Assessing the stability of maximum entropy prediction for rill erosion modelling

Soil erosion management requires providing appropriate solutions that can be achieved with knowing soil erosion situation. The aim of this study, modeling rill erosion potentially by using maximum entropy (MaxEnt) and investigation of its robustness to knowing about rill erosion susceptibility in the Golgol watershed, Ilam province. To this purpose, different geo-environmental factors were selected to be employed in the modeling process. In addition, 157 rill erosion events were recorded by a global positioning system (GPS). These events were then classified into two classes of training and validation with a ratio of 70:30. To evaluate model robustness, these classifications were repeated three times, and therefore, three sample datasets (D1, D2, and D3), were prepared. The area under receiver operating characteristics (AUC) curve was used for evaluating the performance of the model. Regarding the robustness results, all of the datasets obtained good AUC values and all of them were robust for both the goodness-of-fit (RAUC =1.3) and prediction performance (RAUC =3.1). In other words, the results demonstrated that the model remained quite stable when the calibration and validation data were changed. In addition, we found that the MaxEnt model is capable to produce rill erosion susceptibility map. Furthermore, based on the sensitivity analysis, it found that the most important components in rill erosion susceptibility modeling are lithology and distance from stream. The adopted methodology can be useful as an efficient approach for land use planning and erosion risk management.