Research On Spatial Fine Distribution Of Urban Population Based On Multi-source Data

Using gridded population map to reveal the distribution and quantity of population in geographical space is of great significance in the fields of urban planning,disaster prevention and mitigation,resource structure optimization,regional economic development and so on.The traditional population data is based on the minimum administrative unit statistics of the census data,the census data has limitations,its accuracy to villages and towns(streets),can not understand a more fine population distribution,at this stage,all kinds of population data set population surveying and mapping mostly use coarse resolution night light(NTL)data for large-scale population spatialization.In the aspect of fine resolution population mapping,the existing results of population spatialization mostly use the fusion of multi-source data(such as POI data,DEM data,etc.)to study the population distribution in provincial and municipal areas,without taking into account the impact of spatial heterogeneity,but the population distribution pattern and spatial distribution are complex and uneven,often affected by the aggregation effect,seriously affecting the results of population spatialization.In this paper,through the study of the zoning of Zhengzhou under the zoning mechanism,Zhengzhou is divided into agglomeration and non-agglomeration areas,and the spatialization of population data is studied according to different population agglomeration categories.This paper mainly adopts random forest algorithm and multiple linear regression model,based on the human geography conditions of the study area,five kinds of data including digital elevation model,land use type,interest point,road network and vegetation coverage are selected as the influencing factors of population distribution in this study,and 14 main factors are selected as influencing factors of population distribution through correlation analysis.The population spatialization on 150x150 m grid scale in agglomeration and non-agglomeration areas of Zhengzhou was studied in different agglomeration areas.The spatial pattern of the results of population spatialization on grid scale is analyzed and discussed.The main work of this paper includes the following aspects:(1)Multi-source geographic data and population data factor extraction,through Pearson correlation analysis,select the POI index with greater correlation,and eliminate the smaller indicators.POI data select five POI indicators,such as catering distribution,science and education culture distribution,and government agencies and social groups distribution for this study.Through the correlation analysis of multi-source geographic data and population data and multiple collinearity test,Zhengzhou is divided into agglomeration area and non-agglomeration area based on the zoning mechanism of population agglomeration degree,and a grid of 150 × 150 m is constructed according to different agglomeration degree categories.Nine factors,such as vegetation coverage,slope and rail transit core density,are selected for this study,a total of 14 factors are used to calculate the average value of each factor in each township(street)pass network.(2)Based on multi-source geographic data,four kinds of grid population refinement models are established based on random forest algorithm and multiple linear regression algorithm,and the accuracy of the model is evaluated.the random forest algorithm model performs well in different agglomeration areas,and the multivariate linear regression model performs well in agglomeration areas,but poor in nonagglomeration areas.(3)Based on the comparison and characteristic analysis of population spatialization models,the optimal model is selected to visualize the grid-scale population data,and the contribution weight of multisource geographic data to the model under different agglomeration degrees is discussed.in the agglomeration area,the distribution of government agencies and their social groups in POI data makes the greatest contribution to the model,the slope aspect of topographic factors contributes the least to the model,and the distribution of buildings in non-agglomeration areas makes the greatest contribution to the model.Vegetation coverage makes the least contribution to the model.(4)The spatial pattern analysis of the grid-scale population data distribution under different population agglomeration degree categories shows that the population in the agglomeration area has obvious spatial correlation,and the population distribution has a very strong spatial agglomeration effect.The population distribution in non-agglomeration areas shows the trend of random distribution,which confirms the correctness of the modeling based on the zoning mechanism based on population agglomeration degree classification.