Soil wind erodibility estimation by tree-based frameworks

The comprehension and subsequent modeling of soil wind erodibility are hindered by the intricate nature of the erosion processes and the scarcity of empirical data. It is often essential to use rigorous methods to identify connected patterns between soil erodibility magnitudes and their determinants. This work used single machine learning (ML) techniques, namely Random Forest Regression (RF) and Extremely Randomized Trees (ERT), to establish robust predicting models for the wind erodibility (E) of soil, using a comprehensive dataset obtained from literature. The dataset, including ۱۱۸ data points on E, was obtained from academic publications. Upon assessing all components, this research opted to allocate ۷۰% (۸۲) of the dataset for training and ۳۰% (۳۶) for testing. Data suggests that both RF and ERT can reliably predict E. Each model was assessed for credibility and reliability via improvement percentage, logical analysis, assessment criteria, and scoring system. The ERT model surpasses the alternative model in achieving its primary objective. Models of RF and ERT enable the prediction of soil wind erodibility for infrastructure, agriculture, and climate adaptation. These models improve environmental protection, resilience, and sustainability.

کلیدواژه ها:

soil ، wind erodibility ، estimation ، tree algorithms ، recursive feature elimination