地理加权回归在人口空间分布研究中的应用

Application of Geographical Weighted Regression in Population Spatial Distribution Research

针对基于多元线性回归的人口数据空间化方法忽略研究区间空间异质性的问题,采用分区回归和地理加权回归改进。以西南地区为例,以县级为研究单位,利用全球地表覆盖数据(GlobeLand30),对西南地区2010年县级人口统计数据进行空间化模型研究。采用二种不同方式进行改进,其一是利用影响人口空间分布的因素进行人口一致性分区,依据分区结果分别进行基于地表覆盖类型的统计分析建模;其二是引入局部回归建模思想的地理加权回归分析方法,利用局部最小二乘方法进行逐点估计。最后,基于误差理论,对人口数据空间化的结果进行误差比较和模型精度分析。研究结果表明,运用分区回归模型和地理加权回归模型提高了人口数据空间化的数值精度和空间精度。

To solve the problem of spatial heterogeneity in population based on multivariate linear regression,we propose the improvement by partitioning regression and geographical weighted regression.This paper takes the southwest region as an example,uses the global land cover data(GlobeLand30)and carries on the spatial model research to the county level demographic data of the southwest region in 2010.On the basis of traditional methods,two different ways are used to improve.One is the population consistency partition using the influence factors of population distribution,and the statistical analysis modeling of land cover types according to the zoning results;the second is the introduction of the local geographical weighted regression modeling thought regression(GWR)analysis method,using local least squares method point estimate.At last,the results of error comparison and model accuracy analysis of the spatial distribution of population data are analyzed by using the error theory.The results show that the numerical and spatial precision of population data are improved by the use of zonal regression model and geographically weighted regression model.