PowerScour: tracking electrified settlements using satellite data.

Access to electricity is crucial for poverty reduction and economic growth. However, almost 759 million people still do not have access to electricity. More than 90% are located in the global South, where low-income countries struggle to provide clean, reliable, and affordable energy sources to ameliorate the basic living standards. Even though there are many opportunities to provide basic electrification in these settings, the lack of reliable and updated information about electrification has become one of the main challenges for policymakers and developers to better plan grid extensions and prioritize communities with higher needs. The increasing availability of remote sensing data has created opportunities to obtain information about electricity access at a larger and quicker scale. Using ground truth data of 57 k distribution transformer locations from Kenya, we present a processing pipeline to validate and compare state-of-the-art techniques that use very high resolution (VHR) daytime imagery (50 cm Digital Globe) or low-resolution nightlight (NTL) imagery (450 m VIIRS-DNB) to identify electricity access. Further, we propose a supervised-learning approach called PowerScour that outperforms three techniques from the commercial, scientific and public fields. By assessing the trade-offs between temporal and spatial resolution and comparing population and settlement patterns, we find that PowerScour improves the F1-score of existing techniques by up to ≈27% in deep rural areas. In Kenya, our model correctly identified ≈73.4% of places with and without access to electricity between 2013 and 2017. All data processing and modeling scripts are available at https://github.com/santiagocorrea/PowerScour.