Development Of The Remote Sensing Datasets Of The Annual Nighttime Light In China From 1992 To 2021

Nighttime light remote sensing satellite data records the artificial light source at night on the earth surface,which can be used as an important reference to measure human social and economic activities,and has been more and more applied in monitoring human activities and urban changes.Among them,DMSP-OLS and NPP-VIIRS data are the most widely used.However,both kinds of data have defects in time series.OLS data remain from 1992 to 2013,while VIIRS data start from April 2012.At the same time,differences between OLS and VIIRS data sensors make it difficult to use data together.Problems with outliers caused by OLS data not being calibrated on board,image to image continuity,and VIIRS data not filtering stray light,etc.need to be further addressed.In view of the wide application of nighttime light data and some problems of existing data,this paper optimized data correction and data set construction process based on DMSP-OLS from 1992 to 2013 and NPP-VIIRS nighttime light data from 2013 to 2021.The annual nighttime light image datasets of 1992-2021 old series is constructed,and the reliability of the new datasets is verified by economic parameter data.Based on the correction method between DMSP-OLS data,reference images and pseudo invariant regions were selected to conduct internal calibration for a total of 34images acquired by multiple satellites during 1992-2013.Subsequently,the situation where multiple satellite images existed in the same year was corrected to ensure the uniqueness and continuity of data in the same year.In view of the existing research on the processing method of NPP-VIIRS outliers is reasonably optimized,the image is detected by using the outlier recognition method based on the spatial location relationship and the Digital Number(DN)attribute,combined with sliding Windows of different sizes.Google Earth images are used to analyze the outliers identified in detail,and at the same time,a reasonable threshold range is selected to remove and interpolate the outliers of the image,which improves the accuracy of outliers identification and enhances the rationality of data processing.Considering the different scales between DMSP-OLS and NPP-VIIRS and the unclear corresponding relationship between DN values,the coefficient of variation(CV)of each pixel of OLS and VIIRS images in 2013 was calculated,and the common stable pixel was extracted as modeling data.Then,the mean value of DN value coordinates in DMSP-OLS images corresponding to NPP-VIIRS pixels was obtained.By establishing the mapping relationship with DN values between the image,the interference caused by multiple mapping relationship in data matching to image fitting was solved.At the same time,Bi Dose Resp model was selected to establish the relationship for data fitting to obtain the coefficient.The goodness of fit(R~2)obtained is 0.98,meeting the conversion needs of Nighttime Light(NTL),and the model is applied to transform NPP-VIIRS images from 2013 to 2021,generating time series Nighttime Light datasets.For the generated NTL data set,R language was used to make statistics on the long sequence noctilucent data,and extract three light indicators,namely,the total amount of nighttime light in each year,the average brightness intensity of nighttime light and the comprehensive light index.At the same time,economic parameter data such as Gross Domestic Product(GDP)and Electric Power Consumption(EPC)were used for validation,and good goodness of fit was established to verify the accuracy of the constructed NTL data set.It also provides reference for the future application of light image data set.The developed nighttime light remote sensing datasets have a certain reference value for monitoring and analyzing human activities and urban changes.