Streamlining Hydrological Models with Improved Parameter Learning Techniques

The Science

Accurate prediction of streamflow is vital for water resource usage and management. Given the growing complexity of modern hydrological models, a key question arises: How many streamflow observations are needed to reliably estimate model parameters? This study addresses that question by systematically evaluating the influence of observation period, the role of streamflow information content, and the gauge location in learning the parameters of a fully integrated hydrological model, called the Advanced Terrestrial Simulator. This study uses data from the Upper Neversink River Watershed in Delaware, which had streamflow records from multiple gauges over multiple years. The observed streamflow data were compared against the performance of model parameter learning through a knowledge-informed deep learning technique. The performance highly correlates with the information encoded in the observations. The success of this approach underscores the potential of using information theory to identify and narrow down the observation period for parameter learning, thus facilitating the reduction of the computational budget.

The Impact

Identifying streamflow observations for model calibration has long been a challenge in hydrology. The increasing computational demands of the watershed model further complicate this issue. This research highlights the potential of leveraging metrics that have been evaluated using information theory to select the most informative time period from a data series before undertaking extensive watershed modeling, thereby significantly reducing the computational costs. Moreover, such analysis could offer valuable insights for fieldwork, particularly in determining the optimal duration of gauge operation. Most importantly, the methodology is transferable to other types of watershed observations, such as the riverine biogeochemistry process.

Summary

Integrated hydrological modeling is gaining popularity due to its mechanistic representation of the surface and subsurface processes. However, estimating the parameters of such process-based models can be computationally expensive if careful consideration is not given to the length of streamflow observations used during model calibration. This study evaluated the influence of the calibration period, the role of streamflow information content, and the gauge location in parameter learning and calibration of a fully integrated hydrological model, called the Advanced Terrestrial Simulator. The Upper Neversink River Watershed within the Delaware River Basin, where streamflow observations were available at 11 gauges with varying record lengths, was used as the focus area. A recently proposed knowledge-informed deep learning technique was leveraged for parameter estimation. To assess the impact of the observation period and gauge location, model parameters learned on scenarios using different chunks of streamflow observations. Results show that the basin outlet discharge prediction is mostly improved when using at least four years of observations for parameter estimation. Further, the performance of the calibrated Advanced Terrestrial Simulator run correlates with the information content of the observed streamflow, suggesting that the information-theoretic metrics could be indicators for selecting the observation period for parameter estimation. Finally, we find that observations from an informative gauge can be used in learning parameters to predict the streamflow at a nearby gauge, which would potentially lower the computational expense by reducing the watershed domain used in calibration.

Contact

Xingyuan Chen, Pacific Northwest National Laboratory, Xingyuan.Chen@pnnl.gov

Funding

This work was funded by the ExaSheds project, which was supported by the Department of Energy Office of Science, Biological and Environmental Research program, Earth and Environmental Systems Sciences Division, Data Management Program.