Prediction of Saturated Hydraulic Conductivity Using Artificial Neural Networks and Tree Boost Methods

The hydraulic conductivity of saturated soil is a critical parameter in the design of efficient drainage systems, reflecting the soil's ability to transmit water based on pore size and structure. This study explores the use of Artificial Neural Networks (ANNs) and tree boost models to predict the field saturated soil hydraulic conductivity (Kfield) of sandy soils influenced by saline and alkaline conditions. Two ANN models were used, namely a Multilayer Perceptron (MPNN) and a General Regression Neural Network (GRNN). Soil samples were taken from El-Nubaria and Sinai in Egypt's western delta, and their physical and chemical properties were tested in the laboratory. Statistical analyses were performed to assess the relationships between these properties and Kfield. Model training and evaluation were conducted using cross-validation and five evaluation metrics. The results show that the tree boost model outperformed the ANN model, demonstrating superior accuracy and predictive capability, indicating that tree boost algorithms could be very useful for estimating Kfield in situations where data are limited or the soil is complicated.