PNNL

Abstract

This report summarizes the initial results and outcomes of a physics-informed, data-driven groundwater level (GWL) mapping capability for the Hanford Site. GWL mapping at Hanford is typically conducted annually and requires a significant amount of computational and expert resources, and it does not allow assessment of the informational value of specific monitoring wells. The proposed method produces spatially and temporally resolved fields consistent with sparse, irregularly sampled, and nonuniformly distributed well measurements. Implemented successfully, this capability will allow rapid mapping of groundwater levels and provide an opportunity to optimize monitoring activities (both location and sampling frequency) based on data information value evaluation. The approach integrates a diffusion-based generative model – trained on MODFLOW simulation data from the Plateau-to-River (P2R) model – with score-based data assimilation (SDA), allowing observation-conditioned mapping without retraining for each monitoring-network layout. The diffusion prior captures dominant spatiotemporal structure – including regional gradients, barrier effects, and capture zones – while SDA infuses current observations to generate full-domain GWL fields that are both physically consistent and measurement-informed. Uncertainty is represented through an ensemble of reconstructions, each reflecting independent realizations of the diffusion model conditioned on observed data. A geospatial sampling pipeline harmonizes unstructured simulation outputs and point-well data onto a common rasterized grid, providing consistent model-ready input throughout the workflow. Preliminary results show stable reconstruction accuracy [mean absolute error (MAE) ˜ 0.12–0.24 m], with spatial uncertainty concentrated in high-gradient or data-sparse areas. A remove-well demonstration quantifies marginal information value and illustrates how targeted adjustments to the monitoring network can reduce reconstruction error. A prior-only forecast establishes a baseline for error growth when no new observations are assimilated, providing guidance on monitoring frequency and scenario screening (e.g., pump-and-treat operational or hydrologic perturbations). Work planned for fiscal year 2026 will focus on fine-tuning the diffusion–SDA model hyperparameters, expanding robustness testing, developing the monitoring-network assessment workflow, completing formal quality assurance and validation, and preparing a manuscript. Together, these developments offer a scalable, physics-aware, and uncertainty quantified mapping capability to support adaptive groundwater management at Hanford.