China’s rapid economic rise is one of the most important events of the last century. However, rigorous empirical work analyzing the reforms that led to this miraculous growth is scarce, due in part to the lack of reliable, disaggregated Chinese data. We propose to harness an untapped source of historical economic data: declassiﬁed photographs from the US CORONA spy satellite program. From 1960 to 1972, the CORONA satellites collected high-resolution daytime images from almost the entire land area of the Earth. By applying state-of-the-art deep learning methods to predict economic outcomes from this satellite imagery, we hope to generate new data to better understand China’s historic economic transformation. Moreover, we hope to develop a deep learning toolkit that can later be ﬂexibly applied to measure historical economic growth around the world.
We will train a convolutional neural network (CNN) to predict survey-based measures of village-level economic outcomes from historical satellite imagery. The CORONA satellite imagery, available for purchase by the public from the US Geological Survey, contains multiple cross-sections for most major regions (including China), beginning in 1960 and ending in 1972.
Several hundred Chinese village-level estimates of economic outcomes are freely available through the Contemporary Chinese Village Gazetteer Dataset (CCVG). By aggregating up our village estimates, we also hope to compare our model’s predictions to ofﬁcial provincial-level economic data from the Chinese National Bureau of Statistics.
Copyright 2022. All Rights Reserved
Design & Dev by Wonderland Collective