Toward Artificial General Intelligence in Hydrogeological Modeling With an Integrated Latent Diffusion Framework

Abstract

Deep learning models have been extensively applied to various aspects of hydrogeological modeling. However, traditional approaches often rely on separate task-specific models, resulting in time-consuming selection and tuning processes. This study develops an integrated Latent Diffusion Model (LDM) framework to address four key hydrogeological modeling tasks: aquifer heterogeneity structure generation, surrogate modeling for flow and transport, and direct inversion of aquifer heterogeneity structure. Using a consistent architecture and hyperparameters, the LDM demonstrates robust multi-task processing capabilities, accurately capturing aquifer heterogeneity, enabling rapid predictions of hydraulic head and solute transport, and efficiently performing direct inversion without iterative simulations. By integrating multiple tasks within a single framework, LDM eliminates the need for task-specific models or extensive parameter optimization, offering an efficient and adaptive general solution for deep learning-based hydrogeological modeling. Its generalization across diverse objectives underscores its potential as a cornerstone for advancing Artificial General Intelligence in hydrogeological modeling.