Science 69

822019 Science 69

httpslidepdfcomreaderfullscience-69 114

A GIS-based comparative study of frequency ratio analytical hierarchy processbivariate statistics and logistics regression methods for landslide susceptibilitymapping in Trabzon NE TurkeyA Yalcina S Reisb AC Aydinogluc T Yomralioglu c

a Department of Geological Engineering Aksaray University 68100 Aksaray Turkeyb Department of Geodesy and Photogrammetry Engineering Aksaray University 68100 Aksaray Turkeyc Department of Geomatic Engineering Istanbul Technical University 34469 Istanbul Turkey

a b s t r a c ta r t i c l e i n f o

Article historyReceived 23 January 2009Received in revised form 21 April 2010Accepted 26 January 2011

KeywordsLandslideGISFrequency ratio modelAnalytical hierarchy processBivariate statisticLogistics regression model

Over the last few decades many researchers have produced landslide susceptibility maps using differenttechniques including the probability method (frequency ratio) the analytical hierarchy process (AHP)bivariate multivariate logistics regression fuzzy logic and arti cial neural network In addition a number of parameters such as lithology slope aspect land cover elevation distance to stream drainage densitydistance to lineament seismicity and distance to road are recommended to analyze the mechanism of landslides The data quality is a very important issue in landslide studies and more accurate results will beachieved if the data is adequate appropriate and drawn from a wide range of parameters The aim of thisstudy was to evaluate the susceptibility of the occurrence of landslides in Trabzon province situated in northeast Turkey This wasachieved using thefollowing ve methods thefrequency ratio modelAHP thestatisticalindex ( Wi) weighting factor ( Wf ) methods and the logistics regression model incorporating a GeographicalInformation System (GIS) and remote sensing techniques In Trabzon province there has been an increasingoccurrence of landslides triggered by rainfall These landslides have resulted in death signi cant injurydamage to property and local infrastructure and threat of further landslides continues In order to reduce theeffects of this phenomenon it is necessary to scienti cally assess the area susceptible to landslide To achievethis landslide susceptible areas were mapped the landslide occurrence parameters were analyzed using vedifferent methods The results of the ve analyses were con rmed using the landslide activity map containing50 active landslide zones Then the methods giving more accurate results were determined The validationprocess showed that the Wf method is better in prediction than the frequency ratio model AHP the statisticalindex (Wi) and logistics regression model

copy 2011 Elsevier BV All rights reserved

1 Introduction

Landslides are amongst the most damaging natural hazards inmountainous regions Every year hundreds of people all over theworld lose their lives in landslides furthermore there are largeimpacts on the local and global economy from these events Over thepast 25 years many government and international research institu-tions across the world have invested considerable resources inassessing landslide susceptibilities and in attempting to producemaps portraying their spatial distribution ( Guzzetti et al 1999 ) InTurkey landslides are the second most common natural hazard afterearthquakes ( Ildir 1995 ) and the Eastern Black Sea region isespecially affected This region exhibits mountainous topographicalfeatures and is frequently subjected to heavy precipitation This

combination results in the region being prone to extensive and severelandslides In Turkey during the last 50 years natural hazards causedlosses related to housing amounting to an estimated US$155 billionThe annual economic loses emanating from landslides are about US$80 million and the majority of the losses are in the Eastern Black Searegion ( Yalcin 2007 ) A number of different methods for landslidesusceptibility mapping have been utilized and suggested The processof creating these maps involves several qualitative or quantitativeapproaches Early attempts de ned susceptibility classes by thequalitative overlaying of geological and morphological slope-attri-butes to landslide inventories ( Nielsen et al 1979 ) More sophisti-cated assessments involved for example AHP bivariate multivariatelogistics regression fuzzy logic and arti cial neural network analysis(Carrara 1983 van Westen 1997 Lee and Min 2001 Ercanoglu andGokceoglu 2004 Lee et al 2004 Komac 2006 Yalcin 2008 )

Landslide susceptibility mapping may be de ned as qualitative orquantitative and direct or indirect ( Guzzetti et al 1999 ) Qualitativemethods are subjective they represent the susceptible levels in

Catena 85 (2011) 274 ndash 287

Corresponding author Tel +90 382 2801357 fax +90 382 2801365E-mail address ayalcinaksarayedutr (A Yalcin)

0341-8162$ ndash see front matter copy 2011 Elsevier BV All rights reserved

doi101016jcatena201101014

Contents lists available at ScienceDirect

Catena

journa l homepage wwwelseviercom loca te ca tena

822019 Science 69


descriptive expressions and depend on expert opinions The mostcommon types of qualitative methods basically use landslideinventories to recognize sites of comparable geological and geomor-phologic characteristics that are susceptible to failure Howeverweights of the parameters are determined from the knowledge of specialists on the subject and the area The designating weights arethus highly personal and may be include some virtual admissionQuantitative methods are based on numerical expressions of the

relationship between controlling factors and the landslides There aretwo types of quantitative methods deterministic and statistical(Aleotti and Chowdhury 1999 ) Deterministic methods are based onslope stability studies expressed in terms of the safety factor ( Re ceand Capolongo 2002 Zhou et al 2003 ) The statistical approachesanalyze the historical link between landslide-controlling factors andthe distribution of landslides Quantitative methods may be used todecrease the personality and bias in the weight assessment processTherefore more realistic susceptibility maps can be produced from anobjective measure of values During the past few years quantitativemethods have been implemented for landslide susceptibility zonationstudies in different regions ( Clerici et al 2002 Suzen and Doyuran2004 Ercanoglu and Gokceoglu 2004 Yesilnacar and Topal 2005Kanungo et al 2006 Yalcin and Bulut 2007 Garciacutea-Rodriacuteguez et al2008 Nefeslioglu et al 2008 etc)

The aim of this study was to use widely-accepted models astatistical method (frequency ratio model) a multi-criteria decisionmaking approach (AHP) bivariate and multivariate approaches(logistic regression) and evaluate their performances

Frequency ratio model is based on the observed associationsbetween allocation of landslides and each associated factors of landslide occurrence to display the correlation between landslidelocations and the parameters controlling landslide occurrence in thearea ( Lee 2005) Therefore the method gives very good results fordetermining the landslide inventory with rigorous accuracy Theweights of the parameters and the decision alternatives used inproducing landslide susceptibility map are determined with the AHPWhen these weights are determined both the comparison of theparameters relative to each other and determination of the effectvalues of the decision alternatives namely the sub-criteria are basedon a landslide inventory map obtained with the help of aerial photosand satellite images As a result of dual comparisons a pair-wisecomparison matrix is obtained for each parameter and sub-criteriaConsequently the weight values were determined correctly for thereal land data It has been shown that the use of the AHP methodproduces a practical and realistic result to de ne the factor weights inthe landslide susceptibility model In statistical models usingbivariate or multivariate techniques for landslide susceptibilityanalysis is widespread ( Nandi and Shakoor 2009 ) There are anumber of ways to apply bivariate and multivariate statistics to assesslandslide susceptibility of a region More than a few instabilityparameter variables are used in the present bivariate approach thein uence of each variable on the occurrence of landslide is evaluated

independently and the variables are combined in the form of a uniqueequation ( Conoscenti et al 2008 Nandi and Shakoor 2009 ) Inmultivariate approaches logistic regression was detected to be themost appropriate approach for the present study In this analysisspatial distribution of landslides is assessed on the basis of interactionof onlystatistically signi cant instability factordata insigni cant dataare excluded from consideration Additionally logistic regressionanalysis is free of data distribution issues and can handle a variety of datasets such as continuous categoricaland binary common types of instability factor data used in landslide studies ( Dai et al 2001 Leeand Min 2001 Lee and Sambath 2006 Akgun et al 2008 Nandi andShakoor 2009 ) Anyeffort to make certain landslide susceptibility in aregion needs proper validation Con rmation should establish thequality (ie consistency robustness degree of tting and prediction

skill) of the proposed susceptibility estimate The excellence of a

landslide susceptibility model can be ascertained using the samelandslide data used to obtain the susceptibility estimate or by usingindependent landslide information not available to construct themodel ( Chung and Fabbri 2003 Guzzetti et al 2005 2006 ) In thispaper we provide a comprehensive validation of a landslidesusceptibility model prepared through ve different methods for theTrabzon City ( Fig 1)

2 Description of the study area

The study area in Trabzon province consisted of approximately4660 km 2 located between 39deg 15 prime and 40deg 15 prime west ndash east longitudesand 41deg 8 prime and 40deg 30 prime north ndash south latitudes in the middle of EasternBlack Sea region ( Fig 1) Altitudes reach 3400 m in parts of the regionand steep slopes are very common The climate is characteristic of theBlack Sea region with temperate climate summers and a rainy seasonnormally lasting from September to April Nevertheless the rainfallregime is irregular with some periods of rare precipitation with long-lasting heavy rains Very intense precipitation has caused disastrous ash oods in river basins and many landslides on slopes ( Reis andYomralioglu 2006 Yalcin 2008 ) According to observations result 22yearly in Trabzon province February is the coldest month with anaverage temperature of 67 degC the hottest month is August with antemperature average of 232 degC The annual precipitation of Trabzoncity is 838 mm and precipitations disperse every month symmetricalThe main commercial agricultural products in the region are hazel-nuts and green tea Apart from the agricultural areas the other mainland cover types are forest and pasture The population of theprovince was 740569 in 2007 year ( TUIK 2008)

3 Thematic data layers

Thestudy began with the preparation of a landslide inventorymapbased on extensive eld work a previous inventory map and satelliteimages Furthermore the following seven possible landslide causinglayers lithology slope aspect land cover elevation distance tostream anddistance to road were analyzed for landslide susceptibilitymapping using the logistic probability method (frequency ratiomethod mdash FRM) the analytical hierarchy process (AHP) bivariate(Wi and Wf ) and logistics regression (LR) methods Finally thesusceptibility maps produced from the ve different methods werecompared and evaluated using validation data sets

31 Landslide inventory map

Themaps show the locationsand properties of landslides that havemoved in the past These slope failures were related to geologicaltopographical and climatic conditions thus they can often facilitatethe prediction of locations and conditions of future landslides For thisreason it is important to determine the location and area of thelandslide accurately when preparing the landslide susceptibility

maps Landslide susceptibility assessment is performed in a range of phases The initial phase is identifyingand evaluating landslide-proneareas and constructing a landslide inventory map for future useLandslide inventory mapping is the systematic mapping of existinglandslides in a region using different techniques such as eld surveyair photosatellite image interpretation and literature search forhistorical landslide records A landslide inventory map provides thespatial distribution of locations of existing landslides The landslidesin the study area were determined by comprehensive eld surveysThe landslides which are currently inde nite in characteristics andboundaries were identi ed using old dated satellite images As aresult the satellite images were very useful in determination of landslides inventory map ( Yalcin and Bulut 2007 ) In this study thesusceptibility mapping started with the preparation of an inventory

map of 250 landslides from eld studies a previous inventory map

275 A Yalcin et al Catena 85 (2011) 274 ndash 287

822019 Science 69


and satellite image analyses from Quickbird ( Fig 2) Also to con rmthe practicality of producing ve susceptibility maps 50 activelandslides zones were determined separately from the inventory

map In the Trabzon region the rough topography susceptibleweathering units and the temperate climate means that many newlandslides appear from time to time as a result of heavy rainfalls High

Fig 1 Location map of the study area

Fig 2 Landslide inventory map


822019 Science 69


intensity rainfalls produce ash oods which cause shallow land-slides So the landslides in this region show shallow translationalcharacteristics The eld surveys and drilling data with previousstudies were used to determine the depth of the weathering zones inrocks and the groundwater table The weathering zone has reached tomaximum approximately 20 m However the groundwater tablewasnt observed 30 m depth and thus there were no landslides withdeeper slip surfaces

32 Lithology map

Lithology is one of the most important parameters in landslidestudies because different lithological units have different susceptibil-ity degrees ( Dai et al 2001 Yesilnacar and Topal 2005 Yalcin andBulut 2007 Garcia-Rodriguez et al 2008 Nefeslioglu et al 2008 )The landslideeventa component of thegeomorphological research isrelated to the lithological characteristics of the land It is extensivelyaccepted that lithology signi cantly in uences the occurrence of landslides because lithological variations often lead to a difference inthe strength andpermeability of rocks andsoilsThe lithologymaps of the study area were differentiated into eight lithological units As aresult of the aerial distributions analysis performed according to thelithological units most landslides (2562) are located within basaltandesite pyroclastics and intercalations of mudstone sandstone andtuffs ( Table 1)

33 Slope map

The major parameter of slope stability analysis is the slope angle(Lee and Min 2001 ) Slope angle is very regularly used in landslidesusceptibility studies since landsliding is directly related to slopeangle ( Dai et al 2001 Cevik and Topal 2003 Lee 2005 Yalcin 2008Nefeslioglu et al 2008 ) The slope map of the study area was dividedinto ve slope categories ArcGIS 92 analysis was performed todiscover in which slope group the landslide happened and the rate of occurrence was observed The landslide percentage in each slopegroup class is determined as a percentage of slopes The resultindicates that most of landslides (4926) occur when the percentageof the slope more than 50 ( Table 1)

34 Aspect

Aspect is also considered an important factor in preparinglandslide susceptibility maps ( Cevik and Topal 2003 Lee 2005Yalcin and Bulut 2007 Galli et al 2008 ) Aspect associatedparameters such as exposure to sunlight drying winds rainfall(degree of saturation) and discontinuities may affect the occurrenceof landslides ( Suzen and Doyuran 2004 Komac 2006 ) Theassociation between aspect and landslide is shown with aspectmaps Aspect regions are classi ed in nine categories according to theaspect class as at ( minus 1deg) north (0deg ndash 225deg 3375deg ndash 360deg) northeast

(225deg ndash 675deg) east (675deg ndash 1125deg) southeast (1125deg ndash 1575deg) south(1575deg ndash 2025deg) southwest (2025deg ndash 2475deg) west (2475deg ndash 2925deg)and northwest (2925deg ndash 3375deg) Analyses were performed usingaspect and landslide inventory maps to determine the distributionof landslides according to the aspect class and the percentage of landslides that occurred in each aspect class ( Table 1)

35 Elevation

Elevation is useful to classify the local relief and locate points of maximum and minimum heights within terrains To calculatelandslide densities for different relief classes the relief map wasdivided into seven altitude classes on 500-m basis and the study areareveals that the elevation ranges from 0 to 3500 m above mean sea

level However landslides below 1500 m are dominant (8928) due

to the lithological character of the units that have pyroclasticcompositions

36 Land cover

The effect of land cover on slope stability can be clari ed by anamount of hydrological and mechanical effects Land cover acts as ashelter and reduces the susceptibility of soil erosion landslides andthe get water on action of the precipitation Vegetation extensivelychanges soil hydrology by increasing rainfall interception in ltrationand evapo-transpiration Interception and evapo-transpiration de-crease the quantity of water that reaches the soil and is stored in itThey dont play a vital function during the short heavy rainfall eventsgenerally required to trigger shallow landslides but they can be of

importance for the long term evolution of water in soil and thus for

Table 1Frequency ratio values of the landslide-conditioning parameters

Parameter Classes of totalarea (a)

of landslidearea (b)

Frequencyratio (ba)

Geologya Alv 386 154 040Pl 239 223 094Ev 3663 1317 036Kru 5465 8002 146 Jkr 011 000 000

Jlh 067 000 000γ 2 163 304 187Kk 007 000 000

Slope () 0ndash 10 1372 241 01810ndash 20 694 571 08220ndash 30 1107 1179 10730ndash 50 2682 3083 115N 50 4145 4926 119

Aspect Flat 1055 072 007North 1239 874 071Northeast 1290 1265 098East 1186 1714 145Southeast 924 1117 121South 868 849 098Southwest 1054 1146 109West 1173 1636 140Northwest 1212 1327 110

Elevation (m) 0ndash

500 2377 2533 107500ndash 1000 2223 2967 1331000 ndash 1500 1985 3428 1731500 ndash 2000 1769 636 0362000 ndash 2500 1396 435 0312500 ndash 3000 233 000 0003000 ndash 3500 017 000 000

Land cover Tea 109 169 156Hazelnut 1539 2315 150Deciduous 3799 4285 113Coniferous 231 165 072Mix wood 517 477 092Rocky 114 102 089Pasture 1861 803 043Agriculture 1714 1648 096Settlement 116 035 030

The distance tostream (m)

0ndash 25 2920 2821 09725ndash 50 1068 962 09050ndash 75 1971 2037 10375ndash 100 1126 1128 100100ndash 150 917 980 107150ndash 200 975 1038 106200ndash 250 1023 1033 101

The distance toroad (m)

0ndash 25 1989 2109 10625ndash 50 2001 2020 10150ndash 75 2007 1952 09775ndash 100 2004 1946 097100ndash 125 1998 1972 099

a AlvmdashAlluvium PlmdashPliocene continental units Ev mdashEocene volcanic facies Krumdash

Basalt andesite pyroclastics and intercalations of sandstone clayey limestone andsiltstone Jkr mdash Jurassicndash Cretaceous units Jlh mdashLias units γ 2 mdashKaccedilkar granites Kkmdash

Carboniferous units


822019 Science 69


initial moisture conditions when an extreme event occurs Rootsincrease soil permeability and thus in ltration and conductivitywhich cause greater accumulation of water in the soil during bothshort term events and long rainfall periods The vegetation cover alsointroduces some mechanical changes through soil reinforcement andslope loading The increase in soil strength due to root reinforcementhas great potentialto reduce the rate of landslide occurrence ( Wu andSwanston 1980 Blijenberg 1998 Cannon 2000 Begueriacutea 2006 )

Several researchers ( Ercanoglu and Gokceoglu 2004 Tangestani2004 Reis and Yomralioglu 2006 Yalcin 2007 ) have emphasized theimportance of land cover on slope stabilities In this study a singledate image of Landsat ETM+ (Path 173 Row32) onOctober 19 2000was used to generate the land cover types Using the image afterextracting an application area of approximately 120times90 km coveringthe administrative boundaries of Trabzon province other studies asrequired were implemented on this area The Landsat ETM+ imagehas six multi-spectral bands with 28 m resolution one thermal bandwith a 60 m resolutionand a panchromatic band with 15 m resolution(Reis and Yomralioglu 2006 ) The study area was divided into nineland cover classes ( Table 1) being mostly covered with deciduouspasture and agriculture areas Landslides are largely observed indeciduous and hazelnut areas The deciduous areas include differenttree types of tree growth such as brake thicketand small wood Thesetypes obstruct the surface ow of precipitations and this increases thepore water pressure of soil thus the potential of the occurrence of landslides has increased in these areas

37 Distance to stream

Distance to stream is one of the controlling factors for the stabilityof a slope The saturation degrees of the materials directly affect slopestability The proximity of the slopes to the drainage structures is alsoimportant factor in terms of stability Streams may negatively affectstability by eroding the slopes or by saturating the lower part of material until the water level increases ( Dai et al 2001 Saha et al2002 ) In this respect the relation streams and groundwater are alsoimportant Groundwater exchanges directly the characteristics of surface water by sustaining stream base ow Groundwater affectssurface water by providing moisture for riparian vegetation andcontrolling the shear strength of slope materials thereby affectingslope stability and erosion processes Low river ow during periods of no rain or snowmelt input is called base ow which represent thenormal condition of rivers Groundwater provides base ow foressentiallyall rivers and has a major effecton the amount of waterandchemical composition of rivers In smaller low-order streamsgroundwater also provides much of increased discharge during andimmediately following storms The effect of streams to landslideincreases all of these events The study area was divided into sevendifferent buffer ranges Primary streams and secondary streams werebranched and the proximity buffers were constructed for intervals of 100 ndash 250 m although extra classeswere de ned for 0 ndash 25 m 25 ndash 50 m

50ndash 75 m and 75 ndash 100 m

38 Distance to roads

The road density is one of the causal factors for landslides and isparallel to the effect of the distance to streams The load in the toe of slope can be reduced by road-cuts A drop-down road section maybehave like a wall a net source a net sink or a corridor for water owand depending on its location in the mountains this type of road isusually a contributing factor in causing landslides ( Ayalew andYamagishi 2005 Yalcin 2008 ) The study area was divided into vedifferent buffers categorized to designate the in uence of the road onthe slope stability The landslide percentage distribution wasdetermined according to the buffer zones by comparing the map of

the distance to the road and the landslide inventory ( Table 1)

4 Landslide susceptibility analyses

In this study the landslide susceptibility analyses were imple-mented using the methods of frequency ratio analytical hierarchyprocess bivariate ( Wi and Wf ) and logistics regression In order toachieve this landslide factors related to the causes of landslideoccurrence in the study area such as the geology slope aspectelevation land cover distance to streams and distance to roads layers

were used The Digital Elevation Model (DEM) was digitized from 125000 scaled Standard Topographic Maps and the contours on thesemaps are drawn at 10 m intervals The DEM of the study area wascreated using ArcGIS 92 software 10times10 m pixel dimensions of thelandslide and parameter maps were chosen Precipitation data wasnot included in the susceptibility analyses because it was approxi-mately same over the whole area Seismic data was also discountedbecause the study area is far away from seismic activity Landslideareas were determined using previous inventory map and Quickbirdsatellite images Furthermore the landslide data were achieved andcon rmed in the eld studies

41 Frequency ratio method

When evaluating the probability of landsliding within a speci cperiod of time and within a certain area it is of major importance torecognize the conditions that can cause the landslide and the processthat could trigger the movement The correlation between landslideareas and associated factors that cause landslides can be allocatedfrom the connections between areas without past landslides and thelandslide-related parameters In order to prepare the landslidesusceptibility map quantitatively the frequency ratio method wasimplemented using GIS techniques Frequency ratio methods arebased on the observed associations betweendistribution of landslidesand each landslide-related factor to expose the correlation betweenlandslide locations and the factors in the study area Using thefrequency ratio model the spatial associations between landslidelocation and each of the factors contributing landslide occurrencewere derived The frequency is calculated from the analysis of therelation between landslides and the attributed factors Therefore thefrequency ratios of each factors type or range were calculated fromtheir relationship with landslide events as shown in Table 1 Thefrequency ratio was calculated for sub-criteria of parameter and thenthe frequency ratios were summed to calculate the landslidesusceptibility index ( LSI ) (Eq 1) ( Lee and Talib 2005 )

LSI = Fr 1 + Fr 2 + Fr 3 + + Fr n eth1THORN

where Fr is rating of each factors type or rangeAccording to thefrequency ratio method the ratio is that of the area

where the landslide occurred to the total area so that a value of 1 is anaveragevalue If the value is N 1 itmeans the percentageof the landslideis higher than theareaand refers to a highercorrelationwhereasvalues

lower than 1 mean a lower correlation ( Akgun et al 2007 )The geological characteristics of the study area are very important

factors in susceptibility analyses There are eight classes of lithologicalunits in the study area Kru-basalt andesite pyroclastics and intercala-tions of sandstone clayey limestone and siltstone ( Kru) and Kaccedilkargranites ( γ 2) units werefound to be more susceptible lithology Kru andγ 2 include 146 187 of the higher frequency ratio respectively

The slope angle is a one of the most important factors controllingslope stabilities and landslides mostly occur at certain critical slopeanglesMild slopes areestimated to have a lowfrequency for shallow-domiciled landslides because of the minor shear stresses commonlyrelated to low slopes Frequency ratio analyses showed that a slopeangle in a range of 20 ndash 50 and N 50 shows high probability of landslide occurrence As expected a low gradient indicated a low

frequency ratio in a range of 0 ndash 10 giving a 018 ratio ( Table 1)


822019 Science 69


Like slope aspect is another important parameter in preparinglandslide susceptibility maps In the study area landslides generallyoccurred on east-southeast and west-northwest-southwest side slopesThe aspectassessments showed that landslides were not likely to happenon the slope surfaces The assessment of the aspect factor on east-facingslopes shows high probability (145) of landslide occurrence ( Table 1)

The elevation ndash landslide analyses showed that landslides mostlyoccurred from sea level to 1500 m in particular the frequency ratio is

very high in the elevation range of 1000ndash

1500 m ( Table 1 ) The resultsare related to geological characteristics because the areas in theelevationrangeof 0 ndash 1500 m aregenerallyoverlaidto volcanic units asrhyolite rhyodacite dacite andesite and pyroclastics

The land cover type is very important for landslide studiesespecially the areas that are covered with intense vegetation As in teaplantations intensely vegetated areas exhibit more saturation andgreater instabilities than forest Land cover analyses showed thatlandslides commonly occurred in the tea and hazelnut areas thefrequency ratio being 156 and 150 respectively ( Table 1)

Thedegree of soil saturation isone of thecontrolling factors forslopestabilities The rivers rose to water content of soil until water level andaround in the slope The connection between landslides and distance tostreams gives reverse values Normally the distance from streamsaugments the landslideconstituting should be declines However in thestudy the distance from streams increases the landslide constitutingascends ( Table 1) The reason for this is related to the topographicalmodi cation resulting from the caving of the slopes in the study areaand thus retrogressive failures were formed in the slopes

A road constructed alongside slopes causes a decrease in the load onboth the topography and on the heel of slope Tension cracks may becreated as a resultof anincrease instress on the backof the slope becauseof changes in topography and the decrease of load ( Yalcin 2008 ) Thedistance from roads increases the landslide constituting declines in thestudy andthis is compatible with what is expectedThe distance to roadsanalyses showed that landslides usually occurredat the distance range of 0ndash 50 m ( Table 1) On completion of the analyses the frequency ratio of

each layers classes was determined and a landslide susceptibility map(Fig 3) was produced by the LSI map using Eq (1)

42 Analytical hierarchy process (AHP)

The AHP improved by Saaty (1980) supplies a exible and easilyunderstood way of analyzing complicated problems The AHP is amulti-objectivemulti-criteria decision-makingapproach that enables

the user to arrive at a scale of preferences drawn from a set of alternatives The AHP gained wide application in site selectionsuitability analysis regional planning and landslide susceptibilityanalysis ( Ayalew et al 2005 ) The AHP is a problem-solvingconstruction and a methodical process for representing the elementsof any problem ( Saaty and Vargas 1991 ) To apply this approach it isnecessary to break a complex unstructured problem down into itscomponent factors arrange these factors in a hierarchic order assignnumerical values to subjective judgments on the relative importanceof each factor and synthesize the judgments to determine thepriorities to be assigned to these factors ( Saaty andVargas 2001 ) Oneset of models was enhanced using the values from the statistics tomanually describe the relationships betweenthe different parametersaccording to the AHP methodology and later these values wereimported into the AHP matrixes The other set of models wasdeveloped by automatically importing the calculated relationshipvalues of different factors basedon the statistical values into the AHPmatrixes ( Table 2) The pair-wise comparison matrix was created bymaking dual comparisons made in this context The weights arecalculated from the pair-wise comparison matrix undertaking aneigenvalues and eigenvectors calculation It has been demonstratedthat the eigenvector corresponding to the largest eigenvalue of thematrix provides the relative priorities of the factors ie if one factorhas preference its eigenvector component is larger than that of theother The components of the eigenvector sum to unity Thus a vectorof weights is obtained which re ects the relative importance of thevarious factors from the matrix of paired comparisons

Fig 3 The landslide susceptibility map produced by FR


822019 Science 69


Table 2Pair-wise comparison matrix factor weights and consistency ratio of the data layers

Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights

Lithology(1) Alv 1 0057(2) Pl 2 1 0097(3) Ev 5 4 1 0178(4) Kru 9 8 6 1 0486(5) Jkr 12 13 14 19 1 0037

(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

Consistency ratio 0038Slope ()

(1) 0 ndash 10 1 0043(2) 10 ndash 20 2 1 0068(3) 20 ndash 30 4 2 1 0123(4) 30 ndash 50 6 5 3 1 0288(5) N 50 8 7 5 2 1 0479

Consistency ratio 0021Aspect

(1) Flat 1 0024(2) North 4 1 0064(3) Northeast 5 2 1 0112

(4) East 6 3 2 1 0200(5) Southeast 5 2 1 12 1 0112(6) South 4 1 12 13 12 1 0064(7) Southwest 5 2 1 12 1 2 1 0112(8) West 6 3 2 1 2 3 2 1 0200(9) Northwest 4 2 1 12 1 2 1 12 1 0110

Consistency ratio 0008Land cover

(1) Tea 1 0039(2) Hazelnut 5 1 0207(3) Deciduous 7 2 1 0349(4) Coniferous 1 15 17 1 0042(5) Mix wood 2 14 17 2 1 0062(6) Rocky 1 15 17 1 12 1 0042(7) Pasture 3 13 16 3 2 3 1 0093(8) Agriculture 4 12 13 4 3 4 2 1 0144(9) Settlement 12 17 19 13 13 13 14 16 1 0022

Consistency ratio 0074Elevation (m)

(1) 0 ndash 500 1 0179(2) 500 ndash 1000 2 1 0254(3) 1000 ndash 1500 3 2 1 0360(4) 1500 ndash 2000 13 14 15 1 0104(5) 2000 ndash 2500 14 15 16 12 1 0079(6) 2500 ndash 3000 18 19 19 17 16 1 0023(7) 3000 ndash 3500 18 19 19 17 16 1 1 0023

Consistency ratio 0065Distance to stream (m)

(1) 0 ndash 25 1 0323(2) 25 ndash 50 14 1 0080(3) 50 ndash 75 12 3 1 0208(4) 75 ndash 100 13 2 13 1 0114

(5) 100 ndash 150 14 1 12 12 1 0084(6) 150 ndash 200 13 1 12 1 1 1 0095(7) 200 ndash 250 13 1 12 1 1 1 1 0095

Consistency ratio 0016 Distance to road (m)

(1) 0 ndash 25 1 0394(2) 25 ndash 50 12 1 0234(3) 50 ndash 75 13 12 1 0124(4) 75 ndash 100 13 12 1 1 0124(5) 100 ndash 125 13 12 1 1 1 0124

Consistency ratio 0002Data layers

(1) Lithology 1 0386(2) Slope 13 1 0230(3) Aspect 14 13 1 0129

(4) Elevation 14 13 1 1 0098


822019 Science 69


Normally the determination of the values of the parametersrelative to each other is a situation that is dependent on the choices of the decision-maker However in this study both the comparison of the parameters relative to each other and the determination of thedecision alternatives namely the effect values of the sub-criteria of the parameters (weight) were based on the comparison of landslideinventory maps which were constructed using eld studies previousinventory map and satellite image with the other data layers ( Yalcin2008 ) In AHP an index of consistency known as the consistency ratio(CR) is used to indicate the probability that the matrix judgmentswere randomly generated ( Saaty 1977 )

CR = CI=RI eth2THORN

where RI is the average of the resulting consistency index dependingon the order of the matrix given by Saaty (1977) and CI is theconsistency index and can be expressed as

CI = λ max minus neth THORN= nminus 1eth THORN eth3THORN

where λ max is thelargest orprincipaleigenvalueof thematrixand canbeeasily calculated from the matrix and n is the order of the matrix

For all the models where the AHP was used the CR (ConsistencyRatio) was calculated If the CR values were greater than 01 themodels were automatically discarded Using a weighted linear sumprocedure ( Voogd 1983 ) the acquired weights were used to calculatethe landslide susceptibility models ( Komac 2006 ) As a result of theAHP analyses the landslide susceptibility map was produced forTrabzon province ( Fig 4) In the study lithology slope and aspect arefound to be important parameters for the study area whereasdistance to streams and roads were of lesser importance

43 Bivariate statistics method

In this study landslide susceptibility analyses were implementedusing statistical bivariate methods namely the statistical index ( Wi)method ( van Westen 1997 ) and the weighting factor ( Wf ) method(Cevik and Topal 2003 ) For this reason geology slope aspectelevation land cover distance to streams anddistance to roads layerswere used in the analyses

TheWimethod is based onstatisticalcorrelation(mapcrossing) of thelandslide inventory map with attributes of a different parameters mapThe map crossing results in a cross-table which can be used to calculatethe density of landslides per parameter class A standardization of these

Fig 4 The landslide susceptibility map produced by AHP

Table 2 (continued )

Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(5) Land cover 15 14 13 1 1 0083(6) Dist to stream 17 15 14 13 13 1 0037(7) Dist to road 17 15 14 13 13 1 1 0037

Consistency ratio 0038


822019 Science 69


density values can be obtained by relating them to the overall density inthe entire area ( Oztekin and Topal 2005 ) In this study the Wi values foreach class of each parameter map were obtainedquantitativelyusing the

following formula suggested by van Westen (1997)

Wi = lnDensclassDensmap

= lnNpix Sieth THORNNpix Nieth THORNSNpix Sieth THORNSNpix Nieth THORN

eth4THORN

where

Wi Weight given to a certain parameter classDensclass Landslide density within the parameter classDensmap Landslide density within the entire mapNpix(Si) Number of pixels that contain landslide in a certain

parameter classNpix(Ni) Total number of pixels in certain parameter classSNpix(Si) Number of pixels all landslide

SNpix(Ni) Total number of all pixels

Thenthe Wi value ofeachattributewascalculated( Table3 ) Finally alllayers were overlaid and a resulting susceptibility map was obtained(Fig 5) The Wi susceptibility map was divided into equal classesaccording to the totalnumber of elementsThe classes are very low lowmoderate high and very high susceptibility However in the statisticalindex method it is considered that each parameter map has an equaleffect on landslides which may not be the case in reality ( Oztekin andTopal 2005 ) Therefore a weighting factor ( Wf ) for each parameter map

was produced For this purpose

rst the Wi value of each pixel wasdetermined by the statistical index method then all pixel values withinthe landslide zones belonging to each layer were summed By using themaximum and minimum of all layers the results were stretched ( Cevikand Topal 2003 ) Finally the weighting factor ranging from 1 to 100 foreach layer was determined by the following formula

Wf =TWivalueeth THORNminus MinTWivalueeth THORN

MaxTWivalueeth THORNminus MinTWivalueeth THORNAtilde100

where

Wf Weighting factor calculated for each layerTWivalue Total weighting index value of cells within landslide bodies

for each layerMinTWivalue Minimum total weighting index value within selected

layersMaxTWivalue Maximum total weighting index value within selected

layers

By executing this formula the weighting factor ( Wf ) values of eachlayer weredetermined ( Table 3 ) Forthe analyses the Wf value foreachlayer was multiplied by the Wi value of each attribute and nally allparameter maps were summed up to yield the nal landslidesusceptibility map from the Wf method ( Fig 6) The associationbetween pixel value and cumulative pixel count mainly yielded vesusceptible zonesnamely mdashvery low low moderate highandvery highAccording to the results of the Wf method elevation is found to be themost important parameter for the landslides in the study area

44 Logistic regression method

Logistic regression permits one to type a multivariate regressionrelationship between a dependent variable and several independentvariables Logistic regression which is one of the multivariate analysismodels is helpful for forecasting the presence or absence of acharacteristic or outcome based on the values of a set of predictorvariables The advantage of logistic regression is that through theaddition of a suitable link function to the usual linear regressionmodel the variables may be either continuous or discrete or anycombination of both types and they do not necessarily have normaldistributions ( Lee 2005) In the present situation the dependentvariable is a binary variable representing presence (1) or absence (0)

of a landslide Where the dependentvariable is binary the logistic linkfunction is applicable ( Atkinson and Massari 1998 )

In the landslide susceptibility studies logistic regression model is onethe acceptable methods to characterize the association between thepresence or absence of a landslide the dependent variable and a set of independent parameters including geology slope and land cover(Ayalew and Yamagishi 2005 ) Presence (1) and absence (0) coef cientscanbeutilizedto calculate approximateratios foreachof the independentvariables Logistic regression analysis is generally used in earth sciencesand explained as a linear equation given below ( Lee 2005)

Y = Logit peth THORN= lnp

1minus p eth5THORN

Y = C 0 + C 1 X 1 + C 2 X 2 + + C n X n eth6THORN

Table 3Distribution of landslide for various data layers Wi and Wf values of each attribute

Parameter Classes Landslide area () Wi Wf

Geology Alv 154 minus 0920 9116Pl 223 minus 0066Ev 1317 minus 1023Kru 8002 0381 Jkr 000 0000 Jlh 000 0000

γ 2 304 0625Kk 000 0000

Slope () 0ndash 10 241 minus 1741 475710ndash 20 571 minus 019520ndash 30 1179 006330ndash 50 3083 0140N 50 4926 0173

Aspect Flat 072 minus 2687 6136North 874 minus 0349Northeast 1265 minus 0020East 1714 0369Southeast 1117 0190South 849 minus 0022Southwest 1146 0084West 1636 0333Northwest 1327 0091

Elevation (m) 0 ndash 500 2533 0064 10000

500ndash

1000 2967 02891000 ndash 1500 3428 05461500 ndash 2000 636 minus 10222000 ndash 2500 435 minus 11652500 ndash 3000 000 00003000 ndash 3500 000 0000

Land cover Tea 169 0445 3741Hazelnut 2315 0408Deciduous 4285 0120Coniferous 165 minus 0333Mix wood 477 minus 0080Rocky 102 minus 0115Pasture 803 minus 0840Agriculture 1648 minus 0039Settlement 035 minus 1189

The distance to stream (m) 0 ndash 25 2821 minus 0035 15025ndash 50 962 minus 010450ndash 75 2037 003375ndash 100 1128 0002100ndash 150 980 0067150ndash 200 1038 0063200ndash 250 1033 0010

The distance to road (m) 0 ndash 25 2109 0059 10025ndash 50 2020 001050ndash 75 1952 minus 002875ndash 100 1946 minus 0029100ndash 125 1972 minus 0013


822019 Science 69


where p is the probability thatthe dependent variable ( Y ) is1 p( 1 minus p)is the so-called odd or frequency ratio C 0 is the intercept and C 1 C 2 hellip C n are coef cients which measure the contribution of the independentfactors ( X 1 X 2 hellip X n ) to the variations in Y (Lee 2005)

The spatial association between landslide inventory and thelandslide factors maps (geology slope aspect land use elevationdistance to stream and distance to road) was assessed using thelogistic regression method The statistical assessment was carried out

Fig 5 The landslide susceptibility map developed using Wi method

Fig 6 The landslide susceptibility map developed using Wf method


822019 Science 69


using an IDRISI GIS environment In this assessment a logisticregression equation was obtained as shown in Eq (7) and the LR map was produced ( Fig 7)

Y = minus 47485 + 0 000929 AtildeGeologyminus 0052129 AtildeLanduse

+ 0 000503 AtildeAspect

+ 0 050838 AtildeSlopeminus

0006434 AtildeRoad

minus 0239024 AtildeElevation minus 0098631 AtildeStream eth7

THORN

The statistical results of the logistic regression method are abridgedin Table 4 A key starting point could be the model chi-square value of which provides the usual signi cance test for logistic regression It is adifference between minus 2ln L (L=likelihood) for the best- tting modeland minus 2lnL 0 forthe null hypothesisin whichallthe coef cients are setto0 and measures the improvement in t that the independent variablesbring to the regression The high value for the model chi-squareindicates that theoccurrence of landslides is far lesslikelyunder thenullhypothesis (without landslide in uencing parameters) than the fullregressionmodel (where theparametersare included)The goodness of t is analternative to themodel chi-square forassessingthe signi cance

of LR models The calculation is based on the difference between theobserved and the predicted values of the dependent variable Thesmaller this statistic the better t it indicates ( Ayalew and Yamagishi2005 )

The pseudo R2 equal to 1 indicates a perfect t whereas 0 shows norelationship When a pseudo R2 is greater than 02 it shows a relativelygood t ( Clark andHosking1986 )The pseudo R2 in this study is00799In addition a disjunctive approach which is much easierto interpret isto look at how well the model actually predicts the dependent variableIn this case IDRISI uses the so-called Relative Operating Characteristic(ROC) tocomparea Boolean map of ldquoreality rdquo (thepresence orabsenceof landslides) with the probability map The ROC value ranges from 05 to1where 1 indicatesa perfect t and 05represents a random t ( Ayalewand Yamagishi 2005 Akgun and Bulut 2007) A value of 07502 wasobtained in this study which can be taken as a sign of good correlationbetween the independent and dependent variables

According to Eq (7) the geology slope and aspect coef cients arepositive the land use elevation distance to road and distance tostream coef cients are negative This means that the geology slopeand aspect are positively related to the occurrence of a landslidewhereas land use elevation distance to road and distance to streamindicate a negative relation with the landslide occurrence in the studyarea In particular the coef cient that belongs to the parameterldquoslope rdquo strongly departs from 0 and led to the inference that the

topographical slope has a higher effect on the development of landslides than any other parameter

5 Results and comparative analysis

The landslide susceptibility maps were prepared using vedifferent weighting procedures in a GIS-based approach The areaand percentage distribution of the susceptibility classes in the studyarea were determined as a result of the ve different methods To testthe reliability of the landslide susceptibility maps produced by thefrequency ratio AHP Wi and Wf methods and logistic regression alandslide activity map of fty active zones of recent landslides and thesusceptibility maps were compared In these comparisons the area onthe landslide activity map that shows where the landslides occurred is

matchedwith the landslide susceptibility maps Then the distribution

Fig 7 The landslide susceptibility map produced by LR

Table 4Summary statistics of the logistic regression model

Statistics Value

Number of sampled observations 471900 a

2ln L 188596872ln L 0 17353213Goodness of t 285841938Pseudo R2 00799ROC 07502a

(Using 50 m cell size was used to LR analysis)


822019 Science 69


of the actual landslide areas is determined according to the landslidesusceptibility zonesThe landslide susceptibility map has a continuousscale of numerical values and there is a need to separate these valuesinto susceptibility classes There are several mathematical methodsfor the classi cationsusceptibility degrees ( Ayalewet al 2004 Suzenand Doyuran 2004 ) The standard deviation classi er is proposedwhen the histogram of data values exhibits a normal distribution(Suzen and Doyuran 2004 ) As a result the standard deviationclassi er was used since the data values in the landslide susceptibilitymaps obtained using the FRM AHP Wi Wf and LR show a normaldistribution ( Fig 8)

According to the landslide susceptibility map produced from thefrequency ratio method 800 of the total area is found to be of verylow landslide susceptibility Low moderate and high susceptiblezones represent 2238 3693 and 2579 of the total arearespectively The very high landslide susceptibility area is 691 of the total study area

The landslide susceptibility map generated with AHP whichincluded 958 of total area is determined to be of very low landslidesusceptibility Low and moderate susceptible zones make up 1422and 1946 of the total area respectively The high and very highsusceptible zones values are close to each other 2900 and 2774respectively

The landslide susceptibility map created in accordance with thestatistical index ( Wi) contains 984 of the total area which isdesignated to be of very low landslide susceptibility The value is nearto the very low category in relation to the FR and AHP methods Lowmoderate and high susceptibility zones constitute 1995 3646 and3258 of the total area respectively At only 118 the percentage of very high susceptibility area is very small

The landslide susceptibility map produced through the weightingfactor ( Wf ) method involves different values from the other methodsThe very low and low susceptibility areas are very small percentagesat 013 and 116 respectively but the percentages of moderatehigh and very high susceptibility areas are 2310 5324 and2238 respectively

The logistic regression method showed different results whilecontaining high percentages values for the low susceptibility zone(2188) in the LR method the low susceptibility zones percentage inthe FR AHP Wi and Wf methods show small values such as 5181354 321and 176 respectively Thevery lowsusceptible area isdenoted at a value of 266 The moderate high and very highsusceptibility zones show 3289 3094 and 1164 of the wholeareas respectively ( Fig 8) According to the LR method it isdetermined that the geology slope and aspect coef cients arepositive the land use elevation distance to road and distance tostream coef cients are negative This means that the geology slopeand aspect are positively related to the occurrence of landslideswhereas land use elevation distance to road and distance to streamindicate a negative relationshipwith landslideoccurrence in the studyarea

For the veri cation procedure the ve susceptibility maps were rst divided into ve classes based on standard deviations of thecorresponding histograms ( Fig 9) Next they were crossed with thelandslide activity map containing fty active landslide zones Fig 9presents a histogram that summarizes the result of the entire processThe highand very high susceptibility zones (4 and 5) found by the FRAHP Wi and Wf LR methods contain 6098 6271 6256 9329and 4258 of the active landslide zones respectively Fig 9 showsthat the extent of the active landslide zones located in the very high

1 2 3 4 5

Susceptibility classes

N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0

Fig 8 Bar graphs showing the relative distribution of susceptibility levels when the susceptibility maps are classi ed on the basis of standard deviations

0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n

d s l i d e z o n e s

( ) FRAHPWiWfLR

Fig 9 A histogram showing the amount of active landslide zones that fall into the various classes of the FR AHP Wi Wf and LR susceptibility maps


822019 Science 69


susceptibility class is higher in the map of the Wf than the FRAHP Wiand LR maps 4059 of the active landslide zones fall into the veryhigh susceptibility class on the Wf map This value falls to 168 in thecase of the susceptibility map produced by the Wi method Besides6088 of the active landslide zones coincide with the highsusceptibility class on the Wi map The moderate zones (3) of theFR Wi and LR methods include about 30 of the active landslidezones In addition while covering 20 of the moderate class from the

AHP method the Wf method includes only 5 approximately The low(2) and very low susceptible (1) zones contain less than 75 of theactive landslide zones in the FR Wi and Wf methods The values are1711 and 2453 for AHP and LR respectively According to usinganalyses methods the very low susceptible zones include less than 5of the active landslide zones all of the methods (see Fig 9) From theanalysis in Fig 9 it is easy to conclude that the very high and highsusceptibility classes of the Wf map captured the locations of theactive landslide zones (9329) better than the correspondingcounterparts of the FR AHP Wi and LR maps at 6098 62716256 and4258 respectively This might be due to the factthat theWf method presents much more distinct and homogeneous values forwide study areas such as this application Furthermore this might bederived from the landslide inventory map because this map wasproduced as a result of a very high sensitivity study and eld check

6 Discussion and conclusion

The reasons for landslides are many complex convoluted andevery so often unknown Although the basic factors related tolandslides can be observed during eld studies aerial photos andsatellite images interpretations some factors remain closed So as todetermine whether there are closed parameters affecting theoccurrence of landslides several geomorphometrical parameterswere entered into the analyses Most geomorphometrical factors aresubjective and hard to measure quantitatively in the eld Thereforeit may be dif cult to understand their contributions to the landslideoccurrence mechanism Since landslides are among the mostdangerous natural disasters for many years research institutionsworldwide have attempted to assess the landslide hazard determinethe risk and to show its spatial distribution In this context this studyundertook comprehensive research on slope stability assessment andlandslide susceptibility mapping in a part of Trabzon province inTurkey The region is continually at risk of landslides followingprecipitation since the topography and lithological materials are of the very best t to create landslides It is known that the role of precipitation as the triggering mechanisms of landslides is stronglyin uenced by the landscape dynamic and geology

This is primarily because of the problems inherited from landslideinventory maps and the absence of universal guidelines to selectcausal factors In this study the landslide inventorymap waspreparedin such a way that it includes shallow landslides and in considerationof the fact that geology slope and aspect of materials are important

parameters for susceptibility mapping various methods An attemptwas made to differentiate the concepts of landslide susceptibilitymapping Five of the available approaches for landslide susceptibilitymapping used in this study were FR AHP statistical index ( Wi)weighting factor ( Wf ) and LR with the Wf map gives the best resultsTo con rm the practicality of the results the ve susceptibility mapswere compared with 50 active landslidezones The resultwas that theactive landslide zones coincided with a high percentage for the highand very high susceptibility class in the FR AHP Wi and Wf maps butthe values of LR were not in agreement 9329 of these landslidezones fall into the high and very high susceptibility classes of the Wf map The FR AHPWi and LR maps contained 6098 6271 6256and 4258 of the landslide zones respectively In the FR and Wimethods the geology parameter is positively associated with the

occurrence of landslides According to the AHP method the geology

slope and aspect parameters upwards of land use elevation distanceto stream and distance to road are implicated in the occurrence of landslide in the study area The elevation and geology factors arepositively associated with the occurrence of landslide in the Wf method In addition in the LR method the geology slope and aspectare positively associated with the occurrence of landslide whereasland usedistance to stream elevation distance to road appear to havea negative relation with landslide occurrence in the study area Thus it

can be concluded that when

eld conditions and characteristics areaccurately determined by professional expertise the Wf method givesbetter results over larger areas as in this study

References

Akgun A DagS Bulut F 2008Landslide susceptibility mapping fora landslide-pronearea (Findikli NE of Turkey) by likelihood-frequency ratio and weighted linearcombination models Environ Geol 54 1127 ndash 1143

Aleotti P Chowdhury R 1999 Landslide hazard assessment summary review andnew perspectives Bull Eng Geol Env 58 21 ndash 44

Atkinson PM Massari R 1998 Generalized linear modelling of susceptibility tolandsliding in the central Appennines Italy ComputerampGeoscience 24 (4)373ndash 385

Ayalew L Yamagishi H Ugawa N 2004 Landslide susceptibility mapping using GIS-based weighted linear combination the case in Tsugawa area of Agano RiverNiigata Prefecture Japan Landslides 1 73 ndash 81

Ayalew L Yamagishi H 2005 The application of GIS-based logistic regression forlandslide susceptibility mapping in the Kakuda ndash Yahiko Mountains Central JapanGeomorphology 65 15 ndash 31

Ayalew L Yamagishi H Marui H Kanno T 2005 Landslides in Sado Island of JapanPart II GIS-based susceptibility mapping with comparisons of results from twomethods and veri cations Eng Geol 81 432 ndash 445

Begueriacutea S 2006 Changes in land cover and shallow landslide activity a case study inthe Spanish Pyrenees Geomorphology 74 196 ndash 206

BlijenbergH 1998 Rolling Stones Triggering and Frequency Of Hillslope Debris Flowsin the Bachelard Valley Utrecht University Utrecht Southern French Alps

Cannon SH 2000 Debris ow response of southern California watersheds burned bywild re In Wieczorec GF Naeser ND (Eds) Debris Flow Hazards MitigationMechanics Prediction and Assessment Balkema Rotterdam pp 45 ndash 52

Carrara A 1983 A multivariate model for landslide hazard evaluation Math Geol 15403ndash 426

Cevik E Topal T 2003 GIS-based landslide susceptibility mapping for a problematicsegment of the natural gas pipeline Hendek (Turkey) Environ Geol 44 949 ndash 962

Chung C-JF Fabbri AG 2003 Validation of spatial prediction models for landslidehazard mapping Nat Hazards 30 (3) 451 ndash 472

Clark WA Hosking PL 1986 Statistical Methods for Geographers John Wiley andSons New York 518 pp

Clerici A Perego S Tellini C Vescovi P 2002A procedurefor landslidesusceptibilityzonation by the conditional analysis method Geomorphology 48 349 ndash 364

Conoscenti C Ciprioano DM Rotigliano E 2008 GIS analysis to assess landslidesusceptibility in a uvial basin of NW Sicily (Italy) Geomorphology 94 325 ndash 339

Dai FC Lee CF Li J Xu ZW 2001 Assessment of landslide susceptibility on thenatural terrain of Lantau Island Hong Kong Environ Geol 43 (3) 381 ndash 391

Ercanoglu M Gokceoglu C 2004 Use of fuzzy relations to produce landslidesusceptibility map of a landslide prone area (West Black Sea Region Turkey) EngGeol 75 (3ndash 4) 229 ndash 250

Galli M Ardizzone F Cardinali M Guzzetti F Reichenbach R 2008 Comparinglandslide inventory maps Geomorphology 94 268 ndash 289

Garciacutea-Rodriacuteguez MJ Malpica JABenito B Diacuteaz M 2008Susceptibilityassessmentof earthquake-triggered landslides in El Salvador using logistic regressionGeomorphology 95 172 ndash 191

Guzzetti F Carrara A Cardinalli M Reichenbach P 1999 Landslide hazardevaluation a review of current techniques and their application in a multi-scalestudy Central Italy Geomorphology 31 181 ndash 216

Guzzetti F Reichenbach P Cardinali M Galli M Ardizzone F 2005 Landslidehazard assessment in the Staffora basin northern Italian Apennines Geomorphol-ogy 72 272 ndash 299

Guzzetti F Reichenbach P Ardizzone F Cardinali M Galli M 2006 Estimating thequality of landslide susceptibility models Geomorphology 81 166 ndash 184

Ildir B 1995 Turkiyede heyelanlarin dagilimi ve afetler yasasi ile ilgiliuygulamalarInOnalp A (Ed) Proceedings of 2nd National Landslide Symposium Turkey SakaryaUniversity pp 1 ndash 9

Kanungo DP Arora MK Sarkar S Gupta RP 2006 A comparative study of conventional ANN black box fuzzy and combined neural and fuzzy weightingprocedures for landslide susceptibility zonation in Darjeeling Himalayas Eng Geol85 347ndash 366

Komac M 2006A landslide susceptibility model usingthe analytical hierarchyprocessmethod and multivariate statistics in perialpine Slovenia Geomorphology 74 (1 ndash

4) 17ndash 28Lee S Min K 2001 Statistical analysis of landslide susceptibility at Yongin Korea

Environ Geol 40 1095 ndash 1113Lee S Choi J Min K 2004 Probabilistic landslide hazard mapping using GIS and

remote sensing data at Boun Korea Int J Remote Sens 25 (11) 2037 ndash 2052


822019 Science 69


Lee S 2005 Application of logistic regression model and its validation for landslidesusceptibility mapping using GIS and remote sensing data Int J Remote Sens 26(7) 1477 ndash 1491

Lee S Talib JA 2005 Probabilistic landslide susceptibility and factor effect analysisEnviron Geol 47 982ndash 990

Lee S Sambath T 2006 Landslide susceptibility mapping in the Damrei Romel areaCambodia using frequency ratio and logistic regression models Environ Geol 50(6) 847 ndash 855

Nandi A Shakoor A 2009 A GIS-based landslide susceptibility evaluation usingbivariate and multivariate statistical analyses Eng Geol 110 11 ndash 20

Nefeslioglu HA Gokceoglu C Sonmez H 2008 An assessment on the use of logistic

regression and arti cial neural Networks with different sampling strategies for thepreparation of landslide susceptibility maps Eng Geol 97 171 ndash 191Nielsen TH Wrigth Rh Vlasic TC Spangle WE 1979 Relative slope stability and

land-use planning in the San Francisco Bay region California US Geological SurveyProfessional paper 944

Oztekin B Topal T 2005 GIS-based detachment susceptibility analyses of a cut slopein limestone Ankara mdashTurkey Environ Geol 49 124 ndash 132

Re ce A Capolongo D 2002 Probabilistic modeling of uncertainties in earthquakeinduced landslide hazard assessment Comput Geosci 28 735 ndash 749

Reis S Yomralioglu T 2006 Detection of current and potential hazelnut (corylus)plantation areas in Trabzon North East Turkey using GIS amp RS J Environ Biol 27(4) 653 ndash 659

Saaty TL 1977 A scaling method for priorities in hierarchical structures J MathPsychol 15 234 ndash 281

Saaty TL 1980 The Analytical Hierarchy Process McGraw Hill New YorkSaaty TL Vargas GL 1991 Prediction Projection and Forecasting Kluwer Academic

Publishers DordrechtSaaty TL Vargas GL 2001 Models Methods Concepts and Applications of the

Analytic Hierarchy Process Kluwer Academic Publisher Boston

Saha AK Gupta RP Arora MK 2002 GIS-based landslide hazard zonation in theBhagirathi (Ganga) valey Himalayas Int J Remote Sens 23 (2) 357 ndash 369

Suzen ML Doyuran V 2004Data driven bivariatelandslidesusceptibilityassessmentusing geographical information systems a method and application to Asarsuyucatchment Turkey Eng Geol 71 303 ndash 321

Tangestani MH 2004 Landslide susceptibility mapping using the fuzzy gammaapproach in a GIS Kakan catchment area southwest Iran Aust J Earth Sci 51439ndash 450

TUIK 2008 Prime MinistryRepublicof Turkey Turkish Statistical Institute httpwwwturkstatgovtr cited 15 September 2007

van Westen CJ 1997 Statistical landslide hazard analysis ILWIS 21 for Windows

application guide ITC Publication Enschede pp 73 ndash 84Voogd H 1983 Multicriteria Evaluation for Urban and Regional Planning Pion LtdLondon

Wu YH Swanston DN 1980 Risk of landslides in shallos soils and its relations toclearcutting in southeastern Alaska For Sci 26 495 ndash 510

Yalcin A 2007 Environmental impacts of landslides a case study from East Black Searegion Turkey Environ Eng Sci 24 (6) 821 ndash 833

Yalcin A Bulut F 2007 Landslide susceptibility mapping using GIS and digitalphotogrammetric techniques a case study from Ardesen (NE-Turkey) NatHazards 41 201 ndash 226

Yalcin A 2008 GIS-based landslide susceptibility mapping using analytical hierarchyprocess and bivariate statistics in Ardesen (Turkey) comparisons of results andcon rmations Catena 72 1 ndash 12

YesilnacarE Topal T 2005 Landslide susceptibility mappinga comparison of logisticregression and neural networks methods in a medium scale study Hendek region(Turkey) Eng Geol 79 251 ndash 266

Zhou G Esaki T Mitani Y Xie M Mori J 2003 Spatial probabilistic modeling of slope failure using an integrated GIS Monte Carlo simulation approach Eng Geol68 373ndash 386


822019 Science 69


descriptive expressions and depend on expert opinions The mostcommon types of qualitative methods basically use landslideinventories to recognize sites of comparable geological and geomor-phologic characteristics that are susceptible to failure Howeverweights of the parameters are determined from the knowledge of specialists on the subject and the area The designating weights arethus highly personal and may be include some virtual admissionQuantitative methods are based on numerical expressions of the

relationship between controlling factors and the landslides There aretwo types of quantitative methods deterministic and statistical(Aleotti and Chowdhury 1999 ) Deterministic methods are based onslope stability studies expressed in terms of the safety factor ( Re ceand Capolongo 2002 Zhou et al 2003 ) The statistical approachesanalyze the historical link between landslide-controlling factors andthe distribution of landslides Quantitative methods may be used todecrease the personality and bias in the weight assessment processTherefore more realistic susceptibility maps can be produced from anobjective measure of values During the past few years quantitativemethods have been implemented for landslide susceptibility zonationstudies in different regions ( Clerici et al 2002 Suzen and Doyuran2004 Ercanoglu and Gokceoglu 2004 Yesilnacar and Topal 2005Kanungo et al 2006 Yalcin and Bulut 2007 Garciacutea-Rodriacuteguez et al2008 Nefeslioglu et al 2008 etc)

The aim of this study was to use widely-accepted models astatistical method (frequency ratio model) a multi-criteria decisionmaking approach (AHP) bivariate and multivariate approaches(logistic regression) and evaluate their performances

Frequency ratio model is based on the observed associationsbetween allocation of landslides and each associated factors of landslide occurrence to display the correlation between landslidelocations and the parameters controlling landslide occurrence in thearea ( Lee 2005) Therefore the method gives very good results fordetermining the landslide inventory with rigorous accuracy Theweights of the parameters and the decision alternatives used inproducing landslide susceptibility map are determined with the AHPWhen these weights are determined both the comparison of theparameters relative to each other and determination of the effectvalues of the decision alternatives namely the sub-criteria are basedon a landslide inventory map obtained with the help of aerial photosand satellite images As a result of dual comparisons a pair-wisecomparison matrix is obtained for each parameter and sub-criteriaConsequently the weight values were determined correctly for thereal land data It has been shown that the use of the AHP methodproduces a practical and realistic result to de ne the factor weights inthe landslide susceptibility model In statistical models usingbivariate or multivariate techniques for landslide susceptibilityanalysis is widespread ( Nandi and Shakoor 2009 ) There are anumber of ways to apply bivariate and multivariate statistics to assesslandslide susceptibility of a region More than a few instabilityparameter variables are used in the present bivariate approach thein uence of each variable on the occurrence of landslide is evaluated

independently and the variables are combined in the form of a uniqueequation ( Conoscenti et al 2008 Nandi and Shakoor 2009 ) Inmultivariate approaches logistic regression was detected to be themost appropriate approach for the present study In this analysisspatial distribution of landslides is assessed on the basis of interactionof onlystatistically signi cant instability factordata insigni cant dataare excluded from consideration Additionally logistic regressionanalysis is free of data distribution issues and can handle a variety of datasets such as continuous categoricaland binary common types of instability factor data used in landslide studies ( Dai et al 2001 Leeand Min 2001 Lee and Sambath 2006 Akgun et al 2008 Nandi andShakoor 2009 ) Anyeffort to make certain landslide susceptibility in aregion needs proper validation Con rmation should establish thequality (ie consistency robustness degree of tting and prediction

skill) of the proposed susceptibility estimate The excellence of a

landslide susceptibility model can be ascertained using the samelandslide data used to obtain the susceptibility estimate or by usingindependent landslide information not available to construct themodel ( Chung and Fabbri 2003 Guzzetti et al 2005 2006 ) In thispaper we provide a comprehensive validation of a landslidesusceptibility model prepared through ve different methods for theTrabzon City ( Fig 1)

2 Description of the study area

The study area in Trabzon province consisted of approximately4660 km 2 located between 39deg 15 prime and 40deg 15 prime west ndash east longitudesand 41deg 8 prime and 40deg 30 prime north ndash south latitudes in the middle of EasternBlack Sea region ( Fig 1) Altitudes reach 3400 m in parts of the regionand steep slopes are very common The climate is characteristic of theBlack Sea region with temperate climate summers and a rainy seasonnormally lasting from September to April Nevertheless the rainfallregime is irregular with some periods of rare precipitation with long-lasting heavy rains Very intense precipitation has caused disastrous ash oods in river basins and many landslides on slopes ( Reis andYomralioglu 2006 Yalcin 2008 ) According to observations result 22yearly in Trabzon province February is the coldest month with anaverage temperature of 67 degC the hottest month is August with antemperature average of 232 degC The annual precipitation of Trabzoncity is 838 mm and precipitations disperse every month symmetricalThe main commercial agricultural products in the region are hazel-nuts and green tea Apart from the agricultural areas the other mainland cover types are forest and pasture The population of theprovince was 740569 in 2007 year ( TUIK 2008)

3 Thematic data layers

Thestudy began with the preparation of a landslide inventorymapbased on extensive eld work a previous inventory map and satelliteimages Furthermore the following seven possible landslide causinglayers lithology slope aspect land cover elevation distance tostream anddistance to road were analyzed for landslide susceptibilitymapping using the logistic probability method (frequency ratiomethod mdash FRM) the analytical hierarchy process (AHP) bivariate(Wi and Wf ) and logistics regression (LR) methods Finally thesusceptibility maps produced from the ve different methods werecompared and evaluated using validation data sets

31 Landslide inventory map

Themaps show the locationsand properties of landslides that havemoved in the past These slope failures were related to geologicaltopographical and climatic conditions thus they can often facilitatethe prediction of locations and conditions of future landslides For thisreason it is important to determine the location and area of thelandslide accurately when preparing the landslide susceptibility

maps Landslide susceptibility assessment is performed in a range of phases The initial phase is identifyingand evaluating landslide-proneareas and constructing a landslide inventory map for future useLandslide inventory mapping is the systematic mapping of existinglandslides in a region using different techniques such as eld surveyair photosatellite image interpretation and literature search forhistorical landslide records A landslide inventory map provides thespatial distribution of locations of existing landslides The landslidesin the study area were determined by comprehensive eld surveysThe landslides which are currently inde nite in characteristics andboundaries were identi ed using old dated satellite images As aresult the satellite images were very useful in determination of landslides inventory map ( Yalcin and Bulut 2007 ) In this study thesusceptibility mapping started with the preparation of an inventory

map of 250 landslides from eld studies a previous inventory map


822019 Science 69







822019 Science 69



32 Lithology map


33 Slope map


34 Aspect



35 Elevation




36 Land cover






Frequencyratio (ba)


Jlh 067 000 000γ 2 163 304 187Kk 007 000 000












Carboniferous units


822019 Science 69























822019 Science 69















822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69







822019 Science 69



32 Lithology map


33 Slope map


34 Aspect



35 Elevation




36 Land cover






Frequencyratio (ba)


Jlh 067 000 000γ 2 163 304 187Kk 007 000 000












Carboniferous units


822019 Science 69























822019 Science 69















822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69



32 Lithology map


33 Slope map


34 Aspect



35 Elevation




36 Land cover






Frequencyratio (ba)


Jlh 067 000 000γ 2 163 304 187Kk 007 000 000












Carboniferous units


822019 Science 69























822019 Science 69















822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69























822019 Science 69















822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69















822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69



Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights


(6) Jlh 12 13 14 19 1 1 0037(7) γ 2 2 12 13 18 2 2 1 0070(8) Kk 12 13 14 19 1 1 12 1 0037

















(4) Elevation 14 13 1 1 0098


822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69













Factors 1 2 3 4 5 6 7 8 9 10 11 12 Weights





822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69






eth4THORN

where









where




layers







1minus p eth5THORN





γ 2 304 0625Kk 000 0000




500ndash






822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69







822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69






0006434 AtildeRoad


THORN











Statistics Value





822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69









1 2 3 4 5


N u m

b e r o

f p

i x e l

FRAHPWiWfLR

3E+07

3E+07

2E+07

2E+07

1E+07

5E+06

0


0

10

20

30

40

50

60

70

1 2 3 4 5


A c

t i v e l a n


( ) FRAHPWiWfLR



822019 Science 69












References






























822019 Science 69





























822019 Science 69












References






























822019 Science 69





























822019 Science 69





























Documents

Science 69