Spatialization of population based on Xgboost with multi-source data

Abstract

Aiming at the problem that demographic data cannot visually and clearly show the true distribution of population and cannot be combined with other environmental resource spatial data for analysis. This paper takes Chongqing as an example, selects nighttime light data etc. as variable factors affecting population distribution. Using the Xgboost model to build a regression model on the county level, and generates the population data of 100m in Chongqing in 2010. The accuracy of the population spatialization results and three public data sets were compared on the township scale. Finally, based on the importance of the variable factors of the Xgboost model, the influencing factors of the spatial distribution of Chongqing’s population were explored. The results show that the root mean square error in this paper is significantly better than the other three population data sets, the absolute value error is significantly better than the GPW data set and the Chinese kilometer grid data set, and slightly better than the World Pop data set. Through the analysis of the importance of variable factors, it is found that the distance from construction land is the most important indicator, and the nighttime light data, residential area and POI data all play an important role in population distribution of Chongqing.