Surrogate modeling for Critical Zone Simulations in the Venetian High plain

Simulating critical zone processes in three-dimensional hydrological models like CATHY (Catchment Hydrology) can be computationally complex, especially when applied to large-scale systems such as the Venetian Plains in Northern Italy. To address this, we develop a surrogate modeling framework that leverages Proper Orthogonal Decomposition (POD) and Gaussian Process Regression (GPR) to approximate variably saturated moisture dynamics under simplified boundary conditions. The model setup imposes no-flow conditions on lateral boundaries and includes only drainage at the base to reduce computational and structural complexity. POD is used to extract dominant spatial-temporal modes from high-fidelity simulations, while GPR maps input parameters to the modal coefficients for efficient emulation. This reduced-order approach enables rapid, reliable simulation of subsurface dynamics, facilitating scenario analysis and uncertainty quantification in critical zone modeling with limited data and computational resources.