NPP-VIIRS与加权回归的人口网格化方法改进——以福建省为例

The Improvement of Population Gridding Method Based on NPP-VIIRS and Weighted Regression:Taking Fujian Province as an Example

基于GIS空间分析,利用NDVI修正NPP-VIIRS夜间灯光数据过饱和与溢出效应(定义为NANI模型)后作为自变量,采用直接面向格网的加权个案和按县域中有人口分布的网格数加权最小二乘法进行回归;进而按镇乡尺度零误差调整,实现福建省人口普查数据空间网格化的方法改进,并用随机抽取乡镇域样本验证改进后效果.结果表明:改进后的人口网格化既考虑了县域规模差异的影响,又避免了建模过程中行政区划与网格之间的尺度变换.该结果通过了P=0.001的显著性检验,网格人口与乡镇域人口数的平均和加权平均的绝对误差、圴方根误差、相对误差均比方法改进前小,显著提高了人口网格化的精度.

Based on GIS spatial analysis,using NDVI vegetation index data to correct NPP-VIIRS nighttime light data have oversaturation and spillover effects(defined as the NANI model),and then the data results as independent variables,using directly oriented grid-oriented weighted cases and the number of grids with population distribution by county the weighted least squares method was used for regression improvement,and the population census data in Fujian Province is spatially gridded,and the improvement effect was verified by random sampling of township samples.The experimental results showed that the improved population gridding not only considered the influence of county scale differences,but also avoided the scale transformation between adminwastrative divisions and grids in the modeling process.The improved gridding results pass P=0.001.The significance test of,the mean absolute error,square root error,and relative error of grid population and township population were smaller than those before improvement,which significantly improved the accuracy of population gridding.