The United States measures gross domestic product (GDP) over short time horizons for the nation, but it does not produce high-frequency measures of changes in economic activity for smaller geographic areas. This inhibits analyses of how communities adjust to economic shocks related to business cycles, technological progress, climate change and other events. Existing data sources are limited not only in their spatial resolution, but also in their temporal frequency.
Gordon Hanson and colleagues argue that advances in remote sensing and machine learning will allow them to conduct economic analyses at the local level. They will use daytime satellite imagery to measure household income and population at very high spatial and temporal resolutions. They will also apply the latest advances in general-purpose deep-learning algorithms to predict income and population changes for localities using only spectral imagery from satellites.