基于随机森林优化模型的高分辨率人口密度研究

人口密度图是揭示人地关系的重要基础数据。基于随机森林的人口密度模型显著改善了人口密度图的质量，但存在区群谬误、影响因子遴选欠缜密、混淆人口分布规律等问题。本研究以石家庄为实验区，以河北省为模型应用区，依据自然区划、地貌区划和城乡分布，编制综合禀赋分区图；利用村常住人口数据，编制最小粒度人口密度图；从自然、经济和制度禀赋等维度，遴选人口密度影响因子；以公顷网格为采样单元，分区开展随机抽样，克服了样本采集单元面积显著大于输出单元面积（区群谬误）的问题，并避免影响因子数据产生可塑性面积单元问题，改善了模型输入数据的质量；借助随机森林模型，开展分组对照实验，分区筛选人口密度影响因子，分区推敲随机采样规模，分区编制人口密度图，避免在人口密度图中混淆人口分布法则。本研究将改善人口密度模型的再测信度和结构效度，推动人口密度影响机制、人口分布规律等基础理论研究，为编制全球高分辨率人口密度图提供中国方案。

The population density map is an important basic data to reveal the Man-land relationship. The population density algorithm based on the random forest model significantly improves the quality of the population density map, but there are problems such as ecological fallacy, inadequate selection of influence factors, and confusion of population distribution rules. This study uses Shijiazhuang as the experimental area and Hebei Province as the model application area. Based on natural zoning, landform zoning, and urban-rural distribution, a comprehensive endowment zoning map is prepared. Using the village resident population data, the minimum grain population density map is prepared. It selects the population density influencing factors from the dimensions of natural, economic, and institutional endowment; It uses the hectare grid as the sampling unit and conduct random sampling by area, which overcomes the problem that the area of the sample collection unit is significantly larger than the output unit area (ecological fallacy), and avoids modifiable areal unit problem generated by the impact factor data, which improves the quality of model input data; With the help of random forest model, grouped control experiments are carried out, it screens population density impact factors by district, consider random sampling size by district, compile population density map by district, which overall avoids confusing population distribution rules in the population density map. This study will improve the test-retest reliability and structural validity of the population density model, promote basic theoretical research on population density impact mechanisms, and population distribution laws, and thus provide a Chinese solution for the preparation of a global high-resolution population density map.