Global GDP Prediction With Night-Lights and Transfer Learning

Abstract

Nighttime lights (nightlights) data are useful in predicting gross domestic product (GDP), a key economic indicator used by policymakers and economists. A persistent problem in such prediction is that nightlights under-represent economic activity in rural areas. Attempting to disaggregate nightlights using urban and rural regions is problematic as the urban&#x2013;rural dichotomy is increasingly tenuous due to changing economic structures. In response, this article presents a regionalization approach which is data driven. Utilizing transfer learning, we trained a model which takes fine spatial resolution daytime satellite sensor imagery and learns an optimal regionalization to disaggregate VIIRS nightlights for GDP prediction. To make national scale inference feasible, we formulated a novel Monte Carlo importance sampling scheme and then performed a single-year cross-sectional study across 178 countries using <inline-formula><tex-math notation="LaTeX">$178\,000$</tex-math></inline-formula> images. This achieved an <inline-formula><tex-math notation="LaTeX">$R^{2}$</tex-math></inline-formula> between predicted and actual <inline-formula><tex-math notation="LaTeX">$\text{log}_{10} \text{GDP}$</tex-math></inline-formula> of 0.86 on the validation set and 0.89 on the whole study area. To benchmark, we performed a subnational study over 396 US counties using <inline-formula><tex-math notation="LaTeX">$98\,500$</tex-math></inline-formula> images in which our method outperformed comparable methods. Interpreting the regionalization, we found that the utility of the urban&#x2013;rural dichotomy is not supported by the model and argue that seeing the nightlights of some land types as representative of the overall economy is a better way to understand the model.