A Machine Learning Approach for Soil Wind Erodibility Prediction

Comprehending and modeling soil wind erodibility is challenging due to the complex nature of wind-driven erosion and the limited availability of empirical observations. Therefore, it is crucial to apply robust analytical approaches capable of revealing meaningful relationships between wind erodibility values and their controlling factors. In this study, two single machine learning models-Random Forest Regression (RF) and Extremely Randomized Trees (ERT) were employed to develop reliable predictive frameworks for soil wind erodibility (E). The models were trained and tested using an extensive literature-derived dataset comprising ۱۱۸ observations of E collected from published sources. After evaluating the dataset components, ۷۰% (۸۲ samples) was used for model training and the remaining ۳۰% (۳۶ samples) for testing. The results indicate that both RF and ERT can effectively estimate E. Model performance was further evaluated using indicators related to improvement, logical assessment, credibility/reliability criteria, and a scoring procedure. Overall, the ERT model outperformed the RF model in meeting the main research objective. The proposed RF and ERT models can be used to predict soil wind erodibility for applications in infrastructure planning, agricultural management, and climate adaptation strategies, thereby supporting environmental protection, increased resilience, and long-term sustainability.

کلیدواژه ها:

Machine Learning ، Soil ، Wind Erosion ، Predictive Modeling