Spatial modelling of gully erosion in the Ardib River Watershed using three statistical-based techniques

Abstract

Gully erosion threatens land sustainability. Gullies trigger considerable erosion, damaging agricultural land, infrastructure and urban areas; thus, predicting and modelling gully susceptibility is of utmost concern. In particular, such a model is urgently required in semiarid areas where soil loss from gullies is high. Three predictive models are evaluated to assess gully erosion susceptibility mapping (GESM) in Semnan Province, Iran. The index of entropy (IOE), frequency ratio (FR) and certainty factor (CF) models are combined with remote sensing and geographic information system techniques to predict gully erosion. The collation of data from geographic resources identified 287 gullies in the study area. These areas were then randomly divided into 2 groups for calibration (70% or 201 gullies) and validation (30% or 86 gullies). Pairwise linear dependency amongst geoenvironmental factors was also assessed. A total of 16 factors were screened for modelling. Four performance metrics, namely, true skill statistic (TSS), area under the receiver operating characteristic (AUROC) curve, seed cell area index (SCAI) and modified SCAI (mSCAI), were used to evaluate the prediction accuracy and robustness of each model using validation datasets. Bootstrapped replicates were considered in estimating the accuracy and robustness of each model by varying gully/no-gully samples. The IOE results indicated that elevation, lithology and slope angle promoted favourable conditions for gully erosion in the study area. The results showed that the IOE model performed better than the FR and CF models for all three validation datasets (AUROCmean = 0.874 and TSSmean = 0.855). This finding was also confirmed in terms of stability and robustness (RTSS = 0.024 and RAUROC = 0.023). The SCAI and mSCAI results showed that all the models exhibited acceptable accuracy, but IOE demonstrated superior performance. Accordingly, IOE was used as the reference model for the study area, indicating that 19.75% and 9.44% of the study area are included in the predicted high and very high susceptibility classes, respectively. Considering the accuracy of GESM, IOE is a reliable tool for decision-making, management and land use planning within the region.