A hybrid hydrologic modelling framework with data-driven and conceptual reservoir operation schemes for reservoir impact assessment and predictions

Toward Improved Parameterizations of Reservoir Operation in Ungauged Basins: A Synergistic Framework Coupling Satellite Remote Sensing, Hydrologic Modeling, and Conceptual Operation Schemes

“…In our study, the parameter calibration data for both reservoirs consist of a total of 73 data points, yet the calibrated D22 scheme yields relatively satisfactory storage and release simulations during not only the entire period (Figures 6 and 7) but also during extreme hydrologic events (Figure 8). In comparison, data‐driven reservoir operation models are often trained against thousands of daily, continuous operation data (Dong et al., 2023; Yang et al., 2019). While it may be possible to train these models using daily values interpolated from remotely sensed operation data, its feasibility and accuracy require future investigation.…”

“…Our previous studies (Dong et al., 2022, 2023) have demonstrated that the storage anomaly based reservoir operation scheme of Equation is able to represent the multi‐purpose reservoir operation. When reservoir storage is below $${V}_{d}$$ (Equation ), the release is set as a minimum value for downstream ecological and emergency socio‐economic use, and the water stored in reservoirs is often released through pumping or emergency sluices.…”

“…Conceptual reservoir operation schemes estimate the release of reservoirs by establishing empirical functions between release, storage, inflow, and/or water demand, which include simpler ones that are designed for large‐scale applications with minimum data requirements (Hanasaki et al., 2008; Shin et al., 2019; Solander et al., 2016) and more sophisticated ones that require long‐term operation data for parameter calibration (Wu & Chen, 2012; G. Zhao et al., 2016). Data‐driven models, on the other hand, deduce reservoir releases by training machine learning regression algorithms on extensive reservoir operation data (Dong et al., 2023; Yang et al., 2019). Both models have advantages and disadvantages.…”

A Hierarchical Temporal Scale Framework for Data‐Driven Reservoir Release Modeling

“…(2022), Dong et al. (2023), and Brunner and Naveau (2023), this study omits reservoir storage simulation, which is a frequently pursued research objective in the development of reservoir operation models. It is because this study aims at investigating the significance of multi‐timescale information in data‐driven reservoir operation modeling.…”

“…The feasibility of data‐driven reservoir simulations can be further boosted through the use of hybrid strategies that combine rule‐based or conceptual operation schemes with machine learning techniques (Dong et al., 2023; Gangrade et al., 2022). By leveraging expert knowledge in the form of appropriate feature engineering (Yang et al., 2016, 2017), and by incorporating reservoir storage dynamics derived from a range of advanced sensing techniques (T. Chen et al., 2022; Y. Chen et al., 2022; Eilander et al., 2014; Sorkhabi et al., 2022; Van Den Hoek et al., 2019), it is possible to use DDMs to better reconstruct downstream flow in data‐sparse regions, using meteorological forcing and inflow generated by hydrological models.…”