Because of the nature of the data (resettlements for dam construction) two thirds of the data is from China. But interestingly according to Figure 7 [1] the discrepancy exists even in countries with normally very meticulous record keeping, like Sweden and Germany. I find this surprising.

The root cause analysis also falls short for those cases: Germany's bureaucracy might be underfunded but fundamentally requires every resident to register at their place of residence. There is also no large-scale conflict or violence in Germany, no regions are really remote or hard to reach, all rural areas are electrified (important for satellite-based night-time light counting by GRUMP and WorldPop). The only satisfying root causes left are about satellite counting methods being too coarse or badly tuned to accurately count rural area, or rural living patterns possibly not fitting the assumptions of these models. But it's weird that these errors are so similar between so many different methods, not all of which even use satellite data

https://www.nature.com/articles/s41467-025-56906-7/figures/7