Research On Population Spatialization Based On Random Forest Model And Multi-source Geographic Big Data

The study of population spatialization is an important way to realize the delicacy management of urban space,and it is the only way for coordinated development of resources and environment in villages and towns.By reducing the dimensionality of the population data at the administrative level to a fine-grained grid scale,it is possible to dig deeper into the spatial distribution characteristics of the population density and the heterogeneity of geographic space in the region.As we all know,our country's population census is conducted every ten years,counting with the township as the smallest unit.However,with the continuous development of computer science and geography,its data accuracy can no longer meet the data requirements of modern geographic scientific research.Population spatialization is based on national demographic data,the idea of multi-source data fusion and the technical means of data mining,through specific statistical methods to decompose large-scale population data to the corresponding grid.In addition,population spatialization recreates national demographic data with a more precise spatial scale,which is helpful to grasp the population distribution pattern in fine granularity.Furthermore,it provides a data basis for the research on the coordinated development of villages and towns and the fine management of urban resources.In this paper,Hubei Province was the study area,we combined the two emerging geographic big data including points of interest and building outlines with the traditional geographic data.Then,we extracted the features related to population settlements.And the spatial data set of population data in Hubei Province is constructed.Later,we predicted the population distribution that on the grid scale through the model of machine learning.Experimental results and model performance of population spatialization in Hubei province had been evaluated accurately.Finally,the population spatialization results are matched with the grid data,and the 1000-meter population distribution map of Hubei Province is output through the ARCGIS platform.Further,the spatial autocorrelation analysis of the population concentration degree of Hubei Province is made through the Moran index.The main research results of this article can be divided into the following aspects.First,3D image data of baidu map building was obtained through the PYTHON programming language based on the idea of multi-source data fusion.Building contour data is used for the first time in the study area at the provincial level to construct the spatial feature data set of population in Hubei Province through the secondary development and feature extraction of the building outline data and fusion with other features.Second,We used the demographic data of Hubei Province in 2010 as training data and predicted the population density distribution of population grid in Hubei province through the random forest model that based on machine learning models and multi-source geographic big data.The goodness-of-fit value of the model with high accuracy is 0.902 indicating that the model has a strong generalization ability.Third,the spatial distribution of population in Hubei Province is visualized.We successfully connected the population spatialization research of Hubei Province with grid data,and outputs a population distribution map on a 1000-meter grid scale in Hubei Province.Simultaneously,the concentration degree of population distribution in Hubei Province is analyzed via spatial auto-correlation analysis.