Bounding the Null Space: Interval-Based Uncertainty Quantification for Non-Identifiable Groundwater Models

Abstract

Groundwater models are routinely non-identifiable: sparse subsurface observations leave many combinations of parameters, states, and boundary conditions equally consistent with the available data. Existing uncertainty quantification (UQ) methods address this by exploring a finite set of model realizations, but incomplete exploration can systematically underestimate the true range of admissible solutions. We propose a fundamentally different approach based on Optimization-based Bound Tightening (OBBT), which represents uncertainty directly as intervals and tightens them by extremizing variables over a constraint system encoding physical laws and observations. This yields guaranteed outer bounds on all uncertain variables without sampling, side-stepping the exploration problem entirely. To apply OBBT to groundwater flow, we discretize Darcy's law using a finite-volume scheme and handle the resulting bilinear terms through McCormick relaxations. We show that these relaxations can break the sign coupling between fluxes and head gradients, permitting non-physical rotational flow and failing to provide sufficient information for effective bound tightening. We identify flow sign prescription and irrotationality constraints as effective remedies and characterize their respective strengths and limitations. We demonstrate the framework on three numerical examples - a 1D steady-state model, a 2D steady-state model across four experimental configurations, and a 2D transient model on a hexagonal grid - and discuss computational performance, scalability, and directions for future research. OBBT offers a conservative, deterministic, and physically grounded alternative to ensemble-based UQ, with natural connections to null space theory and data assimilation.