Potential Erosion Mapping Using Machine Learning Methods (Case Study: Rud-e-Faryab Basin, Bushehr Province, Iran)

Abstract

Abstract Purpose. The requirement of soil erosion management is to provide appropriate solutions which can be obtained by recognizing the state of soil erosion. The purpose of the study is to model the potential erosion using 10 environmental variables and 10 models in BIOMOD-2 package in R software and to evaluate the stability of the model in order to be aware of the sensitivity of erosion in Rud-e-Faryab basin in Bushehr province. Methods. In addition, the location of five dominant erosions of the basin was recorded using GPS. These events were classified into two groups of training and validation with a ratio of 70 to 30. In order to evaluate the stability, the BIOMOD-2 package model was repeated 10 times. The efficiency of the model was evaluated using ROC, KAPPA and TSS. Results. According to stability results, the best models are determined as GLM model with the accuracy of above 70% in low-rated waterway erosion, RF model with the accuracy of above 80% in medium-rated waterway erosion, ANN and SRE models with 100% accuracy in medium-rated rill erosion, MARS model with accuracy above 60% in high-rated rill erosion and GLM model with accuracy above 80% in medium-rated stream bank erosion according to all three validation parameters. Conclusion. These results indicate high agreement with studies which have used the vector machine models as an effective tool in preparing the potential erosion map of watersheds. In general, it can be said that vector machine methods are a helpful new approach for land use planning and erosion risk management.