Building-level urban population mapping based on SDGSAT-1 nighttime light and multisource geospatial data

High-resolution population maps play a critical role in addressing the growing risks of urban disasters. This study develops a transferable, building-scale population spatialization framework for residential areas, entirely using freely accessible open data. The framework avoids dependence on costly or sensitive fine-grained demographic datasets and overcomes the limitations of census data, which are updated infrequently and available only at coarse spatial scales. Using 10-meter SDGSAT-1 NTL data, we applied a statistical modeling approach directly at the community level within residential areas, effectively resolving the scale inconsistency that often arises when coarse-scale models are downscaled to finer resolutions. We further introduced a Building Residential Weight index that integrates building capacity, occupancy rate, and functional attributes. This factor enables the population of each community to be proportionally allocated to its buildings, producing a detailed and realistic building-level population distribution. Model evaluation experiments demonstrate that the Random Forest algorithm achieved the highest modeling accuracy in this study, with an R2 of 0.779, representing an improvement of more than 0.55 compared with widely used global population datasets such as WorldPop, LandScan, and GHS-Pop. The generated building-level population distribution maps provide a high-resolution spatial foundation for megacity disaster risk management, resource allocation, and urban planning.