Estimation Of GDP Based On Long Time Series Of Nighttime Light Images

Gross domestic product(GDP)is one of the basic indicators to measure the economic development of a country or region,while DMSP-OLS and NPP-VIIRS nighttime light data are closely related to economic development and are widely used to estimate socioeconomic parameters.The time coverage of the DMSP-OLS data covers from 1992 to 2013,while the NPP-VIIRS data is available from 2012.Therefore,this paper proposes a simple data assimilation method to integrate DMSP and NPP nighttime light data into a long-term sequence of nighttime light datasets,which will help to study long-term socioeconomic development patterns.First,in this paper,the DMSP/OLS nighttime satellite data of China from 1992to 2013 were used to find the relationship between GDP and nighttime satellite data.The nighttime satellite imageries were corrected with mutual correction,saturation correction and continuity correction through the invariant target region method.Then the lighting information of Mainland China and 31 provincial regions were extracted and models,including linear,quadratic polynomial,power function and exponential regression model,between GDP and light information,were tested to find the optimal one.The results showed that 1)The corrected DMSP/OLS nighttime satellite data is more stable and continuous than uncorrected data;2)There is a strong correlation between corrected DMSP/OLS night light dataset and GDP;3)Exponential model,with0.97 of R2 and 11.32%of MARE,was the most suitable one for predicting GDP of Mainland China.;4)Provincial models of long time series are better than the annual provincial administrative regions models.Also,the exponential function model was optimal for four municipalities and the top six provincial economic entities,and the quadratic polynomial model was optimal for other administrative regions,whose R2 is above 0.95 and MARE is about 10%.Second,this paper preprocesses the 2012-2017 NPP nighttime lighting data to remove background noise,highlighting outliers and negative values,and constructs relationship between TNL of DMSP nighttime light data and TNL of NPP nighttime light data at the provincial scale in 2012 and 2013.The results showed that the power model was optimal,R~2 was 0.9417,and the two nighttime light datasets were integrated by a power model to form a 1992-2017 long-term sequence nighttime light dataset.Long-term sequence TNL of the dataset continues on both provincial and provincial scales.In addition,the average relative errors using NPP nighttime data to simulation TNL of DMSP in 2012 and 2013 were 15.63%and 16.51%,respectively,and the accuracy was relatively high,which verified the feasibility of this data assimilation method.Then,we use linear,quadratic polynomial,exponentiation,and exponential models to model the relationship between TNL and GDP of China Mainland from1992 to 2017.The results show that the best GDP predictive model in Mainland China of 1992-2017 is the power model,R~2 is 0.9795,and the average relative error is12.23%,while the correlation coefficient of the quadratic,linear,and exponential models is high,but the MARE is larger.It has been impossible to accurately predict China's GDP.Considering that a single model is increasingly unable to reflect the relationship between TNL and GDP,this paper uses a simple combination model(power and quadratic item combination model,two quadratic item combination models,two power combination models,quadratic items and power combination models)to study the relationship between TNL and GDP in Mainland China from1992 to 2017.The results show that the quadratic power quadratic combination model is the best,R~2 is 0.9938,MARE is 6%,the precision is higher,and it is better than a single exponential model fitting.Finally,the paper uses four single models and power and quadratic combination model to study the long-term sequential model between TNL and GDP for each provincial administrative region from 1992 to 2017.The best single model for each province is a power model,with R~2 at 0.95.About 15.91%of the MARE.Except for Tianjin,Tibet,Guangxi,Sichuan and Chongqing,the power and quadratic combination model of other provinces is higher than the single model,and the correlation coefficient is higher.The relative error between the GDP forecast value and the statistical GDP is smaller and can better reflect the relationship between TNL and GDP.In addition,on the provincial level,the NPP nighttime lighting image simulating GDP accuracy(R~2 is above 0.8)is better than the DMSP nightlight image simulating GDP accuracy(R~2 is about 0.75).