We propose a methodology for predicting subnational economic development using daytime satellite imagery. We collected high-resolution satellite images and corresponding ground truth data (e.g. buildings, roads) for over 28 million 1km x 1km grid cells covering 25 European and 7 African countries. We first use a standard random forest model to identify a subset of features from the ground truth data that best predict fixed capital at the EU NUTSII level. We then trained a convolutional neural network to extract the relevant features from each daytime satellite image. The resulting measures were used to predict fixed capital at the EU-NUTSII level. Finally, the predicted capital values are used in a standard development accounting framework to construct regional GDP per capita. The correlation between our constructed regional GDP and the official GDP figures is around 0.7. Thus our methodology can be used to predict subnational and national GDP for countries where these data are either missing or unreliable.

A Zoom link will be provided once you register for this event.