Multitemporal land surface water mapping in Poyang Lake with the local temporal spectral mixture model

Inland surface water mapping is crucial for water resource management and understanding the hydrological cycle. However, the high seasonal variability of surface water necessitates continuous monitoring. MODIS data, with its high temporal frequency, is increasingly used for intra-annual and large-scale surface water mapping. This study focuses on Poyang Lake in China, where the low spatial resolution of MODIS data poses challenges due to spectral mixture effects. A sub-pixel surface water fraction model based on a water index regression was applied. Considering spatial heterogeneity and seasonal variability, different spectral mixture models (global vs. local and single vs. temporal) were evaluated. The local temporal model performed best, with an RMSE of approximately 20%, compared to 70% for the worst global single model. Errors in the local temporal model (RMSE 20%–26%) were attributed to overestimation from water-mud or water-vegetation mixtures and underestimation from shallow, turbid water. This method enables accurate large-scale modeling of surface water dynamics using low-resolution, multi-temporal images.