Gully erosion susceptibility assessment using machine learning techniques, Shazand watershed, Iran

Abstract

Soil, as a valuable natural resource, provides a large number of services and plays an important role in the environment and world economy. Soil degradation and erosion reduce the quality and quantity the soil and are important natural and anthropogenic processes that affect many countries. Water erosion is the most common type of soil degradation in the world and Asia has about fifty percent of the total water erosion area of the world. Gullies are a typical erosion type and gully formation is an important process of soil erosion and degradation in semi-arid and arid areas, especially areas impacted by human activities and land uses. Because of arid and semi-arid climate, piping and gully erosion is an active phenomenon in the agricultural lands, bare land and rangeland areas of the Shazand watershed, Markazi Province, central Iran. The goal of this research was to identify the priority conditioning factors of gully erosion, map the susceptibility of the Shazand watershed to gully erosion and compare some of the applied machine learning techniques based on their accuracy. Prioritization of conditioning factors using a random forest (RF) algorithm demonstrated that distance from the roads, altitude, and rainfall have the greatest impact on gully occurrence in the Shazand watershed. The RF, boosted regression tree (BRT), functional discriminant analysis (FDA), generalized linear model (GLM), and mixture discriminant analysis (MDA) algorithms were applied to create gully erosion susceptibility maps in the study area. The receiver operating characteristic curve (ROC) and area under the curve (AUC) performance metrics were used to validate susceptibility maps. The AUC values of 0.850, 0.831, 0.760, 0.751, and 0.758 were achieved for the RF, BRT, FDA, GLM, and MDA algorithms, respectively. Due to the negative and destructive effects of gully erosion, its management and control is a critical component in the management of natural resources and land uses. The susceptibility maps of gully erosion prepared in this study are a substantial information resource for decision makers, planners, and engineers concerned with human impacts on natural resources and land uses. The areas identified with high and very high erosion susceptibility in the Shazand watershed need more care to mitigate the consequences of gully erosion and soil degradation.