Gridded population mapping of Chinese mainland (1982–2020) using a geographically and temporally neural network‑weighted nonlinear regression

Since the Reform and Opening-up, the population pattern of Chinese mainland has experienced a profound spatiotemporal restructuring. However, current mainstream open-access Gridded Population Data (GPD), including WorldPop, LandScan Global, and Global Human Settlement Population (GHS-POP), are limited by historical data gaps or the potential uncertainties introduced by estimating historical periods using recent static auxiliary data. Moreover, current population regression modeling methods typically fail to simultaneously consider nonlinearity and spatiotemporal nonstationarity, thereby limiting the accuracy of Gridded Population Mapping (GPM). To overcome these challenges, we propose a novel model, named Geographically and Temporally Neural Network Weighted Nonlinear Regression (GTNNWNR), to generate 1 km × 1 km GPD for Chinese mainland from 1982 to 2020 by integrating multi-source dynamic auxiliary data (i.e. building area, nighttime light, and land use data). Results demonstrate that GTNNWNR outperforms baseline models in county-level fitting performance, with the generated GPD exhibiting the highest accuracy at the township level. Furthermore, our dataset shows significantly higher overall accuracy compared to mainstream products. Notably, by integrating fully dynamic auxiliary variables, our dataset achieves superior temporal stability in accuracy compared with GHS-POP. Spatiotemporal analysis indicates that from 1982 to 2020, despite a significant increase in the area and population proportion of high-density regions, the macroscopic pattern across the Hu Line remained highly stable. In conclusion, this study contributes to improving the GPM method and provides researchers and policymakers with valuable datasets and insights to support sustainable development in the Chinese mainland. Moreover, the GTNNWNR model developed in this study exhibits strong spatiotemporal modeling and predictive potential for various geographic processes and environmental phenomena beyond GPM applications.