ARTICLE IN PRESS

International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

www.elsevier.com/locate/ijrmms

Technical note

Forecasting potential rock slope failure in open pit mines

using the inverse-velocity method
N.D. Rosea, O. Hungrb,
a
Piteau Associates Engineering Limited, 215-260 West Esplanade, North Vancouver, B.C., Canada
b
Department of Earth and Ocean Sciences, University of British Columbia, 6339 Stores Rd., Vancouver, Canada
Accepted 28 July 2006
Available online 2 October 2006

Keywords: Inverse-velocity; Displacement monitoring; Slope failure prediction; Contingency planning

1. Introduction The inverse-velocity method, developed by Fukuzono [7]

in 1985, provides a useful tool for interpretation of
Assessment of rock slope failure mechanisms requires an instrument data with the objective of anticipating eventual
understanding of structural geology, groundwater and slope failure. Astonishingly, even though this approach
climate, rock mass strength and deformability, in situ was developed based on laboratory tests more than 20
stress conditions and seismicity. Stress relief associated years ago, it does not appear to have been applied for real-
with mining excavation leads to elastic rebound and time slope failure prediction in the mining industry until
ground relaxation displacements that dissipate with time, 2001 when it was used to predict the first of three large-
a process that is often referred to as time-dependent scale slope failures presented in this paper. Experience with
deformation [1,2]. With continuing excavation, regressive the inverse-velocity approach for large-scale slope move-
slope displacements may occur in a cyclical accelerating/ ments in poor and fair quality rock masses, has led to
decelerating fashion. As strain levels increase, strain accurate prediction of three large failures ranging in size
softening may lead to plastic (non-recoverable) deforma- from 1 to 18 million cubic metres (M m3). This paper
tion and progressive failure development [3]. Displacement presents the data and methodology used in these case
rate (velocity) is commonly considered the best indicator of histories, but also provides more general discussion of the
the failure process. inverse-velocity method as a tool for interpreting displace-
Monitoring is used in mines in order to anticipate ment monitoring results, its advantages and limitations.
possible acceleration or failure of a moving slope mass. The original examples presented here display kinematic
Available instruments include precise survey stations and behaviour that is remarkably consistent, making the
prisms, wire and rod extensometers, inclinometers, tilt- application of the method seem rather simple. The authors
meters, Global Positioning System (GPS) devices and recognize that such conditions do not always exist. For this
geophones to record the intensity of ground noise. reason, a detailed discussion has been included of the types
Comprehensive descriptions of surface and subsurface of variation of displacement rates that can precede failure
monitoring, data collection and assessment methods are of large rock slides. The discussion is illustrated by typical
included in Dunnicliff [4], Turner and Shuster [5] and examples drawn from the published literature, involving
Wyllie and Mah [6]. The common approach towards mine slopes as well as natural rock slopes.
interpretation of monitoring data is to convert measure-
ments to rates (velocities), assuming that, in general, a 2. Inverse-velocity method
slope failure will be preceded by increasing rates of
displacement, strain or micro-seismic activity. The concept of inverse-velocity for predicting slope
failure time was developed by Fukuzono [7] based on
Corresponding author. previous Japanese work and on large-scale well-instrumen-
E-mail address: oHungr@eos.ubc.ca (O. Hungr). ted laboratory tests simulating rain-induced landslides in

1365-1609/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijrmms.2006.07.014
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 309

co-exist and interact, as described in the following

paragraphs (see also [11]).

3.1. Measurement error and random ‘‘noise’’

INVERSE VELOCITY (1/V)

In large open pits, considerable distances may be

required to survey prisms in the direction opposing slope
movement. Optical refraction and diurnal temperature and
pressure effects can result in appreciable error, even with
the most accurate monitoring systems. A certain amount of
filtering is required to differentiate displacement measure-
Failure ments and obtain estimated movement rates, free of
‘‘noise’’. Unfortunately, no general rule regarding the
degree of filtering can be given, as it depends on the
measurement time interval, as well as the type and quality
of the measurements. A reasonable way to approach this
t0 t tf
issue, used in Section 4, is to analyse a time series of
TIME measurements to a fixed target positioned at a distance
Fig. 1. Inverse-velocity versus time relationships preceding slope failure— similar to the survey prisms and determine the optimal
after Fukuzono [7]. filtering method by trial and error to remove random
variation, without concealing trends.
The simplest filtering, used for the data reported in
soil. The conditions simulated in the laboratory were Section 4, averages the displacement rate over the last
considered to be characteristic of accelerating creep (i.e., several (‘‘n’’) readings from a densely sampled time series.
slow continuous deformation) under gravity loading. Another method uses linear regression analysis of groups
When the inverse of observed displacement time rate of measurement points, beginning with the most recent
(‘‘inverse-velocity’’) was plotted against time, its values measurement and counting backwards. The two alternative
approached zero as velocity increased asymptotically algorithms are described in the Appendix A. With either
towards failure. A trend-line through values of inverse- method, the degree of filtering increases as the chosen
velocity versus time could be projected to the zero value on length of interval and number of included measurements
the abscissa (x-axis), predicting the approximate time of increase. When applied to a fixed reference point, the
failure, as shown in Fig. 1. Fukuzono presented three types corrected displacement rate should tend to zero. When
of plots fitted to the laboratory data (i.e., concave, convex applied to a group of slope measurements, it will give the
or linear), defined by the following equation: estimated current displacement rate (velocity).
Provisions can also be used to reduce the effect of
V 1 ¼ ½Aða  1Þ1=a1 ðtf  tÞ1=a1 , (1) instrument error, including:
where t is time, A and a are constants and tf is the time of
failure. In the laboratory measurements preceding failure, a  measurement of slope movements at approximately the
was found to range between 1.5–2.2. As shown in Fig. 1, same time of day to reduce diurnal effects;
the curve of inverse-velocity is linear when a ¼ 2, concave  minimizing the number of surveyors to reduce the
when ao2 and convex when a42. Based on the results of influence of operator error;
laboratory testing, Fukuzono concluded that a linear trend  using only displacement readings measured in the line of
fit through inverse-velocity data usually provided a reason- sight or ‘‘slope distance’’, thereby reducing the influence
able estimate of failure time, shortly before failure. of horizontal and vertical angular error of a total
Several authors showed theoretically that, given ductile, station, which tends to be greater. This is not meant to
accelerating creep occurring under constant effective stress diminish the importance of also deriving complete
conditions in soil, rock and other materials, the inverse- movement vectors that are useful in aiding interpreta-
velocity plot would in fact be expected to be linear [8–10]. tion of the movement mechanism. Also, in some cases,
the dominant vector component may not coincide with
the sight distance (e.g., vertical).
3. Variation of displacement rates preceding slope failure

Fukuzono’s experiments were carried out using model 3.2. Local movements
sand slopes under closely controlled laboratory conditions
in a rainfall simulator. Obviously, such conditions rarely A large rock slide rarely moves as a completely rigid
exist in natural or mining rock slopes. Under typical field block. Fragments of various sizes, often situated around
conditions, six types of variation of displacement velocity the perimeter of a central coherent or semi-coherent mass,
 ARTICLE IN PRESS
310 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

can be subject to localized movements such as minor slides imposed on a progressive and approximately linear (on the
or topples, bulging or buckling of beds, rotations and inverse plot) trend of acceleration. Vibert et al. [12] fitted a
possibly abrupt, local deformation adjustments caused by linear trend line to the mean inverse-velocity during the last
crack opening. The result can be random, chaotic two years of observations in mid-1986 and predicted failure
displacement that is of itself unsuitable for detection of in 1988 or 1989. Target 9 on the slide reached its maximum
any large-scale trends. It is not unusual for local move- velocity of 11 cm day1 at the low point of the annual cycle
ments to contradict the movement trend of the main slide in the fall of 1987. However, by then, an approximately
body, both in space and time. Crosta and Agliardi [11], circular rupture surface developed through the hinge zone
dealing with a complex natural rock slope failure compris- of the toppling, the cumulative displacement approached
ing both sliding and toppling, present several examples of 100 m, the toe thrust into the soil deposits of the valley
chaotic displacement records, obviously representing local floor and the slide began to self-stabilize [13]. The
deformations. It is necessary to have a sufficiently large movement rate trend reversed at this point, although the
number and wide distribution of measuring devices to seasonal waves continued.
allow those records affected by local movements to be Another example of a record characterized by cyclic
identified and separated from records that are more variation, this time in a shorter time frame, is the 6 M m3
representative of the overall trend. On-going re-assessment wedge failure at Liberty Pit in Utah, described by [3] and
of the failure mechanism and its relation to the geometry shown in Fig. 3. A change in trend was observed at 40 days,
and structure of the slope is important and this also helps corresponding to the onset of the failure process (termed
to optimize the placement of instruments in view to ‘‘progressive’’ stage by Zavodni and Broadbent [3]). The
avoiding confusion caused by local movements. prominent one to two-week cycles coincided with blasting
activity in the pit and were superimposed on a concave trend
3.3. Periodic variation that, nevertheless, pointed to the eventual failure date and
became approximately linear during the last 25 days (Fig. 3c).
Displacement rates will be affected by periodically Fig. 3 shows that the cycling variation pattern is easier to
changing factors such as precipitation, snowmelt, freezing interpret on the inverse-velocity plot. The pattern is clear
of groundwater inflow or discharge paths, mining activity enough on the direct velocity-time plot (Fig. 3a) and
and blasting vibration. Periodic changes appear as wavi- indicates very significant accelerations in response to
ness of the record, both on direct velocity-time plots and blasting vibration at the beginning of each cycle. However,
inverse-velocity-time plots, superimposed on the longer- only the inverse-velocity plot (Fig. 3b) clarifies the
term trend line. significance of these accelerations: the amplitude of the
Fig. 2 shows the only published example known to the cycles is of the same order as the distance to the horizontal
authors of failure forecast using the Fukuzono inverse- axis, indicating that each period of blasting had the
velocity method. The example is the 49 M m3 massive potential to trigger failure. The overall trend is also easier
flexural topple at La Clapiére in the Maritime Alps, France to extrapolate on the inverse plot, as it approaches linearity
[12]. The landslide exhibited distinct annual cycles, super- within the final weeks before failure.

3.4. Displacement rate trends

While the above-described phenomena complicate the

picture, in many cases a clear trend can be discerned that
allows prediction of the failure date to be made. The
original examples presented in Section 4 later in this paper
illustrate cases in which the trend is dominant and
consistent, over-riding all other types of variability. In
these cases, the trend was also approximately linear over
periods of several days to several weeks preceding failure.
Crosta and Agliardi [11] re-analysed rock slide monitoring
data reported in the literature using a non-linear curve
fitting technique based on Eq. (1), applied over a variety of
time periods and found that for five cases out of seven the
trend line was approximately linear (a2), while it was
concave (ao2) for the remaining two cases.

3.5. Trend changes

Fig. 2. Long-term inverse-velocity plot of displacements, Target 9 on the
La Clapiére slide (data from [12]). The straight dashed line is the
prediction fit used by Vibert et al. [12]. The cross symbols are subsequent A linear inverse-velocity versus. time trend may well be
measurements [13]. fundamental for progressive creep failure of rock and other
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 311

An excellent example of a failure process controlled by

observable external factors is the 1963 Vaiont landslide
[14,15]. Fig. 4 summarizes the history of observations over
the last 2 years preceding the slide, based on data presented
by Müller [14]. The time series plotted include precipitation
(a), water level in the reservoir (b), displacement rate (c)
and its inverse (d). The last graph shows that reasonably
systematic trends can be observed, but only within certain
discrete ‘‘compartments’’ of the record. The first accelera-
tion, culminating at 400 days (October, 1962) is clearly
influenced by the second episode of reservoir filling, but
was probably also aggravated by the heavy precipitation at
360 and 370 days (late October, 1962). The trend reversed
with coincident reservoir lowering and decreased precipita-
tion. It is interesting to note from Fig. 4d that a failure
warning of approximately 7 days could have been issued at
that time. The warning would have been subsequently
cancelled, once the trend reversal was confirmed.

Fig. 3. Displacement velocity (a) and inverse-velocity (b) plots from the
liberty pit mine failure (data from [3]), (c) enlargement of the final 25 days
inverse-velocity plot.

materials [9], provided that the failure has a ductile

character and that there are no external conditions
influencing the process. However, the latter condition is
rarely satisfied in rock slopes. The form of any observed
movement trend can be influenced by changes in conditions
occurring in the background. The changes may be sudden,
such as an earthquake or a blast-related shaking episode
that can damage a key element in the structure of a
potential rupture surface, triggering accelerating creep
(cf. Fig. 3 at 40 days). At other times, a change from a
slow, regressive movement trend into progressive acceler-
ating creep can occur without an obvious trigger [3]. Fig. 4. Synchronized time series of observations over two years before the
catastrophic failure of the Vaiont Slide (data from [14]): (a) precipitation
A trend change can also set in gradually, as a result of
over 10-day periods; (b) elevation of reservoir water surface; (c)
varying climate or drainage conditions, infiltration of water displacement velocity and (d) inverse-velocity. Vertical lines 1–3 mark
into tension cracks, change in slope geometry due to on- trend changes induced by reservoir filling and precipitation. The arrow
going excavation and similar. shows the time of failure on October 9, 1963.
 ARTICLE IN PRESS
312 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

A slightly concave accelerating trend began with the

third cycle of reservoir filling at 520 days, its beginning
marked by the vertical line number 1 on the figure (May,
1963). Its concavity may reflect the convex shape of the
filling curve within this period. A distinct reduction in the
slope of the trend line occurred when the filling stopped
(vertical line 2). The new, slower trend continued for 50
days, but was sharply interrupted by the arrival of heavy
rains at 650 days, marked by line 3 (August, 1963).
The new trend turned abruptly towards the axis, moder-
ated slightly with the desperate attempt to lower the
reservoir during the last 20 days, but then progressed to
failure at 23:39 on October 9, 1963, with well-known
catastrophic consequences. The trend of the final 50 days
could be reasonably well represented by a linear fit [16].
These monitoring results lend further support to the
conclusion of Hendron and Patton [15] that the final
trigger was a combination of high-reservoir level and high
precipitation. Fig. 5. Displacement record of four extensometers on the Libby Dam
abutment wedge failure (data from [8]). The lines end on the date of
Not all processes influencing the shape of the inverse- sudden, extremely rapid failure of the wedge in early 1970.
velocity graph can be as easily visualized as those at Vaiont
and some driving factors of this type may remain hidden.
However, this case history illustrates that attempting to
find a ‘‘universal’’ equation to fit a trend over long periods 4. Case examples of predicted failure time using inverse-
of time is unlikely to succeed. Any ‘‘fit’’ to the inverse- velocity
velocity curve is valid only within its specific compartment
and must be modified as soon as a trend change is Three case examples are presented from two large open
confirmed. The authors recommend linear extrapolation of pit mines that illustrate the use of inverse-velocity for
data over varying lengths of time, looking for a consistent predicting failure time. Examples 1 and 3 occurred in 2001
trend and noting and re-evaluating any departures from it, and 2005 at Barrick Gold’s Betze-Post open pit mine
combined with observation of various controlling factors located in the Carlin Trend, northeastern Nevada. Exam-
and the developing failure mechanism. In other words, firm ple 2 occurred at another large mining operation in western
‘‘prediction’’ of the failure time probably cannot be made United States. Each of these cases involved instabilities
over a long period of time. But the inverse-velocity method that were monitored using manual and robotic total station
provides a powerful means to examine developing trends, survey of reflective prisms, and wireline extensometers. In
make and systematically revise interim predictions and each case, failure predictions approximated the failure time
rapidly and quantitatively assess the significance of to the day of the actual failure. Predictions of failure time
apparent trend changes. for the 1, 2 and 18 M m3 failures were forecasted 2 weeks, 5
days and 3 months prior to failure, respectively.
3.6. Brittle failure At both mining operations, survey distances of over 1 km
resulted in accuracies in survey measurements of about
All of the above discussion relates to cases where large- 710–15 mm, or greater. As a result, different methods of
scale, ductile failure occurred, involving high stress levels data smoothing were required to provide reasonable
relative to rock strength. Brittle rock failure in tension or resolution of slope movement trends.
shear may be a controlling factor in other cases, especially
at lower stress levels in smaller slides and involving strong 4.1. Case Example 1
rock mass. A good example is the 1971 slide of a 50,000 m3
wedge of quartzitic argillite at Libby Dam in Montana [17]. Example 1 involves a 550 m high slope failure on the
A set of extensometer displacement records reproduced in southeast wall of the Betze-Post open pit in August 2001.
Fig. 5 shows three step-like, nearly instantaneous displace- The southeast wall is situated in Jurassic granodiorite rocks
ments on centimetre scale over a four year period, followed that are argillically (clay) altered along major gouge-filled
by sudden failure. The failure mechanism involved shearing faults and shear zones. As a result, groundwater is highly
of a highly strength-asymmetric wedge comprising smooth compartmentalized and has a significant influence on slope
and rough discontinuities, combined with tensile failure stability. A description of the structural geology and
at the head of the slide. The timing of this type of hydrogeology of the southeast wall is included in [18].
failure cannot be anticipated by means of displacement Complex wedge deformations on the upper southeast
monitoring. wall began as early as 1993. The failure mechanism
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 313

involved shearing along deep wedge intersections that Approximately 6 months after the start of the 2SE
plunged moderately towards the open pit, causing upward Layback, slope deformations began to develop in the mid
heaving along shallow in-slope dipping faults that were to upper slope and continued for another two years,
oriented obliquely to the slope face. Stability analyses were through to completion of the ultimate wall. Nine survey
carried out to define the required slope geometry to prisms were used to monitor the main complex wedge area
maintain a minimum factor of safety (FOS) of 1.2 under with locations shown in Fig. 6. Total displacements of up
partially depressurized conditions. Between 1993 and 1998, to 1 m were encountered in the upper slope over the course
slope stability was managed on the First East and Second of mining, defining an average movement rate of about
East Laybacks on the southeast wall with a combination of 2 mm day1. Slope deformations responded cyclically to
engineered waste rock buttresses, offload cuts, step-outs, mining and seasonal precipitation, but remained regressive
horizontal drain holes and vertical wells. throughout mining. No remedial changes were required to
In 1998, a re-design of the southeast wall was required the 2SE Layback mine plan and ore recovery was
due to instability that had occurred in two adjacent areas successfully completed in January 2001 to one additional
of the southeast wall. A detailed description of the re- bench below the final bottom target elevation.
design for one of those areas is presented in [18]. As part of Approximately 5 months after mining was completed,
the 2SE Layback design, 12 nested complex wedges ranging the southeast wall began to exhibit signs of progressive
in size from approximately 1–10 M m3 were analysed using failure development, as was recognized from weak accel-
a three-dimensional version of Bishop’s simplified method erations in the slope monitoring data. Inverse-velocity
[19] to satisfy a minimum FOS of 1.2 for the ultimate slope. graphs were developed and the potential slope failure time

Fig. 6. Example 1: Map of upper southeast wall showing complex wedge geometries, prism locations (solid dots), mean azimuth and plunge angles of
displacement vectors (short solid lines) and plane intersections of the interpreted multi-planar complex wedge rupture surface (dashed lines).
 ARTICLE IN PRESS
314 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

1.0 35 Predicted velocity curv es (based

30 on inverse-velocity fits ) compared
with actual velocity data

VELOCITY (cm/day)
0.9
25

0.8 20

15
INVERSE VELOCITY (days/cm)

0.7 10

5
0.6
0
0.5 -45 -40 -35 -30 -25 -20 -15 -10 -5 0
TIME BEFORE FAILURE (DAYS)

0.4

S-189
0.3 S-190
S-205
S-219
0.2 S-220
S-221
0.1 S-249 Regression coefficient (R2) =99%
S-262 for all inverse-velocity fits
S-265
0.0
45 40 35 30 25 20 15 10 5 0
TIME BEFORE FAILURE (DAYS)

18-Jul-01 25-Jul-01 1-Aug-01 8-Aug-01 15-Aug-01 22-Aug-01 29-Aug-01

DATE

Fig. 7. Example 1: Plot of 6-day average inverse-velocity and velocity (predicted curves versus actual values on inset graph) versus time for nine prisms
(time 0 was the observed time of failure). Prism numbers correspond to Fig. 6.

was initially predicted to occur approximately 2–3 months angle (crest-to-toe) was 271 over a slope height of 450 m.
later. As displacement rates increased, a clear inverse- The failure encompassed an overall slope height of 550 m.
velocity trend developed and began to converge on a failure The angle defined from the original crest to the toe was 241.
time of August 29, 2001. Fig. 7 is a plot of inverse-velocity The failure episode lasted several hours as a series of nested
versus time showing the trends of nine survey prisms complex wedge failures and rock avalanches. Fig. 6
located at various elevations on the slope (Fig. 6) over the illustrates the total vector displacements for nine survey
last 6 weeks of data prior to failure. Targets were prisms approximately one week before failure. Following
monitored using a robotic total station at 2 h intervals. failure, a maximum total displacement of about 400 m was
Data filtering was achieved by calculating six-day average estimated for the failure mass. The Fahrböschung angle,
slope distance velocities to reduce the effects of instrument defined from the crest of the back-scarp to the toe of the
error in low-level velocity values (i.e., n ¼ 72 in Eq. (A1) in landslide deposit, was estimated at 22.51. The failure
the Appendix A). Linear regression was then applied to the occurred on an ultimate pit wall approximately 8 months
entire data set of inverse-velocities. The regression coeffi- after completion of mining and had no adverse impacts on
cients (R2) were 99% for all nine prisms, indicating a the open pit mining operation.
consistent, linear trend. The data began to diverge from the
best-fit lines closer to the failure time (Fig. 7). But, since 4.2. Case Example 2
this was an optimistic change, it was decided to base the
predictions conservatively on projecting the best-fit linear Example 2 involves a failure of a 365 m high slope, with
trends through 2–4 weeks of data rather than adjusting 210 m of consolidated tertiary alluvial sediments overlying
predictions based on the shorter-term trend. weathered and altered intrusive bedrock. High ground-
An 18 M m3 (47 million ton) failure occurred on the water conditions, a highly fractured rock mass and the
southeast wall of the Betze-Post pit on August 29, 2001. occurrence of steeply in-slope dipping faults, resulted in
The failure initiated in a 345 m high section of the upper deep-seated toppling deformations that propagated into
bedrock slope. Prior to failure, the maximum overall slope the overlying sediments. Slope deformations had occurred
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 315

0.14
250
Predicted velocity curves (based
on inverse-velocity fits) compared
200 with actual velocity data

VELOCITY (cm/day)
0.12

150

0.10 100
INVERSEVELOCITY(days/cm)

50
0.08
0
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
0.06 TIME BEFORE FAILURE (DAYS)

0.04

0.02
Regression coefficient (R2) = 83 to 99%
for all inverse-velocity fits

0.00
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
TIME BEFORE FAILURE (DAYS)

Fig. 8. Example 2: Plot of inverse-velocity and velocity (predicted curves versus actual values on inset graph) versus time for five prisms and one wireline
extensometer (line) (time 0 was observed time of failure).

in a regressive fashion in response to mining over a one (IRA) of 441. The head scarp of the failure extended
year period before signs of progressive failure were approximately 50–75 m behind the pit crest, defining a
recognized. Fahrböschung angle of 351. Boulders of up to 3 m in
Fig. 8 is a plot of data for five survey prisms and one diameter rolled to a maximum distance of 130 m from the
wireline extensometer, showing linear best-fit trends original toe of the slope, but were arrested by a safety
projecting to a failure time that was forecasted from one impact berm. The total duration of the failure stage was
to two weeks prior to failure. Velocities for five prisms were several minutes.
calculated based on incremental total vector survey read-
ings. The wireline extensometer was measured at 15-min 4.3. Case Example 3
intervals via radio telemetry, but velocities were calculated
on a 24-h basis (n ¼ 96) to reduce the effect of error and Example 3 involves a 2 M m3 (5 million ton) failure that
instrument resets. The wireline extensometer data provided occurred over a 120 m slope height on the southwest wall of
the most accurate prediction of slope failure time. The R2 the Betze-Post pit on May 22, 2005. The failure occurred
linear regression coefficient for the wireline extensometer approximately two weeks after a low shear strength
data was 99%, as compared to 83–99% for the five survey lithologic contact was daylighted with an average dip of
prisms. about 171 towards the open pit. The final trend was
Mining activities were successfully stopped with ad- progressive in nature and followed 115 mm of rain in 11
vanced warning of impending slope failure. No adverse days, and a peak rainfall event of 53 mm in 24 h, 6 days
impacts were experienced by the mining operation. The prior to failure. The failure occurred as a result of the
failure occurred on an interim slope and contingency adverse orientation of low shear strength materials at the
planning provided alternate mining faces. lithologic contact and pore pressures associated with
Approximately 1 M m3 (2 million ton) of material from infiltration of up to 4000 l per minute of surface water in
the upper slope was deposited by the landslide on the tension cracks at the pit crest. As shown in Fig. 9,
working level, forming a debris lobe that extended consistent inverse-velocity trends were defined 4–5 days
approximately 100 m from the toe of the slope. Prior to before failure, based on 2-day average inverse-velocity
failure, the slope was mined at an interramp slope angle values, based on slope distance survey measurements taken
 ARTICLE IN PRESS
316 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

0.25 300
Predicted velocity curves (based
on inverse-velocity fits) compared

VELOCITY (cm/day)
with actual velocity data
200
0.20
INVERSE VELOCITY (days/cm)

100

0.15
0
4 3 2 1 0
TIME BEFORE FAILURE (DAYS)
0.10

0.05

Regression coefficient (R2) = 73 to 91%

for all inverse-velocity fits
0.00
4 3 2 1 0
TIME BEFORE FAILURE (DAYS)

18-May-05 19-May-05 20-May-05 21-May-05 22-May-05 23-May-05

DATE

Fig. 9. Example 3: Plot of inverse-velocity and velocity (predicted curves versus actual values on inset graph) versus time for three prisms (time 0 was the
observed time of failure).

at 2 h intervals (n ¼ 24). Prior to this, slope movements had ton) instability on the northeast wall of the Betze-Post open
behaved in a regressive fashion so that only a relatively pit that developed in late 2000 and continued to deform at
short interval provided a consistent trend, which was, controlled rates, through to completion of mining in
nevertheless used for a failure prediction. January 2003. During mining, slope deformations were
Fig. 10 is a photo approximately 2 months after the managed by implementing well placed offloading cuts,
failure. The failure occurred on an interim wall and had no step-outs, mid-slope waste rock buttresses and temporarily
adverse impacts on mining activities. Remediation included splitting Layback development. A detailed account of the
diversion of surface water near the pit crest and a stepout engineering geology, hydrogeology, design and develop-
at the failure toe. Preceding failure, the slope was mined at ment of the Northeast Layback is included in [20].
an IRA of 381. The Fahrböschung angle was determined to From early May to the end of June 2002, slope
be 271, and the maximum runout distance from the original movements within the Midnight/Pats complex wedge
mined toe was about 90 m. The maximum total displace- exhibited slope accelerations. Slope movements were
ment of the failure mass was estimated to be about 140 m. related to low angle shearing along the low shear strength
The failure event, observed from a distance, occurred over (i.e., f ¼ 91, c ¼ 35 kPa) Carlin waxy silt unit. Fig. 11 is a
a period of about 1 min. graph of inverse-velocity versus time for six prisms located
on the upper slope. Again, a consistent linear inverse-
4.4. Case Example 4 velocity trend was identified, with an indicated potential
failure time of early to mid August 2002.
A fourth case example is presented that illustrates the use Based on this observation, a decision was made to
of inverse-velocity to plan and implement remedial stabilize the slide by unloading the active part of the
measures to mitigate possible slope failure, using selected complex wedge. Stability analyses carried out using the
threshold movement rates based on forecasted failure three-dimensional Bishop’s simplified method (Fig. 12)
times. This example involves a 3–10 M m3 (5–18 million were used to determine the volume of material required to
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 317

stabilize the wedge. It was estimated that up to 360,000 ton Carlin waxy silt unit to stabilize the slope. It was estimated
of material would have to be excavated at the pit crest, and that up to one month could be required to implement the
approximately 90,000 ton placed at the daylight level of the remedial measures, without causing significant disruption
to the mine plan. Threshold movement rates were selected
for remediation activities such that construction and
operation would be completed with a two-week buffer
period prior to failure, if it were to occur.
Approximately 3 days after construction started, a
significant trend change could be discerned in the inverse-
velocity plot of the monitoring data, as the slope began
decelerating. A threshold movement rate of about
2.5 cm day1, or an inverse-velocity of about 0.4 days/cm,
was used to schedule construction and excavation. As seen
on Fig. 11, remediation continued over a period of about
three weeks until slope displacement rates stabilized to
acceptable levels. Periodically throughout the remainder of
mining, additional crest offloading was required to main-
tain velocities below threshold values. Fig. 13 is a photo
showing the upper portion of the Midnight/Pats complex
wedge (see Fig. 12) as crest offload mining was taking
place. The Northeast Layback was successfully completed
in January 2003. Failure of the upper northeast wall was
Fig. 10. Example 3: Photo of 2 M m3 Southwest wall failure at the Betze- mitigated by implementing the remedial measures dis-
Post open pit that occurred May 22, 2005 (case Example 3). cussed above.

Stabilization of Slope Displacements

1.5
INVERSEVELOCITY(days/cm)

Buttress and Offload

Remedial Measures
1

0.5

Regression Coefficient( R2) = 96 to 99%

for all Inverse-Velocity Fits.
0
105 100 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 5 0
TIME BEFORE FAILURE (DAYS)

1-May-02 16-May-02 31-May-02 15-Jun-02 30-Jun-02 16-Jul-02 31-Jul-02 15-Aug-02

DATE

Fig. 11. Example 4: plot of inverse-velocity versus time for six prisms on the upper northeast wall. Note the trend change in inverse-velocity in response to
slope remediation.
 ARTICLE IN PRESS
318 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

Midnight Fault Pats Fault

Carlin waxy silt unit

Scale
100m

Fig. 12. Example 4: Complex wedge on the upper northeast wall of the Betze-Post open pit involving the Midnight/Pats faults and Carlin waxy silt
(3 M m3).

observation that the inverse-velocity plot often approaches

linearity, especially during the final stages before failure,
enhances the utility of the technique. In principle, the same
approach can be used for extrapolating any unstable
phenomena, for example the frequency of micro-seismic
events recorded during the onset of a rock slope failure.
The following general rules are suggested for use of the
method.

(a) The method must not be applied in isolation, without

being accompanied by qualitative observations of slope
behaviour, collection of data and on-going analysis of
the structure of the slope, rock mass condition, stress
and groundwater regime. Displacement monitoring is
only one component of a complex process that
Fig. 13. Example 4: Photo of the upper northeast wall of the Betze-Post
open pit showing the location of crest offloading in the Midnight/Pats comprises slope stability management.
complex wedge. (b) The method cannot be used for rock slides dominated
by brittle failure, although the identification of such
cases is at present a matter of judgment. Particular care
5. Conclusion should be exercised when dealing with relatively small
failures in strong rock.
Can the time of rock slope failure be predicted from (c) The monitoring data must be processed to remove the
displacement monitoring results? Obviously, in view of the effects if instrument error (see Appendix A) and
mix of review and original data presented in this paper, we eliminate records distorted by local movements.
cannot answer an unequivocal ‘‘yes’’. However, Fukuzo- (d) Failure forecasting relies on the identification of
no’s inverse-velocity method is a powerful tool that consistent trends. The possibility of trend changes,
significantly improves our ability to interpret monitoring driven by observable or unknown factors, must always
data and estimate the timing of the failure process. The be kept in mind. Monitoring must be continued as long
primary advantage of the method is that inverse-velocity- as possible prior to failure. The results must be
time plots, whether linear or not, are much easier to constantly re-evaluated and any established best-fit
extrapolate towards the failure limit than the usual functions must be revised in view of the latest data. If a
hyperbolic curves recorded by accelerating deformations stable trend exists, as shown in the four original
prior to failure, trending to a vertical asymptote. The examples presented in Section 4, predictions can be
 ARTICLE IN PRESS
N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320 319

given and their reliability can be confirmed by contributed to the sound engineering judgment that was
continuing to evaluate the most recent data until the used in the successful application of this approach.
time of failure. If trend changes occur, the predictions Particular thanks are given to Bob Sharon, Mark Rantapaa,
must be promptly revised. It must be accepted that false Tracey Miller, John Cash, Dave Pierce, Joe Gallegos and
alarms may occasionally be given. ‘‘Optimistic’’ trend Jorge Armstrong. Critique by an anonymous reviewer has
changes can sometimes be deliberately ignored as led to an improvement of the manuscript.
shown in Example 1 of Section 4. ‘‘Pessimistic’’ trend
changes must be evaluated and acted upon promptly. Appendix A
(e) The treatment of cyclic changes depends on the
magnitude of their amplitude, relative to the distance Part 1: Two alternative data filtering methods
from the horizontal axis on the inverse-velocity plot. If
the ratio between these two quantities is high, it may be Displacement monitoring results are collected as a time
necessary to assume that the low point of any given series, t1yti, d1ydi, where ti and di are the most recent
cycle may produce sudden rupture. time and displacement, respectively. The simplest filtering
(f) Data fitting using non-linear inverse-velocity trend method consists of using only every nth observation to
lines may provide a more accurate assessment of some calculate rate. Thus, the current displacement rate is
longer-term trends, but is more complex, which may
d i  d in
limit practical use. The authors recommend the use of vi ¼ . (A.1)
linear fits, updated on an ongoing basis to identify ti  tin
trend curvature or to signal the onset of trend changes. The second method calculates velocity as the slope of a
Since trend curvature may result from a particular linear regression line, plotted through the last n observa-
manner of change of unknown driving factors, it does tion points:
not seem advisable to look for a uniquely shaped std
mathematical function. Linear fits facilitate easy vi ¼ , (A.2)
s1
prediction of anticipated displacements, which can
then be used for rapid confirmation of a consistent where
2 !2 3
trend (or otherwise). The prediction equations are
1 4X i
1 X
i
derived in the second part of the Appendix A. s1 ¼ t2j  tj 5 (A.3)
n  1 j¼in n j¼in
Possibly for the first time in the published literature,
and
successful a priori forecasts of large mine rock slope " ! !#
failures using the inverse-velocity method have been 1 Xi
1 X
i X
i
documented. As a tool helping interpretation of displace- std ¼ tj d j  tj dj . (A.4)
n  1 j¼in n j¼in j¼in
ment monitoring results, the Fukuzono method appears to
be indispensable for rock slide cases of the kind described
in Section 4 and very helpful for many other cases.
Part 2: Anticipated displacement data, based on a linear
6. Experience database trend

In an ongoing effort to increase the database of Linear extrapolation has the advantage of being easy to
experience with the inverse-velocity approach, the authors execute and to modify when deviations occur. Once linear
of this paper would like to extend an invitation for others trends are identified in slope movement data and fit with a
to share their experience using this method. If necessary, line with a slope A and intercept v0, predicted failure time
confidentiality can be maintained by omitting reference to can be calculated as follows:
the location or source of the data. Correspondence to this
effect can be made via electronic mail to Mr. Nick Rose at 1
tf ¼ þ t0 . (A.5)
nrose@piteau.com or Dr. Oldrich Hungr at ohungr@ Av0
eos.ubc.ca. Predicted velocities can be plotted versus time by taking
the inverse of Eq. (A.5), such that:
Acknowledgements  1
1
vPredicted ¼  Aðt  t0 Þ . (A.6)
The authors of this paper are grateful to the management v0
of Barrick Goldstrike Mines Ltd., Barrick Gold Corp. and By integrating the above equation, predicted relative
the anonymous mining company that provided permission displacements can be calculated as follows:
to present the slope monitoring data and information     
included in this paper. Recognition is given to the staff at 1 1
d Predicted ¼ ln  ln  Aðt  t0 Þ . (A.7)
both mines that collected the slope monitoring data and v0 v0
 ARTICLE IN PRESS
320 N.D. Rose, O. Hungr / International Journal of Rock Mechanics & Mining Sciences 44 (2007) 308–320

Predicted velocity and relative displacement curves can [11] Crosta GB, Agliardi F. Failure forecast for large rock slides
be compared to the actual slope monitoring data to by surface displacement measurements. Can Geotech J 2003;40:
176–91.
determine whether the predicted trend is stable or whether
[12] Vibert C, Arnould M, Cojean R, Le Cleach M. Essai de prevision de
changes in slope behaviour may be occurring. rupture d’un versant montaineaux. In: Bonnard C, editor. Proceed-
ings of fifth international symposium on landslides, Lausanne, vol. 1.
References Rotterdam: A.A. Balkema; 1988. p. 789–92.
[13] CETE: Centre d’Etudes Techniques de l’Equipment, Centre de Nice.
[1] Martin DC. Time dependent deformation of rock slopes. PhD thesis, Glissement de La Clapiere, Hydrologie et Cinematique, Donnes
University of London, August 1993. Sommaires, Unpublished report, 1990.
[2] Zavodni ZM. Time-dependent movements of open-pit slopes. In: [14] Muller L. The rock slide in the Vajont valley. Felsmech Ingenieurgeol
Hustrulid A, Carter MK, Van Zyl DJA, editors. Slope stability in 1964;2:148–212.
surface mining. Littleton, Colorado: Society for Mining, Metallurgy, [15] Hendron AJ, Patton FD. The Vaiont Slide, a geotechnical analysis
and Exploration, Inc.; 2000. p. 81–8. based on new geologic observations of the failure surface. US Army
[3] Zavodni ZM, Broadbent CD. Slope failure kinematics. Canadian Corps. of Engineering,Technical report 85, 1985. 104pp.
Institute of Mining, Metal Petroleum (CIM) Bull 1980;73(16):69–74. [16] Voight B. Materials science law applies to time forecasts of slope
[4] Dunnicliff CJ. Geotechnical instrumentation for monitoring field failure, vol. 3. Landslide news, Tokyo: Japan Landslide Society; 1989.
performance. New York: Wiley; 1988 577pp. p. 8–10.
[5] Turner AK, Schuster RL, editors. Landslides, investigation and [17] Voight B. Wedge rockslides, Libby Dam and Lake Kookanusa,
mitigation. transportation research board special report 247. Montana. In: Voight B, editor. Rock slides and avalanches.
Washington, DC.: National Research Council; 1996 673pp. Developments in geotechnical engineering, 14b. Amsterdam: Elsevier;
[6] Wyllie DC, Mah CW. Rock slope engineering, 4th ed. New York: 1979. p. 281–315.
Spon Press; 2004 431pp. [18] Rose ND, Sharon RP. Practical rock slope engineering designs at
[7] Fukuzono T. A new method for predicting the failure time of a slope. Barrick Goldstrike. In: Hustrulid A, Carter MK, Van Zyl DJA,
In: Proceedings of the fourth international conference and field editors. Slope stability in surface mining. Littleton, Colorado: Society
workshop on landslides. Tokyo: Japan Landslide Society; 1985. for Mining, Metallurgy, and Exploration, Inc.; 2000. p. 213–8.
p. 145–50. [19] Hungr O. Geotechnical Research Inc. CLARA-W – slope stability
[8] Voight B. A method for prediction of volcanic eruption. Nature analysis in two or three dimensions for IBM compatible micro-
1988;332:125–30. computers. User’s manual. Vancouver, BC; 1994. 77pp.
[9] Voight B. A relation to describe rate dependent material failure. [20] Sharon, R., Rose, N. Rantapaa, M. Design and development of the
Science 1989;243:200–3. Northeast Layback of the Betze-Post open pit. In: Proceedings, SME
[10] Kilburn CRJ, Petley DN. Forecasting giant, catastrophic slope collapse: annual meeting, Salt Lake City, Utah, Society for Mining,
lessons from Vajont, Northern Italy. Geomorphology 2003;54:21–32. Metallurgy, and Exploration, Inc., 2005., Pre-print 05-09, 10pp.