كاربرد شاخص هاي كمي ژيومورفومتريك در شناسايي پهنه‌هاي مستعد زمين لغزش با استفاده از مدل SVM (مطالعه موردي: آزادراه خرم‌آباد – پل‌زال)

Geomorphomety quantitative indices used in the identification of zones prone to landslides using the model SVM

شناخت و نحوه پراكنش فضايي لندفرمي به‌منظور درك و ارزيابي تحول آن ها، مطالعات پايداري دامنه اي و برنامه ريزي منطقه اي آن ها در آينده، از نيازهاي اساسي در علم ژيومورفولوژي كاربردي است. اهميت مطالعه لندفرم ها به حدي است كه امروزه موضوع مطالعه ژيومورفومتري به‌عنوان زيررشته‌اي از ژيومورفولوژي ارايه مي شود. ژيومورفومتري‌ به اندازه‌گيري كمي سطوح و لندفرم‌ها بر اساس تغييرات ارتفاعي تحت تاثير تابع فاصله مي‌پردازد. شاخص‌هاي ژيومورفومتريك، ويژگي‌هاي شكل عوارض زميني را به‌صورت كمي بيان مي‌دارند. اين تحقيق با تاكيد بر استفاده از شاخص‌هاي ژيومورفومتريك و الگوريتم SVM به شناسايي سطوح مستعد زمين لغزش در آزادراه خرم آباد – پل زال به‌عنوان يكي از راه‌هاي ارتباطي مهم كشور مي‌پردازد. شاخص‌هاي مورداستفاده شامل شيب، جهت شيب، ليتولوژي، وضعيت گسل‌ها، شبكه زهكشي و كاربري اراضي است كه به همراه شاخص‌هاي ژيومورفومتريك شامل انحناي پروفيل، انحناي پلان و انحناي كلي با استفاده از رويكرد هوش مصنوعي و توابع خطي و چندجمله‌اي الگوريتم SVM در شناسايي سطوح مستعد لغزش استفاده‌شده است. نتايج تحقيق نشان مي‌دهد استفاده از شاخص‌هاي ژيومورفومتريك، انحناي پلان، پروفيل و كلي توانسته ويژگي‌هاي شكلي سطوح را به‌صورت كمي مشخص نمايد و درنتيجه نقش مهمي در افزايش دقت شناسايي سطوح مستعد لغزش داشته است. با توجه به به‌هم‌ريختگي و زبري سطوح لغزشي شاخص‌هاي ژيومرفومتريك كارايي خوبي در شناسايي پهنه‌هاي لغزشي داشته‌اند. ارزيابي دقت با استفاده از داده‌هاي پيمايش زميني ضمن تاكيد اين مطلب نشان مي‌دهد تابع چندجمله‌اي در شناسايي سطوح مستعد لغزش، دقت بيشتري نسبت به تابع خطي الگوريتم SVM داشته است كه علت آن مي‌تواند رفتار غيرخطي وقوع لغزش در منطقه مطالعاتي است و بنابراين تابع غيرخطي الگوريتم بهتر توانسته احتمال وقوع لغزش را مشخص نمايد.

Introduction Landslides are usually complex systems and prediction of their the occurrence sensitivity needs to environmental various datas such geomorphometric indicators. Occurred landslides Khorramabad - Paul Zal freeway in Iran southwestern and Lorestan province considers threat serious for zone social and economic stability. Following the construction of this main road that the central and western regions connects to the Persian gulf coastal and plain Khuzestan؛ new trenching in susceptible geological formations, has increased landslides potential and slope unstable has become as a serious and inevitable problem. Therefore the landslide prediction is essential for preventing new landslides and activation older landslide for reducing the risk of area. The aim of this study is evaluate to SVM method in order to identification landslide of susceptible areas with emphasis on the use of geomorpholometric parameters. Methodology In this study, Khorramabad, Bidrubeh and Khalil Akbar basic topographic maps with scale 1: 50,000 and Khorramabad and Balarud geological maps with scale 1: 100,000 and 90 m SRTM digital elevation model have been used. land use layer has been prepared By using Landsat TM images. (PAN) IRS images with numbers (66/48 A), (66/47 C) related to years 2003 and 2005 have been used for detecting landslides before road construction. In addition, for the completion of landslides distribution layer in fieldwork ,altitude and location of new landslides After the road building operation have been recorded by using GPS system and with satellite images and topographic maps have been compered. Also geomorphometric indicators have been calculated by using computational relationship in analyzes geomorphometric part LANDSERF software and outputs have been got with raster format in Arc GIS software. Results and Discussion In order to detect landslides sensitivity, the 39 characteristics of landslides occurred in the study area were harvested. Random samples of landslide areas to implement the model was divided to 2 part for training and validation of the model. The modeling of landslide suscepttible zones in the study area were obtained by using the landslide effective criteria (slope, aspect, lithology, land use fault conditions and drainage) and geomorphometric indicators (profile curvature, plan curvature and total curvature) and polynomial and linear functions in models of SVM. Sensitivity values is between 0 and 1, and the potential and susceptibility landslides in the study area is increasing by approaching to 1 value. The results shows that more than 30 percent of the study area is located in areas with the amount more than 0.8 persent Which shows the high sensitivity of the area to the landslide. Conclusion In this study, in addition to other effective variables, geomorphometric indicators for more accurate surface identification of surface roughness and their effect on landslide and finally, increased accuracy of modeling has been used. With using SVM model the assessment model was done with analyzing the twenty percent of taken samples. The results shows the modeling based on a polynomial function with an overall accuracy of 89% compared to linear function with an overall accuracy of 84% have more accuracy and more realistic performance. With due attention to the behavior of the landslide control variables in the study area is nonlinear and complex؛ polynomial function Compared to linear function has had a better modeling ability and its zonation results is more accommodate with ground truth.