Spatially explicit reconstruction of the population distribution in the Tuojiang River Basin during 1911–2010 using random forest regression

Reconstruction of a century-long historical population distribution on the regional scale is essential to understanding human–environment interactions and socio-environmental problems. This study uses a random forest regression method based on county-level census data, natural suitability, and socio-economic factors to reconstruct the spatiotemporal population distribution of the Tuojiang River Basin from 1911 to 2010 at a 1-km resolution. By comparing the reconstruction results with the actual 2010 population density at the township level, we found that the population spatialization results are congruent with the actual population density distribution, with a determination coefficient ( R 2 ) of 0.84. Compared with three widely used gridded population products, the WorldPop population dataset, the gridded population of the world (GPW), and the grid population dataset of china (CHP), our population map showed the highest accuracy. High population density areas, primarily distributed in the districts and counties in well-developed areas, were mainly centered around the Chengdu Plain and its lower reaches where the landscape meets the Yangtze River. Such areas are characterized by moderate elevations and ideal topographical and climatic conditions. In contrast, the high-altitude areas of the southern and southwestern of the Tuojiang River Basin have relatively low population density. Our reconstructive analysis has the potential to benefit regional socio-economic development and planning.