Application of morphometric indices in optimization of landslide susceptibility zonation maps using probabilistic methods

Application of morphometric indices in optimization of landslide susceptibility zonation maps using probabilistic As a geomorphic hazard, landslide incurs great deals of financial damages and casualties every year, and directly and/or indirectly contributes to large economic costs in different areas. Given that numerous factors contribute to the occurrence of a landslide, in order to prepare more accurate zonation maps, it is necessary to use more information layers and evaluate various factors that can possibly contribute to the occurrence of the event before adopting the existing models to zonation the susceptibility map. This phenomenon is a natural hazard which is affected by the land surface shape (morphology). As such, when it comes to susceptibility analysis of landslides in a particular area, not only common factors in zonation, but also morphometric features of ground surface are important and should be evaluated. Geomorphometric indices, which together comprise the so-called morphotectonic framework, can be used in studies on many geomorphologic events and natural hazards. These indices express quantitatively characteristics of hillsides which are susceptible to landslide. In this research, a total of 18 factors contributing to the occurrence of landslides in Fahlian watershed were identified and evaluated; these included slope, slope aspect, slope length, altitude, distance to fault, distance to river, precipitation, litology, landuse, general curvature, Plan curvature, profile curvature, Normalized Difference Vegetation Index (NVDI), topographic position index (TPI), Length and Slope Factor (LSF), Terrain Ruggedness Index (TRI), Stream Power Index(SPI), and topographic wetness index (TWI). In order to prepare layers of the effective factors, in addition to 1:100,000 geological maps, 1:50,000 topographic maps, digital elevation model (DEM: ASTER), satellite images, and aerial photographs, ArcGIS, Global Mapper, Surfer, and ENVI 4.5 software packages were utilized. Given the focus of this research on the application of morphometric indices to optimize zonation map of susceptibility to landslide, the indices were extracted. Land surface characteristics (i.e. morphometric, hydrologic, and climatic properties, etc.) and land features (watersheds, stream networks, landforms, etc.) were extracted using digital models of ground surface (DEM) and parameterization software. Subsequently, using Dempster-Shafer probabilistic models, evidence of weight, and the morphometric indices, zonation map of susceptibility to landslide was prepared for Fahlian River watershed. Finally, using receiver operating characteristics(ROC), both models were validated. Based on the weights related to the role of each unit of factor layers and their order of priority and importance in the occurrence of landslide, and upon combining factor maps and landslide distribution maps and calculating the weight of each level based on the relationships related to the Dempster-Shafer model in GIS environment, it was found that, with a belief function weight of 0.87 and minimum disbelief weight, slope > 40% imposes the largest contributions into the occurrence of landslide across the watershed. At lower slopes, other forces such as the friction between soil particles and other hillside material usually dominate over driving forces such as gravity. In contrast, on highly sloping hillsides, due to the dominance of shear stress over resisting force, one may end up with increased probability of the occurrence of a landslide. Moreover, based on the obtained results, with a belief weight of 0.77, TRI > 14 was the second most effective factor on the occurrence of landslides across the studied watershed. It was further found that, Stream Power Index < 1.2, concave curvatures, precipitation < 750 mm/year, TPI < -4.2, profile curvature of 0.3 – 4.2, TWI of -1.5 to 2.5, forest lands, surface curvature of -5 to -2.99, distance to fault of 0 to 500 and Pabdeh – Gurpi and formation (with belief function values of 0.68, 0.63, 0.60, 0.57, 0.49, 0.49, 0.47, 0.46, 0.38, and 0.37, respectively) imposed large contributions into regional landslides.
According to the evidence of weight model, the class of TRI >14 (final weight: 2939.32) was found to the most effective factor on the occurrence of landslide across the region. Following a similar trend of reasoning, the class of slopes higher than 40% (final weight: 2611.21) was the second most important factor, which are in agreement with the results of Javadi et al. (2014) and Teymoori-Yanseri et al. (2017). Moreover, in their research, Pourghasemi et al. (2011) referred to the slope as the second most important factor contributing to the occurrence of landslide. In this model, NVDI > 0.6 (final weight: 400.60) is identified as the third most important factor. Following the land use, Stream Power Index > 1.2, TPI < -4.2, TRI of 7 – 14, profile curvature of 0.1 – 0.3, NVDI of 04 – 0.6, precipitation > 750 mm, and Pabdeh-Gurpi Formation imposed the largest impacts (sorted in the order of effectiveness, with final weights of 2037.60, 1925.99, 1803.48, 1793.34, 1722.40, 1494.60, and 1340.28, respectively). Final results of the present research showed that, in both of the models, slopes higher than 40% and TRI > 14 exhibited the highest weights and played the most significant roles in the occurrence of landslide across the region. Moreover, based on the obtained results, 82.59% of the landslides across the watershed in an area of 547.82 hectare had occurred in postures. Based on the results of Dempster-Shafer model, very low, low, intermediate, high, and very high susceptibility classes covered 23.85% (961.34 km2), 31.82% (1282.49 km2), 21.72% (875.63 km2), 16.41% (661.45 km2), and 6.20% (249.97 km2) of the entire region, respectively.
Moreover, the results obtained from the evidence of wight model model shows that 25.29% (1019.59 km2), 30.98% (1248.82 km2), 21.28% (857.64 km2), 15.68% (631.93 km2), and 6.77% (272.90 km2) of the entire susceptibility zonation map are composed of zones of very low, low, intermediate, high, and very high susceptibility, respectively. Results of evaluating the models using ROC showed that, the Dempster-Shafer model provides higher prediction accuracy (0.79) than the evidence of weight model (0.76). Considering quantitative results of validation, the combination of Dempster-Shafer model with morphometric indices is herein introduced as an appropriate model for zonation susceptibility to landslide.
Keywords: landslide, Dempster-Shafer model, evidence of wight model, morphometry, Fahlian Watershed.