چکیده:
در این پژوهش یک مدل ترکیبی نوین بهمنظور افزایش دقت تهیه نقشه حساسیت زمینلغزش در حوضه دزعلیا، استان اصفهان که یک منطقه حساس نسبت به زمینلغزش میباشد ارائهشده است. بدین منظور در ابتدا با استفاده از مطالعه ادبیات تحقیق، تفسیر عکسهای هوایی و خصوصیات منطقه مطالعاتی ۲۳ فاکتور مؤثر در زمینلغزش شامل فاکتورهای ژئومورفولوژیکی، زمینشناختی، هیدرولوژیکی و محیطی انتخاب گردید، سپس با استفاده از مدل AHP < /span> به غربالگری پارامترها پرداخته شد و تعداد ۱۲ پارامتر بهمنظور اجرای مدل انتخاب گردید. با توجه به اینکه میزان تأثیر پارامترها در زمینلغزش در بخشهای مختلف یک حوضه یکسان نمیباشد بهمنظور رفع این مشکل از مدل رگرسیون وزنی جغرافیایی بهمنظور قطعهبندی حوضه موردمطالعه استفاده گردید و حوضه با استفاده از ۳ پارامتر لیتولوژی، TPI و انحنای سطح به ۲۵ قطعه تقسیم گردید و سپس مدل SVM-FR برای هر یک از قطعهها اجرا گردید و درنهایت از تلفیق قطعهها، نقشه نهایی پهنهبندی حساسیت زمینلغزش حاصل گردید. از ۸۴ زمینلغزش موجود در منطقه ۷۰ درصد (۵۹ زمینلغزش) بهمنظور اجرای مدل و ۳۰ درصد (۲۵ زمینلغزش) بهمنظور صحت سنجی مورداستفاده قرار گرفت. بهمنظور بررسی دقت و صحت مدل، به مقایسه مدل با مدلهای SVM-FR و FR با استفاده از منحنی ROC پرداخته شد و نتایج نشان داد مدل ترکیبی دارای دقت پیشبینی بالاتری (۸۵۱/۰) نسبت به مدل SVM-FR (۷۴۲/۰) و مدل FR (۷۱۴/۰) میباشد. بر اساس نتایج حاصل از مدل ترکیبی ۵۱/۳۴۵۰ هکتار (۷۴/۲۰ درصد) از منطقه مطالعاتی در رده خطر زیاد و ۹۴/۴۴۱ هکتار (۶۶/۲ درصد) در رده خطر خیلی زیاد قرار دارد. با توجه به تأثیر شگرف مدل GWR در بالا بردن دقت نقشههای حساسیت زمینلغزش، استفاده از آن در پژوهشهای مربوط به زمینلغزش توصیه میگردد.
Introduction :The landsides are one of the most common catastrophic natural dangers that occur in many regions of world and cause to hundreds milliard dollar economic loss and hundreds thousand mortality and injuries yearly. During recent years, many governments and research institutions have invested in assessing the landslide risks for preparing the maps that indicate the landslides spatial distribution (Gusset et al 1999). Regardless the obtained progresses for identifying, measuring, forecast and warn systems of landslide, but yet the losses result from landslide are increasingly in the worldwide (Schuster 1996). The harmful losses of landside is relatively very high in Iran and have been caused to very much mortality. For example, In Spring of 1377, the Abikar village in the Bazoft region of chaharmahal bakhtiyari buried under high volume of soil and stone and all of its habitants died (Ali mohammadi and colleagous 1388). In this research, a modern hybrid approach of GWR-SVM-FR have been used. It’s worthy to note that different degrees of parameters effect maybe occur in studied region so that with local changing in one basin, the effect of one parameter will change (Emer, Duzgun 2010). For removing this problem in this research, the geographical weight regression model have been used for preparing landslide sensitive map that is an innovation and evolution for preparing the landslide sensitive maps. Methodiligy:The used data in this research for extracting the environmental parameters including: geology map in 1:100000 scale, topography map in 1:50000 scale, aerial images with high quality in 1:40000 scale, satellite images ETM+ , ASTER data with 30m accuracy, precipitation data of climatology station in a 30 years period. Many researchers have confirmed the correlation between various environmental factors with landslide event (liu et al 2004). According to these researches and studied region specifications 23 environmental factor selected for predicting the landslide sensitive regions including geomorphological, geological, hydrological and environmental factors. In the next step, for screening the parameters, the AHP method have used (figure 15). 12 parameters, among 23 parameters selected for executing the model .In the next stage, the studied region divided to 25 divisions by using GWR method and the numbers of each 12-fold parameters in each division determined by using SVM method. The FR method used for determining the sub-criterion weight too. In the last step, after executing the SVM-FR model for each division, the maps merged altogether and the final map of landslide sensitive zoning obtained.. Results and discussionThe results of criterion ranking in 25-fold regions with SVM model indicated that in the most regions, the lithology, TPI and surface curvature parameters have had too effect in landslide occurring that conform with research results (Ahmad Abadi and Rahmati 1394, Regmi, Jaafari et al 2014). The results of assessing the landslides sensitive according to frequency ratio model indicated that the effect amount of every class of effective factors in landslide have been specified according to accumulation of occurred landslide in that class (table 2). Just as indicated in table 2, the height parameter in 3190-4000 m class with frequency ration of 2/131 have had the grates effect in landslide occurring, and the frequency ratio will decrease with decreasing the height. Pachauri& Pant 1992 suggested that higher heights indicate more sensitive to landslide. and etc. Table 2 indicate the results of relation of other parameters with landslides by using frequency ratio model. After calculating the weight of 12-fold parameters for every 25-fold regions by using back vector machines method and multiply it on the weight of mentioned factors classes by using frequency method, the weight maps added together and the landslide sensitive map obtained for every regions. Finally, the final map of landslide sensitive obtained by integrating the 25-fold regions maps. Conclusionresults indicated that hybrid model GWR-SVM-FR, have the higher accuracy (0/158) relative to SVM FR (0/247) and FR model (0/417). The results of this research conform with sabokbar et al 2016 investigation that believe GWR algorithm is very effective in upgrading the accuracy of landslide sensitivity. According to final results of hybrid method, in the studied region 138/89 hec (%5) locate in very low risk level, 33296/73 hec (%14/26) in low risk, 25894/52 hec (%92/79) in moderate risk, 543/15 hec (2/47 %) in high risk and 144/49 hec (% 2/66) is in very high risk.
خلاصه ماشینی:
بدين منظور، به هر واحد محاسباتي در منطقه مطالعاتي يک ضريب GWR با استفاده از استخراج نوع کرنل و انتخاب معيارهاي مناسب اختصاص داده شد و واحدهايي که داراي ضريب مشابه بودند بهعنوان ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ 1 Conoscenti et al 2 Hong et al 3 Chen et al 4 Chen et al 5 artificial neural network (ANN) 6 Lee et al 7 Gomez and Kavzoglu 8 Wang and Sassa 9 Melchiorre et al 10 Li et al 11 Zare et al 12 Dou et al 13 Binh et al 14 support vector machine (SVM) 15 Yao et al 16 Erener and Düzgün 17 Marjanović et al 18 Xu et al 19 Ballabio and Sterlacchini 20 Pradhan 21 Pourghasemi et al 22 San 23 Yao et al 24 Peng et al 25 Hong et al 26 geographically weighted regression (GWR) 27 Erener and Düzgün 28 Sabokbar et al 29 Xianyu et al يک واحد در نظر گرفته شدند.