Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

GIS-based spatial prediction of landslide susceptibility using logistic regression model

GIS-based spatial prediction of landslide susceptibility using logistic regression model Geomatics, Natural Hazards and Risk Vol. 2, No. 1, March 2011, 33–50 GIS-based spatial prediction of landslide susceptibility using logistic regression model SEYEDEH ZOHREH MOUSAVI{, ATAOLLAH KAVIAN*{, KARIM SOLEIMANI{, SEYED RAMEZAN MOUSAVI{, and ATAOLLAH SHIRZADI{ {College of Natural Resources, Sari Agricultural Sciences and Natural Resources University, Sari, Iran {College of Natural Resources, University of Kurdistan, Sanandaj, Iran (Received 1 August 2010; in final form 17 August 2010) In the present study, logistic regression analysis has been used to create a landslide hazard map for Sajarood basin, Northern Iran. At first, an inventory map of 95 landslides was used to produce a dependent variable, a value of 0 for absence and 1 for presence of landslides. The effect of causative parameters on landslide occurrence was assessed by the corresponding coefficient that appears in the logistic regression function. The interpretation of the coefficients shows that the road network plays the major role in determining landslide occurrence. Elevation, slope curvature, rainfall and distance to fault were excluded from the final analysis, because these variables do not significantly add to the predictive power of the logistic regression. After running the final probability function into Arc/view 3.2 software, a landslide susceptibility map has been produced. The accuracy assessment shows an overall accuracy of the landslide susceptibility map to be 85.3%. An area of 53.01% is found to be located in a very low, 18.33% in low, 20.96% in moderate and 7.7% in high-risk regions. The proposed 2 2 susceptibility map was tested using –2LL, Cox and Snell R , Nagelkerk R and Roc procedure, and it is found to be very reliable. 1. Introduction Landslides are one of the most destructive natural hazards in mountainous areas and play an important role in landform evolution and cause serious hazards in the world. Landslides occur when unstable masses of soil and stone are affected by earthquakes, heavy rainfall and various human activities. Damage to residential regions, economic losses and human fatalities caused by landslides are increasing worldwide. Landslides represented approximately 9% of the natural disasters that occurred worldwide during the 1990s (Gomez and Kavzoglu 2005, Yilmaz 2009). Because of mainly mountainous topography, tectonic and seismic activities, various geological and climatic conditions, population increase and land-use change, numerous landslides have occurred in Iran in recent decades (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009, Mousavi et al. 2009). Every year, landslides cause economic *Corresponding author. Email: a.kavian@sanru.ac.ir Geomatics, Natural Hazards and Risk ISSN 1947-5705 Print/ISSN 1947-5713 Online ª 2011 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/19475705.2010.532975 34 S.Z. Mousavi et al. damage to roads, railway tracks, power lines, irrigation channels, facilities of gas petroleum extraction and refinery factories and industrial centres, dams and natural or artificial reservoirs, forests, range lands, farms and villages in many provinces in Iran. In February 1998, one of the biggest wayside landslides on Emamzade Ali- Abask region of Haraz road in Mazandaran, Iran occurred, resulting in damage to more than 600 m of road, the whole village, many restaurants, several power lines and 10 fish production ponds (National Geosciences Database of Iran, htpp:// www.ngdir.ir). Several factors, such as geology, hydrology, hydrogeology, topo- graphy and morphology, climate and weathering affect slope instability and cause landslides (Soeters and van Westen 1996, Chau et al. 2004b, Peart et al. 2005, Domı´nguez-Cuesta et al. 2007, Garfi et al. 2007). Elevation, slope angle and slope aspect are the main topographic factors in landslide occurrence (Dai and Lee 2002, Ohlamcher and Davis 2003, Chau and Chan 2005, Nefeslioglo et al. 2008). Slope curvature is another causative factor on mass movement that a few researchers such as Can et al. (2005), Lee and Sambath (2006) and Greco et al. (2007) have considered the influence of on mass movements. On the other hand, significant detachments in slopes occur after heavy or long-term rainfall and water infiltration into the cracks (Crosta 1998, Corominas and Moya 1999, Ayalew et al. 2005, Dahal and Hasegawa 2008). Tectonics and the activity of large active faults can also be considered causative factors on landslide occurrence (Ayalew et al. 2005). In addition, streams could cause landslides by eroding the slope or saturating lower horizons of ground while increasing water level in their channels (Saha et al. 2002, Yaclin 2008, Kelarestaghi and Ahmadi 2009). Human activities such as road construction are included as significant factors on landslide occurrence in mountainous lands (Greco et al. 2007, Lee and Pradhan 2007). Land cover and land use conditions and also their changes during sequential periods are considered effective parameters on landslide occurrence (Dai et al. 2004, Lee and Sambath 2006, Greco et al. 2007). Lithological and structural variations, which affect the physical properties of slope- farming materials such as strength and permeability, widely influence landslide occurrence (Garcia-Rodriguez et al. 2008, Yaclin 2008). Nowadays, considering these causative factors, it is necessary to create landslide susceptibility maps for a region for the sake of effective management of soil, water and natural resources. The necessity of predicting landslide occurrences has led to the development of many empirical and statistical models with emphasis on the use of Geographic Information Systems (GIS). Landslide susceptibility analysis by providing useful information for catastrophic damage reduction can assist us with the development of policies for land-use management. GIS as the basic analysis tool for landslide hazard mapping can be effective for spatial data management and manipulation of the analysis. Landslide susceptibility mapping depends on complicated knowledge of slope movements and their controlling factors. The process of providing a landslide susceptibility map involves numerous quantitative and qualitative models. Some qualitative methods depend on classifying and weighing the causative factors which may evolve into semi-quantitative methods, Analytic Hierarchy Process (AHP) (Kelarestaghi and Garaee 2007, Yaclin 2008) and Weighted Linear Combination (WLC) (Ayalew et al. 2004). Since qualitative or semi-quantitative methods change with an expert’s knowledge, they are often useful for regional studies (Guzzetti et al. 1999). Quantitative methods are based on numerical statements of the correlation between causative factors and landslides. Deterministic models (Gokceoglu and Aksoy 1996) and statistical methods are included in quantitative approaches to make GIS-based spatial prediction of landslide susceptibility 35 indirect hazard mapping. Bivariate statistical analyses (Cevik and Topal 2003, Yaclin 2008, Kelarestaghi and Ahmadi 2009) involve the comparison of landslide inventory map with maps of causative factors to rank their classes based on their roles in landslide occurrence. Multivariate statistical approaches include discriminate analyses (Baeza and Corominas 2001, Carrara et al. 2003), artificial neural network (Gomez and Kavzoglu 2005, Nefeslioglu et al. 2008, Pradhan and Lee 2009a, 2010a,b, Yilmaz 2009), fuzzy logic (Ercanoglu and Gokceoglu 2004, Kanungo et al. 2006, Pradhan et al. 2009), and logistic regression that is the most common statistical method used in earth sciences (Jade and Sarkar 1993, Wieczorek et al. 1996, Guzzetti et al. 1999, Dai et al. 2001, Dai and Lee 2002, Ohlmacher and Davis 2003, Chau et al. 2004a, Ayalew et al. 2005, Can et al. 2005, Chao and Chan 2005, Davis et al. 2006, Duman et al. 2006, Greco et al. 2007, Lee and Pradhan 2007, Nefeslioglu et al. 2008, Pradhan et al. 2008, Yilmaz 2009). Landslide risk analysis is also carried out by Pradhan and Lee (2009b) using artificial neural network approach. Recently, Pradhan (2010) and Pradhan et al. (2010b) have used applied logistic regression, frequency ratio and a neural network model at three different locations in Malaysia. Their results have been used to validate landslide susceptibility mapping. In Iran, numerous efforts have been made to map landslide hazard sites, but new statistical approaches such as logistic regression have not been incorporated. Furthermore, because of extensive land-use changes and road construction in the northern forests of Iran (Kelarestaghi and Jafarian 2010), which makes hill slopes susceptible to landslide occurrence (Kelarestaghi and Garaee 2007), it is extremely necessary to model and map landslide spatial distribution using new quantitative approaches. The present study is carried out by identifying some main physical factors contributing to the landslide occurrence and incorporating them in logistic multiple regression, by which regional slope instability in the Sajarood basin, Northern Iran, was modelled. This involved the identification and mapping of a group of natural and human factors that are directly or indirectly correlated with slope instability. 2. Methodology 2.1 Study area The northern part of Iran, Sajarood watershed in Mazandaran which has suffered from several landslide damages, was chosen as the study area to evaluate landslide distribution and susceptibility (figure 1). In the study area, topographical, geological and anthropogenic characteristics have contributed to the occurrence of landslides (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009). The region is mostly forest land and most of the landslides occurred on cut slopes or embankments alongside the 0 00 roads (Kelarestaghi et al. 2007). The study area is located between 52834 48 – 0 00 0 00 0 00 52842 36 E and 36812 00 –36825 12 N, covering an area of 162.88 km . Geologically, the study area consists mainly of senozoaek and neogen deposits including conglomerate with sandstone and silty marl horizons, mudstone, limy marl and marly limestone that cover large parts of the region. Quaternary sediments include old alluviums and young alluvial fans and debris that are mostly found in river sides. The climate of the study area is humid and semi-humid in summer, mild and dry in winter. Average annual precipitation is 617 mm over the period 1982– 2006. The region is mostly covered by forest with dominant type of Crataegus 36 S.Z. Mousavi et al. Figure 1. Landslide locations in the study area. oxyacantha, Parrotia persica, Acer platanoides, Quercus castanifolia and Fagus orientalis that in lower elevations have been changed to orchard and farmlands. There are many landslides all over the region, which is undergoing significant change due to land-use changes and road construction, especially at lower elevations. There are many long landslides with high depth extension and their longitudinal and latitudinal deformations are less than their depth deformations. 2.2 Landslides inventory Preparing a landslide inventory for a certain region constitutes the first step of data production. In this study, landslides were detected from the interpretation of 1:40,000 scale aerial photographs. These locations were checked and verified by field work (figures 1 and 2). A series of field surveys helped with checking the sizes and shapes of landslides and identification of the types of movements and the materials involved. Field observation revealed that in the study area most landslides are rotational and have occupied large areas alongside the roads. The materials involved in many landslides are a mixture of soil (including marl, silty marl, siltstone, limy siltstone and mudstone), gravels and cobbles. In the forests of northern Iran most of the landslides have occurred at lower elevations and the type of mass movements toward higher elevations are inclined to falls and topples, therefore they are not included in the landslide inventory map. This is partially due to the fact that most residential and rural areas, forest roads and finally sites of human activities are focused on lower parts of this region. A total of 95 landslides were mapped and subsequently digitized and rasterized in Arc/view 3.2a. The study area was divided into a grid with a 306 30 m cell, occupying 583 rows and 310.4 columns, totalling 180,987 grid cells. 2.3 Landslide-influencing parameters In logistic regression, the more independent variables are included, the more complete the model will be, but only when they play a major role in determining the dependent variable (Ayalew et al. 2005). The occurrence of landslides in a special area is a function of direct and indirect natural and human factors. In fact, there are no universal guidelines for selecting landslide-related factors. Generally, it is GIS-based spatial prediction of landslide susceptibility 37 Figure 2. Field photographs of the landslides. acceptable that in a GIS-based study, any selected factor must be operational (it has a certain degree of affinity with the dependent variable), complete (it is fairly represented all over the study area), non-uniform (it varies spatially), measurable (it can be expressed by any of the different types of measuring scales) and non- redundant (its effect should not account for double consequences in the final result). In this research, all factors in relevant to landslides were selected based on study and analysis of LANDSAT ETMþ satellite imagery of the study area and also previous works conducted on forests in northern Iran (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009) considering five mentioned criteria. The factors contributing to the landslide occurrence in the study area were elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology that their maps were digitized and rasterized in Arcview3.2a. First a digital elevation model (DEM) with a resolution of 30 m was created from the 1:50,000 scale topographic map. Based on DEM, slope angle, slope aspect and slope curvature were calculated. The curvature represents the morphology of the topography. A positive curvature indicates that the surface is upwardly convex at that cell and a negative curvature indicates that the surface is upwardly concave at that cell. A value of zero indicates that the surface is flat. By exerting the equation of rainfall gradient with elevation to the DEM and classifying it, the precipitation map was calculated into 9 classes of 40 mm intervals. The distance to drainage and road were calculated using the topographic database. Buffering intervals around digitized streams and roads on a 1:50,000 scale topographical map, distance to road and 38 S.Z. Mousavi et al. drainage maps was created. Using the geology database, the distance to faults was extracted with buffering around digitized faults on 1:100,000 scale geological map. The geology from the digitized 1:100,000 scale geological maps was classified into 12 different geological units related to Terias, Kertase, Tertiary and Quaternary years. Land-use data were classified from a LANDSAT ETMþ image using unsupervised classification method. The land-use map has been classified into five classes such as intense forest, intense range, forest–orchard lands, orchard–agricultural lands and orchards. 2.4 Landslide susceptibility modelling using logistic regression Logistic regression, which is a multivariate analysis model, employs the use of independent variables to predict the probability of a dichotomous event, such as 0 and 1 or true and false (Dai and Lee 2002). In the case of landslide susceptibility mapping, the aim of logistic regression is to find the best-fitting model to describe the relationship of the presence or absence of landslides (dependent variable) with a set of independent parameters such as slope angle, aspect, curvature, lithology and land use (Dai et al. 2001, Ohlmacher and Davis 2003, Ayalew and Yamagishi 2005, Davis et al. 2006, Lee and Pradhan 2007, Nefeslioglu et al. 2008, Pradhan et al. 2008, Yilmaz 2009). Quantitatively, the relationship between the probability of land sliding and the independent variables can be expressed as: p ¼ ð1Þ 1 þ e where p is the probability of land sliding and z is the weighted linear combination of the independent variables z ¼ log itðpÞ¼ Ln ¼ c þ c x þ  þ c x ð2Þ 0 1 1 n n 1  p where p/1 – p is the so-called odds or likelihood ratio. c is the constant of the equation and c , c ... , c are the coefficients of variables x , x ... , x .As z 1 2 n 1 2 n varies from –? to þ? the probability (p) varies from 0 to 1 on an S-shaped curve. The closer predicted probability of any landslide raster point to 1, the more probable a landslide is to occur and the closer probability to 0, the less probable is landslide occurrence. The advantages of logistic regression are that despite making an appropriate link function to the usual linear regression models, this technique does not assume linearity of relationship between the independent variables and does not assume variables having equal statistics variances, and in general has less- stringent requirements. Another advantage is that the variables can be measured in a nominal, ordinal, interval or ratio manner of any continuous or discrete or combination of both types. Since the application of logistic multiple regression is to connect slope-instable factors with a binary dependent variable (presence or absence of landslide), in addition to 95 landslide pixel points, another 95 raster points were selected randomly all over the region to generate non-landslide locations. The value of 1 will be given to the classes of independent variables comprising presence of landslide and the value of 0 to the rest of the classes. The GIS-based spatial prediction of landslide susceptibility 39 elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology of these 190 points are extracted using GIS and input into the forward stepwise logistic regression to perform statistical analyses. Stepwise logistic regression is most often used in situations where the ‘important’ independent variables are not known, and their associations with the outcome are not well understood (Garcı´a-Rodrı´guez et al. 2008). There are two basic forms of stepwise logistic regression: forward inclusion and backward elimination. In forward logistic regression, not all independent variables are included in the model at first and at each step variables are evaluated for entry to model one by one if they contribute sufficiently to the regression equation. Just the opposite occurs in backward logistic regression, in which all independent variables are initially included in the model. In the present analysis, a likelihood ratio test based on the maximum partial likelihood estimates was used for determining whether variables should be added to the model. It involves estimating the model with each variable entered in turn and looking at the change in the logarithm of likelihood when each variable is added. If the observed significance level is less than the probability for remaining in the model (0.05 in this study), the variable is entered into the model and the model statistics are recalculated to see if any other variables are eligible for entry (Dai and Lee 2002, Dai et al. 2004). Finally, it becomes a model excluding all insignificant independent variables and coefficients are allocated to the independent variable classes, correlated with the dependent variable. If a coefficient is positive, its transformed log value will be greater than one, meaning that the event is more likely to occur. If a coefficient is negative, the latter will be less than one and the odds of the event occurring decreases. A coefficient of 0 has a transformed log value of 1, and it does not change the odds one way or the other (Ayalew et al. 2005). After imposing given coefficients to independent variables classes, weighted linear combination (z) was provided. The logistic multiple regression model was then transmitted into the Arc/view GIS and the primary landslide susceptibility map with the probability (p) ranging from 0 to 1 was prepared. This map was then divided into 20 classes with 0.05 equal probability intervals. After overlapping with the landslide inventory map, the histogram representing the frequency of landslide and non-landslide occurrence versus the probability classes was plotted. Finally, based on the histogram, the susceptibility range was subjectively classified into four categories (Dai and Lee 2002, Dai et al. 2004) and a final landslide susceptibility map representing a very low, low, moderate and high probability of landslide occurrence was prepared. In this study, all statistical analyses developing the model have been applied in SPSS 15 software. 2.5 Validation of the susceptibility map A standard analysis in logistic regression to check the validity of the model is called ‘percent correct prediction test’ based on classification plot of 0 and 1 for the observed groups and the predicted probability. In the logistic regression model, the probability of landslide occurrences can be calculated for all raster points of landslide occurrence and non-landslide occurrence, so all data points (both landslide and non-landslide points) can be used as a check for the coherence of the hazard map (Chau and Chan 2005). If the predicted probability of any landslide raster point is larger than 0.5 and for any non-landslide point is less than 0.5, the prediction is 40 S.Z. Mousavi et al. considered to be successful. A classification plot of 0 and 1 is obtained according to the diagram that plots the predicted probability of all landslide and non-landslide points. For a successful model, 0 points should appear close to the lower end of the plot (or p ¼ 0), whereas 1 should appear close to the upper end (or p ¼ 1) (Chau et al. 2004a). Another way to investigate the reliability of the hazard analysis is to consider 2 2 2 Nagelker R , Cox and Snell R and –2Log likelihood factors. Similar to the R in 2 2 linear regression, Cox and Snell R and Nagelkerk R are also correlation coefficients for logistic regression analysis. The theoretical values of these coefficients are from 0 to 1 and a higher value represents more accuracy of the model. For the case of linear regression, R 40.9 is considered a good indicator of a reasonable fit, while in logistic regression these coefficients can be relatively small and do not necessarily invalidate the model (Chau et al. 2004a). –2Log likelihood functions like k and a smaller value of it implies a higher reliability of the model. In this research we have conducted another series of logistic regression analysis by removing each of the causative factors separately and the model is the best accuracy fitted when Nagelkerk R , Cox and Snell R are the largest and –2Log likelihood is the smallest under conditions where all independent variables are included. This is a standard technique in logistic regression analysis to examine the statistical significance of each of the selected independent variables (Hosmer and Lemeshow 2000, Menard 2001). Another way to investigate the reliability of the produced susceptibility map is to compare the known landslide location data with the landslide susceptibility map (a cross-validation technique) and verification would be performed in terms of success rate (Lee 2004, Lee and Sambath 2006). To obtain the relative rank, the landslide susceptibility index values of all cells in the study area were sorted in descending order. Then, using 100 subdivisions of logistic regression values of all cells as the x-axis and cumulative percentage of landslide occurrence in the classes as the y-axis, the success rate curve was drawn (figure 6). According to the curve, a model is considered reliable where the smaller degree of fit was distributed in the low and very low susceptibility classes, and the higher values of degree of fit were scattered in the high and very high susceptibility classes of the landslide susceptibility maps produced by the model. To compare the quantitative result, the areas under the curve (AUC) were recalculated as the total area is 1, which means perfect prediction accuracy. The AUC is a good indicator to check the prediction performance of the model and the largest AUC, varying from 0.5 to 1.0, is the most ideal model. 3. Results 3.1 Model development Using aerial photos interpretation and field work, 95 landslide cases were detected in the study area. A landslide inventory map was produced in Arc/view 3.2.a software (figure 1). The maps of causative factors on landslide occurrence , such as elevation, slope gradient, slope aspect, slope curvature, precipitation, distance to fault, distance to drainage, distance to road, land use and geology were produced based on databases including 1:50,000 scale topographic map, 1:100,000 scale geological map and LANDSAT ETMþ image. The landslide densities computed for the classes of causative factors are shown in figure 3. After the categorical variables were coded, a forward stepwise logistic regression model was constructed based on the independent variables as defined above. Independent variables of elevation, slope curvature, rainfall and distance from fault, GIS-based spatial prediction of landslide susceptibility 41 Figure 3. Landslide density for each class of a variable. (a) Elevation, (b) slope gradient, (c) slope aspect, (d) slope curvature, (e) distance to fault, (f) rainfall, (g) distance to drainage, (h) distance to road, (i) land use, and (j) geology. because of not having statistical correlation at 95% significant level in predicting slope instability, were excluded from the model and coefficients are allocated to the independent variable classes, correlated with the dependent variable (table 1). 42 S.Z. Mousavi et al. Table 1. Coefficient values for logistic regression in the case of each factor. Independent parameter Class Coefficient Significant Slope gradient 0–5872.287 0.000 Slope aspect Northwest 3.628 0.011 Distance to drainage 400–600 1.725 0.007 Distance to road 0–100 4.682 0.000 100–200 2.505 0.000 Land use High-density forest 72.15 0.001 Geology Silty marl, sandstone, siltstone 2.092 0.000 Marl, limey siltstone, silty marl, 1.655 0.008 sandy limestone, mudstone Constant 72.649 0.005 Since the highest positive significant coefficient of the model (b ¼ 4.682) belongs to 0–100 m distance from the road, closeness to the road is introduced as the most significant causative factor on landslide occurrence in this research, so that assuming the rest of the factors are constant, for 1 m closeness to the road, landslide occurrence probability will be e4.682 or 107.983 times. Coefficients of significant variables have been substituted in equation (2) to develop linear combination (z) as: Z ¼2:649 þ 4:682ðÞ D:R þ 2:505ðÞ D:R þ 3:628ðÞ S:A þ 2:092ðÞ G þ 1:655ðÞ G 1 2 9 7 10 þ 1:725ðÞ D:D  2:150ðÞ L:U  2:287ðÞ S:G ð3Þ 3 1 1 Inserting the linear combination (z) into the equation (1), the final equation of logistic regression was developed as: pðlandslideoccurenceÞ ð2:649þ4:682ðD:R1Þþ2:505ðD:R2Þþ3:628ðS:A9Þþ2:092ðG7Þþ1:655ðG10Þþ1:725ðD:D3Þ2:150ðL:U1Þ2:287ðS:G1Þ 1 þ e ð4Þ where p is landslide occurrence probability which ranges from 0 to 1; D.R is the first class of distance to road; D.R is the second class of distance to road; S.A is the ninth class of slope aspect; G is the seventh class of geology; G is the tenth class of 7 10 geology; D.D is the third class of distance to drainage; L.U is the first class of land 3 1 use; S.G is the first class of slope gradient. After providing the primary landslide susceptibility map in Arc/view 3.2a software and dividing it into 20 classes with 0.05 equal probability intervals, the landslide occurrence probability histogram versus the frequency of presence and absence of landslide occurrence was provided (figure 4). Then the range of the probability of land sliding was subjectively classified into four classes: 0–0.05, 0.05–0.25, 0.25–0.40 and 0.40–1, representing a very low, low, moderate and high probability of land sliding, respectively (figure 5). Results showed that 53.01% of the area was located in very low hazard, 18.33% in low hazard, 20.96% in moderate hazard and 7.7% of area is located in high hazard regions. GIS-based spatial prediction of landslide susceptibility 43 Figure 4. Histogram of predicted landslide susceptibility. Figure 5. Landslide susceptibility map of the study area. 3.2 Reliability of the susceptibility map The results of correct prediction test in the form of classification plot of 0 and 1 for the observed points and the predicted probability are given in table 2. The successful rate of prediction for the 95 landslide points is 82.1%, whereas that for the non- landslide reference points is 88.4%. Thus, the total successful rate is 85.3% which is considered acceptable. Figure 6 plots the frequency versus the predicted probability for all landslide (1) and non-landslide (0) points. Non-landslide points appear close to the lower end of the plot (or p ¼ 0), whereas landslide points close to the upper end (or p ¼ 1), indicating that the present logistic regression model is considered quite reliable. Table 3 showed that –2log likelihood is the smallest (141.176) and Cox and 2 2 Snell R and Nagelkerk R coefficients are the largest (0.437 and 0.630, respectively) if all variables of slope gradient, slope aspect, distance to drainage, distance to road, land use and geology are included. It means that the logistic regression model is considered as successful and the selected independent variables by the model are 44 S.Z. Mousavi et al. Table 2. Classification table and statistical performance of the model. Predicted Absence of Presence of Model Observed landslide (0) landslide (1) accuracy (%) Absence of landslide (0) 84 11 88.4 Presence of landslide (1) 17 78 82.1 Overall accuracy (%) 85.3 Figure 6. Success rate curve showing landslide susceptibility index rank (y-axis) occurring in cumulative percent of landslide occurrence (x-axis). Table 3. The comparison of the performance of the logistic regression models for various assumptions of the number of independent variables. 2 2 Independent variables –2log likelihood Cox and Snell R Nagelkerke R All six 141.761 0.437 0.630 No slope gradient 170.494 0.387 0.516 No slope aspect 155.660 0.424 0.565 No distance from drainage 155.757 0.433 0.577 No distance from road 229.475 0.164 0.218 No land use 154.290 0.437 0.582 No geology 148.061 0.405 0.607 significant statistically. Considering the success rate curve (figure 6), the logistic regression model shows high accuracy when compared with known events. For example, 90–100% (10%) class of the study area where the landslide hazard index had a higher rank could account for 43% of all the landslides. In addition, the 80– 100% (20%) class of the study area where the landslide hazard index had a higher rank and could explain 54% of the landslides. Furthermore, the AUC value of 0.7740 represents that the prediction accuracy of the susceptibility map is 77.40% that is highly desirable. GIS-based spatial prediction of landslide susceptibility 45 4. Discussion The statistical analysis of logistic regression has been applied to landslide susceptibility mapping in various studies and its reliability has been confirmed. Brenning (2005) studied different approaches and introduced logistic regression with stepwise variable selection as a proper method for the prediction of landslide susceptibility. Lee and Pradhan (2007) also used a logistic regression model to assess landslide hazard caused by rainfall. The verified results were compared with the results from a probabilistic model. It was proved that a logistic regression model is more successful than a probabilistic model in hazard prediction. In this research, logistic regression is used to prepare a landslide susceptibility map for Sajarood basin, northern Iran using a total of 95 landslide and 95 non-landslide data points. Ten causative factors on landslide occurrence such as elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology were taken as independent parameters. Depending on the independent parameters considered, the landslide inventory map and the statistical approach used, the best predictor parameters and the predicted probability map of a logistic regression can vary considerably. There are no guidelines for choosing predictor factors. In fact they can vary according to the study area characteristics. The first results of the logistic regression were the model statistics and coefficients, which were useful to assess the accuracy of the regression function and the role of parameters on the presence or absence of landslides. Elevation, slope curvature, rainfall and distance to fault were considered insignificant in predicting slope instability and thus excluded from the model. The landslide susceptibility map representing a very low, low, moderate and high probability of landsliding was the second outcome of the regression process incorporating with GIS. According to the map classification, most parts of Sajarood area are found in very low and low susceptible zones. A few land parts in proximity to roads and residential parts are susceptible at moderate and high scale. The low- and mid-altitude elevations of the area facing the roads in forest lands covered by relatively weak rocks and soils such as silty marl, siltstone and mudstone are classified as high or medium susceptible to the process of landslides. As a last step, the reliability of hazard map was confirmed through accuracy assessment approaches. Considering coefficients estimated for the logistic regression (table 1), the ‘closeness to roads’ parameter was found to have the strongest relationship with landslide occurrence. Ayalew et al. (2005) have introduced a ‘proximity to roads’ parameter as the most important factor on landslide occurrence in Kakuda-Yahico, central Japan. They declared that most of the landslides were located in range 0–100 m from roads. Also Lee and Sambath (2006), Greco et al. (2007) and Kelarestaghi and Ahmadi (2009) have emphasized the adverse effect of road construction on landslide occurrence in their studies. Northwest slope aspects, due to having more rainfall and moisture retention and then cutting down normal pressure and shear strength of soils, were introduced as an important interior causative factor on landslides in logistic regression in the study area. Garcia-Rodriguez et al. (2008), in their research on making a susceptibility map using a logistic regression model in El Salvador, illustrated the importance of terrain roughness, especially north- and northwest-facing slopes as a key factor within the model. Table 2 indicates that geological units belonging to the Myosen era consisting mostly of marl, silty marl, siltstone, limy siltstone and mudstone are more susceptible to landslide. Because of having fine particles of clay and marl, after absorbing 46 S.Z. Mousavi et al. moisture, these units soon reach liquid limit and flow through the slope. Can et al. (2004), Lee (2004) and Nefeslioglu et al. (2008) emphasized the causative role of geological units on mass movements in their research. In general in the study area, in addition to natural parameters including ‘slope gradient’, ‘slope aspect’ and ‘closeness to drainage’, human activities played a major and more important role on landslide. Due to improper human activities such as principal road construction in forests causing instability and incoherence in slopes and also land-use changes such as changing natural forests to farmlands specially, in the lower parts of the region, the number of landslides in the region are significantly increasing. Hence, the produced susceptibility map may be accepted as the basis for landslide risk management studies to be applied in the study area. 5. Conclusion Landslide susceptibility mapping has shown a great deal of importance in hazard management and land-use planning. In this manner, there are various GIS-based qualitative and quantitative techniques that are useful for analysing the relationship between landslides and their influence factors. GIS methodology contributes by manipulating data and performing the necessary analysis in a short time very cheaply and it also contributes to the number of different scenarios that can be presented graphically. In this study, GIS-based logistic regression model is used to prepare a susceptibility map for Sajarood basin, northern Iran. Model statistics and coefficients indicate that proximity to roads has the strongest relationship with landslide occurrence, whereas slope gradient and land-use parameters with negative coefficients showed a negative correlation with landslide occurrence. According to the produced landslide susceptibility map, the proportion of medium and high susceptible zones is far smaller than the low and very low counterparts. This is in agreement with the observation that landslides are more common in low- and mid- altitude slopes than in highlands. Nevertheless, this study illustrates that rural and residential regions in low- and mid-altitude slopes of forest lands, northern Iran, are strongly biased toward landslides because of human activities, especially unprincipled road constructions and land-use changes. Therefore, the government must exert prompt laws for avoiding unprincipled uses of forest lands. The results of this study indicate that it is necessary to integrate and combine the GIS and statistics because each has only a few functions when applied separately. Finally, the produced landslide susceptibility map can provide a cheap and comprehensive assessment of the likelihood of future failures, which can be useful to planners and decision-makers during site selection and rebuilding process and future zoning issues. Acknowledgements The authors wish to express their sincere thanks to Iranian National Cartographic Centre (INCC), Iran Meteorological Organization (IMO), National Geographical Organization, Forests, Range and Watershed management Organization (FRWO) for providing various data sets for this research. Also, the remarks made by two anonymous reviewers on the earlier version of the paper have greatly improved the current version. GIS-based spatial prediction of landslide susceptibility 47 References AYALEW, L. and YAMAGISHI, H., 2005, The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda–Yahiko Mountains, central Japan. Geomorphology, 65, pp. 15–31. AYALEW, L., YAMAGISHI, H. and UGAWA, N., 2004, Landslide susceptibility mapping using GIS-based weighted linear combination, the case in Tsugawa area of Agano River, Niigata Prefecture, Japan. Landslides, 1, pp. 73–81. AYALEW, L., YAMAGISHI, H., MARUI, H. and KANNO, T., 2005, Landslides in Sado island of Japan: part II. GIS-based susceptibility mapping with comparisons of results from two methods and verifications. Engineering Geology, 81, pp. 432–445. BAEZA, C. and COROMINAS, J., 2001, Assessment of shallow landslide susceptibility by means of multivariate statistical techniques. Earth Surface Processes and Landforms, 26, pp. 1251–1263. BRENNING, A., 2005, Spatial prediction models for landslide hazards: review, comparison and evaluation. Natural Hazards and Earth System Sciences, 5, pp. 853–862. CAN, T., NEFESLIOGLU, H.A., GOKCEOGLU, C., SONMEZ, H. and DUMAN, Y., 2005, Susceptibility assessments of shallow earth flows triggered by heavy rainfall at three catchments by logistic regression analysis. Geomorphology, 82, pp. 250–271. CARRARA, A., CROSTA, G. and FRATTINI, P., 2003, Geomorphological and historical data in assessing landslide hazard. Earth Surface Processes and Landforms, 28, pp. 1125–1142. CEVIK, E. and TOPAL, T., 2003, GIS-based landslide susceptibility mapping for a problematic segment of the natural gas pipeline, Hendek (Turkey). Environmental Geology, 44, pp. 949–962. CHAU,K.T.and CHAN, J.E., 2005, Regional bias of landslide data in generating susceptibility maps using logistic regression for Hong Kong Island. Landslides, 2, pp. 280–290. CHAU, K.T., SZE, Y.L., FUNG, M.K., WONG, W.Y., FONG, E.L. and CHAN, L.C.P., 2004b, Landslide hazard analysis for Hong Kong using landslide inventory and GIS. Computers & Geosciences, 30, pp. 429–443. CHAU, K.T., TANG, Y.F. and WONG, R.H.C., 2004a, GIS-based rock fall hazard map for Hong Kong. Rock Mechanic, 41(3), pp. 1–6. COROMINAS, J. and MOYA, J., 1999, Reconstructing recent landslide activity in relation to rainfall in the Llobregat river basin, Eastern Pyrenees, Spain. Geomorphology, 30, pp. 79–93. CROSTA, G., 1998, Regionalization of rainfall thresholds: an aid to landslide hazard evaluation. Environmental Geology, 35(2–3), pp. 131–145. DAHAL, R.K. and HASEGAWA, S.H., 2008, Representative rainfall thresholds for landslides in the Nepal Himalaya. Geomorphology, 100, pp. 429–443. DAI, F.C. and LEE, C.F., 2002, Landslide characteristics and slope instability modelling using GIS, Lantau Island, Hong Kong. Geomorphology, 42, pp. 213–228. DAI, F.C., LEE, C.F., LI, J. and XU, Z.W., 2001, Assessment of landslide susceptibility on the natural terrain of Lantau Island, Hong Kong. Engineering Geology, 40, pp. 381–391. DAI, F.C., LEE, C.F., THAM, L.G., NG, K.C. and SHUM, W.L., 2004, Logistic regression modelling of storm-induced shallow land sliding in time and space on lantau island, Hong Kong. Bulletin of Engineering Geology and Environment, 63, pp. 315–327. DAVIS, J.C., CHANG, C.H.J. and OHLMACHER, G.C., 2006, Two models for evaluating landslide hazards. Computers & Geosciences, 32, pp. 1120–1127. DOMINGUEZ-CUESTA, M., JIMENEZ-SONCHEZ, M. and BERREZUETA, E., 2007, Landslide in the central coalfield (Cantabarian Mountains, NW Spain): geomorphological feature conditioning factors and meteorological implication in susceptibility assessment. Geomorphology, 89, pp. 1–12. DUMAN, T.Y., CAN, T., GOKCEOGLU, C., NEFESLIOGLU, A.H. and SONMEZ, H., 2006, Application of logistic regression for landslide susceptibility zoning of Cekmece Area, Istanbul, Turkey. Environmental Geology, 51, pp. 241–256. 48 S.Z. Mousavi et al. ERCANOGLU, M. and GOKCEOGLU, C., 2004, Use of fuzzy relations to produce landslide susceptibility map of a landslide prone area (West Black Sea Region, Turkey). Engineering Geology, 75, pp. 229–250. GARCIA-RODRIGUEZ, M.J., MALPICA, J.A., BENITO, B. and DIAZ, M., 2008, Susceptibility assessment of earthquake-triggered landslides in El Salvador using logistic regression. Geomorphology, 95, 172–191. GARFI, G. and BRUNO, D.E., 2007, Fan morphodynamics and slope instability in the Mucone River basin (Sila Massif, southern Italy): significance of weathering and role of land use changes. Catena, 50, pp. 181–196. GOKCEOGLU, C. and AKSOY, H., 1996, Landslide susceptibility mapping of the slopes in the residual soils of the Mengen region (Turkey) by deterministic stability analyses and image processing technique. Engineering Geology, 44, pp. 147–161. GOMEZ, H. and KAVZOGLU, T., 2005, Assessment of shallow landslide susceptibility using artificial neural networks in Jabonosa River Basin, Venezuela. Engineering Geology, 78, pp. 11–27. GRECO, R., SORRISO, V. and CATALANO, E., 2007, Logistic regression analysis in the evaluation of mass movement’s susceptibility case study: Calabria, Italy. Engineering Geology, 89, pp. 47–66. GUZZETTI, F., CARRARA, A., CARDINALI, M. and REICHENBACH, P., 1999, Landslide hazard evaluation: a review of current techniques and their application in a multi-scale study, Central Italy. Geomorphology, 31, pp. 181–216. HOSMER, D.W. and LEMESHOW, S. (Eds.), 2000, Applied Logistic Regression, 2nd edition, 373 pp (New York, NY: Wiley). JADDA, M., SHAFRI, H.M., MANSOR, S.H., SHARIFIKIA, M. and PIRASTEH, S., 2009, Landslide susceptibility evaluation and factor effect analysis using probabilistic-frequency ratio model. European Journal of Scientific Research, 33, pp. 654–668. JADE, S. and SARKAR, S., 1993, Statistical models for slope stability classification. Engineering Geology, 36, pp. 91–98. KANUNGO, D.P., ARORA, M.K., SARKAR, S. and GUPTA, R.P., 2006, A comparative study of conventional, ANN black box, fuzzy and combined neural and fuzzy weighting procedures for landslide susceptibility zonation in Darjeeling Himalayas. Engineering Geology, 85, pp. 347–366. KELARESTAGHI, A. and AHMADI, H., 2009, Landslide susceptibility analysis with a bivariate approach and GIS in northern Iran. Arabian Journal of Geosciences, 2, pp. 95–101. KELARESTAGHI, A. and GARAEE, P., 2007, Landslide hazard mapping using AHP technique in Vastan basin. Geographical Researches, 87(4), pp. 49–68 [in Persian]. KELARESTAGHI, A., HABIBNEJAD, M., AHMADI, H., 2007, Assessment of landslide occurrence related to landuse change and road construction. Geographical Researches, 62, pp. 81– 91 [in Persian]. KELARESTAGHI, A. and JAFARIAN, Z., 2010, Land use/cover change and driving force analyses in parts of northern Iran using RS and GIS techniques. Arabian Journal of Geosciences. DOI: 10.1007/s12517-009-0078-5. LEE, S., 2004, Application of likelihood ratio and logistic regression models to landslide susceptibility mapping using GIS. Environmental Management, 34, pp. 223–232. LEE, S. and PRADHAN, B., 2007, Landslide hazard mapping at Selangor Malaysia using frequency ratio and logistic regression models. Landslides, 4, pp. 33–41. LEE, S. and SAMBATH, T., 2006, Landslide susceptibility mapping in the Damrei Romel area, Cambodia using frequency ratio and logistic regression models. Environmental Geology, 50, pp. 847–855. MENARD, S.W. (Ed.), 2001, Applied Logistic Regression Analysis, 2nd edition, 111 pp (Thousand Oaks, CA: Sage). GIS-based spatial prediction of landslide susceptibility 49 MOUSAVI KHATIR, S.Z., KELARESTAGHI, A. and HASHEMZADEH ATOEI, A., 2009, Statistical analysis of some morphometric characteristics and effective factors on slope instability in parts of Babolrood watershed. Journal of Water and Soil Conservation, 16(2), 85– 103 [in Persian]. NEFESLIOGLU, H.A., GOKCEOGLU, C. and SONMEZ, H., 2008, An assessment on the use of logistic regression and artificial neural networks with different sampling strategies for the preparation of landslide susceptibility maps. Engineering Geology, 97, pp. 171–191. OHLMACHER, G.C. and DAVIS, J.C., 2003, Using multiple logistic regression and GIS technology to predict landslide hazard in northeast Kansas USA. Engineering Geology, 69, pp. 331–343. PEART, M.R., NG, K.Y. and ZHANG, D.D., 2005, Landslides and sediment delivery to a drainage system: some observations from Hong Kong. Asian Earth Sciences, 25, pp. 821–836. PRADHAN, B., 2010, Remote sensing and GIS-based landslide hazard analysis and cross- validation using multivariate logistic regression model on three test areas in Malaysia. Advances in Space Research, 45, pp. 1244–1256. PRADHAN, B. and LEE, S., 2009a, Delineation of landslide hazard areas using frequency ratio, logistic regression and artificial neural network model at Penang Island, Malaysia. Environmental Earth Science. doi:10.1007/s12665-009-0245-8. PRADHAN, B. and LEE, S., 2009b, Landslide risk analysis using artificial neural network model focusing on different training sites. International Journal of Physical Sciences, 3(11), pp. 1–15. PRADHAN, B. and LEE, S., 2010a, Regional landslide susceptibility analysis using back pro- pagation neural network model at Cameron Highland, Malaysia. Landslides, 7, pp. 13–30. PRADHAN, B. and LEE, S., 2010b, Landslide susceptibility assessment and factor effect analysis: backpropagation artificial neural networks and their comparison with frequency ratio and bivariate logistic regression modeling. Environmental Modelling and Software, 25, pp. 747–759. PRADHAN, B., LEE, S. and BUCHOITHNER, M.F., 2009, Use of geospatial data for the development of fuzzy algebraic operators to landslide hazard mapping: a case study in Malaysia. Applied Geomatics, 1, pp. 3–15. PRADHAN, B., LEE, S. and BUCHROITHNER, MF., 2010a, Remote sensing and GIS-based landslide susceptibility analysis and its cross-validation in three test areas using a frequency ratio model. Photogrammetrie and Fernerkun, 1, pp. 17–32. PRADHAN, B., LEE, S. and BUCHROITHNER, MF., 2010b, A GIS-based back-propagation neural network model and its cross application and validation for landslide susceptibility analyses. Computers, Environment and Urban Systems, 34, pp. 216–235. PRADHAN, B., LEE, S., MANSOR, S., BUCHROITHNER, MF., JALLALUDDIN, N. and KHUJAIMAH, Z., 2008, Utilization of optical remote sensing data and geographic information system tools for regional landslide hazard analysis by using binomial logistic regression model. Applied Remote Sensing, 2, pp. 1–11. SAHA, A.K., GUPTA, R.P. and ARORA, M.K., 2002, GIS-based landslide hazard zonation in the Bhagirathi (Ganga) valley, Himalayas. International Journal of Remote Sensing, 23, pp. 357–369. SOETERS, R. and VAN WESTEN, C.J., 1996, Slope instability recognition, analysis and zonation. In A.K. Turner and R.L. Schuster (Eds.), Landslides Investigation and Mitigation, Transportation Research Board, National Research Council, Special Report 247, pp. 129–177 (Washington, DC: National Academy Press). WIECZOREK, G.F., GORI, P.L., JAGER, S., KAPPEL, W.M. and NEGUSSEY, D., 1996, Assessment and management of landslide hazards near Tully Valley landslide, Syracuse, New York, USA. In Proceeding of 7th International Symposium on Landslides, Trondheim, pp. 411–416 (Rotterdam: Balkema). 50 S.Z. Mousavi et al. YACLIN, A., 2008, GIS-based landslide susceptibility mapping using analytical hierarchy process and bivariate statistics in Ardesen (Turkey): comparisons of results and confirmations. Catena, 72, pp. 1–12. YILMAZ, I., 2009, Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: a case study from Kat landslides (Tokat, Turkey). Computers and Geoscience, 35, pp. 1125–1138. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png "Geomatics, Natural Hazards and Risk" Taylor & Francis

GIS-based spatial prediction of landslide susceptibility using logistic regression model

Download PDF
18 pages

Loading next page...
 
/lp/taylor-francis/gis-based-spatial-prediction-of-landslide-susceptibility-using-GgYcfuzw9t

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Taylor & Francis
Copyright
Copyright Taylor & Francis Group, LLC
ISSN
1947-5713
eISSN
1947-5705
DOI
10.1080/19475705.2010.532975
Publisher site
See Article on Publisher Site

Abstract

Geomatics, Natural Hazards and Risk Vol. 2, No. 1, March 2011, 33–50 GIS-based spatial prediction of landslide susceptibility using logistic regression model SEYEDEH ZOHREH MOUSAVI{, ATAOLLAH KAVIAN*{, KARIM SOLEIMANI{, SEYED RAMEZAN MOUSAVI{, and ATAOLLAH SHIRZADI{ {College of Natural Resources, Sari Agricultural Sciences and Natural Resources University, Sari, Iran {College of Natural Resources, University of Kurdistan, Sanandaj, Iran (Received 1 August 2010; in final form 17 August 2010) In the present study, logistic regression analysis has been used to create a landslide hazard map for Sajarood basin, Northern Iran. At first, an inventory map of 95 landslides was used to produce a dependent variable, a value of 0 for absence and 1 for presence of landslides. The effect of causative parameters on landslide occurrence was assessed by the corresponding coefficient that appears in the logistic regression function. The interpretation of the coefficients shows that the road network plays the major role in determining landslide occurrence. Elevation, slope curvature, rainfall and distance to fault were excluded from the final analysis, because these variables do not significantly add to the predictive power of the logistic regression. After running the final probability function into Arc/view 3.2 software, a landslide susceptibility map has been produced. The accuracy assessment shows an overall accuracy of the landslide susceptibility map to be 85.3%. An area of 53.01% is found to be located in a very low, 18.33% in low, 20.96% in moderate and 7.7% in high-risk regions. The proposed 2 2 susceptibility map was tested using –2LL, Cox and Snell R , Nagelkerk R and Roc procedure, and it is found to be very reliable. 1. Introduction Landslides are one of the most destructive natural hazards in mountainous areas and play an important role in landform evolution and cause serious hazards in the world. Landslides occur when unstable masses of soil and stone are affected by earthquakes, heavy rainfall and various human activities. Damage to residential regions, economic losses and human fatalities caused by landslides are increasing worldwide. Landslides represented approximately 9% of the natural disasters that occurred worldwide during the 1990s (Gomez and Kavzoglu 2005, Yilmaz 2009). Because of mainly mountainous topography, tectonic and seismic activities, various geological and climatic conditions, population increase and land-use change, numerous landslides have occurred in Iran in recent decades (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009, Mousavi et al. 2009). Every year, landslides cause economic *Corresponding author. Email: a.kavian@sanru.ac.ir Geomatics, Natural Hazards and Risk ISSN 1947-5705 Print/ISSN 1947-5713 Online ª 2011 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/19475705.2010.532975 34 S.Z. Mousavi et al. damage to roads, railway tracks, power lines, irrigation channels, facilities of gas petroleum extraction and refinery factories and industrial centres, dams and natural or artificial reservoirs, forests, range lands, farms and villages in many provinces in Iran. In February 1998, one of the biggest wayside landslides on Emamzade Ali- Abask region of Haraz road in Mazandaran, Iran occurred, resulting in damage to more than 600 m of road, the whole village, many restaurants, several power lines and 10 fish production ponds (National Geosciences Database of Iran, htpp:// www.ngdir.ir). Several factors, such as geology, hydrology, hydrogeology, topo- graphy and morphology, climate and weathering affect slope instability and cause landslides (Soeters and van Westen 1996, Chau et al. 2004b, Peart et al. 2005, Domı´nguez-Cuesta et al. 2007, Garfi et al. 2007). Elevation, slope angle and slope aspect are the main topographic factors in landslide occurrence (Dai and Lee 2002, Ohlamcher and Davis 2003, Chau and Chan 2005, Nefeslioglo et al. 2008). Slope curvature is another causative factor on mass movement that a few researchers such as Can et al. (2005), Lee and Sambath (2006) and Greco et al. (2007) have considered the influence of on mass movements. On the other hand, significant detachments in slopes occur after heavy or long-term rainfall and water infiltration into the cracks (Crosta 1998, Corominas and Moya 1999, Ayalew et al. 2005, Dahal and Hasegawa 2008). Tectonics and the activity of large active faults can also be considered causative factors on landslide occurrence (Ayalew et al. 2005). In addition, streams could cause landslides by eroding the slope or saturating lower horizons of ground while increasing water level in their channels (Saha et al. 2002, Yaclin 2008, Kelarestaghi and Ahmadi 2009). Human activities such as road construction are included as significant factors on landslide occurrence in mountainous lands (Greco et al. 2007, Lee and Pradhan 2007). Land cover and land use conditions and also their changes during sequential periods are considered effective parameters on landslide occurrence (Dai et al. 2004, Lee and Sambath 2006, Greco et al. 2007). Lithological and structural variations, which affect the physical properties of slope- farming materials such as strength and permeability, widely influence landslide occurrence (Garcia-Rodriguez et al. 2008, Yaclin 2008). Nowadays, considering these causative factors, it is necessary to create landslide susceptibility maps for a region for the sake of effective management of soil, water and natural resources. The necessity of predicting landslide occurrences has led to the development of many empirical and statistical models with emphasis on the use of Geographic Information Systems (GIS). Landslide susceptibility analysis by providing useful information for catastrophic damage reduction can assist us with the development of policies for land-use management. GIS as the basic analysis tool for landslide hazard mapping can be effective for spatial data management and manipulation of the analysis. Landslide susceptibility mapping depends on complicated knowledge of slope movements and their controlling factors. The process of providing a landslide susceptibility map involves numerous quantitative and qualitative models. Some qualitative methods depend on classifying and weighing the causative factors which may evolve into semi-quantitative methods, Analytic Hierarchy Process (AHP) (Kelarestaghi and Garaee 2007, Yaclin 2008) and Weighted Linear Combination (WLC) (Ayalew et al. 2004). Since qualitative or semi-quantitative methods change with an expert’s knowledge, they are often useful for regional studies (Guzzetti et al. 1999). Quantitative methods are based on numerical statements of the correlation between causative factors and landslides. Deterministic models (Gokceoglu and Aksoy 1996) and statistical methods are included in quantitative approaches to make GIS-based spatial prediction of landslide susceptibility 35 indirect hazard mapping. Bivariate statistical analyses (Cevik and Topal 2003, Yaclin 2008, Kelarestaghi and Ahmadi 2009) involve the comparison of landslide inventory map with maps of causative factors to rank their classes based on their roles in landslide occurrence. Multivariate statistical approaches include discriminate analyses (Baeza and Corominas 2001, Carrara et al. 2003), artificial neural network (Gomez and Kavzoglu 2005, Nefeslioglu et al. 2008, Pradhan and Lee 2009a, 2010a,b, Yilmaz 2009), fuzzy logic (Ercanoglu and Gokceoglu 2004, Kanungo et al. 2006, Pradhan et al. 2009), and logistic regression that is the most common statistical method used in earth sciences (Jade and Sarkar 1993, Wieczorek et al. 1996, Guzzetti et al. 1999, Dai et al. 2001, Dai and Lee 2002, Ohlmacher and Davis 2003, Chau et al. 2004a, Ayalew et al. 2005, Can et al. 2005, Chao and Chan 2005, Davis et al. 2006, Duman et al. 2006, Greco et al. 2007, Lee and Pradhan 2007, Nefeslioglu et al. 2008, Pradhan et al. 2008, Yilmaz 2009). Landslide risk analysis is also carried out by Pradhan and Lee (2009b) using artificial neural network approach. Recently, Pradhan (2010) and Pradhan et al. (2010b) have used applied logistic regression, frequency ratio and a neural network model at three different locations in Malaysia. Their results have been used to validate landslide susceptibility mapping. In Iran, numerous efforts have been made to map landslide hazard sites, but new statistical approaches such as logistic regression have not been incorporated. Furthermore, because of extensive land-use changes and road construction in the northern forests of Iran (Kelarestaghi and Jafarian 2010), which makes hill slopes susceptible to landslide occurrence (Kelarestaghi and Garaee 2007), it is extremely necessary to model and map landslide spatial distribution using new quantitative approaches. The present study is carried out by identifying some main physical factors contributing to the landslide occurrence and incorporating them in logistic multiple regression, by which regional slope instability in the Sajarood basin, Northern Iran, was modelled. This involved the identification and mapping of a group of natural and human factors that are directly or indirectly correlated with slope instability. 2. Methodology 2.1 Study area The northern part of Iran, Sajarood watershed in Mazandaran which has suffered from several landslide damages, was chosen as the study area to evaluate landslide distribution and susceptibility (figure 1). In the study area, topographical, geological and anthropogenic characteristics have contributed to the occurrence of landslides (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009). The region is mostly forest land and most of the landslides occurred on cut slopes or embankments alongside the 0 00 roads (Kelarestaghi et al. 2007). The study area is located between 52834 48 – 0 00 0 00 0 00 52842 36 E and 36812 00 –36825 12 N, covering an area of 162.88 km . Geologically, the study area consists mainly of senozoaek and neogen deposits including conglomerate with sandstone and silty marl horizons, mudstone, limy marl and marly limestone that cover large parts of the region. Quaternary sediments include old alluviums and young alluvial fans and debris that are mostly found in river sides. The climate of the study area is humid and semi-humid in summer, mild and dry in winter. Average annual precipitation is 617 mm over the period 1982– 2006. The region is mostly covered by forest with dominant type of Crataegus 36 S.Z. Mousavi et al. Figure 1. Landslide locations in the study area. oxyacantha, Parrotia persica, Acer platanoides, Quercus castanifolia and Fagus orientalis that in lower elevations have been changed to orchard and farmlands. There are many landslides all over the region, which is undergoing significant change due to land-use changes and road construction, especially at lower elevations. There are many long landslides with high depth extension and their longitudinal and latitudinal deformations are less than their depth deformations. 2.2 Landslides inventory Preparing a landslide inventory for a certain region constitutes the first step of data production. In this study, landslides were detected from the interpretation of 1:40,000 scale aerial photographs. These locations were checked and verified by field work (figures 1 and 2). A series of field surveys helped with checking the sizes and shapes of landslides and identification of the types of movements and the materials involved. Field observation revealed that in the study area most landslides are rotational and have occupied large areas alongside the roads. The materials involved in many landslides are a mixture of soil (including marl, silty marl, siltstone, limy siltstone and mudstone), gravels and cobbles. In the forests of northern Iran most of the landslides have occurred at lower elevations and the type of mass movements toward higher elevations are inclined to falls and topples, therefore they are not included in the landslide inventory map. This is partially due to the fact that most residential and rural areas, forest roads and finally sites of human activities are focused on lower parts of this region. A total of 95 landslides were mapped and subsequently digitized and rasterized in Arc/view 3.2a. The study area was divided into a grid with a 306 30 m cell, occupying 583 rows and 310.4 columns, totalling 180,987 grid cells. 2.3 Landslide-influencing parameters In logistic regression, the more independent variables are included, the more complete the model will be, but only when they play a major role in determining the dependent variable (Ayalew et al. 2005). The occurrence of landslides in a special area is a function of direct and indirect natural and human factors. In fact, there are no universal guidelines for selecting landslide-related factors. Generally, it is GIS-based spatial prediction of landslide susceptibility 37 Figure 2. Field photographs of the landslides. acceptable that in a GIS-based study, any selected factor must be operational (it has a certain degree of affinity with the dependent variable), complete (it is fairly represented all over the study area), non-uniform (it varies spatially), measurable (it can be expressed by any of the different types of measuring scales) and non- redundant (its effect should not account for double consequences in the final result). In this research, all factors in relevant to landslides were selected based on study and analysis of LANDSAT ETMþ satellite imagery of the study area and also previous works conducted on forests in northern Iran (Jadda et al. 2009, Kelarestaghi and Ahmadi 2009) considering five mentioned criteria. The factors contributing to the landslide occurrence in the study area were elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology that their maps were digitized and rasterized in Arcview3.2a. First a digital elevation model (DEM) with a resolution of 30 m was created from the 1:50,000 scale topographic map. Based on DEM, slope angle, slope aspect and slope curvature were calculated. The curvature represents the morphology of the topography. A positive curvature indicates that the surface is upwardly convex at that cell and a negative curvature indicates that the surface is upwardly concave at that cell. A value of zero indicates that the surface is flat. By exerting the equation of rainfall gradient with elevation to the DEM and classifying it, the precipitation map was calculated into 9 classes of 40 mm intervals. The distance to drainage and road were calculated using the topographic database. Buffering intervals around digitized streams and roads on a 1:50,000 scale topographical map, distance to road and 38 S.Z. Mousavi et al. drainage maps was created. Using the geology database, the distance to faults was extracted with buffering around digitized faults on 1:100,000 scale geological map. The geology from the digitized 1:100,000 scale geological maps was classified into 12 different geological units related to Terias, Kertase, Tertiary and Quaternary years. Land-use data were classified from a LANDSAT ETMþ image using unsupervised classification method. The land-use map has been classified into five classes such as intense forest, intense range, forest–orchard lands, orchard–agricultural lands and orchards. 2.4 Landslide susceptibility modelling using logistic regression Logistic regression, which is a multivariate analysis model, employs the use of independent variables to predict the probability of a dichotomous event, such as 0 and 1 or true and false (Dai and Lee 2002). In the case of landslide susceptibility mapping, the aim of logistic regression is to find the best-fitting model to describe the relationship of the presence or absence of landslides (dependent variable) with a set of independent parameters such as slope angle, aspect, curvature, lithology and land use (Dai et al. 2001, Ohlmacher and Davis 2003, Ayalew and Yamagishi 2005, Davis et al. 2006, Lee and Pradhan 2007, Nefeslioglu et al. 2008, Pradhan et al. 2008, Yilmaz 2009). Quantitatively, the relationship between the probability of land sliding and the independent variables can be expressed as: p ¼ ð1Þ 1 þ e where p is the probability of land sliding and z is the weighted linear combination of the independent variables z ¼ log itðpÞ¼ Ln ¼ c þ c x þ  þ c x ð2Þ 0 1 1 n n 1  p where p/1 – p is the so-called odds or likelihood ratio. c is the constant of the equation and c , c ... , c are the coefficients of variables x , x ... , x .As z 1 2 n 1 2 n varies from –? to þ? the probability (p) varies from 0 to 1 on an S-shaped curve. The closer predicted probability of any landslide raster point to 1, the more probable a landslide is to occur and the closer probability to 0, the less probable is landslide occurrence. The advantages of logistic regression are that despite making an appropriate link function to the usual linear regression models, this technique does not assume linearity of relationship between the independent variables and does not assume variables having equal statistics variances, and in general has less- stringent requirements. Another advantage is that the variables can be measured in a nominal, ordinal, interval or ratio manner of any continuous or discrete or combination of both types. Since the application of logistic multiple regression is to connect slope-instable factors with a binary dependent variable (presence or absence of landslide), in addition to 95 landslide pixel points, another 95 raster points were selected randomly all over the region to generate non-landslide locations. The value of 1 will be given to the classes of independent variables comprising presence of landslide and the value of 0 to the rest of the classes. The GIS-based spatial prediction of landslide susceptibility 39 elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology of these 190 points are extracted using GIS and input into the forward stepwise logistic regression to perform statistical analyses. Stepwise logistic regression is most often used in situations where the ‘important’ independent variables are not known, and their associations with the outcome are not well understood (Garcı´a-Rodrı´guez et al. 2008). There are two basic forms of stepwise logistic regression: forward inclusion and backward elimination. In forward logistic regression, not all independent variables are included in the model at first and at each step variables are evaluated for entry to model one by one if they contribute sufficiently to the regression equation. Just the opposite occurs in backward logistic regression, in which all independent variables are initially included in the model. In the present analysis, a likelihood ratio test based on the maximum partial likelihood estimates was used for determining whether variables should be added to the model. It involves estimating the model with each variable entered in turn and looking at the change in the logarithm of likelihood when each variable is added. If the observed significance level is less than the probability for remaining in the model (0.05 in this study), the variable is entered into the model and the model statistics are recalculated to see if any other variables are eligible for entry (Dai and Lee 2002, Dai et al. 2004). Finally, it becomes a model excluding all insignificant independent variables and coefficients are allocated to the independent variable classes, correlated with the dependent variable. If a coefficient is positive, its transformed log value will be greater than one, meaning that the event is more likely to occur. If a coefficient is negative, the latter will be less than one and the odds of the event occurring decreases. A coefficient of 0 has a transformed log value of 1, and it does not change the odds one way or the other (Ayalew et al. 2005). After imposing given coefficients to independent variables classes, weighted linear combination (z) was provided. The logistic multiple regression model was then transmitted into the Arc/view GIS and the primary landslide susceptibility map with the probability (p) ranging from 0 to 1 was prepared. This map was then divided into 20 classes with 0.05 equal probability intervals. After overlapping with the landslide inventory map, the histogram representing the frequency of landslide and non-landslide occurrence versus the probability classes was plotted. Finally, based on the histogram, the susceptibility range was subjectively classified into four categories (Dai and Lee 2002, Dai et al. 2004) and a final landslide susceptibility map representing a very low, low, moderate and high probability of landslide occurrence was prepared. In this study, all statistical analyses developing the model have been applied in SPSS 15 software. 2.5 Validation of the susceptibility map A standard analysis in logistic regression to check the validity of the model is called ‘percent correct prediction test’ based on classification plot of 0 and 1 for the observed groups and the predicted probability. In the logistic regression model, the probability of landslide occurrences can be calculated for all raster points of landslide occurrence and non-landslide occurrence, so all data points (both landslide and non-landslide points) can be used as a check for the coherence of the hazard map (Chau and Chan 2005). If the predicted probability of any landslide raster point is larger than 0.5 and for any non-landslide point is less than 0.5, the prediction is 40 S.Z. Mousavi et al. considered to be successful. A classification plot of 0 and 1 is obtained according to the diagram that plots the predicted probability of all landslide and non-landslide points. For a successful model, 0 points should appear close to the lower end of the plot (or p ¼ 0), whereas 1 should appear close to the upper end (or p ¼ 1) (Chau et al. 2004a). Another way to investigate the reliability of the hazard analysis is to consider 2 2 2 Nagelker R , Cox and Snell R and –2Log likelihood factors. Similar to the R in 2 2 linear regression, Cox and Snell R and Nagelkerk R are also correlation coefficients for logistic regression analysis. The theoretical values of these coefficients are from 0 to 1 and a higher value represents more accuracy of the model. For the case of linear regression, R 40.9 is considered a good indicator of a reasonable fit, while in logistic regression these coefficients can be relatively small and do not necessarily invalidate the model (Chau et al. 2004a). –2Log likelihood functions like k and a smaller value of it implies a higher reliability of the model. In this research we have conducted another series of logistic regression analysis by removing each of the causative factors separately and the model is the best accuracy fitted when Nagelkerk R , Cox and Snell R are the largest and –2Log likelihood is the smallest under conditions where all independent variables are included. This is a standard technique in logistic regression analysis to examine the statistical significance of each of the selected independent variables (Hosmer and Lemeshow 2000, Menard 2001). Another way to investigate the reliability of the produced susceptibility map is to compare the known landslide location data with the landslide susceptibility map (a cross-validation technique) and verification would be performed in terms of success rate (Lee 2004, Lee and Sambath 2006). To obtain the relative rank, the landslide susceptibility index values of all cells in the study area were sorted in descending order. Then, using 100 subdivisions of logistic regression values of all cells as the x-axis and cumulative percentage of landslide occurrence in the classes as the y-axis, the success rate curve was drawn (figure 6). According to the curve, a model is considered reliable where the smaller degree of fit was distributed in the low and very low susceptibility classes, and the higher values of degree of fit were scattered in the high and very high susceptibility classes of the landslide susceptibility maps produced by the model. To compare the quantitative result, the areas under the curve (AUC) were recalculated as the total area is 1, which means perfect prediction accuracy. The AUC is a good indicator to check the prediction performance of the model and the largest AUC, varying from 0.5 to 1.0, is the most ideal model. 3. Results 3.1 Model development Using aerial photos interpretation and field work, 95 landslide cases were detected in the study area. A landslide inventory map was produced in Arc/view 3.2.a software (figure 1). The maps of causative factors on landslide occurrence , such as elevation, slope gradient, slope aspect, slope curvature, precipitation, distance to fault, distance to drainage, distance to road, land use and geology were produced based on databases including 1:50,000 scale topographic map, 1:100,000 scale geological map and LANDSAT ETMþ image. The landslide densities computed for the classes of causative factors are shown in figure 3. After the categorical variables were coded, a forward stepwise logistic regression model was constructed based on the independent variables as defined above. Independent variables of elevation, slope curvature, rainfall and distance from fault, GIS-based spatial prediction of landslide susceptibility 41 Figure 3. Landslide density for each class of a variable. (a) Elevation, (b) slope gradient, (c) slope aspect, (d) slope curvature, (e) distance to fault, (f) rainfall, (g) distance to drainage, (h) distance to road, (i) land use, and (j) geology. because of not having statistical correlation at 95% significant level in predicting slope instability, were excluded from the model and coefficients are allocated to the independent variable classes, correlated with the dependent variable (table 1). 42 S.Z. Mousavi et al. Table 1. Coefficient values for logistic regression in the case of each factor. Independent parameter Class Coefficient Significant Slope gradient 0–5872.287 0.000 Slope aspect Northwest 3.628 0.011 Distance to drainage 400–600 1.725 0.007 Distance to road 0–100 4.682 0.000 100–200 2.505 0.000 Land use High-density forest 72.15 0.001 Geology Silty marl, sandstone, siltstone 2.092 0.000 Marl, limey siltstone, silty marl, 1.655 0.008 sandy limestone, mudstone Constant 72.649 0.005 Since the highest positive significant coefficient of the model (b ¼ 4.682) belongs to 0–100 m distance from the road, closeness to the road is introduced as the most significant causative factor on landslide occurrence in this research, so that assuming the rest of the factors are constant, for 1 m closeness to the road, landslide occurrence probability will be e4.682 or 107.983 times. Coefficients of significant variables have been substituted in equation (2) to develop linear combination (z) as: Z ¼2:649 þ 4:682ðÞ D:R þ 2:505ðÞ D:R þ 3:628ðÞ S:A þ 2:092ðÞ G þ 1:655ðÞ G 1 2 9 7 10 þ 1:725ðÞ D:D  2:150ðÞ L:U  2:287ðÞ S:G ð3Þ 3 1 1 Inserting the linear combination (z) into the equation (1), the final equation of logistic regression was developed as: pðlandslideoccurenceÞ ð2:649þ4:682ðD:R1Þþ2:505ðD:R2Þþ3:628ðS:A9Þþ2:092ðG7Þþ1:655ðG10Þþ1:725ðD:D3Þ2:150ðL:U1Þ2:287ðS:G1Þ 1 þ e ð4Þ where p is landslide occurrence probability which ranges from 0 to 1; D.R is the first class of distance to road; D.R is the second class of distance to road; S.A is the ninth class of slope aspect; G is the seventh class of geology; G is the tenth class of 7 10 geology; D.D is the third class of distance to drainage; L.U is the first class of land 3 1 use; S.G is the first class of slope gradient. After providing the primary landslide susceptibility map in Arc/view 3.2a software and dividing it into 20 classes with 0.05 equal probability intervals, the landslide occurrence probability histogram versus the frequency of presence and absence of landslide occurrence was provided (figure 4). Then the range of the probability of land sliding was subjectively classified into four classes: 0–0.05, 0.05–0.25, 0.25–0.40 and 0.40–1, representing a very low, low, moderate and high probability of land sliding, respectively (figure 5). Results showed that 53.01% of the area was located in very low hazard, 18.33% in low hazard, 20.96% in moderate hazard and 7.7% of area is located in high hazard regions. GIS-based spatial prediction of landslide susceptibility 43 Figure 4. Histogram of predicted landslide susceptibility. Figure 5. Landslide susceptibility map of the study area. 3.2 Reliability of the susceptibility map The results of correct prediction test in the form of classification plot of 0 and 1 for the observed points and the predicted probability are given in table 2. The successful rate of prediction for the 95 landslide points is 82.1%, whereas that for the non- landslide reference points is 88.4%. Thus, the total successful rate is 85.3% which is considered acceptable. Figure 6 plots the frequency versus the predicted probability for all landslide (1) and non-landslide (0) points. Non-landslide points appear close to the lower end of the plot (or p ¼ 0), whereas landslide points close to the upper end (or p ¼ 1), indicating that the present logistic regression model is considered quite reliable. Table 3 showed that –2log likelihood is the smallest (141.176) and Cox and 2 2 Snell R and Nagelkerk R coefficients are the largest (0.437 and 0.630, respectively) if all variables of slope gradient, slope aspect, distance to drainage, distance to road, land use and geology are included. It means that the logistic regression model is considered as successful and the selected independent variables by the model are 44 S.Z. Mousavi et al. Table 2. Classification table and statistical performance of the model. Predicted Absence of Presence of Model Observed landslide (0) landslide (1) accuracy (%) Absence of landslide (0) 84 11 88.4 Presence of landslide (1) 17 78 82.1 Overall accuracy (%) 85.3 Figure 6. Success rate curve showing landslide susceptibility index rank (y-axis) occurring in cumulative percent of landslide occurrence (x-axis). Table 3. The comparison of the performance of the logistic regression models for various assumptions of the number of independent variables. 2 2 Independent variables –2log likelihood Cox and Snell R Nagelkerke R All six 141.761 0.437 0.630 No slope gradient 170.494 0.387 0.516 No slope aspect 155.660 0.424 0.565 No distance from drainage 155.757 0.433 0.577 No distance from road 229.475 0.164 0.218 No land use 154.290 0.437 0.582 No geology 148.061 0.405 0.607 significant statistically. Considering the success rate curve (figure 6), the logistic regression model shows high accuracy when compared with known events. For example, 90–100% (10%) class of the study area where the landslide hazard index had a higher rank could account for 43% of all the landslides. In addition, the 80– 100% (20%) class of the study area where the landslide hazard index had a higher rank and could explain 54% of the landslides. Furthermore, the AUC value of 0.7740 represents that the prediction accuracy of the susceptibility map is 77.40% that is highly desirable. GIS-based spatial prediction of landslide susceptibility 45 4. Discussion The statistical analysis of logistic regression has been applied to landslide susceptibility mapping in various studies and its reliability has been confirmed. Brenning (2005) studied different approaches and introduced logistic regression with stepwise variable selection as a proper method for the prediction of landslide susceptibility. Lee and Pradhan (2007) also used a logistic regression model to assess landslide hazard caused by rainfall. The verified results were compared with the results from a probabilistic model. It was proved that a logistic regression model is more successful than a probabilistic model in hazard prediction. In this research, logistic regression is used to prepare a landslide susceptibility map for Sajarood basin, northern Iran using a total of 95 landslide and 95 non-landslide data points. Ten causative factors on landslide occurrence such as elevation, slope gradient, slope aspect, slope curvature, rainfall, distance to fault, distance to drainage, distance to road, land use and geology were taken as independent parameters. Depending on the independent parameters considered, the landslide inventory map and the statistical approach used, the best predictor parameters and the predicted probability map of a logistic regression can vary considerably. There are no guidelines for choosing predictor factors. In fact they can vary according to the study area characteristics. The first results of the logistic regression were the model statistics and coefficients, which were useful to assess the accuracy of the regression function and the role of parameters on the presence or absence of landslides. Elevation, slope curvature, rainfall and distance to fault were considered insignificant in predicting slope instability and thus excluded from the model. The landslide susceptibility map representing a very low, low, moderate and high probability of landsliding was the second outcome of the regression process incorporating with GIS. According to the map classification, most parts of Sajarood area are found in very low and low susceptible zones. A few land parts in proximity to roads and residential parts are susceptible at moderate and high scale. The low- and mid-altitude elevations of the area facing the roads in forest lands covered by relatively weak rocks and soils such as silty marl, siltstone and mudstone are classified as high or medium susceptible to the process of landslides. As a last step, the reliability of hazard map was confirmed through accuracy assessment approaches. Considering coefficients estimated for the logistic regression (table 1), the ‘closeness to roads’ parameter was found to have the strongest relationship with landslide occurrence. Ayalew et al. (2005) have introduced a ‘proximity to roads’ parameter as the most important factor on landslide occurrence in Kakuda-Yahico, central Japan. They declared that most of the landslides were located in range 0–100 m from roads. Also Lee and Sambath (2006), Greco et al. (2007) and Kelarestaghi and Ahmadi (2009) have emphasized the adverse effect of road construction on landslide occurrence in their studies. Northwest slope aspects, due to having more rainfall and moisture retention and then cutting down normal pressure and shear strength of soils, were introduced as an important interior causative factor on landslides in logistic regression in the study area. Garcia-Rodriguez et al. (2008), in their research on making a susceptibility map using a logistic regression model in El Salvador, illustrated the importance of terrain roughness, especially north- and northwest-facing slopes as a key factor within the model. Table 2 indicates that geological units belonging to the Myosen era consisting mostly of marl, silty marl, siltstone, limy siltstone and mudstone are more susceptible to landslide. Because of having fine particles of clay and marl, after absorbing 46 S.Z. Mousavi et al. moisture, these units soon reach liquid limit and flow through the slope. Can et al. (2004), Lee (2004) and Nefeslioglu et al. (2008) emphasized the causative role of geological units on mass movements in their research. In general in the study area, in addition to natural parameters including ‘slope gradient’, ‘slope aspect’ and ‘closeness to drainage’, human activities played a major and more important role on landslide. Due to improper human activities such as principal road construction in forests causing instability and incoherence in slopes and also land-use changes such as changing natural forests to farmlands specially, in the lower parts of the region, the number of landslides in the region are significantly increasing. Hence, the produced susceptibility map may be accepted as the basis for landslide risk management studies to be applied in the study area. 5. Conclusion Landslide susceptibility mapping has shown a great deal of importance in hazard management and land-use planning. In this manner, there are various GIS-based qualitative and quantitative techniques that are useful for analysing the relationship between landslides and their influence factors. GIS methodology contributes by manipulating data and performing the necessary analysis in a short time very cheaply and it also contributes to the number of different scenarios that can be presented graphically. In this study, GIS-based logistic regression model is used to prepare a susceptibility map for Sajarood basin, northern Iran. Model statistics and coefficients indicate that proximity to roads has the strongest relationship with landslide occurrence, whereas slope gradient and land-use parameters with negative coefficients showed a negative correlation with landslide occurrence. According to the produced landslide susceptibility map, the proportion of medium and high susceptible zones is far smaller than the low and very low counterparts. This is in agreement with the observation that landslides are more common in low- and mid- altitude slopes than in highlands. Nevertheless, this study illustrates that rural and residential regions in low- and mid-altitude slopes of forest lands, northern Iran, are strongly biased toward landslides because of human activities, especially unprincipled road constructions and land-use changes. Therefore, the government must exert prompt laws for avoiding unprincipled uses of forest lands. The results of this study indicate that it is necessary to integrate and combine the GIS and statistics because each has only a few functions when applied separately. Finally, the produced landslide susceptibility map can provide a cheap and comprehensive assessment of the likelihood of future failures, which can be useful to planners and decision-makers during site selection and rebuilding process and future zoning issues. Acknowledgements The authors wish to express their sincere thanks to Iranian National Cartographic Centre (INCC), Iran Meteorological Organization (IMO), National Geographical Organization, Forests, Range and Watershed management Organization (FRWO) for providing various data sets for this research. Also, the remarks made by two anonymous reviewers on the earlier version of the paper have greatly improved the current version. GIS-based spatial prediction of landslide susceptibility 47 References AYALEW, L. and YAMAGISHI, H., 2005, The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda–Yahiko Mountains, central Japan. Geomorphology, 65, pp. 15–31. AYALEW, L., YAMAGISHI, H. and UGAWA, N., 2004, Landslide susceptibility mapping using GIS-based weighted linear combination, the case in Tsugawa area of Agano River, Niigata Prefecture, Japan. Landslides, 1, pp. 73–81. AYALEW, L., YAMAGISHI, H., MARUI, H. and KANNO, T., 2005, Landslides in Sado island of Japan: part II. GIS-based susceptibility mapping with comparisons of results from two methods and verifications. Engineering Geology, 81, pp. 432–445. BAEZA, C. and COROMINAS, J., 2001, Assessment of shallow landslide susceptibility by means of multivariate statistical techniques. Earth Surface Processes and Landforms, 26, pp. 1251–1263. BRENNING, A., 2005, Spatial prediction models for landslide hazards: review, comparison and evaluation. Natural Hazards and Earth System Sciences, 5, pp. 853–862. CAN, T., NEFESLIOGLU, H.A., GOKCEOGLU, C., SONMEZ, H. and DUMAN, Y., 2005, Susceptibility assessments of shallow earth flows triggered by heavy rainfall at three catchments by logistic regression analysis. Geomorphology, 82, pp. 250–271. CARRARA, A., CROSTA, G. and FRATTINI, P., 2003, Geomorphological and historical data in assessing landslide hazard. Earth Surface Processes and Landforms, 28, pp. 1125–1142. CEVIK, E. and TOPAL, T., 2003, GIS-based landslide susceptibility mapping for a problematic segment of the natural gas pipeline, Hendek (Turkey). Environmental Geology, 44, pp. 949–962. CHAU,K.T.and CHAN, J.E., 2005, Regional bias of landslide data in generating susceptibility maps using logistic regression for Hong Kong Island. Landslides, 2, pp. 280–290. CHAU, K.T., SZE, Y.L., FUNG, M.K., WONG, W.Y., FONG, E.L. and CHAN, L.C.P., 2004b, Landslide hazard analysis for Hong Kong using landslide inventory and GIS. Computers & Geosciences, 30, pp. 429–443. CHAU, K.T., TANG, Y.F. and WONG, R.H.C., 2004a, GIS-based rock fall hazard map for Hong Kong. Rock Mechanic, 41(3), pp. 1–6. COROMINAS, J. and MOYA, J., 1999, Reconstructing recent landslide activity in relation to rainfall in the Llobregat river basin, Eastern Pyrenees, Spain. Geomorphology, 30, pp. 79–93. CROSTA, G., 1998, Regionalization of rainfall thresholds: an aid to landslide hazard evaluation. Environmental Geology, 35(2–3), pp. 131–145. DAHAL, R.K. and HASEGAWA, S.H., 2008, Representative rainfall thresholds for landslides in the Nepal Himalaya. Geomorphology, 100, pp. 429–443. DAI, F.C. and LEE, C.F., 2002, Landslide characteristics and slope instability modelling using GIS, Lantau Island, Hong Kong. Geomorphology, 42, pp. 213–228. DAI, F.C., LEE, C.F., LI, J. and XU, Z.W., 2001, Assessment of landslide susceptibility on the natural terrain of Lantau Island, Hong Kong. Engineering Geology, 40, pp. 381–391. DAI, F.C., LEE, C.F., THAM, L.G., NG, K.C. and SHUM, W.L., 2004, Logistic regression modelling of storm-induced shallow land sliding in time and space on lantau island, Hong Kong. Bulletin of Engineering Geology and Environment, 63, pp. 315–327. DAVIS, J.C., CHANG, C.H.J. and OHLMACHER, G.C., 2006, Two models for evaluating landslide hazards. Computers & Geosciences, 32, pp. 1120–1127. DOMINGUEZ-CUESTA, M., JIMENEZ-SONCHEZ, M. and BERREZUETA, E., 2007, Landslide in the central coalfield (Cantabarian Mountains, NW Spain): geomorphological feature conditioning factors and meteorological implication in susceptibility assessment. Geomorphology, 89, pp. 1–12. DUMAN, T.Y., CAN, T., GOKCEOGLU, C., NEFESLIOGLU, A.H. and SONMEZ, H., 2006, Application of logistic regression for landslide susceptibility zoning of Cekmece Area, Istanbul, Turkey. Environmental Geology, 51, pp. 241–256. 48 S.Z. Mousavi et al. ERCANOGLU, M. and GOKCEOGLU, C., 2004, Use of fuzzy relations to produce landslide susceptibility map of a landslide prone area (West Black Sea Region, Turkey). Engineering Geology, 75, pp. 229–250. GARCIA-RODRIGUEZ, M.J., MALPICA, J.A., BENITO, B. and DIAZ, M., 2008, Susceptibility assessment of earthquake-triggered landslides in El Salvador using logistic regression. Geomorphology, 95, 172–191. GARFI, G. and BRUNO, D.E., 2007, Fan morphodynamics and slope instability in the Mucone River basin (Sila Massif, southern Italy): significance of weathering and role of land use changes. Catena, 50, pp. 181–196. GOKCEOGLU, C. and AKSOY, H., 1996, Landslide susceptibility mapping of the slopes in the residual soils of the Mengen region (Turkey) by deterministic stability analyses and image processing technique. Engineering Geology, 44, pp. 147–161. GOMEZ, H. and KAVZOGLU, T., 2005, Assessment of shallow landslide susceptibility using artificial neural networks in Jabonosa River Basin, Venezuela. Engineering Geology, 78, pp. 11–27. GRECO, R., SORRISO, V. and CATALANO, E., 2007, Logistic regression analysis in the evaluation of mass movement’s susceptibility case study: Calabria, Italy. Engineering Geology, 89, pp. 47–66. GUZZETTI, F., CARRARA, A., CARDINALI, M. and REICHENBACH, P., 1999, Landslide hazard evaluation: a review of current techniques and their application in a multi-scale study, Central Italy. Geomorphology, 31, pp. 181–216. HOSMER, D.W. and LEMESHOW, S. (Eds.), 2000, Applied Logistic Regression, 2nd edition, 373 pp (New York, NY: Wiley). JADDA, M., SHAFRI, H.M., MANSOR, S.H., SHARIFIKIA, M. and PIRASTEH, S., 2009, Landslide susceptibility evaluation and factor effect analysis using probabilistic-frequency ratio model. European Journal of Scientific Research, 33, pp. 654–668. JADE, S. and SARKAR, S., 1993, Statistical models for slope stability classification. Engineering Geology, 36, pp. 91–98. KANUNGO, D.P., ARORA, M.K., SARKAR, S. and GUPTA, R.P., 2006, A comparative study of conventional, ANN black box, fuzzy and combined neural and fuzzy weighting procedures for landslide susceptibility zonation in Darjeeling Himalayas. Engineering Geology, 85, pp. 347–366. KELARESTAGHI, A. and AHMADI, H., 2009, Landslide susceptibility analysis with a bivariate approach and GIS in northern Iran. Arabian Journal of Geosciences, 2, pp. 95–101. KELARESTAGHI, A. and GARAEE, P., 2007, Landslide hazard mapping using AHP technique in Vastan basin. Geographical Researches, 87(4), pp. 49–68 [in Persian]. KELARESTAGHI, A., HABIBNEJAD, M., AHMADI, H., 2007, Assessment of landslide occurrence related to landuse change and road construction. Geographical Researches, 62, pp. 81– 91 [in Persian]. KELARESTAGHI, A. and JAFARIAN, Z., 2010, Land use/cover change and driving force analyses in parts of northern Iran using RS and GIS techniques. Arabian Journal of Geosciences. DOI: 10.1007/s12517-009-0078-5. LEE, S., 2004, Application of likelihood ratio and logistic regression models to landslide susceptibility mapping using GIS. Environmental Management, 34, pp. 223–232. LEE, S. and PRADHAN, B., 2007, Landslide hazard mapping at Selangor Malaysia using frequency ratio and logistic regression models. Landslides, 4, pp. 33–41. LEE, S. and SAMBATH, T., 2006, Landslide susceptibility mapping in the Damrei Romel area, Cambodia using frequency ratio and logistic regression models. Environmental Geology, 50, pp. 847–855. MENARD, S.W. (Ed.), 2001, Applied Logistic Regression Analysis, 2nd edition, 111 pp (Thousand Oaks, CA: Sage). GIS-based spatial prediction of landslide susceptibility 49 MOUSAVI KHATIR, S.Z., KELARESTAGHI, A. and HASHEMZADEH ATOEI, A., 2009, Statistical analysis of some morphometric characteristics and effective factors on slope instability in parts of Babolrood watershed. Journal of Water and Soil Conservation, 16(2), 85– 103 [in Persian]. NEFESLIOGLU, H.A., GOKCEOGLU, C. and SONMEZ, H., 2008, An assessment on the use of logistic regression and artificial neural networks with different sampling strategies for the preparation of landslide susceptibility maps. Engineering Geology, 97, pp. 171–191. OHLMACHER, G.C. and DAVIS, J.C., 2003, Using multiple logistic regression and GIS technology to predict landslide hazard in northeast Kansas USA. Engineering Geology, 69, pp. 331–343. PEART, M.R., NG, K.Y. and ZHANG, D.D., 2005, Landslides and sediment delivery to a drainage system: some observations from Hong Kong. Asian Earth Sciences, 25, pp. 821–836. PRADHAN, B., 2010, Remote sensing and GIS-based landslide hazard analysis and cross- validation using multivariate logistic regression model on three test areas in Malaysia. Advances in Space Research, 45, pp. 1244–1256. PRADHAN, B. and LEE, S., 2009a, Delineation of landslide hazard areas using frequency ratio, logistic regression and artificial neural network model at Penang Island, Malaysia. Environmental Earth Science. doi:10.1007/s12665-009-0245-8. PRADHAN, B. and LEE, S., 2009b, Landslide risk analysis using artificial neural network model focusing on different training sites. International Journal of Physical Sciences, 3(11), pp. 1–15. PRADHAN, B. and LEE, S., 2010a, Regional landslide susceptibility analysis using back pro- pagation neural network model at Cameron Highland, Malaysia. Landslides, 7, pp. 13–30. PRADHAN, B. and LEE, S., 2010b, Landslide susceptibility assessment and factor effect analysis: backpropagation artificial neural networks and their comparison with frequency ratio and bivariate logistic regression modeling. Environmental Modelling and Software, 25, pp. 747–759. PRADHAN, B., LEE, S. and BUCHOITHNER, M.F., 2009, Use of geospatial data for the development of fuzzy algebraic operators to landslide hazard mapping: a case study in Malaysia. Applied Geomatics, 1, pp. 3–15. PRADHAN, B., LEE, S. and BUCHROITHNER, MF., 2010a, Remote sensing and GIS-based landslide susceptibility analysis and its cross-validation in three test areas using a frequency ratio model. Photogrammetrie and Fernerkun, 1, pp. 17–32. PRADHAN, B., LEE, S. and BUCHROITHNER, MF., 2010b, A GIS-based back-propagation neural network model and its cross application and validation for landslide susceptibility analyses. Computers, Environment and Urban Systems, 34, pp. 216–235. PRADHAN, B., LEE, S., MANSOR, S., BUCHROITHNER, MF., JALLALUDDIN, N. and KHUJAIMAH, Z., 2008, Utilization of optical remote sensing data and geographic information system tools for regional landslide hazard analysis by using binomial logistic regression model. Applied Remote Sensing, 2, pp. 1–11. SAHA, A.K., GUPTA, R.P. and ARORA, M.K., 2002, GIS-based landslide hazard zonation in the Bhagirathi (Ganga) valley, Himalayas. International Journal of Remote Sensing, 23, pp. 357–369. SOETERS, R. and VAN WESTEN, C.J., 1996, Slope instability recognition, analysis and zonation. In A.K. Turner and R.L. Schuster (Eds.), Landslides Investigation and Mitigation, Transportation Research Board, National Research Council, Special Report 247, pp. 129–177 (Washington, DC: National Academy Press). WIECZOREK, G.F., GORI, P.L., JAGER, S., KAPPEL, W.M. and NEGUSSEY, D., 1996, Assessment and management of landslide hazards near Tully Valley landslide, Syracuse, New York, USA. In Proceeding of 7th International Symposium on Landslides, Trondheim, pp. 411–416 (Rotterdam: Balkema). 50 S.Z. Mousavi et al. YACLIN, A., 2008, GIS-based landslide susceptibility mapping using analytical hierarchy process and bivariate statistics in Ardesen (Turkey): comparisons of results and confirmations. Catena, 72, pp. 1–12. YILMAZ, I., 2009, Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: a case study from Kat landslides (Tokat, Turkey). Computers and Geoscience, 35, pp. 1125–1138.

Journal

"Geomatics, Natural Hazards and Risk" – Taylor & Francis

Published: Mar 1, 2011

There are no references for this article.