Applied AI and Machine Learning Surrogate Models

EKI uses ML / AI–based surrogate models to make integrated groundwater and subsidence modeling faster, more scalable, and more useful for real-world decision making. While fully integrated hydrologic models provide critical physical insight, their computational demands often limit how efficiently scenarios can be evaluated. EKI’s ML / AI surrogates emulate the most computationally intensive components of these models, enabling rapid testing of management alternatives, near-real-time “what-if” analyses, and scalable evaluation across pumping strategies, climate conditions, and regulatory constraints—while preserving consistency with the underlying physics.

Sample project 1 –

EKI supported a confidential client with a large, fully integrated GSFLOW–PRMS groundwater model that provided strong physical insight but required extremely long run times, limiting its practical use for scenario evaluation and decision support. To address this challenge, EKI developed an ML / AI–based surrogate modeling framework trained on ensembles of GSFLOW–PRMS simulations. The surrogate models replicated key system responses with high fidelity while reducing computational requirements by orders of magnitude. This approach enabled rapid evaluation of pumping and climate scenarios, scalable ensemble analysis, and efficient exploration of management alternatives that were previously impractical. In addition, the surrogate framework was used to identify dominant controls on system behavior, helping the client better understand the relative influence of climate drivers, recharge processes, and groundwater stresses. The result was a modeling workflow that preserved the value of the original integrated model while making it faster, more transparent, and more actionable for planning and decision making.

Sample project 2  –

EKI supported a confidential client in evaluating land subsidence risk and drivers across a complex groundwater system where observed deformation reflected a combination of historical groundwater conditions, ongoing pumping, regional influences, and management actions. To support prediction and attribution, EKI applied ML / AI–based modeling approaches that integrated groundwater levels, pumping records, geologic and hydrostratigraphic information, and InSAR subsidence data. These models were used to predict subsidence responses under alternative management scenarios and to quantify the relative contributions of different drivers, helping distinguish between legacy effects and contemporaneous stresses. The analysis provided clear, data-driven insight into subsidence mechanisms, supported defensible interpretation of observed trends, and enabled more informed evaluation of management strategies, regulatory compliance, and long-term risk mitigation.

Outcome

EKI’s use of ML/AI-based surrogate models to integrate groundwater and subsidence modeling help clients extract significantly more value from existing model investments and supports timely, transparent, and defensible planning and regulatory decisions.