Mapping and Preparing a Susceptibility Map of Gully Erosion Using the MARS Model

Abstract

Preparing and mapping gully erosion (GE) is a basic instrumentation to land use projecting and reducing destruction of the land. The purpose of the current investigation was to assess gully erosion spatial modeling using multivariate adaptive regression spline (MARS) model in Maharlou watershed, Fars Province, Iran. The current study is consisted from two important parts including (1) recognizing dependent and variables, e.g., gully erosion inventory map (GEIM) and gully effective agents, and (2) running a famous machine learning algorithm named the MARS in order to gully erosion mapping. Gully erosion inventory map is randomly separated into two categories: training and validation datasets. Then, nine causative factors including land use, distance from rivers, clay percent, geology, pH, NDVI, drainage density, distance from roads, and slope direction are recognized, and their maps are classified in the ArcGIS. Also, the GESM was created using the MARS model in the R statistical environment. The outcomes of the MARS technique of the 30% of the unused gully points used in the modeling procedure based on the ROC curve. Results demonstrated that the ultimate gully erosion map had a top precision with AUC values 96.3% for accuracy data set.