We used high-resolution night time satellite imagery to assess the luminosity of sub-national regions. The ‘brightness’ of a place at night time has been found to correlate with economic activity.
This data could then be used to measure the local economic impact of COVID as well as to understand how lockdown policies impacted local economies.
We developed a proprietary method of cleaning and processing night time satellite imagery (a widely used proxy for sub-national economic activity in academic studies). This involved: cleaning snowfall (snow reflects light), water bodies (water reflects light), and cloud cover (clouds obscure light).
We subsequently built a range of machine learning models (tree based models and neural networks) to train a model on the relationship between luminosity and GDP at a national level. We then used sub-national luminosity to predict sub-national GDP.
The accuracy of our estimates was 98%. This was calculated by comparing our models’ prediction of annual sub-national GDP in 2019 in Europe to official data.
The data was then used by local and national policymakers to assess the impact of COVID and subsequent lockdown policies on local (i.e. sub-national) economies.