quantitative assessment of landslide hazard along transportation lines using historical records

Landslides (2011) 8:279–291DOI 10.1007/s10346-011-0252-1Received: 25 March 2010Accepted: 9 January 2011Published online: 27 January 2011© Springer-Verlag 2011

Pankaj Jaiswal I Cees van Westen I Victor Jetten

Quantitative assessment of landslide hazardalong transportation lines using historical records

Abstract In this paper, a quantitative landslide hazard model ispresented for transportation lines, with an example for a road andrailroad alignment, in parts of Nilgiri hills in southern India. Thedata required for the hazard assessment were obtained fromhistorical records available for a 21-year period from 1987 to 2007.A total of 901 landslides from cut slopes along the railroad androad alignment were included in the inventory. The landslideswere grouped into three magnitude classes based on the landslidetype, volume, scar depth, and run-out distance. To calculatelandslide hazard, we estimated the total number of individuallandslides per kilometer of the (rail) road for different returnperiods, based on the relationship between past landslides(recorded in our database) and triggering events. These weremultiplied by the probability that the landslides belong to a givenmagnitude class. This gives the hazard for a given return periodexpressed as the number of landslides of a given magnitude classper kilometer of (rail) road. The relationship between the totalnumber of landslides and the return period was established usinga Gumbel distribution model, and the probability of landslidemagnitude was obtained from frequency–volume statistics. Theresults of the analysis indicate that the total number of landslides,from 1- to 50-year return period, varies from 56 to 197 along therailroad and from 14 to 82 along the road. In total, 18 hazardscenarios were generated using the three magnitude classes andsix return periods (1, 3, 5, 15, 25, and 50years). The hazardscenarios derived from the model form the basis for future directand indirect landslide risk analysis along the transportation lines.The model was validated with landslides that occurred in the year2009.

Keywords Landslide inventory . Historical record . Magnitudeclass . Quantitative hazard assessment . Transportation line .

Nilgiri

IntroductionTransportation lines such as roads and railroads in mountainousterrain are often prone to landslides which may occur on cut ornatural slopes. Any damage to these facilities due to landslidescan result in consequences, which can be direct or indirect.

Along transportation lines, landslides may occur frequentlyfrom untreated cut slopes (Dai and Lee 2001). Landslidesinitiating from cut slopes are generally small in size but poseconsiderable risk to life and property. To quantify risk along aroad or a railroad, we require the estimation of the frequency oflandslides and the degree of loss to specific elements at riskresulting from the specified landslide magnitude (AGS, AustralianGeomechanics Society and Sub-committee on Landslide RiskManagement 2000; Fell et al. 2005, 2008; van Westen et al. 2006).Researchers have used different statistical models to estimatefrequency of landslides, such as those based on the magnitude–frequency relationship of landslides (Hungr et al. 1999), exceed-ance probability of rainfall threshold (Jaiswal and van Westen

2009), and exceedance probability of landslides based on Poissonor binomial probability models (e.g., Coe et al. 2000; Guzzetti etal. 2002, 2005). The models based on rainfall threshold provide anestimate of the probability of occurrence of one or more rainfallevents that can trigger landslides, whereas those based on theoccurrence of past landslides provide exceedance probability ofone or more landslides that can occur in an area. Both threshold-and landslide-based models are ultimately used to obtain theprobability of occurrence of one or more landslides in a specifiedtime period. For transportation lines, specific information on theexpected number of landslides and their annual probability ofoccurrence (or return period) are important for estimating directand indirect risk. This information is used to estimate theprobability of a landslide hitting a moving vehicle or a commuter(AGS, Australian Geomechanics Society and Sub-committee onLandslide Risk Management 2000; Wilson et al. 2005) and tocalculate the blockage time by estimating the total volume ofdebris on the transportation line.

An estimate of the frequency of a specific number oflandslides per unit area can be made if a relationship can beestablished between the number of landslides triggered by anevent and the probability of occurrence of this event. To establishsuch a statistical relationship, we require a continuous record oflandslide inventory. Some technical offices, such as road andrailroad maintenance departments, or geotechnical offices, pro-duce inventories containing continuous records of landslideoccurrences within a region or along transportation lines. Mostoften they refer to landslides from cut slopes and fills alongtransportation lines (Fell et al. 1996). The main advantage of suchrecords is that they are prepared shortly after the occurrence of alandslide-triggering event thereby recording most of the landslidesin an area, but the major limitation remains the availability andaccessibility of such data. The unavailability of a continuous recordof landslides is one of the major drawbacks in the quantitativeassessment of landslide hazard (van Westen et al. 2006).

In this, we assessed hazard from landslides originating from cutslopes along a road and a railroad in the Nilgiri hills of Tamil Nadu,India. We used the historical data from the railroad maintenanceunit and geotechnical office to obtain a substantially completelandslide inventory on cut slopes. The landslide hazard descriptorfor roads and railroads, as recommended by Joint TechnicalCommittee on landslides and Engineered Slopes, JTC-1 guidelines(Fell et al. 2008), is expressed as the number of landslides of a givenmagnitude (area or volume) per annum per kilometer of cut slopes.We used this descriptor in our analysis and have tried to quantify thehazard based on historical information.

The study areaThe study was carried out along a transportation corridor insouthern India. Figure 1 shows the location of the 17-km-longrailroad and 24-km-long national highway road (NH-67) con-necting Mettupalayam and Coonoor in the Nilgiri hills, in the

Cees J. van Westen I Victor Jetten

Landslides 8 & (2011) 279

Original Paper

western Tamil Nadu region of southern India. The railroad wasconstructed in the late nineteenth century and became opera-tional in 1899.

Geologically, the transportation lines expose gneisses belong-ing to the Charnockite Group of Archaean age (Seshagiri andBadrinarayanan 1982), overlain by weathered soil, exposed alongcut slopes of the road, railroad, and landslide scarps. The gneisses arelight bluish gray in color and medium to coarse grained and showonly crude foliation on fresh surfaces. However, when weatheredthese rocks show well-developed foliation planes. The regional strikeof the foliation is ranging from ENE–WSW to E–W direction withmoderate to steep dips. The sub-tropical climate and intense physicaland chemical weathering have resulted in a thick yellowish toreddish brown soil. Borehole data from a location south of Katterireveals that the maximum thickness of the weathering profile isabout 32 m (Seshagiri and Badrinarayanan 1982). Along the road andthe railroad, the weathering soil thickness varies from less than ameter to 20 m, as observed in cut slopes.

The area has two distinct landforms, namely moderatelydissected plateau remnants to the west of Marapallam and highlydissected denudational slopes to the east of Marapallam (Seshagiriand Badrinarayanan 1982). The elevation difference of the trans-portation lines, from Kallar farm to Coonoor, is about 1,300 m. Thearea forms part of the Coonoor river basin and is characterized by asub-dendritic drainage pattern.

Both transportation lines pass mostly through reserved forestand tea plantations. To the west of Marapallam, the road and therailroad go through gentle slopes covered by tea plantations, whereasto the east of Marapallam, they are passing through steep slopescontaining patches of forest.

Generation of the landslide inventory along transportation linesThe landslide inventory along the road and the railroad wasprepared by interpreting historical records available with differentorganizations in the Nilgiri area, which include a railroadmaintenance register, a summary table of landslides along therailroad, and several technical reports.

Data sources and types of informationThe main continuous record of landslides along the railroad wasavailable from the railroad maintenance register (locally called“railway slip register”). Since 1992, data were present in a paperform recorded and maintained by the Southern Railway office atCoonoor. The railway slip register is updated soon after theoccurrence of a landslide-triggering event and is used fortendering contracts for railroad clearance. It contains data onthe spatial distribution of landslide debris on the railroad. Thetype of landslide data compiled from the register is shown inTable 1. The location of landslides is referenced with respect to thenearest telephone posts, which are used as permanent markersalong the railroad by the railroad authorities. The averagedistance between two telephone posts is 50 m. The landslidedescription contains the type of material (e.g., earth mixed withboulder), the total volume of debris on the railroad, and the dateof occurrence.

From 1987 to 1991, landslide data were only available as a“landslide summary table” maintained by the Southern Railwayoffice, which contains the spatial distribution of landslide debrison the railroad in different months. Landslides prior to 1987 werenot available for the study area. The type of information onlandslides available in the “landslide summary table” is shown in

Fig. 1 Location of the road and the railroad alignment in the Nilgiri hills of southern India. Black circles are the location of landslides on cut slopes. The numbersindicate the kilometer marks along the railroad and the road

Original Paper

Landslides 8 & (2011)280

Table 1. The summary table does not contain information on thedate of occurrence of landslides, type of material involved, andtotal volume of debris on the railroad. It only provides thelocation of landslides with respect to the nearest telephone post aswell the month and year of landslide occurrence.

Apart from the above-mentioned sources, landslide informa-tion was also collected from published and unpublished technicaldocuments of landslide investigations. Most of the reportscontained detailed geotechnical investigation of major landslidesin the Nilgiri area. Some reports contain inventories of landslidesalong the road (NH-67). The oldest inventory record was from theyear 1987. All reports provide some basic information onlandslides but the type of information varies depending on thepurpose of investigation and the time and resources that wereavailable to carry out the investigation. Table 1 gives the type oflandslide data compiled from the technical reports.

Method used for inventory mappingAfter selecting the data sources, all relevant data pertaining tolandslides on cut slopes was compiled in a spreadsheet, whichformed the basis for field mapping where all the data related toeach specific location were checked and updated. All landslidesites compiled from the historic archives were visited in order toidentify the landslide scars based on the description provided inthe historical records. The data on landslide volume were used toinfer the size of landslides. The morphological signatures left bylandslides on slopes, such as scarp faces and concave slopes, wereused to identify the shape of landslides. Some of the landslidescars and deposition areas were not clearly discernable due to theremoval of debris and remedial works. To overcome this, five

criteria were adopted for the identification of landslides: (1)changes in the slope morphology characterized by slope concavityor surface irregularities, (2) presence of scarp faces with contrastin vegetation cover with the surrounding areas, (3) presence ofretaining walls or other slope stabilization structures, (4) presenceof removed debris on the down slope part, and (5) informationfrom local people.

After identifying the location of landslides, they were mappedon a 1:10,000 scale topographic map. The initiation (source) anddepositional area of landslides were separately marked. Themorphological parameters (landslide scar length, width, anddepth) were plotted after carefully measuring them in the fieldusing a meter tape. Additional data such as type of landslide, run-out distance, present land use, thickness of weathering profile,probable cause, and damage details were also added to theinventory. The mapped landslides were digitized as polygons orpoints and entered in a geo-database of Arc GIS. Table 1 shows thetype of landslide information obtained after the field check.

Data were compiled from the historical records covering a 21-year period from 1987 to 2007. Out of the 901 landslides, 565landslides (63%) were obtained from railway slip registers (from1992 to 2007), 220 (24%) from the railroad landslide table (from1987 to 1991), and 116 (13%) from technical reports (from 1987 to2007). Through field mapping, it was possible to identify andmap 67% of the compiled landslides of the railway slip registerand technical reports. Some of the smaller landslides with avolume less than 20 m3 were not identifiable in the field due topossible reactivations which have obliterated the earlier morphol-ogy. The location and volume of these small landslides weretherefore taken directly from the original source data. Since they

Table 1 Types of information on landslides compiled from the historical records (railway slip register, railroad landslide table, and technical reports) and inventoryobtained after field check

Railway slip register Railroad landslide table Technical reports Inventory after field check

Date Kilometer mark Location of slide Identification number

Between stations Telephone post Data source Location of slide

Kilometer mark Month Latitude of slide Type of slide

Telephone Post Longitude of slide Date of occurrence

Volume of earth Type of slide Length of slide (m)

Volume of boulders Fresh (date) Width of slide (m)

Damage detail Reactivated (date) Scar depth (m)

Restoration date Slope Area (m2)

Remarks, if any Soil type Volume (m3)

Rainfall in mm Run-out distance (m)

Length of slide (m) Land cover/land use

Width of slide (m) Regolith thickness (m)

Height of slide (m) Probable cause

Depth of slide (m) Remarks (including damages, etc.)

Crown height (m)

Probable cause

Remarks (including damages, overburdenthickness, etc.)


were small and located along the road or the railroad, it waspresumed that most of the released material from these landslideswas accumulated on the road and the railroad. Therefore, themeasured volume from the maintenance records was considered agood representation of the size of these small landslides. Thelocation of the mapped landslides on cut slopes, indicated aspoints, along the railroad and the road is shown in Fig. 1. Thenumber of recorded landslides along the road is relatively lessthan along the railroad. The relative lack of data along the roadmight be due to the fact that smaller landslides are not reportedas they do not cause damage to the road itself.

Landslide characteristicsThe landslides are shallow translational debris slides and mostlytriggered by retreating monsoon rainfall during the period fromOctober to December, with a maximum (54%) during the monthof November. Figure 2 shows the annual distribution of recordedlandslides. Landslides occur annually (except in 1995) with anaverage rate of 43 landslides per year. In 1995, the rainfall was lessthan the required threshold value for landslide initiation, andhence, no landslide occurred (Jaiswal and van Westen 2009).Major landslide activities were observed during 1987, 1992, 2001,and 2006. Landslides occur both as a first time or reactivatedfailures. Reactivations were common on cut slopes that providemore surface area for slope failures to retrograde.

Examples of the type of landslides that are common along theroad (Fig. 3a) and the railroad (Fig. 3b) are shown in Fig. 3. On cutslopes, these landslides initiate as a small slope failure due toexcavation and develop into landslides of larger dimension withpassage of time. The slided material mostly consists of weatheringsoil mixed with debris of weathered gneissic rock. Failures arecommon during rains due to the buildup of pore water pressureson the contact of the weathering soil and weathered rock exposedin cut slopes (Iverson 2000). Even though the majority of theslides are small, they are capable of causing extensive damage tothe infrastructure properties and the temporal blockage of the(rail) road thus resulting in direct and indirect losses.

Table 2 summarizes the main characteristics of landslides oncut slopes along the railroad and the road. The inventory from therailroad landslide table is not added in the statistics shown inTable 2 because of the unavailability of data on volume.Landslides are generally small in size (median volume <200 m3)but occur with a high frequency. In some sections of the railroad,the density of slope failures exceeds 50 landslides per kilometerand for the road ten landslides per kilometer.

From historical records, we have differentiated 110 landslideevents. The term “landslide events” is used here for days whenone or more landslides were triggered by rainfall. Out of 110landslide events, the actual date was only known for 86 events.For 24 events, the month of occurrence was available (taken fromthe railroad landslide table). It is assumed that in the latter case,only one event had occurred each month. Table 3 provides themain characteristics of all landslide events that have affected thearea in the 21 years from 1987 to 2007. The number of landslidesper event varies from one to 148, and the number of events peryear varies from one to 11.

Since the railroad maintenance record is prepared specificallyfor tendering contracts of debris clearance and is updated soonafter each landslide-triggering event, we are certain that itcontains practically all landslides that occurred along the railroad.The presence of data on a substantial fraction of small landslides(median volume ~20 m3, see Table 2) and the availability of arecord of 110 events in 21 years time (average ~5.2 events peryear), including events that triggered even one landslide, clearlyindicate that the landslide inventory along the railroad issubstantially complete at least for the time period from 1987 to2007.

For the road, on the contrary, the technical reports providedinformation on only seven events. From the description given insome of the technical reports, it is evident that few events thathave resulted in small landslides along the road are not recordedas they did not cause damage to the road. However, for two events(i.e., 14 December 1987 and 14 November 2006), a completelandslide investigation was carried out so that a substantiallycomplete inventory was prepared.

Method for landslide hazard assessmentThe presence of a fairly complete landslide inventory allowed toestimate the probability of occurrence of landslides (hazard)along cut slopes by analyzing the total number of landslidesexpected per kilometer for different return periods and multi-plying this by the probability that the landslides belong to a givenmagnitude class.

For the hazard calculation, we assumed that the probabilityof landslide occurrence can be calculated directly from thecomplete landslide inventory by analyzing the number ofexpected landslides per kilometer of the (rail) road for differenttriggering events. We have used the Gumbel method forfrequency–magnitude analysis in which the magnitude is repre-sented as the number of landslides per kilometer. The volume ofexpected landslides was analyzed separately using the volume–frequency analysis. Given the limitations in the available data, weassumed that the volume estimation is independent of thenumber of landslides triggered by the event.

Probability of landslide volumeIt is recommended in the JTC-1 guidelines (Fell et al. 2008) thatlandslide hazard should also include information on the landslidemagnitude. Hungr (1997) argued that in literature, no uniquemeasure of landslide magnitude is available and proposed touse landslide damage (destruction) as a measure of landslidemagnitude. Damage caused by a landslide largely depends onthe velocity and impact pressure of the mass, but theseparameters are extremely difficult to obtain and to integrateFig. 2 Annual distribution of recorded landslides from 1987 to 2007

Original Paper


in the hazard zoning. Therefore, researchers have used onlylandslide area or volume as a proxy for landslide magnitude (e.g., Guzzetti et al. 2005). In this study, the availability of dataon volume allows using it as a proxy for landslide magnitude.Velocity was not considered, as most of the landslides in thearea are fast moving debris slides. For the probabilitycalculation, we grouped all landslides on cut slopes into threemagnitude classes (i.e., M-I, M-II, and M-III) based onlandslide type, volume, and other characteristics such as scardepth, run-out distance, and depth of accumulated debris(Table 4).

In order to estimate the probability of landslide volume,we first studied the probability density distribution of alllandslides for which data on volume were available. Theprobability density p(VL) was computed using the equationgiven by Malamud et al. (2004) as:

pðVLÞ ¼ 1NLT

dNL

dVLð1Þ

where δNL is the number of landslides with volumes between VL

and VL+δNL and NLT is the total number of landslides.Figure 4 shows the probability density distribution computed

using Eq. 1. The observed probability distribution shows a distinct“rollover effect” for landslide volumes less than 10 m3. Therollover may be due to the fact that not all of the very small-scalelandslides, particularly along the road, were reported.

In many studies, the relationship of landslide size andfrequency shows a similar distribution pattern as observed in

Fig. 4, with a power law for landslides of large sizes and a linearcurve for those whose size is small (e.g., Stark and Hovius 2001;Guthrie and Evans 2004; Malamud et al. 2004; Brardinoni andChurch 2004; Catani et al. 2005). The studies concluded that therollover or flattening of the curve is either a real effect reflectingslope stability processes (Guthrie and Evans 2004) or it could alsobe due to the incompleteness of the inventory (Malamud et al.2004; Brardinoni and Church 2004; Catani et al. 2005) as smallerlandslide deposits are either removed or are not discernible. Someauthors have fitted various functional relationships to the rollover(Stark and Hovius 2001; Malamud et al. 2004), while few explainthe distribution with two power law trends having different slopevalues (Brardinoni and Church 2004). Brardinoni and Church(2004) have also shown an increase in the frequency of smalllandslides when the photo-interpreted inventory was integratedby intensive field-based inventory.

In this study, the analysis of the landslide inventory suggeststhat the number of landslides and the range of volume varyaccording to landslide events. To investigate the frequency–volume distribution further, we selected landslide inventoriesfrom the railway slip register only because of the availability ofcontinuous records of landslides, and we analyzed the probabilitydensity distribution for each year separately. Ideally, if a sufficientnumber of landslides are available for every landslide event, thenit is better to consider landslide event inventories for suchanalysis. However, in this case, the number of landslides triggeredby individual landslide events is not sufficient for plotting theprobability density for each event individually as some eventscontain only a few landslides. We therefore considered the totallandslides per year in the analysis. We computed probabilitydensity distribution curves for each year separately using Eq. 1.Thus, for the 15-year data (1992 to 2007), we obtained 15curves. Based on the analysis of the obtained curves, threedistinct distribution patterns of the probability density wereobserved (Fig. 5): type I pattern, where the probability densitydistribution of volumes shows a power law distribution for thewhole data range (e.g., 1993, 1994, 1997, 1998, 1999, 2001, 2002,2003, and 2007); type II pattern, where the distribution exhibitsdistinct rollover for volumes less than 30 m3 (e.g., 1992, 2004,and 2005); and type III pattern, where distribution shows aflattening of the curve for volume less than 100 m3 (e.g., 1996,2000, and 2006).

The three distinct patterns shown in Fig. 5 indicate thatthe probability distributions are not the same in all yearsthough all landslides were triggered by rainfall events and

Fig. 3 Examples of landslides on cutslopes along the road (a) and therailroad (b)

Table 2 Main characteristics of the landslide inventory along the railroad and theroad

Attributes Railroad Road

Total number of landslides nr 565 116

Total volume of landslides m3 52,500 41,600

Volume of smallest landslide m3 2 2

Volume of largest landslide m3 3600 5250

Average volume of landslides m3 90 360

Median volume of landslides m3 20 160

Standard deviation oflandslide volume

m3 270 700

Average landslide density nr/km 33 4


Table 3 Main characteristics of the multi-temporal landslide inventory

Event year Date of landslide event (date/month (number oflandslides))

Total landslide (nr) Landslide volumeMin (m3) Max (m3) Mean (m3)

1987a Oct (23), Nov (19); Dec (31); 14/12 (43)b 116c 50b 1,050b 238b

1988a Sept (1); Nov (2); Dec (1) 4 – – –

1989a Feb (1); Mar (1); Jul (1); Sep (3); Oct (1); Nov (22); Dec (2) 31 – – –

1990a Jan (5); Feb (1); Mar (3); Oct (29); Nov (10) 48 – – –

1991a Jan (6); Apr (1); May (3); July (1); Oct (2); Nov (46) 59 – – –

1992 14/11 (7); 15/11 (27); 16/11 (17); 20/11 (3); 21/11 (14) 68 6 900 77

1993 10/11 (6); 11/11 (23) 29 10 5,250 324

1994 11/11 (5) 05 2 3,600 883

1995 No event – – – –

1996 17/12 (40) 40 2 2,000 120

1997 3/7 (1); 11/7 (1); 16/10 (1); 10/11 (1); 27/11 (16) 20 6 1,944 133

1998 8 /1 (1); 18/8 (1); 27/9 (1); 9/10 (2); 13/10 (1); 9/11 (2);2/12 (16); 11/12 (5); 12/12 (1)

30 3 240 54

1999 8/2 (1); 14/4 (1); 16/8 (1); 2/9 (1); 8/10 (1); 16/10 (23);6/11 (1); 23/11 (2); 24/11 (1); 29/11 (2)

34 2 900 53

2000 25/9 (1); 14/11 (1); 15/11 (2); 22/11 (1); 24/11 (19); 1/12(2); 31/12 (1)

27 2 600 49

2001 1/1 (1); 31/7 (1); 25/10 (1); 26/10 (3); 27/10 (22); 16/11(4); 17/11 (25); 23/11 (3); 24/11 (1); 25/12 (13); 27/12 (7)

81 2 1,200 53

2002 5/5 (1); 8/10 (19); 2/11 (2); 5/11 (12) 34 4 270 36

2003 20/3 (1); 30/4 (1); 13/5 (1); 29/5 (1); 19/8 (1); 10/11 (1) 06 2 77 17

2004 5/5 (21); 17/10 (1); 20/10 (3); 22/10 (4); 31/10 (14); 2/11(3); 7/11 (2); 9/11 (6); 13/11 (2)

56 5 200 27

2005 7/10 (4); 22/10 (1); 7/11 (2); 13/11 (3); 25/11 (6) 16 2 100 22

2006 17/10 (4); 18/10 (9); 21/10 (2); 22/10 (5); 2/11 (8); 4/11(2); 13/11 (1); 14/11 (148)

179 3 5,000 286

2007 8/4 (7); 27/10 (11) 18 2 420 66

Total 901

a Landslides data taken from the railroad landslide table that only contains information on the number and month of occurrence of landslide (month (nr))b Data based on technical report pertaining landslide inventory along the roadc Out of 116 landslides 78 slides are taken from the railroad landslide table for which volume is not known

Table 4 Landslide volume classes for landslides on cut slopes and the corresponding probabilities based on the analysis of the frequency distributions, separated in twogroups (less than 100 slides per year or more than 100 per year)

Sizeclass

V, range(m3)

Sd (m) RD (m) Ad (m) P, range (average) Potential consequences<100 slides/year

≥100 slides/year

M-I <102 <1 <10 1 0.85 0.39 Minor or no damage to road, rail or house;one can escape unhurt

M-II 102–103 <2 10–50 <2 0.13 0.53 Damage rail and non-RCC structures; minor ormajor injuries; society accept risk

M-III >103 2–8 >50 <5 0.02 0.08 Complete damage to rail and buildings; injuriesmajor or even death in some cases; societylives and tolerates risk

V landslide source volume, Sd depth of scar, RD run-out distance, Ad depth of accumulated debris, P probability of occurrence based on number of landslides per annum

Original Paper


occur on cut slopes along the railroad. Differences are alsoobserved in the maximum volume of landslides. The maximumvolume in most years lies between 100 and 1,000 m3, except in2003 and 2005 where it is less than 100 m3. These variationscould be due to changes in the local terrain conditions and theamount of rainfall, as observed also by Dai and Lee (2001) in acase study of Hong Kong. The authors showed a dependencyof landslide volume and the intensity of rainfall and indicatedthat the number of landslides varies according to the rainfallintensity. In this study, it is observed that in a year if alllandslides are triggered by a low intensity rainfall, then thetotal number of landslides is expected to be few and of smallvolume range in comparison to a very high intensity rainfall.Also the rainfall distribution may vary significantly during oneevent, as was also evidenced in a major event that occurred in2009.

The actual frequency distribution of landslide volumes oncut slopes thus remains unclear, and it is difficult to designspecific frequency distributions of landslide volume for differ-ent triggering events. Due to the large variability in thefrequency of landslide volumes, it is evident that a singleprobability density distribution is not enough to explain theprobability of landslide volumes for all triggering events.Therefore in this study, we calculated the frequency percentageof landslide volumes in different years, and the average valuewas taken for the hazard analysis. We decided to use two setsof probabilities for years with a threshold of 100 landslidesbecause the event inventories indicate that if rainfall triggersless than 100 landslides in a year, then the majority (>55%) hasa volume less than 100 m3. These are the events that occurmore frequently, have lower intensity, and trigger more smalllandslides. But for events resulting in more than 100 landslides(e.g., 14 November 2006), the inventory constitutes of land-slides with a volume greater than 100 m3. Such events occurless frequently, but due to the high rainfall amounts, they cantrigger more landslides with large volumes.

The frequency analysis of landslide volumes in differentyears allowed us to conclude that for years with less than 100landslides, the probability of occurrence of landslides with

volumes less than 100 m3 varies from 0.5 to 1 (average=0.85).For volumes ranging from 100 to 1,000 m3, the probabilitiesvaried from 0.01 to 0.33 (average 0.13) and for volumes largerthan 1,000 m3 from 0 to 0.16 (average=0.02). For years withmore than 100 landslides, the following probability values wereused: 0.39 for <100 m3, 0.53 for 100 to 1,000 m3, and 0.08 for>1,000 m3.

Estimation of landslide frequency and return periodThe number of landslides varies significantly in differentsections of the railroad and the road. For example, duringthe period from 1987 to 2007, a total of 785 landslides haveoccurred along the railroad of which the lowest number wasrecorded at km 26 (14 landslides) and the highest at km 12 (101landslides). The number of landslides per year per kilometervaries from 1 to 25 along the railroad. The maximum numberof landslides in a year was recorded in 2006 (25 landslides atkm 11). The inventory suggests that a frequency of landslidesof less than three landslides per year per kilometer is commonin many sections, but extreme events (e.g., >10 landslides peryear per kilometer) occur less frequently. Thus, it is possible touse the number of landslides per kilometer per year as anindication of the magnitude of the triggering events andcalculate the frequency of occurrence based on the completelandslide inventory.

It is possible to relate the magnitude of extreme events totheir frequency of occurrence through the use of probabilitydistributions, such as the Gumbel extreme value distribution(Gumbel 1958). The Gumbel function is frequently used inhydrological applications to model extreme events. In landslidestudies, it was used to model return periods of landslide eventsthrough the analysis of extreme precipitation and extremeground water changes (e.g., Miller 1988; Odorico and Fagher-azzi 2003; Frattini et al. 2009). In a study focusing on thehazard along linear infrastructures such as roads or railroads,the information on the expected number of landslides that canoccur per unit length in a year and their probability ofoccurrence are essential for estimating the risk. Furthermore,traffic disruption time and expected indirect loss can be

Fig. 4 Probability density distribution of landslide volume. The black line is thefitted power trend to the linear portion of the curve (power scaling exponent is −1.8)

Fig. 5 Three types of pattern observed in the probability density distribution oflandslide volume for different years from 1992 to 2007


modeled if information on the number of landslides per unitlength is available. The cumulative probability distribution ofthe Gumbel extreme model can be applied in a landslide studyto model the probability of occurrence of the number oflandslides (NL) equal to or less than some value n. The modelcan be expressed as:

P NL � nð Þ ¼ e�e� aþnð Þ=c ð2Þ

where α and c are two parameters of the Gumbel distribution. Bythe method of moments, the parameters are evaluated as (e.g.,Chow et al. 1988):

a ¼ gc� � ð3Þ

c ¼ffiffiffi

6p

p� ð4Þ

where γ=0.57721 is a Euler’s constant, μ is the mean, and σ isthe standard deviation. For a specified time interval in a year,Eq. 2 can be rewritten for the value (NL) equal to or greater thansome value n as:

P NL � nð Þ ¼ 1T¼ 1� e�e� aþnð Þ=c ð5Þ

Two methods are commonly used for fitting distributions tothe Gumbel model and using them for frequency analysis: theplotting position method and the frequency factor method. Theformer is a straightforward plotting technique to obtain thedistribution function by use of certain “plotting position”formulas (Chow et al. 1988). The technique is to arrange the datain increasing or decreasing order of magnitude and to assignorder number R to the ranked values. The Weibull formula iscommonly used to obtain the plotting position, which for P(NL≥n)can be expressed as:

P ¼ Rmþ 1

ð6Þ

where R is the rank and m is the total number of observations.When R is ranked from lowest to highest, P is an estimate of P(NL≤n); when the rank is from highest to lowest, P is P(NL≥n).Equation 6 can be plotted on a probability paper to represent thecumulative probability distribution. The graph is designed in sucha way that it directly gives the return period and the correspond-ing magnitude of an event. The probability paper tends tolinearize the distribution so that the plotted data can be easilyanalyzed for extrapolation or comparison purposes. When usingthe graphical method, it is recommended not to extrapolate thedata more than two times the period of record; otherwise, theuncertainty in prediction will be high (Viessman et al. 1989).

To obtain the probability or return period of a certainnumber of landslides per unit length (kilometer) along thetransportation corridor exceeding a given value, the yearly totalnumber of landslides from cut slopes was selected for eachkilometer section for the period 1987 to 2007. The total number ofobservations (m) is 21 years. The values were ranked from low to

high, such that the lowest rank (1) was assigned to the lowest datavalue and the highest rank (21) to the highest data value. Usingthe plotting position method, the data were plotted on aprobability paper and a curve was fitted to the plotted points.

An example of the result of the Gumbel plot for km 14 of therailroad is shown in Fig. 6. In km 14, the maximum number oflandslides triggered in a year was recorded in 2001 (11), and yearswith no landslide were 1988, 1995, and 1997. The fitted straight linein the figure can be extrapolated to make an estimate of thenumber of landslides in different return periods. As an example,for km 14, the figure gives nine landslides for a 15-year returnperiod and 12 landslides for a 50-year return period.

Along the railroad, the Gumbel analysis was carried out foreach of the 17-km sections producing 17 Gumbel plots. Along theroad, data on the occurrence of landslides were not available forevery kilometer, and therefore, the Gumbel analysis was per-formed on two larger sections: a section with a length of 10 km (S-I, from km 390 to 400) and a section of 14 km length (S-II, fromkm 400 to 414). The two sections were selected on the basis of thedifference in geomorphological setting and the density of land-slide scars, which is the percentage of length of the road coveredby landslide scars. In section S-I, the average density is 14 km−1,which is about three times higher than in section S-II. The averagelandslide density for the entire road is about 9 km−1. The locationof landslides and kilometer marks along the railroad and the roadare shown in Fig. 1. For each section of the railroad and the road,the expected number of landslides with T1, T3, T5, T15, T25, and T50year return periods was estimated.

Figure 7 shows results of the Gumbel analysis for eachkilometer length along the railroad represented as a smooth linegraph. Results for the two sections along the road are given inTable 5. Results indicate that one or more landslides can occur onaverage once in three or more years. A 4-km stretch of therailroad (from km 10 to 13) is relatively more prone to landslides,as is the S-I section (from km 390 to 400) along the road. Total 56,84, 140, 164, and 197 landslides are expected to occur along theentire railroad, and about 14, 28, 55, 66, and 82 landslides areexpected along the entire road in T3, T5, T15, T25, and T50 yearreturn period, respectively.

The results indicate that the maximum number of landslidesper kilometer is expected toward the east of Burliyar. This sectionof the transportation lines also has the maximum probability ofexperiencing rainfall events that can trigger one or more land-slides (Jaiswal and van Westen 2009).

Landslide hazard assessmentThe specific hazard, expressed as the probability of a givennumber of landslides with a particular volume to occur along aspecific section of the road or railroad, was obtained by multi-plying the probability of having a certain number of landslidesper unit length resulting from the Gumbel analysis, with theresults of the volume–frequency analysis explained earlier. Intotal, 18 specific hazard scenarios were generated using combina-tions of the three volume classes (M-I, M-II, and M-III) and sixreturn periods (T1, T3, T5, T15, T25, and T50 year).

The results of three hazard scenarios, related to the threelandslide volume classes defined in Table 4, for T50 year returnperiod are given in Table 6 (for the railroad) and Table 7 (for theroad). The hazard categories H-I, H-II, and H-III show the

Original Paper


number of landslides of volume classes M-I, M-II, and M-III,respectively, that can occur per kilometer of cut slopes in a 50-year return period. The analyses show that on average once in50 years (annual probability of 0.02), the entire railroad will beaffected by 76.7, 104.3, and 16 landslides and the road by 32, 43.4,and 6.6 landslides of H-I, H-II, and H-III hazard, respectively.

Validation of landslide hazard modelValidation of results of a predictive model is absolutely essential inorder to make the model applicable for practical purposes (Chungand Fabbri 2003). Validation should be performed using landslidedata that are independent from those used in the developing themodel. The most suitable validation dataset is the one that includeslandslides that occurred after the period considered in the modeling.In our case, wewere able to use the landslide inventory from 2009 forthe validation of the hazard model.

Between 8 and 10 November 2009, an intense rainfall eventtriggered 136 landslides along the road and the railroad within thestudy area. The 3-day cumulative rainfall was recorded as 202 mm atKallar Farm, 387 mm at Burliyar, 560 mm at Hillgrove, 865 mm atKatteri farm, and 937 mm at Coonoor. The amount of rainfall showsa dramatic decrease from West (Coonoor) to East (Kallar Farm) ofthe study area. A landslide inventory was prepared from the railwayslip register and landslide technical reports. Field mapping wascarried out in February 2010 to spatially map the recorded landslides.Figure 8 shows the spatial distribution of landslides (which are allshallow debris slides) on cut slopes (represented as points), andTable 8 gives the statistics of these landslides. Out of the total136 slides, 71 occurred on cut slopes along the railroad and 65along the road. About 80 are reactivated old landslides and 56occurred for the first time. Prior to November 2009, sevenmore landslides have occurred along the railroad, which makesa total of 143 landslides in 2009.

The availability of data on landslides of the 2009 eventprovided an opportunity to compare the probability of land-slide volume used in the hazard modeling. For cut slopes, theprobability of occurrence of the landslides in 2009 was foundto be 0.61 for <100 m3, 0.35 for 100 to 1,000 m3, and 0.04 for>1,000 m3, which is different from the values used in thehazard model on cut slopes for years with more than 100landslides (0.39 for <100 m3, 0.53 for 100 to 1,000 m3, and 0.08for >1,000 m3). The occurrence of a higher proportion of smalllandslides (<100 m3) in this event is probably due to the factthat most landslides occurred between Marapallam and Katteri,which has rather gentle slopes covered by tea plantation. Thedifference in values is mainly due to the decreasing trend inrainfall from west to east, whereas in the eastern sections ofthe road and railroad, larger volume landslides are expected tooccur based on the 21-year record. The percentage of landslides

Fig. 6 Gumbel plot for km 14 of therailroad used to establish relationshipbetween total number of landslidesand return period

Fig. 7 Total number of landslides in different return periods (T1 to T50 years)along the railroad obtained using a Gumbel distribution


of larger volume would have been more if the same amount ofrainfall that triggered landslides between the Marapallam andKatteri section would have affected the area east of Burliyarwhere slopes are steep and cut slopes are high.

The validation of a hazard model for different temporalscenarios is difficult because for the single year of 2009, it isnot possible to have information on landslide events withdifferent return periods. Using the landslide event of 2009, weattempted to validate the results of the Gumbel analysis (thenumber of landslides per kilometer of cut slopes in differentreturn periods). Figure 9 compares the results of the Gumbelanalysis (Fig. 7) with the occurrence of landslides per kilometeralong the railroad in 2009. The figure indicates that for thesections of the railroad of km 17, 19, and 20, the number oflandslides that occurred in 2009 corresponds to a returnperiod of 25 or more years. The amount of rainfall thattriggered landslides in this section of the railroad (i.e., 865 mmaround Katteri farm) has never occurred in the period

considered in the modeling (i.e., 1987 to 2007) but was knownto have occurred in 1979 resulting in floods and triggeringmany landslides around the Coonoor area (Seshagiri andBadrinarayanan 1982).

Figure 9 also shows that in most of the other railroadsections (between km 10 and 17 and in the sections fromkm 20 to 26), the number of landslides that occurred in 2009corresponds to a much lower return period (5 years or less;Fig. 9). Although the sections from km 10 to 17 are more proneto landslides as indicated in Fig. 7, the amount of rainfall thattriggered landslides in these sections in 2009 was compara-tively less than those recorded around km 19. In fact therainfall decreased sharply from Coonoor toward the east. Theamount of rainfall recorded in Burliyar on 10 November 2009was 192 mm. Rainfall events with this amount have actuallyoccurred three times in the past 21-year period (1987 to 2007),which corresponds to a recurrence interval of 7 years.

A total 78 landslides occurred along the entire railroad in2009, which is comparable to the result of the Gumbel analysisfor a 5-year return period (i.e., 84 landslides). It is reasonableto expect here that if the entire area would have received theamount of rainfall that triggered landslides around Katterifarm (corresponding to a landslide event with a 25-year returnperiod), then the total number of landslides would have beenmuch higher (around 160) as predicted by the Gumbel analysisfor a scenario with a 25-year return period. In fact, only asmall section actually received a very high rainfall in 2009 (i.e.,around Katteri farm), and since in most sections the rainfallwas less therefore, as expected, the total number of landslideswas also less along the entire railroad.

Along the road, the occurrence of 65 landslides iscomparable with the model output for a 25-year return period(66 landslides). The occurrence of more landslides betweenkm 390 and 414 (S-II) triggered by extremely high rainfallcorresponds to an event similar to the one from 1979.

DiscussionGiven the completeness of the historical landslide informationespecially along the railroad for the past 21 years, we were able toapply a landslide hazard assessment method that is based only onthe historical information, without including statistical analysis ofpossible contributing factors or carrying out physically basedmodeling. Such conventional hazard assessment methods haveproven to be difficult to predict spatiotemporal occurrence oflandslides with different magnitudes. Most of the landslide hazardmodels presented in the literature are actually susceptibilitymodels providing an estimate of “where” landslides are expected,rather than giving also temporal and size probability (Brabb 1984;Guzzetti et al. 2005). The proposed method allowed us todetermine quantitative landslide hazard along a road and a

Table 5 Number of landslides in different return periods (T1 to T50) along the road

Road section Number of landslides per kilometerT1 T3 T5 T15 T25 T50

S-I (total length 10 km) 0 1.01 2.04 4.08 4.93 6.12

S-II (total length 14 km) 0 0.29 0.53 1.00 1.20 1.48

Table 6 Landslide hazard along the railway line in T50 year return period

Kilometer mark Number of landslides per kilometerH-I H-II H-III

10 8.2 11.2 1.7

11 10.1 13.7 2.1

12 8.9 12.2 1.8

13 7.1 9.7 1.5

14 4.7 6.5 1.0

15 4.8 6.5 1.0

16 4.1 5.6 0.9

17 3.8 5.2 0.8

18 3.3 4.5 0.7

19 2.7 3.7 0.6

20 2.6 3.5 0.5

21 2.8 3.7 0.6

22 3.4 4.6 0.7

23 2.7 3.7 0.6

24 2.8 3.7 0.6

25 2.5 3.4 0.5

26 2.2 2.9 0.4

Total 76.7 104.3 16

H-I, H-II, and H-III are the hazard related to landslide of M-I, M-II, and M-III, indicated inTable 4, respectively

Original Paper


railroad estimated directly from the frequency of past slopefailures. In fact, landslide hazard assessment approaches usingfrequency of past landslides not necessarily require prior land-slide susceptibility analysis if a complete inventory is available(Corominas and Moya 2008). In such cases, the probability ofoccurrence of landslides in a given area can be estimated directlyfrom the past landslides, especially when we are interested toknow the hazard over particular sections of the (rail) road andnot for every location individually.

According to the definitions given by Varnes (1984), thelandslide hazard should provide information on the probability ofoccurrence of landslides and their magnitudes in a given area.Along a transportation line such as a road or railroad, thisdefinition of hazard is more appropriate if it includes informationon the probability of occurrence of a certain number of landslidesof a given size over a given unit length (e.g., per kilometersection). The number of landslides is required if direct risk toroad, vehicles, commuters, and the indirect risk due to trafficdisruption is to be analyzed. The number of landslides per unitlength is an important input required to estimate the probabilityof a landslide hitting a vehicle (AGS, Australian GeomechanicsSociety and Sub-committee on Landslide Risk Management 2000;Wilson et al. 2005), and if it is multiplied by the probability thatthe landslide will be of a given volume, then the total volume oflandslide deposits expected and the time required to remove thedebris can be estimated.

All the information on landslides and rainfall were obtainedfrom historical records, which were fairly well maintained in thispart of India. The availability of continuous landslide recordsalong the railroad allowed to make a substantially completelandslide inventory, at least for the available time period. Never-

theless, the use of information from historical records iscomplicated and prone to errors. Errors may arise from manysources, such as the difficulty in translating maintenance recordsinto actual landslide events as different sources record landslideswith different perspectives, which may not be directly related tolandslide studies. Other sources of error are related to thedifficulty in recognizing past landslides in the field and themethod used in plotting landslides onto topographic base map.

In historical records, if the description of a landslide containsall the necessary information such as landslide occurrence date,type, location, and size and accompanied by photographs ormaps, then the identification of the landslide is easy and mappingis subjected to very little error. Some technical records containeddetailed description of the landslides along with illustrations. Fora small landslide where all its material is accumulated on therailroad, the recorded volume is a good indication of the landslidesize, and therefore, identification of the landslide and its size issubjected to minimum error. Where data on volumes areincomplete then other indications were used with a higheruncertainty. In case of the unavailability of data on volume (e.g.,from the railroad landslide table), the record was only used forobtaining the number of landslides per section of the (rail) roadand not for spatially mapping of landslide scars.

In landslide modeling, the Gumbel model has been used toanalyze the frequency of extreme events in spite of certainlimitations and assumptions, of landslides being stochasticprocesses and random distributed independent events, as dis-

Table 7 Landslide hazard along the road in T50 year return period

Road section Number of landslides per kilometerH-I H-II H-III

S-I (total length 10 km) 2.39 3.24 0.49

S-II (total length 14 km) 0.58 0.78 0.12

H-I, H-II, and H-III are the hazard related to landslide of M-I, M-II, and M-III, respectively

Fig. 8 Location of landslides (black circles) along the road and the railroad triggered between 8 and 10 November 2009

Table 8 Statistics of landslides triggered between 8 and 10 November 2009

Attributes

Total landslides on cut slopes nr 136

Volume of smallest landslide m3 2

Volume of largest landslide m3 2,500

Average volume of landslides m3 170

Median volume of landslides m3 50

Standard deviation of landslide volume m3 320


cussed by Miller (1988) and Odorico and Fagherazzi (2003). Theseassumptions should be considered when interpreting and usingthe results of the probability model. In this study, the use ofGumbel model was found suitable for estimating the frequency ofslope failures that are small in size and occur repeatedly from cutslopes at a number of locations. The Gumbel model was appliedto the available time series data and predicts the expectedfrequency of occurrence of extreme events, also with returnperiods that are larger than the time period used for the inputdata. It is important to realize that these extensions are only asvalid as the data used and uncertainty could be high ifextrapolation is done more than twice the length of the availabletime series. The Gumbel analysis is sensitive to outliers, and if theinventory contains recorded events related to much larger returnperiods, one should be careful with analyzing the results. In thisstudy, we estimated probabilities only up to a 50-year returnperiod, which is slightly more than twice the 21-year period forwhich data were available. The records also indicate that land-slides occur frequently in the area, and therefore, it was prudenthere to consider scenarios based on the lower return period, i.e., 1,3, and 5 years.

The result of the Gumbel analysis is also prone to uncer-tainty. The degree of uncertainty can be low if the inventorycontains records for all landslides such as those available for therailroad. The availability of a substantially complete inventoryalong the railroad provided a better assessment of the frequencyof slope failures. Along the road, we were not able to perform aGumbel analysis for every kilometer section because of theincompleteness of the landslide records. However, the results ofthe analysis along the road can be acceptable because theinventory at least contained the large and damaging landslides.

The validation with the data from 2009 illustrated that thehigh spatial variability of the rainfall along the (rail) road caninduce uncertainty in the result of the hazard analysis. Highrainfall amounts occurring on certain sections of the road and therailroad may trigger more landslides locally, similar to the oneobserved in 2009. The result is then that the same triggeringevents will be related to different return periods, due to the spatialvariation of triggering rainfall. However, over a longer period oftime, rainfall amounts do not vary significantly in the study area.The average rate of occurrence of extreme events might beincreased due to climate change; however, this has not beenproven yet for the area based on historical rainfall records.

The results of the hazard assessment are also prone touncertainty due to the non-cyclical nature of landslide events.Slopes that are subjected to landslides once might behavedifferently under similar rainfall conditions. Also human inter-vention along the road in the form of the construction ofretaining walls and drainage works will have an influence. Therainfall event with a 50-year return period (i.e., as characterizedby the number of landslides per kilometer caused by this event)may cause less landslides when it happens later again.

ConclusionsIn our earlier work, we estimated the temporal probability ofoccurrence of one or more rainfall events that can triggerlandslides along the transportation lines (Jaiswal and vanWesten 2009), based on rainfall thresholds for the same area.However, such an analysis does not provide information onthe frequency of landslides in different sections of the roadand the railroad. The method presented here does allow us tomake an estimation of the expected number of landslides perunit length and return period.

The proposed hazard model can also forecast landslidesdirectly for the entire route instead of per kilometer section.However, we recommend carrying out hazard assessmentindividually for different sections since rainfall amounts canvary locally to the extent that the same rainfall event canrepresent a 30-year event in the west and only a 7-year eventin the east, as observed in 2009.

The inclusion of the proposed landslide volume classes inthe hazard assessment will help in assessing landslide riskalong the road and the railroad. Ideally, landslide magnitudeshould be quantified based on absolute values of landslidevelocity, intensity, peak discharge, etc. but such parameters arevery site specific and vary with local condition such as channelgeometry, terrain roughness, and land use and are, therefore,difficult to obtain and integrate in the hazard analysis. Due tothis limitation and the complexity of landslide phenomena, theproposed magnitude classification is considered the bestsolution for this study.

This study aimed to quantify landslide hazard along atransportation corridor where landslides are small in size butoccur frequently. Hazard assessment including small-sizedlandslides is important if these occur along roads andrailroads. One could argue that small slides are actually localslips which may not be of much significance. This could betrue if such slides are on natural slopes, but along a road or arailroad, even small slips can be disastrous such as resulting ina road accident or train derailment. Small landslides, in fact,constitute the largest part of landslide inventories, particularlyif inventories contain cut slope failures as well. For example,out of the 2,811 landslides reported in the GeotechnicalEngineering Office database of Hong Kong for the period from1992 to 1997, about 90% were of volume less than 50 m3 (Daiand Lee 2001).

Temporal prediction of few small-sized slides in a givenarea is difficult, but in the Gumbel extreme value method, it ispossible to estimate the probability of occurrence and returnperiod of the larger (extreme) events based on past records.This information will help to manage landslide risk morerationally.

Fig. 9 Comparison of model output of the Gumbel analysis and landslides thatoccurred in 2009 for different sections of the railroad

Original Paper


The study of landslide volume–frequency distribution showedlarge variation both in the number of landslides triggered in a year aswell as the range of volumes. It was observed that these changes arerelated to the amount of rainfall, and therefore, the estimation of theprobability of landslide size based on the frequency percentagerelated to different magnitudes of rainfall events can be a workablesolution.

The hazard model performed reasonably well in forecastinglandslides, as evidenced from the validation result. However, weneed to validate the model with landslide events of different returnperiods, and for this, we have to test it for events that will happen inthe near future to improve the model.

AcknowledgmentsWe acknowledge the help of Southern Railway, Geo-technical Celland Tea Estates of Coonoor, Tamil Nadu, India for the relevantdata and support. The research was carried out under the UnitedNations University—ITC School on Disaster Geo-InformationManagement (www.itc.nl/unu/dgim).

References

AGS, Australian Geomechanics Society and Sub-committee on Landslide Risk Manage-ment (2000) Landslide risk management concepts and guidelines. Aust Geomech 35(1):49–92

Brabb EE (1984) Innovative approach to landslide hazard and risk mapping. Proceedingsof the 4th international symposium on landslides, Toronto, vol. 1, pp 307–324

Brardinoni F, Church M (2004) Representing the landslide magnitude-frequency relation:Capilano River basin, British Columbia. Earth Surf Process Land 29:115–124

Catani F, Casagli N, Ermini L, Righini G, Menduni G (2005) Landslide hazard and riskmapping at catchment scale in the Arno River basin. Landslides 2:329–342

Chow VT, Maidment DR, Mays LR (1988) Applied hydrology. McGraw-Hill, New YorkChung CJ, Fabbri AG (2003) Validation of spatial prediction models for landslide hazard

mapping. Nat Hazards 30:451–472Coe JA, Michael JA, Crovelli RA, Savage WZ (2000) Preliminary map showing landslide

densities, mean recurrence intervals, and exceedance probabilities as determinedfrom historic records, Seattle, Washington. USGS Open-File report 00-0303. Availableat: http://pubs.usgs.gov/of/2000/ofr-00-0303, cited on 15 July 2008

Corominas J, Moya J (2008) A review of assessing landslide frequency for hazard zoningpurposes. Eng Geol 102:193–213

Dai FC, Lee CF (2001) Frequency–volume relation and prediction of rainfall-inducedlandslides. Eng Geol 59:253–266

Fell R, Finlay PJ, Mostyn GR (1996) Framework for assessing the probability of sliding ofcut slopes. In: Senneset K (ed) 7th international symposium on landslides, vol 1.Balkema, Trondheim, pp 201–208

Fell R, Ho KKS, Lacasse S, Leroi E (2005) A framework for landslide risk assessment andmanagement. In: Hungr O, Fell R, Couture R, Eberthardt E (eds) Landslide riskmanagement. Taylor and Francis, London, pp 3–25

Fell R, Corominas J, Bonnard C, Cascini L, Leroi E, Savage WZ (2008) Guidelines forlandslide susceptibility, hazard and risk zoning for land use planning. Eng Geol102:85–98

Frattini P, Crosta G, Sosio R (2009) Approaches for defining thresholds and returnperiods for rainfall-triggered shallow landslides. Hydrol Process 23:1444–1460

Gumbel EJ (1958) Statistics of extremes. Columbia University Press, New YorkGuthrie RH, Evans SG (2004) Magnitude and frequency of landslides triggered by a

storm event, Loughborough Inlet, British Columbia. Nat Hazards Earth Syst Sci4:475–483

Guzzetti F, Malamud BD, Turcotte DL, Reichenbach P (2002) Power-law correlations oflandslide areas in central Italy. Earth Planet Sci Lett 195:169–183

Guzzetti F, Reichenbach P, Cardinali M, Galli M, Ardizzone F (2005) Probabilisticlandslide hazard assessment at the basin scale. Geomorphology 72:272–299

Hungr O (1997) Some methods of landslide hazard intensity mapping. In: Cruden DM,Fell R (eds) Landslide risk assessment. Balkema, Rotterdam, pp 215–226

Hungr O, Evans SG, Hazzard J (1999) Magnitude and frequency of rock falls and rockslides along the main transportation corridors of south-western British Columbia. CanGeotech J 36:224–238

Iverson RM (2000) Landslide triggering by rain infiltration. Water Resour Res 36:1897–1910

Jaiswal P, van Westen CJ (2009) Estimating temporal probability for landslideinitiation along transportation routes based on rainfall thresholds. Geomorphol-ogy 12:96–105

Malamud BD, Turcotte DL, Guzzetti F, Reichenbach P (2004) Landslide inventories andtheir statistical properties. Earth Surf Process Land 29(6):687–711

Miller SM (1988) A temporal model for landslide risk based on historical precipitation.Math Geol 20(5):529–542

Odorico PD, Fagherazzi S (2003) A probabilistic model for rainfall-triggered shallowlandslides in hollows: a long term analysis. Water Resour Res 39(9):1–14

Seshagiri DN, Badrinarayanan B (1982) The Nilgiri landslides. GSI Misc Pub No 57Stark CP, Hovius N (2001) The characterization of landslide size distributions. Geophys

Res Lett 28(6):1091–1094van Westen CJ, Asch TWJ, Soeters R (2006) Landslide hazard and risk zonation—why is

it still so difficult? Bull Eng Geol Environ 65:67–184Varnes DJ (1984) Landslide hazard zonation: a review of principles and practice.

UNESCO, ParisViessman W, Lewis GL, Knapp JW (1989) Introduction to hydrology. Harper Collins, New

YorkWilson RA, Moon AT, Hendrickx M (2005) Application of quantitative risk assessment to

the Lowrence Hargrave Drive Project, New South Wales, Australia. In: Hungr O, FellR, Couture R, Eberthardt E (eds) Landslide risk management. Taylor and Francis,London, pp 589–598

P. Jaiswal ())Geological Survey of India,Bandlaguda, Hyderabad, Andhra Pradesh, Indiae-mail: [email protected]

P. Jaiswal : C. J. van Westen : V. JettenITC,University of Twente,Hengelosestraat 99, 7500 AA, Enschede, The Netherlands


quantitative assessment of landslide hazard along transportation lines using historical records

Documents