Quantcast

Gully erosion susceptibility mapping: the role of GIS-based bivariate statistical models and their comparison

Research paper by Omid Rahmati, Ali Haghizadeh, Hamid Reza Pourghasemi, Farhad Noormohamadi

Indexed on: 22 Feb '16Published on: 22 Feb '16Published in: Natural hazards (Dordrecht, Netherlands)



Abstract

Gully erosion is a key issue in natural resource management that often has severe environmental, economic, and social consequences. The objective of the present study is to assess the capability of weights-of-evidence (WofE) and frequency ratio (FR) models for spatial prediction of gully erosion susceptibility and characterizing susceptibility conditions at Chavar region, Ilam province, Iran. At first, a gully erosion inventory map is prepared, using multiple field surveys. In total, of the 63 gullies which have been identified, 44 (70 %) cases are randomly algorithm selected to build gully susceptibility models, while the remaining 19 (30 %) cases are used to validate the models. The effectiveness of gully erosion susceptibility assessment via GIS-based models depends on appropriate selection of the conditioning factors which play an important role in gully erosion. Learning vector quantization (LVQ), one of the supervised neural network methods, is employed in order to estimate variable importance. In this research, the selected conditioning factors are: lithology, land use, distance from river, soil texture, slope degree, slope aspect, plan curvature, topographic wetness index, drainage density, and altitude. Finally, validation of the gully dataset which has not been utilized during the spatial modeling process is applied to validate the gully susceptibility maps. The receiver operating characteristic curves for each gully susceptibility map (i.e., produced by WofE and FR) are drawn, and the areas under the curves (AUC) are calculated. The results show that the gully erosion susceptibility map produced by the frequency ratio model (AUC = 78.11 %) functions well in prediction compared with the WofE model (AUC = 70.07 %). Furthermore, LVQ results reveal that distance from river, drainage density, and land use are the most effective factors.