A Novel Approach for Cloud-Free MODIS NDSI Reconstruction on the Tibetan Plateau Combining Spatiotemporal Cube and Environmental Features

Abstract

Snow cover is essential for the hydrological cycle and ecological balance of the Tibetan Plateau (TP). The normalized difference snow index (NDSI) is a widely used indicator for snow detection, yet extensive cloud cover often disrupts the spatiotemporal continuity of MODIS NDSI data. Given the close link between snow cover and environmental conditions, introducing environmental factors provides a novel perspective on reconstruction. Here, we developed a LightGBM-based NDSI reconstruction method that integrates the spatiotemporal cube with environmental features—meteorological, topographical, and geographical—along with a spatiotemporal reliability assessment. This method generated a robust, long-term, cloud-free MODIS NDSI dataset over the TP (daily, 500 m). Through simulation experiments, we evaluated the numerical, spatial, and classification accuracy of our method. Results showed that this method achieved high accuracy with averaged coefficient of determination ( $R^{2}$ ), mean absolute error (MAE), and root-mean-square error (RMSE) of 0.81, 0.090, and 0.138, respectively, while classification metrics overall accuracy (OA), $F1$ -score (FS), commission error (CE), and omission error (OE) of 0.94, 0.82, 0.038, and 0.20, respectively. Notably, incorporating snow-related environmental features resulted in superior metric accuracy, image quality, and spatial detail compared to spatiotemporal interpolation (SI) alone. Furthermore, the proposed method demonstrated higher accuracy during snow cover periods and in high-altitude regions on the TP. This novel approach to NDSI reconstruction enhances the understanding of snow accumulation and melting processes on the TP, offering a robust data foundation for climate change monitoring and hydrological modeling.