Reservoir Storage Retrieval From Multi-Source Satellite Remote Sensing and Artificial Intelligence: A Comprehensive Review

Reservoirs are critical infrastructure for global water, food, and energy security, yet their storage dynamics remain poorly characterized across most of the world. Accurate and timely knowledge of reservoir storage is essential for sustainable water management, climate adaptation, and transboundary water diplomacy. Reservoir storage is a primary state variable for assessing water security and calibrating hydrological models, yet transforming satellite observations into reliable storage estimates remains methodologically fragmented. The integration of multi-source satellite remote sensing with artificial intelligence (AI) is widely seen as a paradigm shift. The core advance, as revealed by this review, lies not in algorithmic improvement alone, but in the systematic re-engineering of the inversion chain from fragmented sensor inputs to integrated storage estimates. AI functions as an integrative layer: it stitches together altimetry, optical, and SAR data from disparate sources, learns reservoir-specific water level-area-storage relationships without requiring bathymetric priors, and embeds physical constraints into data-driven models. Applications at global, transboundary, and national scales have demonstrated both the potential and the persistent blind spots of current approaches. Small reservoirs remain largely unmonitored, most models operate with limited explainability, and uncertainty quantification is routinely marginalized. Looking forward, the deepest integration of next-generation missions like SWOT with physics-informed AI represents the most credible path toward a globally scalable and genuinely intelligent reservoir monitoring system.