Using modern census and environmental factor data, this study first identified the environmental factors that significantly affect the population distribution through Geodetector analysis and then constructed a population spatial distribution model based on the random forest regression algorithm. Finally, with this model and historical population data that were examined and corrected by historians, gridded population distributions with a spatial resolution of 10 km by 10 km in the traditional cultivated region of China (TCRC, hereafter) were reconstructed for six time slices from 1776 to 1953. Using the reconstruction dataset, the spatiotemporal characteristics of the population distribution were depicted. The results showed that (1) the environmental factors that significantly affected the population density differences among counties in the TCRC mainly consisted of elevation, slope, relief amplitude, distances to the nearest prefectural and provincial capitals, distance to the nearest river and the climatology moisture index. (2) Using the census data of 1934 counties in the TCRC in 2000 and the abovementioned environmental factor data, a random forest regression algorithm-based population spatial distribution model was constructed. Its determination coefficient (R^{2}) is 0.81. In 88.4% of the counties (districts), the relative errors of the model predictions were less than 50%. (3) From 1776 to 1953, the total population in the study area showed an uptrend. Prior to 1851, the population increased mainly in the Yangtze River Delta. During this period, the number of grid cells in which the population densities were greater than 500 persons per km^{2} increased from 292 to 683. From 1851 to 1953, the population increased extensively across the study area. In the North China Plain and the Pearl River Delta, the number of grid cells in which the population densities were greater than 500 persons per km^{2} increased from 36 to 88 and from 4 to 35, respectively. The spatial clustering pattern of the population distribution varied temporally. The potential reasons included the shifts in economic development hot spots, traditional beliefs, wars, famine, and immigration policies. (4) Between our reconstructions and the HYDE dataset, there are large differences in the data sources, selected environmental factors and modeling methods. As a consequence, in comparison to our reconstructions, there were fewer populations in the eastern area and more populations in the western area from 1776 to 1851 and more populations in urban areas and fewer populations in rural areas after 1851 in the HYDE dataset.

Xuezhen ZHANG ^{1,5}, Fahao WANG ^{1,2}, Weidong LU ^{3}, Shicheng LI ^{4}, Jingyun ZHENG ^{1,5,*}