Predicting Hydraulic Conductivity using Machine Learning techniques

Abstract The hydraulic conductivity of water-saturated soils (Ks) is a critical parameter influencing water infiltration, runoff, and drainage. However, collecting this value is always costly and time-consuming in large scale measurements. This study employs state-of-the-art machine learning (i.e., the Multi-Layer Perceptron (MLP)) to predict K values in different spatial distribution by utilizing groundwater head values generated from the VSAFT2 groundwater flow model. An ablation study is conducted to assess the contributions of various inputs, including groundwater head values, time step, and well information in the groundwater flow model. The results show a good fit between observation and prediction, and R2 values are above 0.8 for training and testing. The uncertainty of the data is evaluated with different scenarios to find the reasonable parameters. The use of advanced machine learning techniques, demonstrates their efficiency in predicting Ks, would help in efficient spatial-temporal measurement of hydraulic conductivity, indicating valuable insights of using machine learning in the field of groundwater prediction.