Research On Population Spatialization Method Based On Multi-source Data

Demographic data provode support to the analysis of regional natural resources and social factors.Mastering the detailed population data of the administrative unit has played an important role in understanding the region's economy,human geography and local policies.Currently,the demographic spatialization is based on the population statistic of administrative unit obtained by the statistical bureau of the state or local government.Hence,the temporal and spatial resolution of statistical population data is low,and it is impossible to display the aggregation situation of population in the administrative unit.This articel aims to discover a method of using auxiliary data to achieve high temporal and spatial resolution population spatialization and explore the relationship between people and natural resources,the environment,and economic growth.However,the Tencent location big data compensate for the disadvantadges of the aforementioned data.With the assisstence of multi-source data,achieve the 500*500m grid scale population spatialization of Xinjiang Uygur Region using Geographically weighted regression,XGBoost,Random Forest and modified Random Forest,also simulate the spatial aggregation and impact factor of Xinjiang Uygur Region.The contributions of this articel:(1)Based on the existing studies,this paper elaborates the data that can be used in the study of population spatial distribution,including night light,land cover,POI data,etc.Due to the massive spatial geographic data generated by Tencent users' network requests of location services,it contains rich location and attribute information and can reflect the dynamic change of population.Therefore,This paper use this data as an important factor to reflect the spatial population.(2)Use four methods for contrast experiment of population spatial visualization,which include geographically weighted regression,XGBoost,random forest and improved random forest.These four methods are used in the population spatial experiment on the 500*500m grid in Aksu,Xinjiang.In the process of improving random forest,features are randomly selected from multiple feature subspaces.The feature space is divided to multiple subregions by the importance of feature according to Fisher ratio.This method as a better performance when processing data sets with high dimentional and redundant characteristics.According to the accuracy comparison of the four models,the modified random forest get higher accuracy and has strong practicability on the experiment of spatial population.At the same time,software was bulided up by Python to display the accuracy comparation of three machine learning methods.(3)The spatial distribution of population was explored and the influencing factors of population distribution were analyzed,by comparing the population spatial data.The density of population is higher in the residential land near water source bucause of drought,so the rivers play an important role in Aksu area.The government of Aksu Prefecture of Xinjiang Uygur Autonomous Region can applypopulation spatial localization result into improving the ability of public resource allocation,industrial agglomeration and urban construction planning,and the results can also provide support for the proposal of local economic policies and population introduction policies.The quantitative analysis of Aksu in Xinjiang Uygur Autonomous Region is expected to provide some reference for urban vitality analyse of similar cities.