Research On Nighttime Light Remote Sensing Identification Method Based On Spatiotemporal Structural Characteristics Of Coastal Urbanization In The Yangtze River Delta

The rapid advancement of urbanization has reshaped the types of natural surface cover,resulting in ecological and environmental pollution and social problems such as reduced surface runoff,water pollution,urban heat islands and urban waterlogging.To assess the impact of urbanization on ecosystems and regional sustainability,and to mitigate the negative environmental impacts of urbanization in my country,it is necessary to comprehensively understand the characteristics and changes of urban patterns,and rationally adjust and optimize urban land use patterns.With nighttime artificial lights applied to most buildings and infrastructure,the unique perspective of nighttime lights enables the detection of human activity dynamics that cannot be captured during the day.Therefore,nighttime light remote sensing data has been widely used in urban spatiotemporal pattern evolution and human activity monitoring.The Defense Meteorological Satellite Program Operational Linescan System(DMSP-OLS)nighttime light data and the Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite(NPP-VIIRS)nighttime light data are two widely used nighttime light datasets.However,the differences between DMSP-OLS and NPP-VIIRS in terms of their spatial resolutions and sensor design,which requires a cross-sensor calibration of these two datasets on which to conduct long-term urbanization evolution process analysis.In this paper,the three fitting methods of power function model,geographic weighted regression(GWR)model and Sigmoid function model are compared.The accuracy evaluations show that the Sigmoid function model has the highest fitting accuracy,the goodness of fit R~2 is 0.90,and the RMSE(root mean square error)was 6.29,the SNDI(normalized coefficient of difference index)were 6.02×10~4,and the spatial profile analysis and histogram pixel-by-level comparison analysis showed that the Sigmoid function model had the highest consistency with the original DMSP-OLS data.A set of long-term DMSP-OLS night light data generated based on the Sigmoid function model has excellent spatial pattern and temporal consistency,which is similar to DMSP-OLS data(hereinafter referred to as DMSP-OLS-like data).Furthermore,the generated DMSP-OLS-like data can provide a way to monitor the dynamics of population and socioeconomic activities over a longer period than other remote sensing data,land use data or statistics.The main research contents include:(1)Due to the lack of on-orbit radiometric calibration of DMSP-OLS data,data in different years and between different sensors in the same year are incompatible with each other.Select the method of step-by-step correction to adjust the pixel with the minimum times of modification,and establish a stepwise quadratic polynomial regression model between the data to be calibrated and the basic calibration data,and adjust the DMSP-OLS data for years to be comparable and usable in the next fitting state,and ensure that the range of the sum of the brightness of the pixels before and after the modification remains basically unchanged.(2)The NPP-VIIRS data may be disturbed by the noise of extreme events such as aurora,fire,etc.Therefore,the highest radiation value in the center of Shanghai in the annual NPP-VIIRS data is used as the threshold,and the outliers are screened out and used in the surrounding eight neighborhoods.Maximum value replacement until all cell radiance values are within the threshold.(3)Power function fitting,geographically weighted regression fitting and Sigmoid function model fitting were performed using the DMSP-OLS data and NPP-VIIRS data of the overlapping year of 2013,respectively.The first two methods both use Gaussian low-pass filtering to smooth the NPP-VIIRS data.The difference is that the power function uses a 13×13 pixel moving window,while the geographically weighted regression fit uses a 5×5 pixel moving window.The smoothing method of the sigmoid function model is different from the previous two methods,using the kernel density method and setting the size of the circular moving window to 5 times the spatial resolution of NPP-VIIRS,i.e.,2500 m.The accuracy verification show that the Sigmoid function model fitted best,and this method of kernel density space aggregation will not cause blanks at the edge of the image,which is the most reasonable image smoothing method.(4)By constructing urbanization level indexes based on nighttime lights,analyze the spatiotemporal changes of the light index,area index,city index and its growth rate in Jiangsu,Zhejiang,Shanghai and Anhui from 2000 to 2020.The rapid increase in the area index is the main determinant of the urbanization level,indicating that the urbanization process in the Yangtze River Delta region were mainly in the form of area expansion before 2020.(5)Hot spot analysis and clusters and outliers analysis of the SUM,MEAN,MAJORITY and STD of night lighting statistics at district and county levels show that the urbanization level and the rate of urbanization in the central urban area of Shanghai,Nanjing and Hangzhou are relatively faster,and showing a trend of continuous radiation to the outer urban areas.