Modeling the probability of gully occurrence and investigating the spatial effects of its drivers using the boosted tree regression

 Gully erosion is a subtype of water erosion which makes agricultural lands impracticable during its development. Given the geographical and environmental conditions, various factors contribute to the development and expansion of gully erosion. In this study, due to the extensive expansion of gully erosion in Jafarabad Moghan, and damaging the agricultural and rangelands, the probability of gully occurrence and the spatial effects of its drivers has been investigated.

 In this study, using a boosted regression tree model, the effect of the following factors on the gully occurrence were investigated; slope, aspect, plan curvature, altitude, clay content of horizon A, clay content of horizon B, sand content of horizon A, sand content of horizon B, surface organic matter, distance from roads and distance from rivers. Then, the susceptibility map of the gully erosion was created.

 The results showed that the distance from river, the percentage of sand of horizon A, the percentage of clay of horizon A, and the surface organic matter with 16.3%, 13.1%, 11.4% and 10.7% respectively, were the most important influential factors of gully occurrence. On the other hand, aspect (4.5%) and elevation (5.5%) were the least important ones which could be due to the lack of significant elevation shift in the region. Based on the susceptibility map, 10/63% of the study area was classified as very sensitive to the gully erosion. The AUC value for the boosted tree regression model was 0.81 which indicates a good model performance in the prediction of areas sensitive to gully erosion.

 The results of this study shows the critical influence of surface soil properties on the gully erosion. Considering the fact that the greatest effect on the probability of gully erosion is related to distance from the river and surface soil properties, it is possible to manage the lands susceptible to gully erosion by effective management of streams and existing gullies, and also by training the farmers and increasing their knowledge regarding gully erosion, land management and sustainable agriculture. The results indicate the suitability of the boosted regression trees for the similar studies.