Predicting saturated hydraulic conductivity by pedo-transfer function and spatial methods in calcareous soils

Modeling soil saturated hydraulic conductivity (Ks) plays an important role to design and manage irrigation methods. Developing and evaluating accurate pedo-transfer functions (PTFs) for predicting difficult-to-measure soil properties such as Ks is an essential factor. This study aimed to develop aspatial models such as MLR (Multiple linear regression)-, ANN (Artificial neural network)-, and GMDH (Group method of data handling)-based PTFs and spatial methods such as Ordinary Kriging (OK), MLR-Kriging (MLR-K), ANN- Kriging (ANN-K), and GMDH-Kriging (GMDH-K) in GIS (Geographic information system) for predicting Ks in semi-arid soils. Soil infiltration was measured with a double-ring approach at 124 points with three replications at the field scale. After that, soil Ks was obtained by fitting the Green and Ampt infiltration model on soil infiltration data. In addition, for each point, easily measurable soil attributes such as texture, calcium carbonate equivalent, bulk density, soil moisture, saturated percent, organic matter, and gravel contents were measured. The results showed that the ANN-based PTFs yielded the better results with the highest E (0.605) and the lowest RMSE (0.055) than the MLR-based PTFs (E = 0.341 and RMSE = 0.068) and GMDH-based PTFs (E = 0396 and RMSE = 0.065) for predicting Ks parameters. In addition, the hybrid spatial methods in this study, which indude GMDH-K, MLR-K, and ANN-K provided more reliable estimation than the OK method. In overall, the best estimation of spatial method was ANN-K method, which had the highest E value (0.728) and the smallest RMSE value (0.044) for predicting the Ks parameters. (C) 2021 Elsevier B.V. All rights reserved.