STARS: A novel gap-filling method for SDGSAT-1 nighttime light imagery using spatiotemporal and spectral synergy

Abstract

The Sustainable Development Goals Satellite 1 (SDGSAT-1), equipped with the Glimmer Imager (GLI), provides high-resolution nighttime light (NTL) data across multiple spectral bands. Thus, it can notably monitor human dynamics and light pollution with enhanced spectral and spatial resolution. However, cloud cover and low-quality observations often contaminate the SDGSAT-1 GLI NTL data, limiting its effectiveness. This challenge is addressed by the development of a novel method, namely the SpatioTemporal And spectRal gap-filling method for Sdgsat-1 (STARS) GLI NTL images, which combines spatiotemporal and spectral information to generate cloud-free NTL images with satisfactory pixel brightness and continuity. STARS is the first method to effectively address gap-filling in multiband NTL data using RGB spectral information, even with irregular time intervals and limited image inputs. Compared with traditional methods such as the temporal gap-filling method and the mean-weighted gap-filling method, the Cloud Removing bY Synergizing spatioTemporAL information (CRYSTAL) method, and the spatial and temporal adaptive reflectance fusion model (STARFM), which do not specifically account for the differences in light source variations in multi-band NTL data, STARS demonstrates superior performance (higher R-squared (R²) and lower root-mean-square error (RMSE)) in simulations across seven global cities, demonstrating its effectiveness in filling cloud-induced gaps in multi-band NTL data. On average, STARS achieves R² values for the gap-filling results compared to the actual values of 0.79, 0.78, and 0.70 in the RGB bands, respectively. The cloud-free images produced by STARS extend the time series of the SDGSAT-1 GLI NTL data, supporting multitemporal quantitative analysis. In cloudy regions like Tianjin, China, STARS effectively captures dynamic changes in NTL before and after the Spring Festival, closely matching human activity patterns from Baidu Maps, both spatially and temporally. Overall, STARS offers an innovative and effective approach for gap-filling multiband NTL data, with potential applications in similar datasets.