Landslide susceptibility assessment of the part of the North Anatolian Fault Zone 1 ( Turkey ) by GIS-based frequency ratio and index of entropy models

Landslide susceptibility assessment of the part of the North Anatolian Fault Zone 1 (Turkey) by GIS-based frequency ratio and index of entropy models 2 3 Gökhan Demir 1 4 5 1 Department of Civil Engineering, Faculty of Engineering, Ondokuz Mayıs University, 6 Samsun, Turkey 7 Correspondence: gokhan.demir@omu.edu.tr 8 9 Abstract: In the present study, landslide susceptibility assessment for the the part of the 10 North Anatolian Fault Zone is made using index of entropy models within geographical 11 information system. At first, the landslide inventory map was prepared in the study area 12 using earlier reports, aerial photographs and multiple field surveys. 63 cases (69 %) out 13 of 91 detected landslides were randomly selected for modeling, and the remaining 28 14 (31 %) cases were used for the model validation. The landslide-trigerring factors, 15 including slope degree, aspect, elevation, distance to faults, distance to streams, distance 16 to road. Subsequently, landslide susceptibility maps were produced using frequency 17 ratio and index of entropy models. For verification, the receiver operating characteristic 18 (ROC) curves were drawn and the areas under the curve (AUC) calculated. The 19 verification results showed that frequency ratio model (AUC=75.71%) performed 20 slightly better than index of entropy (AUC=75.43%) model. The interpretation of the 21 susceptibility map indicated that distance to streams, distance to road and slope degree 22 play major roles in landslide occurrence and distribution in the study area. The landslide 23 susceptibility maps produced from this study could assist planners and engineers for 24 reorganizing and planning of future road construction. 25 26


Introduction
Among various natural hazards, landslides are the most widespread and damaging.
This study aims to develop landslide susceptibility maps of the part of the North Anatolian Fault Zone, southeast Resadiye-Koyulhisar Turkey, (Fig. 1), using index of entropy (IOE) model.To achieve this, index of entropy analysis methodology, to obtain landslide susceptibility map using the geographic information system was developed, applied, and verified in the study area.short distances due to the effect of the NAFZ, it is generally to the northeast (Gokceoglu et al. 2005).Landslides are common natural hazards in the seismically active North Anatolian Fault Zone (NAFZ), which is 1,100 km in length and is moving westward with the rate of 2.5 cm every year according to geological and GPS data (Demir et al 2013) The latest catastrophic event occurred on March 17, 2005 at Kuzulu (Sivas) in the valley.The landslide was initiated within highly weathered volcanics in the mode of sliding and then transformed to an earth flow.It killed 15 people, and more than 30 houses and a mosque were buried and damaged by the earth flow material.A second but smaller landslide originated from the same source areas after 4 days and caused additional damages (Gokceoglu et al. 2005;Ulusay et al. 2007;Yılmaz 2009).After the main event, the governor needed landslide susceptibility maps of the landslide area.

Data production
The study began with the preparation of a landslide inventory map based on field work, earlier reports and satellite images.Landslide inventory maps show the areal distribution of existing landslide areas and their characteristics.These maps indicate the landslides, which are perceptible on site (Cevik and Topal 2003).In total, 321 landslides were mapped (Fig. 2) and subsequently digitized for further analysis.The mapped landslides cover an area of 47.81 km2, which constitutes 6.68 % of the entire study area.From these landslides, 63 (69 %) randomly selected instabilities were taken for making landslide susceptibility models and 28 (31 %) were used for validating the models.The number of landslide-conditioning factors may range from only a few numbers to several (Mohammady et al. 2012;Pourghasemi et al. 2012d;Papathanassiou et al. 2012).The selection of these factors mainly depends on the availability of data for the study area and the relevance with respect to landslide occurrences (Papathanassiou et al.2012).According to the Ayalew and Yamagishi (2005), in GIS-based studies, the selected factors should be operational, complete, non-uniform, measurable, and nonredundant.We prepared six thematic data layers representing the following landslide conditioning factors: slope degree, aspect, elevation, distance to faults, distance to streams, distance to road (Figure 3).The main parameter of the slope stability analysis is the slope degree(S.Lee and K. Min 2001).Because the slope degree is directly related to the landslides, it is frequently used in preparing landslide susceptibility maps[S.Lee, J. H. Ryu, J. S. Won and H. J. Park (2004), M. Ercanoglu, C. Gokceoglu, T. W. J. Van Asch (2004, A. Clerici, S. Perego, C. Tellini andP. Vescovi (2002) .For this reason, the slope degree map of the study area is prepared from the digital elevation model (DEM) and divided into nine slope classes with an interval of 5°(Figure 3).Aspect and elevation also were extracted from the DEM.Aspects are grouped into 9 classes such as flat (-1), north (337.5°-360°,0°-22.5°),northeast (22.5°-67.5°),east (67.5°-112.5°),southeast (112.5°-157.5°),south (157.5°-202.5°),southwest (202.5°-247.5°),west (247.5°-292.5°),and northwest (292.5°-337.5°).In the study area, the elevation ranges between 500 and 2,625 m.The elevation values were divided into nine classes (Figure 3).The distance from faults, road and stream is calculated at 100m intervals using the geological map (Figure 9).An important parameter that controls the stability of a slope is the saturation degree of the material on the slope (Yalçın 2008, Yalçın andBulut 2007).The closeness of the slope to drainage structures is another important factor in terms of stability.Streams may adversely affect stability by eroding the slopes or by saturating the lower part of material resulting in water level increases (Pourghasemi et al. 2012a, Gökçeoğlu 1996, Saha et al. 2002).For this reason, ten different buffer zones were created within the study area to determine the degree to which the streams affected the slopes.A road constructed can cause a disturbance of the slopes that lead to increase in stress on the back of the slope, because of changes in topography and decrease of load on toe, some tension cracks may develop.Although a slope is balanced before the road construction, some instability may be happened because of negative effects of excavation.In the current study many landslides were recorded along the roads in the study area that is due to road construction.The distance from roads was calculated and reclassified into ten classes.

Landslide Susceptibility Analysis a. Application of Index of Entropy Model
In this study index of entropy model was used for landslide susceptibility analysis using six landslide conditioning factors.
The entropy indicates the extent of the instability, disorder, imbalance, and uncertainty of a system (Yufeng and Fengxiang, 2009).The entropy of a landslide refers to the extent that various factors influence the development of a landslide (Pourghasemi et al., 2012b;Jaafari et al., 2013).Several important factors provide additional entropy into the index system.Therefore, the entropy value can be used to calculate objective weights of the index system.The equations used to calculate the information coefficient Wj representing the weight value for the parameter as a whole (Bednarik et al., 2010;Constantin et al., 2011)   The final landslide susceptibility map was prepared by the summation of weighted products of the secondarily parametric maps.The final landslide susceptibility maps using index of entropy model was developed using the following equation: where Y is the value of landslide susceptibility (Fig. 4).The result of this summation is a continuous interval of values from 0.7297 to 6.1861, which represents the landslide susceptibility index.A natural break classification method was used to divide the interval into four classes and a susceptibility map was prepared (Bednarik et al., 2010;Constantin et al., 2011;Erner et al., 2010;Falaschi et al., 2009;Ram Mohan et al., 2011;Xu et al., 2012aXu et al., , 2012b)).According to the landslide susceptibility map generated with the IOE model (Fig. 4 and Table 1), it was found that 24.87% and 23.50% of the total landslides falls in the very low and low susceptibility zones respectively.Moderate, high, and very high susceptible zones represent 20.37%, 16.42%, and 14.83% of the landslides pixels, respectively.

b. Application of Frequency ratio method
Frequency ratio method is a simple and understandable probabilistic model, and the model is based on the observed relationships between distribution of landslides and each landslide-causative factor, to reveal the correlation between landslide locations and the factors in the study area (Lee and Pradhan, 2007).To calculate the frequency ratio, the ratio of landslide occurrence to non-occurrence (Regmi et al., 2013) was calculated for each factor's class.Therefore, the frequency ratio for each factor's class was calculated from its relationship with landslide events.The frequency ratio is defined as shown in Equation ( 8).Therefore, the greater the ratio above unity, the stronger the relationship between 202 landslide occurrence and the given factor's class attribute, and the lower the ratio below 203 unity, the lesser the relationship between landslide occurrence and the given factor's 204 class attribute (Lee and Pradhan, 2006;Yalcin et al., 2011) The LSI map was reclassified using the equal interval method in GIS, and as a result, 210 the study area was divided into five susceptibility classes: very low, low, moderate high, 211 and very high (Fig. 5).According to this landslide susceptibility map, 24.67 % of the

Validation of Landslide Susceptibility map
Landslide susceptibility maps without validation are less meaningful (Chung and Fabbri by using receiver operating characteristics (ROC) (Akgun et al., 2012;Tien Bui et al., 2012a, b, 2013;Regmi et al., 2014;Ozdemir and Altural, 2013).The ROC curve is a useful method for representing the quality of deterministic and probabilistic detection and forecast systems.The ROC plots the different accuracy values obtained against the whole range of possible threshold values of the functions, and the ROC serves as a global accuracy statistic for the model, regardless of a specific discriminate threshold (Pourghasemi et al., 2012).In the ROC curve, the sensitivity of the model (the percentage of existing landslide pixels correctly predicted by the model) is plotted against 1-specificity (the percentage of predicted landslide pixels over the total study area) (Mohammady et al., 2012;Jaafari et al., 2013).The area under the ROC curve (AUC) represents the quality of the probabilistic model to reliably predict of the occurrence or non-occurrence of landslides.A good fit model has an AUC values range from 0.5-1, while values below 0.5 represent a random fit.The success rate results were obtained by comparing the landslide training data with the susceptibility maps (Fig. 6).
AUC plot assessment results showed that the AUC values were 0.7571 and 0.7543 for FR and IOE models and the training accuracy were 75.71 and 75.43 %, respectively.
From the results of the AUC evaluation, it is seen that both the success rate curve show almost similar result.All the models employed in this study showed reasonably good accuracy in predicting the landslide susceptibility of the study area.

Conclusion
In this study, the accepted models, frequency ratio and index of entropy, within a GIS environment for the aim of LSM have been used.For this purpose, six trigerring factors, i.e., slope degree, aspect, elevation, distance to faults, distance to streams and distance to road were used.A total of 91 landslides were identified and mapped.Out of which, 63 (69 %) were randomly selected for generating a model and the remaining 28 (31 %) were used for validation purposes.In this study, five landslide susceptibility classes, i.e., very low, low, moderate, high, and very high susceptibility for landsliding, were derived with equal interval method.The validation has been determined by using the ROC method in which the accuracy of the LS maps produced by the frequency ratio and index of entropy models was 0.757 and 0.754, respectively for success rate technique.
This shows that all the models employed in this study showed reasonably good accuracy in predicting the landslide susceptibility of the part of the North Anatolian Fault Zone (Turkey).All susceptibility zones require further engineering geological and geotechnical considerations.The increasing population pressure has forced people to concentrate their activities on steep mountain slopes.Thus, to safeguard the life and property from landslides, the susceptibility map can be used as basic tools in land management and planning future construction projects in this area.The landslide susceptibility map produced in this study can be used for optimum management by decision makers and land use planners, and also avoidance of susceptible regions in study area.Akgun, A., (2012).A comparison of landslide susceptibility maps produced by logistic regression, multi-criteria decision, and likelihood ratio methods: a case study at İzmir, Turkey.Landslides 9 (1), 93-106.http://dx.doi.org/10.1007/s10346-011-0283-7.

Figure
Figure 1.Study Area

Figure 3 .
Figure 3. Conditioning Factors(a.slopedegree, b.distance to stream, c.distance to road, d.elevation, e.distance to faults,f.aspect.) b are the domain and landslide percentages, respectively, (Pij) is the probability density, Hj and Hj max represent entropy values, Ij is the information coefficient and Wj represents the resultant weight value for the parameter as a whole.

Figure 6 .
Figure 6.Success rate curves of FR and IOE models of the study area.