Soil information on a regional scale: Two machine learning based approaches for predicting saturated hydraulic conductivity

Derivation of physically based soil hydraulic parameters in New Zealand by combining soil physics and hydropedology

“…This assures that K s is correlated to the pore size distribution described by θ(ψ). The advantage of using physical approaches to derive K s is that it does not require a large observed K s database compared to, for example, using a random forest algorithm as used in the European Pedotransfer functions (T oth et al, 2015;Szab o et al, 2021) or by using eXtreme Gradient Boosting in Austria (Zeitfogel et al, 2023). The limited number of samples in New Zealand as well as the very large spatial variability of K s due to the very young, alluvial and volcanic soils (Hewitt et al, 2021), makes it very problematic to rely on statistical approaches.…”

Predicting Hydraulic Conductivity using Machine Learning techniques

“…In recent years, deep learning methods are used widely in terms of water resources management, especially groundwater. Some studies have applied different machine learning techniques (e.g., Random forest (RF), Quantile random forest (QRF), Cubist (Cu), and Decision tree regression (DTr), Multilayer Perceptron (MLP-GA) and Support Vector Machine (SVM-GA), Radial basis function neural networks (RBFNN), Multi layer perceptron neural networks (MLPNN), Adaptive neurofuzzy inference system (ANFIS) and Multiple-linear regression (MLR), Convolutional neural network (CNN), Extreme Gradient Boosting (XGBoost) and feed-forward neural network (FNN), the k-nearest neighbors (KNNs), support vector regression (SVR), and boosted regression tree (BRT) in different regions, such as the border between Belgium, and the Netherlands, the Sheonath basin (Chhattisgarh, India), Khuzestan province (Iran), Austria, or global scale (Gupta et al, 2021;Rezaei et al, 2023;Singh et al, 2022;Arshad et al, 2013;Zhou et al, 2020;Zeitfogel et al, 2023;Araya and Ghezzehei, 2019). Gupta et al (2021) used the RF to generate Ks mapping and compared results with availability of remotely sensed surrogate information, showing the Ks could be modeled without bias using Covariate-based GeoTransfer Functions.…”

“…Zhou et al (2020) used deep learning techniques to build the functional mapping between the hydraulic conductivity field and the longitudinal macro-dispersivity, and produces better macro-dispersivity estimation for the conductivity field with smaller variance. Zeitfogel et al (2023) used machine learning models (i.e., XGBoost and FNN) to predict Ks, and help to reduce current gaps in soil data availability for Austria. Araya and Ghezzehei (2019) used machine learning and a large database of over 18,000 soils to develop new pedotransfer functions (PTFs) for estimating Ks, indicating that PTF models based on boosted regression tree algorithm obtained the best result.…”