Interpretable Physics-Informed Neural Networks Groundwater Level Simulation Model: A Case Study of Linyi, Shandong Province, China

Entitled “Interpretable Physics-Informed Neural Networks Groundwater Level Simulation Model: A Case Study of Linyi, Shandong Province, China" for possible publication in Water Resources Research. metadata: The "metadata" folder contains the raw data used in the paper, including groundwater monitoring well data, precipitation data, hydrological parameter data, and more. In this study, we utilized data from 11 monitoring wells. For each monitoring well, we used its location information, monthly-scale groundwater level data, corresponding monthly-scale precipitation data, storage coefficient, hydraulic conductivity, and time information. simulated_dataset: The "simulated_dataset" folder contains the normalized training dataset used as input for both the MLP and PINNs. codes: The "codes" folder contains all the code files, including the PINNs code, MLP code, model accuracy analysis, SHAP importance analysis, and the code for result output. All the codes were executed in Python. output: The "output" folder contains the results of the study, including the trained PINNs model, MLP model, PINNs inversion model, as well as the simulated regional groundwater levels, temporal groundwater levels, model importance information, and parameter inversion results produced by the PINNs model.