Modelling hydraulic conductivity of porous media using machine learning techniques

ABSTRACT Investigating hydraulic conductivity (K) is crucial for aquifer studies and groundwater flow modelling. The main objectives of the current study are to investigate the effectiveness of artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), Gaussian process regression (GPR), and random forest (RF) algorithms in estimating K using data from 270 borehole soil samples, collected along the Beas riverbank in Kangra district, Himachal Pradesh, India. For the K estimation, the study utilizes the grain size parameters, i.e., d10, d50, coefficient of uniformity (Cu), and porosity (n) as input parameters. The performance evaluation of the developed models was assessed using the statistical parameters. While the performance of each model is quite satisfactory, the present study establishes the efficacy of the GPR model during validation having a determination coefficient of 0.985. The root mean square errors for ANN, ANFIS, GPR, and RF were 0.019, 0.017, 0.00853, and 0.019, respectively. The techniques used in the study offer precise K-prediction abilities that facilitate groundwater management and contaminant transport analysis. The GPR model in the study outperforms other models in estimating the K of soil samples and serves as an efficient tool for managing soil water and solute transport.