انجمن مخاطره شناسی ایران مدیریت مخاطرات محیطی 2423-415X 6 4 2019 12 22 Evaluation and Spatial Prediction of Landslide Hazard in Mountainous Road of Sanandaj-Kamyaran using Advanced Data Mining Algorithms ارزیابی و پیش‌بینی مکانی مخاطرۀ زمین‌لغزش در جادۀ کوهستانی سنندج-کامیاران با استفاده از الگوریتم‌های پیشرفتۀ داده‌کاوی 317 340 75388 10.22059/jhsci.2020.294693.527 FA صالح میرزانیا کارشناسی ارشد رشتۀ مخاطرات محیطی، گروه ژئومورفولوژی، دانشکدۀ منابع طبیعی، دانشگاه کردستان هیمن شهابی استادیار گروه ژئومورفولوژی، دانشکدۀ منابع طبیعی، دانشگاه کردستان 0000-0001-5091-6947 Journal Article 2019 12 25 Introduction Communication as one of the most important elements of modern civilization provides the background for economic and social development, and development in different regions. Road construction is one of the major causes of landslides in mountainous areas. Landslides are natural disasters that cause a lot of financial and life losses in the country, annually. Identifying high risk areas can reduce the damages and be effective on land development policies. In various studies, different factors and conditioning factors have been considered for the occurrence of landslides. On the other hand, landslide susceptibility mapping is the first and most important step in preventing and controlling of landslides. The roads of Kurdistan province are constantly witnessing mass movements including landslides and rock fall due to the mountainous and climatic conditions. These landslides causing tens of thousands of dollars of damage each year. The Sanandaj-Kamyaran main road is also one of the areas with high hazardous potential due to its location, and variety of environmental variables including climatic, tectonic, lithology and land cover conditions. Hence, spatial prediction of mass movements and landslide susceptibility mapping on the Sanandaj-Kamyaran mountainous road using advanced data mining algorithms such as weight of evidence (WOE) and evidential belief function (EBF) is essential. Materials and methods In this study according to previous studies and regional conditions, fourteen conditioning factors including slope, aspect, elevation, distance to river, river density, distance to fault, distance to road, land use, soil type, curvature, lithology, normalized difference vegetation index (NDVI), stream power index (SPI),  and topographic wetness index (TWI) were used to landslide hazard potential map. Also, two developed data mining models including EBF and WOE were used to extraction of landslide susceptibility mapping. The EBF model is based on the Dempster–Shafer Theory of Evidence. Therefore, to implement the EBF model, the layers of the conditioning factors were transformed into evidential data layers and then integrated using knowledge of the spatial relationships between the landslide occurrences and factors influencing the land sliding in order to generate a predictive landslide susceptibility Index (LSI) map. One of the advantages of this model is that both the predicted landslide and flooding zone outputs exist within the same degree of uncertainty. The EBF model is composed of four functions, namely: Bel (degree of Belief), Dis (degree of Disbelief), Unc (degree of Uncertainty) and Pls (degree of Plausibility). Four maps of Bel, Dis, Pls, and Unc were used for the assessment of the fourteen factors influencing landslide. The  Weights of Evidence  (WOE)  is  a  statistics  method  can  uses  in  probability  condition  to  assess parameters  which  are  influence  on  one  or  more  other  phenomena.  It establishes a relationship between factors and it uses intersected with among variable. Weight of evidence origin was of Bayes’ theorem that predicts variable from combine parameter maps. Finally, the receiver operating characteristic (ROC) curve and the area under the curve (AUC) were used for validation of the two achieved landslide susceptibility map. Discus and Results The altitudes of the study area are minimum 1293 m and maximum 2288 m. The mapping of altitude levels of the study area and the use of WOE as well as the use of EBF model showed that the altitude of 1300-1450 m had the greatest impact on landslide occurrence in the study area. Investigation of aspect map based on WOE showed that the highest and the lowest frequency of landslide occurred was in the northwest and southwest direction, respectively, but according to the EBF model, the highest and the lowest frequency of landslide occurred was in the northwest and flat directions, respectively. Furthermore, evaluation of river density map in WOE and EBF models showed that medium and low river densities had the most impact on landslide occurrence, respectively, but high and very high river density had least effect on landslide occurrence, respectively. Investigation of the information layers in the WOE model showed that TWI with very low class, SPI with very low class, and distance to fault in very high class had the most impact on landslide occurrence, but in EBF model, TWI with middle class and very high class had the highest and least impact on landslide occurrence, respectively. Also, SPI at very low class and distance to fault at very high class had the greatest impact on landslide occurrence. According to the evaluation criterion used in this study (ROC) and validation data, the WOE function model performed better than the EBF model. Conclusion The findings of this research showed that the advanced data mining algorithms based on their structure have sufficient accuracy in spatial predicting of landslide in the study area. In general, it can be said that a rigorous landslide susceptibility map can help managers especially in natural hazard management section in identifying landslide sensitive areas for disaster management. The field survey is a difficult approach for the preparation of the landslide inventory map, especially for elevation which often affects landslide distribution. Landslides that occur in high altitude areas are often lost, because of the difficulty of accurate field surveys. Thus, it is recommended that identification of landslide locations should be based on high-resolution satellite images. زمین‌لغزش‌ها از بلایای طبیعی هستند که سالانه خسارت‌های مالی و جانی زیادی را در کشور ایجاد می‌کنند. شناخت مناطق پرخطر می‌تواند در کاهش خسارت‌ها و تصمیم‌گیری در زمینۀ سیاست‌های توسعۀ اراضی مؤثر باشد. هدف این پژوهش، پیش‌بینی مکانی خطر زمین‌لغزش در محدودۀ جادۀ ارتباطی سنندج-کامیاران در استان کردستان است. در این تحقیق، پهنه‌بندی خطر زمین‌لغزش با استفاده از الگوریتم‌های پیشرفتۀ داده‌کاوی از قبیل مدل تابع شواهد قطعی (EBF) و شواهد وزنی (WOE) انجام گرفت. ابتدا 79 زمین‌لغزش با استفاده از پیمایش میدانی شناسایی شد. سپس این نقاط به‌طور تصادفی به‌منظور تهیۀ مدل و اعتبارسنجی به‌ترتیب به دو گروه آموزش (70 درصد، 55 نقطه) و اعتبارسنجی (30 درصد، 24 نقطه) تقسیم شدند. در این پژوهش، با توجه به تحقیقات قبلی و شرایط منطقه، چهارده عامل مؤثر در وقوع زمین‌لغزش‌ها شامل شیب، جهت شیب، ارتفاع، فاصله از رودخانه، تراکم رودخانه، فاصله از گسل، فاصله از جاده، کاربری اراضی، جنس خاک، انحنای شیب، لیتولوژی، شاخص تفرق پوشش گیاهی ((NDVI، شاخص توان آبراهه (SPI) و شاخص رطوبت توپوگرافی (TWI) برای پهنه‌بندی پتانسیل خطر زمین‌لغزش در نظر گرفته شدند. همچنین در این پژوهش با استفاده از منحنی ویژگی عملگر نسبی (ROC)، عملکرد هر دو مدل بررسی شد.نتایج تحلیل منحنی ویژگی عملگر نسبی نشان داد که مدل‌های WoE و EBF به‌ترتیب دارای مقدار AUC 89/0 و 79/0 هستند؛ بنابراین مدل WoE نسبت به EBF دارای بیشترین مقدار AUC بوده و بهترین مدل برای پیش‌بینی مکانی خطر زمین‌لغزش در منطقۀ پژوهش است. افزون‌بر آن، نتایج نشان داد که الگوریتم‌های پیشرفتۀ داده‌کاوی بنا به ساختار خود از دقت کافی در پیش‌بینی مکانی زمین‌لغزش در منطقۀ پژوهش برخوردارند. https://jhsci.ut.ac.ir/article_75388_2683f10bde07984e524f91f1dcd1d7d5.pdf