Physics-Informed Convolutional Decoder (PICD): A novel approach for direct inversion of heterogeneous subsurface flow

Abstract

In this study, we present the development and application of the physics-informed convolutional decoder (PICD) framework for inverse modeling of heterogenous groundwater flow. PICD stands out as a direct inversion method, eliminating the need for repeated forward model simulations. The framework leverages both data-driven and physics-driven approaches by integrating monitoring data and domain knowledge (governing equation, boundary conditions, and initial conditions) into the inversion process. PICD utilizes a convolutional decoder to effectively approximate the spatial distribution of hydraulic heads, while Karhunen Loeve expansion (KLE) is employed to parameterize hydraulic conductivities. During the training process, the stochastic vector in KLE and the parameters of the convolutional decoder are adjusted simultaneously, ensuring that the predictions align with available measurements and adhere to domain-specific knowledge. The final optimized stochastic vectors correspond to the estimation of hydraulic conductivities, and the trained convolutional decoder demonstrates the ability to predict the evolution and distribution of hydraulic heads in heterogeneous fields. To validate the effectiveness of the proposed PICD framework, various scenarios of groundwater flow are examined. Results demonstrate the framework's capability to accurately estimate heterogeneous hydraulic conductivities and to deliver satisfactory predictions of hydraulic heads, even with sparse measurements. The proposed PICD framework emerges as a promising tool for efficient and informed groundwater flow inverse modeling.