Research On Population Spatialization Based On XGBoost And Multi-source Data

Population is a social entity with rich content and complex relationship.The development of population is usually accompanied by various social phenomena and social problems.Accurate spatial distribution of population has important application value and scientific significance for government departments at all levels to solve such problems as urban planning and disaster risk response.The traditional census data has some limitations,such as long updating period,unable to make clear the spatial distribution of population in each census unit,and difficult to combine with other geographical environment data.Therefore,it is of great significance to achieve high-resolution population spatialization based on census data.This thesis takes Chongqing as an example to discuss from two aspects of population spatialization and grid scale suitability.First,this thesis selects multi-source data such as night light data,land use data,and point of interest data as variable factors indicating population distribution,and uses the XGBoost model and dasymetric mapping to spatialize the population data of Chongqing in 2010 at 100 m grid scale;then based on the importance of the variable factors of the XGBoost model to explore the relationship between the spatial distribution of Chongqing’s population and the characteristic variables;finally,the grid scale suitability is explored,and the most suitable grid scale for Chongqing is determined by constructing grid scale suitability indicators.The main research results are as follows:1.The introduction of point of interest data can effectively reduce the weight of night lights and provid e richer information,thereby improving the problem of night lighting "oversaturation",showing more differences in population distribution in areas with strong night lights,and improving the accuracy of population spatialization simulation.2.The overall simulation accuracy of population spatialization based on the XGBoost model and multi-source data proposed in this thesis is higher,with about 81.7%,which is about 3% higher than the simulation accuracy of the random forest model.Research shows that the XGBoost model and multi-source data have a better simulation effect in population spatialization.3.Due to the grid scale effect,this thesis constructs the grid scale suitability index through two aspects: heterogeneity and numerical accuracy.For heterogeneity,the difference in population distribution has the best effect on the two grid scales of 200 m and300m.For the numerical accuracy,the relative error value reaches the minimum value of28.8% at a grid scale of 300 m.Comprehensive analysis shows that the most suitable grid scale for Chongqing is 300 m.