Eigenvector Spatial Filtering Modelling And Spatial-temporal Analysis Of China GDP Based On Nighttime Light

Gross Domestic product(GDP)is an important reference value for assessing regional economic development and managing production patterns between regions.Large-scale GDP fitting research has become a hot research topic in recent years.There are two main trends in exploring how to fit regional GDP with high accuracy: one is to fit GDP with the convenience and relevance of nighttime light remote sensing data;the other is to use machine learning algorithms such as least squares and neural networks to improve the accuracy of fitting GDP.However,there are several shortcomings in previous studies,many fitting methods ignore the spatial correlation of GDP distribution,which often shows that the model residuals have spatial autocorrelation phenomenon.On the other hand,few studies have compared the applicability of different nighttime lighting data for fitting regional economies at different scales,the applicability of the model to different study scales,etc.This study corrects the deficient points in the previous studies,and the main research contents are as follows.(1)Multi-scale GDP time-series regression based on nighttime lights and eigenvector spatial filtering(ESF)method.In order to attenuate the interference of spatial autocorrelation phenomenon on GDP fitting analysis,this study introduces ESF method into China multi-scale time-series GDP fitting with Chinese prefecture-level cities as the main research scale,and performs three-year time-series regressions in2013,2015,and 2017.To investigate the applicability of different nighttime lights,the effects of three kinds of nighttime light data on the fitting accuracy are compared,namely two NPP-VIIRS nighttime light monthly synthetic data and Liaojia-1 data.The results show that the ESF method can effectively eliminate the spatial autocorrelation in the model residuals,and significantly improve the multifaceted accuracy of the model compared with the least squares method and the spatial error model.The accuracy difference between the three different nighttime lighting data comparisons is not obvious,and the choice of the model plays a crucial role.According to the results of different scales,the ESF method does not eliminate spatial correlation well for provincial areas with a small number of cells,so it should be used with caution.(2)Economic contribution analysis of nighttime lighting and population based on ESF method and eigenvector spatially filtered unconditional quantile regression(SFUQR).According to the results of multiple models on the time-series fitting of China’s municipal GDP,the coefficient sums of population and nighttime lighting factors are increasing,indicating the increasing importance of resident population and regional construction for local economic activities.The eigenvector spatial filtering factor is introduced into the unconditional quantile regression to further analyze the effects of regional resident population and local nighttime lighting on cities with different levels of economic development.The selected study scale is a prefecture-level city in inland China,and the results for the three years 2013,2015,and 2017 show that the economic contributions of resident population and average night light brightness are positive for cities with different levels of economic development,i.e.,they always promote urban economic development.However,the two factors promote cities with different GDP quartiles in opposite trends,with the correlation between night light and GDP increasing and population showing a decreasing trend as the quartiles increase.And this trend is the same for all three years.Considering that nighttime lighting is related to urbanization progress,urban construction and other factors,this study can provide reference for cities with different development bases.(3)Development of a time-series GDP-nocturnal light Web analysis system for prefecture-level cities in China.Based on the aforementioned GDP fitting and analysis results of Chinese prefecture-level cities,we analyzed and designed the system in terms of requirement analysis and overall design,and developed the interface and map operation,query of model fitting results and true values,and query of potential spatial patterns of Chinese prefecture-level cities with 342 municipal administrative units as application examples.It is used for the integration and display of time-series GDP analysis results of Chinese prefecture-level cities.