97,067 59.8 18.5 41.4 West 4 Upper Nadowli 49,007 28,794 97,305 63.7 23.0 40.7 West 5 Northern Mamprusi 48,170 26,651 194,035 64.5 29.3 35.2 West 6 Upper Bawku 48,118 18,435 96,600 60.8 25.9 34.9 East West 7 Upper Lawra 44,933 27,619 107,566 59.0 24.2 34.8 West 8 Upper Bawku 44,928 37,110 206,721 66.9 32.2 34.7 East Municipal 9 Volta South 44,595 14,591 101,194 96.7 64.7 32.0 Tongu 10 Upper Lambus- 40,170 9,419 60,343 53.5 22.0 31.5 West sie 41 Overall, the results suggest that new data on nighttime light output and the location of human settlements can be reliably used to generate high resolution energy access estimates at higher spatial resolutions than previously possible. 42 What are the New Features in the Global Night Lights Platform? A new visualization platform, the Global Night Lights platform, was developed during this work. The flagship showcase for the platform is the Global Night Lights website, which provides a birds-eye view of a country’s electrical infrastructure. Using state-of-the-art mapping and visualization libraries, it shows the current level of electrification as derived from the Night Lights global dataset. A powerful Application Program Interface (API) allows users to see how that level of electrification has evolved over the course of many years, seeks to enable energy planners, policymakers and other stakeholders a better targeting and monitoring of electrification interventions. The Global Night Lights platform was launched at the World Bank Headquarters in Washington, DC, in November 2018. Improving and Iterating The Global Night Lights platform is a successor and upgrade of the first version of this project, India Night Lights. The Global Night Lights visualization platform and backend infrastructure improve upon its predecessor in key ways. Whereas the India Night Lights project used custom levels based on administrative boundaries to limit the amount of data shown on screen at once (in consideration of load times and hardware limitations), the Global Night Lights platform leans on updated mapping libraries to intelligently transition more data into view as the user scrolls and zooms. The resulting website (as shown in Figure 19) has a more fluid interface that encourages users to spend time exploring the data. Beyond improving the underlying mapping technology, the team developed a wholly new mapping strategy to address the scaling issues of visualizing a global dataset. Instead of plotting individual points as vectors, a resource-hungry process, we built a pipeline to pre-process the nighttime light data as a tiled image, or raster layer. This raster layer gave us greater flexibility to adjust the visual presentation and impact of the night time lights. Besides improving the speed and responsiveness of the platform, the visual result is a mosaic that more closely mimics what this data would look like to the naked eye, peering down from a satellite. 43 Figure 19: Screenshot of Global Night Lights website. Delivering Data at Scale As was the case with India Night Lights, the Global Night Lights visualization platform is a window into the wealth of data assembled by the World Bank in conjunction with NOAA and the University of Michigan. The team processed this data over the course of hundreds of machine hours, extracting numerical light values from individual pixels of time-series from NOAA satellite imagery spanning many years. Once this data was extracted, our academic partners at the University of Michigan normalized the data, removing spikes and outliers caused by seasonal variations and data abnormalities. They delivered the data to Development Seed in a compressed, normalized format of rich, numerical information representing the night time light outputs of a given country. Delivering this data and making it accessible to researchers and academics is a core concern of our work. It is our hope that this data can be used not only to produce more research and analysis, but in other visualization and mapping applications as well. This meant the data had to be “application-ready,” a 44 higher standard than what is often applied to open datasets where some preprocessing is expected. Figure 20 shows how data arrives and is prepared for use. CSV CSV observations positions Command Line Tool API on AWS Postgres DB Beanstalk on Amazon RDS Nightlights Website on Github Pages Map Tiles on Mapbox Figure 20: Architecture diagram for Global Night Lights platform. To meet these goals, we built a powerful yet simple API. A major value-add of our work for researchers and decision makers is the historical data; allowing you to see the development of night time light infrastructure in a given settlement and comparing it across other settlements in a given country, which is unprecedented. Our API focused on providing this value while simplifying some of the India-specific features of the India Night Lights API. As a result, we offer an easier, more digestible window into this powerful historic dataset. 45 Data Service Compared to India Night Lights, our hosting needs are much more demanding for the Global Night Lights platform as we needed server infrastructure capable of handling multiple countries worth of data. We also knew that the process would be gradual: though we would start with a small subset of countries, we wanted to build a system that could handle a truly global dataset, meaning it had to scale with little effort. To do this, we built a new data service, built from the ground up and leveraging Amazon Web Services (AWS), an industry leader in providing cloud hosting and computation resources. The Global Night Lights API uses Elastic Beanstalk, a fee-based solution that is as flexible for small datasets as it is for gigantic ones. The data itself is hosted on Amazon Relational Database Service (RDS), an AWS-managed relational database built off the flexible open source Postgres database. Using a managed solution for the database and an auto-scaling API platform allows us to support the full breadth of available night time data, as well as the full global dataset once that comes online. Improved Data Model A new data model was designed, optimizing the way that the data is stored and ingested into the database. In India Night Lights, every record contained information about the observations and the full location. While this is a straight-forward data model, it is also one with a lot of duplication. To better deal with the scale-up of the platform, the data model was revised to track positions (locations) and observations separately. This change not only reduces the size of the dataset significantly but also improves the response time from the API. Flexible Data Ingest To facilitate the way model data is loaded into the system, we developed a well-documented command line tool. This allows anybody with access to the production database to ingest data for new countries. It imports both positions and observations; and performs several basic validations to ensure the quality of the data in the database. This includes checks for the proper structure of the headers, and whether coordinates are well formatted. The command line tool allows positions and observations to be added over time. This is especially useful to update the database once new observations are generated. 46 Scalable API We also set up a new, simple API { "type": "FeatureCollection", to serve the Global Night Lights "features": [{ data. In this version, we used auto- "type": "Feature", "geometry": { scaling technology that allows us to "type": "Point", "coordinates": [-1.36992582, 8.15081722] support multiple countries' worth of }, data. The Global Night Lights API "properties": { "id": "s100451", is truly a platform that can support "x": "-1.36992582", "y": "8.15081722", global nightlights data, at a per-unit "dist2010": 720, cost scale below what was paid for "prp_sets": "0.04444444", "prp_sets_dist1": "0.875", the India Night Lights platform. "prp_sets_dist3": "0.33333333", "data": [{ This means accessible nightlights "rade9": "0.05272794", data not only for the website, but "rad": "8.5929698900000009", "li": "0", for other users, such as researchers, "lc_type": 9, "total_hh": 2481, academics, and government "e_hh": 865, planners. The API contains a single "e_rate": "0.34865", "scanned_at": "2012-04-23T01:04:00.000Z" endpoint that allows people to }, { "rade9": "0.24038182", search for the observations nearest "rad": "8.57699871", to a set of coordinates. The "li": "0", "lc_type": 9, response of the API is structured in "total_hh": 2481, "e_hh": 865, Geo JavaScript Object Notation "e_rate": "0.34865", (GeoJSON) and contains the "scanned_at": "2012-04-24T00:47:00.000Z" }] location and metadata of the } }] observations, as shown in Figure } 21. Figure 21: Sample GeoJSON query response. Extracting High-Resolution Settlement Data A primary challenge with measuring night time light levels is where to look. The specialized satellites that provide the raw imagery detects electric lighting, but also see reflections caused by clouds or moonlight; fires and gas flares; and non-settlement electric lighting, such as that from a fishing vessel. To improve our outcomes and the usefulness of our data, we had to focus on only those areas that are known to contain human settlements. 47 In India Night Lights, this settlement data was provided in the form of village locations by the national government. At a global scale, we could not rely on such privileged access. Instead, we turned again to satellites, using a satellite-derived data source that uses AI and high-resolution imagery to detect buildings. The settlement data is a product is the work of multiple public and private institutions, a group that includes Facebook. Using the data, we are able to assess electrical output only from those areas that have a high probability of containing permanent settlements, and thus potentially also electrical infrastructure. Constantly Improving From the beginning of this work, we had as a goal a global dataset. While the current visualization platform encompasses a small set of countries, it is truly capable of the larger task. Improving on some of the India-specific aspects of the first version, discovering a scalable settlement data layer, and upgrading our infrastructure to utilize cloud-native hosting solutions has allowed us to lay the groundwork. In conjunction with our partners at the University of Michigan, we have also laid out a common data transfer format that can be inserted directly into the database, and shared database credentials. As night time data for additional countries comes online, we will have no issue extending the visualization platform to show this data. 48 What Possibilities Do We Now Have? When official data are accurate, the thresholding approach (Method One) provides consistent geocoded information about the likely location of electrified settlements, which enables the quick assessment of electrification progress in a country. It also becomes possible to rapidly identify locations that have not yet been reached, enabling maps to easily visualize and better understand how access to electricity varies beyond the single national level electricity access rate of a country. For each country, the electrification access rate can be visualized by two complementary maps: (i) one that shows the status of electrification, as shown for Kenya in Figures 7 and 8 above; and (ii) another one that depicts areas for additional focus, as shown for Kenya in Figure 22 (below) – the darker the area, the higher the priority. Figure 22: Kenya – High Priority Areas by Concentration of Unlit Settlements In addition, these satellite-based layers can help refine the outputs of the least-cost geospatial planning exercise currently supported by the Bank/ESMAP in many countries. The figures below provide maps of the classification of electrified settlements for selected countries using the 2016 VIIRS nighttime lights data and the high-resolution settlement layer. Higher resolution images for 33 countries (where high 49 resolution settlements data was available at the time of analysis) are provided in Annex 1. With these maps, it is now possible to compare countries across the globe, assess electrification policies and programs and derive lessons on program effectiveness. Figure 23: Comparison between Kenya and other African Countries Figure 24: Comparison between Rwanda, Tanzania, Uganda, South Africa, Philippines and Sri Lanka 50 Figure 25: Comparison between Haiti, Burkina Faso, Algeria, Cambodia, Guatemala and Indonesia For electricity utilities, the night lights layer can also provide important contributions for the improvement of their operations. In the context of Kenya, for instance, the comparison between the electrified settlements derived from satellite imagery, the coverage of the medium voltage (MV) electricity distribution network and the GIS-based utility meter database systems (MDS) can help to locate potential gaps in service provision or data coverage. These insights can be derived by overlapping the MV electricity distribution and the night lights map (Figure 26 below), which then can lead to the identification of information gaps in service areas and ultimately to updates of the customer database. 51 Figure 27: Electric infrastructure and night lights 52 Conclusion Energy is critical to boosting shared prosperity and reducing extreme poverty. Yet despite extensive planning and monitoring efforts, many countries still lack reliable data on precisely which settlements are electrified and how the quality of service provision varies. Five years ago, the idea of using satellite imagery and night lights data for energy operations was a promising but as yet impractical idea. Since that time, numerous advances have established the reliability and usefulness of using satellite-based approaches to better track the reliability, quality and sustainability of electricity access and service provision. The methods introduced here demonstrate how new combinations of daytime and nighttime satellite data can be used to estimate electricity access at the level of individual settlements, an unprecedented level of spatial resolution. Moreover, the reliance on computational methods and autonomously collected data allow for rapid extensibility and replication of estimates to new geographic areas and temporal periods. The new data on High Resolution Electricity Access (HREA) provides insights about who has electricity and who does not at a level not possible with standard national-level access rates. During the launch of the Kenya National Electrification Strategy on December 6, 2018, Honorable Charles Keter, Cabinet Secretary, Ministry of Energy, presented slides of nighttime light over time and underscored the value of this innovative approach in his opening statement “You can see the satellite images from 2012. That is where we were. You can see how Kenya is dark. 2017, you can see the difference. Those are some of the facts that show where we’ve come from and where we are heading.” The present work thus affirms that the proliferation of satellite data, the development of new analytical tools, and the advancement of computational techniques relying on artificial intelligence and machine learning, can open new frontiers in development work. The rapid growth of data approaches that emphasize inter-operability and visualization can also accelerate global efforts to enhance electricity access and reach SDG 7. Beyond energy, the information generated and posted on the Global Night Lights Platform appears extremely useful for experts and researchers based on the high number of queries and platform visits to have access to the datasets. Several priorities motivate our next steps. To make the HREA indicators and methods as useful and accessible as possible, we will continue to integrate data into energy data platforms across the World Bank, including as geospatial layers for direct input into platforms such as www.energydata.info, the 53 Least-Cost Geospatial Electrification Planning Tool and the Global Electrification Platform (GEP) led by the SE4ALL team. As data coverage and availability grows, a priority for the team is to expand HREA estimates to cover more countries and to update the indicators for 2018 and 2019. This expansion will require additional computational resources and programming work to update the online visualization platform. As with all our efforts to date, we will continue to emphasize validation and reliability checks against other metrics of electricity use, including surveys, censuses, and other studies, as appropriate. Our team continues to refine the development of our indicators on power supply reliability. The computational estimates of frequency lit for all settlements provides a plausible indicator of power supply stability and outages but will need further validation against “ground-truth” data on voltage interruptions. Given the growing recognition of the impacts of power outages and reliability issues on social and economic welfare in much of the world, this is a priority area of concern for utilities, regulators and other stakeholders. We are also committed to building in-country expertise in the use of the HREA data. In December 2018, we led training sessions requested by the Government of Kenya to support the planning team charged with implementation of its National Electrification Strategy. We trained participants in the use of satellite data to support their access and monitoring efforts and provided full and open access to all the nighttime lights data and derived HREA indicators for Kenya. Using this model, we aim to conduct training and data sharing sessions with other interested government partners and organizations. With the miniaturization of satellites, which has significantly brought down their costs, and the proliferation of commercial businesses offering satellite services, the team believes that time has now come to finance, launch and operate a dedicated satellite to track and monitor electricity access. Based on the initial team estimates, the launch of a dedicated satellite for electricity access would cost less than US$40-50 million, which is a negligible amount compared to the billions of US$ dollars spent worldwide for electrification projects. However, it would take the leadership of an institution such as the World Bank to reach that goal. With the outreach done by the team over the years, it appears that such a decision rests entirely on the willingness of the Bank Management Team, and, of the Energy and Extractive Industries Global Practice to create the conducive environment and mobilize the resources for mainstreaming this disruptive approach into energy operations globally. 54 Annex 1: Classification of Electrified Settlements in 33 countries (Method One) Sub-Saharan Africa: Benin, Botswana, Burkina Faso, Central African Republic, Ghana, Guinea, Ivory Coast, Kenya, Liberia, Madagascar, Malawi, Mali, Mauritania, Mozambique, Nigeria, Rwanda, Sierra Leone, South Africa, Tanzania, Uganda, and Zambia 55 56 57 58 59 60 61 62 63 64 65 Middle East and North Africa: Algeria and Tunisia 66 South Asia: Sri Lanka 67 East Asia and Pacific: Cambodia, Indonesia, Philippines and Thailand 68 69 Latin America and the Caribbean: Argentina, Guatemala, Haiti, Mexico and Puerto Rico 70 71 72 Annex 2: Data Appendix This appendix contains the subnational data underlying both the scatterplot comparisons of electrification rate measures and the main validation comparisons described in the report. While the satellite data on settlements and light output are produced at a consistent pixel level across countries, census-derived data has country-to-country differences that we describe below. These include the type of administrative unit used, yearly availability of data, the wording of census questions, and several other idiosyncrasies. The census-based electrification rate (column 6) is calculated by dividing the total number of households (column 4) by the number of households with electricity (column 5). Columns labelled “Satellite” (columns 3 and 7) are derived from satellite data as described in the report. 73 Ghana Census Year: 2010 VIIRS Year: 2012 Administrative Unit: District. Ghana has 170 Districts following a redrawing of administrative regions in 2010. Administrative Unit Boundaries: Available from IPUMS. Census data description: The data are from the 2010 Population and Housing Census completed by the Ghana Statistical Service. The census covered the universe of residents of Ghana, though the data used here is based on a 10 percent extraction from IPUMS due to restrictions on data access from the Ghanaian Statistical Service. Census question: “What is the main source of lighting for your dwelling?” (H08) Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Ghana AMA 2077329 53708 47314 0.881 1 Ghana Abura-Asebu-Kwamankese 132775 3194 1905 0.596 0.875 Ghana Adaklu Anyigbe 55982 1495 574 0.384 0.506 Ghana Adansi North 113770 2584 1614 0.625 0.699 Ghana Adansi South 130513 2859 739 0.258 0.323 Ghana Adenta 86235 2257 1499 0.664 1 Ghana Afigya Kwabre 229944 3295 2316 0.703 0.895 Ghana Agona East 103323 2279 935 0.41 0.665 Ghana Agona West 123187 3120 1931 0.619 0.746 Ghana Ahafo Ano North 110481 2268 771 0.34 0.37 Ghana Ahafo Ano South 110802 2885 856 0.297 0.418 Ghana Ahanta West 107442 3002 1871 0.623 0.821 Ghana Ajumako-Enyan-Essiam 167920 3879 2318 0.598 0.851 Ghana Akatsi 142785 4060 1141 0.281 0.65 Ghana Akwapim North 158641 3549 2032 0.573 0.785 Ghana Akwapim South 116710 3333 2072 0.622 0.938 Ghana Akyemansa 106814 2311 1224 0.53 0.629 Ghana Amansie Central 73615 2497 847 0.339 0.497 Ghana Amansie West 146183 3132 1846 0.589 0.579 Ghana Aowin or Suaman 129977 3333 1265 0.38 0.183 Ghana Asante Akim North 142401 3416 2125 0.622 0.672 Ghana Asante Akim South 127215 2832 1148 0.405 0.598 74 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Ghana Ashaiman 105466 5185 4437 0.856 1 Ghana Asikuma-Odoben -Brakwa 120423 2933 1181 0.403 0.624 Ghana Assin North 170417 3902 1766 0.453 0.588 Ghana Assin South 122963 2564 975 0.38 0.601 Ghana Asunafo North 129318 2928 1356 0.463 0.399 Ghana Asunafo South 109279 2134 684 0.321 0.345 Ghana Asuogyaman 107579 2616 1742 0.666 0.869 Ghana Asutifi 112764 2473 1103 0.446 0.508 Ghana Atebubu 131938 2171 939 0.433 0.478 Ghana Atiwa 135797 2825 1684 0.596 0.664 Ghana Atwima Kwanwoma 101468 2268 1541 0.679 0.987 Ghana Atwima Mponua 132622 2854 755 0.265 0.419 Ghana Atwima Nwabiagya 189785 3625 2493 0.688 0.955 Ghana Awutu Senya 207687 5063 3402 0.672 0.977 Ghana Bawku Municipal 206721 3330 1113 0.334 0.312 Ghana Bawku West 96600 1602 250 0.156 0.285 Ghana Bekwai Municipal 146140 2995 1935 0.646 0.939 Ghana Berekum 147264 3286 2452 0.746 0.735 Ghana Bia 121772 2762 809 0.293 0.209 Ghana Biakoye 92052 1662 759 0.457 0.502 Ghana Birim Central Municipal 149751 3939 2337 0.593 0.799 Ghana Birim North 94764 2301 1168 0.508 0.726 Ghana Birim South 160100 3157 1715 0.543 0.791 Ghana Bole 79530 1135 248 0.219 0.36 Ghana Bolgatanga Municipal 141920 2797 1455 0.52 0.695 Ghana Bongo 88125 1547 185 0.12 0.119 Ghana Bosome Freho 55565 1423 628 0.441 0.682 Ghana Bosumtwi 109434 2505 1626 0.649 0.947 Ghana Builsa 99099 1742 312 0.179 0.214 Ghana Bunkpurugu Yonyo 149496 1832 192 0.105 0.348 Ghana Cape Coast 196312 4344 3697 0.851 0.99 Ghana Chereponi 57127 729 148 0.203 0.242 Ghana Dangme East 140745 2957 1718 0.581 0.662 Ghana Dangme West 136210 2980 1609 0.54 0.875 Ghana Dormaa East 70077 1231 687 0.558 0.465 Ghana Dormaa Municipal 166236 3975 1769 0.445 0.513 Ghana East Akim 190669 4681 2857 0.61 0.902 75 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Ghana East Gonja 198587 2161 629 0.291 0.176 Ghana Effutu 77652 1919 1442 0.751 0.998 Ghana Ejisu Juaben 167863 3509 2331 0.664 0.947 Ghana Ejura Sekye Dumasi 85516 1698 804 0.473 0.545 Ghana Ellembelle 85063 2010 1435 0.714 0.634 Ghana Fanteakwa 115791 2884 1097 0.38 0.454 Ghana Ga East 311554 7051 5526 0.784 1 Ghana Ga West 436601 7137 5780 0.81 1 Ghana Garu Tempane 141936 1812 198 0.109 0.099 Ghana Gomoa East 275999 6019 3997 0.664 0.976 Ghana Gomoa West 124838 3733 2349 0.629 0.918 Ghana Gonja Central 87950 1370 185 0.135 0.232 Ghana Gushiegu 115336 1155 210 0.182 0.362 Ghana Ho 394433 8216 5149 0.627 0.835 Ghana Hohoe 268528 7291 4286 0.588 0.736 Ghana Jaman North 93549 1754 716 0.408 0.551 Ghana Jaman South 91615 2173 1220 0.561 0.453 Ghana Jasikan 36488 1700 708 0.416 0.469 Ghana Jirapa 97067 1554 277 0.178 0.186 Ghana Jomoro 164348 3731 2457 0.659 0.798 Ghana Juabeso 109051 2633 917 0.348 0.302 Ghana Kadjebi 56275 1470 649 0.441 0.414 Ghana Karaga 107030 799 133 0.166 0.288 Ghana Kasena Nankana East 103536 2086 583 0.279 0.48 Ghana Kasena Nankana West 86915 1314 168 0.128 0.263 Ghana Keta Municipal 154086 4189 1671 0.399 0.824 Ghana Ketu North 104417 3075 897 0.292 0.772 Ghana Ketu South 154882 4247 1802 0.424 0.882 Ghana Kintampo North 107392 2090 735 0.352 0.43 Ghana Kintampo South 79645 1682 330 0.196 0.389 Ghana Komenda-Edina-Eguafo-Abirem 152312 3925 2694 0.686 0.944 Ghana Kpandai 73435 1815 499 0.275 0.354 Ghana Krachi East 135743 2452 1057 0.431 0.271 Ghana Krachi West 142964 2575 937 0.364 0.299 Ghana Kumasi Metropolitan Assembly 2495950 53304 46916 0.88 1 (KMA) Ghana Kwabre East 51906 2909 2256 0.776 1 Ghana Kwaebibirem 193373 5169 3096 0.599 0.714 76 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Ghana Kwahu East 68051 2030 876 0.432 0.707 Ghana Kwahu North (Afram Plains) 268125 5763 1225 0.213 0.264 Ghana Kwahu South 77652 1896 896 0.473 0.655 Ghana Kwahu West 91388 2465 1502 0.609 0.744 Ghana Lambussie 60343 830 143 0.172 0.238 Ghana Lawra 107566 1798 488 0.271 0.241 Ghana Ledzokuku or Krowor 290617 6414 5719 0.892 1 Ghana Lower Manya Krobo 81640 2340 1668 0.713 0.851 Ghana Mampong Municipal 91842 2049 1249 0.61 0.673 Ghana Mamprusi West 194035 2051 545 0.266 0.287 Ghana Mamprusi East 130330 1446 539 0.373 0.473 Ghana Mfantsiman 213778 5359 3493 0.652 0.886 Ghana Mpohor-Wassa East 116414 3442 1396 0.406 0.413 Ghana Nadowli 97305 1680 345 0.205 0.226 Ghana Nanumba North 141445 1885 539 0.286 0.507 Ghana Nanumba South 148654 1349 367 0.272 0.4 Ghana New Juaben Municipal 200489 5346 4322 0.808 0.98 Ghana Nkoranza North 75765 1512 702 0.464 0.429 Ghana Nkoranza South 113568 2336 1299 0.556 0.567 Ghana Nkwanta North 76769 1082 472 0.436 0.3 Ghana Nkwanta South 123426 2464 864 0.351 0.217 Ghana North Dayi 100965 2566 1475 0.575 0.874 Ghana North Tongu 152824 3615 1177 0.326 0.507 Ghana Nzema East 57753 1521 690 0.454 0.299 Ghana Obuasi Municipal 159551 4347 3792 0.872 0.992 Ghana Offinso Municipal 77226 1622 900 0.555 0.699 Ghana Offinso North 47714 1181 539 0.456 0.445 Ghana Prestea or Huni Valley 170418 4443 2503 0.563 0.515 Ghana Pru 142734 2481 865 0.349 0.351 Ghana Saboba 70002 946 180 0.19 0.288 Ghana Savelugu Nanton 167954 1525 626 0.41 0.537 Ghana Sawla-Tuna-Kalba 100882 1650 249 0.151 0.151 Ghana Sefwi-Akontombra 98995 1976 426 0.216 0.268 Ghana Sefwi-Bibiani-Ahwiaso-Bekwai 130865 2926 2021 0.691 0.839 Ghana Sefwi-Wiawso 146652 3322 1760 0.53 0.634 Ghana Sekondi-Takoradi 629906 15301 13320 0.871 0.996 Ghana Sekyere Afram Plains 98956 2141 849 0.397 0.412 77 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Ghana Sekyere Central 71741 1560 510 0.327 0.548 Ghana Sekyere East 35846 1483 867 0.585 0.819 Ghana Sekyere South 134998 2067 1246 0.603 0.771 Ghana Sene 138381 2440 772 0.316 0.396 Ghana Shama 172714 2150 1388 0.646 0.964 Ghana Sissala East 72580 920 315 0.342 0.261 Ghana Sissala West 40988 773 302 0.391 0.36 Ghana South Dayi 47324 1210 604 0.499 0.803 Ghana South Tongu 101194 2382 812 0.341 0.588 Ghana Suhum-Kraboa Coaltar 162192 4273 1718 0.402 0.528 Ghana Sunyani Municipal 111963 3065 2286 0.746 0.881 Ghana Sunyani West 106517 2082 1357 0.652 0.742 Ghana Tain 116999 2160 968 0.448 0.466 Ghana Talensi Nabdam 98190 2259 188 0.083 0.249 Ghana Tamale Metro 404642 6308 4652 0.737 0.921 Ghana Tano North 91142 1983 1044 0.526 0.591 Ghana Tano South 74719 1737 934 0.538 0.609 Ghana Tarkwa Nsuaem 77239 2307 1610 0.698 0.761 Ghana Techiman 214765 5020 3509 0.699 0.821 Ghana Tema 516110 10412 8276 0.795 1 Ghana Tolon Kumbugu 99465 1273 481 0.378 0.427 Ghana Twifo-Heman-Lower Denkyira 119656 2745 1480 0.539 0.472 Ghana Upper Denkyira East 69556 1522 834 0.548 0.489 Ghana Upper Denkyira West 69874 1481 762 0.515 0.537 Ghana Upper Manya Krobo 74648 1747 373 0.214 0.253 Ghana Wa East 102928 1183 105 0.089 0.085 Ghana Wa Municipal 87693 2002 1376 0.687 0.776 Ghana Wa West 75535 1288 115 0.089 0.122 Ghana Wassa Amenfi East 94918 2096 891 0.425 0.317 Ghana Wassa Amenfi West 148986 3870 1640 0.424 0.239 Ghana Weija (Ga South) 477419 13498 9152 0.678 0.999 Ghana Wenchi 86430 2025 1040 0.514 0.584 Ghana West Akim 211597 4967 2611 0.526 0.657 Ghana West Gonja 98625 1332 381 0.286 0.303 Ghana Yendi 238416 2318 610 0.263 0.364 Ghana Yilo Krobo 86493 2209 1159 0.525 0.596 Ghana Zabzugu Tatali 147767 1448 271 0.187 0.303 78 Cote D’Ivoire Census Year: 2014 VIIRS Year: 2014 Administrative Unit: The most disaggregated unit available from the census corresponds to the former 58 départements of Côte d’Ivoire prior to administrative reorganization in 2005. Administrative Unit Boundaries: Available from GADM. Census data description: Census data are from the 2014 Population and Housing Census. Census question : “Principal mode d’éclairage.” (Q51) Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Côte d’Ivoire Abengourou 391835 72518 20377 0.281 0.542 Côte d’Ivoire Abidjan 4698470 1016386 533611 0.525 0.966 Côte d’Ivoire Aboisso 387805 67025 16306 0.243 0.475 Côte d’Ivoire Adiake 65063 16402 6667 0.406 0.779 Côte d’Ivoire Adzope 388029 68167 20744 0.304 0.569 Côte d’Ivoire Agboville 284875 56032 17434 0.311 0.634 Côte d’Ivoire Agnibilekrou 169298 32102 6909 0.215 0.637 Côte d’Ivoire Alepe 84289 21110 5138 0.243 0.612 Côte d’Ivoire Bangolo 308760 60983 2821 0.046 0.15 Côte d’Ivoire Beoumi 126207 25623 3225 0.126 0.534 Côte d’Ivoire Biankouma 191484 44224 4808 0.109 0.302 Côte d’Ivoire Bocanda 157618 20417 1517 0.074 0.566 Côte d’Ivoire Bondoukou 390439 88013 9925 0.113 0.442 Côte d’Ivoire Bongouanou 360609 55021 11957 0.217 0.672 Côte d’Ivoire Bouafle 417108 71793 9031 0.126 0.585 Côte d’Ivoire Bouake 755307 129698 61817 0.477 0.807 Côte d’Ivoire Bouna 266038 42088 3232 0.077 0.228 Côte d’Ivoire Boundiali 257833 32968 5394 0.164 0.51 Côte d’Ivoire Dabakala 191552 32957 2033 0.062 0.315 Côte d’Ivoire Dabou 228693 45598 21920 0.481 0.819 Côte d’Ivoire Daloa 685467 120039 24778 0.206 0.544 Côte d’Ivoire Danane 460461 90130 12236 0.136 0.36 Côte d’Ivoire Daoukro 160339 24929 5296 0.212 0.652 Côte d’Ivoire Dimbokro 91369 15184 6905 0.455 0.726 Côte d’Ivoire Divo 638484 102731 19772 0.192 0.477 Côte d’Ivoire Duekoue 420881 69810 7391 0.106 0.39 79 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Côte d’Ivoire Ferkessedougou 474209 70440 10174 0.144 0.375 Côte d’Ivoire Gagnoa 601384 108002 21397 0.198 0.768 Côte d’Ivoire Grand-Bassam 216539 38726 18123 0.468 0.675 Côte d’Ivoire Grand-Lahou 144549 24757 2543 0.103 0.359 Côte d’Ivoire Guiglo 405704 82798 6753 0.082 0.367 Côte d’Ivoire Issia 338379 52905 7457 0.141 0.493 Côte d’Ivoire Jacqueville 49107 11115 5180 0.466 0.617 Côte d’Ivoire Katiola 238526 40210 10969 0.273 0.548 Côte d’Ivoire Korhogo 772531 131374 20539 0.156 0.598 Côte d’Ivoire Lakota 196180 38547 3809 0.099 0.6 Côte d’Ivoire M'Bahiakro 154346 25665 2518 0.098 0.406 Côte d’Ivoire Man 547389 104008 26553 0.255 0.506 Côte d’Ivoire Mankono 387162 59781 4190 0.07 0.242 Côte d’Ivoire Odienne 290058 32011 9296 0.29 0.41 Côte d’Ivoire Oume 282883 43044 6819 0.158 0.664 Côte d’Ivoire Sakassou 99036 15395 3411 0.222 0.64 Côte d’Ivoire San-Pedro 630588 121608 15889 0.131 0.446 Côte d’Ivoire Sassandra 377303 69788 3996 0.057 0.167 Côte d’Ivoire Seguela 276526 41978 8798 0.21 0.371 Côte d’Ivoire Sinfra 238409 34098 4952 0.145 0.568 Côte d’Ivoire Soubre 952346 171327 13552 0.079 0.454 Côte d’Ivoire Tabou 212165 39269 5976 0.152 0.23 Côte d’Ivoire Tanda 273348 65069 7610 0.117 0.484 Côte d’Ivoire Tengrela 120528 13393 2460 0.184 0.517 Côte d’Ivoire Tiassale 247042 48804 9715 0.199 0.561 Côte d’Ivoire Tiebissou 192030 32490 3353 0.103 0.824 Côte d’Ivoire Touba 180585 32875 7534 0.229 0.392 Côte d’Ivoire Toulepleu 56717 12228 2158 0.176 0.438 Côte d’Ivoire Toumodi 150865 27109 7245 0.267 0.758 Côte d’Ivoire Vavoua 395935 56467 4884 0.086 0.208 Côte d’Ivoire Yamoussoukro 353882 71137 30652 0.431 0.94 Côte d’Ivoire Zuenoula 234465 33130 4983 0.15 0.452 80 The Philippines Census Year: 2015 VIIRS Year: 2015 Administrative Unit: Province. The Philippines has 81 Provinces, but our comparisons rely on boundaries reflecting 88 Provinces based on the units used in the 2015 census. Administrative Unit Boundaries: Available from HDX. Census data description: The data are from the 2015 Census of Population completed by the Philippines Statistical Authority. The census covered the universe of Filipino households. We aggregate census data up from the smallest administrative unit (Municipality) as this poses a more direct comparison to the census data from other countries in this report. Census question: “What type of fuel does your household use for lighting?” (H1) Country Admin Unit Unit Pop Total HHs HHs w/ Elec rate Elec rate (HRSL) (Census) Elec (Census) (Satellite) (Census) Philippines Abra 242473 52929 46491 0.878 0.294 Philippines Agusan del Norte 672785 152253 134795 0.885 0.651 Philippines Agusan del Sur 748987 141797 110682 0.781 0.331 Philippines Aklan 572284 135758 127526 0.939 0.587 Philippines Albay 1276681 276352 245806 0.889 0.671 Philippines Antique 548379 121010 110602 0.914 0.439 Philippines Apayao 126553 25560 17343 0.679 0.076 Philippines Aurora 208265 49410 44874 0.908 0.281 Philippines Basilan 162330 71649 39627 0.553 0.189 Philippines Bataan 715050 166152 159743 0.961 0.978 Philippines Batanes 13469 4761 4695 0.986 0.438 Philippines Batangas 2571694 533377 516216 0.968 0.926 Philippines Benguet 824051 196825 187923 0.955 0.755 Philippines Biliran 157863 38518 35154 0.913 0.548 Philippines Bohol 1234340 332313 291428 0.877 0.426 Philippines Bukidnon 1526296 283596 206939 0.73 0.397 Philippines Bulacan 3042824 687054 669259 0.974 0.968 Philippines Cagayan 1213144 272893 248064 0.909 0.433 Philippines Camarines Norte 573968 134587 120541 0.896 0.475 Philippines Camarines Sur 2011351 395822 338741 0.856 0.588 Philippines Camiguin 80171 20080 18874 0.94 0.378 Philippines Capiz 738054 171515 150948 0.88 0.435 81 Country Admin Unit Unit Pop Total HHs HHs w/ Elec rate Elec rate (HRSL) (Census) Elec (Census) (Satellite) (Census) Philippines Catanduanes 237311 105450 94441 0.896 0.32 Philippines Cavite 3541223 861598 833413 0.967 0.993 Philippines Cebu 4595860 1055646 968705 0.918 0.855 Philippines City of Isabela 98773 22916 19961 0.871 0.459 Philippines Compostela Valley 775993 157491 116947 0.743 0.367 Philippines Cotabato 1413239 305027 214111 0.702 0.277 Philippines Cotabato City 288067 58866 54320 0.923 0.958 Philippines Davao del Norte 1114829 204122 173575 0.85 0.735 Philippines Davao del Sur 2163682 566646 516396 0.911 0.809 Philippines Davao Occidental 292832 70737 30748 0.435 0.127 Philippines Davao Oriental 545609 121819 90830 0.746 0.313 Philippines Dinagat Islands 109074 37978 29612 0.78 0.237 Philippines Eastern Samar 409451 108758 98124 0.902 0.484 Philippines Guimaras 173926 35225 27931 0.793 0.319 Philippines Ifugao 207484 43281 36693 0.848 0.11 Philippines Ilocos Norte 597718 139336 137511 0.987 0.571 Philippines Ilocos Sur 684984 159432 155555 0.976 0.674 Philippines Iloilo 2368401 615304 568552 0.924 0.591 Philippines Isabela 1645592 380175 357572 0.941 0.482 Philippines Kalinga 239468 42435 38339 0.903 0.195 Philippines La Union 762304 170515 162440 0.953 0.682 Philippines Laguna 3019508 750497 727442 0.969 0.974 Philippines Lanao del Norte 1010677 213074 175274 0.823 0.525 Philippines Lanao del Sur 689312 172526 122639 0.711 0.428 Philippines Leyte 1800268 477896 395179 0.827 0.585 Philippines Maguindanao 727623 194507 110902 0.57 0.227 Philippines Marinduque 211230 51603 44444 0.861 0.271 Philippines Masbate 856947 193273 126393 0.654 0.269 Philippines Misamis Occidental 603974 173394 155385 0.896 0.474 Philippines Misamis Oriental 1562564 345891 309715 0.895 0.783 Philippines Mountain Province 164844 35175 32896 0.935 0.101 Philippines NCR, City of Manila, First 467645 4406532 4343279 0.986 1 District Philippines NCR, Fourth District 3279100 901443 888643 0.986 1 Philippines NCR, Second District 4712489 1090873 1073673 0.984 1 Philippines NCR, Third District 2908977 668014 655948 0.982 1 Philippines Negros Occidental 2887266 726004 631129 0.869 0.65 82 Country Admin Unit Unit Pop Total HHs HHs w/ Elec rate Elec rate (HRSL) (Census) Elec (Census) (Satellite) (Census) Philippines Negros Oriental 1356591 294538 195330 0.663 0.459 Philippines Northern Samar 598316 151572 129233 0.853 0.42 Philippines Nueva Ecija 2169718 464592 443440 0.954 0.713 Philippines Nueva Vizcaya 464940 112223 97600 0.87 0.393 Philippines Occidental Mindoro 485972 82425 65106 0.79 0.31 Philippines Oriental Mindoro 841773 197254 170253 0.863 0.342 Philippines Palawan 1053456 303480 233359 0.769 0.293 Philippines Pampanga 2428302 527132 513793 0.975 0.949 Philippines Pangasinan 3001416 670157 636434 0.95 0.818 Philippines Quezon 2082084 478135 417215 0.873 0.624 Philippines Quirino 203941 43506 37469 0.861 0.187 Philippines Rizal 2761198 605530 579309 0.957 0.952 Philippines Romblon 260366 114338 102389 0.895 0.156 Philippines Samar 692625 172440 151149 0.877 0.388 Philippines Sarangani 528182 124313 84964 0.683 0.319 Philippines Siquijor 90978 21375 17706 0.828 0.274 Philippines Sorsogon 755939 149993 138247 0.922 0.466 Philippines South Cotabato 1482896 350408 314587 0.898 0.722 Philippines Southern Leyte 353380 125053 111932 0.895 0.43 Philippines Sultan Kudarat 863469 172879 127971 0.74 0.315 Philippines Sulu 475516 143806 56495 0.393 0.235 Philippines Surigao del Norte 447605 140577 121282 0.863 0.61 Philippines Surigao del Sur 563970 164756 147371 0.894 0.409 Philippines Tarlac 1307976 308220 295817 0.96 0.856 Philippines Tawi-Tawi 166182 67529 30096 0.446 0.179 Philippines Zambales 794953 202617 190049 0.938 0.832 Philippines Zamboanga del Norte 1012470 232643 167173 0.719 0.309 Philippines Zamboanga del Sur 1843463 433137 347057 0.801 0.549 Philippines Zamboanga Sibugay 625197 150889 117470 0.779 0.263 83 Rwanda Census Year: 2012 VIIRS Year: 2012 Administrative Unit: Districts. Rwanda has 30 districts. Administrative Unit Boundaries: Available from HDX. Census data description: Census data is derived from the 2012 Population and Housing Census completed by the National Institute of Statistics of Rwanda. Estimates here are based on a 10 percent sample. Column 3 of the following table was not provided by the Institute of Statistics and is therefore calculated based on the supplied total households (Column 4) and electrification rate (Column 5). Census question: “What is the main source of energy the household uses for lighting?” (H12) Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Rwanda Bugesera 403506 85369 8195 0.096 0.069 Rwanda Burera 340412 73624 4638 0.063 0.029 Rwanda Gakenke 348050 79760 2871 0.036 0.023 Rwanda Gasabo 631315 137146 82425 0.601 0.68 Rwanda Gatsibo 511428 96320 11655 0.121 0.048 Rwanda Gicumbi 422904 86075 5251 0.061 0.064 Rwanda Gisagara 345997 77259 1854 0.024 0.035 Rwanda Huye 357290 77915 10441 0.134 0.157 Rwanda Kamonyi 376841 80468 6196 0.077 0.082 Rwanda Karongi 347571 73326 5206 0.071 0.075 Rwanda Kayonza 406105 80517 13366 0.166 0.091 Rwanda Kicukiro 360210 77238 58624 0.759 0.592 Rwanda Kirehe 385863 77879 9345 0.12 0.023 Rwanda Muhanga 332794 75207 10980 0.146 0.174 Rwanda Musanze 395465 84756 18646 0.22 0.206 Rwanda Ngoma 380160 79647 7885 0.099 0.067 Rwanda Ngororero 357461 78963 3711 0.047 0.034 Rwanda Nyabihu 303875 65855 6190 0.094 0.034 Rwanda Nyagatare 576034 105365 19493 0.185 0.061 Rwanda Nyamagabe 366948 74848 5239 0.07 0.052 Rwanda Nyamasheke 405160 82054 6564 0.08 0.062 Rwanda Nyanza 357649 77522 5116 0.066 0.083 Rwanda Nyarugenge 319217 72280 52331 0.724 0.33 Rwanda Nyaruguru 316571 63613 1781 0.028 0.016 Rwanda Rubavu 442985 88849 23900 0.269 0.459 84 Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Rwanda Ruhango 351412 76968 6619 0.086 0.061 Rwanda Rulindo 301638 67453 4115 0.061 0.061 Rwanda Rusizi 418339 83180 17052 0.205 0.213 Rwanda Rutsiro 352242 71267 2566 0.036 0.012 Rwanda Rwamagana 354714 74175 10904 0.147 0.164 85 Sri Lanka Census Year: 2012 VIIRS Year: 2012 Administrative Unit: Municipality. At the time of the 2012 census Sri Lanka had 331 municipalities. Administrative Unit Boundaries: Available from HDX. Census data description: Data are derived from the 2012 Census of Population and Housing carried out by the Sri Lankan Department of Census and Statistics, originally reported at the ward level. We aggregate census data up from the smallest administrative unit (Ward) as this poses a more direct comparison to the census data from other countries in this report. Census question: “Principal type of lighting” (H4) Country Admin Unit Unit Pop Total HHs HHs w/ Elec rate Elec rate (Satellite) (Census) Elec (Census) (Satellite) (Census) Sri Lanka Addalachchenai 35192 9946 8591 0.864 0.918 Sri Lanka Agalawatta 36912 9649 8804 0.912 0.364 Sri Lanka Akkaraipattu 30419 9562 8869 0.928 0.933 Sri Lanka Akmeemana 76942 19736 18585 0.942 0.973 Sri Lanka Akurana 59299 13546 12749 0.941 0.825 Sri Lanka Akuressa 56875 13595 12529 0.922 0.23 Sri Lanka Alawwa 65237 17262 15246 0.883 0.509 Sri Lanka Alayadiwembu 22435 5975 4095 0.685 0.808 Sri Lanka Ambagamuwa 205159 50811 46054 0.906 0.55 Sri Lanka Ambalangoda 56348 15032 14447 0.961 0.993 Sri Lanka Ambalantota 74830 18804 16825 0.895 0.672 Sri Lanka Ambanganga Korale 15242 4502 3391 0.753 0.023 Sri Lanka Ambanpola 24120 6552 4816 0.735 0.22 Sri Lanka Ampara 35421 10023 8630 0.861 0.579 Sri Lanka Anamaduwa 39593 10879 7439 0.684 0.208 Sri Lanka Angunakolapelessa 49838 12826 11210 0.874 0.21 Sri Lanka Arachchikattuwa 42272 11465 9559 0.834 0.796 Sri Lanka Aranayaka 69132 18548 16455 0.887 0.388 Sri Lanka Athuraliya 29499 8180 7587 0.928 0.237 Sri Lanka Attanagalla 182801 45213 43517 0.962 0.991 Sri Lanka Ayagama 31348 8415 5871 0.698 0.063 Sri Lanka Badalkumbura 40857 10863 8150 0.75 0.053 Sri Lanka Baddegama 76053 19672 18081 0.919 0.927 86 Sri Lanka Badulla 70108 18835 18011 0.956 0.806 Sri Lanka Balangoda 82215 21423 18109 0.845 0.466 Sri Lanka Balapitiya 68479 17142 16340 0.953 0.996 Sri Lanka Bamunakotuwa 39448 9719 8487 0.873 0.507 Sri Lanka Bandaragama 112702 27263 26526 0.973 1 Sri Lanka Bandarawela 65155 16774 16065 0.958 0.641 Sri Lanka Beliatta 57560 14453 13473 0.932 0.446 Sri Lanka Bentota 48899 12659 11877 0.938 0.989 Sri Lanka Beruwala 164526 36483 34561 0.947 1 Sri Lanka Bibile 41129 10924 7679 0.703 0.166 Sri Lanka Bingiriya 62872 17243 15100 0.876 0.605 Sri Lanka Biyagama 191912 49008 47962 0.979 1 Sri Lanka Bope-Poddala 52378 12463 12074 0.969 1 Sri Lanka Bulathkohupitiya 47022 12670 10612 0.838 0.346 Sri Lanka Bulathsinhala 65371 17261 14698 0.852 0.464 Sri Lanka Buttala 54375 14170 11263 0.795 0.251 Sri Lanka Chilaw 55758 16608 15118 0.91 0.985 Sri Lanka Colombo 45446 68245 65133 0.954 1 Sri Lanka Damana 40950 10057 7630 0.759 0.131 Sri Lanka Dambulla 74811 19235 15469 0.804 0.489 Sri Lanka Dankotuwa 62936 17067 16033 0.939 1 Sri Lanka Dehiattakandiya 64091 15907 12703 0.799 0.15 Sri Lanka Dehiovita 82987 21068 17813 0.846 0.585 Sri Lanka Dehiwala 41643 22691 22552 0.994 1 Sri Lanka Delft 2986 930 232 0.249 0.237 Sri Lanka Delthota 30996 7628 6580 0.863 0.437 Sri Lanka Deraniyagala 46286 12387 8842 0.714 0.202 Sri Lanka Devinuwara 45316 11836 11321 0.956 0.867 Sri Lanka Dickwella 54402 13484 12733 0.944 0.882 Sri Lanka Dimbulagala 78142 21922 16655 0.76 0.322 Sri Lanka Divulapitiya 148297 38841 35713 0.919 0.996 Sri Lanka Dodangoda 63567 16116 14888 0.924 0.909 Sri Lanka Doluwa 47685 13131 11499 0.876 0.451 Sri Lanka Dompe 156810 39526 37153 0.94 1 Sri Lanka Eheliyagoda 74314 18171 16293 0.897 0.63 Sri Lanka Ehetuwewa 26255 7318 5604 0.766 0.125 Sri Lanka Elahera 44570 12310 9711 0.789 0.274 Sri Lanka Elapatha 40004 9916 8323 0.839 0.312 Sri Lanka Ella 44797 11823 10682 0.903 0.302 87 Sri Lanka Elpitiya 65316 17371 16094 0.926 0.685 Sri Lanka Embilipitiya 137386 36263 29659 0.818 0.51 Sri Lanka Eravur Pattu 81232 17057 8926 0.523 0.73 Sri Lanka Eravur Town 11295 6147 5577 0.907 1 Sri Lanka Galenbindunuwewa 44381 11593 9653 0.833 0.059 Sri Lanka Galewela 72137 19169 15933 0.831 0.304 Sri Lanka Galgamuwa 56934 15600 12075 0.774 0.321 Sri Lanka Galigamuwa 76866 19667 17824 0.906 0.818 Sri Lanka Galle Four Gravets 87239 22613 22200 0.982 1 Sri Lanka Galnewa 34938 9098 7704 0.847 0.188 Sri Lanka Gampaha 200269 51499 50452 0.98 1 Sri Lanka Ganewatta 41109 11146 8831 0.792 0.191 Sri Lanka Ganga Ihala Korale 55688 13903 12618 0.908 0.437 Sri Lanka Giribawa 32570 9034 5728 0.634 0.049 Sri Lanka Godakawela 76596 19861 16310 0.821 0.445 Sri Lanka Gomarankadawala 8120 2184 1379 0.631 0.213 Sri Lanka Gonapeenuwala 22888 5754 5456 0.948 0.991 Sri Lanka Habaraduwa 63765 15644 14836 0.948 1 Sri Lanka Hakmana 29571 8025 7531 0.938 0.374 Sri Lanka Haldummulla 38223 10270 7863 0.766 0.256 Sri Lanka Hali-Ela 92932 24057 21943 0.912 0.243 Sri Lanka Hambantota 60511 14718 12851 0.873 0.769 Sri Lanka Hanguranketha 90457 24144 20762 0.86 0.245 Sri Lanka Haputale 49414 12629 11813 0.935 0.641 Sri Lanka Harispattuwa 94823 22603 21669 0.959 0.886 Sri Lanka Hatharaliyadda 31084 8129 7295 0.897 0.133 Sri Lanka Hikkaduwa 102360 25544 24552 0.961 1 Sri Lanka Hingurakgoda 71049 17558 15617 0.889 0.581 Sri Lanka Homagama 247790 62127 60888 0.98 1 Sri Lanka Horana 112137 28075 26721 0.952 1 Sri Lanka Horowpothana 34870 8583 6244 0.727 0.093 Sri Lanka Ibbagamuwa 88591 23227 20140 0.867 0.413 Sri Lanka Imaduwa 46369 11449 10753 0.939 0.884 Sri Lanka Imbulpe 59169 15959 13184 0.826 0.355 Sri Lanka Ingiriya 55185 13921 12666 0.91 0.922 Sri Lanka Ipalogama 38012 10413 9250 0.888 0.352 Sri Lanka Irakkamam 14641 3331 2720 0.817 0.777 Sri Lanka Island North (Kayts) 12451 2306 1105 0.479 0.23 Sri Lanka Island South (Velanai) 18032 3953 1373 0.347 0.142 88 Sri Lanka Ja-Ela 208453 53271 52229 0.98 1 Sri Lanka Jaffna 46546 11151 9845 0.883 1 Sri Lanka Kaduwela 253586 64834 63713 0.983 1 Sri Lanka Kahatagasdigiliya 38754 9519 7396 0.777 0.138 Sri Lanka Kahawatta 46095 11276 9213 0.817 0.419 Sri Lanka Kalawana 51968 13824 10264 0.742 0.118 Sri Lanka Kalmunai 20946 10658 10314 0.968 1 Sri Lanka Kalmunai Tamil Division 10052 7224 6684 0.925 1 Sri Lanka Kalpitiya 85440 20330 15695 0.772 0.909 Sri Lanka Kalutara 158736 36743 35648 0.97 1 Sri Lanka Kamburupitiya 42945 10488 9977 0.951 0.353 Sri Lanka Kandaketiya 23948 6268 4738 0.756 0.089 Sri Lanka Kandavalai 12702 5697 67 0.012 0.049 Sri Lanka Kandy Four Gravets & Gangawata 141975 37230 36318 0.976 0.993 Korale Sri Lanka Kanthalai 50195 12051 10333 0.857 0.655 Sri Lanka Karachchi 42103 15308 2651 0.173 0.323 Sri Lanka Karainagar 8829 2519 1441 0.572 0.216 Sri Lanka Karaitheevu 9187 4341 3951 0.91 0.999 Sri Lanka Karandeniya 64048 17462 16222 0.929 0.984 Sri Lanka Karuwalagaswewa 25217 6848 4464 0.652 0.22 Sri Lanka Katana 225665 71059 68703 0.967 1 Sri Lanka Katharagama 18558 4757 3716 0.781 0.786 Sri Lanka Kattankudy 28155 10907 10623 0.974 1 Sri Lanka Katuwana 46601 12338 10733 0.87 0.191 Sri Lanka Kebithigollewa 21110 5642 4324 0.766 0.069 Sri Lanka Kegalle 91043 23753 22031 0.928 0.711 Sri Lanka Kekirawa 59066 15488 13679 0.883 0.546 Sri Lanka Kelaniya 116541 34087 33284 0.976 1 Sri Lanka Kesbewa 246067 63497 62874 0.99 1 Sri Lanka Kinniya 65094 14880 11907 0.8 0.937 Sri Lanka Kiriella 33583 8698 7586 0.872 0.211 Sri Lanka Kirinda Puhulwella 20896 5271 4954 0.94 0.135 Sri Lanka Kobeigane 37359 10118 8192 0.81 0.447 Sri Lanka Kolonna 46353 12590 9194 0.73 0.051 Sri Lanka Kolonnawa 128156 45446 44229 0.973 1 Sri Lanka Koralai Pattu West (Oddamavadi) 11319 6025 5313 0.882 0.984 Sri Lanka Koralai Pattu (Valachchenai) 23021 5360 3562 0.665 0.996 Sri Lanka Koralai Pattu Central 23774 6832 5645 0.826 0.987 89 Sri Lanka Koralai Pattu North (Vaharai) 20982 5292 1169 0.221 0.147 Sri Lanka Koralai Pattu South (Kiran) 31569 6634 2041 0.308 0.643 Sri Lanka Kotapola 62871 16695 14602 0.875 0.282 Sri Lanka Kotavehera 21571 6045 3832 0.634 0.107 Sri Lanka Kothmale 96946 23849 20479 0.859 0.423 Sri Lanka Kuchchaveli 36099 8506 6206 0.73 0.798 Sri Lanka Kuliyapitiya East 54001 13167 11962 0.908 0.534 Sri Lanka Kuliyapitiya West 78834 21539 19951 0.926 0.715 Sri Lanka Kundasale 132444 32201 30534 0.948 0.898 Sri Lanka Kurunegala 77662 20325 19071 0.938 0.888 Sri Lanka Kuruvita 96019 24257 21316 0.879 0.709 Sri Lanka Laggala-Pallegama 12121 3501 2316 0.662 0.066 Sri Lanka Lahugala 9037 2382 1816 0.762 0.392 Sri Lanka Lankapura 35484 9944 8645 0.869 0.47 Sri Lanka Lunugala 30567 8443 6230 0.738 0.266 Sri Lanka Lunugamvehera 33993 8494 6756 0.795 0.412 Sri Lanka Madampe 50633 13208 11932 0.903 0.995 Sri Lanka Madhu 10425 1932 471 0.244 0.061 Sri Lanka Madulla 32221 8216 4041 0.492 0.053 Sri Lanka Madurawala 34038 9020 8322 0.923 0.888 Sri Lanka Mahakumbukkadawala 19306 5456 2820 0.517 0.114 Sri Lanka Mahaoya 21327 5398 2218 0.411 0.088 Sri Lanka Mahara 205677 52838 51245 0.97 1 Sri Lanka Maharagama 146548 50293 49828 0.991 1 Sri Lanka Mahawewa 49821 13962 12801 0.917 0.999 Sri Lanka Mahawilachchiya 17811 4546 2871 0.632 0.077 Sri Lanka Mahiyanganaya 77215 20558 15776 0.767 0.433 Sri Lanka Maho 58697 16596 12619 0.76 0.427 Sri Lanka Malimbada 34828 8928 8534 0.956 0.646 Sri Lanka Mallawapitiya 51852 13790 12830 0.93 0.837 Sri Lanka Manmunai North 67150 21515 18991 0.883 0.999 Sri Lanka Manmunai Pattu (Araipattai) 28737 8130 6688 0.823 0.959 Sri Lanka Manmunai South & Eruvil pattu 61367 14468 11686 0.808 0.995 Sri Lanka Manmunai South-West 25073 6197 2226 0.359 0.673 Sri Lanka Manmunai West 30230 7368 1773 0.241 0.339 Sri Lanka Mannar Town 47183 11768 9065 0.77 0.72 Sri Lanka Manthai East 3247 1761 355 0.202 0.059 Sri Lanka Manthai West 9989 3778 535 0.142 0.062 Sri Lanka Maritimepattu 22205 7878 2351 0.298 0.481 90 Sri Lanka Maspotha 38438 9409 8782 0.933 0.728 Sri Lanka Matale 76767 18731 17451 0.932 0.712 Sri Lanka Matara Four Gravets 110418 28040 27469 0.98 0.984 Sri Lanka Mathugama 83124 20879 19275 0.923 0.874 Sri Lanka Mawanella 114578 28124 26091 0.928 0.675 Sri Lanka Mawathagama 67243 17262 15921 0.922 0.716 Sri Lanka Medadumbara 59308 16097 14230 0.884 0.284 Sri Lanka Medagama 36048 9516 6446 0.677 0.109 Sri Lanka Medawachchiya 41340 10656 8342 0.783 0.262 Sri Lanka Medirigiriya 67257 18124 14552 0.803 0.218 Sri Lanka Meegahakivula 19872 5410 3835 0.709 0.086 Sri Lanka Mihinthale 33137 7959 6973 0.876 0.481 Sri Lanka Millaniya 55125 13644 12683 0.93 0.972 Sri Lanka Minipe 54281 14178 11734 0.828 0.275 Sri Lanka Minuwangoda 186135 46658 44883 0.962 1 Sri Lanka Mirigama 168899 43183 40034 0.927 0.995 Sri Lanka Moneragala 51159 12744 8807 0.691 0.354 Sri Lanka Moratuwa 76912 42466 41672 0.981 1 Sri Lanka Morawewa 10287 2350 1338 0.569 0.205 Sri Lanka Mulatiyana 52671 13108 12256 0.935 0.177 Sri Lanka Mundal 59347 15705 12142 0.773 0.786 Sri Lanka Musali 7890 2050 1296 0.632 0.454 Sri Lanka Muttur 61042 14440 9083 0.629 0.853 Sri Lanka Nachchaduwa 22678 5828 5074 0.871 0.636 Sri Lanka Nagoda 51888 14154 12682 0.896 0.281 Sri Lanka Nallur 69240 16476 14887 0.904 0.992 Sri Lanka Nanattan 16484 4386 2767 0.631 0.323 Sri Lanka Narammala 56183 15297 13847 0.905 0.745 Sri Lanka Nattandiya 63297 16908 15923 0.942 1 Sri Lanka Naula 31320 8512 6926 0.814 0.079 Sri Lanka Navithanveli 19569 4812 3496 0.727 0.952 Sri Lanka Nawagattegama 14904 4205 2951 0.702 0.054 Sri Lanka Negombo 118839 35190 34111 0.969 1 Sri Lanka Neluwa 27469 7743 6721 0.868 0.224 Sri Lanka Nikaweratiya 41321 11061 8916 0.806 0.375 Sri Lanka Ninthavur 27429 7236 6759 0.934 0.985 Sri Lanka Nivithigala 60632 15430 13039 0.845 0.281 Sri Lanka Niyagama 35490 9499 8781 0.924 0.293 Sri Lanka Nochchiyagama 46491 11995 10038 0.837 0.266 91 Sri Lanka Nuwara Eliya 205750 49165 44286 0.901 0.786 Sri Lanka Nuwaragam Palatha Central 58356 14905 12720 0.853 0.565 Sri Lanka Nuwaragam Palatha East 66172 16537 15650 0.946 0.923 Sri Lanka Oddusuddan 11017 3998 203 0.051 0.217 Sri Lanka Okewela 19746 4807 4434 0.922 0.253 Sri Lanka Opanayake 26617 7194 5940 0.826 0.285 Sri Lanka Pachchilaipalli 8684 2209 67 0.03 0.153 Sri Lanka Padavi Sri Pura 12480 3197 2198 0.688 0.116 Sri Lanka Padaviya 23437 6219 3906 0.628 0.219 Sri Lanka Padiyathalawa 19184 4851 1942 0.4 0.099 Sri Lanka Padukka 68125 17144 16308 0.951 0.964 Sri Lanka Paduwasnuwara East 33146 8903 7737 0.869 0.155 Sri Lanka Paduwasnuwara West 60479 16946 15362 0.907 0.506 Sri Lanka Palagala 31311 8432 6671 0.791 0.188 Sri Lanka Palindanuwara 51437 13675 10922 0.799 0.141 Sri Lanka Pallama 24837 6922 4788 0.692 0.264 Sri Lanka Pallepola 29570 8108 6868 0.847 0.049 Sri Lanka Palugaswewa 15005 4026 3374 0.838 0.27 Sri Lanka Panadura 173900 44315 43405 0.979 1 Sri Lanka Pannala 126334 33836 30910 0.914 0.794 Sri Lanka Panvila 26866 7076 6022 0.851 0.192 Sri Lanka Pasbage Korale 61698 15353 13748 0.895 0.511 Sri Lanka Pasgoda 59807 16159 14380 0.89 0.131 Sri Lanka Passara 48485 12955 11242 0.868 0.207 Sri Lanka Pathadumbara 85926 22130 21234 0.96 0.828 Sri Lanka Pathahewaheta 60843 14776 13577 0.919 0.368 Sri Lanka Pelmadulla 90691 22909 19655 0.858 0.67 Sri Lanka Pitabeddara 51531 13319 11988 0.9 0.222 Sri Lanka Polgahawela 66773 16972 15645 0.922 0.619 Sri Lanka Polpithigama 76923 21917 14013 0.639 0.114 Sri Lanka Poonakary 18413 4846 5 0.001 0.017 Sri Lanka Poratheevu Pattu 33930 9528 4154 0.436 0.644 Sri Lanka Pottuvil 32985 8678 6652 0.767 0.798 Sri Lanka Pujapitiya 64120 14352 13515 0.942 0.602 Sri Lanka Puthukudiyiruppu 11710 6127 928 0.151 0.002 Sri Lanka Puttalam 73301 19747 17096 0.866 0.942 Sri Lanka Rajanganaya 33169 8953 7339 0.82 0.231 Sri Lanka Rambewa 35314 9173 7419 0.809 0.252 Sri Lanka Rambukkana 83232 22184 20462 0.922 0.52 92 Sri Lanka Rasnayakapura 23018 6136 4001 0.652 0.125 Sri Lanka Rathmalana 49608 21466 21142 0.985 1 Sri Lanka Ratnapura 118308 30385 26194 0.862 0.62 Sri Lanka Rattota 49976 14310 12440 0.869 0.248 Sri Lanka Rideemaliyadda 52638 13822 8655 0.626 0.15 Sri Lanka Ridigama 84600 23343 19333 0.828 0.215 Sri Lanka Ruwanwella 67100 16188 14355 0.887 0.636 Sri Lanka Sainthamarathu 33531 6198 6039 0.974 1 Sri Lanka Sammanthurai 50658 15084 13002 0.862 0.991 Sri Lanka Seethawaka 119341 28739 27002 0.94 0.978 Sri Lanka Seruvila 16416 4231 2592 0.613 0.624 Sri Lanka Sevanagala 42913 11202 9002 0.804 0.294 Sri Lanka Siyambalanduwa 55287 13830 6653 0.481 0.079 Sri Lanka Sooriyawewa 44780 11453 9285 0.811 0.391 Sri Lanka Soranathota 22776 6090 5065 0.832 0.232 Sri Lanka Sri Jayawardanapura Kotte 35686 27583 27166 0.985 1 Sri Lanka Tangalle 72328 19163 17823 0.93 0.669 Sri Lanka Thalawa 56981 15068 12786 0.849 0.266 Sri Lanka Thamankaduwa 80657 21124 19320 0.915 0.691 Sri Lanka Thambalagamuwa 28232 7220 5772 0.799 0.72 Sri Lanka Thambuttegama 39391 10372 8610 0.83 0.254 Sri Lanka Thanamalvila 27042 7462 5131 0.688 0.145 Sri Lanka Thawalama 33981 8902 7744 0.87 0.054 Sri Lanka Thenmaradchy (Chavakachcheri) 70119 16198 10296 0.636 0.607 Sri Lanka Thihagoda 34638 8602 8252 0.959 0.621 Sri Lanka Thimbirigasyaya 56969 54176 53047 0.979 1 Sri Lanka Thirappane 25580 6793 5462 0.804 0.183 Sri Lanka Thirukkovil 24608 6912 4209 0.609 0.782 Sri Lanka Thumpane 37027 9635 8899 0.924 0.405 Sri Lanka Thunukkai 5710 2347 583 0.248 0.155 Sri Lanka Tissamaharama 68523 17539 15412 0.879 0.607 Sri Lanka Trincomalee Town and Gravets 84476 22999 20342 0.884 0.992 Sri Lanka Udadumbara 22428 6518 4994 0.766 0.11 Sri Lanka Udapalatha 94114 23318 21243 0.911 0.716 Sri Lanka Udubaddawa 51241 14257 13123 0.92 0.897 Sri Lanka Udunuwara 115427 27517 26395 0.959 0.805 Sri Lanka Uhana 62871 14795 12313 0.832 0.472 Sri Lanka Ukuwela 68882 17167 15540 0.905 0.606 Sri Lanka Uva Paranagama 78431 20948 18958 0.905 0.327 93 Sri Lanka Vadamaradchy East 10737 3432 507 0.148 0.106 Sri Lanka Vadamaradchy North (Point Pedro) 45326 11981 9465 0.79 0.996 Sri Lanka Vadamaradchy South-West 47850 11239 8512 0.757 0.966 (Karaveddy) Sri Lanka Valikamam East (Kopay) 72170 17166 12388 0.722 0.924 Sri Lanka Valikamam North 32844 6939 4141 0.597 0.877 Sri Lanka Valikamam South (Uduvil) 51458 12609 9965 0.79 0.998 Sri Lanka Valikamam South-West (Sandilipay) 54310 12460 9536 0.765 0.912 Sri Lanka Valikamam West (Chankanai) 46342 10964 7848 0.716 0.781 Sri Lanka Vavuniya 110724 27447 21057 0.767 0.763 Sri Lanka Vavuniya North 9046 2920 243 0.083 0.18 Sri Lanka Vavuniya South 12751 3738 2754 0.737 0.171 Sri Lanka Vengalacheddikulam 33829 7173 4753 0.663 0.449 Sri Lanka Verugal Eachchilampattu 15463 2948 1091 0.37 0.115 Sri Lanka Walallavita 55227 14659 12751 0.87 0.304 Sri Lanka Walapane 98435 28486 23362 0.82 0.398 Sri Lanka Walasmulla 43979 11186 10069 0.9 0.167 Sri Lanka Wanathavilluwa 17613 4651 2802 0.602 0.436 Sri Lanka Warakapola 115190 29550 26455 0.895 0.796 Sri Lanka Wariyapola 61059 16994 14789 0.87 0.534 Sri Lanka Wattala 132772 43648 42236 0.968 1 Sri Lanka Weeraketiya 42937 10799 9569 0.886 0.237 Sri Lanka Weerambugedara 35178 9347 8404 0.899 0.388 Sri Lanka Weligama 72086 17708 16867 0.953 0.989 Sri Lanka Weligepola 30694 8996 6206 0.69 0.025 Sri Lanka Welikanda 34638 9324 6274 0.673 0.279 Sri Lanka Welimada 103446 26047 23893 0.917 0.46 Sri Lanka Welioya 5453 1528 927 0.607 0.264 Sri Lanka Welipitiya 53527 12985 12266 0.945 0.729 Sri Lanka Welivitiya-Divithura 29020 7838 7197 0.918 0.777 Sri Lanka Wellawaya 62817 16453 12524 0.761 0.318 Sri Lanka Wennappuwa 65794 18806 18156 0.965 1 Sri Lanka Wilgamuwa 30615 8220 5904 0.718 0.226 Sri Lanka Yakkalamulla 44322 11872 10683 0.9 0.389 Sri Lanka Yatawatta 30883 7974 6907 0.866 0.03 Sri Lanka Yatinuwara 107872 27305 26244 0.961 0.723 Sri Lanka Yatiyanthota 59613 15965 13036 0.817 0.429 94 Tanzania Census Year: 2012 VIIRS Year: 2012 Administrative Unit: District. Tanzania has 160 Districts. Administrative Unit Boundaries: Available from HDX. Census data description: The census data on electricity access is derived from the 2012 Population and Housing census conducted by the Tanzanian National Bureau of Statistics. Census question: “What is the main source of energy used for Lighting?” (45) Country Admin Unit Unit Pop Total HHs HHs w/ Elec Elec rate Elec rate (HRSL) (Census) (Census) (Census) (Satellite) Tanzania Arusha 349435 NA NA 0.307 0.463 Tanzania Arusha Urban 533162 NA NA 0.548 0.95 Tanzania Babati 367927 NA NA 0.089 0.089 Tanzania Babati Urban 110974 NA NA 0.303 0.433 Tanzania Bagamoyo 371758 NA NA 0.177 0.299 Tanzania Bahi 258528 NA NA 0.055 0.021 Tanzania Bariadi 527773 NA NA 0.089 0.087 Tanzania Biharamulo 374918 NA NA 0.07 0.054 Tanzania Buhigwe 299134 NA NA 0.051 0.003 Tanzania Bukoba 333552 NA NA 0.054 0.038 Tanzania Bukoba Urban 148424 NA NA 0.546 0.924 Tanzania Bukombe 283464 NA NA 0.063 0.05 Tanzania Bunda 387622 NA NA 0.127 0.172 Tanzania Busega 249262 NA NA 0.062 0.052 Tanzania Butiam 281646 NA NA 0.088 0.079 Tanzania Chake Chake 109806 NA NA 0 0.517 Tanzania Chamwino 388487 NA NA 0.047 0.041 Tanzania Chato 450955 NA NA 0.055 0.063 Tanzania Chemba 274088 NA NA 0.054 0.031 Tanzania Chunya 352236 NA NA 0.099 0.127 Tanzania Dodoma Urban 479416 NA NA 0.338 0.563 Tanzania Gairo 227695 NA NA 0.085 0.073 Tanzania Geita 1011334 NA NA 0.14 0.107 Tanzania Hai 246942 NA NA 0.318 0.585 Tanzania Hanang 326799 NA NA 0.11 0.1 Tanzania Handeni 336567 NA NA 0.055 0.118 95 Tanzania Handeni Mji 96111 NA NA 0.24 0.371 Tanzania Igunga 465807 NA NA 0.116 0.082 Tanzania Ikungi 320079 NA NA 0.081 0.019 Tanzania Ilala 1620313 NA NA 0.625 0.962 Tanzania Ileje 143036 NA NA 0.063 0.055 Tanzania Ilemela 376594 NA NA 0.451 0.792 Tanzania Iramba 269300 NA NA 0.102 0.056 Tanzania Iringa 281833 NA NA 0.08 0.069 Tanzania Iringa Urban 180650 NA NA 0.588 0.778 Tanzania Itilima 392652 NA NA 0.077 0.009 Tanzania Kahama 617715 NA NA 0.073 0.061 Tanzania Kahama Township Authority 285720 NA NA 0.225 0.493 Tanzania Kakonko 196600 NA NA 0.03 0.015 Tanzania Kalambo 248664 NA NA 0.041 0.021 Tanzania Kaliua 493051 NA NA 0.042 0.028 Tanzania Karagwe 385203 NA NA 0.08 0.091 Tanzania Karatu 270947 NA NA 0.17 0.148 Tanzania Kaskazini A 119257 NA NA 0 0.622 Tanzania Kaskazini B 100833 NA NA 0 0.378 Tanzania Kasulu 499991 NA NA 0.048 0.005 Tanzania Kasulu Township Authority 244605 NA NA 0.131 0.242 Tanzania Kati 86102 NA NA 0 0.327 Tanzania Kibaha 87401 NA NA 0.262 0.207 Tanzania Kibaha Urban 158014 NA NA 0.374 0.698 Tanzania Kibondo 308269 NA NA 0.069 0.082 Tanzania Kigoma 250953 NA NA 0.049 0.071 Tanzania Kigoma Urban 251108 NA NA 0.402 0.901 Tanzania Kilindi 300105 NA NA 0.041 0.028 Tanzania Kilolo 244342 NA NA 0.098 0.085 Tanzania Kilombero 480698 NA NA 0.164 0.28 Tanzania Kilosa 517963 NA NA 0.135 0.252 Tanzania Kilwa 206420 NA NA 0.137 0.122 Tanzania Kinondoni 2189085 NA NA 0.688 0.992 Tanzania Kisarawe 118013 NA NA 0.127 0.17 Tanzania Kishapu 310865 NA NA 0.101 0.062 Tanzania Kiteto 303299 NA NA 0.083 0.095 Tanzania Kondoa 316350 NA NA 0.102 0.087 Tanzania Kongwa 361901 NA NA 0.084 0.115 Tanzania Korogwe 279933 NA NA 0.086 0.179 96 Tanzania Korogwe Township Authority 79350 NA NA 0.37 0.689 Tanzania Kusini 41851 NA NA 0 0.635 Tanzania Kwimba 440822 NA NA 0.143 0.042 Tanzania Kyela 261087 NA NA 0.134 0.27 Tanzania Kyerwa 370276 NA NA 0.048 0.013 Tanzania Lindi 215533 NA NA 0.092 0.039 Tanzania Lindi Urban 83557 NA NA 0.324 0.436 Tanzania Liwale 105759 NA NA 0.129 0.054 Tanzania Longido 139264 NA NA 0.066 0.082 Tanzania Ludewa 147613 NA NA 0.1 0.047 Tanzania Lushoto 563898 NA NA 0.071 0.072 Tanzania Mafia 49959 NA NA 0.211 0.241 Tanzania Mafinga Township Authority 58749 NA NA 0.524 0.761 Tanzania Magharibi 469908 NA NA 0 0.946 Tanzania Magu 323828 NA NA 0.204 0.161 Tanzania Makambako Township 106966 NA NA 0.28 0.472 Authority Tanzania Makete 104162 NA NA 0.088 0.064 Tanzania Manyoni 363114 NA NA 0.092 0.12 Tanzania Masasi 281728 NA NA 0.04 0.058 Tanzania Masasi Township Authority 116857 NA NA 0.166 0.355 Tanzania Maswa 391296 NA NA 0.099 0.054 Tanzania Mbarali 353218 NA NA 0.095 0.207 Tanzania Mbeya 351442 NA NA 0.089 0.168 Tanzania Mbeya Urban 469346 NA NA 0.471 0.913 Tanzania Mbinga 412437 NA NA 0.09 0.087 Tanzania Mbogwe 243644 NA NA 0.057 0.005 Tanzania Mbozi 544323 NA NA 0.095 0.111 Tanzania Mbulu 381516 NA NA 0.103 0.078 Tanzania Meatu 348080 NA NA 0.108 0.041 Tanzania Meru 304607 NA NA 0.23 0.412 Tanzania Micheweni 116389 NA NA 0 0.206 Tanzania Missenyi 231601 NA NA 0.068 0.107 Tanzania Misungwi 422372 NA NA 0.162 0.066 Tanzania Mjini 261233 NA NA 0 1 Tanzania Mkalama 214510 NA NA 0.105 0.031 Tanzania Mkinga 127457 NA NA 0.105 0.106 Tanzania Mkoani 106798 NA NA 0 0.18 Tanzania Mkuranga 267518 NA NA 0.095 0.224 97 Tanzania Mlele 342906 NA NA 0.049 0.023 Tanzania Momba 240930 NA NA 0.029 0.047 Tanzania Monduli 197210 NA NA 0.162 0.211 Tanzania Morogoro 323881 NA NA 0.053 0.074 Tanzania Morogoro Urban 356574 NA NA 0.53 0.866 Tanzania Moshi 530965 NA NA 0.307 0.58 Tanzania Moshi Urban 219878 NA NA 0.714 1 Tanzania Mpanda 216750 NA NA 0.041 0.03 Tanzania Mpanda Urban 123712 NA NA 0.351 0.713 Tanzania Mpwapwa 355295 NA NA 0.091 0.096 Tanzania Mtwara 254523 NA NA 0.035 0.027 Tanzania Mtwara Urban 123086 NA NA 0.333 0.781 Tanzania Mufindi 301502 NA NA 0.091 0.073 Tanzania Muheza 233614 NA NA 0.153 0.229 Tanzania Muleba 603517 NA NA 0.083 0.068 Tanzania Musoma 206258 NA NA 0.063 0.003 Tanzania Musoma Urban 151624 NA NA 0.49 0.92 Tanzania Mvomero 383372 NA NA 0.107 0.198 Tanzania Mwanga 141106 NA NA 0.341 0.484 Tanzania Nachingwea 201874 NA NA 0.13 0.144 Tanzania Namtumbo 228847 NA NA 0.059 0.026 Tanzania Nanyumbu 170101 NA NA 0.024 0.03 Tanzania Newala 232020 NA NA 0.068 0.089 Tanzania Ngara 370540 NA NA 0.058 0.127 Tanzania Ngorongoro 205710 NA NA 0.056 0.031 Tanzania Njombe 97268 NA NA 0.09 0.047 Tanzania Njombe Urban 148049 NA NA 0.324 0.331 Tanzania Nkasi 337107 NA NA 0.037 0.039 Tanzania Nyamagana 482129 NA NA 0.428 0.887 Tanzania Nyang'hwale 185577 NA NA 0.071 0.004 Tanzania Nyasa 170768 NA NA 0.048 0.029 Tanzania Nzega 583986 NA NA 0.094 0.081 Tanzania Pangani 59740 NA NA 0.177 0.245 Tanzania Rombo 287036 NA NA 0.219 0.153 Tanzania Rorya 300580 NA NA 0.069 0.036 Tanzania Ruangwa 145431 NA NA 0.116 0.092 Tanzania Rufiji 232705 NA NA 0.098 0.219 Tanzania Rungwe 382639 NA NA 0.115 0.133 Tanzania Same 316629 NA NA 0.233 0.267 98 Tanzania Sengerema 769095 NA NA 0.142 0.074 Tanzania Serengeti 302962 NA NA 0.105 0.062 Tanzania Shinyanga 387923 NA NA 0.075 0.016 Tanzania Shinyanga Urban 186954 NA NA 0.378 0.613 Tanzania Siha 135105 NA NA 0.164 0.183 Tanzania Sikonge 216350 NA NA 0.087 0.078 Tanzania Simanjiro 212587 NA NA 0.155 0.216 Tanzania Singida 265059 NA NA 0.076 0.034 Tanzania Singida Urban 177978 NA NA 0.385 0.517 Tanzania Songea 194549 NA NA 0.057 0.035 Tanzania Songea Urban 255539 NA NA 0.408 0.696 Tanzania Sumbawanga 365213 NA NA 0.03 0.022 Tanzania Sumbawanga Urban 255447 NA NA 0.252 0.439 Tanzania Tabora Urban 262810 NA NA 0.458 0.673 Tanzania Tandahimba 256559 NA NA 0.059 0.067 Tanzania Tanga Urban 303549 NA NA 0.504 0.862 Tanzania Tarime 387465 NA NA 0.11 0.267 Tanzania Temeke 1804339 NA NA 0.599 0.974 Tanzania Tunduma 113718 NA NA 0.362 0.852 Tanzania Tunduru 338945 NA NA 0.091 0.107 Tanzania Ukerewe 376194 NA NA 0.156 0.068 Tanzania Ulanga 319699 NA NA 0.065 0.071 Tanzania Urambo 243241 NA NA 0.119 0.146 Tanzania Uvinza 450486 NA NA 0.043 0.021 Tanzania Uyui 482077 NA NA 0.078 0.016 Tanzania Wanging'ombe 183617 NA NA 0.085 0.037 Tanzania Wete 118445 NA NA 0 0.369 99