AI for Good

Chapter 11Identifying Solar Energy Locations in India

—Anthony Ortiz and Joseph Kiesecker

Executive Summary

To avoid the most catastrophic impacts of climate change, countries will have to reduce greenhouse gas emissions by 50 percent by 2030 and achieve net zero emissions by 2050. With 73 percent of global emissions tied to energy use, a rapid transition to renewable energy is urgently needed. Solar energy is positioned to be a major contributing source of the energy transition and India is a burgeoning hub for solar energy. However, the land used for solar energy, farms has other uses, such as biodiversity conservation, agriculture, or urban development. We first sought to use artificial intelligence to identify solar farms to estimate when those projects were built. Then, we overlaid that information onto land-use maps so that policymakers could understand the land-use impact of previously developed solar projects. Using the locations of current solar farms, we predicted areas suitable for development to help developers and government make informed decisions about the siting of future projects.

To do this work, we developed a semantic segmentation model that used lower-resolution (10 meters) satellite imagery that is freely available worldwide. We integrated pixel-wise labels and hard negative mining techniques to identify 1,363 solar farms in India, including 1,035 that had never before been mapped. Our model had a mean accuracy of identification of solar farms of 92 ...