The value of observed reservoir storage anomalies for improving the simulation of reservoir dynamics in large-scale hydrological models

<p>Human-managed reservoirs alter water flows and storage, impacting the hydrological cycle. Modeling reservoir outflow and storage, which affect water availability for humans and freshwater ecosystems, is challenging because they depend on human decisions. In addition, access to data on reservoir inflows, outflows, storage, and operational rules is very limited. Consequently, large-scale hydrological models either exclude reservoir operations or use calibration-free algorithms to model reservoir dynamics. Nowadays, estimates of reservoir storage anomalies based on remote sensing are a potential resource for calibrating the release algorithms for many reservoirs worldwide. However, the impact of calibration against the storage anomaly on simulated reservoir outflow and absolute storage is unclear. In this study, we address this by using in situ outflow and storage data from 100 reservoirs in the USA (ResOpsUS dataset) to calibrate three reservoir operation algorithms: the well-established Hanasaki algorithm (CH) and two new storage-based algorithms, the Scaling algorithm (SA) and the Weighting algorithm (WA). These algorithms were implemented in the global hydrological model WaterGAP, with their parameters estimated individually for each reservoir and four alternative calibration targets: monthly time series of (1) the storage anomaly, (2) estimated storage (calculated based on the storage anomaly and GRanD reservoir capacity), (3) storage, and (4) outflow. The first two variables can be obtained from freely available global datasets, while the latter two variables are not publicly accessible for most reservoirs. We found that calibrating against outflow did not result in skillful storage simulations for most of the 100 reservoirs and only slightly improved outflow simulations compared to calibration against the three storage-related targets. Compared to the non-calibrated Hanasaki algorithm (DH), calibrating against both the storage anomaly and estimated storage improved the storage simulation, whereas the outflow simulation was only slightly improved. Calibration against the storage anomaly yielded skillful storage simulations for 64 (39), 68 (45), and 66 (45) reservoirs in the case of CH, SA, and WA, respectively, during the calibration (validation) period, compared to just 16 (15) for DH. Using estimated storage instead of the storage anomaly does not offer any added benefit, primarily due to inconsistencies in the observed maximum water storage and storage capacity data from GRanD. The default parameters of the Hanasaki algorithm rarely matched the calibrated parameters, highlighting the importance of calibration. Using observed inflow rather than simulated inflow has a greater impact on improving the outflow simulation than calibration, whereas the opposite is true for the storage simulation. Overall, the performance of the SA and WA algorithms is nearly equal, and both outperform the CH and DH algorithms. Moreover, incorporating downstream water demand into the reservoir algorithms does not necessarily improve modeling performance due to the high uncertainty in demand estimation. Therefore, to improve the modeling of reservoir storage and outflow in large-scale hydrological models, we recommend calibrating either the SA or the WA reservoir algorithm individually for each reservoir against the remote-sensing-based storage anomaly, unless in situ storage data are available, and improving the reservoir inflow simulation.</p>