I do a reasonable amount of empirical work with VIIRS VNP46A* data (the remote sensing platform and data product(s) used for the nighttime light data analysis).
GDP estimates from nighttime light data are possible but often inaccurate. Before you even get into the larger issues, there are a lot of small things, such as cloud coverage, ground coverage changes, lunar illumination, &c... that have to be accounted for.
It's great when you're trying to track urban development. It's good when you need quick ground truth estimates (e.g., you're looking at an area impacted by war or lockdown and you want to get a rough guess of the actual impact), and it's ok when you are looking at a country with an economic model (manufacturing/export-led) where changes in nighttime light volumes can tell you something.
It's iffy if you're trying to compare a lot of different countries, with different economic models, across different time spans, and you don't have the time to accurately assess whether nighttime light data is a good proxy for GDP in each of the many regions/country <-> economic model combinations you're looking at.
they're not trying to estimate GDP though (absolute value), they're trying to estimate GDP increase (relative value). I agree estimating the absolute value would be erroneous. However, estimating an increase is less so. Other issues you mention are circumstantial - and that's why companies doing this estimation have specific experts on politics, economy, and so on. It's not just some geek with python going gdp[country] = bright_pixels * 0.18
>It's not just some geek with python going gdp[country] = bright_pixels * 0.18
With respect to this study, it more or less is. Go read Martinez original study from 2017 and lol at the methadology. It's rudimentary application of Henderson et. al. original work on predicting economic activity with nighttime light as proxy with all the caveates parent comment highlighted - not expert estimation. Martinez updated his study occasionally with newer data, for click bait articles like this (started with Washington Post in 2018), with no modification in methodology as far as I'm aware (have not tracked down latest revision to compare). As someone who follows PRC development closely, NBER released a paper using light data with more sophisticated analysis also from 2017 stating PRC growth being understated. NBER is one of the most influential economic think tanks. To compare credential wank as proxy for expertise, author of this paper Martinez, is assisant prof at Chicago Harris School of Public Policy. NBER paper was conducted by two current staff economists at the FED and a titled prof of economics prof at Columbia. But guess which study gets repeatedly posted in western MSM.
GDP estimates from nighttime light data are possible but often inaccurate. Before you even get into the larger issues, there are a lot of small things, such as cloud coverage, ground coverage changes, lunar illumination, &c... that have to be accounted for.
It's great when you're trying to track urban development. It's good when you need quick ground truth estimates (e.g., you're looking at an area impacted by war or lockdown and you want to get a rough guess of the actual impact), and it's ok when you are looking at a country with an economic model (manufacturing/export-led) where changes in nighttime light volumes can tell you something.
It's iffy if you're trying to compare a lot of different countries, with different economic models, across different time spans, and you don't have the time to accurately assess whether nighttime light data is a good proxy for GDP in each of the many regions/country <-> economic model combinations you're looking at.