Research On Population Spatialization In Northeast Black Soil Region Based On Random Forest Model

Accurate population spatial distribution information is of great significance for regional infrastructure construction,resource allocation and disaster assessment.There are not only densely distributed areas,but also broad areas with relatively sparse population distribution in the northeast black soil region.The state of population distribution is unbalanced and uneven,with strong spatial heterogeneity.Therefore,it is of great significance to simulate the spatial distribution of population in the northeast black soil region.In this study,the northeast black soil region is taken as the research area,and elevation and slope,air temperature,land use,road network,interest points,location of the community,night light and other data are selected as factors affecting population distribution from the perspectives of natural environment and social economy.First,based on the seventh population census data in 2020,combined with the district-county scale administrative division data,the random forest model,extreme gradient lifting model and BP neural network model were used to model and compare the fitting accuracy of the models.Finally,the random forest model was selected to realize the population grid in the northeast black soil region.The population spatial distribution data of 500m×500m is obtained.Secondly,the accuracy of the model was compared and evaluated with the existing World Pop,Land Scan and GPW spatial distribution data from the perspective of the whole and subdivided population density interval,so as to verify the reliability of population gridding by using random forest model.Then the characteristics of population distribution are analyzed from the perspective of space and quantity.Finally,the factors affecting population distribution are discussed from qualitative and quantitative perspectives.The main research conclusions are as follows:(1)The R~2 of population spatial model based on random forest model and characteristic index database is 97.20%.The accuracy evaluation results show that each index is superior to Land Scan,GPW and World Pop population density data from the perspective of the whole and subdivided population density interval,and the random forest model has obvious advantages.On the whole,random forest model has certain advantages in the application of population spatialization in the vast area with sparse population distribution.(2)From the perspective of spatial distribution,the spatial distribution of population in the northeast black soil region presents an obvious agglomeration effect,showing a pattern of multi-center and marginal.In terms of number,the average population density of the whole region was lower than that of China in the same year,and there were fewer grids with higher population density.In the unit grid population classification statistics of northeast black soil,the cumulative ratio of total population in each unit grid population interval is always smaller than the cumulative value of area.Overall,the population distribution characteristics of the northeast black soil region are as follows:elevation within 500m,slope below 2°,temperature between 4℃and12℃,density of POI nuclei between 0 and 53 per square kilometer,distance from road within 10km,and brightness value of night light below 50.(3)In the correlation analysis between population density and various indicators,the comprehensive information of POI kernel density and the mean value of night light brightness showed a strong positive correlation.Combining the results of correlation analysis and characteristic importance evaluation,it is found that the natural environment factors determine whether there is population in a certain area,while the social and economic factors determine the number and pattern of population distribution in this area.