Professional Documents
Culture Documents
Jordi Corominas Jose Moya Alberto Ledesma Antonio Lloret Josep A. Gili
Abstract In active landslides, the prediction of acceleration of movement is a crucial issue for the design and performance of warning
systems. The landslide of Vallcebre in the Eastern Pyreenes, Spain,
has been monitored since 1996 and data on rainfall, groundwater
levels and ground displacements are measured on a regular basis.
Displacements observed in borehole wire extensometers have shown
an immediate response of the landslide to rainfall episodes. This rapid
response is likely due to the presence of preferential drainage ways.
The occurrence of nearly constant rates of displacement in coincidence with steady groundwater levels suggests the presence of viscous
forces developed during the movement. An attempt to predict both
landslide displacements and velocities was performed at Vallcebre by
solving the momentum equation in which a viscous term (Bingham
and power law) was added. Results show that, using similar rheological parameters for the entire landslide, computed displacements
reproduce quite accurately the displacements observed at three selected wire extensometers. These results indicate that prediction of
displacements from groundwater level changes is feasible.
Keywords Translational slide . Viscous behaviour . Prediction
of displacements . Eastern Pyrenees . Spain
Introduction
Traditional strategies in landslide hazard management have been
mostly oriented at avoiding dangerous sites and stabilizing unstable slopes. Regional and urban planning can be powerful tools to
ensure that development occurs in the safest places. In that respect,
planners and decision-makers have benefited from the information
provided by hazard and risk maps. However, development in many
regions has occurred, and indeed continues to occur without hazard
maps being available. Because of this, there are numerous examples
of development built either on or near large landslides. In such cases,
remedial measures are often unaffordable, while moving the population to more stable slopes can pose a considerable societal problem.
As a consequence, the coexistence of human activities with natural
hazards must be considered with the risk being mitigated through
the use of hazard management strategies.
In landslide threatened areas, solutions usually focus on the implementation of mitigation measures that avoid the slope failure or
divert the moving mass away from vulnerable elements, or protect
or reinforce the threatened elements. However, in some places, to
stabilise a landslide may simply be too costly in financial and/or environmental terms. When mitigation is not feasible, it is important
to consider the implementation of warning systems that may, at least,
avoid damage and/or loss of human lives. To be effective and reliable, warning systems require a sound knowledge of the behaviour
of the landslide, including the mechanism, the potential triggers and
their respective thresholds (i.e. rainfall intensity and duration), the
Landslides 2 . 2005
83
Original Article
(Corominas et al. 1999). In some of these movements, the search for
reliable rainfall thresholds has been unsuccessful (Noverraz et al. 1998;
Flageollet et al. 1999) and alternative approaches are required to assess
the conditions leading to the reactivation. Petley et al. (2002) analyzed
inverse rate curves and demonstrated that landslides occurring in
existing failure surfaces show an increase of the rate of deformation
up to a constant rate (steady-state behaviour) for any given stress state
material combination. On the other hand, the use of hydrological
and mechanical models is restricted because of the complexity of
these landslides. Existing models remain limited when considering
groundwater changes and rates of displacement. Some attempts to
examine large landslides have included the use of simplified empirical
models that combine hydrological and stability analyses (Van Asch
and Buma 1997) and the use of viscous constitutive laws for the
simulation of continuous displacement patterns (Vulliet 2000).
Continuously moving landslides require a dynamic analysis instead of a classical static approach. The temporal scale is fundamental, and any modelling attempt should consider this. In addition to
these conceptual difficulties, the analysis of the response to both short
and medium-term rainfall scenarios with hydrological and mechanical models requires long records of high quality data that include
rainfall, groundwater levels and landslide displacements.
In this paper we present an examination of movement of the Vallcebre landslide, a slow-moving landslide activated by rainfall. A simulation of the displacement pattern in relation to groundwater record
by means of a dynamic, physical model has been attempted. The
work has been complemented by an extensive field survey and laboratory investigation to allow the best possible modelling of landslide
behaviour. That has included an analysis of the rheology of the sliding material and of the influence of the viscous component on the
motion.
Fig. 1 General view of the Vallcebre translational slide. The movement is bounded, in the background, by the high scarp of Serra de La Llacuna (center-right), in the foreground by the Vallcebre
torrent, which runs from the right to the left in an incised gully, and laterally by the light grey limestone outcrops
84
Landslides 2 . 2005
Fig. 3 Geological cross-sections of the Vallcebre landslide (the location of the profiles appears in Fig. 2)
Landslides 2 . 2005
85
Original Article
Fig. 4 Photograph of the toe of the Vallcebre landslide which is being continuously undermined and eroded by the Vallcebre torrent. Local slope failures are observable in the front
ground surface and the trees are tilted backwards due to the development of a graben along with a slight rotation of the head. In
contrast, within the units, the ground surface is only disturbed by
minor fissures, scarps less than 50 m long, and by some cracking
of the walls of farm-houses standing on the landslide. The direction of both the transverse scarps and grabens, indicates a movement towards the north-west (see Fig. 2). A secondary direction of
movement, towards the Torrent Llarg, is also suggested by the trend
of the escarpment of the upper slide unit, but this has not been
validated to date. The most active area is the lower unit, which is
bounded, at the south-western side, by the torrents of Vallcebre and
Llarg and, at its north-eastern side, by a well developed lateral shear
surface.
The geological structure of the landslide has been obtained by
means of an intense geological and geomorphological assessment,
which has included mapping of the surficial exposures, geophysical
surveys and drilling. The mobilised material consists of a sequence
of continental sediments. From the bottom to the top it includes
(Fig. 3): (i) densely fissured shales, 1 to 6 m thick, showing distinct
slickensides; (ii) locally, gypsum lenses up to 5 m thick and some tens
of meters width which are predominant in the southern side of the
landslide; and (iii) clayey siltstones rich in veins and micronodules of
gypsum with thickness of up to tens of meters. In addition to these
layers, the extension zones located at the toe of the scarps have been
filled with colluvium composed of boulders and gravel with a silty
matrix.
Gypsum lenses are affected by solution processes and have developed some karst features like pipes and springs. These pipes run
86
Landslides 2 . 2005
runs roughly parallel to the ground surface except in the area close
to the contact between the intermediate and lower units where the
presence of the inverse fault produces a bedrock threshold and a decrease of the landslide thickness. The depth of the failure surface is
not constant. Inclinometric readings showed that the lower slide unit
has a thickness of 10 to 15 m, whereas the intermediate unit reaches
a thickness of at least 34 m in the northern side and between 14 and
19 m in the southern one.
Since 1996, systematic logging of rainfall, groundwater level
changes, and landslide displacements has been carried out every
20 min. Piezometric readings have indicated that changes in groundwater levels occur quickly. The extensometer has recorded sudden
changes in displacement rates that can be directly related to the fluctuations of the water table governed by rainfall.
Measurements of surface displacements using differential GPS have
been used to complement the measurements of the inclinometers
and wire extensometers. A total of 30 points were positioned on the
landslide surface for periodic control. These points included reference
points, stable points adjacent of the landslide, and targets within the
landslide mass (buildings, outcropping rock blocks, steel rods and
upper ends of the boreholes). Real Time Kinematics and Fast Static
were used to analyze the GPS observations (Gili et al. 2000). Fourteen
GPS campaigns were carried out from December 1995 to February
1998. During this period horizontal displacements up to 1.6 m and
subsidence of the landslide surface of up to 0.35 m were observed.
The wire extensometer measurements show that the landslide has
never completely stopped moving since we started the continuous
monitoring in November 1996, although velocities reduced significantly during dry periods (Fig. 5). On the other hand, the history
of displacement of the extensometers reflects that different parts of
the landslide mass move synchronically but with a different rate of
displacement. Extensometer S2 has shown the fastest displacements,
with maximum recorded rates of up to 50 mm/week. The other extensometers standing on this slide unit, the S5, S9 and S11, have exhibited
rates lower than those of S2, although they were of the same order of
magnitude.
At borehole S6, placed on the intermediate slide unit, the velocity
is significantly smaller, indicating that, compared to the lower unit,
the intermediate unit is less active.
Determination of the landslide parameters
Tests on undisturbed samples obtained by drilling were carried out
to determine parameters and soil characteristics to be used in the
numerical analyses. Special attention was paid to tests for determining
the shear strength of the soil involved in the slip surface.
Basic identification tests and different types of shear tests were
performed on 14 undisturbed samples. Shear tests were carried out
predominantly on samples of the fissured shale unit where the slip
surface is located; in particular, some tests were conducted on observed pre-existing shear surfaces (Fig. 6). Table 1 shows the location
where the samples were taken and the type of tests performed on each
sample.
The average Atterberg limits obtained on two samples of clayey
siltstone and seven samples of fissured shales are indicated in
Table 2.
Material
S-2
15
15.2
Fissured shale
S-3
9.5
S-4
9.5
S-6
44.5
19.6
10.3
10.7
4.0
8.9
44.1
Fissured shale
Fissured shale
Fissured shale
Clayey siltstone
Clayey siltstone
Fissured shale
S-7
34
S-9
S-11
S-12
S-13
S-14
14.5
8
17.5
13.5
13.5
25.4
34.2
15.2
6.5
11.2
11.4
8.7
Fissured shale
Fissured shale
Fissured shale
Clayey siltstone
Clayey siltstone
Clayey siltstone
Clayey siltstone
Existing shear
surface
Undisturbed
Undisturbed
Undisturbed
Undisturbed
Existing shear
surface
Undisturbed
Undisturbed
Undisturbed
Undisturbed
Undisturbed
Undisturbed
CU triaxial test
Remoulded
Undisturbed
Remoulded
Remoulded
Remoulded
Remoulded
Undisturbed
Remoulded
Remoulded
Remoulded
Remoulded
Remoulded
Undisturbed
Landslides 2 . 2005
87
Original Article
Fig. 6 Slickensided shear surface within
the fissured shales
Liquid limit
Plastic limit
Plasticity index
41.8
54.5
22.1
35.2
19.7
19.3
Table 3 Peak shear strength parameters of fissured shale and clayey siltstone
88
c (kPa)
()
0<n<100 kPa
100<n<700 kPa
0.0
53
38.7
15.1
Landslides 2 . 2005
borehole S6, the specimen was prepared and mounted in the shear
box such that a pre-existing shear surface was the slip surface in the
shear box. In addition, direct shear tests were also performed on five
samples of fissured shale and five samples of clayey siltstone.
Ring shear tests were performed on ten samples using a Bromhead ring shear apparatus. Specimens were prepared by remoulding
the specimens previously tested in the shear box to a water content similar to their liquid limit. A total slip displacement of about
150 cm was applied in each test using displacement rates between 0.02
and 0.3 mm/min. Finally, consolidated-undrained (CU) triaxial tests
Table 4 Shear strength parameters derived from different tests : Minimum strength
measured in direct shear, ring shear and triaxial tests for fissured shale; minimum
strength measured in direct shear for clayey siltstone and residual strength measured on
a pre-existing shear surface in fissured shale
Material
c (kPa)
()
Fissured shale
0<n<200 kPa
200<n<700 kPa
0<n<400 kPa
0<n<800 kPa
0.0
44.0
0.0
0.0
23.4
11.8
33.0
7.8
Clayey siltstone
Shear surface
Table 5 Depth of piezometers and range of groundwater level position in boreholes S2,
S4, S5, S6, S9 and S11
Depth to groundwater level (m)
Borehole
Piezometer depth (m)
Minimum
Maximum
Fig. 8 Piezometric records of boreholes S2, S4, S5, S6, S9 and S11
S2
S4
S5
S6
S9
S11
8.7
7.7
8.8
19.8
9.9
7.9
0.63
0.96
3.50
3.59
1.28
0.96
6.28
4.47
6.2
11.71
5.81
5.05
Landslides 2 . 2005
89
Original Article
For this analysis, the Bishop and Janbu methods of slices have
been used by means of the STABL code, which was developed at
Purdue University, USA (Lovell 1988). The stability analyses have been
performed considering the two extreme groundwater levels recorded
in the piezometers. In this case the observed groundwater depths
ranged between 1 and 6 m in the lower unit and 10 and 12 m in the
intermediate unit. A flow parallel to the slope surface was assumed
to compute pore water pressures from water table depth. Finally,
a direct relation between water table position and global factor of
safety is defined. We have determined the factor of safety for a range
of strength parameters and water table positions. Four groups of limit
equilibrium analyses were performed:
Fig. 9 Piezometric record (blue line) and landslide velocity (black line) at borehole S2
and pipes rather than by soil porosity. The role of the karstic network in the gypsum lenses is unclear but all the observed features are
very shallow (up to 3 m depth), which is well above of the normal
groundwater level fluctuation. Thus it might play only a secondary
role.
Despite the rapid response of the piezometers, peak water levels
are attained at slightly different times depending on the permeability
of the adjacent material. Two basic types of responses to rainfall
have been observed, depending on the location of the piezometers.
These located in tension zones, such as S5, show a smaller variation
in groundwater level (ranging between 0.5 and 2 m) and a faster
drainage compared to the piezometers located elsewhere (for example
S2, S4 and S11). The latter piezometers experience changes of 2 to 5 m
and a slower rate of groundwater level decrease. The behaviour of
piezometer S5 is consistent with the presence of high permeability
cracks in the tension zone (graben). Consequently, it can be inferred
that cracks act as preferential drainage paths within the landslide.
In addition to the rapid response to precipitation inputs, all
piezometers show a defined level below which the groundwater table
decreases very slowly. This level may be observed during the periods
FebruaryApril 1997 and SeptemberOctober 1997, in which no or
negligible rain was recorded in the area (Fig. 8).
A close relationship between the groundwater level changes and
landslide activity was measured using a wire displacement meter
at borehole S2 (Fig. 9). There exists a strong level of synchronism
between the two records. In addition, the rate of displacement is
strongly correlated with the water table data. The exception to this is
the event of January 1997 (increment of velocity without increment
of water table), which may be caused by other factors (i.e. toe erosion
by the Vallcebre torrent).
Stability conditions
An analysis of the stability of the landslide was performed using 2-D
limit equilibrium method. This gives a global safety factor which is
difficult to relate to the creep behaviour seen in Vallcebre, where
movements occur almost continuously. However the determination
of both driving and resisting forces provides an insight into the magnitude of the unbalanced force and the degree of stability of the
landslide.
90
Landslides 2 . 2005
n = 20 kN m3
n = 22 kN m3
Clayey siltstone
c
Fissured
c
shale
These results also agree with the field measurements, which show
higher velocities for the lower part. In fact, it is assumed that movement starts on the lower unit, which has a factor of safety close to 1,
and then a tension area and some cracks develop at the head scarp of
this unit, which is actually the toe of the medium part of the landslide. This movement of the lower unit generates, after some days, a
movement increase on the intermediate part as well. This behaviour
has been considered in the group of analyses in which the equilibrium of the intermediate part without the lower unit has been studied
(Case 4).
However, the analysis of the lower unit (Case 2) also shows that for a
low groundwater level, the factor of safety is still very low (F.S. = 0.79),
which does not correspond with the very low rate of displacement
observed. This suggests that either the analysis is not appropriate or
other forces must be taken into account.
On the other hand, Fig. 10 shows an interesting relationship between observed velocities and the depth of water table at borehole S2
for the period considered (modified from Fernandez-Pombo 1998).
A cubic curve can be fitted to the data, giving a good regression
coefficient. The points at a water table depth of 5 m and velocities
over 5108 m/s refer to the event of December 1996January 1997
mentioned above.
Colluvium
c
Factor of safety
Position of water table
Low
High
11.8
7.8
0
0
14.7
14.7
1.19
1.02
1.14
0.98
11.8
7.8
0
0
14.7
14.7
1.17
0.79
1.07
0.72
11.8
7.8
0
0
14.7
14.7
1.18
1.05
1.13
1.01
11.8
7.8
0
0
14.7
14.7
1.14
1.02
1.10
0.99
The position of the groundwater level at a depth of 6.22 m approximately, corresponds to a stable situation. If a back analysis of stability
is performed for this case in the lower unit, values of the effective
friction angle of = 14 and of cohesion c = 0 for the fissured shale
are obtained. This, again, does not agree with the value obtained from
laboratory tests, which range from 7.8 to 11.8 . Note that this small
range of friction angle provides a change in factor of safety from 0.79
to 1.17.
The value of = 14 obtained assuming limit equilibrium for
that position of water table is difficult to explain if only static conditions are considered. Although laboratory data could include some
uncertainties (i.e. due to sampling) we do not have any evidence of
such high strength for the slip surface. Consequently, we understand
that beside frictional resisting forces, additional resisting forces (i.e.
viscous forces) are necessary to stop the movement. Overall, a limit
equilibrium analysis can not simulate the continuous movement of
the landslide, and a different mechanical analysis must be used instead.
Modelling of landslide displacements
The close relationship existing between landslide velocity and position of groundwater level at borehole S2 suggests that it is possible
to perform a simulation of the landslide displacements from data
recorded in the piezometers. Figure 10 shows that a black box regression analysis is feasible for the simulation of displacements, but we
preferred to adopt a physical approach using the momentum equation. We analysed the dynamics of local points of the landslide in
which uniform conditions for the geometry (i.e. infinite slope conditions) can be reasonably assumed. In addition to that, a viscous
component has also been taken into account. A brief description of
the model and its application to Vallcebre is presented in this section,
which also includes a critical appraisal of the approach considered.
Viscous behaviour of the landslide
The dynamics of the landslide are governed by the difference between
destabilising forces (F), that depend basically on weight and slope,
which are constant, and resisting forces (F r ), that are sensitive to
water pressure at the slip surface. The momentum equation can be
written as:
Fig. 10 Velocities versus water table depths at borehole S2 for November 1996 to October
1998. Data correspond to mean daily values (modified from Fernandez-Pombo 1998)
F F r = ma
(1)
Landslides 2 . 2005
91
Original Article
where m is the mass and a the acceleration. For a local point of the
landslide where infinite slope conditions apply, resisting forces can
be estimated using Mohr-Coulomb criterion, depending on cohesion
and friction. Forces are computed over a unit surface, and therefore
shear stresses are considered in what follows:
[c + ( pw ) tan ] = ma
(2)
92
Landslides 2 . 2005
(4)
where is the specific weight of the sliding mass. Note that in Eq. (4),
forces are expressed per unit area of slip surface.
Viscous forces are usually dependent on the strain rate of the shear
zone. For a Bingham type model, this relation is linear and becomes:
v = v /z
(5)
Fig. 11 Water table depths and displacements measured by the wire extensometer, plotted
against time at borehole S2 during spring and summer 1997
(3)
= ma +
dv
v
v
=m
+
z
dt
z
(6)
Pore water pressure was not measured directly, but it was estimated
from readings of depth of groundwater level. Assuming a parallel flow
to the slope surface:
pw = w cos2 h = w cos2 (l Dw )
(7)
x
,
> 0
(8)
v x1
A
(9)
(10)
Fig. 13 Measured and predicted displacements and velocities at borehole S2 using Bingham
and power law models. The results of the two models are represented by a same line because
they are almost the same and indistinguishable at the scale of the graphs
could apply. A fourth borehole, S5, located close to the graben and
affected by some rotational component was also used. The parameters of the viscous models and the angle of friction were estimated
independently for each borehole.
Figure 13 presents a comparison between measured and predicted
displacements and velocities, using the Bingham and power law models for borehole S2. It can be observed that both models provide a
good simulation of the measured behaviour. Table 7 shows the estimated values of material parameters. The values predicted by the two
models are almost the same, because when the exponent in the power
law is near one, as occurs in this case, the two models are equivalent, and the value of /z is equal to /A. Table 7 also indicates the
goodness of fit. The latter is expressed as the root mean squared error
(RMSE).
Simulated displacements and velocities for the other two analysed
boreholes are not as good as for borehole S2. Power law model provides better fits than the Bingham model for the borehole S9, although
the difference between the velocities predicted by the two models is
small (Fig. 13 and Table 7). For borehole S11, the predictions of the
models are very similar.
Landslides 2 . 2005
93
Original Article
Table 7 Parameters obtained
from back analysis and goodness
of fitted displacements and velocities
(expressed as root mean squared
errors, RMSE's)
94
Borehole
S2
S9
S11
31.5
7.3
1.51 107
0.55 106
1.0023
12.7
1.05 108
12.7
1.05 108
21.5
6.1
1.52 107
0.44 106
1.39
21.2
1.48 108
14.7
1.36 108
32.0
6.2
1.60 107
4.06 106
1.83
36.4
1.69 108
36.2
1.59 108
Displacement (mm)
Velocity (m/s)
Displacement (mm)
Velocity (m/s)
Fig. 14 Measured and predicted displacements and velocities at borehole S9 using Bingham
and power law models (dashed line corresponds to a period without measured data)
Fig. 15 Measured and predicted displacements and velocities at borehole S5 using Bingham
and power law models
Note that the parameters obtained through back analysis are similar
in all boreholes. On the one hand, friction angles of 6 to 7 are close
to the residual value measured in the laboratory for the fissured clay
(7.8 degrees). On the other hand, the thickness of the shear zone, a
parameter difficult to estimate in practice, is around 0.3 m, a value
obtained by analysing the behaviour of the wire extensometer records
(Corominas et al. 2000).
For the Bingham model, the viscosity calibrated was around
1.5107 kPa-s in the three boreholes. These values are within the
range of the viscosities obtained in other studies (Vulliet and Hutter 1988; Angeli et al. 1996). However, parameters estimated for the
power law provide small variations in the three boreholes, in particular the value of parameter A in borehole S11 is almost an order
of magnitude higher than the same parameter obtained in other
boreholes.
The agreement between computed and predicted displacements using the estimated parameters is generally good, particularly for borehole S2. Nevertheless at that particular point, some of the events have
Landslides 2 . 2005
not been reproduced by the model. This is the case of the peak of velocity in January 1997 without rising of water table, already mentioned.
A condition that seems to have a strong influence on the quality
of the simulation is the assumption of infinite slope. In that respect,
at borehole S2, located in the centre of the lower unit and, in a lesser
extent, at boreholes S9 and S11, these conditions may apply (Figs. 2
and 3).
Instead, for borehole S5, the infinite slope conditions are clearly
not valid. This borehole is located near the head scarp of the lower
unit where the ground surface has some rotational component of
movement. We obtained spurious results, as negative velocities or
null displacements for periods in which the landslide was moving
(Fig. 15).
The former shows that extension of the results to other local points
must be always performed carefully. The use of local models seems
to be more interesting for conceptual purposes, when the mechanical
behaviour is not known in advance and general trends are investigated as a first step for developing future global models. In this case,
the use of a viscous component in the dynamics of the landslide
has proved to be useful for a consistent analysis of the records of
displacements and water pressures. Obviously these analyses are limited to local conditions and thus, a representative set of boreholes
must be selected. For a particular landslide, those types of records
and the kind of analyses described here may become a promising
tool to develop alarm systems and to develop future procedures for
predicting landslide behaviour at a global scale (i.e., using coupled
finite element models including flow and mechanical equations).
Conclusions
The nearly continuous monitoring of the Vallcebre landslide has
allowed the observation of some characteristics of the landslide behaviour that would otherwise have gone unnoticed. The landslide is
very sensitive to rainfall, and cracks have been shown to be preferential infiltration and drainage pathways.
Our monitoring network has shown the dependence of the rate of
displacement on groundwater level; from this we infer the existence of
a viscous component in the landslide behaviour. On the other hand,
the continuous records have been used to both calibrate and validate
the analyses performed.
From the point of view of the limit equilibrium stability conditions,
the Vallcebre landslide has a factor of safety of about one. However,
when critical conditions are reached, failure does not occur instantaneously, because only a small acceleration of the mass is produced.
Thus, a full analysis needs to consider the dynamics of the landslide, instead of assuming a binary approach: stability versus failure.
The models used in the time dependent simulations are based on
this dynamic approach, rather than just considering static limit
equilibrium.
The simulations of the landslide dynamics by taking into account
a viscous component were performed using parameters obtained
by back-analysis of the recorded data. This simulation is successful
because the parameters back-analyzed are consistent with field and
laboratory data obtained in an independent manner. Moreover, those
values obtained were similar for all the boreholes considered. Thus,
the landslide behaviour at selected locations seems to be well reproduced using the procedure outlined in the paper. Next steps in this
analysis will include the examination of the behaviour of the landslide from a global perspective, instead of considering only motion
in individual boreholes. That approach, however, will likely require
a considerable computational effort, because a 3D analysis and a
Landslides 2 . 2005
95
Original Article
Vulliet L, Hutter K (1988) Some constitutive laws for creeping soil and for rate-dependent sliding at
interfaces. In: Swoboda (ed) Numerical methods in geomechanics (Innsbruck 1988). Balkema,
Rotterdam, pp 495502
Wieczorek GF (1987) Effect of rainfall intensity and duration on debris flows in central Santa Cruz
Mountains, California. Geological Society of America. Rev Eng Geol 7:93104
Wilkinson PL, Anderson MG, Lloyd DM (2002) An integrated hydrological model for rain-induced
landslide prediction. Earth Surf Proc Landforms 27:12851297
96
Landslides 2 . 2005