We integrate our data-driven reservoir operation model with a linear river routing scheme. This approach enables us to simulate long-term hydrological drought dynamics and assess water supply shortages at the river basin scale. Our simulations rely on runoff data generated by the Variable Infiltration Capacity (VIC) hydrological model. We choose a selection of intensively regulated river basins across the Contiguous United States (CONUS) for our analysis. Our objectives are multifold: (1) to examine the spatial and temporal patterns of water supply droughts, (2) to explore the interaction between hydrological droughts and water supply shortages, particularly under the influence of reservoir operations, (3) to evaluate the risk of water supply droughts under various water management strategies, including both the status-quo as represented by our data-driven model and forecast-informed operations, and (4) to develop a comprehensive drought index that considers both water availability for supply and demand, incorporating reservoir storage as a potential buffer against drought.

Seasonal soil freezing and thawing processes significantly influence runoff generation dynamics during cold periods, affecting various hydrological and agricultural systems, including flood generation, soil erosion, and plant health. Representing frozen soil conditions in land surface or hydrological models is therefore crucial. While fully distributed models implement the process by solving energy-mass balance equations to obtain soil temperature profiles, parsimonious models using “snow tanks” or frozen ground states can provide suitable modeling solutions with reduced computational demands. However, even these parsimonious approaches to representing frozen ground typically require some additional complexity through additional inputs or surface energy balance calculations. This study evaluates the applicability of a simplified soil temperature prediction model that determines frozen/unfrozen ground states using only air temperature and snow cover data, reducing model complexity. We first validate the model performance using AmeriFlux network in-situ measurements across the United States and Canada. Furthermore, we provide a comprehensive assessment at the global scale with ERA5-LAND reanalysis data (1980-2020). The model demonstrates robust performance globally, achieving an average true frozen rate of 0.90 and false frozen rate of 0.06. We also investigate the model performance by month, and, while monthly analyses show drops in model performance for certain months, these lower scores are primarily due to the limited number of freeze-thaw events during these periods, which makes the model appear less accurate than it actually is. In terms of spatial performance, the model shows reduced accuracy in mountainous regions, including the Tibetan Plateau, Rocky Mountains, and Andes, suggesting the need for region-specific parameter calibration in orographic settings. Nevertheless, this parsimonious soil temperature model demonstrates significant potential as a computationally efficient solution for incorporating frozen ground effects in distributed hydrological models with simple conceptual runoff generation schemes. 

This work has been submitted to Journal of Hydrology.

Accurately representing daily reservoir operations in large-scale hydrological and water resource modeling remains challenging due to both the complex and unclear nature of real-world operations and very limited availability of operation records for many reservoirs worldwide. To address this gap, this study introduces MODROM (MOdular Data-driven Reservoir Operation Model), a parsimonious reservoir parameterization scheme that conceptualizes reservoir operations through simple operation modules and their seasonal transitions. These operation modules are designed to be simple and parsimonious for easier generalizing from data-rich to data-scarce reservoirs. MODROM is calibrated using high-quality long-term operation records from more than 400 data-rich reservoirs across the contiguous United States (CONUS), and a Random Forest model is developed to provide calibrated parameters for data-scarce reservoirs based on a suite of static reservoir characteristics. Results demonstrate MODROM’s strong and robust performance when calibrating using all available data for each reservoir, though the performance generally declines for reservoirs with larger regulation capacity. The generalization performance is strong under favorable sampling conditions but is affected by sampling uncertainty due to the limited reservoir dataset. Benchmarking against existing models shows that MODROM offers distinct advantages in generalizing parameters to data-scarce reservoirs using readily available static reservoir characteristics, with the potential for global-scale application.

This work has been submitted to Journal of Advances in Modeling Earth Systems.

This study investigates whether improvements could be achieved for the United States National Water Model (NWM) by using a data-driven reservoir operation simulation algorithm, that is, the generic data-driven reservoir operation model (GDROM), to replace the existing reservoir representation in NWM for reservoir routing in streamflow simulation. The evaluations of NWM versus NWM + GDROM are conducted using 41-year NWM retrospective simulation products during 1979 to 2020. The results show that NWM + GDROM exhibits significant improvement to NWM in both the entire Contiguous U.S. (CONUS) and various regions. It is found that the accuracy of reservoir inflow and storage values affects the NWM + GDROM improvement. When driven by observed storage (which is usually more accurate than modeled storage) or more accurately simulated inflow, NWM + GDROM produces more significant improvements in reservoir outflow simulations in all the study CONUS regions. Especially, since high flows are more accurately simulated by NWM than low flows, NWM + GDROM has larger improvement for the simulation of high flows than low flows.

This work has been published on Journal of the American Water Resources Association. (link)

We assess the overall watershed system representation via fully coupling a generic reservoir operation model with a conceptual rainfall-runoff model. The performance of the coupled model is evaluated comprehensively by examining watershed outflow simulations, model parameter values, and a key internal flux of the watershed model (here reservoir inflow). Five published generic reservoir operation models are coupled with a watershed rainfall-runoff model, and results are compared across the coupled models and one additional model called ResIgnore that ignores reservoir operation. Traditional loosely coupled watershed hydrologic models (where calibrated inflow is routed through reservoir operation models) are used as baselines to examine the differences in simulation performance and parameterization obtained from the fully coupled models. We find that fully coupling the Generic Data-Driven Reservoir Operation Model (GDROM) and the Dynamically Zoned Target Release (DZTR) reservoir operation models with the rainfall-runoff model obtains robust simulations of watershed outflow with realistic parameterization, suggesting that they can be reliably integrated into large-scale hydrological models for simulating streamflow in heavily dammed watersheds. Our results also show that compared to ResIgnore, the fully coupled watershed models more accurately simulate the entire distribution of watershed outflow, obtain more realistic values of model parameters, and simulate reservoir inflow with higher accuracy. Finally, we note that the prediction intervals of watershed outflow obtained from the GDROM- and DZTR-based fully coupled models consistently envelop observed watershed outflow across the study watersheds, indicating that GDROM and DZTR can be suitable reservoir components of large-scale hydrology models.

This work has been published on Water Resources Research. (link)

Reservoir operations are influenced by hydroclimatic variability, reservoir characteristics (i.e., size and purpose), policy regulation, as well as operators' experiences and justification. Data-driven reservoir operation models based on long-term historical records shed light on understanding reservoir operation rules and patterns. This study applies generic data-driven reservoir operation models (GDROMs) developed for 452 data-rich reservoirs with diversified operation purposes across the CONUS to explore typical operation rules and patterns. We find that the operating policies of any of these reservoirs can be modeled with a small number (1–8) of typical operation modules. The derived modules applied to different conditions of the 452 reservoirs can be categorized into five basic types, that is, constant release, inflow-driven piecewise constant release, inflow-driven linear release, storage-driven piecewise constant release, and storage-driven nonlinear (or piecewise linear) release. Additionally, a joint-driven release module, constructed from these five basic types, has been identified. The analysis further shows the module application transition patterns featuring operation dynamics for reservoirs of different operation purposes, sizes, and locations. The typical module types can be used as “Lego” bricks to build operation models, especially for data-scarce reservoirs. These module types and their application and transition conditions can inform Standard Operation Policy (SOP) and Hedging Policy (HP) with specific inflow, storage, and/or both conditions.

This work has been published on Water Resources Research. (link)

The GDROM-derived inventory of empirical operation rules is on HydroShare. 

We presents an interpretable machine learning-based reservoir operation model. The hidden Markov model and decision tree are coupled to derive representative operation modules for a reservoir; a decision tree model is then used to identify the application and transition conditions for the operation mdoules. These two procedures result in our reservoir operation model that is featured by 1) using a few input variables (inflow, storage, DOY, and PDSI); 2) inheriting merits of decision trees but dramatically reducing model complexity; 3) adopting a consistent and transparent structure (i.e., better interpretability than other machine learning models); and 4) showing a better performance than traditional decision tree models, especially in storage simulation. GDROM is developed for 450+ reservoirs with diverse operation purposes in different regions of the Contiguous United States (CONUS), and the testing procedure shows comparable accuracy in release simulation to other machine learning models; among these reservoirs, 15 are selected for detailed analysis with diverse operational purposes and regulation capacities, from different USGS Water Regions. GDROM presents a ready-to-use reservoir operation model that can be incorporated into a watershed hydrological simulation model.

This work has been published on Advances in Water Resources (link).