Global patterns of reservoir normal water storage capacities estimated by statistical and machine learning models

Reservoir water storages are increasingly recognized as important tools for adapting to water-related hazards under climate change, particularly through their normal water storage (NWS). However, a critical data gap on regarding global reservoir NWS hinders our understanding of their spatial patterns and impacts on water management. This study aims to bridge this data gap by estimating the NWS for 7,250 global reservoirs using an ensemble approach that includes linear regression, machine learning, and satellite-based methods. We train and evaluate these models based on a representative sample of 801 reservoirs with reported NWS data and multiple static attributes. Among the evaluated methods, the Gradient Boosted Decision Trees model utilizing various reservoir features performed best (achieving a coefficient of determination of 0.95 and a mean absolute percentage error of 12%), and was subsequently selected to estimate the NWS for unrecorded reservoirs. Our analysis of the global reservoir NWS ratio (NWSR) at individual, basin, and national scales reveals significant spatial variations, with a global mean NWSR of 72%. The distribution of NWSRs varies considerably across different reservoir groups categorized by physical size, primary purpose, and climatic region. At the national and basin scales, the USA, Canada, China, Russia, and Brazil, along with the Yenisey, Nile, Mississippi, Parana, and Yangtze basins, account for substantial portions of the global NWS capacity. The developed dataset and the analyzed spatial patterns provide vital empirical inputs for water resource management, large-scale hydrological modeling, and climate change adaptation, ultimately supporting effective planning for sustainable freshwater supplies.