ARTICLE IN PRESS

Geomorphology 1321 (2002) 1 – 18

www.elsevier.com/locate/geomorph

Landslide inventory in a rugged forested watershed:

a comparison between air-photo and field survey data
Francesco Brardinoni *, Olav Slaymaker 1, Marwan A. Hassan 2
Department of Geography, University of British Columbia, Vancouver, BC, Canada V6T 1Z2

Received 1 August 2002; received in revised form 17 November 2002; accepted 29 November 2002

Abstract

Landslide inventories are routinely compiled by means of aerial photo interpretation (API). When examining photo pairs, the
forest canopy (notably in old-growth forest) hides a population of ‘‘not visible’’ landslides. In the present study, we attempt to
estimate how important is the contribution of landslides not detectable from aerial photographs to the global mass of sediment
production from mass failures on forested terrain of the Capilano basin, coastal British Columbia. API was coupled with
intensive fieldwork for identification and measurement of all landslides. A 30-year framework was adopted. We show that ‘‘not
visible’’ landslides can represent up to 85% of the total number of failures and account for 30% of the volume of debris
mobilised. Such percentages display high sub-basin variability with rates of sediment production varying by one order of
magnitude between two sub-basins of the study area. This is explained qualitatively by GIS-based analysis of slope frequency
distributions, drainage density, and spatial distribution of surficial materials. Such observations find further support in the
definitions of transport-limited and supply-limited basins. As a practical consideration to land managers, we envisage that
supplementary fieldwork for landslide identification is mandatory in transport-limited systems only. Fieldwork has
demonstrated that gully-related failures have a greater importance than one could expect from API.
D 2002 Elsevier Science B.V. All rights reserved.

Keywords: Landslide identification; Air-photo interpretation; Interbasin variability; Forested terrain; Coastal British Columbia

1. Introduction to evaluating landslide hazard for land use planning

and development (Wieczorek, 1984), is routinely
Despite of the great advances in remote sensing performed by means of aerial photo interpretation
technology, compilation of landslide inventories in (API) and limited fieldwork. When examining photo
rugged forested terrain, the first fundamental phase pairs, the forest canopy (notably in old-growth forest)
hides a population of ‘‘not visible’’ landslides (Fig. 1).
Reid and Dunne (1996) assert that most landslides are
* Corresponding author. Tel.: +1-604-822-2269; fax: +1-604- readily identified on aerial photographs. They
822-6150. acknowledge the existence of a minimum landslide
E-mail addresses: [email protected] (F. Brardinoni),
size visible on photographs that must be determined
[email protected] (O. Slaymaker), [email protected]
(M.A. Hassan). with the aid of field measurements. The advice given
1
Tel.: +1-604-822-3246; fax: +1-604-822-6150. is that the frequency of occurrence of smaller and ‘‘not
2
Tel.: +1-604-822-5894; fax: +1-604-822-6150. visible’’ landslides must be estimated. The problem

0169-555X/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
doi: 10.1016/S0169-555X(02)00355-0
 ARTICLE IN PRESS
2 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

many slides were missed in the forested areas. A good

reference datum against which to evaluate manage-
ment impact was simply not available (F.J. Swanson,
United States Department of Agriculture, Pacific
Northwest Research Station, Oregon, personal com-
munication, 2001). As a consequence, the Oregon
Department of Forestry initiated a large project (Robi-
son et al., 1999, to date the only study) to systemati-
cally compare ground-based landslide inventories
with air-photo-based inventories. About half of the
landslides in recent plantations were visible in photos
at the 1:6000 scale. In mature and old-growth forest,
the air photos revealed 5% (at most) of the ground-
surveyed slides. Such percentages should not surprise
the reader, as landslide size in the Oregon Coast
Range is typically smaller than any of the air-photo
thresholds reported in the literature (e.g., O’Loughlin,
1972; Rood, 1984; Schwab, 1988; Rollerson et al.,
2001). However, there are two limitations to Robison
et al. (1999) study: (i) they counted and measured
only those failures that impacted stream channels; and
(ii) the study focused on areas considered to have
Fig. 1. Sketch diagram showing canopy displacement and visible experienced the most severe impacts from the 1996
portion of a landslide as seen from an aerial photograph (modified
after Pyles and Froehlich, 1987).
storms.
The main purpose of this study is to critically
evaluate the use of air photographs in the study of
discussed here is that there are no clear guidelines for landslide incidence in both clearcut and forested
the estimation procedure required. terrain. Our goal is to determine how large a propor-
Most of the published literature addresses the tion of failures is missed, both in terms of number of
question of landslide identification in empirical ways events and volume of mobilised debris, at a given air-
(e.g., O’Loughlin, 1972; Rood, 1984; Schwab, 1988; photo scale. Specifically, we aim to see whether the
Rollerson et al., 2001). In fact, under the assumption proportion of ‘‘missed’’ events is constant throughout
that the incidence of ‘not visible’ landslides is negli- a single physiographic region (the Pacific Ranges of
gible to the whole picture of the inventory, such the Coast Mountains) so that our findings can be
studies exclude events below a certain size (threshold) generalised to the regional scale. Lastly, we will test
that is considered still to be assuredly photo-detect- whether identification of ‘‘not visible’’ events can
able in recent harvested cut-blocks and in primary affect (i) estimated rates of sediment delivery to
forest. Interestingly enough, in British Columbia, this streams; and (ii) conclusions about the impact of
threshold appears to have increased over time. logging on slope stability.
O’Loughlin (1972) considered debris slides and ava-
lanches larger than 93 m2; Rood (1984), events larger
than 200 m2; Schwab (1988), larger than 400 m2 and, 2. Study area
Rollerson et al. (2001), larger than 500 m2.
The importance of the question of ‘‘not visible’’ The study was conducted in the Capilano River
landslides became apparent after the 1996 floods in basin, a 198-km2 watershed located in West Vancou-
Oregon. Some environmental groups did a flyover ver, British Columbia. The site was selected for the
inspection and reported high slide frequency in cut following reasons: (i) forest harvesting has been
and roaded areas, while others were concerned that performed with well-defined and documented techni-
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 3

ques over the last 50 years; (ii) it is a water supply

area for Greater Vancouver, hence a wide range of
thematic maps and detailed air-photo coverage are
available; (iii) geology is relatively homogeneous; and
(iv) it is the most easily accessible forested watershed
that meets our experimental requirements.
The watershed (Fig. 2) lies within the Pacific
Ranges of the Coast Mountains (Holland, 1976).
The most striking feature of the Capilano valley is
rugged topography, with slopes typically steeper than
35j and steepness generally increasing with elevation.
The rugged landscape is the result of the combined
effects of tectonic uplift, rock strength, and glacial
erosion. Bedrock consists primarily of intrusive igne-
ous rocks—granodiorite, quartz diorite, diorite, and
lesser amounts of gabbro and migmatite. Limited
metamorphic-dominated formations are present in
places (i.e., Gambier Group and Twin Islands Group;
Roddick, 1965). Gentle and moderate slopes, espe-
cially at mid to low elevations, are mantled by glacial
till, which is the most extensive surficial material and
constitutes the primary source of fine sediments.
Surficial materials deposited on hillslopes in post-
glacial time consist primarily of colluvium.
The climate in the Pacific Ranges of the Coast
Mountains consists of two main seasons. A wetter
one, from October through May, is characterised by an
almost uninterrupted series of large-scale oceanic
storms that encroach upon the coast. A much drier
regime characterises June through September, when
only intense small-scale oceanic storms disrupt the
persistent high-pressure domain. Accordingly, in the
Capilano watershed (at the reservoir outlet), about
80% of the annual precipitation falls during the wet
season. Minimum monthly totals occur in July and
August, accounting for about 4% of the normal annual
precipitation.
Convergence in valleys and topographic uplift tend
to increase precipitation abruptly north of the Lower
Mainland, where fronts encounter the Pacific Ranges.
Annual precipitation ranges from 2000 mm at Cleve-
land Dam to 5000 mm at Mt. Hollyburn (1325 m) and
Mt. Strachan (1454 m; Fig. 2). Elevation also affects
the snowfall proportion of total precipitation. Meas-
urements from snow courses show that at elevations
of around 1000 m the normal maximum water equiv-
alent in the snow pack is at least 1600 mm (Greater Fig. 2. Map of the Capilano watershed indicating the location of the
Vancouver Regional District, 1999). ground-surveyed areas.
 ARTICLE IN PRESS
4 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

3. Methods watershed, and an indefinite number of landslide scars

have been rehabilitated; and (iv) only those mass
3.1. Field and air-photo surveys wasting events that occurred during the last 30 years
were considered (Fig. 4). The temporal framework was
The study entailed the compilation of a landslide determined on the basis of earlier studies (Smith et al.,
inventory through API. The scale of aerial photogra- 1983, 1986; Rood, 1984; Reid and Dunne, 1996) and
phy available ranges from 1:12,000 to 1:15,000. of the uncertainty involved in the identification of old,
Photo sets date from 1968, 1976, 1984, 1992, and small, slope failures (e.g., Wieczorek, 1984).
1996 (plus helicopter flyovers in July 2000). Within Minimum ages of landslides were estimated in the
the watershed, intensive ground checking was con- field by means of dendro-chronology, branch level
ducted successively on Sisters and East Cap Creeks counts on saplings, and presence of 1- to 2-year
(Fig. 2). The two subwatersheds were selected for litterfall on the scar; and on aerial photos through
their topographic, land use, and logistic attributes: stereoscopic inspection of sequential photo sets. Photo
Sisters Creek is the closest location presenting areas interpretation was conducted with a SOKKIA MS27
of old extensive logging (Fig. 3A) on steep terrain stereoscope (3
and 8
magnification) and was
(class IV and V polygons; cf. Table 1), while East Cap supervised by J.M. Ryder and Associates. Each mass
Creek is the most easily accessible site with recent movement event, when possible, was partitioned into
cut-blocks (Fig. 3B) on similarly steep slopes. These initiation, transportation, and deposition zone. The
basins present the opportunity to conduct meaningful technique adopted to evaluate the mobilised volume
comparisons among undisturbed forest, recent ‘‘care- associated with each failure was that described by
ful’’ clearcut, and old extensive logging. Land use Rood (1984).
covers in the surveyed areas of the two creeks are Episodic and recurrent mass failures were differ-
reported in Table 2. entiated and then classified according to whether they
Detailed fieldwork was conducted with the pur- were active or dormant (Zaruba and Mencl, 1969;
pose of identifying and measuring every distinguish- Erskine, 1973; Wieczorek, 1984; Yanai and Usui,
able slope failure within each traversed terrain 1987). This distinction is relative to a specific time
polygon. Following British Columbia Terrain Stabil- frame (cf. Flageollet, 1996, for an extended discussion
ity Classification guidelines of the British Columbia on the time dimension in the study of landslides). In
Forest Practices Code (British Columbia Ministry of our case, dormant (inactive or fossil) mass movements
Forests, 1995, cf. Table 1), J.M. Ryder and Asso- were taken as those that, although still visible on air
ciates compiled the terrain stability and surficial photos, had occurred more than 30 years ago. These
material mapping of the watershed. The classification last were not taken into consideration. Especially at
was conducted mainly by API, with only 25% of places of recurring landslide occurrence (i.e., gully-
terrain polygons ground-checked at the time of map channels and -sidewall), scars were not always the
compilation. In every study polygon, we walked result of a one-off event, indeed sometimes they
along the entire gully network and covered the whole resulted from the occurrence of overlapping, repeated
area of the polygon. We can assert with confidence events, or from remobilisation of antecedent failure
that we detected all existing landslides within each deposits. In the former case, when recognisable, only
polygon. the scar of the most recent event was measured, and
The inventories were carried out with the following events associated to debris remobilisation were not
operational specifications: (i) by examining ground considered. Following these criteria, a quantitative
classified in British Columbia stability classes III, IV, database was obtained from the integration of field-
and V (Table 1), terrain polygons were selected in such work and API (Fig. 4).
a way that every slope aspect was proportionally
represented in each sub-basin; (ii) all traversed, ‘‘not 3.2. Statistical analysis
visible’’ events were systematically measured; (iii)
road-related landslides were not included because The land base was subdivided into polygons ac-
there has been continuous road maintenance in the cording to British Columbia terrain classification
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 5

Fig. 3. Aerial photographs taken in 1996 showing part of (A) Sisters Creek and (B) East Cap Creek basins. Note the higher degree of landscape
dissection in Sisters Creek when compared to East Cap Creek. Examples of types of forest cover are marked with capital letters: A = recent cut-
blocks, B = old-growth forest, C = old extensive clearcut. Examples of visible mass movement scars are marked with numbers: 1 = debris slides
in recent cut-blocks, 2 = debris flows in old-growth forest reaching Strachan Creek, 3 = debris slide on old clearcut crossing a logging road and
reaching Sisters Creek.
 ARTICLE IN PRESS
6 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

Table 1
British Columbia Terrain Stability Classificationa
Terrain stability class Subjective rating system Interpretation
I Flood plains and level to undulating coastal plain areas No significant stability problems exist
Most terrain with slopes < 20%
II Most gently sloping (20 – 40%), poorly to well drained, Very low likelihood of slides following
lower slope landforms logging or road construction
Moderately sloping (40 – 60%), well to rapidly drained Minor slumping is expected along road cuts,
surficial deposits especially for 1 or 2 years following
construction
III Moderately sloping (40 – 60%), imperfectly to poorly Minor stability problems can develop
drained surficial deposits that are not marine or Timber harvesting should not significantly
lacustrine reduce terrain stability; there is a low
Level to gently sloping (0 – 40%), imperfectly likelihood of landslide initiation following
to poorly drained, deep marine clays and lacustrine logging
deposits Minor slumping is expected along road cuts
Moderately sloping, deeply gullied surficial Low likelihood of landslide initiation
deposits that are not of lacustrine or marine origin following road building
IV Steeply sloping (>60%), well drained, deeply Expected to contain areas with a moderate
gullied surficial deposits likelihood of slide initiation following
Steeply sloping, poorly drained surficial deposits logging or road construction
Moderately sloping, deeply gullied, or imperfectly
to poorly drained lacustrine or marine deposits
V Any areas where natural landslide scars are visible Expected to contain areas with high
on air photographs or in the field likelihood of slide initiation following
Very steeply sloping (>70%), imperfectly to logging or road construction
poorly drained, deeply gullied surficial deposits
a
Modified from British Columbia Ministry of Forests, 1995.

guidelines (Table 1). A polygon was considered as an were minimised by comparing polygons of equal land
experimental unit, and each polygon that received the use and stability (according to the British Columbia
same classification (‘‘treatment,’’ i.e., survey type, Terrain Stability Classification).
land use, stability class) was regarded as a replication. A two-step statistical analysis was performed.
Two different types of survey to calculate landslide First, one-way ANOVA-like tests were conducted to
density and the associated amount of mobilised vol- evaluate how the survey method affected landslide
ume of debris were conducted: (i) exclusively by API; frequency per unit area (LS/ha) and amount of mobi-
and (ii) by API coupled with intensive field survey. lised debris (m3/ha). Second, the experimental design
Control on experimental error was achieved by select- was made more complex (two-way ANOVA-like
ing study sites with homogeneous climate, geological tests) so that the location effect (Sisters vs. East
setting, and vegetation cover. Confounding factors Cap) on survey capability could be incorporated.
We took advantage of replications. The greater the
number of replications, the larger the number of
Table 2
Land use areas in East Cap and Sisters Creeks
degrees of freedom in the error term and the smaller
the experimental error, which represents the unex-
Land use East Cap Creek Sisters Creek
(ha) (ha) plained variability of the dependent variable. Two-
way layouts possess a greater explanatory power than
Old growth (undisturbed) 409.3 (61.3%) 271 (44.1%)
Old growth (including fire) 578.8 (86.6%) 345.1 (56.1%) one-way analyses as they incorporate both factors
Old harvesting 0 (0) 261.3 (42.5%) (location and survey type) over the entire database.
Recent harvesting 89.2 (13.4%) 8.6 (1.4%) In this way, treatment interactions can be evaluated,
Fire 169.5 (25.3%) 74.1 (12%) and unexplained variability (experimental error) is
Total surveyed 668 615
minimised.
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 7

ation are the factors that determined which macro-

scopic variables could be extracted from the DEM.
These were elevation, slope gradient, slope aspect,
drainage density, and spatial partitioning of surficial
materials across terrain stability classes. Elevation
affects the amount of precipitation, rate of physical
weathering (e.g., freeze –thaw activity and magnitude
and persistence of snow cover), and the hydrologic
regime of a basin. Slope gradient is one of the most
important geomorphic factors for shallow mass move-
ment processes (Sidle et al., 1985). Slope gradient (b)
appears in the Mohr – Coulomb equation (Eq. (1),
infinite slope analysis’ case) and directly affects both
normal (r) and tangential (s) components of the shear
stress:
S ¼ C þ DC þ ðr  lÞtan/V ð1Þ
Fig. 4. The 30-year window framework adopted to obtain the where l is the pore water pressure, C is the effective
quantitative database of landslides. soil cohesion, DC is the cohesion caused by root
systems, and /V is the angle of shearing resistance.
However, the assumptions of ANOVA were not all Slope aspect influences terrain exposure to storm
met. The assumptions of normality of distributions and fronts. It also affects fluctuations of pore water pres-
homogeneity of variances are not satisfied. For this sure (l) and alternation of weathering environment
reason, two one-way nonparametric methods (Krus- (oxidising/reducing) brought on by wet/dry and/or
kal – Wallis and Median tests) were added (Table 3). freeze/thaw cycles. Increase in pore water pressure
The Median test is regarded as the most appropriate of alters slope stability by reducing the effective normal
the three options presented as it is particularly useful stress and thus soil shear strength. Comparisons of
where the scale contains artificial limits and many elevation, slope gradient, and aspect between the two
cases fall at either extreme of the scale. In this study, creeks were made by inspecting both the average
more than 50% of observations (polygons) have a values and the DEM cell frequency distributions.
value of zero (no landslides). Drainage density (Dd) describes the degree of
topographic dissection of the landscape. It is defined
3.3. GIS-based analysis by the ratio between the total length of the channel
network and its drainage area. Specifically, in moun-
GIS technology has been intensively used in tainous areas of coastal British Columbia, a greater Dd
landslide studies as a tool for modelling and map- corresponds with a higher number of zero- and first-
ping landslide hazard (e.g., Terlien et al., 1995; order streams (gullies), ephemeral streams that natu-
Carrara et al., 1999; Donati and Turrini, 2002). Here, rally experience recurrent failures (i.e., debris flows/
GIS-based analysis was explored as a way of resolv- torrents) on a decennial time scale. Therefore, higher
ing the unexpected contrasts that East Cap and Dd means higher probability of mass failure occur-
Sisters Creeks exhibited in terms of landslide density rence. The scale of TRIM data (1:20,000) does not
and denudation rate (Table 4). A 25-m gridded capture the entire stream network. No study has yet
Digital Elevation Model (DEM) was derived from investigated the relationship between measured (on
the 1:20,000 Terrain Resource Information Manage- TRIM maps) and actual Dd. The former certainly
ment (TRIM) digital maps of the Province of British underestimates the actual Dd, but the associated error
Columbia. is assumed to be the same for both sub-basins.
The nature and quality of information available, Mapped Dd was used as a proxy of the actual Dd of
ease of GIS-extraction, and power of physical explan- the field-surveyed landscape.
 ARTICLE IN PRESS
8 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

Table 3
Analysis of variance and nonparametric tests of landslide density and volume
(a) One-way ANOVA-like of the entire database (survey)
Factor Dependent variable One-way ANOVA Kruskal – Wallis ANOVA Median test
Significance
Survey Volume (m3/ha) ns 0.0003 0.0001
Number (LS/ha) 0.002 0.0000 0.0000

(b) Two-way ANOVA-like of the entire database (survey – location)

Factor Dependent variable Full factorial experiment Factorial experiment in Friedman test
a fixed block design
Significance
3
Survey Volume (m /ha) ns ns 0.000
Location 0.002 0.002 n/a
Survey
location ns n/a
Survey Number (LS/ha) 0.003 0.001 n/a
Location 0.001 0.001 n/a
Survey
location 0.011 n/a

(c) One-way ANOVA-like: Survey within locations

Factor Location Dependent variable One-way ANOVA Kruskal – Wallis Median test
ANOVA
Significance
Survey Sisters m3/ha ns 0.0003 0.0001
LS/ha 0.002 0.0000 0.0001
East Cap m3/ha ns ns ns
LS/ha ns ns ns

The study of the spatial distribution of surficial englacial or supraglacial conditions. It is therefore still
materials across slope stability classes focused partic- adjusting to a subaerial, nonglacial environment. Con-
ularly on glacial deposits (or till) and post-glacial versely, colluvium deposits, products of subaerial
deposits (or colluvium). The former is considered to mass movement, are arranged in more stable config-
be less stable, having been deposited under subglacial, urations. Typically, in each terrain polygon, the nature

Table 4
Sediment yields from landsliding and landslide densities during an f 30-year window (1968 – 2000) as obtained from both air-photo
interpretation coupled and not coupled with detailed field survey
Field Number of LS Mobilised volume (m3) LS density Annual LS Denudationa Yieldb
surveyed APIc Field A&Fd API Field A&F (#LS/ha) density (m3/ha) (m3/ha/year)
area (ha) API A&F (#LS/ha/year) API A&F API A&F
API A&F
East Cap 668.4 14 20 34 7869 412 8281 0.021 0.051 0.0007 0.0017 11.7 12.4 .39 .41
Creek (41.2%) (58.8%) (95%) (5%)
Sisters 615 20 117 137 52,480 21,999 74,479 0.033 0.223 0.0011 0.0074 85.3 121.1 2.84 4.04
Creek (14.6%) (85.4%) (70.5%) (29.5%)
Sum 1283.4 34 137 171 60,349 22,411 82,760 0.027 0.133 0.0009 0.0044 46.9 64.5 1.56 2.15
a
Denudation = mobilised volume/surveyed area.
b
Yield=(mobilised volume/surveyed area)/time window.
c
API = Air-photo interpretation.
d
A&F = Air-photo coupled with fieldwork.
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 9

and ratios of relative abundance of surficial materials

are reported (British Columbia Ministry of Environ-
ment, 2002). The analysis aimed to transform that
ratio-like information into a quantitative format. To
achieve this, the area of each polygon was subdivided
according to the relative abundance indicated on the
polygon label. After computing the areas covered by
each material category (i.e., M, C, R, /CM) in every
polygon, these areas were summed according to their
stability class label (i.e., III, IV, and V) and the
relevant histograms were plotted.

Fig. 6. Cumulative volume distributions of landslides in the ground-

surveyed areas obtained by API and field-coupled API.
4. Results

4.1. Landslide visibility and implications landslide scar in recent openings is about 150 m2
(Fig. 5A), but note the gap between the largest ‘‘not
Fig. 5A and B reports areas of visible (from air visible’’ scar (150 m2) and the smallest visible scar
photo) and ‘‘not visible’’ landslides that were meas- (275 m2). This is due to the small number of obser-
ured during field survey in, respectively, recent cut- vations (n = 15), and the critical area for visibility lies
blocks (logged < 15 years ago) and forests older than somewhere between these two values.
50 years. The maximum area of a ‘‘not visible’’ In forests older than 50 years, the largest landslide
scar not detected during API has an area of about 650
m2 (Fig. 5B). One landslide smaller than this (350 m2)
was identified on an air photo; this exemplifies how
the issue of landslide visibility involves various fac-
tors, such as topographic location and landscape
dissection, other than land use. As a first practical
application, then, if one wishes to perform an
unbiased (although incomplete) air-photo-based com-
parison in this area between managed and wild forest,
one should consider only failures that are> 650 m2.
Inevitably, landslides missed during API produce a
supplementary amount of sediment. The plotting of
the cumulative percent volumes against landslide
volume (Fig. 6) shows that additional amounts of
debris inventoried by means of fieldwork in East
Cap and Sisters Creeks range from 10% to 18%
among volume classes. This is clear evidence that
‘‘not-visible’’ failures do contribute significant vol-
umes of debris to the amount of mobilised material
already detected on air photos.
An important supplementary question is whether
missed events contributed significantly to the sedi-
ment loading of montane stream channels. This aspect
Fig. 5. Landslide visibility in relation to landslide area (initiation
has clear management implications in terms of water
plus transportation zone) in (A) recently harvested cut-blocks and quality (e.g., in the Capilano Reservoir) and fish
(B) forests older than 50 years. habitat; and it is also important for estimating the
 ARTICLE IN PRESS
10 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

error associated with remotely sensed sediment budg- and from an applied perspective, it shows that the rate
ets. Landslides have been classified as stream-con- of sediment in-filling of the Capilano Reservoir has
nected, gully-connected, and not connected, depend- been underestimated.
ing on whether debris was delivered to a permanent The combined effect of fieldwork and air-photo-
stream, to a gully, or remained in an unchannelled inventory was to increase the percentage of gully-
portion of the landscape. According to the field data, connected failures and the associated mobilised vol-
about 58% of the mobilised volume (Fig. 7B) enters ume by about four times (from 8.8% to 32.2% and
the channel network directly; and 35% is delivered to from 3.2% to 11.8%, Fig. 7A and B). The proportion
gullies and most probably will be evacuated to the of stream-connected failures decreased by nearly one-
stream network via debris torrent in a decennial time third, and the mobilised volume decreased only 5.5%.
scale (from 10 to 40 years, cf. Thurber Engineering, Percentage of unconnected slides showed minor
1996). However, API indicated that only 3.2% of the changes. Integrating field with air-photo measure-
mobilised volumes were delivered to gullies, and here ments, 72.6% of mobilised volumes (Fig. 7B) were
lies the highest discrepancy between air-photo and delivered to the stream channels. In other words, 49%
field data. The implication is that most of the small of ‘‘not visible’’ events were connected to stream
failures are connected to the drainage network. From a channels and 38% to gullies (Fig. 7A). The gully
process perspective, this finding elucidates the real network appears to have been very stable in the last 30
degree of geomorphic coupling of montane streams; years. Only the formation of one new gully (scoured

Fig. 7. Sediment delivery to streams as obtained by air-photo interpretation (API), fieldwork (FW), and field-coupled air-photo interpretation
(API and FW). Lumped data for East Cap and Sisters Creeks are reported in percentages (A and B). Data for East Cap Creek (C and D) and for
Sisters Creek (E and F) are reported in numbers.
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 11

down to bedrock by a large debris flow) was recorded Table 5

in November 1990 in Sisters Creek. Gullies constitute Denudation and landslide (LS) density ratios
preferential locations for sediment detachment (side- A&F/API Sisters Creek/East
Cap Creek
wall and headwall debris slides). East Cap Sisters
Creek Creek API A&F
4.2. Interbasin variability LS density and 2.4 6.8 1.6 4.4
annual LS
density
Results and observations on landslide sediment
Denudation and 1.05 1.42 7.3 9.8
production deserve a more in-depth analysis; in par- yield
ticular, we need to see whether the lumped database of
East Cap and Sisters Creeks hides any spatial varia-
bility that might prevent such findings from being
generalised to the regional scale. We also seek to landslide density (about 7), while the highest Sisters –
determine whether field surveys lead to comparable East Cap ratios are recorded for denudation (7.3 with
improvements of landslide identification in the two air photo and nearly 10 with field-coupled remote
subareas and whether sediment production and sedi- survey).
ment delivery to streams are homogeneous between A marked interbasin variability was also found in
basins. terms of landslide sediment delivery to streams. In
The contribution of field-detected failures to the East Cap Creek, the proportions between the stream
total number of landslides and, more importantly, to gully and not-connected failures do not vary substan-
the total volume of mobilised debris differs greatly tially between API and field-coupled API (Fig. 7C
between the two sub-basins (Table 4). In East Cap, the and D). By contrast, in Sisters Creek, no gully-
number of ‘‘not visible’’ slides accounts for 58.8% of connected failures were detected from API. When
the total, and the associated amount of debris mobi- field-detected failures were added, gully-connected
lised constitutes only 5%. Conversely, in Sisters, ‘‘not failures accounted for 45% of the total events iden-
visible’’ events represent 85.4% of the slides identi- tified, and 10.3% of the volume of debris mobilised
fied and mobilise 29.5% of the total debris volume. (Fig. 7E and F).
When landslide densities and denudation rates are
computed separately for East Cap and Sisters Creeks, 4.3. ANOVA and nonparametrics
they show a one order of magnitude difference. Land-
slide density is 0.051 #LS/ha in East Cap and 0.223 To this point, we have commented on the general
#LS/ha in Sisters, denudation rates are (respectively) trends obtained from the lumping of the entire data-
12.4 and 121.1 m3/ha. These results demonstrate how base or, at best, split into the two tributary domains. A
large an interbasin variability of landslide activity more thorough analysis of spatial variability can be
exists in the Capilano watershed and, indirectly, made by means of ANOVA and its corresponding
illustrate how inappropriate it would be to generalise nonparametric tests. The two types of surveys are
such figures over this physiographic region (for significantly different for both dependent variables
explanation of interbasin variability, see Section 4.5). (volume and number of landslides) at the 0.01 level
Denudation rates and landslide densities, obtained (Table 3a). The only exception is the one-way
solely from air photos (Table 4), are lower than those ANOVA for volumes of mobilised debris. The large
obtained by coupling intensive fieldwork. The ratio differences in landslide densities and mobilised vol-
between the two types of survey varies with location umes of debris described between the two sub-basins
and with the dependent variable considered: higher in (locations) have prompted us to handle locations as
Sisters Creek, lower in East Cap Creek, higher for fixed blocks. Blocking was justified for both depend-
denudation rates, lower for landslide densities (Table ent variables, as landslide density and mobilised
5). Also, the Sisters – East Cap ratio increases when volumes varied significantly as a function of location
one couples fieldwork to the routine inventory method (Table 3b). A fixed block design, when blocking is
(API). The highest survey ratio is in Sisters Creek for used to increase precision of comparisons, assumes no
 ARTICLE IN PRESS
12 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

block-by-treatment interaction (i.e., location-by-sur- As for the volumes of debris per hectare (m3/ha),
vey), while such interaction is of primary interest they show no significant survey-by-location interac-
when blocking is used to broaden the scope of tion; hence, one can comment on survey and location
inference (León and Mee, 2000). separately. The former is not significant; the latter is
As for the full factorial experiment, the advantage (Table 3b) with Sisters greater than East Cap. The
of having a factorial arrangement of treatments (or nonparametric equivalent of a two-way ANOVA, the
factors) over a block design lies in that one can Friedman test, requires the same assumptions as the
analyse interactions and, where interactions are not fixed block design (i.e., null survey – location inter-
statistically significant, one can reduce the number of action); and, as such, it can be performed only on
treatment means in the multiple comparison test. mobilised volumes of debris (Table 3b). The test
According to the full factorial ANOVA (Table 3b), indicates that at least one of the survey – location
the interaction is significant for landslide density. combinations is different from the others; this is the
Therefore, no comment can be made on locations or field-coupled API conducted in Sisters Creek.
survey separately; instead, having detected the sig- We performed the Tukey’s Honest Significant
nificant interaction, its meaning has to be graphically Difference (HSD, a generalisation of Tukey’s test
evaluated by inspecting the plot of the means (Fig. 8A for unequal sample sizes) multiple comparison test,
and B). The plots show that the coupling of API with which is considered to be of intermediate conserva-
intensive fieldwork has sensibly increased average tism. The test, in line with the Friedman’s results,
values of landslide density and volumes in Sisters reports that the coupling of API with intensive field-
Creek. In East Cap Creek, such increase appears to be work in Sisters Creek gives a significantly greater
negligible. This difference between the two locations amount of mobilised debris per unit area (m3/ha) than
is particularly large in terms of landslide density does any other survey – location combination. This
(significant location-by-survey interaction). result stresses the importance of conducting fieldwork
in the Sisters Creek area.
The database was then split in two location-wise;
hence a series of one-way ANOVA-like tests was
performed. Regardless of the dependent variable
examined, in East Cap, survey types were not sig-
nificantly different (K – W and Median tests); con-
versely, in Sisters Creek field-coupled survey was
significantly greater than the air-photo-based survey
(Table 3c).

4.4. Survey influence on effects of forest management

Moving on to the anthropogenic impacts, the

comparison of the management effects between air-
photo-based and field-coupled surveys (Table 6)
showed minor differences. Fieldwork had the effect
of shifting the old-logging– old-growth ratio toward
unity (no acceleration for both dependent variables).
According to air-photo-derived data, recent logging
produced an acceleration factor of 5.2 (in terms of
landsliding rate—LS/ha/year) as opposed to 2.1
obtained by the field-coupled survey; the effect of
recent logging on denudation rates remained the same
Fig. 8. Survey – Location interaction of (A) landslide density and between survey methods. Globally, as a result of
(B) mobilised debris. Bars indicate standard errors of the mean. fieldwork, absolute numbers change but acceleration
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 13

Table 6
Management effects on rates of landsliding and denudation: comparison between air-photo-based and field-coupled landslide inventories
Land use Annual LS density Yield Management effects Yield
(#LS/ha 1/year 1) (m3/ha 1/year 1) Annual LS density (m3/ha 1/year 1)
API A&F API A&F (#LS/ha 1/year 1) API A&F
API A&F
Sisters Creek Undisturbed forest 0.0023 0.0093 1.92 3.49 1.0 1.0 1.0 1.0
Old logging 0.0005 0.0052 0.66 2.59 0.2 0.6 0.3 0.7
East Cap Creek Undisturbed forest 0.0005 0.0016 0.33 0.34 1.0 1.0 1.0 1.0
Recent logging 0.0026 0.0034 1.46 1.53 5.2 2.1 4.4 4.5

factors remain fairly constant (Table 6). From the land 40j in forested terrain, and landslide initiation was
manager’s perspective, fieldwork application does not found very unusual below such values (Brardinoni et
change the situation; on the other hand, fieldwork al., 2002). Percentages of slope steeper than 35j are
provides the geomorphologist with better sediment significantly different (Z-test), with Sisters Creek hav-
source data to construct a more realistic sediment ing 34% of DEM cells with slope gradient steeper than
budget. 35j and East Cap Creek only 19.5%.
The influence of slope aspect on landsliding
4.5. GIS as a tool for explaining interbasin variability deserves particular attention; several studies have
of landslide density and sediment yield examined this issue in the study area. Approximately
75% of the landslides inventoried by O’Loughlin
In seeking a qualitative explanation for the one (1972) in Howe Sound, Capilano watershed, and
order of magnitude difference in landslide denudation Seymour watershed exhibited a southerly aspect. He
rates between the two study areas, a GIS-based topo- justified such landslide preferential orientation by not-
graphic analysis was performed. Elevation, slope ing that ‘‘north-facing slopes are rocky and broken, a
gradient, slope aspect, drainage density, and spatial condition which discourages landslide formation,
partition of surficial materials across terrain stability while south-facing slopes are relatively uniform and
classes are the variables that were considered. underlain by an extensive unweathered till substratum’’
East Cap Creek is located at a significantly higher (p. 31). This spatial arrangement of surficial materials
elevation (832 m F 2) than Sisters Creek (664 m F 2). is likely to derive from the north-to-south flow that
Accordingly, maximum frequency was reached at the characterised the movement of Pleistocene ice sheets in
600 – 800-m category in Sisters, whereas elevation the Pacific Ranges (Armstrong and Brown, 1954).
frequency peaked at the 1000 –1200-m category in Relations between surficial materials and aspect have
East Cap Creek (Fig. 9A). Higher elevation might been investigated more systematically for the entire
have slightly contributed to a faster evacuation of Capilano watershed (Greater Vancouver Regional Dis-
glacial till deposits from East Cap upper- and mid- trict, 1999). They confirmed that rock dominates north-
slope locations, as higher elevation induces higher erly aspects; and colluvium and till are most extensive
precipitation and greater sediment transport. on west-, south-, and east-facing aspects.
Average slope in Sisters Creek is significantly Heaviest rains in the Pacific Ranges are brought by
higher (31.1j F 0.1) than in East Cap Creek southwesterly air flows; south- and west-facing slopes
(28.1j F 0.1). Although it is difficult to generalise are therefore more directly exposed; north exposures
about a critical slope gradient for slope failure, it is are likely to remain more uniformly damp; and alter-
reasonable to assert that terrain with slopes over 35j nation of wet/dry cycles is enhanced on south-facing
can be regarded as shallow, mass movement prone, slopes. In light of these climatic controls, south- and
when soil is present (Sidle et al., 1985). Data from west-oriented slopes would be expected to be more
Howe Sound and Chapman Creek, located in the same favourable for landsliding. Fig. 9C shows that south-
physiographic region, indicate a critical gradient for facing slopes are of comparable extent in East Cap
sliding of about 30j to 35j in cut-blocks and of 35j to and Sisters Creeks but that west-facing slopes are
 ARTICLE IN PRESS
14 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

In terms of the spatial distribution of surficial

materials throughout stability classes (Fig. 10A and
B), two main points can be made. First, in Sisters
Creek, class V cover is dominant; in East Cap Creek,
class IV covers the majority of the land. Thus, the
B.C. terrain stability classes are consistent with the
outlined discrepancies in denudation rates. Second, if
one considers only class V (unstable) polygons,
colluvium (C) and bedrock (R) cover almost the same
area in the two regions. However, till deposits (M) in
Sisters Creek cover an area that is more than double
the till-mantled terrain in East Cap Creek. In addition,
there are nearly 0.2 km2 of terrain where colluvium
partially covers underlying till (/CM), a configuration
that can be considered as unstable as till itself.
Briefly summarising, the higher denudation rate
from landsliding in Sisters Creek can be explained by
considering that such a sub-basin possesses (i) steeper
slopes: 34% steeper than 35j, whereas in East Cap
Creek, only 20% of the terrain is steeper than 35j; (ii)
major till on class V slopes: it covers a surface of 2 km2
in Sisters Creek vs. just 0.7 km2 in East Cap Creek; and

Fig. 9. Frequency distributions of (A) elevation, (B) slope gradient,

and (C) slope aspects of East Cap and Sisters Creeks.

more abundant in East Cap Creek. East Cap Creek

should, therefore, be more susceptible to landsliding
than Sisters Creek. The evidence, however, is that
landslides appear to be slightly more abundant on
south and east orientations (Greater Vancouver
Regional District, 1999). Thus, unlikely aspect plays
a significant causal role for landsliding in the water-
shed.
Drainage density of the field-surveyed sites is
Fig. 10. Distributions of surficial materials: (A) in Sisters Creek and
greater in Sisters Creek (3.6 km2/km) than in East (B) in East Cap Creek. Histograms reports bedrock (R), colluvium
Cap Creek (2.2 km2/km), partly explaining the greater (C), glacial deposits (M), and glacial deposits partially covered by
denudation rate for the former substudy area. colluvium (/CM).
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 15

(iii) higher drainage density: 3.6 km/km2 in Sisters Analysis of landslide density and denudation rates
Creek compared to 2.2 km/km2 in East Cap Creek. was tackled by considering two main factors affecting
mass failure detection: type of survey and location.
The former is external to the analysed system, and the
5. Discussion latter is an intrinsic property and can be regarded as
the spatial heterogeneity of the system’s propensity to
Estimation of the maximum size of undetectable fail. Intensive ground checking of the study areas has
landslide scars on air photos is of fundamental impor- shown that ‘‘not visible’’ landslides accounted for
tance for both scientists and forest managers. Never- about one-third of the total volume of debris mobi-
theless, little research has addressed the topic (e.g., lised via mass movements during the last 30 years.
Robison et al., 1999); this could be due to under- Even more interestingly, the two subwatersheds
estimation of the relative importance of small failures exhibited very different rates of sediment production
and to the physical- and cost-prohibitive requirements from landsliding (about one order of magnitude dis-
of fieldwork. crepancy), with Sisters Creek being the more active
In Fig. 11, we summarised the factors that have (Table 4). Although these two Capilano River tribu-
been identified to affect landslide visibility during taries possess many similar biophysical characteris-
API. Factors that are manifestly related to the quality tics, they behave in strikingly different ways.
of photography itself [such as nominal scale, sensor The overall survey-by-location picture can be sum-
type (colour, black and white, etc.), weather condi- marised by saying that the same treatment (survey
tions (snow cover, clouds)] are not included. Rela- type) performed differently in different locations
tively large failures that frequently were not detected (blocks). Field-coupled surveys exhibited significantly
in antecedent API were generally located on the lower greater landslide densities and denudation in Sisters
portion of very steep (>40j) old-growth forested Creek. The same variables in East Cap Creek were
slopes. Due to the progressive downhill increase of shown to be rather indifferent to the conduct of field-
pore water pressure, clusters of debris slides are often work as a supplement to API, thus confirming the
found in these locations. On recently harvested cut- simple counts. Friedman’s test showed that field-
blocks, the problem in detecting slides lies in the coupled API in Sisters Creek gives significantly
shallow nature of the events, which facilitates a very greater mobilised volumes than any other survey –
fast process of revegetation. Furthermore, gully- location combination did. This finding emphasises
related failures connected to permanent streams are the importance of intensive ground checking. Solely
more difficult to detect. This is because stream-con- from air photos, landslide sediment production in East
nected failures have their deposition zone (e.g., flow Cap and Sisters Creeks would have not been signifi-
fans) readily washed out by the stream they have cantly different. Apparently, some types of forested
impacted. terrain hide important numbers of ‘‘not visible’’ land-

Fig. 11. Ranking of factors affecting landslide visibility on aerial photographs.

 ARTICLE IN PRESS
16 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

slides, while others do not. In other words, the issue of One of the main issues that puts air-photo-based
‘‘not visible’’ events can be complex even within a landslide studies under suspicion is the relative bias
‘‘medium-size’’ drainage basin. This finding has introduced (because of different conditions of visibil-
important practical implications: in East Cap-like ity) when one compares mass wasting activity in
areas, fieldwork is practically unnecessary and sedi- logged and undisturbed terrain. This case study has
ment budget evaluation requires virtually no correction shown that the discrepancies between air-photo-based
factor to account for not visible events. Conversely, and field-coupled surveys (Table 6) have relatively
Sisters-like areas require intensive fieldwork in order slight management implications. Fieldwork has the
to evaluate the ‘‘invisible’’ volume of mobilised debris. effect of shifting the old-logging – old-growth ratio
One aspect to consider in comparing fieldwork from negative accelerations to near constancy. Man-
landslide detection efficiency is the relative abun- agement effects of recent logging on denudation rates
dance of different land use cover—in our case, the remained constant survey-wise, while landsliding
relative extent of old-growth, old-logged, and recently ratios exhibited a three-time decrease as a result of
logged areas. In Sisters, the presence of a large portion fieldwork coupling. This is due to the higher number of
of old harvested terrain and only a small area of ‘‘hidden’’ slides in old-growth forest than in recent cut-
recently harvested cut-blocks (Table 2) does not blocks, whereas canopies of old-logged and old-growth
favour landslide detection from air photos. Con- forests have a similar effect on landslide visibility.
versely, the absence of the old-logging land use In order to explain the large discrepancy in denu-
category and the large, recently developed clearcut dation rate that was found between the two creeks, we
areas aid remote recognition of mass failures in East performed a GIS analysis. The outcome depicted
Cap (Table 2). We consider historic fires to have no Sisters and East Cap basins as two contrasting geo-
effect on landslide visibility; canopy height in fire- morphic environments (Fig. 3A and B). Several
affected forests was comparable to that in old-logged physiographic characteristics make Sisters basin a
and old-growth forested terrain. highly landslide-prone environment. This includes
A last benefit of conducting intensive fieldwork is steep slopes (one-third >35j), and till-mantled slopes
the recognition that gully-related events have a greater (f 40% class V) are a large source of potential non-
importance than one could expect from API. They coherent material. Also, high drainage density allows
constitute more than one-third (both in terms of efficient evacuation of the sediment load brought into
number of events and of mobilised volumes) of the the channel network via sidewall debris slides and
‘‘missed’’ events, while from air photos alone, they windthrow. Conversely, in East Cap, slopes are gen-
accounted for just below 10% of the total number of erally gentler; till is mainly located on less steep
failures and volumes. From aerial photos and field polygons (class IV); and drainage density is signifi-
traversing of Sisters and East Cap Creeks, the gully cantly lower, thus imparting a lower connectivity to
network appears to have been very stable in the last 30 the system.
years. The stream channel network functions as a These observations are in accordance with the
preferential transportation pathway for debris flows/ framework that subdivides systems into weathering-
torrents through the landscape. Unconnected failures and transport-limited systems (also termed supply-
account for relatively low percentages ( < 16%) of the limited and -unlimited) (Carson and Kirkby, 1972).
total number and volume of landslides (see Fig. 7). In In the first case, the limiting factor is sediment
this mountain environment, the drainage network is production; in the other, sediment mobilisation. In
extremely efficient in evacuating the sediment load. the mountainous, forested environment of coastal
This efficiency is achieved through alternation of a British Columbia, transport-limited basins typically
‘‘normal fluvial regime,’’ where supply-limited have a high density of headwater channels incised
amounts of small material systematically are eroded into thick glacial drift or closely jointed bedrock. This
from the stream channel, and of a ‘‘catastrophic ensures virtually unlimited debris supply, in addition
regime,’’ which entails removal of large quantities to many unstable trigger points for debris slides and
of sediment and woody debris via debris flows/tor- flows. Supply-limited basins denote slower recharge
rents (Nistor, 1996). rates and fewer zones of instability. This is usually
 ARTICLE IN PRESS
F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18 17

due to more massive bedrock or a thinner cover of photo-based and field-coupled surveys, vary very
glacial drift (Bovis and Jakob, 1999). In this sense, little. The impact of recent logging on denuda-
East Cap and Sisters Creeks seem to be good exam- tion rates remained constant survey-wise: land-
ples of supply-limited and transport-limited basins, sliding ratios exhibited a three-time decrease as a
respectively (Fig. 3A and B). result of fieldwork coupling because of the
higher number of ‘‘invisible’’ slides in old-
growth forest. Canopies of forest logged more
6. Conclusions than 50 years ago and those of undisturbed forest
have a similar effect on landslide visibility.
(i) This study provided for the first time a system- (v) Large differences between Sisters and East Cap
atic examination of the assumption that the Creeks in landslide density and denudation rates
relation between the numbers of landslides were qualitatively explained via GIS-based topo-
detectable from air-photo inventories and ground graphic analysis. The outcome depicted the two
surveys is the same as that for landslides that are sub-basins as contrasting geomorphic environ-
‘‘non-visible’’ on air photos. In the Capilano ments. Accordingly, the different landslide
River basin, the number of slope failures that activity between the study areas was clarified
remained undetected from air photos and the with reference to the distinction between supply-
relative volume of debris mobilised showed limited (i.e., East Cap Creek) and transport-
great sub-basin variability. In basins like Sisters limited (i.e., Sisters Creek) basins.
Creek, where below threshold scale failures are
disproportionately extensive, the forest canopy
Acknowledgements
does hide an important population of ‘‘not
visible’’ landslides; in others, such as the East
The research was supported by the U.B.C. Chair of
Cap Creek basin, it does not. It follows that, if
South – North Studies (Professor Olav Slaymaker) and
one wants to obtain complete information for
by scholarships awarded to Francesco Brardinoni by
sediment budget evaluation and terrain stability
the Faculty of Science of the Università Cà Foscari di
assessment, supplementary fieldwork is manda-
Venezia and by the Government of Canada (GOCA).
tory in Sisters Creek-like areas. In East Cap
The Greater Vancouver Regional District, through the
Creek-like areas, the supplementary fieldwork
offices of Mr. Derek Bonin and Mr. David Dunkley,
would not be justifiable.
kindly allowed and facilitated access to the restricted
(ii) Maximum area of ‘‘not visible’’ failures in forest
water supply area. Russell White, Liz-Anne Strik, and
older than 50 years was 650 m2, larger than what
Stephanie Sork assisted with fieldwork. June Ryder
has been assumed in previous studies. In recently
and Michael Church provided many insightful sug-
harvested cut-blocks, the value dropped to about
gestions and comments that greatly improved the
150 m2. Factors that have proven to affect
paper. Chris Ayles and Erik Schiefer commented on an
landslide visibility were land use, gully relation
early draft of the paper. We thank David Alexander,
of failure, slope gradient, valley width, slope
Richard Marston, and an anonymous reviewer for their
position, and stream connection. Lower portions
constructive comments on this paper.
of steep, old-growth forested slopes located in
narrow valleys are most likely to hide a relevant
number of slope failures. References
(iii) Fieldwork has demonstrated that gully-related
events have a greater importance than one could Armstrong, J.E., Brown, W.L., 1954. Late Wisconsin marine drift
detect from API. They constitute more than one- and associated sediments of the Lower Fraser Valley, British
third of the missed events, while from air photos, Columbia, Canada. Bulletin of the Geological Society of Amer-
ica 65, 349 – 364.
they accounted for < 10%. Bovis, M.J., Jakob, M., 1999. The role of debris supply conditions
(iv) This case study has demonstrated that forest in predicting debris flow activity. Earth Surface Processes and
management effects, as perceived between air- Landforms 24, 1039 – 1054.
 ARTICLE IN PRESS
18 F. Brardinoni et al. / Geomorphology 1321 (2002) 1–18

Brardinoni, F., Hassan, M.A., Slaymaker, O., 2002. Complex mass Roddick, J.A., 1965. Vancouver North, Coquitlam and Pitt Lake
wasting response of drainage basins to forest management in Map Areas, British Columbia. Memoir - Geological Survey of
coastal British Columbia. Geomorphology 49, 109 – 124. Canada, vol. 335. Ottawa, Canada.
British Columbia Ministry of Environment, 2002. Web site http:// Rollerson, T., Millard, T., Jones, C., Trainor, K., Thomson, B.,
srmwww. gov.bc.ca/risc/PUBS/TEECOLO/TerClass. 2001. Predicting Post-Logging Landslide Activity Using Ter-
British Columbia Ministry of Forests, 1995. British Columbia rain Attributes: Coast Mountains, British Columbia. Forest
Forest Practices Code. British Columbia Government, Victoria. Service British Columbia, Technical Report, vol. 011. Nanaimo,
216 pp. 20 pp.
Carrara, A., Guzzetti, F., Cardinali, M., Reichenbach, P., 1999. Use Rood, K.M., 1984. An aerial photograph inventory of the frequency
of GIS technology in the prediction and monitoring of landslide and yield of mass wasting on the Queen Charlotte Islands, Brit-
hazard. Natural Hazards 20 (2 – 3), 117 – 135. ish Columbia. Land Management Report, vol. 34. B.C. Ministry
Carson, M.A., Kirkby, M.J., 1972. Hillslope Form and Processes. of Forests, Victoria. 55 pp.
Cambridge Univ. Press, Cambridge, UK. 475 pp. Schwab, J.W., 1988. Mass wasting impacts to forest lands: forest
Donati, L., Turrini, M.C., 2002. An objective method to rank the management implications, Queen Charlotte Timber Supply
importance of the factors predisposing to landslides with the Area. In: Lousier, J.D., Still, G.W. (Eds.), Degradation of Forest
GIS methodology: application to an area of the Apennines Land: Forest Soil at Risk. Proc. 10th B.C. Soil Science Work-
(Valnerina; Perugia, Italy). Engineering Geology 63 (3 – 4), shop, Feb. 1986. Land Management Report, vol. 56. B.C. Min-
277 – 289. istry of Forests, Victoria, pp. 104 – 115.
Erskine, C.F., 1973. Landslides in the Vicinity of the Fort Randall Sidle, R.C., Pearce, A.J., O’Loughlin, C.L., 1985. Hillslope Stabil-
Reservoir, S. Dakota. Prof. Pap. - Geol. Surv. (U.S.), vol. 675, ity and Land Use. Water Resources Monograph Series, vol. 11.
p. 64. Reston, VA. Am. Geophysical Union, Washington, DC. 140 pp.
Flageollet, J.-C., 1996. The time dimension in the study of mass Smith, R.B., Commandeur, P.R., Ryan, M.W., 1983. Natural reve-
movements. Geomorphology 15, 185 – 190. getation, soil development, and forest growth on the Queen
Greater Vancouver Regional District, 1999. Annex CD to GVRD Charlotte Islands. Progress report. Working Paper 7/83. B.C.
Analysis Report Watershed Management Plan #5. Vancouver, Ministry of Forests, Fish/Forestry Interaction Program, Victoria.
BC. 44 pp.
Holland, S.S., 1976. Landforms of British Columbia, a Physio- Smith, R.B., Commandeur, P.R., Ryan, M.W., 1986. Soils, vegeta-
graphic Outline. Bulletin No. 48. BC Ministry of Mines and tion, and forest growth on landslides and surrounding logged
Petroleum Resources, Victoria. 138 pp. and old growth areas on the Queen Charlotte Islands. Land
León, R.V., Mee, R.W., 2000. Blocking multiple sources of error in Management Report, vol. 41. B.C. Ministry of Forests, Victoria.
small analytic studies. Quality Engineering 12 (4), 497 – 502. 95 pp.
Nistor, C.J., 1996. Temporal patterns in the normal-regime fine- Terlien, M.T.J., van Westen, C.J., van Asch, T.W.J., 1995. Deter-
sediment cascade in Russell Creek Basin, Vancouver Island. ministic modelling in GIS-based landslide hazard assessment.
MS thesis, The University of British Columbia, Vancouver. In: Carrara, A., Guzzetti, F. (Eds.), Geographical Information
236 pp. Systems in Assessing Natural Hazards. Kluwer Academic Pub-
O’Loughlin, C.L., 1972. An investigation of the slope stability of lishing, Dordrecht, The Netherlands, pp. 57 – 77.
the steepland forest soils in the Coast Mountains, southwest Thurber Engineering, 1996. Capilano Reservoir Investigation of
British Columbia. PhD dissertation, The University of British Shoreline Instability. Report to Greater Vancouver Regional Dis-
Columbia, Vancouver. 147 pp. trict, Vancouver, BC. 31 pp.
Pyles, M.R., Froehlich, H.A., 1987. Discussion of ‘‘Rates of land- Wieczorek, G.F., 1984. Preparing a detailed landslide-inventory
sliding as impacted by timber management activities in north- map for hazard evaluation and reduction. Bulletin of the Asso-
western California’’ by M. Wolfe and J. Williams. Bulletin of ciation of Engineering Geologists 21 (3), 337 – 342.
the Association of Engineering Geologists 24 (3), 425 – 431. Yanai, S., Usui, G., 1987. The Evaluation of Slope Stability by
Reid, L.M., Dunne, T., 1996. Rapid Evaluation of Sediment Budg- Tephrochronology in Central Hokkaido, Japan. ANZSLIDE
ets. GeoEcology Paperback. Catena Verlag, Reiskirchen, Ger- 87, 5th International Conference and Field Workshop on Land-
many. 164 pp. slides, Christchurch, New Zealand, Conference Proceedings,
Robison, E.G., Mills, K., Paul, J., Dent, L., Skaugset, A., 1999. pp. 165 – 169.
Storm Impacts and Landslides of 1996: Final Report. Forest Zaruba, Q., Mencl, V., 1969. Landslides and Their Control. Aca-
Practices Technical Report, vol. 4. Oregon Department of For- demia and Elsevier, Prague, Czechoslovakia. 205 pp.
estry, Corvallis. 145 pp.