landslide susceptibility zonation with bayes’ theorem (weight of evidence) (case study: siyahrood catchment)

introduction: landslides are amongst the most damaging geologic hazards in the world. they pose a threat to the safety of humankind lives as well as the environment, resources and property. compared with other natural hazards such as volcanic eruptions and floods, landslides cause considerable damage to human beings and massive economic losses (guzzetti, 2005). according to preliminary estimates, about 500 billion riyals annual are caused economic damage in iran by landslide occurrence (hosseinzadeh et al., 1388:27). much literature available on landslide hazard assessment methodologies broadly falls into three main approach groups: qualitative, quantitative and artificial intelligence (ai) approaches. in general, a qualitative approach is based on the subjective judgment of an expert or a group of experts whereas the quantitative approach is based on mathematically rigorous objective methodologies. artificial intelligence (ai) techniques can make use of heuristic knowledge or pattern matching technique as opposed to solving a set of mathematical equations. the ai broadly covers artificial neural networks (ann), expert system, and other heuristic knowledge-based or rules-based techniques. (neaupane and piantanakulchai 2006:281). for the landslide susceptibility mapping can be used a variety of models, such as logistic regression, analytical hierarchy process (ahp), analytic network process (anp), artificial neural network, the bivariate statistical models, lnrf, fuzzy logic models and etc. usually choose the most appropriate approach and model is done based on the data type, the scale of the study area, the scale of analysis and knowledge of researcher. study area: in this study siyahrood catchment has zonation for landslide susceptibility by using weights of evidence models (bayes' theorem). the basin is located in the province of gilan. the catchment area is 437 km and is sub-basins of the sefidrood river. materials and methods: the weight of evidence (wofe) method (bonham-carteretal.,1989) has been used to evaluate shallow-landslide susceptibility and has been tested as a useful spatial data model for various applications including mass-movement studies, mineral research, and groundwater spring mapping (mark and ellen, 995; poli and sterlacchini,2007; barbieri and cambuli, 2009). .it takes into account the relationships existing amongst the occurrence of a supporting evidence (shallow landslides in this study) and the distribution of causal factors (shallow-landslide predisposing factors in this study) the wofe is a statistical method based on the bayes' theorem (denison et al., 2002). this methods first applied to mineral exploration in1988 (bonham-carter et al., 1988) and subsequently, van western (2002) applied the method for landslide susceptibility assessment. bayes’ theorem can be written as: equation (1) p(s│b_i )= (p(b_i│s)×p(s))/(p(b_i)) where p (bi | s) is the conditional probability to have bi given s, p (s) is the prior probability to find s within the study area (as) and p (bi) is the prior probability to find the class bi within the study area as. in this study landslide susceptibility zonation has been done using several natural and anthropogenic parameters such as (lithology, distance from fault, distance from river, rainfall, land slope, aspect, land use, vegetation density (ndvi) and sediment transport index (sti) stream power index (spi) and topographic wetness index (twi )). results and discussion: after the weights classes were obtained using the model for each parameter, weights was applied to each class in the arc map software and eventually with overlay parameters was obtained landslide susceptibility maps. the final maps was classified in 5 susceptibility class using the method of natural breaks (very low susceptibility, low susceptibility, moderate susceptibility, high susceptibility and high susceptibility). according to the results of the model and the map developed in the lithology layer, most of the weight is allocated to class b (old alluvial terraces and high alluvial fans). moderate range among the different classes of land use and north and northwest directions in the aspect parameter have the greatest impact on landslide occurrence. as well as slope of 20-10 degrees and 10-5 degrees, respectively and in the layer distance from the river, 100 meters from the river have the greatest impact in landslide occurrence. conclusion: assessment models with using landslides occurred in the area show that with increasing risk class, landslide density in the class increases and 59 % of landslide, has occurred in very high susceptibility class. while the area of this class compared to total area of the region is only 10.5 percent. although classes with very low susceptibility, low and moderate susceptibility are included approximately 71 percent area of a region, but only a small portion of the landslides occurred (16.7%) in these classes while the roughly 83. 3 percent of landslides occurred in the area are located in the fourth and fifth class (high and very high susceptibility). due to this can be said that the model has a good functionality in the area terms of the prediction of landslides.