Ž .Geomorphology 31 1999 181–216
Landslide hazard evaluation: a review of current techniques and
their application in a multi-scale study, Central Italy
Fausto Guzzetti a,)
, Alberto Carrara b
, Mauro Cardinali a
, Paola Reichenbach a
a
Inst. di Ricerca per la Protezione Idrogeolgica nell’Italia Centrale, CNR, Via della Madonna Alta 126, 06128 Perugia, Italy
b
Centro di studio per L’informatica edi sistemi de telecommunicasioni, CNR, Viale Risiorgimento 2, 40136, Bologna, Italy
Received 30 September 1996; received in revised form 11 March 1997; accepted 1 May 1997
Abstract
In recent years, growing population and expansion of settlements and life-lines over hazardous areas have largely
increased the impact of natural disasters both in industrialized and developing countries. Third world countries have
difficulty meeting the high costs of controlling natural hazards through major engineering works and rational land-use
planning. Industrialized societies are increasingly reluctant to invest money in structural measures that can reduce natural
risks. Hence, the new issue is to implement warning systems and land utilization regulations aimed at minimizing the loss of
lives and property without investing in long-term, costly projects of ground stabilization. Government and research
institutions worldwide have long attempted to assess landslide hazard and risks and to portray its spatial distribution in maps.
Several different methods for assessing landslide hazard were proposed or implemented. The reliability of these maps and
the criteria behind these hazard evaluations are ill-formalized or poorly documented. Geomorphological information remains
largely descriptive and subjective. It is, hence, somewhat unsuitable to engineers, policy-makers or developers when
planning land resources and mitigating the effects of geological hazards. In the Umbria and Marche Regions of Central Italy,
attempts at testing the proficiency and limitations of multivariate statistical techniques and of different methodologies for
dividing the territory into suitable areas for landslide hazard assessment have been completed, or are in progress, at various
scales. These experiments showed that, despite the operational and conceptual limitations, landslide hazard assessment may
indeed constitute a suitable, cost-effective aid to land-use planning. Within this framework, engineering geomorphology may
play a renewed role in assessing areas at high landslide hazard, and helping mitigate the associated risk. q 1999 Elsevier
Science B.V. All rights reserved.
Keywords: landslide; natural hazard; hazard evaluation; statistical modelling
1. Introduction
A hundred years ago, the world population totalled
1.1 billion, and about 5% of people lived in
cities. Today, the population has risen to 5.3 billion
)
Corresponding author. Tel.: q0039-75-5054943; fax: q0039-
75-5051325.
Ž .E-mail address: F.Guzzetti@irpi.pg.cnr.it F. Guzzetti .
and approximately 45% of it is concentrated in urban
areas. The most explosive growth has been in the
developing world, where urban populations have
tripled in the last 30 years. Between 1950 and 1995,
the number of cities with population of more than
one million increased sixfold in the third world
Ž .Helmore, 1996 .
The population growth and the expansion of settlements
and life-lines over hazardous areas are in-
0169-555Xr99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.
Ž .PII: S0169-555X 99 00078-1
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216182
creasing the impact of natural disasters both in the
Ždeveloped and developing world Rosenfeld, 1994;
.Alexander, 1995 . In many countries, the economic
losses and casualties due to landslides are greater
than commonly recognized and generate a yearly
loss of property larger than that from any other
natural disaster, including earthquakes, floods and
Žwindstorms Schuster and Fleming, 1986; Alexander,
1989; Swanston and Schuster, 1989; Olshansky,
.1990; Schuster, 1995a; Glade, 1998 . Casualties due
to slope failures are larger in the developing countries,
whereas economic losses are more severe in the
industrialized world. Both may be increasing because
of the higher value of endangered structures and the
greater number of people potentially involved
Ž .Schuster and Fleming, 1986 .
Third world countries have always had difficulty
affording the high costs involved in controlling natural
hazards through major engineering works and
rational land-use planning. Owing to the economic
recession, many industrialized societies are reluctant
to invest in structural measures to reduce natural
risks. Economic and social considerations suggest
that, even if the recurrence of natural disasters remains
constant — and it may not be the case —
damage caused by catastrophic events is too costly
even for industrialized societies. In other words,
natural catastrophes occur with higher frequency than
our ability to recover from previous events.
The recent trend is towards the development of
warning systems and land utilization regulations
aimed at minimizing the loss of lives and property
damage without investing in long-term, costly proŽjects
of slope stabilization U.S. Geological Survey,
1982; Kockelman, 1986; Schuster and Fleming, 1986;
.IDNHR, 1987; UNDRO, 1991; Schuster, 1995b .
Despite the largely acknowledged need for landslide
planning strategies, few attempts have been made to
introduce landslide hazard considerations in building
Žcodes or civil protection strategies Brabb and Har.
Žrod, 1989 . Notable examples are in France Humbert,
1976, 1977; Antoine, 1977; Godefroy and
.Humbert, 1983; Leroi, 1996 , the San Francisco Bay
Žregion Nilsen and Brabb, 1977; Nilsen et al., 1979;
. ŽBrabb, 1995 and the Los Angeles area IDNHR,
. Ž .1987 in the United States, in Japan IDNHR, 1987 ,
Ž . ŽSweden Ahlberg et al., 1988 and Hong Kong Brand
.et al., 1982; Brand, 1988; Hansen et al., 1995 .
Within this framework, earth sciences, and geomorphology
in particular, may play a relevant role in
assessing areas at high landslide hazard and in helping
to mitigate the associated risk, providing a valuable
aid to a sustainable progress. Tools for handling
Ž .and analyzing spatial data i.e., GIS may facilitate
the application of quantitative techniques in landslide
hazard assessment and mapping.
In this paper, we first introduce the general assumptions,
the mapping unit types, and the most
commonly used hazard evaluation methods. We then
discuss the experience gained from the application of
GIS-based models of hazard and risk due to slopefailures
over test areas in Central Italy, ranging in
size from some tens to some thousands of square
kilometers, and outline the potentials and pitfalls of
the approach. In the light of the results obtained, data
quality, type of terrain-unit and statistical models are
critically evaluated. Lastly, general comments on
data collection, model production and information
transfer are addressed.
2. Definition of landslide hazard
Physical scientists define a natural hazard either
as the probability that a reasonably stable condition
Ž .may change abruptly Scheidegger, 1994 , or as the
probability of occurrence of a potentially damaging
phenomenon within a given area and in a given
Ž .period of time Varnes et al., 1984 . The latter
remains the most widely accepted definition for natural
hazard and for maps portraying its distribution
Žover a region IDNHR, 1987; Einstein, 1988, 1997;
Starosolszky and Melder, 1989; Horlick-Jones et al.,
.1995; Murck et al., 1997 .
The definition incorporates the concepts of magnitude,
geographical location and time recurrence.
The first refers to the ‘‘dimension’’ or ‘‘intensity’’
of the natural phenomenon which conditions its behavior
and destructive power; the second implies the
ability to identify the place where the phenomenon
may occur; the third refers to the temporal frequency
of the event.
Traditionally, earthquake predictive models atŽtempt
to define hazard in terms of magnitude a
.measure of the energy released by a seismic event ,
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 183
affected area, and time recurrence. Ideally, they
largely fulfil the definition of hazard previously mentioned;
unfortunately, scientists are generally unable
to predict with the required accuracy where and
when an earthquake will take place and how severe
it will be. Despite the different meanings of the term
Ž .‘‘flood’’ Baker, 1994 , flood hazard evaluation essentially
consists in the temporal prediction of an
extreme hydrological event of a given magnitude
Ž .peak flow or volume , while, its location and spatial
Ž .extent potentially inundated areas are determined
from other sources of information, such as historical
records and ground morphology.
For landsliding, a conceptual confusion arises
from the use of the same term, landslide, to address
Ž .both the landslide deposit the failed mass and the
movement of slope material or of an existing landŽ
.slide mass Bosi, 1978; Cruden, 1991 . Regional
landslide predictive models generally attempt to
identify where landslides may occur over a given
region on the basis of a set of relevant environmental
characteristics. Under the assumption that slope failures
in the future will be more likely to occur under
the conditions which led to past and present slope
Žmovements Varnes et al., 1984; Carrara et al.,
.1991,1995 , these models provide information on
potentially unstable slopes. Hence they differ from
Ž .maps of landslide deposits landslide inÕentories
which consist of a catalogue of the landslide deposits
present over a region which formed within a generŽ
.ally unknown or unspecified period of time. However,
such models do not directly incorporate time
Ž Ž . Žand magnitude i.e., size Fell, 1994 , speed Cruden
. Žand Varnes, 1996 , kinetic energy Hsu, 1975; Sassa,¨
. .1988 or momentum of the failed mass , hence, they
cannot be correctly defined as hazard models.
Predictive models of landslide movement are generally
confined to single slopes where detailed
geotechnical site investigations attempt to assess
when and to what extent the slope-forming material,
frequently an existing landslide deposit, will move.
Also in this case, the term hazard would be incorrect
since the location of the phenomenon under study
derives from information acquired from other
sources.
Therefore, the application to landsliding of the
term ‘‘natural hazard’’ is difficult and somewhat
inadequate.
The wide spectrum of landslide phenomena and
the complexity and variability of their interactions
Ž .with the environment both natural and human make
the acceptance of a single definition of landslide
hazard unsuitable. For example, very large, fastŽ
.moving landslides e.g., rock avalanches are probably
the most destructive and hazardous mass movements.
Slow-moving, deep-seated failures rarely
claim lives but can cause high property damage.
Fast-moving soil-slip–debris flows triggered by intense
rainfalls are extremely destructive, causing
widespread damage and casualties. Each type of
slope movement pose different threats and may require
a separate assessment, based on distinct definitions
of landslide hazard.
Recurrence, the expected time for the repetition of
an event, is evaluated studying historical records.
Historical data however are seldom available and
difficult to obtain for single landslides or landslide
Žprone areas Guzzetti et al., 1994; Ibsen and Bruns.den,
1996 . In addition, for first-time failures
Ž .Hutchinson, 1988 recurrence is not applicable.
First-time landslides occur at or close to peak strength
values, whereas reactivations occur between peak
and residual conditions. Thus, first-time landslides
provide little information on the behavior of reactivations.
Additionally, each time a landslide occurs, the
topographic, geological and hydrological settings of
the slope change, often dramatically, giving rise to
different conditions of instability. These changes allow
geomorphologists to identify landslides and understand
mechanisms and causes of failures, but limit
their ability to forecast reactivations. Despite the lack
of consensus on the reliability and usefulness of
historic information, some investigators have attempted
the reconstruction of historical records for
single landslides or landslide prone regions. The
results appear to be somewhat encouraging and useful
for the evaluation of landslide hazard at various
Žscales Guzzetti et al., 1994; Ibsen and Brunsden,
.1996; Cruden, 1997; Evans, 1997; Glade, 1998 .
Historical records may be integrated with temporal
data derived from dendrocronology and other dating
techniques which have been used by some investigaŽtors
to date landslide deposits Stout, 1977; DeGraff
.and Agard, 1984; Trustrum and De Rose, 1988 .
Due to the conceptual and operational limitations,
most landslide hazard maps could be better defined
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216184
Ž .as landslide susceptibility maps Brabb, 1984 . Unfortunately,
terms such as susceptibility or propensity
have long been used with different meanings
ranging from landslide-deposits inventory to estimates
of landslide incidence based on the subjective
Žjudgement of the investigator Radbruch-Hall and
Varnes, 1976; Varnes et al., 1984; van Westen,
. Ž1993 . In this paper, the term landslide map or
.landslide inventory map will be used to indicate a
map portraying the distribution of deposition and
erosion areas of gravity-induced mass movements
which may vary in type, age and activity. The term
landslide hazard map will refer to a quantitatiÕe
prediction of the spatial distribution of both landslide
deposits and slopes which are likely to be site of
Ž .failures; whose movement or reactivations will take
place in a way and within a time period defined from
information that is not directly incorporated in the
model.
3. Landslide hazard mapping
Over the past 25 years, government and research
institutions have invested considerable resources in
assessing landslide hazard, and in attempting to proŽduce
maps portraying its spatial distribution lands.lide
hazard zonation . Several different methods and
techniques for evaluating landslide hazard and risk
have been proposed or tested. Inspection of the
literature reveals that a few reviews of the concepts,
principles, techniques and methodologies for landŽslide
hazard evaluation have been proposed Cotecchia,
1978; Carrara, 1983; Brabb, 1984; Crozier,
1984; Hansen, 1984; Varnes et al., 1984; Crozier,
1986; Einstein, 1988; Hartlen and Viberg, 1988;´
.Mulder, 1991; van Westen, 1993, 1994 . Surprisingly,
little work has been done on the systematic
comparison of different techniques, outlining advanŽtages
and limitations of the proposed methods Car.rara
et al., 1992, 1995; van Westen, 1993 ; or to the
critical discussion of the basic principles and underlying
assumptions of landslide hazard evaluation
ŽVarnes et al., 1984; Carrara et al., 1995; Hutchin.son,
1995 . Likewise, only few attempts have been
made to define, conceptually or operationally, landŽslide
risk Yong et al., 1977; Ahlberg et al., 1988;
Bernknopf et al., 1988; Brand, 1988; Carrara et al.,
.1991; Fell, 1994; Cruden and Fell, 1997 .
The majority of papers discuss specific attempts
at the evaluation of landslide hazard in limited areas.
Only a few authors report on long-term projects on
the evaluation of slope instability conditions, and the
related hazard and risk, over large regions. Notable
examples are represented by the work carried out in
San Mateo County, CA, by the US Geological SurŽvey
Nilsen and Brabb, 1977; Brabb et al., 1978;
.Mark, 1992; Brabb, 1995 ; by the proposal made by
the French Bureau des Recherches Geologiques et´
Minieres for a geomorphologically based evaluation`
Žof landslide hazard Humbert, 1976, 1977; Antoine,
1977; Delaunay, 1981; Godefroy and Humbert, 1983;
.Leroi, 1996 ; by the work carried out at the GeotechŽnical
Engineering Office, in Hong Kong Brand,
1988; Brand et al., 1982; Burnett et al., 1985; Hansen
.et al., 1995 ; and by the application of multivariate
statistical techniques in pilot areas of Southern and
ŽCentral Italy Carrara, 1983; Carrara et al., 1991,
.1995 .
At present, there is no agreement either on the
methods for or on the scope of producing hazard
Ž .maps Brabb, 1984; Carrara, 1989; Nieto, 1989 .
Operational and conceptual differences include: general
underlying assumptions; the type of mapping
unit selected for the investigation; and the techniques
and tools favored for the analysis and the hazard
assessment.
3.1. Basic assumptions
Despite the conflicting views among geomorphologists
and engineers, all the proposed methods are
based upon a few, widely accepted principles or
Žassumptions Varnes et al., 1984; Carrara et al.,
1991; Hutchinson and Chandler, 1991; Hutchinson,
.1995; Turner and Schuster, 1995 , namely, the fol-
lowing:
– Slope failures leave discernible morphological
features; most of them can be recognized, classified
and mapped both in the field or through remote
Žsensing, chiefly aerial photographs Rib and Liang,
1978; Varnes, 1978; Hansen, 1984; Hutchinson,
.1988; Dikau et al., 1996 .
– Landsliding is controlled by mechanical laws
that can be determined empirically, statistically or in
deterministic fashion. Conditions that cause land-
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 185
Ž .slides instability factors directly or indirectly linked
to slope failure, can be collected and used to build
Žpredictive models of landslide occurrence Dietrich
.et al., 1995 .
– The past and present are keys to the future
ŽVarnes et al., 1984; Carrara et al., 1991; Hutchin.son,
1995 . As previously mentioned, the principle,
which follows from uniformitarianism, implies that
slope failures in the future will be more likely to
occur under the conditions which led to past and
present instability. Hence, the understanding of past
failures is essential in the assessment of landslide
hazard.
– Landslide occurrence, in space or time, can be
inferred from heuristic investigations, computed
through the analysis of environmental information,
or inferred from physical models. Therefore, a territory
can be zoned into hazard classes ranked according
to different probabilities.
Ideally, evaluation of landslide hazard and its
mapping should derive from all of these assumptions.
Failure to comply to them will limit the applicability
of any hazard assessment, regardless of the
methodology used or the goal of the investigation.
Unfortunately, as will be later discussed, satisfactory
application of all of these principles proves difficult,
both operationally and conceptually.
3.2. The mapping unit
Evaluation of landslide hazard requires the preliminary
selection of a suitable mapping unit. The
term refers to a portion of the land surface which
contains a set of ground conditions which differ from
Žthe adjacent units across definable boundaries Han.sen,
1984 . At the scale of the analysis, a mapping
unit represents domain that maximises internal homogeneity
and between-units heterogeneity. Various
methods have been proposed to partition the landscape
for landslide hazard assessment and mapping
Ž .Meijerink, 1988; Carrara et al., 1995 . All methods
fall into one of the following five groups:
– grid-cells;
– terrain units;
– unique-condition units;
– slope-units; and
– topographic units.
Grid-cells, preferred by raster-based GIS users,
divide the territory into regular squares of pre-defined
size which become the mapping unit of referŽence
Carrara, 1983; Bernknopf et al., 1988; Pike,
1988; van Westen, 1993, 1994; Mark and Ellen,
.1995 . Each grid-cell is assigned a value for each
Ž .factor morphological, geological, of land-use, etc.
taken into consideration. Alternatively, a stack of
raster layers, each mapping a single instability factor,
is prepared.
Terrain units, traditionally favored by geomorphologists,
are based on the observation that in
natural environments the interrelations between materials,
forms and processes result in boundaries
which frequently reflect geomorphological and geological
differences. Terrain units are the base of the
land-system classification approach which has found
application in many land resources investigations
ŽCooke and Doornkamp, 1974; Speight, 1977; Verstappen,
1983; Burnett et al., 1985; Meijerink, 1988;
.Hansen et al., 1995 .
ŽUnique-condition units Bonham-Carter, 1994;
.Chung et al., 1995 imply the classification of each
slope-instability factor into a few significant classes
which are stored into a single map, or layer. By
sequentially overlying all the layers, homogeneous
Ž .domains unique conditions are singled out whose
number, size and nature depend on the criteria used
in classifying the input factors.
Slope-units, automatically derived from high-quality
DTMs, partition the territory into hydrological
Žregions between drainage and divide lines Carrara,
.1988; Carrara et al., 1991 . Depending on the type of
Žinstability to be investigated deep-seated vs. shallow
.slides or complex slides vs. debris flows the mapping
unit may correspond either to the sub-basin or
Žto the main slope-unit rightrleft side of the sub.basin
.
Slope-units can be further subdivided into topographic
units defined by the intersections of contours
and flow tube boundaries orthogonal to contours
Ž .O’Loughlin, 1986 . For each topographic unit, local
morphometric variables and the cumulative drainage
area of all up-slope elements are computed.
Selection of an appropriate mapping unit depends
on a number of factors, namely: the type of landslide
phenomena to be studied; the scale of the investigation;
the quality, resolution, scale and type of the
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216186
thematic information required; and the availability of
the adequate information management and analysis
tools. Each technique for tesselling the territory has
advantages and limitations that can be enhanced or
reduced choosing the appropriate hazard evaluation
method.
3.3. Landslide hazard modelling
Methods for ranking slope instability factors and
assigning the different hazard levels can be qualitatiÕe
or quantitatiÕe and direct or indirect.
Qualitative methods are subjective and portray the
Ž .hazard zoning in descriptive qualitative terms.
Quantitative methods produce numerical estimates
Ž .probabilities of the occurrence of landslide phenomena
in any hazard zone. Direct methods consist
of the geomorphological mapping of landslide hazŽ
.ard Verstappen, 1983 . Indirect methods for landslide
hazard assessment are essentially stepwise. They
require first the recognition and mapping of landŽslides
over a target region or a subset of it training
.area . It follows the identification and mapping of a
group of physical factors which are directly or indiŽrectly
correlated with slope instability instability
.factors . They then involve an estimate of the relative
contribution of the instability factors in generating
slope-failures, and the classification of the land
surface into domains of different hazard degree
Ž .hazard zoning .
The most important methods proposed in the literŽature
can be grouped into few main categories Carrara
et al., 1992; van Westen, 1993; Carrara et al.,
.1995; Hutchinson, 1995 , namely:
– geomorphological hazard mapping;
– analysis of landslide inventories;
– heuristic or index based methods;
– functional, statistically based models;
– geotechnical or physically based models.
Geomorphological mapping of landslide hazard is
a direct, qualitative method that relies on the ability
of the investigator to estimate actual and potential
Žslope failures Humbert, 1977; Godefroy and Humbert,
1983; Kienholz et al., 1983, 1984; Bosi et al.,
1985; Zimmerman et al., 1986; Seeley and West,
.1990; Hansen et al., 1995 . The heuristic approach,
based on the a priori knowledge of all causes and
instability factors of landsliding in the area under
investigation, is an indirect, mostly qualitative
method, that depends on how well and how much the
investigator understands the geomorphological processes
acting upon the terrain. Instability factors are
ranked and weighted according to their assumed or
expected importance in causing mass movements
ŽNilsen and Brabb, 1977; Amadesi and Vianello,
1978; Hollingsworth and Kovacs, 1981; Neeley and
Rice, 1990; Montgomery et al., 1991; Mejıa-Navarro´
.et al., 1994 .
All other approaches are indirect and quantitative.
The analysis of landslide inventories attempts to
predict future patterns of instability from the past
and present distribution of landslide deposits. This is
Žaccomplished by preparing landslide density ‘‘iso.pleth’’
maps, i.e., maps showing the number or
percent of area covered by landslide deposits over a
Žregion Campbell, 1973; Wright, 1974; Wright and
Nilsen, 1974; Wright et al., 1974; DeGraff, 1985;
.Guzzetti et al., 1994 . Statistical, ‘‘black-box’’ approaches
are based on the analysis of the functional
relationships between instability factors and the past
and present distribution of landslides. Various multivariate
statistical techniques have been applied on
various mapping units. The most favored are discriminant
analysis, linear and logistic regression, and
Žneural networks Neuland, 1976; Carrara, 1983; Carrara
et al., 1991; Carrara et al., 1995; Roth, 1983;
Yin and Yan, 1988; Neeley and Rice, 1990; Mark,
.1992; van Westen, 1993, 1994; Chung et al., 1995 .
A statistical model of slope instability is built on the
assumption that the factors which caused slope-failure
in a region are the same as those which will generate
landslides in the future. The general linear model
assumes the form:
LsB qB X qB X qB X q . . . qB X qe0 1 1 2 2 3 3 m m
Žwhere L is the presencerabsence or the area per.centage
of landslides in each sampling unit, the X’s
Ž .are input predictor variables or instability factors
measured or observed for each mapping unit, the B’s
are coefficients estimated from the data through
techniques which are dependent on the statistical tool
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 187
Žselected multiple regression, discriminant analysis,
.etc. , and e represents the model error.
Ž .Lastly, process-based geotechnical models rely
upon the understanding of few physical laws controlŽling
slope instability Okimura and Kawatani, 1987;
Dunne, 1991; Montgomery and Dietrich, 1994; Diet.rich
et al., 1995; Terlien et al., 1995 . These models
Žcouple shallow subsurface flow i.e., the pore pres.sure
spatial distribution , predicted soil thickness,
Žand landsliding of the soil mantle Dietrich et al.,
.1995 . Stability conditions are generally evaluated by
means of a static model, such as the ‘‘infinite slope
model’’, where the local equilibrium along a potential
slip surface is considered.
As previously mentioned, hazard models and
mapping units are conceptually and operationally
Ž .interrelated Carrara et al., 1995 . In the direct hazard
mapping the geomorphological unit of reference
is implicitly defined by the interpreter that maps
those portions of the territory that are subject to
Ž .different geomorphological hazards Hansen, 1984 .
ŽIn all other cases i.e., grid-based modelling,
unique-condition units, slope-units, topographic
.units , the mapping unit is explicitly defined by the
operator. In general, grid-cells are preferred for
Ž .heuristic Pike, 1988; Mejıa-Navarro et al., 1994 ,´
Ž .statistical Carrara, 1983; van Westen, 1994 and
Žphysical or simulation Mark, 1992; Terlien et al.,
.1995 modelling. Unique-condition units have been
Ž .applied to both heuristic van Westen, 1993 and
Žstatistical methods Carrara et al., 1995; Chung et al.,
.1995 . Slope-units and topographic units have been
Ž .used in statistical Carrara et al., 1991; 1995 and
Ž .physically based Montgomery and Dietrich, 1994
models.
4. The Umbria–Marche hazard assessment pro-
ject
In the Umbria and Marche Regions of Central
Ž .Italy Fig. 1 evaluation of landslide hazard was
attempted using a variety of techniques pertaining to
the realms of geology, geomorphology, statistics,
and information technology. Experiments were carried
out at the regional scale, for the entire Umbria–
Ž 2 .Marche territory 18,125 km in size , and at the
Ž 2 . Žlocal scale, in the Tescio 59 km and Carpina 67
2 . Ž .km basins Guzzetti, 1993 . The long-term hazard
assessment project involved:
– the regional evaluation of landslide occurrence,
obtained through the interpretation of mediumŽscale
aerial photographs Guzzetti and Cardinali,
.1989, 1990; Antonini et al., 1993 and the inventory
of historical information on slope movements
Ž .Guzzetti et al., 1994 ;
– a reconnaissance estimate of landslide hazard,
attempted using the regional landslide inventory
and the available, small scale thematic informa-
tion;
– a set of detailed landslide hazard models in test
areas, selected for their lithological, structural and
morphological settings representative of large secŽtors
of the Umbria–Marche territory Carrara et
.al., 1991, 1995 ;
– a conceptual model of landslide occurrence
Ž .Guzzetti et al., 1996 .
Results of these experiments, along with the outcomes
of an international workshop on the application
of GIS technology in assessing natural hazards
ŽReichenbach et al., 1993; Carrara and Guzzetti,
.1995 , encouraged the undertaking of a detailed evaluation
of landslide hazard over the upper section of
Ž 2 .the Tiber River basin 4097 km in size . This
experiment, which is still in progress, is requiring a
great deal of work in data acquisition, storage and
processing and will need a significant amount of
time and funds to be completed.
4.1. Regional setting
The Umbria and Marche Regions are located
Ž .along the Apennines mountain chain Fig. 1 . To the
east of the Apennines, the Umbria Region is drained
by the Tiber River that flows into the Tyrrhenian sea.
The Marche Region, to the west of the Apennines
main divide, exhibits a parallel drainage that flows
into the Adriatic sea.
The study area has a long history of hydrogeological
catastrophes. Reports on landslides go back to
Etruscan and Roman periods, but the first documented
information on slope movements in the hillsides
of Todi and Orvieto dates back to the fourteenth
century. Due to the extent and economic
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216188
Fig. 1. The Umbria–Marche territory in shaded relief. The image was prepared from the Archive of Mean Elevations of Italy with a ground
Ž .resolution of 230=230 m. Sun azimuth angle is 3158, elevation above the horizon is 458. No vertical exaggeration. A Upper Tiber River
Ž . Ž .basin. B Carpina basin. C Tescio basin.
significance of landslides, research on slope movements,
ranging in scale from site specific investigaŽtions
to regional studies, is abundant Guzzetti et al.,
.1996 .
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 189
Different rock types crop out in the area, varying
in strength from hard to weak and soft rocks, that
Žcan be grouped into few lithological domains Fig.
.2A . Hard rocks consist of layered and massive
limestone, cherty limestone, sandstone, pyroclastic
deposits, travertine and conglomerate. Weak rocks
are marl, shale, sand, silty clay and stiff, overconsolidated
clay. Soft rocks are marine and continental
clay, silty clay and shale. The morphological and
structural setting of the area is determined by the
superposition of two tectonic phases. A compressive
phase, late Miocene to early Pliocene in age, was
followed by an extension phase of Pliocene to Recent
age. The compressive deformation produced
large anticlines, corresponding to major divides, and
synclines associated with thrusts and transcurrent
faults. The extensional tectonic phase produced normal
faults that formed intra-mountain basins and
valleys.
The lithological and structural domains are characterized
by a prevalent geomorphological setting
and by typical geotechnical and hydrogeological
properties that control the abundance and pattern of
slope failures. Mass movements, ranging in size
from less than 1 ha to few square kilometers, include:
falls and topples in hard rocks; soil-slips in
the colluvial cover mantling slopes in soft or weak
rocks; rotational slides in homogeneous, mostly soft
rocks; translational slides in well bedded, soft and
hard rocks; earth-flows, complex and compound
slides where alternating hard and soft rocks crop out
Ž .Guzzetti et al., 1996 .
4.2. Regional eÕaluation of landslide occurrence
The regional inventory of landslides can be attempted
through the catalogue of existing informaŽtion
on mass movements bibliographical or historical
catalogue; cf. Nemcok and Rybar, 1968; Rad-´
.bruch-Hall et al., 1982; Brabb, 1984 or by means of
the systematic interpretation of medium- or smallŽscale
aerial photographs reconnaissance inÕentory;
.cf. Brabb, 1984; Hansen, 1984; Wieczorek, 1984 .
For the Umbria and Marche Regions a reconnaisŽsance
inventory of landslide deposits Guzzetti and
.Cardinali, 1989, 1990; Antonini et al., 1993 and a
catalogue of bibliographical information on landŽ
.slides CoGeo, 1994a, 1994b; Guzzetti et al., 1994
were completed in the years 1986–1992.
The reconnaissance mapping was carried out
through the systematic analysis of about 2100 black
and white vertical aerial photographs, at 1:33,000
scale. Landslides were classified according to a simŽ
.plified version of Varnes 1978 classification of
mass movements. In the Marche Region, landslide
relative-age was also estimated. Mapping took 5
manryears and detected about 14,700 landslide deŽ
.posits. Additionally, 9700 small less than 1 ha
Ž .failures, affecting mostly clay about 50% and flyŽ
.sch deposits about 30% , were identified and mapped
as single points. The total mapped landslide area was
1628 km2
, namely, 9% of the Umbria–Marche terriŽ
.tory Fig. 2B . Detailed geomorphological investigations
carried out in pilot areas suggest that this is a
Ž .lower estimate Guzzetti et al., 1996 .
The reconnaissance inventory revealed different
types of landslides. Complex failures, covering 40%
of the total landslide area, showed the largest extent.
Flows were the smallest failures, but in the northern
part of the study area flows exceeding 3 km2
are
Žpresent. In addition, tectonic melanges 40% of the
. Ž .territory and flysch deposits 12% of the territory
were the most landslide-prone rocks, followed by all
other rock types with less than 10% landslide area.
A catalogue of bibliographical information on
slope failures for the Umbria and Marche Regions
was completed for the period 1918–1990 through the
systematic review of four newspapers, the interview
of 24 expert witnesses, and inspection of 180 techniŽ
.cal and scientific reports CoGeo, 1994a, 1994b .
The historical investigation revealed 1485 landslide
events, at 956 different sites, affecting 89 out of 92
Ž .townships in Umbria 97% and 148 out of 246
Ž . Ž .townships in the Marche Region 60% Fig. 2C .
ŽThe analysis of the limited number of failures 35%
.for Umbria and 16% for Marche for which the date
of occurrence was known, showed a higher freŽ
.quency of events in the winter season Fig. 3A .
Additionally, landslide frequency exhibited a correlaŽ
.tion with the general climatic trend Fig. 3B . Landslide
events were found abundant in the period
1950–1969 and rare during the Second World War
Ž .and the post war period 1940–1949 . The latter
reflects the incompleteness of the catalogue rather
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216190
Ž . Ž .Fig. 2. Umbria and Marche Regions. Geological and morphological setting of the territory. A Lithological domains. 1 Lake and alluvial,
Ž . Ž . Ž .post-orogenic sediments; 2 Flysch deposits pertaining to the Macigno Fms.; 3 Flysch deposits of the Marche sequence; 4 Ligurian
Ž . Ž .allochtonous complex; 5 Flysch deposits pertaining to the Marnoso–Arenacea Fm.; 6 Plio-Pleistocene marine and continental deposits;
Ž . Ž . Ž .7 Limestone and Marls pertaining to the Umbria–Marche sequence; 8 Volcanic rocks. B Distribution of landslide deposits mapped
Ž . Ž .through a reconnaissance survey of medium-scale aerial photographs after Guzzetti and Cardinali, 1989; Antonini et al., 1993 . C
Ž .Administrative boundaries townships . Dots report the location of 956 sites affected by mass movements cataloged by the historical
Ž . Ž . Ž . Ž . Ž .inventory completed for the period 1918–1990. D Morphological classification in terrain types. 1 Lowlands; 2 Low hills; 3 Hills; 4
Ž .Mountains; 5 High mountains.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 191
Fig. 3. Umbria and Marche Regions. Frequency of landslide
events cataloged for the period 1918–1990, compared with the
regional climatic trend, expressed by average mean daily discharge
of the Tiber River at the Ponte Nuovo gauging station
Ž . Ž .Perugia . A Average, monthly frequency of landslide events.
Ž .B Frequency of landslide events in 5-year intervals.
Žthan a peculiar climatological condition a dry pe.riod
.
As a preliminary assessment of the regional economic
impact of landslides in the Umbria–Marche
Žterritory, the two inventories reconnaissance and
.bibliographical were compared using the local adŽ
.ministrative boundaries townships as the reference
Ž .mapping unit Fig. 2C . The percentage of landslide
area mapped by the reconnaissance inventory and the
number of events available in the historical catalogue
were counted within the territory of each municipality.
Percentage of landslide area was found ranging
Ž . Žbetween nil 0% , in landslide free areas, to 88% at
.Carpegna , with an average value of 10%. Forty
Ž .percent of townships 125 , corresponding to about
30% of the territory, exhibit a percentage of landslide
area greater than the average. Only 15 townships
had a percentage of landslide area less than
1%. Of these, five, due to the local morphological
Ž .and geological setting i.e., large plains , were found
completely free of landslide deposits. The bibliographical
inventory revealed that 237 townships
Ž .70% experienced from one up to a maximum of 88
Ž .landslide events in 72 years 1918–1990 , with an
Ž .average of five events. For 101 townships 30% no
information on landslides was reported. Further analysis
showed that for only about 10% of the townships,
the morphological and geological setting was
not landslide-prone. In all other cases the lack of
information could not be interpreted as a safety
condition, but the result of the incompleteness of the
historical record.
An attempt was made to test the consistency of
the two regional evaluations of landslide occurrence.
Due to the lack of precision in the location of many
landslides identified by the historical investigation
Ž .Guzzetti et al., 1994 , and the uncertainty associated
Žwith small scale landslide mapping Carrara et al.,
.1992 , a direct map overlay was not appropriate. To
take care of possible mapping errors a ‘‘confidence
Ž .belt’’ a ‘‘buffer’’ was traced around each landslide
Ž .whose width was proportional 10% to landslide
area. Then the distance between each landslide identified
historically to the nearest landslide mapped by
the reconnaissance inventory was computed. It was
Žfound that the density of events number of
2 .eventsrkm that fall directly on landslides mapped
Žby the reconnaissance inventory landslide deposit
.plus 10% confidence belt or within a distance of
500 m, is twice the density of events that lay at a
greater distance. In other words, 70% of historical
events lay on, or within a distance of 500 m to the
nearest mapped landslide.
4.3. Reconnaissance modelling of landslide hazard
The reconnaissance estimate of landslide hazard
for the entire Umbria–Marche territory was attempted
in two ways. At first, an isopleth map,
showing the distribution of landslide density, was
prepared counting the percentage of area affected by
landslides within a circular moving window of about
2 Ž .1 km Fig. 4A . In landslide-prone areas, landslide
Ž 2 .density varies from 0.01 1 harkm up to 1.0,
where the whole area is covered by landslide deŽ
.posits Guzzetti et al., 1994 .
As a second attempt, a statistical model of landslide
hazard was developed using the reconnaissance
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216192
inventory of slope movements and the thematic inŽ
.formation available at small scale Fig. 4B . Geology
was obtained from existing maps at 1:100,000 scale
by grouping the over 100 formations into eight lithoŽ
.logical domains Fig. 2A . Major lithological boundaries
were buffered to capture the instability effect
Žon slopes of contrasting lithologies Guzzetti et al.,
.1996 .
Morphology was estimated through the computaŽtion
of geomorphological parameters i.e., elevation,
terrain gradient, curvature, frequency of slope direc.tion
changes, and elevation relief ratio from a coarse
Ž .230=230 m DTM and an unsupervised cluster
analysis of such morphometric data. Hence, the territory
was simply divided into five terrain types,
namely: lowlands, low hills, hills, mountains and
Ž .high mountains Fig. 2D .
Regional seismicity was obtained from a synoptic
map showing the maximum felt seismic intensity in
Ž .Italy Boschi et al., 1995 . Intensity levels were
Ž .grouped into three classes, namely: low 68–78 ,
Ž . Ž .medium 88–98 , and high 108–118 MCS scale
intensity.
Regional climatic conditions were estimated by
preparing maps of mean annual precipitation and
yearly number of rainy days for the period 1921–
1950. Mean annual precipitation ranges from 570 to
1880 mm, whereas rainy days are between 62 and
124 mm. Both parameters are correlated to elevation.
A simple index expressing the average yearly rainfall
intensity was computed as the ratio between mean
annual rainfall and the yearly number of rainy days.
Index values were ranked into three classes, corresponding
to low, medium and high yearly rainfall
intensity.
Mapping units for the analysis were obtained by
sequentially overlaying the five thematic maps previously
listed. Because the thematic variables are spatially
correlated, of the 1080 possible unique conditions
only 522 actually resulted, for a total of over
Ž .50,000 domains polygons . Each unique condition
was classified as stable or unstable depending on the
percentage of area affected by any type of landslide
deposit. The threshold was selected equal to the
Ž .mean landslide area of the whole territory 9% , that
is, the expected probability to find a landslide deposit
by chance.
Logistic regression was then applied to predict
stable and unstable terrain units using 17 dummy
Ž .0r1 variables corresponding to the classes into
which the five input thematic maps were grouped
Ž .Table 1 . The results of the classification are shown
in Table 2, and the probabilities of landslide occurrence,
grouped into four classes, are displayed in
Fig. 4B. Of the variables entered into the equation,
Ž .those reflecting rock type eight are the most important
in classifying stable and unstable units with a
success nearly equal to 75%. Conversely, seismic
Ž . Ž .zoning two and climatic belts two proved to be
rather poor predictors of landslide distribution. This
Žmight reflect the time-span of the seismic map few
. Ž .centuries and of the rainfall map 30 years . Both
are much shorter than that of the reconnaissance
Žinventory that portrays the result of 10,000 years or
.more of geomorphological history. The limited preŽ
.dictive power of morphological variables four may
be due to the strong correlation at regional scale
Ž . Žbetween morphology Fig. 2D and lithology Fig.
.2A .
ŽBy overlaying the landslide deposits map Fig.
. Ž .2B over the regional hazard model map Fig. 4B ,
Žthe belts at lower probability 0%–20% and 20%–
.40% were found to have a percentage of landslide
Ž .area 4.4% and 6.3%, respectively which is about
one third of that featuring the belts at high hazard
Ž .60%–100% , namely 14.1%; while in the intermediŽ
.ate hazard region 40%–60% landslide area is 8.9%.
Lastly, a preliminary attempt was made to rank
the territory of each municipality into hazard classes
based on the outcome of the reconnaissance hazard
model. It was found that 95 townships have more
than 75% of their territory classified as landslide
Ž . Ž .Fig. 4. Umbria and Marche Regions. Reconnaissance assessment of landslide hazard. A Landslide density map ‘‘isopleth map’’ . Shades
Ž . Ž . Ž .of grey indicate increasing percentage of landslide area, from less than 1% white to 100% black . B Landslide hazard assessment by
Ž . Ž . Ž . Ž . Ž .logistic regression on 522 unique-condition units. Hazard levels are: 1 0%–20% very low ; 2 20%–40% low ; 3 40%–60%
Ž . Ž . Ž .intermediate ; and 4 60%–100% high .
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 193
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216194
Table 1
Umbria and Marche Regions. List of 17 dummy variables entered into the logistic regression model, equation coefficients and their standard
Ž . Ž . Ž .errors S.E. . Grouping variable: unique-condition unit free unstable area F 9% vs. affected unstable area)9% by landslide deposits
Variable Explanation Coefficent S.E.
MAR Marnoso–Arenacea Fm. Flysch deposits 3.081 1.221
AVAN marine clay and sand 3.724 1.235
GESSO Marche flysch deposits 4.225 1.229
UMBRO Umbria–Marche stratigraphic sequence 2.333 1.218
SINE lake and alluvial deposits 1.102 1.244
CERVA Macigno Fms. Flysch deposits 3.244 1.229
VULC volcanic rocks 2.937 1.257
LIGU Ligurian allochtonus sediments 4.732 1.313
DD-DD corridor at the boundary of lithological units 0.919 0.366
MOR-1 lowland y0.894 0.330
MOR-2 low hills 1.144 2.315
MOR-3 hills 0.401 0.309
MOR-5 high mountains 0.481 0.408
CLIMA-1 low rainfall intensity y0.465 0.267
CLIMA-3 high rainfall intensity y0.415 0.251
SEIS-1 low seismicity, 6 to 7 MCS scale 0.403 0.225
SEIS-2 intermediate seismicity, 8 to 9 MCS scale 0.097 0.072
Model y3.322 1.237
Ž .prone high hazard class . Conversely, only 35 townships
have 75% or more of the territory mapped as
Ž .potentially stable low hazard class .
4.4. Landslide hazard modelling in pilot areas
In the Tescio and Carpina tributaries of the Tiber
Ž .River Fig. 1 , detailed hazard evaluations were carried
out testing a variety of data acquisition techniques,
mapping unit types, and information manageŽment
and statistical techniques Carrara et al., 1991,
.1995 .
Both areas are underlain by rocks belonging to the
Umbria–Marche stratigraphic sequence. In the Tescio
basin crop out: to the south, thinly bedded limestone,
marl and shale Late Jurassic to Cretaceous in age; in
the central part, marl and shale Oligocene to Eocene
in age; and, to the north, alternating sandstone, calcarenite
and marl Miocene in age. The latter cover
half of the basin and are affected by numerous
landslides, mostly complex, rotational or translational
slides with a distinct flow component at the
toe. The Carpina basin is underlined by flysch deposits,
Eocene to Miocene in age. In the area crop
out rhythmic sequences of sandstone, calcarenite and
marl in different proportion, marl, shale, and chaotic
Ž .mixtures olistostromes of various rock types. Mass
movements comprise large, very old complex slides
controlled by the local bedding attitude; old to recent
Žslides are abundant where competent beds sandstone
Table 2
Umbria and Marche Regions. Classification of stable and unstable unique-condition units by logistic regression
Unique-condition units correctly classified: 74.8%.
Actual group No. of unique condition units Predicted group membership
Ž . Ž .Group 1 stable units Group 2 unstable units
Ž .Group 1 stable units 278 214 64
Ž .Group 2 unstable units 244 68 176
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 195
Table 3
Tescio basin. List of variables entered in the discriminant function and their relative importance as expressed by the standardized
Ž . Ž .discriminant function coefficient SDFC . Grouping variable: slope-unit free of vs. affected by landsliding after Carrara et al., 1991
Variable SDFC
CINE slope-unit percent of Scaglia Cinerea rock type y0.202
SCHL slope-unit percent of Schlier rock type y0.355
ARCA slope-unit percent of sandstone-rich rock type 0.331
MAXCA product of marl-rich and calcarenite-rich rock types 0.693
DENUD slope-unit percent of uncultivated area 0.314
BOSCO slope-unit percent of forest y0.601
AN slope-unit facing N 0.199
AW slope-unit facing W 0.293
MAGN sub-basin magnitude y0.492
ELV-M slope-unit mean elevation y0.295
FORM slope-unit form perimeterrarea y0.503
RXGR slope-unit surface roughness index y0.260
FRA-TR bedding dipping toward slope-unit free face 0.251
Ž .IDR-A permeable beds sandstone capping impermeable ones 0.545
Ž .IDR-D impermeable beds clay and shale throughout slope-unit 0.840
.and calcarenite are present within mostly marly
rocks; and old to recent shallow soil-slips and flows
take place on soil-mantled slopes.
Detailed thematic data were derived from existing
topographic maps, aerial photographs and field surveys.
Landslide deposits, classified according to relative
age, degree of activity, movement type, estimated
depth and velocity, type of material, and
mapping certainty, were determined by interpreting
aerial photographs of different dates and scales
Ž .1:33,000 and 1:13,000 , and by systematic field
investigations.
ŽUsing high-fidelity DTMs 20=20 or 25=25
.m , drainage-divide networks were automatically
identified and basins were partitioned into sub-basins
and slope-units, each characterized by a wide set of
Žmorphometric and hydrological parameters Carrara,
.1988; Carrara et al., 1991, 1995 .
Geological data were obtained by field mapping
at 1:10,000 scale, aided by photo-geological techŽniques.
Bedding and structural measurements joints,
.cleavage and faults were taken as uniformly as
possible throughout the study areas. This allowed
partitioning of the terrain into structural domains
Ž .i.e., anticline, thrust, graben, etc. as well as in
Ž .constant bedding strike and dip areas. By comparŽing
bedding attitude and slope orientation aspect
.and steepness , slope-units were classified in structural
and bedding attitude classes. To estimate the
hydrological conditions of slopes, the stratigraphic
relations between permeable and impermeable rocks
were estimated in the field and from the lithological
maps. Land-use data were obtained from existing
maps at 1:10,000 scale and through the interpretation
of large scale, color aerial photographs.
In the Tescio basin, for each slope-unit the percentage
of unstable area was derived as the weighted
summation of the landslide area existing in the unit.
Slope-units were defined as landslide-free and landslide-bearing
when the percentage of failed area was
Table 4
Ž .Tescio basin. Classification of stable and unstable slope-units by discriminant analysis after Carrara et al., 1991
Slope-units correctly classified: 83.8%.
Actual group No. of slope-units Predicted group membership
Ž . Ž .Group 1 stable slopes Group 2 unstable slopes
Ž .Group 1 stable slopes 148 128 20
Ž .Group 2 unstable slopes 118 23 95
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216196
Table 5
Carpina basin. List of variables entered in the discriminant function and their relative importance as expressed by the standardized
Ž . Ždiscriminant function coefficient SDFC . Grouping variable: slope-unit free of vs. affected by old to recent slides after Carrara et al.,
.1995
Variable SDFC
CAM slope-unit percent of marly–calcareous sandstone 0.253
ALL-COL slope-unit percent of alluvial–colluvial deposits y0.209
D3 slope-unit percent of N monocline domain 0.215
D11 slope-unit percent of NE transcurrent fault domain 0.122
D12 slope-unit percent of graben domain 0.173
TRR bedding dipping obliquely into the slope y0.217
TFP bedding dipping toward slope free face 0.245
CATA slope-unit percent of cataclastic rocks 0.105
RX variability of across-slope profile y0.338
COC-COV concave–convex slope-unit profile y0.170
IRR irregular slope-unit profile y0.109
MOR-A1 PC reflecting slope length and width 0.354
MOR-B1 PC reflecting slope steepness y0.264
PALEO slope-unit profile inherited from old landsliding 0.268
AC-TM permeable beds capping impermeable ones 0.115
AC-A acquifer in alluvial–colluvial deposits y0.232
S-BO slope-unit percent of forest area y0.204
S-PP slope-unit percent of pasture area 0.126
S-DN slope-unit percent of barren area y0.264
S-SAP slope-unit percent of cultivated area 0.254
less or greater than 2%, respectively. This threshold
was derived from an estimate of average drafting and
digitising errors. Using classified slope-units as the
grouping variable and almost 40 factors as input
predictor variables, stepwise discriminant analysis
was applied in order to predict stable or unstable
slope-units, on the basis of their morphological, geological
and land-use characteristics. The variables
Ž .factors entered into the discriminant function are
listed in Table 3, while the results of the classification
are summarized in Table 4. A test of the statistical
reliability of the model showed that the discrimiŽnant
function was able to classify correctly from
.75% to 82% stable and unstable slopes belonging to
the test set.
In the Carpina basin large, very old complex
slides; old to recent slides; and old to recent, shallow
flows or soil-slips were processed separately. For the
area, three drainage-divide networks of increasing
detail were prepared, partitioning the basin into a
different number of slope-units, namely: 66, 414 and
750 which correspond to an average size of 1.109,
0.162 and 0.090 km2
, respectively. These values
were selected in agreement with the average size of
each landslide group.
Because of the large number of variables availŽ
.able over 60 and their high interrelations, selected
subsets were replaced with their most significant
Ž .principal components PC , through standard princiŽ
.pal components analysis PCA , to reduce redundancy
and to improve numerical stability in the
subsequent analyses. Stepwise discriminant analysis
was performed on each set of 66, 414 and 750
slope-units, setting as grouping variable the presencerabsence
of slope failures belonging to either
the large, very old landslides, or slides, or flows.
Since slope-units are very unequal in size in each
map and uncertainty in the input data is expected to
Ž . Ž . Ž . Ž . Ž .Fig. 5. Carpina basin. Evaluation of landslide hazard. Hazard levels are: 1 0%–40% low ; 2 40%–60% intermediate ; and 3
Ž . Ž . Ž . Ž .60%–100% high ; 4 are landslide deposits. A Landslide hazard assessment by discriminant analysis on 414 slope-units. B Landslide
hazard assessment by discriminant analysis on 2092 unique-condition units.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 197
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216198
Table 6
Ž .Carpina basin. Classification of stable and unstable slope-units by discriminant analysis after Carrara et al., 1995
Slope-units correctly classified: 80.7%.
Actual Group No. of slope-units Predicted group membership
Ž . Ž .Group 1 stable slopes Group 2 unstable slopes
Ž .Group 1 stable slopes 278 228 50
Ž .Group 2 unstable slopes 136 30 106
decrease with slope-unit size, all analyses were
weighted by the log of the slope-unit area. The
results of this threefold analysis can be summarized
as follows.
Ž .To predict successfully at the 92% level the
occurrence of the few large, very old slides, 12
variables entered into the model. They equally reflect
rock composition, structural setting, slope morphometry
and ground water conditions.
In the second analysis carried out on the statistically
more significant group of the old to recent
slides, a wider spectrum of variables entered into the
Ž . Ž .model Table 5 , namely: rock type two , structure
Ž . Ž . Ž .six , morphometry six , water conditions two and
Ž .land-use four . In Fig. 5A, the probabilities of slide
occurrence, grouped into three classes, are displayed
along with the slide deposits. Although the classificaŽtion
power of the model is rather good over 80%,
.Table 6 , too many predictors were needed to obtain
this result. Indeed, an intrinsic limitation of any
multivariate analysis is that as the number of variables
increases the reliability of the model decreases
to some extent.
To predict at the 75% level slope-units affected
by shallow landslides, 20 variables entered into the
discriminant function, of which two regard slope
Table 7
Carpina basin. List of dummy variables entered into the discriminant function and their relative importance as expressed by the standardized
Ž . Ž .discriminant function coefficient SDFC . Grouping variable: unique-condition unit free of unstable area-4.39% vs. affected by old to
Ž .recent slides after Carrara et al., 1995
Variable SDFC
CALC calcareous sandstone and marl y0.195
ARP sandstone and marl y0.141
CAM marly–calcareous sandstone 0.419
PELC marl and calcareous sandstone y0.100
OLI tectonic clayey melange 0.075
ALL-COL alluvial–colluvial deposits y0.216
D3 northern monocline domain 0.307
D7 southern thrust fault domain y0.201
D10 central transcurrent fault domain 0.171
D11 northeastern transcurrent fault domain 0.160
SLO10 slope angle-108 y0.194
SLO25 slope angle between 208–258 y0.151
SLO90 slope angle)258 y0.262
PROF1 concave down-slope profile y0.205
PROF2 rectilinear down-slope profile y0.071
CATA cataclastic rock 0.157
LEN200 slope length-200 m y0.112
LEN400 slope length between 200–400 m 0.286
REG-T bedding dipping into the slope y0.192
FRP-T bedding dipping toward the slope free face 0.231
S-BO forested area 0.220
S-PP pasture area 0.479
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 199
Table 8
Ž .Carpina basin. Classification of stable and unstable unique-condition units by discriminant analysis after Carrara et al., 1995
Unique-condition units correctly classified: 72.7%.
Actual group No. of unique condition units Predicted group membership
Ž . Ž .Group 1 stable units Group 2 unstable units
Ž .Group 1 stable units 1213 893 320
Ž .Group 2 unstable units 879 252 627
material, nine the structure, six the morphometry and
three the land-use type. The rather low percentage of
slope-units correctly classified indicates that input
variables were unable to predict adequately the spatial
distribution of flows in the area. This is not
surprising; a test of randomness of their distribution
proved that in a large portion of the basin shallow
failures were nearly randomly distributed with reŽspect
to the available thematic information Cardinali
.et al., 1994 .
In the Carpina basin, an attempt was also made to
assess landslide hazard due to old to recent slides
using unique-conditions as mapping unit. In order to
apply a multivariate statistical analysis to such an
approach, all input variables were grouped into a few
meaningful classes. For categorical data, such as
rock type and land-use, this operation did not involve
any subjective judgement. For continuous variables,
such as slope angle or length, the selection of the
Ž .number of classes and class limits break points
required a significant amount of guess work guided
by previous knowledge of the causal relationships
between slope failures and instability factors. As a
Ž .result, from eight input dummy 0r1 variables,
Ž .namely: rock type eight classes , structural domains
Ž . Ž .12 classes , fault zones two classes , bedding attiŽ
.tude vs. slope aspectrangle four classes , slope
Ž . Žangle five classes , down-slope profile three
. Ž .classes , slope length four classes and land-use
Ž .three classes a total of 41 classes were derived. To
limit the number of statistically meaningless uniqueconditions,
filtering techniques were applied after
each map overlay step. As a result, the final map had
only 2092 unique-conditions, out of 138,240 possible
cases.
Stepwise discriminant analysis was then applied
using, as the grouping variable, unique-condition
units having a percentage of sliding area lower or
greater than 4.39%, that is half the average instability
percentage of the basin, and, as predictors, the
dummy variables corresponding to the classes into
which the eight input layers were grouped. Under the
assumption that both the errors and uncertainty decrease
with the size of the ground domain, all the
analyses were weighted by the log of the domain
area.
Model results are listed in Table 7 and the probabilities
of slide occurrence, grouped into three classes,
Ž .are displayed in Fig. 5B. Of the 22 dummy 0r1
variables entered into the function, five are lithological,
seven structural, seven morphometrical, and two
concern land-use. The presencerabsence of pasture,
marly–calcareous sandstone, northern monocline domain,
and slope length are the most important in
classifying stable and unstable units with a success
Ž .equal to 73% Table 8 .
The outcome of the two hazard assessments, on
Ž .slope-units model A and on unique-condition units
Table 9
Ž .Carpina basin. Comparison of percentages of area predicted as unstable, intermediate ‘‘unclassified’’ and stable, based on discriminant
Ž .membership probabilities greater than 60%, between 40%–60% and less than 40%. Model A slope-units refers to Tables 5 and 6 and Fig.
Ž . Ž .5A. Model B unique-condition units refers to Tables 7 and 8 and Fig. 5B after Carrara et al., 1995
Ž . Ž .Model A slope-units Model B unique-condition units
Unstable area 31.5% 34.0%
Intermediate area 18.7% 21.1%
Stable area 49.8% 44.8%
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216200
Ž .model B , can be compared. The lists of variables
entered into the discriminant functions for slope-units
Ž . Ž .Table 5 and unique-condition units Table 7 show
that several predictors are in common, albeit with
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 201
Table 10
Upper Tiber River basin. List of 40 variables entered into the discriminant function and their relative importance as expressed by the
Ž .standardized discriminant function coefficient SDFC . Grouping variable: slope-unit free vs. slope-unit affected by deep-seated landslides
Variable SDFC
SAPM Alberese Fm. Limestone and marls 0.345
MAUH Marnoso–Arenacea Fm. Flysch deposits 0.163
ASS Chaotic lithological complex 0.166
FADT alluvial deposits, fans and detritus y0.335
STS Macigno del Mugello Fm. Marly flysch 0.235
STH Macigno del Chianti Fm. Sandy flysch 0.092
MNS Monte Nero Fm. Marl and shale 0.073
LIGH Ligurian allochtonus complex. Ophiolite suite 0.231
MARS Marnoso–Arenacea Fm. Marly, flysch deposits 0.219
DFLS lake deposits, clay and silt 0.082
DFLM lake deposits, silt and sand 0.296
DFLH lake deposits, gravel and cobbles y0.049
LINK LEN channel link length y0.131–
LINK ANG channel link slope 0.164–
ANG STD dispersion of channel link slope y0.129–
SLO ARE slope-unit area 0.153–
R variability of slope profile y0.106
ELEV STD dispersion of elevation 0.588–
SLO LEN slope-unit length 0.422–
COV COC convex–concave slope-unit profile y0.173–
COC COV concave–convex slope-unit profile 0.040–
MOR A1 PC reflecting slope-unit hydrologic position 0.300–
MOR A2 PC reflecting slope-unit hydrologic position 0.189–
MOR B1 PC reflecting slope steepness y0.475–
IRR irregular slope-unit profile y0.064
TR1 slope-unit facing N or NW 0.036
TR2 slope-unit facing NE or E 0.095
STRU1 PC reflecting rock structure and attitude y0.427
STRU3 PC reflecting rock structure and attitude y0.453
REG bedding dipping into the slope y0.389
FRAM bedding dipping toward slope free face 0.537
TRA bedding dipping at right angle to the slope y0.049
CAO chaotic bedding 0.122
NONE undefined bedding y0.116
MASS massive rock types y0.160
AE slope-unit percent of built up area y0.077
SA slope-unit percent of culture with orchard area 0.033
PA slope-unit percent of pasture area 0.275
SS slope-unit percent of cultivated area 0.162
AN slope-unit percent of denuded and unclassified area y0.083
Ž .similar of different coefficients SDCF . Of the 20
variables entered into the slope-unit model, nine,
Ž .equally distributed between lithology two , structure
Ž . Ž . Ž .two bedding attitude three and land-use two ,
Ž . Ž . Ž .Fig. 6. Upper Tiber River basin. A Subdivision of the basin into 5598 slope-units. B Lithological map. 1 Alluvial deposits, fans and
Ž . Ž . Ž . Ž .detritus; 2 Chaotic lithological complex; 3 Limestone and sandstone of the San Marino sequence; 4 Lake deposits; 5 Ligurian
Ž . Ž .allochtonous complex; 6 Flysch deposits of the Marnoso–Arenacea Romagnola sequence; 7 Flysch deposits of the Marnoso–Arenacea
Ž . Ž . Ž .Umbra sequence; 8 Marl and shale of the Monte Nero Fm.; 9 Limestone and marl of the Alberese Fm.; 10 Flysch deposits of the
Ž . Ž . Ž . Ž .Macigno Fms.; 11 Schlier Fm. C Landslide inventory map, only deep-seated landslides are reported. D Bedding attitude map. 1
Ž . Ž . Ž . Ž . Ž . Ž .N–NE; 2 E–SE; 3 S–SW; 4 W–NW; 5 chaotic; 6 undefined; 7 massive.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216202
entered directly into the unique-condition model.
Other variables are comparable, namely: MOR B1,–
a proxy for terrain gradient, incorporates much of the
information of SLO10, SLO25 and SLO90. Likewise,
MOR A1, a proxy for slope length, includes–
LEN200 and LEN400.
The total areas predicted by the two models as
Ž .unstable, intermediate ‘‘unclassified’’ and stable
Ž .are comparable Table 9 . In terms of predictive
Ž .power, model A slope-units is significantly superior
for the higher percentage of classes correctly
Ž .classified 80.7 vs. 72.7 and for the lower proporŽ
.tion of area ‘‘unclassified’’ 18.7% vs. 21.1% ; however,
its spatial resolution is lower than that of model
Ž .B unique-condition units . The average size of a
Ž 2 .slope-unit 0.13 km is five times larger the average
Ž 2 .size of a unique-condition unit 0.03 km .
4.5. PredictiÕe model of landslide hazard for the
upper Tiber riÕer basin
A detailed estimation of landslide hazard over a
large area is currently being attempted in the Upper
Ž .Tiber River Basin Fig. 1 . The long-term experiment
involves: the generation of a high-fidelity DTM;
the production of a revised, 1:25,000 scale landslide
inventory map; the acquisition of lithological, hydrological,
structural and land-use data at 1:25,000 or
1:10,000 scale.
A detailed digital representation of terrain was
generated from contour lines obtained from 1:25,000
scale topographic maps. From a 25=25 m DTM
Ž .totalling 6.5 million heights , nearly 20,000 slopeŽ
.units were generated Fig. 6A . For each slope-unit,
24 morphometric parameters were automatically
computed or subsequently derived.
Lithological, bedding-plane, and landslide inventory
maps were prepared through an extensive interpretation
of 1:33,000 scale, black and white, aerial
photographs and, limited to the outcrop of lake and
continental deposits, of 1:13,000 scale color aerial
Ž .photographs Fig. 6B . Landslides were classified
into shallow failures and deep-seated movements, of
Ž .certain or uncertain identification Fig. 6C . The
lithological map was obtained updating the available
geological maps, at 1:100,000 scale or larger. Attention
was paid to the identification of rock types
particularly prone to landslides, differentiating clay
rich units from more competent rocks. Bedding attitude,
an important factor in controlling landslide
Žtypes and pattern in the region Guzzetti et al.,
.1996 , was mapped identifying areas of constant
Žbedding attitude with respect to the local slope Fig.
.6D . The lithological, bedding attitude and landslide
maps were locally checked against detailed surveys
ŽCarrara et al., 1991; Barchi et al., 1993; Toppi,
1993; Cardinali et al., 1994; Lambrugo and Lattuada;
.1996 . Land-use was obtained assembling the existing
maps at 1:10,000 and 1:25,000 scale.
Lithological, geological, structural, geomorphological
and land-use data are available for the entire
area. However, only for the northernmost part of the
basin, covering about 1132 km2
, thematic data are
validated. For each of the 5598 slope-units pertaining
to this portion of the basin, the percentage of unstable
area was computed adding all deep-seated landslide
area existing in each unit. Area of uncertain
landslides was weighted by a factor of 0.7. Shallow
failures were not taken into consideration.
Ž .Small slope-units less than 10 ha were considered
stable if the total landslide area was less than
Ž .10%. Large slope-units larger than 40 ha were
classified as landslide-bearing if landslide deposits
exceeded 2.5% of the area. For slope-units of interŽ
.mediate size 10–40 ha , the threshold value was set
to 5%.
As for the Carpina basin, selected subsets of the
60 input variables were replaced by their most signifTable
11
Upper Tiber River basin. Classification of stable and unstable slope-units by discriminant analysis
Slope-units correctly classified: 72.0%.
Actual Group No. of unique condition units Predicted group membership
Ž . Ž .Group 1 stable units Group 2 unstable units
Ž .Group 1 stable units 3502 2522 980
Ž .Group 2 unstable units 2096 587 1509
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 203
Ž .Fig. 7. Upper Tiber River basin. Landslide hazard assessment by discriminant analysis on 5598 slope-units. Hazard levels are: 1 0%–40%
Ž . Ž . Ž . Ž . Ž .low ; 2 40%–60% intermediate ; and 3 60%–100% high .
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216204
Ž .icant principal components PC . Factors were computed
for morphometrical and structuralrbedding attitude
variables.
Using the presencerabsence of landslides as the
grouping variable, stepwise discriminant function was
applied to the 5598 slope-units. For a preliminary
prediction of deep-seated landslide hazard, 40 variŽ
.ables entered the discriminant function Table 10 .
Of these, 12 are lithological, 15 morphometrical,
eight express structure or bedding plane attitude, and
five refer to land-use. Variables reflecting slope morphometry
and attitude of bedding were the most
powerful in classifying stable and unstable units with
Ž .a success equal to 72% Table 11 . The probability
of slide occurrence, grouped into three classes, is
displayed in Fig. 7.
The preliminary hazard assessment assigned 42%
of the territory of the Upper Tiber River basin to the
Žhigh probability class, 28% to the intermediate ‘‘un.defined’’
class, and 30% to the low hazard class.
Frequencies of landslide area in each class are 28%,
11% and 4%, respectively.
5. Discussion
Landslide hazard evaluation and mapping rely on
a rather complex body of knowledge of slope movements
and on few basic assumptions, widely accepted
among earth scientists. Ideally, such assumptions
form the conceptual framework within which
the ‘‘rationale’’ on slope movements is applied,
regardless of the hazard evaluation method, the mapping
unit, the scale of the analysis, or the goal of the
investigation. Unfortunately, due to operational and
conceptual constraints, the task is not always feasible
or possible.
Major constraints include: systematically identifying
landslide deposits; correctly understanding the
causes and triggering mechanisms of slope-failures;
obtaining adequate information on the relevant geological,
geomorphological, hydrological, climatological,
etc. instability factors; selecting the most
suitable mapping unit and predictive model; and
acquiring appropriate techniques and tools for data
analysis and modelling.
These constraints pose severe limitations on the
evaluation of landslide hazard. Lack of understanding
and recognition of the main causes of landsliding
prevents any successful hazard evaluation. Deficiency
of adequate information on the instability
factors affects the reliability and effectiveness of the
forecast. Selection of a mapping unit and of a modelling
method affect the way uncertainties in the
input data are dealt with, as well as the model fit and
its reliability. Inadequacy of GIS and modelling software
limits the reliability of the forecast and jeopardize
the practical application of any model.
Some of these limitations can be overcome; others
pose more severe conceptual constraints. The conceptual
limitations and the operational difficulties
will now be discussed in the light of the experience
gained from the Umbria–Marche project.
5.1. Constraints in the application of basic principles
Geomorphological information remains largely
descriptive. Its subjectivity makes it somewhat unsuitable
for engineers, policy makers or developers
in planning land resources, when mitigating the effects
of geological hazards. In the past two decades,
countless landslide maps were produced by geomorphologists.
The reliability of these maps is poorly
documented. This introduces a factor of uncertainty
that cannot readily be evaluated and incorporated in
the subsequent phases of data modelling and in the
transfer and use of this information.
Identification and mapping of landslide deposition–erosional
areas, the first step in any landslide
Ž .hazard assessment Brabb, 1984; Hansen, 1984 , are
indeed difficult, error prone, and subject to uncerŽtainties
largely untested Fookes et al., 1991; Carrara
.et al., 1992; van Westen, 1993 . This is particularly
true for old or inactive slope movements, for landslides
that leave faint morphological signs, for failures
in forested areas, on slopes intensively ploughed,
Žand in recently urbanized areas Brabb, 1984;
Guzzetti and Cardinali, 1989; Brabb, 1995; Hutchin.son,
1995 . Inadequacy in mapping the full extent of
slope movement limits the reliability of any hazard
assessment, particularly, if errors are systematic in
Žrecognizing some types of slope processes Brabb,
.1995 .
Reconnaissance inventories provide a fairly unbiased
spatial coverage but generally lack information
Žon the time of occurrence of failures Cotecchia,
.1978 . This information is available only where a
reconnaissance inventory is completed shortly after a
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 205
particularly damaging meteorological or seismic
event. Attempts at evaluating the ‘‘goodness’’ of
reconnaissance inventories at different scales proved
that errors can be large and related to the experience
of the interpreter, the scale of the inventory and of
the aerial photographs, and the time available for the
Ž .study Carrara et al., 1992; van Westen, 1993 .
Ž .Historical records landslide time-series constitute
the main source for every estimate of landslide
recurrence. Drawbacks of the historical analysis include:
lack of spatial completeness, resolution and
precision; and an undefined over-estimate of events
which caused damage to human structures as opposed
to an under-estimate of failures, even large,
Žwhich took place in unpopulated areas Guzzetti et
.al., 1994; Ibsen and Brunsden, 1996 . The Umbria–
Marche archive inventory largely confirms such bi-
ases.
Identification and mapping of a suitable set of
Ž .instability factors thematic mapping bearing a relationship
with slope failures — such as surface and
bedrock lithology and structure, bedding attitude,
seismicity, slope steepness and morphology, stream
evolution, groundwater conditions, climate, vegetaŽtion
cover, land-use and human activity Carrara et
.al., 1995; Hutchinson, 1995 — require an a priori
Žknowledge of the main causes of landsliding Schus.ter
and Krizek, 1978; Crozier, 1986 . The availability
of thematic data largely varies depending on the
type, scale, and technique for data acquisition. As for
landslide maps, the quality of this information remains
largely undefined. Where thematic data are
gathered manually, by field survey or through the
Žinterpretation of remote sensing data aerial pho.tographs
or satellite images , mismatch between difŽ
.ferent interpreters can be large Carrara et al., 1992 .
Recent visual estimates of the mismatch between
geological maps at different scale and of different
dates in the Umbria Region revealed large discrepancies.
Attempts to evaluate the quality of digital terrain
models, widely used in describing landscape
Žmorphology for slope stability Carrara, 1983; Carrara
et al., 1991, 1995; Pike, 1988; van Westen,
.1993; Dietrich et al., 1995; Mark and Ellen, 1995 ,
proved that even where data are gathered and manipulated
automatically or semi-automatically, errors and
uncertainties can be greater than commonly expected
Ž .Carrara et al., 1997 .
As previously pointed out, predictive hazard models
assume that landslides in the future will take
place under the conditions which led to past and
present instability. This assumption holds true for
factors, such as bedrock lithology, structure and
morphology, which are time-invariant within the
temporal framework of the model. Conversely, it
cannot be extended to environmental factors which
vary with time, such as land-use, human activity and
even climate. Climatological conditions that triggered
mass movements in the past may differ from
present climate, in a way and for an amount that is
usually unknown in quantitative terms. Information
on land-use and human activity can be obtained for
both historical and modern time; however, such instability
factors may vary rapidly in response to
environmental changes or economical needs. Thus,
the use of past environmental settings exhibiting
large temporal variability may lead to erroneous
predictions.
The estimate of the relative contribution of each
physical factor in generating slope-failures, and the
classification of the land surface into domains of
Ž .different hazard degree hazard zoning are a crucial
step. When the main instability factors leading to
slope failure are identified, the understanding of their
complex interactions becomes the next difficult issue
to be accomplished, particularly over large regions
Ž .Hutchinson, 1995; Guzzetti et al., 1996 . AdditionŽally,
the role played by factors leading to i.e., rock
.type, clay content, bedding attitude, etc. or bearing
Ž .i.e., land-use, vegetation cover, etc. a functional
relationship to landslide occurrence in one area may
Žturn out to be very different in other areas Guzzetti
.et al., 1996 .
Quite surprisingly, investigators have invested little
time in the acquisition of terrain information and
in testing innovative mapping techniques. Likewise,
few attempts have been made at the ‘‘regionalization’’
of site specific information and models. This
has limited the use of geotechnical and site specific
Ždata on regional hazard modelling Nieto, 1989;
.Hutchinson, 1995 .
5.2. Selection of a mapping unit
ŽAs previously mentioned, various methods gridcells,
terrain units, unique-condition units, slope-units
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216206
.and topographic units have been proposed and tested
to partition the landscape into mapping units. Each
method has advantages and drawbacks which can be
either enhanced or controlled depending on the hazard
assessment approach used.
Automated cartography and GIS-based spatial operations
have demonstrated their usefulness in partitioning
a territory into mapping units according to
various criteria without the constraints due to traditional,
time-consuming manual work. When appropriate
software is available, the investigator can readily
choose among grid-cells, unique-conditions,
slope-units or other terrain subdivisions without investing
a great deal of time and tedious work. Hence,
the major issue is no longer how to create the
sampling unit, but which unit is the most suitable for
the type of problem to be investigated. Actually, as it
was demonstrated in the Carpina basin, more than
one mapping unit can be tried and the most suitable
for the problem at hand can be used.
Advantages and limitations of grid-cells are
known. Owing to the matrix form of the grid data,
computer implementation is simple and processing is
fast. Since data are regularly spaced, sampling constraints
are relaxed. Drawbacks lay in the absence of
any relation between grid-cells and geological, geomorphological,
or any other terrain information.
The tendency to use smaller and smaller grid-cells
appears unjustified. Spatial inaccuracy is partially
reduced but to cover even small areas an overwhelming
number of grid-cells is required, leading to unmanageable
computer problems and numerical instability
when data have to be processed by statistical
techniques.
Terrain units, which have long been applied in
many land resources investigations on a wide range
of scales, fully exploit the investigator skill in detecting
in the field or on aerial photographs the complex
relations existing between slope-failure and the geomorphological
context. The approach, emphasizing
cataloging, provides much information about the land
but it does little to measure the functional relationships
between instability factors. The main drawback
lays in the intrinsic subjectivity of the method. Different
investigators may classify any given region in
different ways. To partition the landscape into geomorphological-units,
maps portraying all the different
forms and processes are used. These maps use a
variety of classification schemes which are always
complex and frequently inconsistent; conceptually or
spatially.
Unique-condition units are appropriately applied
where it is conceptually or operationally difficult or
impossible to pre-define a physically based mapping
unit or domain. They perform well where thematic
Ž .information layers completely ‘‘fill’’ the territory.
ŽProblems arise where linear features i.e., fault lines
.or lithological boundaries are used in the analysis.
The problem arose in the Umbria–Marche experiment,
where lithological boundaries were buffered to
capture the instability effect of contrasting lithology.
Another weakness is the inherent subjectivity in
factor classification that has to be performed prior to
map overlay. Additionally, by overlying more than
Ž .just few maps five to seven , each with a relatively
Ž .small number of factors 3–10 , thousands of small
domains are generated. Most of these areas result
from errors in data collection and digitisation and are
statistically meaningless. They can be cancelled out
by applying some filtering technique, however loosing
in objectivity in the process. Other areas may
Ž .reflect rare small in size but physically meaningful
Ž .conditions ‘‘outliers’’ that cannot be eliminated.
Since some of the factors into which each input layer
is classified may turn out to be not very significant,
it would be wise to restart the whole map overlay
operation after the reclassification of such layers.
This makes the procedure rather cumbersome.
Since a clear physical relationship exists between
landsliding and the fundamental morphological elements
of a hilly or mountain region, namely drainage
and divide lines, the slope-unit technique seems appropriate
for landslide hazard assessment. In the
Upper Tiber River basin, it was observed that problems
arise where intra-mountain basins or large open
valleys are present. In these areas, slope-units do not
match with the local geomorphological setting bearing
on slope instability. Slope-units can be resized
according to the prevailing failure type and dimension,
partitioning a river basin into nested subdivisions,
coarser for larger landslides and finer for
Ž .smaller failures cf. Carpina basin . Despite this
capability, the tendency of slope-units to identify
relatively large areas into stability types rather than
resolve fine-scale patterns of instability conditions,
limits the applicability of this approach for small,
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 207
shallow landslides such as soil-slips and debris flows
Ž .Montgomery and Dietrich, 1994 .
To overcome this limitation, slope-units can be
further subdivided into topographic units. Due to the
physical relationship between topography and surface
and sub-surface hydrology, the approach appears
most appropriate to predict surface saturation
and the occurrence of topographically controlled
landslides, such as soil-slip–debris flows, in soil
Žmantled topography Montgomery and Dietrich,
.1994 . Limitations refer to: the availability of detailed
contour lines that accurately portray topography,
seldom available over large areas; the assumption
that sub-surface hydrology is directly related to
surface topography; and the related inadequacy to
investigate deep-seated, complex slope failures.
It should be pointed out that too often, the selection
of the mapping unit appears guided more by the
Žtype of software available i.e., raster vs. vector GIS,
.DTM modelling software, etc. , rather than by the
specific requirements of the geomorphological data
to be analysed.
5.3. Landslide hazard modelling
As previously discussed, all methods proposed
and tested to evaluate landslide hazard fall into a few
main categories, namely: direct geomorphological
mapping; analysis of landslide inventories; heuristic
or index based models; functional or statistical models;
and geotechnical or physically based models.
The goodness of direct methods for landslide
hazard mapping relies on the ability of the investigator
to estimate actual and potential slope failures,
taking into account a large number of instability
factors detected in the field or on aerial photographs
Ž .Verstappen, 1983 . In addition, local or peculiar
slope instability conditions can be identified and
assessed. Drawbacks concern the high subjectivity
that characterizes all phases of the geomorphological
investigation. Moreover, the degree of uncertainty
can not be readily evaluated, making it difficult, or
impossible, to compare landslide hazard maps produced
by different investigators, even if they applied
Žthe same ranking criteria Godefroy and Humbert,
.1983 .
Isopleth maps can be readily produced for large
areas; they provide a general overview of landslide
occurrence and may be useful when portraying the
distribution of many failures triggered by severe
storms or seismic events. However, such fairly popular
maps are founded upon the wrong assumption
that landslide presencerabsence is a spatially continuous
variable. Thus, isopleth maps do not incorporate
any relation between slope-failure and landscape,
namely, stable areas, such as flat terrain, can
be ranked as unstable, or isolated outcrops of landslide-prone
clayey rocks may well be classified as
stable.
The reliability of heuristic methods depends
largely on how well and how much the investigator
understands the geomorphological processes acting
upon the terrain. Since this knowledge can be formalized
into rules, the method could take into account
local geomorphological variability or specific
conditions leading to slope failures. Major limitations
refer to the fact that in most cases the body of
knowledge available on the causal relations between
environmental factors and landslides is inadequate
and, most importantly, is essentially dependent on
the experience of the investigator. At present, maps
obtained by this method cannot be readily evaluated
in terms of reliability or certainty. Additionally,
landslide hazard is not directly expressed in terms of
probability, limiting the use for risk evaluation and
economic estimates.
Statistical or probabilistic approaches are based
on the observed relationships between each factor
and the distribution of landslides. Since the instability
determinants and their interrelations are evaluated
on a statistical basis, hazard evaluation becomes an
operation as objective as possible. Black-box models
are conceptually simple but, due to the great complexity
in identifying the slope-failure processes and
the difficulty in systematically collecting the different
factors related to landsliding, the task of creating
a geomorphological predictive model enabling actualrpotential
unstable slopes to be identified over
large areas, is difficult operationally. Errors in mapping
past and present landslides will exert a large
and not readily predictable influence on statistical
models, particularly if errors are systematic in not
Ž .recognizing specific landslide types Brabb, 1995 .
Additionally, being data-driven, a statistical model
built up for one region cannot readily be extrapolated
to the neighboring areas.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216208
Physically based models, being process-driven,
may provide significant insight on the causes and
triggering factors of landslide movements. Their limitations
include: the use of too simple models of
instability evaluation; the fact that very few geotechnical
data can be collected over even small regions at
Ž .reasonable cost Mulder, 1991; Hutchinson, 1995 ;
and the spatial variability of geotechnical factors that
is not controlled for. Additionally, reliable mechanical
models are not yet available for several types of
Žstructurally complex rock units Esu, 1977; Nieto,
.1989 . Despite such limitations, physically based
models show promise for investigating, in quantitative
terms, the influence of instability factors on
Žlandslides Okimura and Kawatani, 1987; Mont.gomery
and Dietrich, 1994; Dietrich et al., 1995 , for
modelling topographically controlled, shallow slope
Ž .failures, for predicting simulate the potential run
Žout path of debris flows Ikeya, 1981; Takahashi et
.al., 1981; Mark and Ellen, 1995 , or where the
elements at stake justify extensive, site-specific investigations,
over limited areas.
5.4. GIS-based statistical modelling
The experience gained from the application, at
various scales, of GIS-based statistical models to
landslide hazard assessment in the Umbria and
Marche Regions allows a few considerations to be
made.
Nowadays, owing to the ever-increasing capabilities
of hardware and software technologies, electronic
geographical data processing is becoming a
common tool in a wide range of research activities
related to the assessment and control of landsliding
Žor other natural catastrophes Wadge, 1988; Soeters
et al., 1991; van Driel, 1991; Carrara, 1993; Carrara
and Guzzetti, 1995; van Westen, 1993; Bonham.Carter,
1994; Kovar and Nachtnebel, 1994 .
A crucial issue in hazard assessment remains that
of the input data, which are fundamentally inadequate
in quantity and quality for the task to be
accomplished. With the diffusion of GIS-driven techniques
the basic data did not change significantly.
The most relevant progress refer to the morphometric
variables derived from DTMs, which in the future
might allow simulating the visual recognition of the
topographic form, the latter being a fundamental
element in any geomorphologic analysis of landslide
Židentification Ollier, 1977; Rib and Liang, 1978;
Pike, 1988; Carrara, 1993; Carrara et al., 1995;
.Howard, 1994; Montgomery and Dietrich; 1994 .
Empirical and process-based models for estimating
Žthe spatial variation in soil attributes chiefly thick.ness
from DTMs proved efficient for predicting
Ž . Žshallow slope instability soil-slip Moore et al.,
.1993; Dietrich et al., 1995 . Attempts at automatically
combining lithological and bedding attitude
Ždata with morphometric parameters terrain gradient
.and aspect to classify the territory into structural or
hydrogeological domains, proved quite satisfactory
Žfor detailed investigations cf. Tescio and Carpina
.basins , but performed less efficiently at the regional
Ž .scale cf. Upper Tiber River basin . Also, the application
of remote sensing techniques to aerial photographs
and satellite imagery to obtain significant
and cost-effective information on instability factors,
remains a future resource, whose potential, yet to be
determined and exploited, appears more promising in
Žunpopulated areas of developing countries Bruns.den,
1993 .
Besides this task, which should constitute a major
research effort in the coming years, more attention
should be paid to the many sources of errors and
uncertainties associated with data acquisition and
manipulation. It has clearly been demonstrated that
landslide mapping is the most error-prone phase of
Žthe whole operation Carrara et al., 1992; van Westen,
.1993 . Likewise, virtually all the instability factors
collected in the field or derived in laboratory through
GIS manipulation, are affected by inaccuracies or
errors whose magnitude cannot readily be estimated
or controlled during the subsequent phase of data
Ž .analysis or modelling Walsh et al., 1987 .
Environmental processes are highly non-linear and
most environmental thematic information is spatially
correlated. Additionally, environmental variables exŽ
.hibit large variances, and peculiar rare or small , but
Ž .significant values outliers . Thus, hazard models
based on the statistical analysis of environment variables
may be affected by large errors and wrong
assumptions, or generate questionable or equivocal
outcomes. Discriminant and regression analyses
would require data derived from a normally distributed
population, an assumption frequently vioŽlated.
In addition, a mixture of continuous i.e.,
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 209
. Želevation and categorical i.e., presencerabsence of
.a rock type variables leads to a solution which is
generally not optimal, namely, it does not minimize
the probability of incorrect predictions. Most importantly,
when the variable set includes good and poor
predictors, that is, some of the input variables do not
bear a clear physical relationship with mass movement,
a statistical stepwise procedure may generate a
linear combination of both types of variables whose
interpretation will eventually give difficult, unreliable
or even meaningless results. Better results could
be obtained by entering into the model only the
variables that the investigator assumes to be the most
significant. However, in general different investigators
will not select the same variables; so the model
becomes dependent on the skill and experience of
the analyst. Since input factors are invariably correlated,
the technique of entering all the available
variables can produce even worse outcomes with
some variables characterized by meaningless coeffiŽ
.cients Carrara et al., 1995 . Hence, correlation between
variables should always be carefully checked
Ž .Bonham-Carter, 1994 . Additionally, variables correlated
to instability conditions in one area may give
rise to stability conditions in a different physiographic
environment. In the Umbria–Marche project
this was found to be the case for land-use data. Thus,
experience on factors bearing a functional relationship
on slope instability should be used with care.
Where input information is highly generalized,
the reliability and usefulness of any predictive model
may be limited. The reconnaissance evaluation carŽried
out for the Umbria and Marche Regions Fig.
.4B; Tables 1 and 2 indicates that a statistical model
based on a set of broad factors which do not reflect
the great variability of conditions leading to slope
failures over a wide region may be fairly successful
in terms of predictive power, but lack adequate
spatial resolution for planning purposes. In addition,
by grouping very different landslide types into a
single class, the model may become physically unre-
liable.
In discriminant analysis and logistic regression,
high and low values of membership probability indicate
hazardous and safe mapping units, respectively.
Values close to 0.5 do not provide any additional
information with respect to the input landslide map.
If this is the case for many sampling units: a large
portion of the region under study will turn out to be
‘‘unclassified’’. Hence, the model could be statistically
sound, but of limited application. In the Umbria–Marche
project, models prepared at various
scales, classified in the intermediate hazard class
Ž .40%–60% probability between 15% and 28% of
the territory.
Any model is unable to correctly classify all
mapping units. If this should occur, the model, once
again, would not provide more information than the
input inventory map. However, misclassifications
Ž .have very different meanings, namely: a a mapping
unit is predicted as unstable, but no landslides were
Ž .found on it by the surveyor; b a mapping unit is
predicted as stable, but slope-failures were mapped
on it. Under the hypothesis that the model is reliable,
the first case is the result of inaccurate mapping or of
a failed mass concealed by erosion or farming activity.
The second case indicates either wrong mapping
or a model which lacks the factors that caused a
landslide in that specific or unique environmental
setting. Regardless of the causes, the first type of
mismatch indicates a mapping unit that has to be
interpreted as hazardous, with a high probability of
failure in the future; while the second is equivocal
and requires further investigation. Ideally, a good
model should minimize the latter type of misclassification.
Conversely, all multivariate procedures yield
an approximately equal proportion of the two types
Žof incorrect predictions cf. Carpina basin; Tables 6
.and 8 .
Owing to these pitfalls, hazard assessment and
mapping by statistical modelling are intrinsically uncertain
operations which nowadays are taking advantage
of the opportunities provided by new technologies,
such as GIS, but are still requiring new efforts
for improving both data quality and model reliability.
Lastly, after a ‘‘black-box’’ model has been built
up and tested, results have to be interpreted in the
light of the local geomorphological setting. This is a
crucial step that often represents one of the most
difficult phases of landslide hazard evaluation.
5.5. Model combination and application
Where various types of landslides take place,
distinct hazard evaluations should be prepared. This
was attempted in the Carpina basin, where different
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216210
hazard models for the three prevailing types of slope
failures were prepared, and in the Upper Tiber River
basin, where provisional hazard modelling was confined
to deep-seated failures. Unfortunately, even if
all hazards can be singled out, assessed and mapped
separately — and this may not always be feasible —
it remains to be understood how to combine them
into a singe hazardrrisk map portraying the spatial
distribution of all endangered areas. At present, it is
not even clear if this is appropriate.
Two conflicting approaches can be followed.
Maps portraying different types of hazard are kept
separate and no attempt is made to compile a general,
holistic hazard evaluation. Alternatively, the
different forecasts are ranked and portrayed in a
single map. Both approaches have advantages and
limitations. The former is preferred where multiple
Ž .hazard evaluations landslide as well as others are
Ž .available Seely and West, 1990; Brabb, 1995 . The
benefit lays in presenting ‘‘simple’’ evaluations, allowing
for various interpretations by decision-makers
and planners, some of which may not be known to
the author of each single hazard assessment. The
later is favored where direct geomorphological hazard
evaluation is attempted. Its main advantage refers
to the possibility of incorporating into a generalized
hazard model or map some of the complex interactions
existing among single hazard evaluations
ŽHumbert, 1977; Godefroy and Humbert, 1983;
Kienholz et al., 1983, 1984; Zimmerman et al.,
.1986 .
A related problem of models combination arises
when more than a single landslide hazard evaluation
is available for the same area, as in the Carpina
basin, where two hazard models based on slope-units
Ž . Ž .Fig. 6A and unique-condition units Fig. 6B were
prepared. This is conceptually equivalent to the situation
where two or more experts are asked for their
opinion on a technical or scientific problem. The
question of which model to prefer or how to combine
different forecasts remains largely unsolved.
Taking the ‘‘worst-case approach’’, that is, choosing
for each site or mapping unit the most catastrophic
forecast, may be too conservative. Also, the choice
Žof the ‘‘simplest’’ and cheapest estimate Hutchin.son,
1995 may not be appropriate. A more sensible
approach would consist in the critical analysis of the
underlying assumptions of each hazard assessment
— if these are clearly stated — and in the evaluation
of external sources of information, such as economical
or other practical constrains.
A still different problem is related to the aggregation
of the results of a landslide hazard evaluation
Ž .one or more hazard models or maps prepared using
Žsome sort of mapping unit grid-cells, slope-units,
.unique-condition units, etc. into a different partition
of the territory, most commonly an administrative or
political subdivision. This step, requested by decision
makers for regional planning purposes, may be
subjective and conceptually troublesome.
Despite the largely acknowledged need of new
tools for planning and policy making, no general
agreement has been reached among earth scientists
and decision makers on the goals and possible use of
landslide hazard evaluations. This may explain why,
despite the fact that numerous models have been
proposed and tested in a variety of physiographic
environments, only in a few cases has knowledge on
landslide hazard become an integral part of building
codes, planning policies, or civil protection regula-
tions.
To limit the discussion to functional models, such
as those presented for the Umbria–Marche territory,
models for the evaluation of landslide hazard can be
prepared with two distinct goals. The first is ‘‘scienŽ
.tific’’ explanatory and aims to explain landslide
phenomena. In the hope of producing more reliable
predictions, it considers a landslide hazard assessment
as a scientific theory, and makes all efforts
Ž .to prove it faulty Popper, 1959 . In this view,
uncertainties and the analysis of errors and residuals
represent useful tools for model refinement and calibration
and for a better understanding of slope
phenomena. The second is an ‘‘engineering’’
Ž .pragmatic approach that, based on the available
information on local slope failures, aims at producing
the ‘‘best’’ possible predictive model. Little
effort is made to improve the understanding on
landsliding. Uncertainties, errors and peculiar conditions,
that make any model somewhat unsuitable for
planners and decision makers, are dealt with pragmatically
by introducing a safety factor. Ideally, both
approaches are needed for a comprehensive evaluation
of landslide hazard. Unfortunately, as has been
discussed, due to practical constraints and conceptual
limitations, this is not often the case.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 211
6. Concluding remarks
Landslides are among the most hazardous natural
disasters. Government and research institutions
worldwide have attempted for years to assess landslide
hazard and risk and to portray its spatial distribution.
Several different methods have been
proposed and tested in a variety of physiographic
environments, with different results. In the Umbria
and Marche Regions, attempts at testing the proficiency
and limitations of multivariate statistical techniques
and of different methodologies for dividing
the territory into suitable terrain units have been
completed, or are in progress, at various scales.
These experiments showed that, despite the operational
and conceptual limitations, landslide hazard
assessment may indeed constitute a suitable, cost-effective
aid to land-use planning, an aid to a sustainable
development both in developed and developing
countries.
Evaluation of landslide hazard aims at the solution
of a complex, ‘‘multi-dimensional’’ problem
that requires expertise pertaining to the earth sciŽences
specifically geomorphology and engineering
.geology , statistics, computer science, physics, information
technology and economics.
The definition of landslide hazard remains a
largely open, ill-formalized question. Landslides are
phenomena with complex feedback varying in scale
from the local to the regional. Their geomorphological
and economic impact ranges from the very short
to the very long term. Despite efforts, landslide
phenomena are still poorly understood, particularly
at the regional scale. Additionally, their interactions
with the economic and human sphere remains a
novel problem to the earth scientists. Knowledge on
slope processes appears insufficient for a comprehensive
and exhaustive evaluation of landslide haz-
ard.
Industrialized societies and developing countries
face increasingly complex problems of planning and
policy making. These are different from the traditional
problems of both pure and applied science
Ž .Funtowicz and Ravetz, 1995; Murck et al., 1997 .
As regards to landslide hazard evaluation, on one
side geomorphology is unable to provide wellfounded
theories for hazard assessment, and on the
other side, environmental issues and policy decisions
challenge geomorphologists with difficult issues. Due
to the uncertainties in data acquisition and handling,
and in model selection and calibration, landslide
hazard evaluation and land-zoning appear out of the
reach of the traditional puzzle-solving scientific approach,
based on experiments and on a generalized
consensus among experts. In general, predictive
models of landslide hazard can not be readily tested
by traditional scientific methods. Indeed, the only
way a landslide predictive map can be validated is
Ž .through time Hutchinson, 1995 . Additionally, as
previously discussed, no general agreement has been
reached on the scope, techniques and methodologies
for landslide hazard evaluation.
Solutions to these challenging problems may come
from a new scientific practice enabling to cope with
large uncertainties, varying experts judgements, and
societal issues risen by hazard evaluation. Within
this framework engineering geomorphology will play
an important role in the future, particularly if geomorphologists
will be able to better formalize and
extend their knowledge on slope processes at different
scales and in different physiographic environ-
ments.
Acknowledgements
The research was supported by CNR-IRPI and
CNR-GNDCI grants. This is CNR-GNDCI paper
number 1876.
References
Ahlberg, P., Stigler, B., Viberg, L., 1988. Experiences of landslide
risk considerations in land use planning in Sweden. 5th International
Symposium on Landslides, Lausanne 2, 1091–1096.
Alexander, E.D., 1989. Urban landslides. Progress in Physical
Geography 13, 157–191.
Alexander, E.D., 1995. A survey of the field of natural hazards
Ž .and disaster studies. In: Carrara, A., Guzzetti, F. Eds. ,
Geographical Information Systems in Assessing Natural Hazards.
Kluwer Academic Publisher, Dordrecht, The Netherlands,
pp. 1–19.
Amadesi, E., Vianello, G., 1978. Nuova guida alla realizzazione
di una carta di stabilita dei versanti. Memorie Societa Geolog-` `
Ž .ica Italiana 19, 53–60, in Italian .
Antoine, P., 1977. Reflextions sur la cartographie ZERMOS et`
bilan des esperiences en cours. Bulletin Bureau des Recherches
Ž . Ž .Geologiques et Minieres 3 2 , 9–20, in French .´ `
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216212
Antonini, G., Cardinali, M., Guzzetti, F., Reichenbach, P., Sorrentino,
A., 1993. Carta Inventario dei Fenomeni Franosi della
Regione Marche ed aree limitrofe. CNR-GNDCI, Publication
Ž .no. 580, 2 sheets, scale 1:100,000. in Italian .
Baker, V.R., 1994. Geomorphological understanding of floods.
Geomorphology 10, 139–156.
Barchi, M., Cardinali, M., Guzzetti, F., Lemmi, M., 1993. Relazioni
fra movimenti di versante e fenomeni tettonici nell’area
del M. Coscerno–M. di Civitella, Valnerina, Umbria. BolletŽ
.tino Societa Geologica Italiana 112, 83–111, in Italian .`
Bernknopf, R.L., Campbell, R.H., Brookshire, D.S., Shapiro, C.D.,
1988. A probabilitic approach to landslide hazard mapping in
Cincinnati, Ohio, with applications for economic evaluation.
Bulletin American Association of Engineering Geologists 25
Ž .1 , 39–56.
Bonham-Carter, G.F., 1994. Geographic Information Systems for
Geoscientists: Modelling with GIS. Pergamon, Ottawa, 398
pp.
Boschi, E., Favali, P., Frugoni, F., Scalera, G., Smaziglio, G.,
1995. Maximum felt intensity in Italy. Istituto Nazionale di
Geofisica. Map at 1:1,500,000 scale.
Bosi, C., 1978. Considerazioni e proposte metodologiche sulla
elaborazione di carte della stabilita. Geologia Applicata ed`
Ž .Idrogeologia, Bari, 13, 246–281, in Italian .
Bosi, C., Dramis, F., Gentili, B., 1985. Carte geomorfologiche di
dettaglio ad indirizzo applicativo e carte di stabilita a base`
geomorfologica. Geologia Applicata ed Idrogeologia, Bari, 20
Ž . Ž .2 , 53–62, in Italian .
Brabb, E.E., 1984. Innovative approaches to landslide hazard
mapping. 4th International Symposium on Landslides, Toronto
1, 307–324.
Brabb, E.E., 1995. The San Mateo County California GIS project
for predicting the consequences of hazardous geologic proŽ
.cesses. In: Carrara, A., Guzzetti, F. Eds. , Geographical Information
Systems in Assessing Natural Hazards. Kluwer Academic
Publisher, Dordrecht, The Netherlands, pp. 234–299.
Ž .Brabb, E.E., Harrod, B.L. Eds. , Landslides: Extent and Economic
Significance. Balkema Publisher, Rotterdam, 385 pp.
Brabb, E.E., Pampeyan, E.H., Bonilla, M.G., 1978. Landslide
susceptibility in San Mateo County, California. US Geological
Survey Misc. Field Studies Map, MF-360, map at 1:62,500
scale.
Brand, E.W., 1988. Special lecture: landslide risk assessment in
Hong Kong. 5th International Symposium on Landslides, Lausanne
2, 1059–1074.
Brand, E.W., Styles, K.A., Burnett, A.D., 1982. Geotechnical land
use maps for planning in Hong Kong. 4th Congress International
Association Engineering Geologists, New Delhi, 1,
145–153.
Brunsden, D., 1993. Mass movements; the research frontier and
beyond: a geomorphological approach. Geomorphology 7, 85–
128.
Burnett, A.D., Brand, E.W., Styles, K.A., 1985. Terrain classification
mapping for a landslide inventory in Hong Kong. 4th
International Conference and Field Workshop on Landslides,
Tokyo, 63–68.
Campbell, R., 1973. Isopleth map of landslide deposits, Point
Dume Quadrangle, Los Angeles County, California; an experiment
in generalising and quantifying areal distribution of
landslides. US Geological Survey Miscellaneous Field Studies
Ž .Map, MF-535 Scale 1:24,000 .
Cardinali, M., Galli, M., Guzzetti, F., Reichenbach, P., Borri, G.,
1994. Relazioni fra movimenti di versante e fenomeni tettonici
nel bacino del Torrente Carpina. Geografia Fisica e Dinamica
Ž .Quaternaria 17, 3–18, in Italian .
Carrara, A., 1983. A multivariate model for landslide hazard
evaluation. Mathematical Geology 15, 403–426.
Carrara, A., 1988. Drainage and divide networks derived from
Ž .high-fidelity digital terrain models. In: Chung, C.F. Ed. ,
Quantitative Analysis of Mineral and Energy Resources. ReiŽ
.del, Dordrecht, pp. 581–597, NATO-ASI Series .
Carrara, A. et al., 1989. Landslide hazard mapping by statistical
methods: a ‘‘black-box’’ model approach. In: Siccardi, F.,
Ž .Brass, R. Eds. , International Workshop on Natural Disasters
in European–Mediterranean Countries, Perugia, 27 June–1
July 1989. CNR-US NFS, pp. 427–445.
Carrara, A., 1993. Potentials and pitfalls of GIS technology in
assessing natural hazards. In: Reichenbach, P., Guzzetti, F.,
Ž .Carrara, A. Eds. , Geographical Information Systems in Assessing
Natural Hazards — Abstracts. CNR, Perugia, pp.
128–137, September 20–22, 1993.
Ž .Carrara, A., Guzzetti, F. Eds. , 1995. Geographical Information
Systems in Assessing Natural Hazards. Kluwer Academic
Publisher, Dordrecht, The Netherlands, 353 pp.
Carrara, A., Cardinali, M., Detti, R., Guzzetti, F., Pasqui, V.,
Reichenbach, P., 1991. GIS Techniques and statistical models
in evaluating landslide hazard. Earth Surface Processes and
Landform 16, 427–445.
Carrara, A., Cardinali, M., Guzzetti, F., 1992. Uncertainty in
assessing landslide hazard and risk. ITC Journal, The Netherlands
2, 172–183.
Carrara, A., Cardinali, M., Guzzetti, F., Reichenbach, P., 1995.
GIS technology in mapping landslide hazard. In: Carrara, A.,
Ž .Guzzetti, F. Eds. , Geographical Information Systems in Assessing
Natural Hazards. Kluwer Academic Publisher, Dordrecht,
The Netherlands, pp. 135–175.
Carrara, A., Bitelli, G., Carla, R., 1997. Comparison of techniques`
for generating digital terrain models from contour lines. InterŽnational
Journal Geographical Information System 11, in
.press .
Chung, Ch.F., Fabbri, A.G., 1995. Multivariate regression analysis
for landslide hazard zonation. In: Carrara, A., Guzzetti, F.
Ž .Eds. , Geographical Information Systems in Assessing Natural
Hazards. Kluwer Academic Publisher, Dordrecht, The
Netherlands, pp. 107–142.
CoGeo, 1994. Progetto AVI — Rapporto di Sintesi: Umbria:
Ž .CNR-GNDCI, 49 pp. in Italian .
CoGeo, 1994. Progetto AVI — Rapporto di Sintesi: Marche:
Ž .CNR-GNDCI, 38 pp. in Italian .
Cooke, R.U., Doornkamp, J.C., 1974. Geomorphology in environmental
management. Clarendon Press, Oxford, 413 pp.
Cotecchia, V., 1978. Systematic reconnaissance mapping and
registration of slope movements. Bulletin International Association
Engineering Geology 17, 5–37.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 213
Crozier, M.J., 1984. Field assessment of slope instability. In:
Ž .Brunsden, D., Prior, D.B. Eds. , Slope Instability. Wiley,
Chichester, pp. 103–142, Chap. 4.
Crozier, M.J., 1986. Landslides: Causes, Consequences and Environment.
Croom Helm Pub, London.
Cruden, D.M., 1991. A simple definition of a landslide. Bulletin
International Association of Engineering Geology 43, 27–29.
Cruden, D.M., 1997. Estimating the risk from landslide historical
Ž .data. In: Cruden, D.M., Fell, R. Eds. , Landslide Risk Assessment.
Proc. Inter. Workshop on Landslide Risk Assessment,
Honolulu, 19–21 February 1997 Balkema, Rotterdam, pp.
177–184.
Cruden, D.M., Fell, R., 1997. Landslide Risk Assessment. Proceedings
International Workshop on Landslide Risk Assessment,
Honolulu, 19–21 February 1997 Balkema, Rotterdam,
371 pp.
DeGraff, J.V., 1985. Using isopleth maps of landslides deposits as
a tool in timber sale planning. Bulletin American Association
of Engineering Geologists 22, 445–453.
Degraff, J.V., Agard, S.S., 1984. Defining geologic hazards for
natural resources management using tree-ring analysis. EnviŽ
.ronmental Geology and Water Sciences 6 3 , 147–155.
Delaunay, J. 1981. Carte de France des zones vulnerables a des`
glissements, ecroulements, affaissements et effrondrements de´
terrain. Bureau de Recherches Geologiques et Minieres. 81´ `
Ž .SGN 567 GEG, 23 pp. in French .
Dietrich, E.W., Reiss, R., Hsu, M.-L., Montgomery, D.R., 1995.
A process-based model for colluvial soil depth and shallow
landsliding using digital elevation data. Hydrological Process
9, 383–400.
Dikau, R., Brunsden, D., Schrott, L., Ibsen, M.-L., 1996. Landslide
recognition. Identification, movements and causes. Wiley,
Chichester, England, 251 pp.
Dunne, T., 1991. Stochastic aspects of the relations between
climate, hydrology and landform evolution. Transaction
Japanese Geomorphological Union 12, 1–24.
Einstein, H.H., 1988. Special lecture: landslide risk assessment
procedure. Proceedings 5th International Symposium on Landslides,
Lausanne 2, 1075–1090.
Einstein, H.H., 1997. Landslide risk — systematic approaches to
assessment and management. In: Cruden, D.M., Fell, R.
Ž .Eds. , Landslide Risk Assessment. Proc. Inter. Workshop on
Landslide Risk Assessment, Honolulu, 19–21 February 1997
Balkema, Rotterdam, pp. 25–50.
Esu, F., 1977. Behavior of slopes in structurally complex formations.
International Symposium on the Geotechnics of Structurally
Complex Formations, Capri 2, 292–304.
Evans, S.G., 1997. Fatal landslides and landslide risk in Canada.
Ž .In: Cruden, D.M., Fell, R. Eds. , Landslide Risk Assessment.
Proc. Inter. Workshop on Landslide Risk Assessment, Honolulu,
19–21 February 1997 Balkema, Rotterdam, pp. 185–
196.
Fell, R., 1994. Landslide risk assessment and acceptable risk.
Ž .Canadian Geotechnical Journal 31 2 , 261–272.
Fookes, P.G., Dale, S.G., Land, J.M., 1991. Some observations on
a comparative aerial photography interpretation of a landslipped
area. Quarterly Journal of Engineering Geology 24,
249–265.
Funtowicz, S.O., Ravetz, J.R., 1995. Planning and decision-making
in an uncertain world: the challenge of post-normal sciŽ
.ence. In: Horlick-Jones Ed. , Natural Risk and Civil Protection.
Commission of the European Communities, E&FN Spon,
London, pp. 415–423.
Glade, T., 1998. Establishing the frequency and magnitude of
landslide-triggering rainstorm events in New Zealand. Environmental
Geology 35, 2–3.
Godefroy, P., Humbert, M., 1983. La cartographie des risques
naturales lies aux movements de terrain et aux seismes. Hy-´ ´
Ž .drogeologie — Geologie de l’Ingenieur 2, 69–90, in French .´ ´ ´
Guzzetti, F., 1993. Landslide hazard and risk by GIS-based multivariate
models. In: Reichenbach, P., Guzzetti, F., Carrara, A.
Ž .Eds. , Geographical Information Systems in Assessing Natural
Hazards–Abstracts, CNR, Perugia, September 20–22, 1993.
pp. 83–91.
Guzzetti, F., Cardinali, M., 1989. Carta Inventario dei Fenomeni
Franosi della Regione dell’Umbria ed aree limitrofe. CNR–
ŽGNDCI, Publication no. 204, 2 sheets, scale 1:100,000 in
.Italian .
Guzzetti, F., Cardinali, M., 1990. Landslide inventory map of the
Ž .Umbria region, Central Italy. In: Cancelli, Andrea Ed. , ALPS
90, 6th International Conference and Field Workshop on
Landslides, Milano. pp. 273–284.
Guzzetti, F., Cardinali, M., Reichenbach, P., 1994. The AVI
Project: A bibliographical and archive inventory of landslides
Ž .and floods in Italy. Environmental Management 18 4 , 623–
633.
Guzzetti, F., Cardinali, M., Reichenbach, P., 1996. The influence
of structural setting an lithology on landslide type and pattern.
Ž .Environmental and Engineering Geology 2 4 , 531–555.
Hansen, A., 1984. Landslide hazard analysis. In: Brunsden, D.,
Ž .Prior, D.B. Eds. , Slope Instability. Wiley, New York, pp.
523–602.
Hansen, A., Franks, C.A.M., Kirk, P.A., Brimicombe, A.J., Tung,
F., 1995. Application of GIS to hazard assessment, with
particular reference to landslides in Hong Kong. In: Carrara,
Ž .A., Guzzetti, F. Eds. , Geographical Information Systems in
Assessing Natural Hazards. Kluwer Academic Publisher, Dordrecht,
The Netherlands, pp. 135–175.
Hartlen, J., Viberg, L., 1988. Evaluation of landslide hazard. 5th´
International Symposium on Landslides, Lausanne 2, 1037–
1058.
Helmore, K., 1996. Cities on the brink try new tactics. Choices 5
Ž .1 , 18–21.
Hollingsworth, R., Kovacs, G.S., 1981. Soil slumps and debris
flows: prediction and protection. Bulletin American AssociaŽ
.tion of Engineering Geologists 18 1 , 17–28.
Ž .Horlick-Jones, T., Amendola, A., Casale, R. Eds. , Natural risk
and civil protection. Commission of the European Communities,
E&FN Spon, London, 554 pp.
Howard, A.D., 1994. A detachment-limited model of drainage
Ž .basin evolution. Water Resources Research 30 7 , 2261–2285.
Ž .Hsu, K., 1975. Catastrophic debris stream Sturzstorm generated¨
by rockfalls. Geological Society of America Bulletin 86, 129–
140.
Humbert, M., 1976. La cartographie en France des zones exposees´
a des risques lies aux mouvements du sols. Cartes ZERMOS.´
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216214
International Association Engineering Geology Bulletin 16,
Ž .80–82, in French .
Humbert, M., 1977. La cartographie ZERMOS. Modalites´
d’etablissement des cartes des zones exposees a des risques´ ´ `
lies aux mouvements du sol et du sous-sol. Bulletin Bureau´
Ž .Recherche Geologique et Mineraire 3, 5–8, in French .
Hutchinson, J.N., 1988. General report: morphological and
geotechnical parameters of landslides in relation to geology
and hydrology. 5th International Symposium on Landslides,
Lausanne 1, 3–35.
Hutchinson, J.N., 1995. Keynote paper: landslide hazard assessŽ
.ment. In: Bell Ed. , Landslides. Balkema, Rotterdam, pp.
1805–1841.
Hutchinson, J.N., Chandler, M.P., 1991. A preliminary landslide
hazard zonation of the Undercliff of the Isle of Wight. In:
Slope Stability Engineering. Thomas Telford, London, pp.
197–205.
Ibsen, M.-L., Brunsden, D., 1996. The nature, use and problems
of historical archives for the temporal occurrence of landslides,
with specific reference to the south coast of Britain,
Ventnor, Isle of Wight. Geomorphology 15, 241–258.
IDNHR, Advisory Committee, 1987. Confronting Natural Disasters.
An International Decade for Natural Hazard Reduction.
National Academy Press, WA, 60 pp.
Ikeya, H., 1981. A method for designation for areas in danger of
debris flows. Erosion and Sediment Transport in the Pacific
Rim Steeplands. IAHS Special Pub. 132, 576–588.
Kienholz, H., Hafner, H., Schneider, G., Tamrakar, R., 1983.
Mountain hazards mapping in Nepal’s Middle Mountains.
ŽMaps of land use and geomorphic damages Kathmandu–
. Ž .Kakani area . Mountain Research and Development 3 3 ,
195–220.
Kienholz, H., Schneider, G., Bichsel, M., Grunder, M., Mool, P.,
1984. Mapping of mountain hazards and slope stability. Mountain
Research and Development 4, 247–266.
Kockelman, W.J., 1986. Some techniques for reducing landslide
hazards. Bulletin American Association Engineering GeoloŽ
.gists 23 1 , 29–52.
Ž .Kovar, K., Nachtnebel, H.P. Eds. , 1994. Application of Geographycal
Information Systems in hydrology and water resource
management. HydroGIS 93, Vienna, April 19–22, 1994, IAHS
Publ. 211, 694 pp.
Lambrugo, B., Lattuada, S., 1996. Relazione fra assetto geologico–morfologico
e distrubuzione dei movimenti franosi
Ž .nell’area di Mercatale Arezzo . Caratterizzazione geotecnica
delle Argille Varicolori. Unpublished Thesis. Dipartimento di
Ž .Scienze della Terra, University of Milano. 152 pp. in Italian .
Leroi, E., 1996. Landslide hazard-risk maps at different scales:
Ž .objectives, tools and developments. In: Senneset Ed. , Landslides.
Balkema, Rotterdam, pp. 35–51.
Mark, R.K., 1992. Map of debris flow probability, San Mateo
County, California, scale 1:62,500, US Geological Survey
Miscellaneous Investigation Map, I-1257-M.
Mark, R.K., Ellen, S.D., 1995. Statistical and simulation models
for mapping debris-flows hazard. In: Carrara, A., Guzzetti, F.
Ž .Eds. , Geographical Information Systems in Assessing Natural
Hazards. Kluwer Academic Publisher, Dordrecht, The
Netherlands, pp. 93–106.
Meijerink, A.M.J., 1988. Data acquisition and data capture through
terrain mapping unit. ITC Journal 1, 23–44.
Mejıa-Navarro, M., Wohl, E.E., Oaks, S.D., 1994. Geological´
hazards, vulnerability and risk assessment using GIS: model
for Glenwood Springs, Colorado. Geomorphology 10, 331–
354.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based
model for the topographic control of shallow landsliding.
Ž .Water Resources Research 30 4 , 1153–1171.
Montgomery, D.R., Wright, R.H., Booth, T., 1991. Debris flow
hazard mitigation for colluvium-filled swales. Bulletin AssociŽ
.ation of Engineering Geologists 28 3 , 303–323.
Moore, I.D., Gessler, P.E., Nielsen, G.A., Peterson, G.A., 1993.
Soil attribute prediction using terrain analysis. Soil Science
Society of America Journal 57, 443–452.
Mulder, HFHM, 1991. Assessment of landslide hazard. Profschrift
ter Verkrijging van Graad van Doctor an de Rijkuniversiteit te
Utrecht, University of Utrecht, 150 pp.
Murck, B.W., Skinner, B.J., Porter, S.C., 1997. Dangerous Earth.
An Introduction to Geological Hazards. Wiley, New York, 300
pp.
Neeley, M.K., Rice, R.M., 1990. Estimating risk of debris slides
after timber harvest in northwestern California. Bulletin AmerŽ
.ican Association of Engineering Geologists 27 3 , 281–289.
Nemcok, A., Rybar, J., 1968. Synoptic Map of Czechoslovak´
Landslide Areas. Geol. Ustar Ceskoslovachia Akademia, Praha.
Neuland, H., 1976. A prediction model of landslips. Catena 3,
215–230.
Nieto, A.S., 1989. Mechanical models and geological observations:
closing the prediction gap. In: Siccardi, F., Brass, R.
Ž .Eds. , International Workshop on Natural Disasters in European–Mediterranean
Countries, Perugia, 27 June–1 July 1989.
CNR-US NFS, pp. 145–164.
Nilsen, T.H., Brabb, E.E., 1977. Slope stability studies in the San
Francisco Bay region, California. Geological Society of America,
Reviews in Engineering Geology 3, 235–243.
Nilsen, T.H., Wright, R.H., Vlasic, C., Spangle, W.E., 1979.
Relative slope stability and land-use planning in the San
Francisco Bay region, California. US Geological Survey Professional
Paper 944, 104 pp.
Okimura, T., Kawatani, T., 1987. Mapping of the potential surŽ
.face-failure sites on granite slopes. In: Gardiner, E. Ed. ,
International Geomorphology 1986, Part I. Wiley, Chichester,
pp. 121–138.
Ollier, C.D., 1977. Terrain classification: methods, applications
Ž .and principles. In: Hails, J.R. Ed. , Applied Geomorphology.
Elsevier, Amsterdam, pp. 277–316.
O’Loughlin, E.M., 1986. Prediction of surface saturation zones in
natural catchments by topographic analysis. Water Resources
Ž .Research 22 5 , 794–804.
Olshansky, R., 1990. Landslide hazard in the United States. Case
studies in planning and policy development. Garland Publishing,
New York, 178 pp.
Pike, R.J., 1988. The geometric signature: quantifying landslideterrain
types from digital elevation models. Mathematical GeŽ
.ology 20 5 , 491–511.
Popper, K.R., 1959. The Logic of Scientific Discovery. Basic
Books, New York, 480 pp.
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216 215
Radbruch-Hall, D.H., Varnes, D.J., 1976. Landslides: causes and
effect. International Association Engineering Geologist Bulletin
14, 205–216.
Radbruch-Hall, D.H., Colton, R.B., Davies, W.E., Lucchitta, I.,
Skipp, B.A., Varnes, D.J., 1982. Landslide overview map of
the conterminous United States. US Geological Survey Professional
Paper 1183, 25 pp.
Ž .Reichenbach, P., Guzzetti, F., Carrara, A. Eds. , Geographical
Information Systems in Assessing Natural Hazards. Abstracts
volume, CNR-IRPI, Perugia, September, 20–22, 1993. 140 pp.
Rib, H.T., Liang, T., 1978. Recognition and identification. In:
Ž .Schuster, R.L., Krizek, R.J. Eds. , Landslides analysis and
control. Washington Transportation Research Board, Special
Report 176. National Academy of Sciences, WA, pp. 34–80.
Rosenfeld, C.L., 1994. The geomorphological dimensions of natural
disasters. Geomorphology 10, 27–36.
Roth, R.A., 1983. Factors affecting landslide susceptibility in San
Mateo County, California. Association of Engineering GeoloŽ
.gists Bulletin 20 4 , 353–372.
Sassa, K., 1988. Special lecture: geotechnical model for the
motion of landslides. 5th International Symposium on Landslides,
Lausanne 1, 37–55.
Scheidegger, A.E., 1994. Hazards: singularities in geomorphic
systems. Geomorphology 10, 19–25.
Schuster, R.L., 1995a. Socioeconomic significance of landslides.
Ž .In: Turner, A.K., Schuster, R.L. Eds. , Landslides, Investigation
and Mitigation. Transportation Research Board Special
Report 247. National Academy Press, WA, pp. 12–35.
Schuster, R.L., 1995b. Keynote paper: recent advances in slope
Ž .stabilization. In: Bell Ed. , Landslides. Balkema, Rotterdam,
pp. 1715–1745.
Schuster, R.L., Fleming, R.W., 1986. Economic losses and fatalities
due to landslides. Bulletin American Association EngiŽ
.neering Geologists 23 1 , 11–28.
Schuster, R.L., Krizek, R.J., 1978. Landslides Analysis and Control.
Washington Transportation Research Board, Special Report
176. National Academy of Sciences, WA, 234 pp.
Seeley, M.W., West, D.O., 1990. Approach to geologic hazard
zoning for regional planning, Inyo National Forest, California
and Nevada. Bulletin American Association of Engineering
Ž .Geologists 27 1 , 23–35.
Soeters, R., Rengers, N., van Westen, C.J., 1991. Remote sensing
and geographical information systems as applied to mountain
hazard analysis and environmental monitoring. 8th Thematic
Ž .Conference Geology Remote Sensing ERIM , April, 29–May,
2 1991, Denver 2, 1389–1402.
Speight, J.G., 1977. Landform pattern description from aerial
photographs. Photogrammetry 32, 161–182.
Starosolszky, O., Melder, O.M., 1989. Hydrology of disasters.
World Meteorological Organisation Technical conference,
Geneva, November 1988. James and James, London, 319 pp.
Stout, M.L., 1977. Radiocarbon dating of landslides in Southern
California. California Geology, 99–105, January 1977.
Swanston, D.N., Schuster, R.L., 1989. Long-term landslide hazard
mitigation programs: structure and experience from other
countries. Bulletin American Association of Engineering GeolŽ
.ogists 26 1 , 109–113.
Takahashi, T., Aschida, K., Sawai, K., 1981. Delineation of debris
flow hazard areas. Erosion and Sediment Transport in the
Pacific Rim Steeplands. IAHS Special Pub. 132, 589–603.
Terlien, M.T.J., van Westen, C.J., van Asch, Th.W.J., 1995.
Deterministic modelling in GIS-based landslide hazard assessŽ
.ment. In: Carrara, A., Guzzetti, F. Eds. , Geographical Information
Systems in Assessing Natural Hazards. Kluwer Academic
Publisher, Dordrecht, The Netherlands, pp. 57–77.
Toppi, A., 1993. Studio geologico–tecnico sullo stato di dissesto
del bacino idrografico del F.sso Bianco–T. Caldaro in Provincia
di Terni. Unpublished Thesis, Dipartimento di Scienze
Ž .della Terra, University of Perugia, Perugia, 103 pp. in Italian .
Trustrum, N.A., De Rose, R.C., 1988. Soil depth–age relationship
of landslides on deforested hillslopes, Taranaki, New Zealand.
Geomorphology 1, 143–160.
Ž .Turner, A.K., Schuster, R.L. Eds. , 1995. Landslides, investigation
and mitigation. Transportation Research Board Special
Report 247. National Academy Press, WA, 673 pp.
UNDRO, 1991. Mitigating Natural Disasters. Phenomena, Effects
and Options. United Nations, New York, 164 pp.
U.S. Geological Survey, 1982. Goals and tasks of the landslide
part of a ground failure hazard reduction program. US Geological
Survey Circular 880, 44 pp.
van Driel, N., 1991. Geographic information systems for earth
science applications. 4th International Conference on Seismic
Zonation, Stanford 1, 469–485.
van Westen, C.J., 1993. Application of Geographical Information
System to landslide hazard zonation. ITC Publication no. 15,
ITC, Enschede, The Netherlands, 245 pp.
van Westen, C.J., 1994. GIS in landslide hazard zonation: a
review with examples from the Colombian Andes. In: Price,
Ž .M.F., Heywood, D.I. Eds. , Taylor and Francis, London, pp.
135–165.
Varnes, D.J., 1978. Slope movements, type and processes. In:
Ž .Schuster, R.L., Krizek, R.J. Eds. , Landslide Analysis and
Control. Transportation Research Board, National Academy
Sciences, Washington, DC, pp. 11–33, Special report 176.
IAEG Commission on Landslides and other Mass-Movements,
Varnes, D.J., 1984. Landslide hazard zonation: a review of
principles and practice. UNESCO Press, Paris, 63 pp.
Verstappen, H.T., 1983. Applied Geomorphology: Geomorphological
Survey for Environmental Development. Elsevier, Ams-
terdam.
Wadge, G., 1988. The potential of GIS modelling of gravity flows
and slope instabilities. International Journal of Geographical
Information Systems 2, 143–152.
Walsh, J.S., Lightfoot, D.R., Butler, D.R., 1987. Recognition and
assessment of error in Geographic Information Systems. PhoŽ
.togrammetric Engineering and Remote Sensing 53 10 , 1423–
1430.
Wieczorek, G.F., 1984. Preparing a detailed landslide-inventory
map for hazard evaluation and reduction. Bulletin Association
Ž .of Engineering Geologists 21 3 , 337–342.
Wright, R.H., Nilsen, T.H., 1974. Isopleth map of landslide
deposits, Southern San Francisco Bay Region, California. US
Geological Survey Miscellaneous Field Studies Map, MF-550
Ž .scale 1:250,000 .
( )F. Guzzetti et al.rGeomorphology 31 1999 181–216216
Wright, R.H., Campbell, R.H., Nilsen, T.H., 1974. Preparation
and use of isopleth maps of landslide deposits. Geological
Society of America, Geology 2, 483–485.
Yin, K.L., Yan, T.Z., 1988. Statistical prediction model for slope
instability of metamorphosed rocks. 5th International Symposium
on Landslides, Lausanne 2, 1269–1272.
Yong, R.N., Alonso, E., Tabba, M.M., Fransham, P.B., 1977.
Application of risk analysis to the prediction of slope instabilŽ
.ity. Canadian Geotechnical Journal 14 3 , 540–543.
Zimmerman, M., Bichsel, M., Kienholz, H., 1986. Mountain
hazards mapping in the Khumbu Himal, Nepal, with prototype
map, scale 1:50,000. Mountain Research and Development 6
Ž .1 , 29–40.