Computers and Geotechnics 37 (2010) 837–845

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Sensitivity analysis applied to slope stabilization at failure

V. Navarro a,*, A. Yustres a, M. Candel a, J. López a, E. Castillo b
a
University of Castilla-La Mancha, Avda. Camilo José Cela s/n, 13071 Ciudad Real, Spain
b
University of Cantabria, Avda. de los Castros s/n. 39005 Santander, Spain

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

Article history: This article discusses how sensitivity analysis is a sound assessment tool for selecting the most efficient
Received 13 November 2009 stabilization method of slopes at failure. A discretized form of the variational approach is used not only
Accepted 3 March 2010 for performing sensitivity analysis but to locate the critical slip surface, i.e., the sensitivity analysis is car-
Available online 24 August 2010
ried out in the same way as it is done in optimization problems. This method supplies a robust formula-
tion and methodology for obtaining the sensitivities of the safety factor with respect to both the soil
Keywords: parameters and the slope profile, stating the slope stabilization design as a relatively simple minimiza-
Landslides
tion problem. Two well known examples, as the Selset landslide and the Sudbury Hill slip are used to
Slopes
Failure
illustrate the application of the method and to highlight both its capabilities and limitations.
Slope stabilization Ó 2010 Elsevier Ltd. All rights reserved.
Design
Clays

1. Introduction In any event, regardless of how the sensitivity analysis is done,

Today, engineers are not completely satisfied with the solutions
to given problems and they also require knowledge of how these
solutions depend on data. Thus, for instance, in minimization prob-
lems, such as those associated with slope stability, it is not enough
to know the optimal value of the objective function (the safety fac-
tor) and the solution (slip line) where the minimum is attained. In
this sense, approaches as the ‘‘safety maps” introduced by Baker
and Leshchinsky [1] provides a valuable information to determine
how and how much specific changes in the parameters of the sys-
tem modify both the optimal objective function value and the opti-
mal solution.
In the field of slope stability, sensitivity analysis is generally
conducted by means of a series of calculations in which each sig-
nificant parameter is varied systematically over its maximum cred-
ible range in order to determine its influence upon the safety factor
[2]. If one is interested in characterizing the variation in safety
when encounter minor modifications in the parameters, these
incremental techniques define an approximation of the safety fac-
tor gradient. In this case, for discrete problems (i.e., slices) the sen-
sitivity may be calculated in a simpler and compact way by using
the techniques that have been developed in the area of non-linear
optimization [3]. When dealing with continuous problems as those
linked to the variational approach of slope stability analysis, the
formulation put forth by Castillo et al. [4] can be used.
* Corresponding author. Address: Edificio Politécnico Avda. Camilo José Cela s/n,
13071 Ciudad Real, Spain. Tel.: +34 926295453; fax: +34 926295391.
E-mail address: vicente.navarro@uclm.es (V. Navarro).
when instability occurs, a sensitivity analysis allows to know
which qualitative or quantitative actions are more appropriate to
stabilize the given slope. Therefore, the sensitivity analysis is a use-
ful tool able to provide a sound assessment for the selection of the
slope stabilization method. Our main objective in this article is to
analyze the use of this sensitivity analysis tool.
2. Conceptual basis of the method
Both when limit equilibrium methods are used, and when the
kinematic approach of limit analysis is applied, if the safety factor
is defined as the ratio of the shear stress of the soil to the shear
stress at failure, slope stability is generally evaluated as a ratio:
Rb  
S G x; yðxÞ; y0 ðxÞ; yG ðxÞ; y0G ðxÞ; uðx; yÞ; p dx
F¼ ¼ Rab ð1Þ
D Q ðx; yðxÞ; y0 ðxÞ; yG ðxÞ; y0G ðxÞ; uðx; yÞ; pÞ dx
a
For the two-dimensional collapse mechanism defined in Fig. 1, a
and b are the x-coordinates of the sliding line end points, yG(x) is
the ground profile (ordinate at point x), y(x) is the ordinate of the
sliding line at point x, and y0 (x) is its first derivative. The u(x, y)
function defines the distribution of the soil water pressure. Finally,
vector p groups all the parameters together. In principle, it could be
a vectorial field (if, for example, the spatial variation of the
strength parameters is taken into account), though what usually
happens is that it contains only a discrete number of parameters
which are constant throughout the entire domain. G and Q are
two functionals that define the actions over the system. When lim-
it equilibrium methods are used, actions are identified with forces

0266-352X/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2010.03.010
838 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845

Nomenclature

a, b x-coordinates of the sliding line end points S stabilizing actions

aW, bW x-coordinates of the water table exit point at the slope s vector of sensitivities
side, and at the slope top t length of the interface between blocks
C associated cost u pore water pressure at the base of the blocks
c cohesion v virtual displacement
D driving actions x abscissa coordinate
E horizontal stabilization force y sliding line
F safety factor y0 first derivative of the sliding line
F+ value of F when the ith component of p is increased by yG ground profile
dpi yG0 first derivative of the ground profile
F- value of F when the ith component of p is decreased by a angle of inclination of the block base with respect to the
dpi horizontal
F* value of F for slopes at collapse Dxk horizontal width of blocks
G function that defines the stabilizing actions dF increase of safety factor
H slope height dF* increase of safety factor for slopes at collapse
l length of the block base dp perturbation in the value of the parameters
n number of blocks e compact notation of the differential operator defined in
np number of components of vector p Eq. (11)
p vector of parameters / angle of internal friction
Q function that defines the driving actions h slope angle

or moments. On the other hand, if the kinematic approach of limit imization can also be done by an iterative method. This is the case
analysis is used, the actions may be identified with the internal dis- of Baker [7], when he applies the dynamic programming algorithm
sipation of energy, and the external work. This procedure is re- iteratively, i.e., assuming a value for Fs, and establishes the critical
ferred to as the ‘‘kinematical approach” in this paper. slip surface yCR (x|FS), by minimizing R[y(x)|FS], obtaining a new esti-
The slope stability problem can be stated as the minimization of mate of FS by applying the Spencer’s procedure to this critical slip
the ‘‘safety functional” (as Baker and Garber [5], termed the quo- surface, and repeating the process until the assumed and resulting
tient F of Eq. (1)) to find the safety factor: values of FS are equal. Although the two preceding methods have
proved its efficiency, for the approach proposed in this paper is of
F ¼ MinyðxÞ;rðxÞ fF½yðxÞ; rðxÞg ð2Þ
interest to minimize the quotient functional directly:
where r(x) represents the distribution of stresses along y(x). To @S @D  
minimize the quotient functional, the Petrov method can be used. @F @ðS=DÞ @pi D  S @pi 1 @S @D
¼ ¼ ¼  F S ð3Þ
Petrov [6] showed that stationary ‘‘points” (functions), of a ratio @pi @pi D2 D @pi @pi
can be obtained by extremizing an auxiliary functional R = R(FS Then, standard optimization packages, as GAMS [8] for example,
)  GFS Q, where FS is the unknown minimum value of the ratio can be used. This allows to convert FS into one more variable that
G/Q, which can be evaluated from the constraint R(FS) = 0. The min- has to satisfy its definition (Eq. (1)). Therefore, it is not needed to
worry about the method of solution, because the method imple-
mented in the software package takes this into account as one con-
xf = b
straint. As a result, in keeping with the proposal of Castillo et al. [9],
αf
the components of the vector of sensitivities s of the objective
y* (x)
function with respect to p can be defined as:
R 
1 b @G @Q

@F D a @pi
 F @p dx
i
x0 = a si ¼ ¼ R   ð4Þ
@pi 1  1 b @G @Q
 F @F dx
αo D a @F
α
where for simplicity, the arguments of the functionals have been
y (x) omitted. The safety factor local sensitivities are defined as the par-
tial derivatives of the safety factor with respect to the parameter
being studied. The partial derivatives are calculated at the optimum
dx value. Thus, these sensitivities provide only a linear approximation
in a neighborhood of the optimal point, and they only indicate the
v0
direction of the action to be taken. Since small property increments
dW
αo φ dpi would produce significant changes in the critical slip surface, the
simultaneous variation of all the variables and functions involved is
φ taken into account, included the slip surface.
v (α)
α φ When idealized examples are analyzed, the slope stability can
be studied analytically, and the sensitivity of the solution can be
φ αf characterized on the basis of the parameters using Eq. (4). In many
vf dl α practical applications, however, these analytical computations can-
v
φ not be carried out. In these cases, the S/D ratio is generally discret-
ized by means of slices (limit equilibrium method) or blocks
Fig. 1. Collapse mechanism, typical slice, and hodograph of the movement. (kinematical approach). When these approaches are adopted, the
 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845 839

problem is usually solved by using numerical tools [10,11]. The the slope. Note that since sensitivities are partial derivatives at
simplest way of approaching s consists on the application of a fi- the optimal point, they indicate which direction to follow. These
nite difference scheme: directions obviously change when one move further from the opti-
mal point, and recalculation is then necessary.
dF F þ  F 
si ¼  ; i ¼ 1  np ð5Þ It should be bear in mind that if the linearization of dF is not
dpi 2dpi adopted, the constraint of Eq. (9) will entail the implicit resolution
where F+ defines the value of F when the ith component of p is in- of the optimization problem related to the computation of a safety
creased by dpi, with F– having a similar definition but with pi factor F equal to F* + dF*. This causes a significant increase of the
decreasing. If vector p contains np components, it would be required computing time, including pre and post-processing. It will become
to solve 2  np minimization problems, in addition to the funda- even more evident if the stabilization project is formulated like a
mental problem for determining F. For this reason, it is more effi- decision-making process (see, for example, [13]), and it will
cient to introduce the following discretization of Eq. (4) into the include cost/benefit analyses that will generally involve risk
numerical solver: assessment and the probabilistic analysis of collapse and the corre-
sponding cost (see, for example, [14]). Then, a huge number of res-
1
Pn  @G @Q

@F D k¼1 @pi  F @pi Dxk olutions of Eq. (9) will be needed. Although these type of studies
k
si ¼ ¼ P   ð6Þ are not within the scope of this article, the methodology put forth
@pi 1  D1 nk¼1 @G @Q
 F @F k Dxk
@F here can indeed be used in such cases.
where n is the number of slices/blocks, and Dxk is the horizontal
width of the kth slice/block. To evaluate Eq. (6) it is necessary
to have previously obtained the values of the partial derivatives 3. Slope model and stabilizing actions considered
oG/opi, oQ/opi, oG/oF, and oQ/oF. To this end, the use of a symbolic
programming code for obtaining closed-form expressions of these Slopes as those illustrated in Fig. 2 are considered, where WT
functions makes this task easier. denotes the position of the water table, and CSS refers to the crit-
Once s is computed, the increase dF undergone by the safety fac- ical slip surface. The collapse mechanism is divided into n blocks.
tor after introducing a perturbation of value dp in the parameters In each block the potential action of a horizontal stabilization force,
can be estimated as: Ei is considered. Also considered is the apparent increment of cohe-
sion, dcj, at the base of each block by soil nailing or micropiling. Not
np
X only will the resistant parameters (cohesion c and the angle of
@F ¼ s  @p ¼ si  @pi ð7Þ
internal friction / when using the Mohr–Coulomb yield condition)
i¼1
be elements of p, but it will also include the basic characterization
This dp action will entail an associated cost C, which may gen- of slope geometry (slope angle, h, and height, H), as well as the
erally be a non-linear function of the variation in the parameters: parameters used to approach the pore water pressure distribution
np
X (exit point at the slope side, aW, and contact with the ground sur-
C ¼ Cð@pÞ ¼ C i ð@pi Þ ð8Þ face at the slope top, bW as indicated in Fig. 2). Consequently, the
i¼1 vector of parameters p will have np = 5 + 3  n components
p = (/1, . . ., /n; c1, . . ., cn; E1, . . ., En; aW, bW; H, h).
Once the cost functions Ci have been defined, the stabilization
The reduction of H and h, simultaneous or not, will allow repro-
problem can be dealt with as if it were a minimization problem:
ducing many of the improvements that are usually adopted in
Min C subject to @F ¼ @F ð9Þ practice. If flattening of the slope angle is considered, one must
@p
proceed with caution, since one must first be sure that there are
i.e., what is looked for is the minimum cost required for a given no deeper-seated slip surfaces that may become mobilized when
improvement dF* in the safety factor. unloading the slope toe [15]. To avoid this problem, the ground sit-
Close to failure, it is usually needed to make a quick decision to uation must be characterized with the utmost care before taking
prevent the collapse evolution. Then, moving from a safety factor any action. In any case, before modifying the slope profile it would
1.00 to a safety factor of 1.05 can be sufficient provided that the be interesting to find out the position of the ‘‘neutral point” [15], as
soil parameters are derived from a well documented back analysis this will help to understand the behavior of the slope. In order to
[12]. In this case, the linear approach (Eq. (7)) can be valid. How- do this, it must be calculated the sensitivity of the system to local
ever, if large changes are done, perhaps additional calculations perturbations of the slope profile. The point where the sensitivity
must replace the linear approach to evaluate the final safety of changes sign will allow to locate the position of this neutral point.

bw

θ
WT

CSS
aw

Δcj
Ei
li

Fig. 2. Rectilinear slope considered in this article.

840 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845

This sensitivity cannot be obtained from Eq. (6). It would be possi- ative displacement between blocks i and i  1, ci represents the
ble to adopt an approach in finite differences which is analogous to cohesion, Wi is the weight, ai is the angle of inclination of the block
the one put forth in Eq. (5). However, even if one only considers the base with respect to the horizontal, Ei is the horizontal stabilizing
same number of potential local perturbations of the slope profile force acting on the block, ui is the pore water pressure at the base
yG(x) as slices/blocks considered in the discrete analysis, i.e., the of the block, [ui] is the average pore water pressure on interface i,
n terms oF/oyG(xi); i = 1,2, . . ., n, it is more efficient to use the var- and /di = atan(tan /i/F). Both the cohesion and the internal friction
iational formulation of Castillo et al. [4] and obtain a sensitivity angle have been assumed to be the same at the contact between
as reported in Castillo et al. [9]: blocks as at the bases of the blocks, although the apparent increase
  of cohesion at the base of each block dcj is considered. Pore water
@F 1
¼ ðeðGÞ  F  eðQÞÞ ; i¼1n ð10Þ pressures have been treated as external forces. Additional vertical
@yG ðxi Þ D x¼xi loads have not been considered.
with e being a compact notation of the differential operator:
2
4. Application examples
@ð
Þ d @ð
Þ d @ð
Þ
eð
Þ ¼  þ ð11Þ
@yG dx @y0G dx2 @y00G As two examples of the formulation and methodology proposed
here, this section presents an analysis of the stabilization of the
where the symbol ‘‘
” represents functions G and Q. Once again, it is
Selset landslide (Fig. 3) [19,20], and the Sudbury Hill slip (Fig. 4)
advisable to obtain e(G) and e(Q) with the help of a symbolic pro-
[20]. Although both slides took place in the past, their stabilization
gramming code.
is discussed as if it were a present-day task.
With respect to water pressure, it has been accepted the
The parameters that characterize the two slopes are given in
Dupuit–Forchheimer assumption. Therefore, for the plane flow un-
Table 1. In both cases six stabilization approaches are considered
der consideration, the free surface has been approximated by
(see Fig. 5): (1) excavation at the slope top, (2) flattening the slope
means of a parabola. The pore water pressure is assumed equal
angle, (3) water table depletion near the slope side (surface drain-
to the vertical distance to the water table, although it is a conser-
age), (4) drainage at a certain distance from the slope (deep drain-
vative estimate, because the computed pressures are higher than
age), (5) installation of passive piles, and (6) use of micropiles or
the actual ones are adopted [16]. Moreover, the incipient collapse
soil nails. As it is usually done, the piling action is introduced into
process is considered to be fully drained where the dilation or con-
the stability calculation as an external stabilizing force [21–23]. In
traction of the soil skeleton does not cause any change in pore
turn, as first approximation, the arrangement of micropiles or soil
water pressure [17].
nails is introduced as the apparent localized cohesion increase dcj
Among the many existing formulations available to define
indicated in Fig. 2.
safety, the kinematical approach proposed by Michalowski [18]
The reduction of H may be achieved not only by excavation at
has been chosen. If the translational collapse mechanism defined
the top, but also by filling at the toe of the slope. In both examples,
in Fig. 1 is adopted, and the safety factor F is defined as the maxi-
however, there are elements (a watercourse in Selset, and a railway
mum available utilization of the shear stress in order to bring the
in Sudbury) which hindered filling at the toe. Therefore, the only
slope to failure, it may be computed as [18]:
option considered was an excavation at the crest of the slope. This
F type of excavation is usually performed as a single terrace with a
Pn
i¼1 ðli v i þ t i ½v i Þci cos/di
horizontal surface and a steep slope at the back [24] (see Fig. 5).
¼ Pn As a working hypothesis, it was assumed that the excavation cov-
i¼1 ðW i sinð a i  / di Þ  E i cosðai  /di Þ þ li ui sin /di Þv i þ t i ½v i ½ui  sin /di

ð12Þ
where subscript ‘‘i” denotes the block number, li is the length of the
block base, ti is the length of the interface between blocks i and
i  1, vi is the virtual displacement, [v]i is the magnitude of the rel-
ered an area whose distance from the slope crest was equal to half
the slope width (toe-to-crest horizontal distance).
As it was mentioned earlier, cost estimation is not a simple task.
Any estimation that does not reflect the real situation of the time
and place where the slides took place may very well lead to conclu-

WT

Average inclination Boulder

of slope 28º Clay
Aluvial SC
12.8 m
Gravel

River
CSS
Lune

CS
Sandstone
Boulder Sandstone and Shale
Clay

Fractured Shale

0 5 10 15 20m
Fig. 3. Selset landslide, adapted from Skempton [20]. Critical slip surface CSS identified in our computations, slip circle SC reported by Skempton [20], and critical slip circle CS
identified by using SLOPE–W (GEO-SLOPE, 2002).
 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845 841

WT Slope Section (1900)

Section after slip

(surveyed 1956) 7.01 m
Brown London Clay

SC
CSS

CS

0 5 10 15 20m
Fig. 4. Sudbury Hill slip, adapted from Skempton [20]. Critical slip surface CSS identified in our computations, slip circle (SC) reported by Skempton [20], and critical slip circle
CS identified by using SLOPE–W (GEO-SLOPE, 2002).

Table 1 kN of imposed resisting effort put forth by Rodriguez Ortiz et al.

Parameters characteristic of the Selset land- [12] to obtain the relative costs r and the cost functions listed in
slide and the Sudbury Hill slip.
Table 2. These functions were introduced into Eqs. (8) and (9) to
Selset Sudbury design the stabilization.
H (m) 12.8 7.01
h (°) 28.0 18.4
4.1. Selset landslide
aW (m) 10.24 1.05
bW (m) 8.00 63.90
cn (kN/m3) 22.3 18.8 The Selset landslide (see Fig. 3) was analyzed and documented
c (kPa) 8.6 2.9 by Skempton and Brown [19], and Skempton [20]. The slide oc-
/ (°) 32 17 curred upstream of the Selset dam in north Yorkshire (England),
F 1.09 1.00
in boulder clay.
Consistent with the parameters of Table 1, the neutral point of
the slope is located at the position indicated in Fig. 6a. This figure
sions that are not likely to be true. Even at the same time and place, also includes the sensitivities with respect to the local variations of
two different companies might come up with different manage- the slope profile (Eq. (10)). In keeping with Hutchinson’s [15] find-
ment costs. Nonetheless, it must be bear in mind that the examples ings, the sensitivity is asymmetrical, showing that corrective fills at
analyzed here have no intention of reproducing real conditions. the toe are preferable (more effective) to corrective cuts at the slide
Rather, the interest is focused on illustrating the proposed method- head. However, as it was indicated earlier, given that the Lune Riv-
ology. However, the examples intend to be as consistent as possi- er is at the toe, it has been only considered the excavation at the
ble. So, the managed costs should not be preposterous. It can be crest.
used, for instance, tables of unit costs like the one proposed in Column 2 of Table 3 lists the sensitivities with respect to the dif-
Seattle by its Department of Planning and Development [25]. How- ferent parameters (in the case of the horizontal forces and the in-
ever, for simplicity, it was decided to use the relative cost for each crease in cohesion, it has been included the sensitivities obtained

D ½D δAW (δbW)

δAW (δaW)

Fig. 5. Stabilization approaches considered in the two examples analyzed in this paper.
842 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845

Table 2 Table 3
Relative cost r of the imposed resisting efforts (adapted from [12]), Sensitivities of the safety factor with respect to different parameters and required
and cost functions used in the computations. changes in the individual parameters to produce an increase of 0.03 in the safety
factor for the Selset and Sudbury cases. Values obtained by using Eq. (4).
# Relative cost Cost functions
  Sensitivity Changes
1 r1 = 1 dH
dH 0:5H þ 2 tan h  r1
 
2 r2 = 2 H2 1 1 Selset Sudbury Selset Sudbury
2 tanðhdhÞ  tan h  r 2
3 r3 = 1.4 dAW ðdaW Þ  r 3 @F/@/ 1.626E+00 2.713E+00 d/ (°) 1.057E+00 6.336E01
4 r4 = 1.4 dAW ðdbW Þ  r 4 @F/@H 2.843E02 3.449E02 dH (m) 1.055E+00 8.697E01
5 r5 = 1.8 dE  r 5 @F/@h 2.079E+00 1.728E+00 dh (°) 8.266E01 9.949E01
6 r6 = 2.3 l @F/@aW 2.018E02 3.048E03 daW (m) 1.486E+00 9.844E+00
cos a dc  r6
@F/@bW 1.787E03 1.388E03 dbW (m) 1.679E+01 2.161E+01
@F/@c1 1.324E03 2.779E03 dc1 (kPa) 2.267E+01 1.080E+01
@F/@E1 1.417E03 2.092E03 dE1 (kN) 2.118E+01 1.434E+01
when action is undertaken at the toe of the slope, which exhibits
maximum sensitivity in both cases; see Fig. 7). The sensitivity val-
ues are not easy to compare, as the units are different. Nonetheless, is unlike to put this measure into practice, it has not been consid-
it is possible to carry out a simple exercise that will facilitate the ered as a measure of stabilization (see Fig. 5). On the other hand, it
comparison and help center stabilization design. Based on Eq. (7) is also clear that although the sensitivities with respect to E and c
it has been calculated what increment each parameter must under- are similar to those with respect to bW, this result is ‘‘misleading”.
go so that by using only it, F will increase from an initial value of In order to obtain an increment of 0.03 in F, bW must increase a va-
1.09 (see Table 1) to 1.12. In this way it has been obtained column lue close to 17 m, which feasibility is questionable for practical
5 of Table 3. As it can be seen, one of the most efficient procedures purposes. Moreover, this calculation is incorrect, because for such
is to act on the angle of internal friction: by increasing it by a little a major change in the system, the linearization assumed when
(near 1°), it is achieved the desired improvement. However, since it using the approach based on Eq. (7) is not valid. Therefore, if the

a 1.20

1.00
neutral point
0.80
S or z/H

0.60

0.40

0.20

Slope Sliding line S

0.00

-0.20
-0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50
x/H

b 1.20

1.00

0.80 neutral point

S or z/H

0.60

0.40

0.20

0.00
Slope Sliding line S
-0.20
-0.50 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50
x/H

Fig. 6. Location of the neutral point (grey dot). Sensitivity S (=[z + s]/H) of the safety factor with respect to the slope profile. (a) Selset landslide; (b) Sudbury Hill slip.
 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845 843

3.0E-03
cohesion Selset horizontal force Selset
2.5E-03 cohesion Sudbury horizontal force Sudbury

2.0E-03

sensitivity 1.5E-03

1.0E-03

5.0E-04

0.0E+00
1
3
5
7
9

11
13
15
17
19

21
23
25
27
29

31
33
35
37
39
block number

Fig. 7. Variation of the sensitivity across the critical slip surface with respect to the increase in cohesion and the application of horizontal forces.

aim is to correctly analyze the effect of this increment in bW, one
must return to the whole problem of minimization based on the
calculus of F and forget about approximations. In addition, it is nec-
essary to point out that the drainage costs shown in Table 2 are
linked to changes in the water table of a more limited scope than
the one causing bW to increase by 17 m. Changes of this type have
a substantially higher cost. In view of the above, it would seem rea-
sonable to rule out the possibility of increasing bW as a stabilization
measure to be applied to the Selset landslide.
The sensitivity has been also obtained applying Eq. (5). To do so,
stability has been calculated by means of the simplified Bishop
method using the program SLOPE/W [26]. As it can be seen in
Fig. 3, the slip circle identified by SLOPE/W is quite similar to the
critical surface identified with our formulation. By applying Eq.
(5), it has been obtained the sensitivities shown in Table 4. In this
table, the increments of the parameters which are necessary to ob-
tain an increase of 0.03 in the factor of safety are also included. If
Tables 3 and 4 are compared, it can be observed that the results ob-
tained with both methods are nearly the same. However, the calcu-
lation effort invested in applying the formulation proposed in this
paper is considerably lower. The process is considerably auto-
mated, making easier the application of methods like the ones
pointed out at the end of Section 2, in which many calculations
must be done.
Once the action 4 has been rejected, it must be chosen the opti-
mal procedure between the other actions outlined in Fig. 5. That is,
the stabilization problem stated in Eq. (9) by assuming dF* = 0.03
must be solved. This is done introducing the values of s consigned
in Table 3 in Eq. (7), and using the cost functions defined in Table 2
for compute the Eq. (8). Afterwards, the Eq. (9) has been solved
with the solver utility from the well known Microsoft Excel spread-
sheet. It has been obtained that the optimal approach that should
be used is to simultaneously reduce H and aW, approach 1 for Selset
in Table 5. In this table the subscript ‘‘ST” refers to the slope toe.
With the exception of the first approach, all other approaches refer
only to the optimum approach if the cost defined in Table 2 of the
procedure 1 of the previous approaches is tenfold. Hence, if the
reduction of H is penalized (assuming that changes in the grade
are not acceptable), incurring a cost that is 10 times greater, the
simultaneous reduction of h and aW turns out to be the most effec-
tive of all actions (approach 2, Table 5). This solution is 23% more
expensive than the previous one. If the reduction of h also becomes
difficult (we have simulated again this situation by raising the cost
by one order of magnitude), the best solution, even though it is
2.87 times more expensive than the one based on the excavation
at the slope top, becomes the use of micropiles, once more, in com-
bination with water table depletion near the slope side (approach
3, Table 5). If the cost of the micropiles also increases tenfold, the
best option is the mixed approach based on the installation of pas-
sive piles combined with a decrease in aW (approach 4, Table 5). In
this case, the stabilization is 3.32 times more expensive than in ap-
proach 1.
If the costs incurred by grading and the arrangement of stabil-
ization elements (piles and micropiles) are 10 times higher than
the value defined in Table 2, the solely reduction of aW will be
the optimal method of choice (approach 5, Table 5). Nevertheless,
it will cost 5.15 times more that the one put forth in approach 1.
Owing to the elevated position of the water table in Selset, in order
to obtain the desired stability only by means of drainage (decrease
in aW or increase in bW), a major action must be undertaken, which
means that this will be a relatively costly (and perhaps slowly)

Table 5
Table 4 Proposed stabilization approaches and associated costs.
Sensitivities of the safety factor with respect to different parameters and required
Approach Procedure 1 Procedure 2 Cost Cost/
changes in the individual parameters to produce an increase of 0.03 in the safety
maximum
factor for the Selset and Sudbury cases. Values obtained by using Eq. (5).
cost
Sensitivity Changes Selset 1 dH = 0.606 m daW = 0.633 m 7.477 1.00
Selset Sudbury Selset Sudbury 2 dh = 0.47° daW = 0.633 m 9.164 1.23
3 dcST = 12.7 kPa daW = 0.633 m 21.44 2.87
@F/@/ 1.547E+00 2.865E+00 d/ (°) 1.111E+00 6.000E01
4 dEST = 12.2 kN daW = 0.633 m 24.8 3.32
@F/@H 3.000E02 3.500E02 dH (m) 1.000E+00 8.571E01
5 daW = 1.486 m – 38.48 5.15
@F/@h 1.948E+00 1.662E+00 dh (°) 8.824E01 1.034E+00
@F/@aW 2.100E02 3.000E03 daW (m) 1.429E+00 1.000E+01 Sudbury 1 dH = 0.870 m – 5.319 1.00
@F/@bW 1.750E03 1.500E03 dbW (m) 1.714E+01 2.000E+01 2 dh = 1° – 9.007 1.69
@F/@c1 1.250E03 2.500E03 dc1 (kPa) 2.400E+01 1.200E+01 3 dcST = 11.0 kPa – 16.47 3.10
@F/@E1 1.500E03 2.000E03 dE1 (kN) 2.000E+01 1.500E+01 4 dEST = 14.3 kN – 25.81 4.85
844 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845
approach. However, given the large quantity of water, drainage is
always a positive solution and constitutes an important comple-
mentary procedure for other stabilization approaches.
4.2. Sudbury Hill slip
The Sudbury Hill slip (see Fig. 4) took place in 1949 [20]. This
slip occurred in weathered London Clay, with resistant parameters
considerably lower than the uniform and almost any weathering
boulder clay of the Selset landslide (see Table 1). In Sudbury, the
water table was substantially lower than in Selset (see Figs. 3
and 4) [20,27,28].
Like in the Selset landslide, both the position of the neutral
point and the sensitivity (see Fig. 6b) indicate that the loading of
the slope toe by filling is preferable to the unloading of its head
by cutting. In this case, however, the sensitivities to local variations
of the slope profile are more symmetrical. Nevertheless, as men-
tioned above, the fact that there is a railway at the slope toe has
prompted to consider grading only by removing soil from the slide
head.
As can be observed in column 3 of Table 3, the low water table
position makes the sensitivity of the safety factor to the drainage
action to be small, similar to the sensitivity of the Selset landslide
with respect to the bW reduction. Therefore (see column 6), a major
drainage is needed to obtain an improvement in the safety factor of
just 3% (of the initial value from 1.00 to 1.03, see Table 1). How-
ever, as discussed earlier, this result is incorrect, because an action
involving a decrease of aW by 9.8 m, or distancing bW by 21.6 m,
means that the water table must be situated below any critical slip
surface. If the safety of the dry slope is calculated, a safety factor
equal to 1.30 will be obtained. Therefore, for this large perturba-
tion, it is clearly erroneous to use the linear approach defined by
Eq. (7). In any event, the low sensitivity with respect to aW and
bW indicate that drainage is not the most effective strategy to ob-
tain a reduced improvement of F. Thus, it was not considered as
a stabilization measure applicable to the Sudbury Hill slip. While
in Selset the water table was so high as to make its stabilization
unfeasible by only using drainage, in Sudbury it was so low that
for the effect to have any impact, there must be a substantial water
table depletion.
Proceeding like in the case of Selset, the sensitivity was also cal-
culated for Sudbury landslide by means of Eq. (5) (see Table 4). It
was also applied the simplified Bishop method, implemented in
the SLOPE/W [26] code. Again, both the sensitivities and the incre-
ments dpi calculated with the aim of achieving an increment of 3%
in the safety factor, are practically equal to those obtained by
applying the proposed methodology in this paper (see Tables 3
and 4). However, as it was previously indicated, the method pro-
posed in this paper could be easily applied.
After solving Eq. (9), the results indicate that the optimal action
to be undertaken to raise the safety factor to an equivalent value of
1.03 consists of reducing H, without changing any of the other
parameters (approach 1 for Sudbury in Table 5). This type of ‘‘sin-
gle action” response will be recurrent in the other hypotheses ana-
lyzed. Thus, if the cost of the excavation at top is multiplied by 10,
flattening the slope becomes the optimal action (approach 2, Table
5), with its cost being 1.69 times higher than that of reducing H. If
the cost of reducing h is also increased tenfold, the installation of
micropiles at the toe of the slope will be the most effective action
(approach 3, Table 5). Its cost, however, is 3.10 times greater than
removing soil from the top of the slope. Lastly, if the cost of the
micropiles is also increased by one order of magnitude, the instal-
lation of passive piles at the slope toe will be the best alternative
(approach 4, Table 5). Here the stabilization cost is 4.85 times high-
er than that of reducing H. A summary of these results is presented
in Table 5.
5. Conclusions
A new sensitivity analysis of slope stability has been presented.
The approach is especially appropriate to be used with computa-
tional modules in which the location of the critical slip surface is
resolved as an optimization problem. In schemes of this type,
which may be interpreted as a discretization of the classic varia-
tional approach to solving the slope stability problem, Castillo
et al. [9] have reported that the sensitivity analysis can be carried
out in the same way as is done with optimization techniques.
Hence, Eqs. (6) and (10) provide a quick way for obtaining the sen-
sitivities. The sensitivities obtained could be introduced into Eq. (7)
to solve Eq. (9), and thus lead to the relatively simple formulation
of slope stabilization design as a minimization problem.
The main drawback to the method stems from the applicability
of Eq. (7). Only in cases where the action to be undertaken does not
entail a major change in the system is it reasonable to assume that
the variation in the safety factor may be calculated by means of the
linear approach defined by Eq. (7). At failure, when urgent actions
must be adopted, it is often of interest to use procedures of this
kind which will result in only a slight increase in F. However, when
sensitivity with respect to a parameter is small, even a slight in-
crease in the safety factor will mean a substantial variation in
the parameter, so the formulation put forth here will not be
applicable.
Acknowledgements
This research was financed in part by a Research Grant awarded
to the 3rd author by the Education and Research Department of the
Castilla-La Mancha Regional Government and the European Social
Fund within the framework of the Integrated Operative Pro-
gramme for Castilla-La Mancha 2000–2006, approved by Commis-
sion Decision C(2001) 525/1. The authors are also indebted to the
Spanish Ministry of Science and Technology (Project BIA2005-
07802-C02-01) for partial support.
References
[1] Baker R, Leshchinsky D. Spatial distribution of safety factors. J Geotech
Geoenviron Eng 2001;127(2):135–45.
[2] Hoek E. Practical rock engineering; 2007. <http://www.rocscience.com/hoek/
PracticalRockEngineering.asp>.
[3] Castillo E, Conejo AJ, Pedregal P, García R, Alguacil N. Building and solving
mathematical programming models in engineering and science. Pure and
applied mathematics. New York: Wiley; 2001.
[4] Castillo E, Conejo AJ, Castillo C, Mínguez R, Ortigosa D. A perturbation approach
to sensitivity analysis in mathematical programming. J Optim Theory Appl
2006;128(1):49–74.
[5] Baker R, Garber M. Theoretical analysis of the stability of slopes. Géotechnique
1978;28(4):395–411.
[6] Petrov IP. Variational methods in optimum control theory. New
York: Academic Press; 1968.
[7] Baker R. Determination of the critical slip surface in slope stability
computations. Int J Numer Anal Methods Geomech 1980;4(4):333–59.
[8] Brooke A, Kendrick D, Meeraus A, Raman R. GAMS users guide. Washington
(DC): GAMS Development Corporation; 1998.
[9] Castillo E, Conejo AJ, Aranda E. Sensitivity analysis in calculus of variations.
Some applications. SIAM Rev 2008;50(2):455–81.
[10] Cheng YM. Location of critical failure surface and some further studies on slope
stability analysis. Comput Geotech 2003;30(3):255–67.
[11] Pham HTV, Fredlund DG. The application of dynamic programming to slope
stability analysis. Can Geotech J 2003;40(4):830–47.
[12] Rodriguez Ortiz JM, Alonso EE, Navarro V. Stabilization of landslides. European
technical committee ERTC 1. International society for soil mechanics and
geotechnical engineering. Unpublished manuscript; 2001.
[13] Turner AK, Schuster RL. Landslides. Investigation and mitigation.
Transportation research board special report 247. Washington (DC): National
Academy Press; 1996.
[14] Abramson LW, Lee TS, Sharma S, Boyce GM. Slope stability and stabilization
methods. 2nd ed. New York: John Wiley & Sons; 2001.
[15] Hutchinson JN. An influence line approach to the stabilization of slopes by cuts
and fills. Can Geotech J 1984;21(3):363–70.
 V. Navarro et al. / Computers and Geotechnics 37 (2010) 837–845
[16] Kim J, Salgado R, Lee J. Stability analysis of complex soil slopes using limit
analysis. J Geotech Geoenviron Eng ASCE 2002;128(7):546–57.
[17] Michalowski RL. Stability charts for uniform slopes. J Geotech Geoenviron Eng
2002;128(4):351–5.
[18] Michalowski RL. Slope stability analysis: a kinematical approach.
Géotechnique 1995;45(2):283–93.
[19] Skempton AW, Brown JD. A landslide in boulder clay at Selset, Yorkshire.
Géotechnique 1961;11(4):280–93.
[20] Skempton AW. The fourth Rankine Lecture: long-term stability of clay slopes.
Géotechnique 1964;14(2):77–102.
[21] NAVFAC. Design manual 7.01-soil mechanics. Naval facilities engineering
command, Alexandria, Virginia; 1986. <www.wbdg.org/ccb/DOD/UFC/ufc_3_
220_10n.pdf>.
[22] Viggiani C. Ultimate lateral load on piles used to stabilize landslides. In:
Proceedings of the 10th international conference on soil mechanics and
foundation engineering, Stockholm, vol. 3; 1981. p. 555–60.
845
[23] Poulos HG. Design of reinforcing piles to increase slope stability. Can Geotech J
1995;32(5):808–18.
[24] Larsson R, Åhnberg H. Long-term effects of excavations at crests of slopes.
Swedish geotechnical institute report no. 61. Linköping: Swedish Geotechnical
Institute Literature Service; 2003.
[25] Department of Planning and Development of Seattle. Seattle landslide study;
2005. <http://www.cityofseattle.net/DPD/Landslide/Study/>.
[26] GEO-SLOPE. Slope/W for slope stability analysis. Version 5. Calgary, Alberta:
GEO-SLOPE International Limited; 2002.
[27] Chandler RJ, Skempton AW. The design of permanent cutting slopes in stiff
fissured clays. Géotechnique 1974;24(4):457–66.
[28] Skempton AW. Slope stability of cuttings in Brown London clay. In:
Proceedings of the 9th international conference on soil mechanics and
foundation engineering, Tokyo, vol. 3; 1977. p. 261–70.