Numerical Modeling of Hydrologic Exchange and Biogeochemistry in Dynamic Rivers
EMSL Project ID
60388
Abstract
The River Corridor SFA project (https://www.pnnl.gov/projects/river-corridor) uses a ModEx (integrated models and experiments) approach to study hydrologic exchange fluxes (HEFs) between river channels and subsurface environments, and the subsequent effect on biogeochemical transformations in watershed systems. High-resolution mechanistic models are applied at reach and watershed scales, based on field and laboratory observational and experimental data, and used together with machine learning (ML) methods to generate a transferable predictive capability. This proposal requests EMSL computational resources (Linux Cluster CPU and GPU nodes) to support high-resolution numerical modeling and machine learning analyses at river corridor, watershed, and basin scales.
We will apply two BER community codes (PFLOTRAN and ATS) to model groundwater-surface water interactions in the Hanford Reach of the Columbia River (PFLOTRAN) and watershed processes in the Yakima River Basin (ATS). These process-rich, high-resolution numerical models (utilizing CPU nodes) will be combined with machine learning (ML) methods (utilizing GPU nodes) for purposes of computational efficiency, model-data integration (parameter estimation), and transferability of understanding across systems.
PFLOTRAN will be applied to the 300 Area sub-reach of the Hanford Reach to evaluate the transferability of ML-based surrogate models developed previously based on high-fidelity simulations in another sub-reach (100H Area). PFLOTRAN simulations will be driven by previously simulated transient pressure fields on the riverbed and will be used to derive spatiotemporal distributions of hydrologic exchange fluxes (HEFs) and subsurface transit time distributions (TTDs) for comparison to predictions based on ML analysis of previous 100H simulation results.
ATS will be applied to selected watersheds within the Yakima River Basin (YRB; expanding on our previous experience modeling the American River watershed). We will perform ensemble simulations of watershed processes (primarily river discharge) for approximately four additional watersheds. Machine learning methods (DNN) will be trained and applied to optimize model parameters to best match field observations. Transfer learning ML methods will then be used to extrapolate those parameter sets to the remaining unsimulated watersheds in the YRB, providing the basis for well-parameterized ATS simulations of the entire basin, which would not otherwise be computationally feasible.
Project Details
Project type
Large-Scale EMSL Research
Start Date
2022-10-01
End Date
N/A
Status
Active
Released Data Link
Team
Principal Investigator
Co-Investigator(s)
Team Members