Mapping Reservoir Water Surface Area in the Contiguous United States Using the High-Temporal Harmonized Landsat and Sentinel (HLS) Data at a Sub-Weekly Time Scale

Traditional remote sensing techniques are limited in providing the necessary spatial and temporal resolution to capture the reservoir dynamics accurately. In this study, we introduce a novel algorithm to generate sub-weekly reservoir surface area time series using the Harmonized Landsat and Sentinel-2 data set across the Continental United States. Our approach addresses common challenges (e.g., cloud contaminations) by integrating a Random Forest classification model with a refined image enhancement algorithm. Validation results against in situ data from 240 reservoirs indicate a high coefficient of determination (R² = 0.98) and relatively low bias (<10%), therefore demonstrating its robustness across reservoirs of varying sizes and climatic conditions. This method not only captures sub-weekly surface area changes, but also provides richer temporal information compared to existing monthly data sets. This enhanced temporal resolution is useful for applications such as reservoir management, hydropower generation, and our overall understanding the transient nature of reservoir dynamics.