Study Of Population Density Based On Random Forest Model

Population density is a measure for studying population distribution laws.Highresolution population density maps are important basic data for improving the level of urban planning,improving emergency management capabilities,and optimizing resource allocation.Population density mapping is often restricted severely by basic data.What kind of basic data there is,people will construct a population density calculation model that matches the basic data.In the era of big data,data on natural endowments,economic endowments,and innovation endowments are extremely rich,providing a data basis for compiling high-resolution population density maps.In the age of artificial intelligence,machine learning algorithms provide non-linear algorithms for compiling population density maps.In this paper,the population density mapping experiment area in Shijiazhuang was sampled at the township and village levels to construct corresponding sample datasets.The population density mapping experiment was carried out with the help of random forest models.Firstly,based on POI data,a population density mapping experiment was carried out.The following population distribution rules are summarized: ? In the population density maps based on the POI kernel density(x_rkmdp,c_rkmdp),the population density "Duneng Ring" generally exists around towns,and the population density decreases rapidly from towns to villages.? In the population density map based on the POI distance(x_rkmdd,c_rkmdd),there are "Point-Axis" structures of population density.The population density near the "Point-Axis" is significantly higher,reflecting that the economic location and traffic location significantly affect the population distribution.Secondly,according to the three dimensions of natural endowments,economic endowments and innovation endowments,the population density influencing factors are selected and the population density mapping experiments are carried out.Normalized difference vegetation index(NDVI),digital elevation model(DEM),terrain fluctuation,distance from river,annual average temperature,annual average precipitation are selected to characterize natural endowments characteristics.Nighttime lights(NTL)imagery,and traffic accessibility are selected to characterize economic endowments characteristics.POI distance combination map and POI kernel density combination map are selected to represent the characteristics of innovative endowments.Natural endowment,economic endowment,and innovation endowment jointly describe the comprehensive endowment space that affects population density.Based on the random forest model,population density maps(x_rkmd,c_rkmd)based on comprehensive impact factors are compiled at the township scale and village scale.Comparison of the six sets of population density maps obtained in this paper and multi-scale population density maps(R=13,R=30,R=100),the sjz_WorldPop map(2007),the sjz_PoiPop map(2010),the sjz_CnPop map(2010)forms the new understandings as follows.? As the sampling accuracy of the training sample dataset is refined from the county level to the village level,the underestimated population density in urban areas and overestimated population density in rural areas have improved significantly.? In theory,because the sampling unit(county,township,and village)and the output unit(100m×100m)do not match,population density mapping algorithms based on random forest models exist "Ecological Fallacy",whether sampling by county,township,or village.Overcoming the "Ecological Fallacy" is the key to improving the population density models based on the random forest models.? By introducing the POI distance data,the "Point-Axis" structures of population density can be characterized.? By introducing the POI kernel density data,the "Duneng Ring" structures of the population density can be characterized.? Natural rivers have a significant impact on population density,and artificial rivers have no significant effect on population density.If natural rivers and artificial rivers are mixed into rivers and brought into random forest models,they will have incorrect interference on the population density map.? Nighttime lights imagery has "Halo Effect",which is replaced by the POI kernel density combination map,which will improve the quality of the population density map.? In order to ensure the computational validity of the population density dataset,it is necessary to introduce POI data differently and selectively.? When there are contradictory population distribution laws in the study area(for example,on both sides of a natural river,the population density will be high if it is located in a mountainous area,and the population density will be low if it is located in a plain.),the random forest model alone cannot overcome them at this time.The study area must be divided into mountainous areas and plains before applying random forests models to overcome them.? The path of population evolution in urban and rural areas is different.Zoning urban and rural areas will further improve the accuracy of population density mapping.Studies show that in the future,high-resolution population density mapping requires natural comprehensive zoning(landform zoning)and urban-rural zoning firstly,and zoning according to comprehensive endowments,and then using random forest models for zoning,which is expected to improve the validity of population density mapping.It is necessary to carefully make various endowment factors.For example,river dataset should only include natural river data,and artificial rivers(canals,jianhe,irrigation channels,etc)should not be introduced.Using the POI kernel density combination map instead of nighttime lights imagery will improve the level of refinement of urban location factors.POI data should be selected in the innovation endowment data according to whether it is conducive to information communication.Relying on the minimum granularity population density data,random sampling will be carried out by districts,and the population density training sample dataset will be constructed by districts.It is expected to overcome the "Ecological Fallacy" and further improve the accuracy of population density mapping.