Multi-Source Nighttime Light Remote Sensing Based Urban Built-up Area Extraction Study

The rapid advancement of urbanization will bring a large expansion of urban built-up areas.Accurate extraction and monitoring of urban built-up areas will help to recognize and understand the development process and spatial characteristics of cities,minimize the negative impacts by future urbanization.Nighttime light remote sensing provides powerful data support for the monitoring of urban built-up areas.Currently,the main types of nighttime light data sets include DMSP/OLS,NPP/VIIRS,and Luojia1-01.The stronger the nighttime light is,the more likely the area is considered as a potential built-up area,The quality of difference nighttime light data sources will have an impact on the extraction of urban built-up areas.However,the existing methods for extracting built-up areas by nighttime light remote sensing lack obvious boundaries between nighttime light data,and there are few comprehensive comparisons of different nighttime light data sources under various built-up area extraction methods.This paper takes Nanchang City as the study area and uses DMSP/OLS,NPP/VIIRS,Luojia1-01 nighttime light data as primary data source.Firstly,the differences in spatial patterns and influencing factors between multiple sources of nighttime light data are clarified,and then different built-up area extraction methods are applied to multiple sources of nighttime light data to evaluate the extraction results.The details of the study are as follows:(i)the spatial pattern of multi-source nighttime light data in urban areas is analyzed by methods such as circle analysis and spatial autocorrelation;(ii)the differences in influencing factors between NPP/VIIRS and Luojia1-01 light intensity are explored by geographic detectors and geographically weighted regression;(iii)statistical data comparison method,support vector machine supervised classification method,Habitat Index method,and Laplace-based improved Sobel edge detection method are used to evaluate the performance of built-up area extraction under different nighttime light data sources.The results show that:(1)The spatial pattern of nighttime lighting in urban areas becomes more complex with the promotion of the data quality.The circle pattern of multi-source nighttime light intensity generally follows the pattern of decreasing from the urban core to the periphery,The higher quality of light data showing greater fluctuations between circles.There is a clear spatial clustering of the three types of nighttime light data in the main urban area of Nanchang,and the hotspot areas will further differentiate with the improvement of the quality of nighttime light data.(2)The influencing factors of nighttime lighting will change with the improvement of data quality.high-value image elements of NPP/VIIRS and Luojia1-01 are mainly distributed on impermeable surfaces,and the light changes of Luojia1-01 under different land cover types are more drastic than those of NPP/VIIRS.POI density,population density,road network density,surface temperature,and building density have a significant influence on NPP/VIIRS and Luojia1-01.NPP/VIIRS and Luojia1-01 light intensity were all positively affected.POI density,population density,and surface temperature have a stronger influence on light intensity in urban areas and a weaker influence in non-urban areas,while the opposite is true for road network density and building density.A ll the factors have a stronger influence on NPP/VIIRS than Luojia1-01.(3)The generalisability of the selected built-up area extraction methods,in descending order of strength,are Habitat Index,Support Vector Machine supervised classification,Statistical Data Comparison Method,and Laplace's improved Sobel edge detection.The DMSP/OLS and NPP/VIIRS are corrected with auxiliary data and form a multi-source nighttime light collection with the original light data.Overall,the Habitat Index performs best,with an average kappa coefficient of 0.809,preserving the structural features within the built-up area.The average kappa coefficient for the support vector machine classification is 0.677,which is prone to misclassification at the edges of built-up areas.The mean kappa coefficient of the Statistical Data Comparison method is0.534,and the higher quality of the lighting data is significantly more accurate with this method.The average kappa coefficient of Laplace's modified Sobel edge detection is0.442,which does not apply to NPP/VIIRS and Luojia1-01 in terms of the morphology and accuracy of the results.