Landslide-Susceptibility Mapping in the

Amahata River Basin, Japan

Masamu Aniya

Institute of Geoscience, University of Tsukuba, Ibaraki 305, Japan

-_
Abstract. This study is concerned with ( I ) defining the terrain parameters and attributes that are
important t o landsliding in the Aniahata River basin near Tokyo, Japan and (2) mapping landslide
susceptibility from these characteristics. After aerial photographic interpretation and field surveys,
ten terrain parameters, which were thought to be important to landsliding, were measured at each
landslide and at the intersections of a 250-meter grid. Failure rate analysis and quantification scaling
type I1 suggest that slope gradient, aspect, slope plan form, break of slope, elevation, and vegetation
are the most critical site characteristics.
L~andslide-susceptibilitymapping was then accomplished by combining the three most basic terrain
parameters: slope gradient, aspect, and slope plan form. First, stable and unstable categories were
separated: then each category was divided into two classes to give four susceptibility classes: high,
moderate. low. and least. These classes represent a qualitative index of the likelihood of a landslide
occurring during heavy rainfall.
Key Words: terrain parameter, terrain attribute, landslide susceptibility, failtire rate analysis, quan-
tification scaling type 11.

P OPULAlION increase and subsequent ur-

banization have demanded t h e develop-
ment of new residential a r e a s around cities.
The general procedure for preparing landslide-
susceptibility maps usually involves three steps:
( I ) selection of critical terrain parameters and
When a city is surrounded by hills and moun- their classes; ( 2 ) production of parameter maps:
tains, as in the San Francisco Bay region, resi- and (3) mapping landslide susceptibility by su-
dential development e n c r o a c h e s o n hill a n d perimposing these parameter m a p s o n e a c h
mountain tops (Nilsen and Brabb 1973). In the other. The study will be concerned with steps
face of housing demands, developers and resi- ( I ) and (3).
dents often ignore o r forget dangers inherent in The selection of critical terrain parameters is
hill and mountain slopes. namely slope failure in many cases based on the results of field in-
and associated phenomena. F o r e x a m p l e , in vestigations of selected landslides. It is rare for
Japan. Kobe with more than one million people all landslides in a study area to be examined in
has been struck several times by slope failures the field, owing to difficulties of access or the
and debris flows caused by heavy rainfalls in the time and effort (cost) required. Consequently,
Rokko Mountains (Aniya 1968). In the U.S. this selected investigation is regarded as equivalent
problem has been particularly serious in areas to sampling. Since accessibility is a primary
such a s the Appalachians and California (Fisher, factor determining the selection of landslides in
Fanaff, and Picking 1968; Woodruff 1971: Wil- rugged mountains, sampling is rarely random,
liams and Guy 1973; Campbell 1975: Bogucki but biased. For this reason the terrain parame-
1976; Scott 1981). Consequently, an effort has ters identified on the basis of field work alone
been made to produce landslide-susceptibility o r as contributing to landsliding may not be cor-
slope-stability m a p s ( B r a b b , Pampeyan, and rect. To avoid such a situation, statistical treat-
Ronilla 1972; Pomeroy and Davies 1975; Lessing ment of data taken from all landslides in the
and Erwin 1977; Nilsen et al. 1979). This kind study area seems desirable. In this respect, re-
of map is essential both for planning residential mote sensing techniques, especially aerial pho-
developments in hilly and mountainous areas tographic interpretation, provide a convenient
and for watershed management. means whereby all landslides can be examined

102
 Landslide Susceptibility Mapping I03

(Poole 1969). The comprehensive data set re- highest point, a 2,529-meter peak, is located at
quired for the selection of critical terrain param- the northwestern corner of the basin; the lowest
eters can be generated for all landslides in the point occurs at the lake level only 7 km to the
study area from remote sensing data in combi- east of the highest peak at an altitude of 440 m.
nation with good topographic maps and other In accordance with the high relief of almost
supplementary data. 2,200 m, the hillslopes are generally very steep,
The aims of this study are (1) to present a with an average gradient of 42".
method that employs aerial photographs and sta- The bedrock is composed mainly of in-
tistical analyses to define terrain parameters and terbedded sandstone, shale, and mudstone of
attributes (classes of parameter) contributing to Cretaceous and Paleogene ages. These strata
landsliding and (2) to introduce a procedure for strike generally north-south and dip to the west
mapping landslide susceptibility. The area at angles ranging from 30" to almost 90". Owing
chosen for this study is the Amahata River to the proximity of a large tectonic zone to the
basin, located about 130 km west of Tokyo (Fig. east, however, overturning and contortion of the
1). This basin includes the Inamata Valley strata are extensive, and the rocks are often
studied previously for similar purposes (Aniya fragmented and weathered to such a degree that
and Endo 1980). The area is covered with black flaky shale can be crumbled in one's fingers.
and white aerial photographs at 1:20,000 scale The depth of regolith ranges up to 100 cm but
taken every five years since 1964 by the Forest is usually less than 50 cm. Soil types are pod-
Agency, Ministry of Agriculture, Forestry and zolic above elevations of 1,500 t o 1,800 m,
Fisheries. Other coverage includes color pho- whereas brown forest soils occur at lower alti-
tographs taken in 1976 by the Land Agency. Uti- tudes. These soils often lack A and/or B hori-
lizing one of the coverages, topographic sheets zons, suggesting active soil movement near the
of 1:10,000 scale with a 10-meter contour in- surface.
terval have been produced. The vegetation is predominantly deciduous
below the 1,500- 1,800 m zone, whereas conifers
Study Area dominate at higher elevations. Large areas have
been logged and are now under silviculture of
The Amahata River basin is a small branch of Japanese pines and cedars. Forestry still re-
the Fuji River system emptying into Suruga Bay mains one of the principal industries in this area.
(Fig. 1). The drainage area is about 97 km' above The climatic record kept for 1959-72 at Ama-
the dam located near the confluence and extends hata village located by the lake shows that the
about 12 km in the north-south direction. The average annual precipitation, occurring mostly
as rainfall, is about 2,200 mm, with a maximum
of 2,614 mm and a minimum of 1,699 mm (Kofu
Regional Meteorological Observatory 1965-72).
STUDY AREA The average annual temperature is 13.1"C.

0r
- M t . Fuji
A
Judging from daily maximum and minimum tem-
peratures, freeze-thaw processes appear to be
active from December through at least the

\
20km 3776m middle of March.
Arnahata River Currently there are in the study area more
Basin than 1,000 landslides, ranging in area from a few
9 7 km2
hundred to several hundred thousand square
meters. The most common type is the planar
slide (Crozier 1973) or debris slide-debris ava-
lanche (Varnes 1978), which has an area up to a
few thousand square meters and an average
depth of 1 to 3 m (Fig. 2). Many of these land-
slides were probably triggered by daily rainfalls
exceeding 300 mm brought by typhoons, partic-
ularly when there has been antecedent rainfall
(Kofu Regional Meteorological Observatory
Figure 1 . Location of study area. 1965-1972).
104 Aniya

Figure 2. Stereopair of ground photos showing a typical landslide found in the study area. Bedrock: mudstone.
sandstone. and shale interbedded. Aspect: 250". Length: 141 ni. Average width at top part: 17 m. Average
gradient: 43". Area: 2.100 m2. Failed volume: 3,500 m' (left one).

Methods fined parameters were produced. and a score

indicating a degree of importance to landsliding
Flo-.v of Study was computed for each class of the defined pa-
rameters. Evaluation of the landslide suscepti-
Aerial photographs were the principal source bility of a slope segment was based on the total
of landslide information, which was then sup- score derived from superimposing these maps.
plemented and corrected by extensive field in-
vestigations (Fig. 3 ) . Landslide distribution
maps were produced from aerial photographs for Delineation of Landslides
landslide measurements. I n order to collect data
o n t e r r a i n c h a r a c t e r i s t i c s , a grid of 2.5 c m Aerial photographs taken at three dates, 1969,
(250 m on the ground) with x and 4' axes aligned 1974, and 1979, were used to map landslides and
east-west and north-south, respectively, was laid were augmented by field investigations. Since all
out. and the terrain characteristics were mea- sets of photographs were taken in May or June,
sured at the grid intersections. After obtaining shortly after new leaves come out, there was no
landslide and grid data sets, iwo types of anal- inconsistency in interpreting the landslides as a
ysis, failure rate (FR) analysis and quantification result of the seasonal variation in vegetation
scaling type I1 ( Q S 11) (Hayashi 1952, 1954), cover. Color photographs taken in 1976 greatly
were used to define terrain parameters and at- enhanced interpretation of the black and white
tributes critical to landsliding. photographs.
Next, terrain classification maps of' the de- Interpreted landslides were transferred onto
 Landslide Susceptibility Mapping 105

( F i e l d Work1
Lands1id e Failure R a t e
Analysis
Distribution
Quantification
Photographs Scaling Type II

Scores o f
-+ S e l e c t e d
t
Attributes Overlay f o r Landslide
Terra in + Each Group o f ---+ Susceptibility
Te r r a i n Attributes Map
Classification -

the topographic maps by a stereo zoom transfer mind, the ten terrain parameters listed in Table
scope. After editing t h e location and delin- 1 were measured at both the landslide sites and
eating the landslides, distribution maps for 1969. the grid intersections, and each was divided into
1974, and 1979 were finalized. The maps for 1974 between three and ten classes to produce a total
and 1979 were then superimposed on those for of 57 attributes. Lithology and structure of the
1969 and 1974, respectively, in order to identify bedrock were not included because, after exten-
the landslides that occurred during the 1969-74 sive field work, there appeared to be no signif-
and 1974-79 periods. Selected terrain parame- icant differences within the study area. Another
ters were subsequently measured at these land- important factor, soil properties, could not be
slides and at the grid intersection points. The incorporated owing t o lack of d a t a . For e x -
number of landslides is 458 for 1969-74 and 276 ample, soil thickness varied so much that inter-
for 1974-79; the number of grid points is 1,562. polation f r o m field d a t a would entail g r e a t
Some grid points fell on landslide sites. uncertainty. Although water catchment area is
important in initiating rain-induced landslides.
this was not included owing to the large mea-
Terrain Parameters and Data Collection s u r e m e n t e r r o r . S o m e o f t h e influence of'
drainage area may be included in the distance-
There are many parameters that are consid- to-ridge variable ( # 5 ) . Rainfall records were also
ered t o b e important t o landsliding. Among omitted because there are only one o r two rain-
them, Nilsen et al. (1979) listed the following as fall recording stations in the entire basin. Again,
the most important: ( 1 ) nature of underlying precipitation might be expected to covary with
bedrock o r unconsolidated deposits; (2) the other variables in Table I , specifically elevation
angle of slope; (3) rainfall; and (4) the presence and aspect.
of older deposits. Nilsen et al. also cited 15 other
parameters; but notably missing from their- fist
were aspect and slope morphology. Beaty (1956) Parameter Measurements and
cited slope exposure (aspect) as an important Their Significance
factor for landsliding.
It is of critical importance for the analyses em- The terrain parameters included in this study
ployed in this study to include parameters that are as follows:
have played a significant role in landsliding and Slope grudient of a landslide was measured by
that can either be measured on the contour map dividing the elevation difference between the
o r identified o n aerial photographs o r on the top and the bottom of the landslide by the
ground in sufficient detail for its areal distribu- horizontal distance. At each grid intersection
tion to be mapped. With these requirements in the number of contours crossing a 5 mm line
~

Slope gradient tdegieeil 0-25 25-30 30-35

Elevation lmt 11-hOO .500-XO(I 800- I .ooo
Aspect N NE E
Drainage h;t\in order 1 >
Distance to ridge lm) 0-20 20-40 40-60
Distance to valley
floor lrn) 0-50 50- 150 150-300
Slope profile form Convex Concave Straight
Slope plan form Divergent Con\ er- Planar
gent
Proximity to break Jurt above Just belou Distant
of slope 1120 rnl (<:XI nil ( i Z 0 rn)
Vegetation Natural Secondary Planted Seconda try Planted Bush Gra\s
forest forest fore\t forest fo I-est (logged1 clogged1
( > 6 ml l'.-.6 m ) 12-6 nil (2-6 ml
__--- ___________

Table 2. Agreement Between Field- and Map-Mea\ured Parameter 5

Correlation coefficienrh
Parameter Regreskin" 01' agreement

Landslide area ( 100 m?) Y = 3.49 + 0.XhX I' =

0.95
Landslide length (m) Y = 2.87 + 0.95x I' 0.91
Aberage gradient (degree$) Y = 23.28 + 0.45x I' = 0.66
Aspect (degree5) Y = 8.27 + 0.99x I' = 0.99
Slope profile form 31 percent
Slope plan form 47 percent
Y is field nieesured a n d A' is map tiirasui-ed.
" The \ample size i-ange\ fl.om 54 to i 7 : tnininiit~n111. = 52. I' = 0 . W . significant ;it 99 pet-cent lcvel
 Landslide Susceptibility Mapping 107

drawn perpendicular t o t h e c o n t o u r s was tially as natural forest, secondary forest, man-

counted and then converted to a gradient. made forest, or logged area. The forest types
Elevrition of a landslide was measured at its top. were further divided into two height classes
Weather and climate conditions vary greatly (2-6 m and over 6 m), and the logged area
at different elevations, and this is reflected in into either bush-covered (height less than 2 m)
differences of soils and vegetation. or grass. In the Amdhata region it is estimated
Aspect w a s r e p r e s e n t e d by eight a z i m u t h a l that 10 to 15 years after logging and planting,
classes. Like elevation, this parameter influ- the regolith-holding power of the root system
ences local climate, including rainfall distri- reaches a minimum. During this time span, a
bution during storms. tree (Japanese pine or cedar) is estimated to
Drainage basin order was determined by the grow to a height of about 6 m on the average.
Strahler (1957) method o n t h e m a p of t h e
drainage system compiled from aerial photo- Accuracy of Map Measurements
graphs and field work. A first-order stream
was defined as the smallest channel with pe- When measurements taken from topographic
rennial water flow. maps are subjected to statistical analyses, the
Distance to ridge was measured as the distance accuracy and reliability of the measurements be-
along the fall line from the top of the landslide come of great concern. Those parameters that
or the grid intersection to the ridge crest. This were measured in the field of landslide sites and
variable is indicative of water catchment area. also could be measured on topographic maps
Distance to vnlley ,floor was measured along the were examined to evaluate measurement accu-
fall line from the top of the landslide or the racies.
grid intersection to the valley floor of the first- They included landslide area, length, average
order stream. This variable is related to the gradient, aspect, and slope morphology. Table 2
size of slope that can collapse. indicates that the map measurements are gcn-
Slope profile and plun form for landslides was erally very good. Discrepancies in the area and
measured by drawing imaginary contours in length measurements could have stemmed from
an effort to reconstruct the original surface of recovery of vegetation, enlargement of the scar,
the top part of the landslide before failure. and/or problems in delineating the landslide area
Slope profile form at the grid intersections on the aerial photographs and transferring it to
was determined from the spacing of the three the map. As for the average gradient, two most
contours closest to the intersection, and slope likely factors giving rise to the discrepancies be-
plan form was determined from the curvature tween the map and the field measurements are
within a distance of 5 mm. Slope morphology the contour interval of 10 m and the delineation
is one of the most important influences on land- of landslide area.
sliding because it affects the movement of sur- The agreement between the map and field data
ficial materials, surficial runoff, and soil water for aspect is outstanding, whereas for slope mor-
(Anderson and Burt 1978: Speight 1980). The phology it is poor. Slope morphology was diffi-
plan form is especially important as it dictates cult to determine in the field because a sound
the area of converging or diverging water flow. perspective of the whole landslide in relation to
Proximity to b r e d of slopc~was determined by the surrounding terrain could not usually be ob-
first identifying b r e a k s of slope (Savigear tained. Consequently, it was felt that map mea-
1965) on the aerial photographs and by then surements are more reliable than field observa-
transferring these features onto the contour t i o n s f o r s l o p e morphology. B a s e d o n t h e
maps. If the top of a landslide was located preceding evaluations, it is concluded that the
within about 2 mm (20 m on the ground) of a map measurements are generally accurate and
break of slope, it was regarded as proximal reliable enough to be employed in the statistical
and classified as either above o r below the analyses.
break. Otherwise it was classified as distant.
Empirical evidence suggests that many land-
slides are located near the break of slope (Wil- Analyses
liams and Guy 1973).
Vegetation was mapped from the 1969, 1974, Two analyses were employed in order to avoid
and 1979 aerial photographs and classified ini- any bias that might be inherent in either one.
I08 Aniya

These were FR analysis and Q-S 11. The first nonlandslide. These two groups cannot readily
one involves computation of the relative fre- be distinguished with respect to parameters XI
quency of each attribute in the landslide and grid and X,. However, if the two groups are pro-
data, and division of each relative frequency for jected on the Z axis, they can be clearly sepa-
the landslide data by that for the grid data. Sup- rated. Location on the Z axis of any point can
pose that F,,,,,represents a frequency in class n be generally written as
of parameter m. then the relative frequency is
given by z = Po + PIX, + P J 2 . (1)
If a point P belongs tojThclass of XI and ithclass
of X,, Equation ( I ) becomes
z,( P ' ) = Po + PJ,] + PJ,,. (2)
What Q-S 11 does is to compute scores of Po,
PIX,,, and P2X2,,where j and i vary from 1 to n
The failure rate is depending upon the number of classes in a pa-
rameter, so that t h e separation of these t w o
groups is maximized. This concept, of course,
can be expanded into rn dimensions ( m param-
Because the grid data supposedly represent the eters). As a large score contributes more than a
terrain characteristics of the whole basin, the small one to the value of Z , it seems more im-
failure rate indicates importance of that attribute portant than a small score. Care must be taken,
to landsliding relative to the basin characteris- however, in interpreting the meanings of these
tics. A value greater than I indicates that the s c o r e s because t h e y may vary considerably
particular attribute contributes to landsliding, when different sizes and kinds of samples are
whereas a value smaller than I signifies that the used.
attribute inhibits landsliding.
The second method. Q-S 11, is essentially the Results
same as discriminant analysis, the only differ-
ence being that Q-S I1 can accommodate quali- Failure Rate Analysis
tative data. Quantification is achieved by using
frequencies as input data. In Figure 4, group A In this analysis all grid data, including those
may be regarded as landslide and group B as that fell on landslide sites, were utilized. Figure
5 illustrates the relative frequency and failure
rate of each attribute. The slope-gradient rela-
tive-frequency graph indicate that both the study
area and the landslides are dominated by slopes
in the 40-45" class. The failure rate graph shows
that this class is most prone to landsliding. With
respect to elevation, a zone of 1,000- 1,600 m is
most susceptible t o landsliding, possibly re-
flecting the precipitation pattern. As for aspect,
east-, south-, and west-facing slopes appear sus-
ceptible, whereas north-facing slopes are fairly
safe. This is because north-facing slopes are in
the rain-shadow when typhoons bring rain.
Drainage basin order does not appear to have a
significant influence on landsliding. The dis-
tance-to-ridge graph indicates that the failure
rate is the greatest in the 60-80 m class, whereas
the failure rate of distance to valley floor reaches
a maximum in the SO- 150 m class. These results
XI i may suggest that there is an optimum length
Figure 4. Concept o f a discriminant function (quan- of slopes for landsliding, in this c a s e about
tification scaling type 11). modified after Davis (1973). 100-200 m. T h e slope-profile-form graph re-
50 1 Slope
- Grid Data
Landslide Data
- -
-........ 1969-1374
1974-1979 Elevation
Aspect
20

... -

I r l 7 1 1 1 1 1 1 1

0-25 -35 -45 -55 55- 0-6W -1000 -1403 -1800 -22002200- N NE E SE S SW W NW

40

30

20
Distance to Ridge

430 1

20
4
,
Distance l o
Valley Floor
zo,
Slope Plan
Form

10 I0

0 0
0-20 -60 -100 100- 0-50-150-300 300-

I-
..___._
_____.----

Figure 5. Results of failure rate analysis.

110 Aniya

Table 3. Top Six Attributes by Failtire Rate Analysis

~

~ . ~

~ __ ___ ____
Fail ti re Failure
Rank 1969-1974 rate 1974- I979 rate
- _ _ _ _ _ ~ ~ _ _
I Pianled forest (2-6 ni)Lit' 2.90 Bush" 3.44
2 Bush,' 2.11 Planted forest (1-6 2.62
3 Convergent,' I .60 South" 1.80
i Eas1 1.57 40-45"" 1.70
i 4-45".' I .57 West 1.51
il So tit h" I .45 Convei-genl" I .47

veals that straight slopes are more prone than rhows the general importance of vegetation. rlope
either convex o r concave ones, and the slope- gradient, aspect and rlope plan form.
plan-form graph indicates that the convergent
plan form is the most favorable for landsliding.
Proximity to break of slope appears a n impor- Q - S 11 Analysis
tant factor. Whether failures occur just below or
just above probably depends upon local condi- The computer program of Q-S I1 has been de-
tions of underlying rocks. Since vegetation veloped for the Japanese version of SPSS. and
changes constantly and drastically when logged. it has been implemented at the computer center
vegetation data at grid intersections were taken of the University of Tsukuba. Analyses were
for 1974 and 1979 to compare with landslide data performed changing the size and set of grid data
for 1969-74 and 1974-79, respectively. The re- in o r d e r to e x a m i n e t h e effect o n a t t r i b u t e
sults clearly indicate the effects oflogging on land- scores. For 1969-74 one-third of the grid data
sliding, ;IS the hush and planted forest (2-6m) was sampled so that the number of grid points
.es have exceptionally high failure rates. Table became approximately the same as the number
3 summarizes the results of the FK analysi.; and of landslides. For 1974-79 the grid data were

Table 4. Top Attributes and Pat-anieters Revealed by CZ-S 11

~_

I Y74- 1979

K.inh Simpling I ' hmpling II Smipling I l l Sampling I" Sampling 11 Sampling I l l

Planted Planted Hmh' Planted

fol-e\t fc lire\ t so1.crt
12-6 in)' (Lf, 1111' 1 2 - 6 nil'
t3il\hL Hu\h' Planted Hlhh'
fore\t
( 2 - 6 nil'
' ..
(,Id\\ h l e ation
~ South Jubl below
10-000 111) hreak i n
\lOpe
('onvergenr' ('onvergent' Lh\tance t o 40-35"
ridge
(60-80 n i l
i()-ii Ju\t bclow Elevation Con\ el-gent
hi-cah in ( 1400-
\lope 1600 n i l
South E.;I\I 40-45"L Elebation
l1-100-
I600 mi
Vegetation Vcgetiition Vege t at ion Slope
giadient
SlLlpe Slope Elevation Vegetation
gi-;idient giudient
Slopc plan Slope plan Slope Elevation
form form gradient
 Landslide Susceptibility Mapping 111

Table 5. Critical Terrain Parameters and Attributes Defined by FR Analysis and Q-S 11
- -

Occurrence period Terrain parameter Attribute

196Y- 1974 Vegetation Planted forest (2-6 m)

Bush
Slope plan form Convergent
Break of slope Just below
Aspect East

1974- 1979 Vegetation Bush

Planted forest (2-6 m)
Slope plan form Convergent
Slope gradient 40-45"
Aspect South
Break of ,lope Just below
- ______

reduced to one-fifth for the same reason. Table The terrain parameters and attributes identified
4 lists these data sets along with the results. Col- for the two periods are similar, probably im-
umns of the top attributes and parameters illus- plying similar circumstances when landslides
trate the problem inherent in this analysis: the occurred. This fact suggests that based on the
change of ranks according to the different sam- results of these two analyses, it might be pos-
ples. Consequently, it seems desirable to regard sible t o m a p t h e landslide susceptibility of
these attributes as a group without ranking for slopes. This would show the area of danger, in
practical purposes. N o n e t h e l e s s , t h e results a relative sense, when daily rainfall exceeds
point t o the important terrain parameters: veg- about 300 mm.
etation, slope gradient, aspect, slope plan form,
elevation, and break of slope. These parameters
are almost identical to those identified by F R Landslide-Susceptibility Mapping
analysis.
The results of these two analyses are sum- Among the six terrain parameters judged sig-
marized in Table 5, which lists the critical pa- nificant in contributing to landslides. slope
rameters and attributes identified by the two gradient. aspect, and plan form can be regarded
analyses. In addition to these, elevation appears as basic components in the landslide-suscepti-
to be an important factor, judging by the partial bility map. Because vegetation changes con-
correlation coefficient for 1974-79 data in Q-S stantly and sometimes drastically, elevation data
I1 analysis. In summary, vegetation, slope gra- can be readily obtained from contour maps but
dient, aspect, slope plan form, break of slope, they usually reflect rainfall distribution, and
and elevation seem to be most important terrain break of slope has a linear rather than an areal
parameters among the ten selected in this study. distribution. Therefore, it seems practical to pre-

Figure 6 . Cumulative frequencies of landslide and nonlandslide groups plotted on the Z axis
I12 Aniya

pare the landslide-susceptibility map from these basic map. By excluding vegetation, the map
three basic terrain parameters, and where nec- can be used in forest management, such as in
essary, vegetation. elevation, and/or break-of- planning logging operations.
slope information may be overlaid to modify the The method of judging the susceptibility of a

Convergent

Planar

Divergent

High
+ Unstable
Moderate Low
- Stable
Least

Figure 7. The landslide-susceptibility classes for all combinations of the slope gradient, aspect. and plan form
attributes.
 Landslide Susceptibility Mapping 113

slope t o landsliding i s of great importance and cation of t h e relative weakness of slopes to

interest. Brabb, Pampeyan, and Bonilla (1972) heavy rain, in that slopes ciassified as having
assigned a numeric value to an attribute that was high susceptibility merely have a higher chance
rather intuitively derived, and they judged sus- of failing than those classified into other classes.
ceptibility by a total value. In another method, Manual compilation of the susceptibility map is
susceptibility is determined by the combination time consuming, which was one reason for lim-
of attributes without assigning values (Van Horn iting the number of parameters to three. In the
1972; Nilsen et al. 1979). In this study it was future, as Newman, Paradis, and Brabb (1978)
decided to employ attribute scores computed by have shown, the use of a drum scanner, digital
Q-S 11, and the program was run for the three computer, and an output device that produces
parameters after combining the 1969-74 and hard c o p y is d e s i r a b l e for p r o d u c i n g m a p s
1974-79 data. Although the scores changed as quickly and effectively from many terrain pa-
a result of lumping the data, the magnitude of rameters.
the change was generally not large enough to
cause a point to switch categories (i.e., from
landslide to nonlandslide or vice versa). Figure Summary and Conclusion
6 shows plots of cumulative frequencies, with
the landslide group starting from the smallest Utilizing the Amahata River basin as a test
s c o r e a n d t h e nonlandslide g r o u p f r o m t h e area, methods of studying terrain parameters
largest score. If the curves of these two groups and attributes that contribute to landsliding and
do not intersect, separation is perfect; however,
of mapping landslide susceptibility have been
they do intersect in this case and so misclassi-
presented. After landslide distribution maps
fications will occur. As for a boundary score that had been prepared from aerial photographs and
would separate the two groups, it appears logical field work had been completed, ten important
to choose the score at which the two curves terrain parameters were measured at landslides
intersect-that is, -0.12-because it is as im- and grid intersections. These data were then
portant to locate correctly stable slopes as un- subjected to failure rate analysis and quantifi-
stable ones. Consequently, points (slope seg- cation scaling type I1 to define the most critical
ments) whose total score of the attributes was terrain parameters and attributes. These include
equal to o r less than -0.12 were classified in slope gradient, aspect, slope plan form, break of
the landslide group (unstable), whereas those slope, elevation, and vegetation.
with scores greater than --0.12 were placed in Based on these analyses, a landslide-suscep-
the nonlandslide group (stable). The overall ex- tibility m a p w a s produced b y combining the
pected accuracy of this judgment can be read three basic terrain parameters: slope gradient,
from Figure 6 as about 60 percent. In order to aspect, and plan form. Classes of these param-
differentiate between very unstable and margin- eters were assigned scores computed by quan-
ally unstable and between very stable and mar- tification scaling. A slope segment having a cer-
ginally stable, scores at which the accuracy of tain class combination of these three parameters
decision would be about 80 percent were se- had a total score on the basis of which its sus-
lected to divide each category into two classes, ceptibility to landsliding was judged as high.
creating four classes of susceptibility: high, moderate, low, or least. If desired, maps of ter-
moderate, low, and least. rain parameters such as break of slope, eleva-
In order to facilitate a rapid manual compila- tion, and/or vegetation c a n b e laid o v e r this
tion of the susceptibility map, attributes with basic map to modify the susceptibility class of
similar s c o r e s w e r e grouped and assigned a the slope.
common score based on the weighted mean. A
matrix of susceptibility was constructed for each
combination of the attribute group (Fig. 7). The
loss of information caused by grouping was min- Acknowledgments
imal compared to the savings in time and effort
in making the final map. A large part of this study was supported by the Kofu
Landslide susceptibility is not a deterministic Sabo Office, Ministry of Construction, Japan. The au-
thor is indebted to Mr. Endo for his assistance at the
classification t h a t will predict exactly w h e n early stage of this study. He is also grateful to the
(with how much rainfall) and where a landslide editors and anonymous reviewers who helped improve
will occur. Rather, it is a probabilistic classifi- the manuscript.
114 Aniya

References 3, ed. D. K. Coates. pp. 245-54. Boulder, Colo.:

Geological Society of America.
Newman, E. B.; Paradis, A. B.; and Brabb, E. E.
Anderson, M. G., and Burt, T. P. 1978. The role of 197X. Feasibility and cost of using a computer to
topography in controlling throughflow generation. prepare landslide susceptibility maps of the San
Etrrrh Srrrhftrc.rProcesses 3:331 -44. Francisco Bay region, California. U . S . Goologictrl
Aniya, M. 1968. Landslides and mudflow in the Su- Surivy Brilletirr 1443, Washington, D.C.
niiyoshi Valley. Rokko Mountains, caused by the Nilsen, T. H., and Brabb, E. E. 1973. Current slope-
concentrated heavy rain in July. 1967. The Hirnrcrn stability studies in the San Francisco Bay region.
G w g r ( i p / i y (Jinhrrn Chiri) 20:454-70. Jorrrncil of R e s e a r c h . U . S . Geo/ogic.crl . S i r n . a y
Aniya, M. and Endo, S. 1980. A study of terrain at- 1:431-37.
tributes contributing to landsliding. 24th fnterncr- Nilsen, T. H.; Wright, R . H.; Vlasic, T. C.; and
tiorrrrl Grwgraphic~nlCongress, Mrrin Se.csion Ah- Spangle, W. E. 1979. Reltcti\~es l o p e .strthility a n d
Sfr(/"t 1:328-29. land-rise plrrnning in thc Son Franc
Beaty, C. B. 1956. Landslides and slope exposure. gion, Californicr. U.S. Geological Survey profes-
Jorrr-nril o f G'coloCyy64:70-74. sional paper 944. Washington, D.C.
Bogucki. D . J . 1976. Debris slides in the M t . Le Pomeroy, J. S., and Davies, W. E. 1975. M a p of .srr.s-
C o n t e area. G r e a t S m o k y Mountains National ceptihility t o lnndsliding. A l l e g h e n y Coririry,
Park. U . S . A . Geogrt~fi.sk(rAnncrler 58A:179-91, Pennsylvtrniu. U.S. Geological Survey miscella-
Brabb, E. E.; Pampeyan, E. H.; and Bonilla, M. B. neous field studies map MF-bsSB, scale 1 :50,000.
1972. Lrrnd.rlidc .srt.sc~eptibility i n Son Mtrteo Poole, D. H. 1969. Slope failure forms: their identifi-
Corrnty, C(r/ifiirnicr. U .S. Geological Survey mis- cation, characteristics, and distribution a s de-
c e l l a n e o u s field s t u d i e s m a p MF-360. s c a l e picted by selected remote sensor returns. Pro-
1 :62,500. ceedings of the 6th I n t r m e i f i o n a l Sytnposirrtn o n
Campbell, R. H. 1975. Soil slips. dchris . f r o ~ , . scrnd Reniote Serising of Environment 2:927-65.
r.triii.ttornr in / l i e SanrLr Monicu Monntcritrs trnd Savigear, R. A. C. 1965. A technique of morpholog-
v i c i n i t y . .sorrrker-n Ctrlifornitr. If . S . Geological ical mapping. Annu/.r of the A.s.soc.itrtion of'Anrrr-
Survey profe\sional paper 85 I . Washington. D.C. ;(,tin Geogrripher.r 5215 14-32.
Crozier. M. J. 1973. Techniques for the morphometric Scott, R. C. 1981. Meteorological aspects of major.
analysis of landslips. Zritschrjfifiir Gaoniorp/rol- debris avalanche occurrences in the Southern Ap-
o g i r 17:78-101. palachians. Vir-giniu Geojirripher 14:27-37.
Davis, J. C. 1973. Slcrti.stic..s m d dutcr rrntrlysis in ,gc- Speight, J . G. 1980. The role of topography in con-
ology. New York: John Wiley. trolling throughflow generation: a discussion.
Fisher, S. P. ; Fanaff, A. S . ; and Picking, L. W. Ear/ii Surfrzce Proc~esses5 : 187-91.
1968. Landslides o f southern Ohio. The Ohio Strahler, A. N. 1957. Quantitative analysis of wa-
Jorrmrrl of S c i e ~ e68:65-80. t e r s h e d g e o m o r p h ol og y. Trri n s a c,tio n s of t I i
(2

Hayashi, C. 1953. O n the prediction of phenomena Atnericrrn Geophysical Union 38913-20.

from qualitstive d a t a and the quantification of Van Horn, R. 1972. RelrrtiivJ slope .st(rhi/ity n i ~ pof
qualitative data from the mathematico-statistical the Sugtrr Horrse yuudrtrnglc, Sol/ Luke County,
point of view. Anncils of'thc l n s t i t i r t e ofStrrti.stic.ci/ Urtrh. ti . S . Geological Survey miscellaneous geo-
.Muthrniutic~.t?:69-98. logical map 1-766-E, scale 1 :24.000.
Varnes, D. J . 1978. Slope movement-types and pro-
-. 1954. Multidimensional quantification-with
cesses. In Landslide: crntrlysis rind w n t r o l , ed.
the applications t o analysis of social phenomena. R . L. S c h u s t e r a n d R . J. K r i e k , p p . 11-33.
Anncrls o f t f i r ftisritrire ofS/rrtis/icnl Mntlrc~tnatics
Highway Research Board special report 176.
5 : 121 -43. Washington, D.C.
Kofu Regional Meteorological Observatory. 1965- Williams, C . P., and Guy, H. P. 1973. B-o.Piotierl and
1972. Mtrnthly t ~ i c ~ t r o r - ~ ~ l o g i creport
~ r r r l of Yrimu- depositional rrspec'ts of hurriccrne Ccrtnille in Vir-
tiushi. Kofu. Yamanashi: Kofu Branch, Japan Me- ginia, 1976. U.S. Geological Survey professional
teorological Association. paper 804. Washington, D.C.
Lessing, P . , and Erwin, P. B. 1977. Landslides in Woodruff, J. F. 1971. Debris avalanches as an ero-
West Virginia. In Lnndslidc,s: Grologic.~lSociety sional a g e n t in t h e A p p a l a c h i a n M o u n t a i n s .
of Anirricrr r c ~ i ~ i ien~engineering
~s g e o l o g y . Vol. Joirrncrl of Geogrupliy 70:399-406.