ABSTRACT:

This repository contains the data and code associated with the paper "Machine Learning Surrogates for Efficient Hydrologic Modeling: Insights from Stochastic Simulations of Managed Aquifer Recharge" by Dai et al. (2025) in the Journal of Hydrology (https://doi.org/10.1016/j.jhydrol.2024.132606). The study evaluates a hybrid modeling framework that combines process-based hydrologic simulations (with the integrated hydrologic code ParFlow-CLM) and machine learning (ML) surrogates to efficiently simulate managed aquifer recharge. This repository includes:

1) Sample ParFlow-CLM output for all three simulation stages
2) PyTorch dataset modules and utility functions that construct PyTorch tensors from raw ParFlow-CLM outputs
3) PyTorch modules to implement each of the eight ML architectures described in the paper (CNN3d, CNN4d, U-FNO3d, U-FNO4d, ViT3d, ViT4d, PredRNN++, and a CNN autoencoder)
4) PyTorch modules for custom layers implemented in each architecture
5) A PyTorch module that implements a normalized L2 loss function
6) Scripts to train and evaluate each surrogate architecture, including the autoencoder

Though this repository only contains sample ParFlow-CLM simulation output, complete ParFlow output files for all simulations used in the paper are available to the public in a separate repository (https://doi.org/10.25740/hj302gv2126).