Original Paper

Landslides (2011) 8:183194

DOI 10.1007/s10346-010-0238-4
Received: 4 January 2010
Accepted: 20 July 2010
Published online: 3 August 2010
Springer-Verlag 2010

Andrew Welsh I Tim Davies

Identification of alluvial fans susceptible to debris-flow

hazards

Abstract We describe and test a method for identifying alluvial fans

likely to be affected by debris ows. It is based on identifying
catchment parameters by geographical information system interrogation of a digital elevation model, using the Melton ratio as the
discriminating parameter. The method was calibrated using data from
debris-ow-generating catchments in Coromandel and the adjacent
Kaimai Ranges, North Island, NZ, and tested against data from the rest
of New Zealand. The procedure is remarkably (but not completely)
reliable for identifying debris-ow-capable catchments, and thus fans,
across the wide range of climates and lithologies in New Zealand
mountains. A case study illustrates the potential of the method for
avoiding future hazards and emphasises the need for a precautionary
approach when eld investigations do not detect evidence for past
debris ows.
Keywords Debris-ow hazards . Hazard area identication .
GIS . Catchment parameters . Geomorphology
Introduction
Alluvial fans are commonly used as sites for development in
mountainous regions because they are gently sloping, with good
views, well-drained and above the ood range of major rivers.
However, many small, steep fans in active mountains are occasionally
affected by debris ows (Jakob and Hungr 2005); these events are
characterised by violent surges of high sediment concentration, often
carrying large boulders and tree trunks, which can be very destructive
and affect any part of a fan surface. Evidently, any fan affected by such
events carries a high risk of severe damage to structures and thus to
life (Davies and McSaveney 2008). It is therefore essential, in
sustainable development of fan areas, that the possibility of debris
ows is investigated.
Such investigations are not simple; ideally, the magnitudefrequency spectrum of debris-ow occurrence would be inferred
from geomorphic data and the risk judged accordingly (Davies and
McSaveney 2008). However, the rst step is to decide whether a
specic fan on which development is proposed is likely to have
suffered debris ows at all in its history; if it has not, no geomorphic
investigations are needed. We present herein a method for preliminary
assessment of likely debris-ow occurrence, based on existing
empirical knowledge of the characteristics of catchments known to
give rise to debris ows (e.g. Wilford et al. 2004). This uses a
geographical information system (GIS) to identify catchments with the
required characteristics in a digital elevation model (DEM). Catchments have much greater relief than fans and are thus able to be
characterised much more accurately from a relatively coarse DEM
than are the more gently sloping fans. The procedure we outline is
suitable for routine use in preliminary assessment of development
proposals.
Background
Herein, we summarise the major outcomes of Welsh (2008). That
study developed a GIS methodology for identifying catchments

with specic morphological parameters relevant to debris-ow

occurrence. Fieldwork tested the ability of some published
discriminators to successfully identify catchments that had been
affected by debris ows.

Alluvial fan hazards

The risks attending development on alluvial fans are described by
Davies and McSaveney (2008). In particular, debris ows pose a
frequently overlooked hazard because of the following:
1. They are not well known outside the research community, and
their risks are often not considered in routine assessment of
development proposals.
2. They occur infrequently in any given drainage (a few times
per century or even millennium), so they may not feature in
local knowledge.
3. Their behaviour is so different from normal streamow that
their consequences are often difcult to believe.
Further, the geomorphic evidence of past debris ows may be
removed by subsequent streamow erosion and sedimentation,
so even expert eld inspection may reveal no sign that debris
ows have occurred.
Debris flows
The causes of debris ows, and their behaviour once initiated, are
known in outline but poorly understood as yet (e.g. Klubertanz et al.
2009; Jakob and Hungr 2005; Iverson 1997). However, it is clear that
occurrence of a debris ow requires large volumes of sediment to be
available, either on slopes or in a stream channel, and steep slopes to
allow rainfall and/or streamow of sufcient intensity to mobilise
the sediment. Thus, we expect that debris ows will occur in
catchments that are (1) steep and therefore (2) generally rather small,
(3) where erosion is active and (4) where intense rainfall or snowmelt
can occur from time to time. The geomorphic signatures of debris
ows include lobate fan deposits (by contrast with the very even fan
surfaces generated by uvial processes), channel-side levees,
anomalously large boulders in channels and on the fan, and scarring
high on streamside and fan trees.
A number of investigators have differentiated between debris
ows and debris oods (also referred to as hyperconcentrated
ows; e.g. Pierson 2005), in terms of their behaviour and hazard
potential. Both differ from normal ood ows in streams by having
very high concentrations (30% by weight; Davies 1988) of
suspended ne sediment, but only debris ows carry large boulders.
In the present work, our focus is on debris ow hazards.
Catchment characteristics
A number of studies have sought to identify catchment and fan
topographic discriminators for debris ows (e.g., Jackson et al.
Landslides 8 & (2011)

183

Original Paper
Fig. 1 Location of the Coromandel/
Kaimai region study area; Matata
(lower right) is the location of
Awatariki and Waitepuru streams,
which are discussed later

1987; Kostaschuk et al. 1986; de Scally et al. 2001; de Scally and Owens
2004; Wilford et al. 2004; Rowbotham et al. 2005). In particular, a
handful of morphometric variables including basin area (Kostaschuk
et al. 1986; de Scally and Owens 2004), Melton ratio (an index of
basin ruggedness; equal to basin relief divided by the square root of
basin areaMelton 1965; Jackson et al. 1987; de Scally and Owens
2004; Wilford et al. 2004) and watershed length (the planimetric
straight-line distance from the fan apex to the most distant point on
the watershed boundaryWilford et al. 2004) have been identied
as capable of identifying and differentiating debris-ow and nondebris-ow basins and their respective fans.

Field studies
We have used catchments in the Coromandel and Kaimai
Ranges of North Island, New Zealand (Fig. 1), to test Melton
ratio and watershed length against geomorphic evidence for
debris-ow occurrence. This area has in the past been affected
by a number of damaging debris-ow events (e.g. Fig. 2 shows
184

Landslides 8 & (2011)

Fig. 2 Devastation and destruction as a result of debris flows in the small town of
Te Aroha, New Zealand, in 1985 (Encyclopaedia of New Zealand 2000)

the result of an event in which 50 homes were damaged and

three people died; Montz 2007).
The study area lies in the northeast of North Island; its mountain
ranges rise to almost 1,000 m above sea level (asl) and cover a total
area close to 6,500 km2. The basement rock is late Jurassic greywacke,
overlain by Oligocene sediments and late Cenozoic sub-aerial
volcanics. Volcanism rst occurred about 20 million years ago near
the northern tip of the Coromandel Peninsula and spread southward
over the next 18 million years to form a chain of volcanoes, the
remnants of which now comprise the Coromandel and Kaimai ranges.
A well-developed NNW and NNE to ENE fault block pattern is evident
in the Coromandel/Kaimai region (Christie et al. 2001; Figs. 3 and 4).
The Coromandel and Kaimai regions are characterised by a diverse
range of vegetation, from low-lying grassland and mixed indigenous
scrub to indigenous and exotic forest (Newsome 1987).
Weather patterns in the Coromandel and Kaimai regions
(and New Zealand in general) are dominated by the easterly
movement of frontal systems from the Tasman Sea (Environment Waikato 2007; Jane and Green 1984; Maunder 1970).

Annual precipitation often exceeds 3,000 mm in the ranges,

with maxima up to 4,500 mm (Environment Waikato 2007).
Extreme rainfalls in the ranges often result from summer
tropical cyclones; winter storms bring heavy rainfall and
occasionally snow to the ranges. Mean annual temperatures
vary between 12C and 14C (515C in winter months and 1525C
in summer); the region receives between 1,800 and 2,000 h of
sunshine per annum (National Institute of Water and Atmospheric
Science 2007).
The main population centres include Thames (population 6,500),
Paeroa (3,700), Te Aroha (3,700), Waihi (3,700), Whitianga (2,500) and
Whangamata (2,500). Tourism, horticulture, farming, shing, forestry
and mining are the main commercial activities in the region; many
smaller settlements associated with these activities occupy locations
close to small streams.
Methodology
Arc View (V.9.2; ESRI 2006) was used to identify catchments
and calculate their parameters (Welsh 2008). The catchment

Fig. 3 DEM of the Coromandel Range

region showing the location of specific
stream sites surveyed in the field
investigation

Landslides 8 & (2011)

185

Original Paper
Fig. 4 DEM of the Kaimai Range
region showing the location of
specific stream sites surveyed in the
field investigation

extent was delineated by the watershed area contributing to

streamow at the location of the most upstream building in
the valley bottom located on the 1:50,000 map of New Zealand
and the corresponding 25-m DEM. Tools (Arc GIS geoprocessing commands) within the spatial analyst toolset in Arc
Toolbox (the user interface used for accessing, organising and
managing geoprocessing tools, models and scripts; ESRI 2006)
were used to investigate the hydrologic characteristics of the
Coromandel/Kaimai study area. These tools were linked within
the Spatial Analyst Model Builder (the interface used to build
and edit geoprocessing models in ArcGIS; ESRI 2006) to
construct a systematic ow diagram or model, enabling the
delineation of watersheds in the study area and extraction of
morphometric parameters associated with them. Input data
were entered as a list into each tool to enable model iteration
and hence processing of large amounts of data automatically
and simultaneously.
The watershed model thus formulated was divided into
186

Landslides 8 & (2011)

two parts: the rst was used to extract a study area subset
from an existing DEM of New Zealand and to delineate a
drainage network for the study region; the second was used to
delineate watersheds in the study area and extract the
morphometric parameters associated with them. Once devised
and run to produce outputs, the extracted morphometric
parameters were exported into a Microsoft Excel spreadsheet
where Melton ratios and watershed lengths for each of the
watersheds were calculated and prepared for subsequent
analysis. Further details of this procedure are presented in
Welsh (2008).
After some preliminary trials, the parameters Melton ratio (R=
Hb/Ab, where Hb is basin relief and Ab basin area, after Melton 1965)
and watershed length (WL, based on the model proposed by Wilford
et al. 2004) were selected as combining ease of use with adequate
discrimination.
Field investigations involved identication and quantication of
a range of geomorphic and sedimentary discriminators (listed later) in

the stream channel. Aerial photograph interpretation involved

searching for the same discriminators on stereo pairs; this was
ultimately unsuccessful due to the inadequate resolution (1 m) of the
available images.

Table 2 Values of R and WL for locations in the study area

Stream location and R/WL category

WL (km)

Category A
Waiwhango stream

0.15

4.31

Hazard thresholds for Melton ratio and watershed length parameters

Melton ratio (R) values for each stream watershed were allocated into
three categories, corresponding to the thresholds for debris ow,
debris ood and uvial phenomena, dened by Jackson et al. (1987)
and Wilford et al. (2004):

Te Puru stream

0.14

6.49

Taruru stream

0.17

6.52

Waitoki stream

0.20

4.07

Putangi stream

0.19

4.05

&
&
&

Matutu stream

0.30

3.35

0.35

3.31

R0.30The threshold below which conventional uvial

processes are generally dominant in a watershed
0.30<R<0.60The range for watersheds that are prone to debris
oods
R0.60The threshold above which watersheds are prone to
debris ows

&
&

Otuturu stream

0.39

1.98

Waiotahi stream

0.46

1.98

Wahine stream (2)

0.47

3.41

In addition, threshold classes were also dened for values

of watershed length (WL) following Wilford et al. (2004):

Gordon stream

0.41

4.14

Stanley stream

0.38

4.40

WL2.7 kmDebris ows can occur in the watershed

WL>2.7 kmConventional uvial processes and/or debris
oods are the dominant processes in the watershed

Category F
Waitoitoi stream

0.68

2.66

Whakanekeneke stream (2)

0.71

1.65

Karaka stream (3)

0.61

1.16

Lipsy stream

0.76

2.21

Moonlight stream

0.95

1.89

Gordon Rd.: unnamed stream

1.51

1.47

This gave rise to six different categorical combinations

(labelled A to F) of R and WL, as shown in Table 1.
Eighteen sites were selected for eld investigation, nine in
each of the Coromandel and Kaimai regions (Figs. 3 and 4),
according to values of R and watershed length obtained for each
stream watershed through the GIS procedure. Of the nine
locations selected for each region, three streams were selected in
category A, three in categories C or D and three in category F. No
category B streams were selected because they were assumed to
show uvial characteristics. No category E streams were selected
as only three watersheds in the study area plot in this category.
Values of R and WL for the selected locations are shown in Table 2.
The locations of the sites investigated are shown in Figs. 3 and 4.
Geomorphic and sedimentary characteristics of debris ows, debris
oods and uvial oods are as follows (e.g. Davies 1997; Jakob and
Hungr 2005):
Debris ows:

&
&
&

Categories C and D
Otohi stream

Steep channel (>5)

Narrow channel, small width-to-depth ratio
Semi-circular to U-shaped channel

Table 1 Category combinations of R and WL

Category

Combination

R0.30; WL>2.7 km

R0.30; WL2.7 km

0.30<R<0.60; WL>2.7 km

0.30<R<0.60; WL2.7 km

R0.60; WL>2.7 km

R0.60; WL2.7 km

(n) indicates watershed derived for an unnamed tributary to the main stream

&
&
&
&
&
&
&
&
&
&
&
&
&

Sinuous terraces formed by ow margins, sloping away from the

channel
Channel scoured to bedrock
Lobate areas of even-age vegetation younger than the
surrounding growth
Old bark scars high on trunks and branches of trees
Coarse deposits beyond the channel on the fan
Substantial depositional lobes in the channel and on the fan surface
Levees of coarse material aligned along the stream on upper fan
Boulders rolled against trees on the channel banks or lodged
high above stream channel
Isolated boulders in the channel and on the fan surface
Unstratied deposits with no structure or imbrication
Angular to sub-angular clasts
A-axes of clasts oriented randomly or parallel to ow
Poorly sorted and matrix-supported deposits
Debris-oods:

&
&
&
&
&
&

Channel has medium to large width-to-depth ratio

Bars, sheets, fans and splays notable at local scale in the channel
Bouldery deposits beyond the channel on the fan
Moderate to poor sorting of deposits
Clast-supported deposits
Deposits have mixed clast orientations: A-axes of large clasts
perpendicular to ow, pebbles and small cobbles parallel to
ow direction
Landslides 8 & (2011)

187

Original Paper
uvial processes and the parameters R and watershed length.
The range in evidence observed for uvial, debris-ow and
debris-ood processes at stream locations in the study area
differs substantially between categories (Figs. 8, 9 and 10; Welsh
2008); we conclude that the categorisation chosen as the basis of
the GIS analysis has succeeded in matching the eld parameters
indicating the dominant processes.

100
Debris-flow
90

Debris-flood
Fluvial

80

60
50
40
30
20
10

tre
am
W
ai
to
ki
St
re
am
Pu
ta
ng
iS
tre
am
M
at
ut
u
St
re
am

S
Ta

ru

ru

ru
Pu
Te

100

Category A stream watershed locations

Debris-flow

40

30

20

10

Landslides 8 & (2011)

St
an
l

ey

St
r

St
r

ea

(2
am

ah

in

ot
W

G
or
do
n

re

ah

hi
to
O

St

St

iS
tre
am

re
am

ea

The evidence recorded for debris ow, debris ood and

uvial processes was quantied by the percentage of the total
noted features corresponding to each criterion for each stream in
each category A, C and D, and F (Figs. 5, 6 and 7).
Linear regressions were carried out to test the relationships
between observed evidence for debris ows, debris oods and

188

50

Meandering or braided channel pattern

Large width-to-depth ratio
Moderate to well-sorted clast-supported well-imbricated deposits
Presence of sedimentary features (e.g. bars, rifes)
Presence of stratication and layering, cross-bedding, ame
structures
Clasts sub-rounded to rounded
A-axes of larger clasts perpendicular to ow
Deposits rare beyond the channel on the fan
Bars, sheets, fans and splays notable at local scale in the channel

60

St
r

&
&
&
&
&
&
&
&
&

70

ai

Fluvial oods:

80

Observed criteria (%)

Weakly imbricated deposits and collapse packing in coarser

sediment fraction
Poorly stratied deposits, comprised of loose mixtures of
coarse gravel and sand
Clasts rounded to sub-angular

Fluvial

O
tu
tu
ru

&
&
&

Debris-flood

90

Fig. 5 Evidence observed for all processes (debris flows, debris floods and fluvial
processes) at stream watershed locations exhibiting R values <0.30 and lengths
>2.7 km (category A)

W
ai
wh
an
go

St

St

re

re

am

am

Application to other known debris-flow streams

In addition to the watershed data derived for stream locations in
the study area, Melton ratios and watershed lengths were also
derived for 18 other stream locations in New Zealand. These were
selected because they are known (e.g. McSaveney et al. 2005; de
Scally and Owens 2004) to have experienced debris ows in the
past (Table 3).
All of these apart from the Awatariki stream and
Waitepuru stream (see Fig. 1) are in the Southern Alps of the
South Island, New Zealand (Fig. 11). R values for the stream
watersheds range from 0.17 at Awatarariki Stream in Matata,
Bay of Plenty, to a maximum of 1.31 at Candys Creek, Otira,
and WL values from 1.15 km at Candys Creek, Otira, to 4.4 km
at Bullock Creek, Fox Glacier. A scatter plot of R against WL
for these stream watershed locations is shown in Fig. 12. The
majority of stream watersheds (11/18; 61%) plot in category F.
The next highest frequencies are observed in categories C (2/18;
11%) and E (2/18; 11%). The R and WL values derived for

ea

Observed criteria (%)

70

Category C & D stream watershed locations

Fig. 6 Evidence observed for all processes at each stream location in categories C and D

100
Debris-flow
Debris-flood

90

Fluvial
80

Observed criteria (%)

70
60
50

40
30
20

Fig. 9 Evidence observed for debris floods (DFld) at stream locations in terms of
categories of Melton ratio

10

am

n
do
or
,G

re
St
ed
U

nn

am

ka
ha
W

am

Li

ig
M

oo
nl

ps

ht

St

St

re

re

am

)
am
re
St
ka

ra
Ka

ne

ke

ne

ai

ke

to

St

ito

re

iS

am

tre

(2

(3

am

Category F stream watershed locations

Fig. 7 Evidence observed for all processes at stream locations in category F

debris-ow watersheds outside the study area compare well

with the study area results (Fig. 12), despite a wide variation in
geological, lithological and meteorological settings. A few of
the watersheds, however, display lower R values and higher WL

Fig. 8 Evidence observed for debris flows (DF) at stream locations in terms of
categories of Melton ratio

values than expected. In particular, Awatarariki (R =0.17) and

Waitepuru (R=0.25) stream watersheds in Matata (Bay of
Plenty, North Island) display abnormally low R values which
fall well below all thresholds identied in the literature and are
consistent with that dened for uvial watersheds (R<0.30;
Jackson et al. 1987; Wilford et al. 2004). Despite this, both
Awatarariki and Waitepuru watersheds have produced large
debris ows in the past, in addition to the major event in 2005
(McSaveney et al. 2005).
It is thus likely that other factors such as the local
topography near the drainage point of the watershed and
sediment supply conditions are more important controls on
the potential for debris ows in these watersheds. For
example, examination of the DEM for Awatarariki Stream
watershed shows a number of small tributaries to lie
adjacent to each other and in relatively close proximity to
the fan. Rainstorm-generated debris ows in these tributaries
may thus have potential to coalesce at stream junctions (as
was the case in the 2005 event; McSaveney et al. 2005). This
may have the effect of forming one large debris ow capable
of owing further than any ow on its own. Furthermore,
the weak, ne-grained, sedimentary and volcanic deposits
that underlie the Matata region (McSaveney et al. 2005) may
cause resultant debris ows to be more uid and hence
capable of owing on shallower gradients than expected.
Essentially, though, the anomalous characteristics of these
two streams remain unexplained. Bullock Creek, Fox Glacier,
exhibits a greater WL (4.46 km) than the 2.7 km upper
threshold for debris-ow watersheds identied by Wilford et
al. (2004). Despite this, it has an R value close to 0.60,
indicating that channel gradients within the basin are
sufcient to maintain momentum in debris ows. It is
possible that glacial conditions within the area may have
had an inuence on the ability of debris ows to reach the
fan in this watershed (e.g. as a result of glacial outbreak
oods; Jackson et al. 1987). In addition to Bullock Creek, a
few other debris-ow watersheds (Halpin CreekWL

Landslides 8 & (2011)

189

Original Paper
The methodology developed thus appears capable of providing
an initial estimate of the potential for debris ows to occur on any
fan whose watershed parameters can be analysed via a commonly
available DEM. With the exception of streams such as Awatariki and
Waitepuru, it appears that any stream with a Melton ratio greater
than about 0.5 should be considered as a potential source of debris
ows. The dashed line representing this criterion in Fig. 12
successfully identies all the South Island debris-ow streams, as
well as all the category F streams studied in Coromandel and Kaimai
(Table 2).

Fig. 10 Evidence observed for fluvial processes (Flu) at stream locations in terms
of categories of Melton ratio

3.18 km; Waterfall CreekWL 3.15 km; and Pipson Creek

WL 3.11 km) exhibit longer watershed lengths than 2.7 km. They
also, however, exhibit high R values (e.g. 0.52, 0.94 and 0.88,
respectively). In light of this, it is possible that a slightly higher
threshold of WL may exist for differentiating debris-ow from nondebris-ow watersheds in the Southern Alps. In practice, this
complication can be avoided by focusing primarily on the Melton
ratio and ignoring the watershed length.

Case studyStony Creek, Westland, New Zealand

The alluvial fan of Stony Creek, near Franz Josef, Westland, New
Zealand, is presently undergoing extensive development as part of
the rapid growth of tourism in the region (Fig. 14). Prior to
development, it was recognised as a potential source of debris
ows based on the work of de Scally and Owens (2004) because of
its location in a region of high and intense rainfall, with faultweakened rock in much of the catchment (Grant 1998; Fig. 13).
Figure 12 conrms this potential on the basis of the present work.
Unpublished eld investigations undertaken by professional
geologists and engineers, however, reported no evidence for
debris ows on the fan or in the streambed, and on that basis
development on a small part of the fan was consented. During
this development, excavations revealed that large boulders were
present in signicant quantities (Fig. 14).
Development has since proceeded, and utilisation of much of
the rest of the fan area as a tourist resort is now proposed (see
http://www.franzalpineresort.co.nz/opport.htm).

Table 3 R and WL derived for stream locations outside the study area

Stream watershed location

Pipson Ck, Makarora

0.89

Waterfall Ck, Lake Hawea

Bullock Ck, Fox Glacier

Yellow Ck, Fox Glacier

1.18

1.53

Stony Ck, Tatare

0.86

2.63

Greyneys Ck, Arthurs Pass

1.07

1.80

Halpin Ck, Arthurs Pass

0.52

3.18

Unnamed Ck, Turiwhate

1.17

1.48

Gr

Turiwate Ck, Turiwhate

1.06

1.58

Gr

10

Grahams Ck, Turiwhate

0.88

2.33

Gr

11

Carew Ck 1, Lake Brunner

1.02

1.77

Gr

12

Candys Ck, Otira

1.31

1.15

13

Unnamed Ck 1, Boyle River

0.92

2.05

14

Unnamed Ck II, Boyle River

1.08

1.97

15

Bullock Ck, Mt Thomas

0.58

1.51

16

Kowhai R, Peel Forest

0.69

2.32

17

Awatarariki Stm, Bay of Plenty

0.17

3.68

18

Waitepuru Stm, Bay of Plenty

0.25

2.38

WL (km)

Rock

Climate

3.11

0.94

3.15

0.57

4.46

Rock type: S schist, G greywacke, Gr granite, V volcanic; annual rainfall: M moderate (4,000 mm), W wet (>5,000 mm), D dry (<2,000 mm)

190

Landslides 8 & (2011)

Fig. 11 South Island locations and

allocated R/WL categories for stream
watersheds outside the study area
known to have produced debris flows

Discussion
The criterion R>0.5 appears to be reasonably successful in
identifying the catchments that are capable of generating debris
ows across wide varieties of lithology, climate and vegetation.
This is an unexpected result, given the fact that debris-ow
occurrence is known to depend on rainfall intensity and sediment
availability, and these can vary dramatically with lithology, uplift
rate and climate. In particular, there is evidence that a threshold
of rainfall intensity must be exceeded to initiate a debris ow (e.g.
Caine 1980) and that this threshold varies widely with topography
and lithology (Baum and Godt 2010). One would expect a
successful predictor to involve all these factors and to be
correspondingly complex. We appear to have identied two
phenomena (the occurrence of debris ows and R>0.5) that both
depend in a similar way on these controlling factors and thus are
closely (but perhaps not causatively) correlated to each other.
Certainly, it is difcult to develop a convincing mechanical
argument that the occurrence of debris ows depends only on R.

The method reported herein failed to identify the two

Matata streams as likely sources of debris ows. This is a
signicant concern and needs to be borne in mind when using
the method. It appears that the Matata debris ows must be
different in some respect to those that occur in Coromandel/
Kaimai and in the Southern Alps, but the nature of the
difference remains unclear; in the sense of the preceding
paragraph, perhaps R and debris-ow occurrence at Matata are
differently related to the governing variables. We therefore
recommend that the present method be applied only in areas
in which the topography resembles that of Coromandel and the
South Alps, which are dominated by approximately angle-ofrepose slopes and sharp or rounded peaks and ridge crests,
rather than in quite different topographies such as the lowelevation (300 m asl) at-topped coastal plateau into which
Awatariki and Waitepuru (Matata) streams are incised (Fig. 15).
Because the GIS technology required for this methodology
is widely available and simple to set up and so can be used by

Landslides 8 & (2011)

191

Original Paper
10.00

Fig. 12 Scatterplot of Melton ratio

and watershed lengths for stream
watersheds outside the study area
known to have experienced debris
flows. 5 = Stony Creek, 17 = Awatariki
Stream, 18 = Waitepuru Stream.
Dashed line is R=0.5

9.00

Watershed length (km)

8.00
7.00

6.00

17

5.00
4.00

3.00
2.00
1.00
0.00
0.00

18

B
0.20

D
0.40

F
0.60

0.80

1.00

1.20

1.40

1.60

Melton's R

persons untrained in geomorphology, it has the potential to

overcome the two primary problems in identifying alluvial fans
with potential debris-ow hazardsthe fact that few people
are skilled in recognising the geomorphic signatures of debris
ows and the fact that evidence may not obviously be present
though the hazard is.
The method does not give any indication of the
magnitude-frequency distribution of debris ows in identied catchments. That would require much more intensive
investigation and indeed may be neither feasible nor
worthwhile. This follows from the deep conundrum of
debris-ow hazard mitigation; debris ows occur relatively
rarely in any given catchment, but they can occur during
any intense rainstorm, and when they do occur they are
potentially lethal. Debris-ow hazard mitigation is not,
fundamentally, a matter of risk management because the
level of acceptable risk to life from natural hazards is 105
106 per year (Finlay and Fell 1997; Gillon 2000), and a
debris ow is always a risk to the life of any person in its
Fig. 13 Catchment and fan of Stony
Creek, Westland, New Zealand,
March 2009. The approximate fan
area now proposed to be developed
is outlined in yellow. Dashed
lines indicate approximate location
of the Alpine fault surface trace.
View is to the south

192

Landslides 8 & (2011)

path. If a debris ow occurs, say, every 100 years or so on a

given fan, and affects say 10% of the fan area, then the risk
to life of a person anywhere on that fan is 103 per year of
occupancy; this is a least two orders of magnitude greater
than acceptable. Structural countermeasures for debris ows
are recognised as both expensive and unreliable (Davies
1997; Takahashi 1981), and avoidance is always both cheaper
and more reliable. It follows that if a fan is identied as a
debris-ow fan, development should not occur under
rational decision-making systems.
The development of Stoney Creek fan described in the
case study raises a number of issues:

&

Lack of evidence for debris-ow activity did not in this case

mean that no activity had occurred; it later became apparent
that sub-surface evidence existed but was not found by the
investigation. This is always a possibility; the present method
may act as an incentive to look more closely if debris-ow
susceptibility is indicated.

warning of intense rain, warning-evacuation systems might be

feasible in some cases (e.g. Davies and Hall 1992), though the time
taken to reliably evacuate a substantial number of people in bad
weather must not be underestimated and neither must the
problem of resistance to evacuation resulting from a series of
false alarms.
The method presented herein, then, appears to have some
potential value as a hazard identication tool. It does not mitigate
the hazard, but until a hazard is identied it cannot be mitigated, so
it is an enabling device. Its success in picking out debris-ow-prone
catchments in a wide range of lithologies and climate zones in New
Zealand mountains may indicate potential value elsewhere in the
world.

Fig. 14 Large boulders (1-m diameter; arrowed) in spoil heaps created by

excavation during development of Stony Creek fan

&

Development pressures make it difcult for local authorities to

constrain, delay or decline consent applications on the basis
that hazards may be present if no evidence has been found on
the ground. Thus, by default, hazards are assumed to be absent
unless denite evidence for their presence can be found. Given
that lives may be at risk when the next debris ow occurs, this
policy is difcult to justify, especially in comparison with the
precautionary principle commonly used to avoid environmental impacts from developments. Again, methods such as
the one described herein could be an incentive for more
searching eld investigations.

Fans already developed are, of course, a common, different

and much more difcult matter. Relocation is always politically
difcult and often socially unacceptable too, so structural
measures may have to be utilised in such cases, although the
design criteria are exceedingly difcult to specify reliably (Davies
1997), to the extent that their likely effect is difcult to quantify.
With the advent of weather radar able to provide advance

Fig. 15 Matata, Bay of Plenty, New Zealand (Google Earth image post-2005),
looking NW. The distance between the two named streams is about 1.5 km. Note
the large number of slope failures in the stream catchments

Summary and conclusions

We have developed and tested a simple GIS-based method for
identication of catchments likely to generate debris ows. Catchment area is dened as the area contributing water ow to the
farthest upstream building on the valley oor. The method uses
Meltons R>0.5 as a discriminator; this was based on study of 18
catchments in the Coromandel and Kaimai Ranges, North Island,
New Zealand, for which the value of R was compared with
geomorphic and sedimentary indicators of the occurrence of past
debris ows, debris oods and water oods. These comparisons
showed that an R value greater than 0.5 indicated accurately those
catchments showing signs of debris-ow occurrence. We also tested
this criterion on two catchments at Matata in the North Island and 16
catchments in the Southern Alps, South Island of New Zealand,
known to generate debris ows and which had lithologies and
climates different from the Coromandel and Kaimai catchments;
only the two Matata catchments (Awatariki Stream and Waitepuru
Stream) had R<0.5, all the others had R>0.5.
The reasons for the anomalously small R values for the Matata
catchments are unknown, but we note that both have morphologies
distinctly different from those of the rest of the catchments
investigated. The Matata catchments are incised into a low-lying
coastal plateau, while the others are set in mountain ranges
characterised by long slopes at about the angle of repose and sharp
or rounded ridge crests. We thus recommend that the methodology
presented should be applied only in the latter type of topography.
We illustrate the value of the technique using a case
example from New Zealand. A fan area was approved for
development on the basis that eld investigations yielded no
evidence for debris-ow occurrence, such as large boulders in
the fan. However, once development started, the site excavation
revealed numerous large boulders; these were not interpreted
as evidence of debris ows, and development is proceeding as
if the debris-ow hazard was not present. This catchment has
R=0.81, so it clearly meets the criterion for possible debrisow occurrence. The use of the present methodology at an
early stage would have indicated the need for more thorough
site investigation, which would have demonstrated the
existence of the hazard and allowed the development plans to
be modied to avoid or mitigate the hazard. In other words,
the hazard could have been signicantly reduced.
We conclude that:
1. The method presented is simple to use and appears reasonably
reliable in identifying mountain catchments liable to generate
Landslides 8 & (2011)

193

Original Paper
debris ows from time to time, in a wide range of climates and
lithologies.
2. It did not identify as debris-ow-prone the two Matata
catchments that generated debris ows in 2005, so it may be
unreliable in terrain that differs in topography from that of
the Coromandel and Southern Alps.
Acknowledgements
We gratefully acknowledge support for this project from the
Earthquake Commission of NZ; in-kind assistance and enthusiastic cooperation from Environment Waikato, NZ; and the
geomorphic advice of Dr. M J McSaveney, GNS Science, Lower
Hutt, NZ. Clive Sabel and Tom Cochrane of the University of
Canterbury both provided invaluable technical advice.

References
National Institute of Water and Atmospheric Science (2007) http://www.niwascience.co.
nz/edu/resources/climate. Accessed 8 March 2007
Baum RL, Godt JW (2010) Early warning of rainfall-induced shallow landslides and
debris flows in the USA. Landslides (in press). doi:10.1007/s10346-009-0177-0
Caine N (1980) The rainfall intensityduration control of shallow landslides and debris
flows. Geogr Ann 62A:2327
Christie AB, Brathwaite RL, Rattenbury MS, Skinner DNB (2001) Mineral resource
assessment of the Coromandel region, New Zealand. Science Report 2001/11,
Institute of Geological and Nuclear Sciences, Lower Hutt, New Zealand
Davies TRH (1988) Debris flow surgesa laboratory investigation. Mitteilung Nr 96,
VAW, ETH-Zurich, Switzerland, p 122
Davies TRH (1997) Using hydro-science and hydro-technical engineering to reduce
debris-flow hazards. In: Proceedings of the First International ConferenceDebrisflow Hazards Mitigation: Mechanics, Prediction and Assessment, pp 787810, San
Francisco, American Society of Civil Engineers
Davies TRH, Hall RJ (1992) A realistic strategy for disaster prevention. In: Proceedings,
Interpraevent 1992, Bern, Switzerland, pp 381390
Davies TRH, McSaveney MJ (2008) Principles of sustainable development on fans. J
Hydrol NZ 47:4365
de Scally FA, Owens IF (2004) Morphometric controls and geomorphic response on fans
in the Southern Alps, New Zealand. Earth Surf Process Land 29:311322
de Scally FA, Slaymaker O, Owens IF (2001) Morphometric controls and basin response
in the Cascade Mountains. Geogr Ann 83A:117130
Environment Waikato (2007) www.ew.govt.nz/enviroinfo/profile/environment/climate.
htm. Accessed 4 May 2007
ESRI (2006) Arc GIS V.9.2 desktop electronic help manual.
Finlay PJ, Fell R (1997) Landslides: risk perception and acceptance. Can Geotech J
34:169188
Gillon MD (2000) Current trends in assessing dam safety. In: Proceedings of Technical
Groups, IPENZ, NZ Society on Large Dams, Wellington, NZ, pp 2536

194

Landslides 8 & (2011)

Grant HI (1998) Late Quaternary and engineering geology of the Franz Josef area, South
Westland. Unpubl. MSc thesis, University of Canterbury, New Zealand,
p 218
Iverson RM (1997) The physics of debris flows. Rev Geophys 35:245296
Jackson LE Jr, Kostaschuk RA, MacDonald GM (1987) Identification of debris-flow
hazard on alluvial fans in the Canadian Rocky Mountains. Rev Eng Geol 7:115
124
Jakob M, Hungr O (eds) (2005) Debris-flow hazards and related phenomena. Springer,
Berlin
Jane TG, Green TGA (1984) Ecological aspects of climate patterns within the Kaimai
Ranges, North Island, New Zealand. NZ J Ecol 7:183197
Klubertanz G, Laloui L, Vulliet L (2009) Identification of mechanisms for landslide type
initiation of debris flows. Eng Geol 109:114123
Kostaschuk RA, Macdonald GM, Putnam PE (1986) Depositional process and alluvial
fandrainage basin morphometric relationships near Banff, Alberta, Canada. Earth
Surf Process Land 11:471484
Maunder WJ (1970) World survey of climatology, vol 13. Elsevier, Amsterdam
McSaveney MJ, Beetham RD, Leonard GS (2005) The 18 May 2005 debris-flow disaster
at Matata: causes and mitigation suggestions. Client Report 2005/71, Institute of
Geological and Nuclear Sciences, Lower Hutt, New Zealand
Melton MA (1965) The geomorphic and paleoclimatic significance of alluvial deposits in
southern Arizona. J Geol 73:138
Montz BE (2007) The effects of flooding on residential property values in three New
Zealand communities. Disasters 16:283298
Encyclopaedia of New Zealand (2000) www.teara.govt.nz
Newsome PFJ (1987) The vegetative cover of New Zealand. National Water and Soil
Conservation Authority, Wellington
Pierson TC (2005) Hyperconcentrated flowtransitional process between water-flow
and debris-flow. In: Jakob M, Hungr O (eds) Debris-flow hazards and related
phenomena. Springer, Berlin
Rowbotham D, de Scally F, Louis J (2005) The identification of debris torrent
basins using morphometric measures derived within a GIS. Geogr Ann
87A:527537
Takahashi T (1981) Estimation of potential debris flows and their hazardous zones: soft
countermeasures for a disaster. J Nat Disaster Sci 3:5789
Welsh AJ (2008) Delineating debris-flow hazards on alluvial fans in the Coromandel and
Kaimai regions, New Zealand using GIS. MSc thesis, University of Canterbury, New
Zealand, 169p + App
Wilford DJ, Sakal ME, Innes JL, Sidle RC, Bergerud WA (2004) Recognition of debrisflow, debris-flood and flood hazard through watershed morphometrics. Landslides
1:6166

A. Welsh
Geoconsult Pty Ltd., Brisbane, Australia
T. Davies ())
Geological Sciences,
University of Canterbury,
Canterbury, New Zealand
e-mail: tim.davies@canterbury.ac.nz