Gully Erosion Modeling Using GIS-Based Data Mining Techniques in Northern Iran: A Comparison Between Boosted Regression Tree and Multivariate Adaptive Regression Spline

Abstract

Land degradation occurs in the form of soil erosion in many regions of the world. Among the different type of soil erosion, high sediment yield volume in the watersheds is allocated to gully erosion. So, the purpose of this research is to map the susceptibility of the Valasht Watershed in northern Iran (Mazandaran Province) to gully erosion. For this purpose, spatial distribution of gullies was digitized by sampling and field monitoring. Identified gullies were divided into a training (two-thirds) and validating (one-third) datasets. In the second step, eleven effective factors including topographic (elevation, aspect, slope degree, TWI, plan curvature, and profile curvature), hydrologic (distance from river and drainage density), man-made (land use, distance from roads), and lithology factors were considered for spatial modeling of gully erosion. Then, Boosted Regression Tree (BRT) and Multivariate Adaptive Regression Spline (MARS) algorithms were implemented to model gully erosion susceptibility. Finally, Receiver Operating Characteristic (ROC) used for the assessment of prepared models. Based on the findings, BRT model (AUC = 0.894) had better efficiency than MARS model) AUC = 0.841) for gully erosion modeling and located in very good class of accuracy. In addition, altitude, aspect, slope degree, and land use were selected as the most conditioning agents on the gully erosion occurrence. The results of this research can be used for the prioritization of critical areas and better decision making in the soil and water management in the Valasht Watershed. In addition, the used models are recommended for spatial modeling in other regions of the worlds.