Digital mapping of soil erodibility factors based on decision tree using geostatistical approaches in terrestrial ecosystem

Abstract

The improper and unconscious use of the soil, a production material for humans, increases the risk of desertification and degradation. This study examines the some erosion susceptibility parameters such as soil erosion factor (USLE-K), dispersion ratio, and aggregate stability by applying various algorithms in decision trees and evaluating the spatial distribution of these properties using multiple interpolation method. While determining statistically significant correlations (P < 0.01) between aggregate stability of soils and mass fractal dimension and mean diameter weight, these features were employed as the root node and inner node in decision trees. The highest estimation rate was determined categorically with the CART algorithm in the prediction phase with decision trees. Aggregate stability was determined by applying two tree depths and mass fractal dimension and silt variables with 81.1% accuracy. In predicting the K factor, the silt and clay variables were estimated at 90.6% by forming nodes. In comparison, the dispersion ratio was calculated at 77.4% accuracy with clay, EC, and mean diameter weight. Soils categorized as moderately erodible according to the USLE-K factor were estimated at 92.3% level of accuracy with decision trees. In the predictive values obtained by numerical data, the highest aggregate stability and the USLE-K factor were obtained with the lowest prediction accuracy. This change was similar in distribution maps. The lowest root means squarer error (8.074; 5.106) was obtained by the simple kriging interpolation method in the distribution maps of the aggregate stability observed and predicted values.