Hydraulic conductivity estimation in Porous Media: Insights from Neural computing

Precise estimation of hydraulic conductivity (K) in porous media is vital for advancing hydrological and subsurface flow investigations. Groundwater experts have increasingly adopted neural computing approaches to indirectly determine K in porous media, offering a more efficient alternative to conventional methods. The research focuses on developing the Feed-Forward neural network (FFNN) and Kohonen Self-organizing maps (KSOM) models to compute the K using easily measurable porous media parameters i.e., grain-size, uniformity coefficient, and porosity. The observed data were split into 70% and 30% for the development and validation phase, respectively. The developed model's performance was examined via statistical indicators, including root mean square error (RMSE), determination coefficient (R²), and mean bias error (MBE). The findings suggest that the FFNN model significantly outperforms the KSOM model in estimating the K value, with the KSOM model achieving only moderate accuracy. During the validation phase, the FFNN model shows a stronger correlation with the measured values, yielding RMSE, R², and MBE values of 0.016, 0.94, and 0.006, while the KSOM model returns values of 0.024, 0.91, and -0.004 respectively. The FFNN model's superior predictive ability makes it a reliable tool for accurate K estimation in aquifers.