From Sediment to Storage: Tracking the World’s Reservoirs

Abstract

Reservoirs formed by dams underpin global water, food, and energy security, yet their long-term viability is increasingly threatened by sedimentation — the gradual infilling of storage capacity by deposited sediment. Despite its importance, the magnitude, spatial variability, and temporal evolution of sediment-driven storage loss remain poorly quantified at global scale, constrained by sparse observations and the absence of scalable monitoring frameworks. This dissertation asks a fundamental question: how much reservoir storage capacity remains worldwide, and how rapidly is it being lost?To answer this, the research develops an integrated framework combining satellite remote sensing, hydrological modeling, machine learning, and curated field observations. Reservoir monitoring is first advanced through RAT 3.0 (Reservoir Assessment Tool), a scalable open-source system that estimates key reservoir dynamics — inflow, outflow, evaporation, storage change, and surface area — from satellite observations alone. Delivered as a Python library with full documentation and continuous integration, RAT 3.0 is designed for broad accessibility and reproducibility. The observational foundation is strengthened by GRILSS (Global Reservoir Inventory of Lost Storage by Sedimentation), a new open-source dataset compiling sedimentation rates and storage loss for over 1,000 reservoirs across major river basins worldwide. Building on GRILSS, a machine learning model — RECLAIM (Reservoir Estimation of Capacity Loss using AI-based Methods) — estimates sedimentation rates by integrating multi-decadal Earth observations with hydroclimatic and geomorphic predictors, achieving an R² of 0.86, RMSE of 4.44 MCM/year, and MAE of 1.65 MCM/year. Critically, RECLAIM enables global-scale estimation without reliance on costly repeated bathymetric surveys. These components converge in GLoRS (Global Long-term Reservoir Sedimentation), the first multi-decadal, observationally constrained assessment of global reservoir sedimentation spanning 1984–2023. Major reservoirs are losing storage at approximately 0.42% per year on average, with nearly one-fifth of capacity already lost across typical large reservoirs. Sedimentation patterns are spatially organized and temporally dynamic, driven by hydroclimatic controls that intensify deposition during extreme events. Together, these tools and datasets transform reservoir sedimentation from a localized, data-scarce problem into a globally observable and quantifiable phenomenon. This work provides a foundation for identifying the most vulnerable reservoirs, designing targeted sediment management strategies, and supporting resilient water infrastructure planning in the face of a changing climate.