ارزيابي خطر زمين‌لغزش با استفاده از فرايند سلسله مراتبي (AHP) ،تحليل شبكه عصبي مصنوعي (ANN) و مطالعات ميداني با رويكرد كاهش ريسك(مطالعه موردي: محور جاده هراز)

Examining the Landslide Risk using Analytic Hierarchy Process (AHP), Artificial Neural Network (ANN) Analysis and Field Studies Aiming for Risk Reduction (Case Study: Haraz Road)

شناسائي پهنه ­هاي حساس متاثر از لغزش و ريزش مهم­ترين بخش در تهيه نقشه زمين‌لغزش و ريزش مي ­باشد. مطالعه حاضر در محور جاده هراز حدفاصل بين امامزاده­ هاشم و شهرستان آمل كه از محور­هاي پرتردد و پرخطر ايران مي ­باشد انجام پذيرفت. در اين پژوهش از دو مدل شبكه عصبي مصنوعي (ANN) و فرآيند سلسله ­مراتبي (AHP) به منظور ارزيابي و مقايسه با وضع موجود و معرفي مدل مناسب استفاده شده است. همچنين وضعيت موجود منطقه به صورت ميداني پايش شده و مناطق متاثر از پديده­ هاي ريزش و لغزش ثبت گرديده است. سطح بندي نقاط از نظر ميزان خطر، تعداد نقاط از نظر عوامل خطرآفرين و اشتراك عوامل مخاطره ­آميز در نقاط ثبت شده، مرحله ديگري از مطالعه ميداني مي ­باشد كه با رويكرد كاهش ريسك انجام پذيرفت. از نقشه ­هاي رقومي ارتفاعي، شيب، جهت شيب، زمين­شناسي، كاربري اراضي، فاصله از گسل، فاصله از راه به عنوان لايه ­هاي اطلاعاتي استفاده شد. در مطالعات ميداني 261 نقطه حادثه‌خيز متاثر از لغزش و ريزش ثبت گرديد. با در نظر گرفتن تعداد نقاط برداشت شده و آمار اشتراك مخاطرات موجود در آن‌ها و همچنين سطح بندي نقاط اين نتيجه بدست آمد كه تمامي نقاط ثبت شده با خطر احتمالي ريزش مواجه است. لازم به­ ذكر است 191 مورد از نقاط ثبت شده با سطح پر خطر تشخيص داده شد. بيشترين خطر در رابطه با اشتراك مخاطرات مربوط به ريزش و روآناب بدست آمد. با توجه به اين كه 88 مورد از نقاط ثبت شده مرتبط با پديده لغزش مي­ باشد، در صورت رخداد، اين مورد بيشترين تاثير را از نظر ميزان آسيب خواهد داشت. در نهايت نتايج حاصل از پژوهش نشان داد كه روش شبكه عصبي مصنوعي با وضع موجود مطابقت بيشتري داشته است. خطاي بدست آمده از روش شبكه عصبي مصنوعي 8 درصد بوده و اين ميزان، قابل قبول بودن تحليل­هاي انجام شده را نشان مي­دهد و اين روش مي ­تواند در مناطق مشابه مورد استفاده قرار گيرد.

Introduction Natural disasters management requires local information to make human societies ready against dangers and reduce the disaster procedure. Hence, evaluation of landslide occurrence in the areas prone to landslide due to geographical condition and human constructions is highly crucial. Identification of rockfall and landslide sensitive zones makes the most important part of preparing map for such areas. The present study is conducted on Haraz road, within the distance of Imamzade Hashem and city of Amol, which is a busy and hazardous road in Iran. Methodology Two models of Artificial Neural Network (ANN) and Analytic Hierarchy Process (AHP) is used to assess and make comparison with the present condition while a proper model is introduced. A field study is also conducted on current condition of the region and landslide sensitive areas are recorded. The next stage of the field study involved ranking parts of the road based on the amount of risk; number of these parts according to risk factors, as well as identifying the common risk factors in the parts recorded aiming for risk reduction. Using digital elevation maps, such layers of information on slope, slope direction, geology, land use and distance from a fault line and road were provided. 261 parts of the road were recorded as Results and discussion The road passes through areas that the points overlooking it often have a significant slope. The slope direction map also demonstrates that such zones overlook Haraz road. Geology of the region indicates frequency of Quaternary alluvium, basaltic tuffs and gray shales as well as sandstones in the middle third of the region. The most likely threat to the road with respect to frequency is related to rockfall. It should be noted that surface runoff produced in slopes overlooking the road during rainfall is both hazardous per se and directs the material down when the water runs. Potential Landslide other than those occurred before and fault are among the threats in the next stage. As the area is active tectonically, crushing of stones and other hard formations in fault zones is more observed and this may play a significant role in triggering slips and landslides. Conclusion Based on the number of detected points and common risks there as well as their ranking , it was concluded that all recorded points faced potential risk of rock fall. It is noteworthy that 191 of the recorded points were identified as hazardous. Most hazardous of all common risks was found as related to rock fall and surface runoff. As 88 cases were related to rock fall, it was found as the most effective factor to create harm. It was finally found that the Artificial Neural Network model is more compatible with the current condition. The error obtained from this method was 8 percent confirming that the analysis was acceptable and the method could be used to study similar areas.