Artificial Neural Network Estimation of Saturated Hydraulic Conductivity

Soil data serve as an important initialization parameter for hydro‐ecological and climatological modeling of water and chemical movement, heat transfer, or land‐use change. Most soil hydraulic properties are difficult to measure and therefore have to be estimated in most cases. Efficient methods for estimating soil hydraulic properties are lacking for tropical soils. This study examines and uses easy‐to‐measure soil properties together with terrain attributes in artificial neural networks (ANNs) to estimate saturated hydraulic conductivity ( K s ), one of the key soil hydraulic properties for two pilot sites in the Volta basin of Ghana. It was observed that good data distribution, range, and amounts are prerequisites for good ANN estimation and, therefore, data preprocessing is important for ANN. With adequate and sensitive data, ANN can be used to estimate K s , using soil properties such as sand, silt, and clay content, bulk density, and organic carbon. Although the inclusion of terrain parameters can improve the estimation of K s using ANN, they cannot be relied on as the sole input parameters as they yield poor results for the scale considered in this study. The source of training data was found to significantly influence the topsoil K s , but the subsoil was not sensitive to training data source.