Spatial mapping of hydrologic soil groups using machine learning in the Mediterranean region

Abstract

Soil and water conservation is one of the major tasks of hydrologic soil groups (HSGs), which play an increasing role in the hydrological process. This research attempts to produce maps of soil classification into four hydrologic groups in the Upper Oum Er-Rbia basin (Morocco), which has many environmental risks due to specific climatic conditions and anthropogenic activities. The study was based on soil physical parameters (texture, organic matter, and bulk density) identified by extensive field surveys and remote sensing data (Digital Elevation Model, Landsat-9, European Reanalysis-5, and Soil Moisture Active Passive). The soil category mapping was prepared using four machine learning (ML) algorithms, namely, random forest (RF), support vector machine (SVM), K-nearest neighbor (K-NN), and naïve Bayes (NB). From 202 hydrologic soil groups, inventory locations were constructed and then randomly divided into a ratio of 70/30% for the training and validating models. Twenty-nine causative factors and indices were prepared and subdivided into four dataset types: whole variables (A), important variables (B), noncollinear variables (C), and HSG-specific variables (D) for HSG classification. The ML models’ performance and robustness were evaluated using six accuracy metrics: Overall Accuracy (OA), Precision, Recall, F1-Score, Kappa coefficient, and Area Under the Curve (AUC) value. The experimental results revealed that the performance of models varies according to the datasets used. Based on the AUC results, all models performed comparably well, with very high accuracies. However, RF had a better classification performance (OA = 0.788), followed by NB (OA = 0.757), SVM, and K-NN (OA = 0.742), especially when using datasets related to HSGs. In addition, according to maps, HSG D had the highest rate and dominated the study area. However, saturated hydraulic conductivity is the main contributor to modeling HSG. Hence, the developed approach and maps produced from this study provide a new contextual planning tool for decision-makers to devise adaptation strategies to avoid environmental risks and conserve soil water resources.