A feasible framework to downscale NPP-VIIRS nighttime light imagery using multi-source spatial variables and geographically weighted regression

Abstract

The cloud-free monthly composite of global nighttime light (NTL) data of the Suomi National Polar-orbiting Partnership with the Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) day/night band (DNB) provides indispensable indications of human activities and settlements. However, the coarse spatial resolution (15 arc sec) of NTL imagery greatly restricts its application potential. This study proposes a feasible framework to downscale NPP-VIIRS NTL using muti-source spatial variables and geographically weighted regression (GWR) method. High-resolution auxiliary variables were acquired from the Landsat 8 OLI/ TIRS and social media platforms. GWR-based downscaling procedures were consequently implemented to obtain NTL at a 100-m resolution. The downscaled NTL data were validated against Loujia1-01 imagery based on the coefficient of determination (R2) and root-mean-square error (RMSE). The results suggest that the data quality was suitably improved after downscaling, yielding higher R2 (0.604 vs. 0.568) and lower RMSE (8.828 vs. 9.870 nW/cm2/sr) values than those of the original NTL data. Finally, the NTL was extendedly applied to detect impervious surfaces, and the downscaled NTL had higher accuracy than the original NTL. Therefore, this study facilitates data quality improvement of NPP-VIIRS NTL imagery by downscaling, thus enabling more accurate applications.