顾及参数空间平稳性的地理加权人口空间化研究

Population spatialization based on geographically weighted regression model considering spatial stability of parameters

近年来,人口空间化的方法理论愈趋成熟,但对人口空间化建模中变量参数的空间平稳性处理却鲜有人关注。以土地利用数据、夜间灯光数据和人口统计数据为数据源,提出一种基于半参数地理加权回归模型(semi-parametric geographically weighted regression,S-GWR)的人口空间化方法,并利用该模型在县级尺度进行常住人口空间化建模,最后以四川省为研究区进行比较论证。在分析变量特征的同时,利用S-GWR模型处理参数变量的空间平稳性,以提高人口估计的精度,最后生成四川省2010年1 km分辨率的人口空间分布图(spatial distribution of population,SDP)。结果表明,S-GWR模型的决定系数为0.903,比传统回归模型表现更好,模型拟合的效果更优。精度验证方面,通过2个常用的人口数据集进行精度对比验证;在县一级,研究区整体SDP的平均误差和每个区县的相对误差都接近于0,比其他2个数据集有更高的精度;在乡镇一级,SDP的平均相对误差、平均绝对误差和均方根误差分别为34.54%,5715.703人和12085.932人,均比其他2个数据集的误差更小,离散度效果更优;从乡镇准确估计个数来看,SDP准确估计的个数最多,达185个。因此,考虑参数的空间平稳性可以提高人口空间化的精度。

The theories on population spatialization tend to be mature in recent years.However,the spatial stability of the variables and parameters used in population spatialization modeling has been scarcely focused on.With the land use data,night-time light data,and demographic data as the data sources,this study proposed a novel precise population spatialization method based on a semi-parametric geographically weighted regression model(S-GWR).Then a permanent population spatialization model on a county scale was built using the method proposed in this study and then was verified using the Sichuan Province as the study area.In this study,the spatial stability of parameters and variables were obtained using the S-GWR model while the characteristics of the variables were analyzed,in order to improve the accuracy of population estimation.Finally,the population spatial distribution map(SDP)with a resolution of 1 km of Sichuan Province in 2010 was formed.The results show that the coefficient of determination coefficient of the S-GWR model was 0.903,which is higher than that of traditional regression models and indicates better fitting effects.The S-GWR model was verified using two commonly used population datasets,and the verification results are as follows.At a county level,the overall average error of the study area and the relative error of each district and county in the SDP all approximated to 0,and thus the SDP was more precise than the other two datasets.At a township level,the mean relative error,mean absolute error,and root mean square error of SDP were 34.54%,5715.703,and 12085.932,respectively,which were all lower than those of the other two datasets.Meanwhile,the SDP showed more favorable dispersion effects than the other datasets.Furthermore,the number of the towns whose population was accurately estimated was 185 in the SDP,which was higher than that in the other two datasets.Therefore,the accuracy of population spatialization can be improved by considering the spatial stability of parameters.