201-5 An Adaptive Implicit-Explicit Algorithm for Simulating Multi-Decadal Land Subsidence Due to Groundwater Pumping in Southern San Joaquin Valley, California

Abstract:

Over the last century, an alarming rate of groundwater overdraft has resulted in severe land subsidence in the California’s San Joaquin Valley. Modeling is often required to manage and mitigate future subsidence, but subsidence models can be complex and computationally expensive. Most of the available physics based one-dimensional deformation models use finite difference methods, which rely either on implicit or explicit formulation. The fully explicit finite difference method is computationally efficient but can become numerically unstable depending on the spatial and temporal discretization and on hydrogeologic parameter values. Maintaining stability may therefore require extremely small-time steps, greatly increasing the total cost of multi-decadal simulations, or a coarser spatial grid, affecting the accuracy. Fully implicit methods are stable, but require solving a system of equations, which increases computational effort. To balance these tradeoffs, we developed an adaptive finite-difference model that uses the explicit formulation when the stability criterion is satisfied and automatically switches to the implicit formulation when necessary. As a result, the framework enabled us to efficiently simulate long term land subsidence in the Southern San Joaquin Valley from 1971 to 2025 while allowing flexible spatial and temporal discretization. The model is designed to simulate the compaction of clay lenses with the upper and the lower aquifers as well as the deformation of the Corcoran Clay at a daily time step. The application is demonstrated through a case study at a high subsidence location in the southern San Joaquin Valley. Model calibration combined three deformation constraints: Sentinel-1 InSAR displacement observations from June 2015 to December 2023, a cumulative deformation estimate derived from ALOS rates for June 2007 to July 2010, and long-term elevation-change records covering 1949 to 2005. Validation was done using January 2024 to December 2025 Sentinel -1 InSAR observations. At the location, the model estimated 6.1 m of cumulative subsidence from 1971 to 2025 and reproduced observed deformation well during calibration (NSE = 0.987, RMSE = 4.38 cm, NRMSE = 3.3%) and validation (NSE = 0.877, RMSE = 1.84 cm, NRMSE = 9.7%). This light-weight python-based modeling framework does not require a 3D groundwater flow model and can be applied elsewhere for improved stability in characterizing land subsidence due to groundwater withdrawals.

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.