Neural networks for the prediction of soil water retention in the upper Cheliff watershed, Algeria

A comprehensive understanding of the water retention properties of soils is imperative for effectively managing these resources and addressing the prevailing paucity of soil data. In recent years, significant attention has been directed towards predicting soil retention properties within the soil physics community. Due to the challenges associated with direct measurement, many researchers have sought to predict this capacity using soil properties that are more readily quantifiable. In this context, the present work aims to find an approach that can contribute to the estimation of water retention and consequently improve the management of water resources, which are in deficit in the inputs. The methodology employed in this study is a modelling-based approach, utilising the method of Artificial Neural Networks (ANN) of the Multilayer Perceptron (MLP) type, which is applied to a selection of soils from the Upper Cheliff catchment area, located in the northwestern region of Algeria. The retention of these soils is predicted using a neural model, and the optimal network architecture is identified through the combination of predictive parameters. The analysis demonstrates the significant learning and prediction capacity of neural networks in relation to retention following textural stratification. However, the ANN model that showed the most remarkable efficacy was established for the clay loam soils at the potential level of -1600 kPa. This model incorporated clay, organic matter, and fine silts, which were identified as the most informative.