See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/309281060

Assessment and validation of GIS-based landslide susceptibility maps: a case

study from Feltrino stream basin (Central Italy)

Article  in  Bulletin of Engineering Geology and the Environment · October 2016

DOI: 10.1007/s10064-016-0954-7

CITATIONS READS

10 571

3 authors:

Marco Sciarra Laura Coco

Italferr Ferrovie dello Stato Università degli Studi G. d'Annunzio Chieti e Pescara
28 PUBLICATIONS   108 CITATIONS    13 PUBLICATIONS   163 CITATIONS   

SEE PROFILE SEE PROFILE

Tullio Urbano
Università degli Studi G. d'Annunzio Chieti e Pescara
14 PUBLICATIONS   48 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

EGU2018 - Workshops and Events for young Geomorphologists - SC1.29/GM12.1 Crowd-solving problems in earth science research (co-organized) View project

PhD Thesis View project

All content following this page was uploaded by Tullio Urbano on 29 December 2017.

The user has requested enhancement of the downloaded file.

Bull Eng Geol Environ
DOI 10.1007/s10064-016-0954-7

ORIGINAL PAPER

Assessment and validation of GIS-based landslide susceptibility

maps: a case study from Feltrino stream basin (Central Italy)
Marco Sciarra1 • Laura Coco1 • Tullio Urbano1

Received: 6 July 2016 / Accepted: 4 October 2016

 Springer-Verlag Berlin Heidelberg 2016

Abstract Landslide susceptibility studies focus on pro- susceptibility model also in the absence of detailed land-
ducing susceptibility maps starting from landslide inven- slide mapping, by considering inventories that are already
tories and considering the main conditioning factors. The available.
validity of susceptibility maps must be verified in terms of
model accuracy and prediction skills. This paper deals with Keywords Landslide susceptibility map  Landslide
a GIS-based landslide susceptibility analysis and relative inventory  GIS  Bivariate statistics  Validation  Abruzzo
validation in a hilly-coastal test-area in Adriatic Central (Italy)
Italy. The susceptibility analysis was performed via
bivariate statistics using the Landslide-Index method and a
detailed (field-based) landslide inventory. Selection and Introduction
mapping of conditioning factors and landslide inventories
was derived from detail geomorphological analyses of the Landslide susceptibility can be defined as the possibility
study area. The susceptibility map was validated using that landslides occur in an area in a certain period of time,
recent (shallow) landslides in terms of both model accuracy based on local conditions (Brabb 1984; Aleotti and
and prediction skills, via Success rate and Prediction rate Chowdhury 1999). Therefore, to understand if a specific
curves, respectively. In addition, a pre-existing official area is prone to landslides, slope movement and condi-
landslide inventory was applied to the model to test whe- tioning (or predisposing) factors need to be known (van
ther it can be used when a detailed (field-based) inventory Westen et al. 2006; Fell et al. 2008).
is not available, thereby extending its usability in similar Many methods have been developed over the past few
physiographic regions. The outcome of this study reveals decades for studying this fundamental topic of geo-envi-
that slope and lithology are the main conditioning factor of ronmental research. The basic idea from different approa-
landslides, but also highlights the key role of surficial ches is to produce a susceptibility map starting from an
deposits in susceptibility assessment, for both their type inventory map of landslides, considering the main geo-
and thickness. The validation results show the effectiveness environmental factors that produce landslides in a study
of the susceptibility model in both model accuracy and area. All susceptibility analyses are based on the principle
prediction skills given the good percentage of correctly that new landslides will occur under the same condition
classified landslides. Moreover, comparison of the sus- that led to past landslides: ‘‘the past is the key to the
ceptibility map with the official Regional landslides future’’ (Varnes 1984; Soeters and van Westen 1996;
inventory proves the possibility of using the developed Aleotti and Chowdhury 1999; Fell et al. 2008).
According to Aleotti and Chowdhury (1999), the
methods for producing landslide susceptibility maps can be
& Laura Coco grouped into two main categories: qualitative and quanti-
lauracoco@libero.it
tative. Qualitative methods are based on evaluations that
1
Department of Engineering and Geology, University ‘G. identify conditioning factors, and eventually the weight
d’Annunzio’ of Chieti, Via dei Vestini 13, 66100 Chieti, Italy attributed to those factors, depending only on expert

123
 M. Sciarra et al.

knowledge and experience (Soeters and van Westen 1996; susceptibility map in terms of model accuracy and pre-
Guzzetti et al. 1999; Fell et al. 2008). Such susceptibility diction skills (Irigaray et al. 1999; Ardizzone et al. 2002;
maps can be produced directly from field geomorphologi- Chung and Fabbri 2003; Remondo et al. 2003; Guzzetti
cal analysis but have the limit of being subjective. To et al. 2006; Petschko et al. 2014). This is an essential phase
overcome this limit, quantitative methods based on specific in landslide susceptibility modeling, as well as in any Earth
and known rules have been developed. These methods Science modeling, to evaluate the ‘‘degree of belief’’ of the
allow the determination of numerical relations that corre- model with respect to available observations (Oreskes et al.
late conditioning factors to landslides, resulting in sus- 1994).
ceptibility maps that can be reproduced objectively and The former issue, i.e., model accuracy, is linked to the
independently of expert opinion (Chung et al. 1995; Aya- capacity of the model to explain the analyzed phenomena
lew and Yamagishi 2005; Felicı́simo et al. 2013). Quanti- and this can depend not only on input data, but also model
tative methods can be of two types: deterministic and form. All input data must not only fully cover the analyzed
statistic. Deterministic methods require knowledge of the area with sufficient accuracy and resolution adequate to the
main physical properties of terrains involved in landslides final product, but also be as complete as possible in order to
and the application of specific mathematical models with include all the possible (conditioning or triggering) factors
the calculation of safety factors. They allow the develop- in the final susceptibility model (Carrara et al. 1999; van
ment of susceptibility models in engineering terms for Westen et al. 2008). Moreover, one of the main data types
slope-specific stability studies and can be applied only to that has to be considered is the landslide inventory (van
small areas. In wider areas and for regional zonation, sta- Westen et al. 2008). Understanding the determinant factors
tistical methods can be used. for landslides in a given region requires good field-based
Statistical methods compare the spatial distribution of knowledge of its geological and geomorphological setting
landslides with some selected conditioning parameters, with special regard to landslides themselves (Guzzetti et al.
producing susceptibility maps in which the territory is 1999). The validity of the (statistical) model deals with the
divided into zones that have a different propensity to be use of correct rules that link landslide occurrence and
subjected to landslides, according to the rank of the con- explanatory variables (Guzzetti et al. 1999; Fell et al. 2008;
sidered factors (Anbalagan 1992; Carrara et al. 1995; Petschko et al. 2014). This requires familiarity with exist-
Aleotti and Chowdhury 1999; Guzzetti et al. 1999; Fell ing modeling methodologies, particularly regarding the
et al. 2008; Yalcin et al. 2011; De Giudi and Scudero 2013; scale of the analysis (van Westen et al. 2006; Cascini
Conoscenti et al. 2015; Dou et al. 2015). In areas without 2008).
landslides, high values of conditioning factors indicate the The last issue, i.e., prediction skills, refers to the power
possibility that landslides will occur in the future. The main of the model to predict where landslides will occur in the
difficulty of this type of landslide zonation is the selection future and entails testing the model with independent
of appropriate conditioning factors (Carrara 1989; Aleotti landslide information (Guzzetti et al. 2006; Von Ruette
and Chowdhury 1999; van Westen et al. 2003; Dou et al. et al. 2011). The most widely used methods for quantita-
2015). According to Fell et al. (2008), conditioning factors tively evaluating this power are Success or Prediction rate
concur with triggering factors in causing landslides. Con- curves (Chung and Fabbri 2003, 2008). Two approaches
ditioning factors are permanent terrain attributes with slow can be followed: the first is to use the same dataset as used
evolution and determine the propensity of slopes to land- in the statistical analysis, split into a calibration subset (for
slides. They usually include lithology, geology, soil type estimating the model coefficients) and one or more vali-
and thickness, slope morphometry, drainage, land cover dation subsets (for assessing its predictive performance)
and land use, among others (e.g., Atkinson and Massari (von Ruette et al. 2011); the second is to compare the final
1998; Lee et al. 2001; Dai and Lee 2002; van Westen et al. susceptibility map to independent data from other events or
2003; Ayalew et al. 2004; Brenning 2005; Gómez and areas, e.g., other landslides that occur after model set-up
Kavzoglu 2005; Irigary et al. 2007; Yilmaz 2010; Pradhan (Irigaray et al. 1999, 2007).
and Lee 2010; Fressard et al. 2014; Pradhan and Kim 2014; This study started from these considerations in order to
Dou et al. 2015). Triggering factors, instead, are temporary perform a landslide susceptibility analysis in a test-area of
conditions that can directly cause landslides, for example Adriatic Central Italy (Feltrino stream basin in Abruzzo
earthquakes, rapid snow melt or rainfall. Landslide sus- Region). Three reasons led us to choosing the Feltrino
ceptibility maps are produced considering conditioning basin as a test-area: a large amount of data available on
factors only, whereas triggering factors determine landslide both landslides and terrain characteristics (morphometric,
hazard maps. geological-geomorphological, land cover and land use);
One of the essential issues in landslide susceptibility (or geo-environmental features representative of the entire
hazard) zoning is to verify the validity of the final Adriatic hilly-coastal area of Central Italy; a deep influence

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

of human activities that makes this area particularly haz- The Apennine piedmont fringes the easternmost car-
ardous. The main aim of this study was to produce a bonate ridges (e.g., Maiella massif) with an abrupt mor-
landslide susceptibility map and validate it in terms of both phologic boundary, and shows cuestas, mesas and a
model accuracy and prediction skills. The landslide sus- plateaux landscape that slopes from *1000 m a.s.l. along
ceptibility assessment was carried out with a GIS-based the mountain front to \100 m a.s.l. close to the coast. To
statistical approach via bivariate analysis using the Land- the East, the area is bounded by a cliffed coast cut by
slide-Index method (Van Westen 1993, 1994; Van Westen alluvial valleys (e.g., Sangro River valley to the south and
et al. 1997). The validation procedures were conducted via Pescara River valley to the north), and by small and sharp
Success rate and Prediction rate curves (Chung and Fabbri valleys of minor streams (such as that of the Feltrino).
2003) using different inventories for testing model accu- Main rivers generally flow in a SW–NE direction within
racy and prediction skills. Moreover, the model was tested wide floodplains, while minor rivers are entrenched within
with a rainfall-induced landslide inventory in order to narrow valleys with different orientations. This landscape
verify whether the model can be used also when a detailed is carved on marine bedrock composed by the Middle
(field-based) inventory is not available, thereby extending Pliocene-Early Pleistocene clayey-sandy sediments, which
its utility in similar physiographic regions. westward unconformably overlie the Miocene-Early Plio-
cene arenaceous-pelitic foredeep sequences (Ori et al.
1986; Crescenti et al. 2004). A Middle Pleistocene sandy-
Materials and methods conglomeratic sequence shows a regressive stage with a
gradual transition from a marine to a subaerial environment
Study area (Cantalamessa and Di Celma 2004). The present structural
setting is marked by a regional homocline gently dipping
The Feltrino stream basin is located in the hilly-coastal NE and affected locally by fractures and low-displacement
area of the central-eastern Apennine (Abruzzo Region), faults, as a response to Plio-Pleistocene uplift and local
between 42180 4000 N and 42110 3100 N latitude, 14200 2500 E tectonics (Ori et al. 1986; Aucelli et al. 1996; D’Alessandro
and 14300 1400 E longitude (Fig. 1). This SW–NE elongated et al. 2008). Furthermore, the interaction between structural
basin slopes from the eastern piedmont of the Maiella framework, tectonic processes, and Quaternary climate and
Massif (400 m a.s.l.) to the Adriatic coast, over an area of glacio-eustatic variations induced fluvial and slope pro-
about 50 km2. It is a 4th order basin in which the drainage cesses, as recorded in a complex sequence of continental
pattern varies from subparallel to trellis (from N–S to NE– deposits (post-orogenic phase) overlying the bedrock.
SW oriented). The basin’s hypsometric integrals are In this frame, most of the Feltrino basin is carved on
moderately high in value (0.47) associated with concave- clay, sandstone and conglomerate units belonging to Plio-
convex hypsometric diagrams, evidencing a deeply incised Pleistocene marine sequences. A conglomerate unit lies on
(up to 200 m relief) and partly enlarged (3–5 km large) the top of the mesa reliefs, whereas sandstone and clay
basin (Piacentini et al. 2015). units are widespread along the valley flanks and bottom.
The area is characterized by a Mediterranean climate, Conglomerate and sands and sandy-clay units, belonging to
with wet winters, dry summers, and periods of high the marine to transition sequences, are scattered to a very
snowmelt and rainfall (spring). The mean annual precipi- small extent near the town of Lanciano (top of the mesa
tation varies between 1000 mm in the Maiella Massif and reliefs). The continental sequences (mainly colluvial and
700 mm along the coast, and the mean annual temperature landslide deposits) outcropping in the basin are strictly
varies from 10–11 C to 16 C (Piacentini et al. 2016). related to the geomorphological evolution of structural
From a geological point of view, the Central Apennine landforms (i.e., structural scarps, mesa and plateaux top
is a NE-verging thrust belt emplaced during the Neogene surfaces). This evolution encompasses four different stages
period that resulted from the incomplete collision of Eur- resulting from the superimposition of major slope gravity
ope and Africa (Carminati and Doglioni 2004; Parotto and and fluvial processes (landslides, soil creep, incision of
Praturlon 2004; Patacca and Scandone 2007). Since the V-shaped valleys) and soil erosion processes (surface
Late Pliocene, the thrust belt has been affected by intense runoff, gullies, sheet erosion), which have progressively
uplift and extensional tectonics inducing the displacement smoothed hillslopes (Piacentini et al. 2015). As a conse-
of the main ridges of the chain, the development of tectonic quence, the bedrock is now largely covered by near-surface
valleys and intermountain basins (e.g., Sulmona basin), and clay-silt-sand-gravel continental deposits ranging in age
the emergence of the western Adriatic area with the sub- from the Upper Pleistocene to the Holocene.
sequent evolution of the Apennine piedmont (Cavinato and The slope gravity landforms of the area are related to a
De Celles 1999; D’Agostino et al. 2001; D’Alessandro large number of landslides, including: rotational and
et al. 2008). translational landslides, shallow landslides, complex and

123
 M. Sciarra et al.

Fig. 1 Study area location map

deep-seated landslides and, to a lesser extent, rock falls, accelerating erosion and shallow landslide distribution
with associated scarps, reverse slopes and gravity trenches (Piacentini et al. 2015; Sciarra 2016).
(active, dormant or inactive). By the term shallow land-
slide, we refer to a surficial mass movement (rapid Methods
earthflow, creeping area, and failure of road cut slopes),
which affects mainly the near-surface deposits above non- The present landslide susceptibility analysis was based on a
weathered bedrock and whose sliding planes are fre- multidisciplinary approach involving GIS (Geographic
quently located along the contact between deposits and Information System) mapping and processing, (bivariate)
bedrock. The type and distribution of slope gravity land- statistical and validation analyses, remote sensing, and geo-
forms are strictly controlled by the lithological and mor- logical and geomorphological field surveys. As many condi-
phostructural setting and, as far as shallow landslides are tioning factors characterize landslide-prone areas at different
concerned, by the poorly vegetated landscape temporal and spatial scales, the experience gained during the
(D’Alessandro et al. 2004; Piacentini et al. 2015). The field-surveys in the Feltrino basin and the well documented
footslopes and the narrow irregular alluvial flats are geological and geomorphological setting (Aucelli et al. 1996;
shaped by the toes of such landslides, which often divert Chiocchini et al. 2006; Piacentini et al. 2015; Sciarra et al.
watercourses. 2016) allowed us to identify seven conditioning factors: slope,
Over recent years, the Feltrino basin has undergone topographic curvature, drainage density, bedrock lithology,
notable anthropogenic activities which caused modifica- near-surface deposits, deposit thickness, and land cover and
tions in the landscape morphology and the evolution of use. As the scope of this work is landslide susceptibility (and
vegetation and land use. These activities are related to not hazard), information on triggering factors, such as rainfall
farming (deforestation, soil ploughing, extensive slope and seismicity, were not taken into account. Bivariate analy-
levelling for new vineyards and olive groves), and building ses for predicting the presence (or the absence) of a phe-
of infrastructure (roads, rails, residential areas), sometimes nomenon (i.e., landslide) were based on the values of predictor

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

variables: we applied the Landslide-Index (LI) method (van subsequently, overlaid the weighted factors (raster-
Westen 1993, 1994; van Westen et al. 1997) for predicting the ized) to produce a preliminary landslide susceptibility
presence (or absence) of landslide inventory maps based on map;
the values of predictor variables (i.e., the selected condition- 3. Validation: first, we evaluated the quantitative mea-
ing factors). In order to assess the accuracy and the prediction surement of model fit (i.e., preliminary susceptibility
skills of the model, we constructed Success rate and Prediction map) with the Success rate curve considering the same
rate curves (Chung and Fabbri 2003). The Success rate curve landslides used for the statistical analysis, and, subse-
was constructed by means of the same landslide data used for quently, we tested the model’s prediction skills with
the statistical analysis of conditioning factors. The Success the Prediction rate curve, comparing the preliminary
rate method, performed by considering independent landslide susceptibility map with an independent inventory that
information not involved in the statistical analysis, produced recorded successive landslide events induced by heavy
the Prediction rate curve. Since the use of the same landslides rainfall events.
for both model construction and verification leads to an
overestimate of the predictive capacity of the susceptibility As a test for extending the utility of the final suscepti-
map (Brenning 2005; Guzzetti et al. 2006), we considered the bility model, we performed an additional validation pro-
Success rate curves as a measure of model accuracy (or fit) cedure using the official regional landslide inventory via
whereas Prediction rate curves were a measure of its predic- Success rate curve. The following sections describe each
tion skills (Chung and Fabbri 2003; van Westen et al. 2003). step of the workflow in detail.
The workflow of this study consisted of three main steps
(Fig. 2): Implementation of the geodatabase
1. Implementation of the geodatabase: we stored the
landslide distribution (inventories) and the selected Since bivariate analyses consider the statistical relation-
landslide conditioning factors as vector and raster ships between landslides and conditioning factors, the first
layers (Table 1); step of this study was the analysis and extraction of
2. Statistical analysis: we used a bivariate statistical detailed landslide inventory maps and landslide condi-
method in order to define the contribution of each tioning factors (slope, topographic curvature, drainage
conditioning factor to landslide susceptibility, and, density, bedrock lithology, near-surface deposits, deposit

Fig. 2 Workflow of the susceptibility analysis

123
 M. Sciarra et al.

Table 1 List of input data (data description, scale, reference and format) used to extract the landslide inventories and the morphometric,
geological/geomorphological, vegetation/anthropogenic features of the study area
Geodatabase layers Data source description Scale References Format

Landslide inventories
Pre-2012 Geomorphological map 1:5 k Piacentini et al. (2015) .shp
2012–2015 Field survey data 1:5 k Sciarra (2016) .shp
PAI Official inventory 1:25 k AdB (2005) .shp
Conditioning factors
Morphometric 5 m cell-size DTM extracted from vector maps 1:5 k Abruzzo Region (2007) Raster
Geological and geomorphological Geomorphological map 1:5 k Piacentini et al. (2015) .shp
Field survey data 1:5 k Sciarra (2016) .shp
Vegetation and anthropogenic Land use map 1:25 k Abruzzo Region (2000) .shp
Color orthophotos 1:5 k Abruzzo Region (2009) .shp
Field survey data 1:5 k Sciarra (2016) .shp

thickness and land use) directly or indirectly linked to slope Varnes (1996) and WP/WLI (1993) that defined the concept
instability. This step was performed in a GIS environment of activity with reference to the originating causes (Dramis
(ArcGIS 10.1) by means of DTM processing, air-photo et al. 2011). We stored the pre-2012 landslides in a vector
analysis and detailed geomorphological field survey. We layer as closed polygon features linked to an attribute table.
created a geodatabase to collect the mapped factors and the It showed the distribution of deposition and erosion (ex-
landslide inventory maps as reported in Table 1. cluding the main scarps when present, see Fig. 4) areas of
gravity-induced mass movements that may vary in type, age
Landslide inventory maps (from late Pleistocene to the present) and activity (active,
quiescent, inactive). Landslides covered a total area of
The landslide inventories of the Feltrino basin consisted of 12.47 km2, corresponding to 24.9 % of the whole basin, and
three maps: the pre-2012 inventory derived from pre-2012 were mainly rotational slides, shallow landslides and com-
surveys, the 2012–2015 inventory from the surveys carried plex movements (85.8, 13.7 and 0.5 % respectively).
out between 2012 and 2015, and the PAI inventory from the
official regional landslides inventory. We used the first in both 2012–2015 Landslide inventory map We implemented the
statistical and validation phases, the second for validation 2012–2015 landslide inventory map with recent landslides
purposes exclusively, and the third for an independent vali- induced by the heavy rains that occurred during the
dation for cases in which a field-based inventory is lacking. 2012–2015 interval, mapped through a detailed (1:5000 scale)
geomorphological field survey reported in Sciarra (2016),
Pre-2012 landslide inventory map The pre-2012 land- according to Piacentini et al. (2015). This inventory map will
slide inventory map included mass movements (occurring be referred to hereafter as the ‘‘pre-2012 inventory’’.
before 2012) that have been extracted from the Geomor- This inventory contains 71 landslides covering a total
phological map of Feltrino Stream basin of Piacentini et al. surface area of 2.13 km2. Heavy rains were distributed in four
(2015). This map was focused on hillslope evolution and main events that occurred on 14–15 September 2012
related near-surface deposits (mainly colluvial, slope, (*250 mm), 1–3 December 2013 (*100 mm), 25–27
landslide and alluvial deposits), and was edited by means February and 4–6 March 2015 (both *200 mm), each
of air-photo analysis, detailed (1:5000) geological and accounting for about one-third of the total mean annual
geomorphological surveys (2011–2012 years) and geo- amount of the area. The triggered slope failures include sev-
physical data (surficial boreholes). These data are particu- eral shallow landslides (i.e., rapid earthflows and creeping
larly representative of near-surface deposits and related areas), one rotational slide, and minor overland flow pro-
geomorphological slope processes (i.e., landslides), which cesses. These movements affected mainly near-surface
describe the occurrence, geometry, typology, and activity deposits and erodible clayey-sand bedrock. In addition to the
of slope instabilities very well. This inventory map will be large number of landslide bodies (i.e., reactivation) well
referred to hereafter as the ‘‘pre-2012 inventory’’. delineated in the pre-2012 inventory (Fig. 4), shallow land-
This inventory (Fig. 3) consists of collected information slides particularly affected agricultural lands with poor veg-
related to 223 landslide bodies as geometrical and etation cover (arable lands, vineyards and olive groves).
alphanumerical features classified according to Cruden and According to Varnes (1978), these flows were characterized

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

Fig. 3 Pre-2012 landslide inventory map

Fig. 4 Examples of different types of landslides in the Feltrino et al. 2015). c, d Shallow landslides (c) and complex movement
Stream basin. a Rotational landslide with expanded inset (b) showing (d) induced by heavy rainfall events that occurred in 2012 and 2015,
detailed view of the main landslide scarp (modified from Piacentini respectively

by a spatially continuous movement in which shear surfaces vector layer as point features (Fig. 5). We located these points
were short-lived, closely spaced, and usually not preserved. where the main detachment occurred (small amphitheaters
Since most of the triggered movements were very small usually in the upper or middle part of the slope), without
(\100 m2 in area), we stored the 2012–2015 landslides in a considering transportation and deposition zones.

123
 M. Sciarra et al.

Fig. 5 2012–2015 landslide inventory map

PAI landslide inventory map We extracted the PAI (2007), and the geological and geomorphological factors
landslide inventory from the ‘‘Piano di Assetto Idrogeo- (i.e., bedrock lithology, near-surface deposits, deposit
logico’’ program (PAI, Hydrogeological Setting Plan) of thickness) from the geomorphological map of the Feltrino
the Abruzzo-Sangro Basin Authority (2005). It will be Stream basin from Piacentini et al. (2015). We extracted
hereafter referred to as the ‘‘PAI inventory’’. the vegetation and anthropogenic factors (i.e., land cover
This official landslide inventory, at 1:25,000 scale, was and use) from the Land Use map of Abruzzo Region (2000)
published in 2005 and contains part of the Italian Landslide at a 1:25,000 scale, calibrated and detailed by field-surveys
Inventory (IFFI Project), a national landslide database (Sciarra 2016) and remote sensing analysis of color
(ISPRA 2007). Inside the Feltrino basin, a total of 124 orthophotos (Abruzzo Region 2009) following the
landslides were mapped according to Cruden and Varnes nomenclature of Corine Land Cover (2012) for obtaining a
(1996) criterion (Fig. 6), of which 25 (20.2 %) were classi- factor layer at 1:5000 scale.
fied as ‘active’ (including active and reactivated phenom- On the basis of the well-known geological and geo-
ena), 92 (74.2 %) as dormant, and 7 (5.6 %) as stabilized morphological features of the study area, and of the data
(including artificially stabilized, abandoned and relict phe- source accuracy and scale, each factor was subdivided into
nomena). Landslides covered a total area of 13.79 km2, classes of values which encompass the whole variability of
corresponding to 27 % of the whole basin, and the main that factor (Fig. 7).
landslide typologies were rotational slides, shallow land-
slides and complex movements (84.45 %, 13.9 % and Slope The steepness of slopes is the major factor used in
1.65 %, respectively). This source of information was based landslide susceptibility studies (e.g., Lee and Min 2001;
mainly on air-photo interpretation and, to a lesser extent, on Cevik and Topal 2003; Ercanoglu et al. 2004; Yalcin and
sample field surveys and a collection of local databases. Bulut 2007; Yalcin et al. 2011; Nefeslioglu et al. 2008) since
landsliding is directly related to slope angle: the steeper the
Conditioning factors slope, the greater the tendency for instability (Hoek and Bray
1981). This factor plays a key role in the dynamics of the
The considered landslide conditioning factors were strictly near-surface processes, strongly influencing not only slope
related to the morphometric, geological, geomorphological, processes but also, indirectly, hydrological processes, drai-
vegetation and anthropogenic features of the study area, nage density, soil erosion, weathering, vegetation cover, and
and were also selected considering the fulfilment of the anthropogenic activity (Evans 1972; Cerdà and Garcı́a-
fundamental requisite to cover the whole Feltrino basin. Fayos 1997; Fox et al. 1997; Bennie et al. 2006; Buccolini
We derived the morphometric factors (i.e., slope, topo- et al. 2007, 2012; Cappadonia et al. 2016).
graphic curvature, drainage density) by processing a 5 m We derived the slope factor using the ‘‘Slope’’ function
cell-size Digital Terrain Model (DTM) extracted from the of the Spatial Analyst tools, and divided the slope values
1:5000 scale Regional Technical Maps of Abruzzo Region (in degrees) into seven classes (Fig. 7) that roughly

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

Fig. 6 ‘‘Piano di Assetto Idrogeologico’’ program (Hydrogeological Setting Plan) of the Abruzzo-Sangro Basin Authority (2005) landslide
inventory map (hereafter, PAI inventory)

Fig. 7 Spatial distribution of the seven conditioning factors: a slope; b topographic curvature; c drainage density; d bedrock; e near-surface
deposits; f deposits thickness; g land cover and use

123
 M. Sciarra et al.

Table 2 Classes and ranges (in

Class Thickness range (m) Morphological unit
degree) of slope factor with
reference to the main 1 0–0.5 Scarps, sub-vertical slopes
morphological units of the
Feltrino Basin 2 0.5–3 Summit of tabular reliefs
3 3.0–5.0 Alluvial plains, upper-middle concave valleys
4 5.0–10 Lower concave valleys, lower hillslopes
5 [10.1 Large landslide bodies, backfill deposits

corresponded to the main morphological units of the Bedrock lithology Lithology is one of the most important
landscape, as reported in Table 2. factors in landslide studies since different lithological units
may be affected by different landslide types and conse-
Topographic curvature The term topographic curvature quently susceptibility degrees (Guzzetti et al. 1996; Dai
refers to the rate of change of slope gradient (profile cur- et al. 2001; Lee and Min 2001; Yesilnacar and Topal 2005;
vature) and/or aspect (planform curvature), usually in a Yalcin and Bulut 2007; Nefeslioglu et al. 2008).
specific direction (Moore et al. 1991; Wilson and Gallant According to Piacentini et al. (2015), the lithology units
2000). Positive values of profile curvatures characterize of the study area are clay, sandstone, conglomerate, con-
convex areas (e.g., top of hills), while negative values glomerate and sands, and sandy-clay. These rocks are lar-
characterize concave areas (e.g., bottom of U-shaped val- gely covered by near-surface deposits and outcrop only for
leys); values around zero indicate flat surfaces (e.g., allu- 0.36, 1.88, 0.61, 0.1 and 0.09 % of the basin’s surface area,
vial plains). The influence of curvature on slope erosion respectively.
processes is mainly related to the convergence (promoting The five bedrock lithological units recognized in the
linear runoff) or divergence (promoting areal runoff) of map of Piacentini et al. (2015) were grouped into three
water flows. main classes (conglomerate, sandstone and clay) (Fig. 7) as
We extracted this parameter from the 5 m DTM using a function of prevalent granulometry and geomechanical
the ‘‘Curvature’’ function of the Spatial Analyst tools. This characteristics (Esu and Martinetti 1965; Chiocchini and
type of curvature combined both the profile and planform Giorelli 1994; Fiorillo 2004).
curvatures and was divided into three main classes (posi-
tive, negative and zero) (Fig. 7). Near-surface deposits Near-surface deposits play a key
role in governing water-groundwater infiltration processes
Drainage density The drainage density represents the (Tromp-van Meerveld and McDonnell 2006; Lanni et al.
degree of fluvial dissection and is defined as the total 2013), and their poor geotechnical properties (usually loose
stream length per unit area (Horton 1932, 1945). Many and highly permeable material) make these particularly
studies have demonstrated how drainage density, strictly prone to erosion (Dietrich and Dunne 1978; Evans 1982;
influenced by slope and underlying lithology, may alter Larsen et al. 2010; Conforti et al. 2011).
slope movement and vice versa (Kirby 1987; Oguchi 1997; In the Adriatic piedmont, as well as in many other
Schlunegger et al. 2006). This factor is particularly physiographic regions, the important bearing of near-sur-
meaningful for the evaluation of slope instability induced face deposits (eluvium, colluvium, slope deposits, soils) on
by critical hydrological conditions, such as along stream the shallow landslides has been well documented (e.g.,
confluence zones. Topography changes at the foot of slopes Nilsen and Turner 1975; Canuti et al. 1985; Crozier 1986;
(or landslides toe) caused by fluvial erosion might affect Montgomery and Dietrich 1994; Gokceoglu and Aksoy
the initiation (or reactivation) of landslides. Along rela- 1996; Atkinson and Massari 1998; Rosso et al. 2006; Lanni
tively small, entrenched and narrow valleys with low-de- et al. 2012; Fressard et al. 2014). These shallow move-
veloped drainage networks, as in the case of the Feltrino ments are mainly caused by the build-up of water pressure
stream valley, drainage density may be representative of into the ground (Campbell 1975; Wilson 1989) and,
the distance from streams, another common conditioning therefore, they are often triggered by rainstorm events
factor in landslide studies (e.g., Dou et al. 2015). (Canuti et al. 1985; Calcaterra et al. 2000; Iverson 2000).
We extracted the drainage network from the Regional These phenomena are poorly understood, and prediction of
Technical Maps of Abruzzo Region (2007) as vector data rainfall induced landslides is problematic (Guzzetti et al.
and derived the drainage density with the ‘‘Kernel Den- 2007).
sity’’ function of the Spatial Analyst tools. The resulting We grouped the eleven near-surface (continental)
pixel-based density values (km/km2) were sorted into six deposit units recognized in the map of Piacentini et al.
classes with the ‘‘Natural break’’ algorithm (Fig. 7). (2015) into six main classes (alluvial/fluvial, backfill,

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

beach, colluvial, eluvial-colluvial and scree slope deposits) Land cover and use Land cover is commonly considered
as a function of prevalent granulometry and geomorpho- to be a landslide conditioning factor since its variability
logical characteristics (Fig. 7). might influence vegetation cover in terms of mechanical
(e.g., slope and soil strength) and hydrological (e.g.,
Deposit thickness The thickness of near-surface deposits overland flow velocity, infiltration) characteristics
exerts a first-order control on both slope stability and (Greenway 1987; van Westen et al. 2003; Reichenbach
hydrological models (e.g., Dietrich et al. 1995; Iverson et al. 2014). Changes in land use distribution and type may
2000; Casadei et al. 2003; Rosso et al. 2006). Several be natural or induced by human activity, or, more com-
field (e.g., Freer et al. 2002; Tromp-van Meerveld and monly, may result from mixed anthropogenic-natural pro-
McDonnell 2006) and numerical (e.g., Hopp and cesses. The use of this factor is particularly significant in a
McDonnell 2009; Lanni et al. 2012) studies have directly densely populated and anthropized region like Italy, where
or indirectly shown that the variability of the depth of landslides caused by anthropogenic disturbances are fre-
near-surface deposits has a strong impact in controlling quently associated with changes in land use and vegetation
connectivity of saturated zones at the deposit-bedrock structure (Marchetti et al. 2001; Canuti et al. 2004; Buc-
interface, and in determining timing and position of colini et al. 2007; Capolongo et al. 2008).
shallow landslide initiation (Okimura 1989; Lanni et al. In this study, we extracted the prevalent destination of
2013). The spatial distribution of these deposits results land use from the Land Use map of Abruzzo Region (2000)
from the interaction of several variables such as geo- modified and updated via remote sensing analysis of col-
morphology, bedrock lithology, climate, biological, ored orthophotos (Abruzzo Region 2009) and field surveys.
chemical and physical processes (Summerfield 1997). As The final map was at a 1:5000 scale and up-to-date to the
a consequence, deposit thickness can be very irregular time of landslide inventory. We grouped the land use and
and its variability is often difficult to estimate (Dietrich cover classes, using the nomenclature of Corine Land
et al. 1995). Cover (2012), into eight classes (artificial surfaces, arable
In our study, we based the evaluation of this factor upon and permanent crops, pastures, heterogeneous agricultural
a geostatistical analysis of morphometric (slope) and geo- areas, forests, shrubs, beaches) (Fig. 7).
logic subsurface (deposits thickness) data according to
relevant literature (Saulnier et al. 1997; Catani et al. 2010). Statistical analysis
This method exploited the information provided by direct
measurements of deposit thickness such as surficial bore- Bivariate statistical analysis for this susceptibility study
holes (about 150) and field-survey observation (scarps, involved the comparison between the independent vari-
road, excavations, quarries, outcrops, road and railroad ables (i.e., the seven conditioning factors) and the distri-
cuttings). The factor was intrinsically extrapolated to the bution of landslides (i.e., the 2012–2015 inventory map)
whole basin through the following linear (pixel-based) (DeGraff and Romesburg 1980; Irigaray et al. 1999, 2007;
relationship between deposit thickness (D) and local slope Clerici et al. 2002; Poli and Sterlacchini 2007; Romeo et al.
(tanb): 2011; Bourenane et al. 2015). In order to rank the corre-
D ¼ 6:34 tan b þ 5:49: sponding classes according to their influence in landslide
formation, to show how much each factor was represented
Lastly, the well-known geomorphological evolution of in the observed landslides, a Landslide Index (LI) was
these deposits and related processes (Piacentini et al. 2015) defined for each class of the conditioning factors (Van
allowed us to define six main classes (Fig. 7) of thickness Westen 1993, 1994; Van Westen et al. 1997). The seven
(0–0.5; 0.5–3; 3–5; 5–10; [10 m), roughly corresponding classified factors and the pre-2012 landslide inventory were
to the morphological units of the landscape (Table 3). firstly rasterized (5 m cell-size raster layers), and then

Table 3 Classes and ranges of

Class Slope range () Morphological unit
deposits thickness factor with
reference to the main 1 0–3 Summit of tabular reliefs, alluvial plains
morphological units
characterizing the Feltrino 2 3.1–5 Lower portion of concave valleys, lower hillslopes
Basin 3 5.1–10 Midslopes
4 10.1–15 Midslopes, v-shaped valley flanks
5 15.1–30 Upper and midslopes
6 30.1–45 Higher part of the slopes
7 [45.1 Scarps, higher part of the slopes

123
 M. Sciarra et al.

processed on GIS platform with the aid of a spreadsheet. In cumulative curves. Finally, we evaluated the area under the
this step, all landslides portrayed in the pre-2012 inventory curve (AUC) for estimating the model performance. The
have been considered without distinction of type or plot charts with the relative AUC values for the three cases
activity. are reported in Fig. 9.
We calculated LI as the ratio between the areal surface The shape of the curves provides information about the
of landslides in the class (Lai) and the total surface of the degree of reliability of the model, with a larger area
considered class (Tai), according to the following rela- meaning higher accuracy achieved, and, therefore, the
tionship (simplified in Romeo et al. 2011): AUC values provide a quantitative evaluation. AUC ranged
LI ¼ Lai =Tai between 0 and 1: higher values (close to 1) suggest higher
model reliability, while lower values (near or less than 0.5)
Then, we calculated the weight (Wj) of each of the suggest that the model is invalid (Chung and Fabbri 2003).
seven classes factors by the following ratio, normalizing
them from 0 to 100:


Wj ¼ LImin;j =LImax;j  100 Results and discussion

where LImin,j was the lowest value of LI for each factor Landslide inventory data (pre-2012 and 2012–2015 maps)
class, and LImax,j was the maximum value of the relative revealed that a total of 294 landslides occurred in the study
factor class. The LI and Wj values for each class are area. All of these landslides fall into three main types of
reported in Table 4. failure: roto-translational slides (64.3 %), shallow land-
The final step in creating the preliminary landslide slides (34 %), and complex movements (1.7 %). As doc-
susceptibility map was to combine all weighted layers into umented by the 2012–2015 inventory, the main triggering
a single preliminary map through the ‘‘Weighted overlay’’ factor for shallow landslides in the study area was the
function of the Spatial Analyst tool in ArcGIS 10.1. The heavy rainfall that occurred during the rainy seasons. In the
resulting landslide susceptibility map (5 m cell-size) other inventories (pre-2012 and PAI maps) there were few
delineated the area into different zones classified by four shallow landslides as they were short-lived landforms that
susceptibility levels: very high, high, medium and low. were recurrently reworked (or destroyed) by agricultural
Subdivision within the classes followed the criterion of practices (e.g., soil plowing) and subsequent slope pro-
natural breaks according to Jenks’ algorithm (Ruff and cesses. In terms of landslide covered areas, the PAI
Czurda 2008; Falaschi et al. 2009; Piacentini et al. 2012; inventory roughly portrayed the same landslides stored in
De Guidi and Scudero 2013). The natural breaks subdivi- the pre-2015 inventory. However, the number of landslides
sion between two classes resulted from jumps of the fre- is underestimated (almost half) due to the smaller scale and
quency distribution of the LI. The grid-cell-based the lower accuracy of remote sensing-based mapping like
preliminary susceptibility map is shown in Fig. 8. that of PAI.

Validation Statistical analysis

We tested the preliminary susceptibility map for both The bivariate statistical analysis allowed us to perform a
model accuracy and prediction skills criteria via Success susceptibility model for the landslides in the study area,
and Prediction rate curves, respectively. Success rate curve proving the relevance of each factor in conditioning the
considered the same landslides involved in the statistical distribution of landslides and the relative weight of each
analysis (i.e., pre-2012 inventory map), whereas prediction factors’ class (Table 4). In this way, we were able to rec-
rate curve considered independent samples of landslides ognize combinations of the seven main factors that con-
(i.e., 2012–2015 inventory map). A supplementary Success ditioned slope instability. In most cases, landslides
rate curve was prepared for the case of the pre-existing occurred in areas characterized by slope angle varying
official landslide inventory (i.e., PAI inventory). Since the from 10 to 15 (LI = 0.395), weakly convex slopes
preliminary map was a pixel-based raster, point features of (LI = 0.319), and with relatively (for the area) moderate
the 2012–2015 inventory were buffered by constructing values of drainage density (1.9–2.6 km/km2; LI = 0.321).
circles of 20 m in diameter around them, in order to ensure Moreover, the bedrock lithology most prone to landslides
overlapping with at least two pixels. was demonstrated to be clayey (LI = 0.417), and, as far as
First, we grouped the landslide inventories into several the near-surface deposits are concerned (lithology and
classes according to their LI values. Subsequently, we thickness), most of the landslides occurred on the colluvial
divided the number of landslide pixels in each class by the deposits (LI = 0.340), especially where thickness ranged
total number of pixels in that class, and plotted the from 5 to 10 m (LI = 0.510). Furthermore, for the land

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

Table 4 Conditioning factors

Factor ID Class LI Wj N pixel
considered in the statistical
analysis with relative classes, Slope ()
landslides index (LI), weights
(Wj) and number of pixels (5 m 1 0–3 0.051 12.87 566,306
cell-size) 2 3.1–5 0.203 51.47 240,550
3 5.1–10 0.365 92.41 575,260
4 10.1–15 0.395 100 337,713
5 15.1–30 0.230 58.14 288,325
6 30.1–45 0.050 12.77 32,562
7 [45.1 0.0004 0.11 295
Topographic curvature
-1 Concave 0.243 76.15 669,295
0 Flat 0.211 66.14 1,012,150
1 Convex 0.319 100 358,133
Drainage density (km/km-2)
1 0–1 0.082 25.69 313,706
2 1.1–1.8 0.240 74.66 822,037
3 1.9–2.6 0.321 100 513,861
4 2.7–3.6 0.254 79.01 321,961
5 [3.7 0.283 88.00 69,275
Bedrock lithology
1 Conglomerate 0.001 0.26 701,977
2 Sandstone 0.265 63.54 460,193
3 Clay 0.417 100 877,704
Near-surface deposits
1 Backfill deposits 0.024 6.93 103,854
2 Alluvial/Fluvial deposits 0.000067 0.01 67,073
3 Beach deposits 0.027 0 168
4 Colluvial deposits 0.340 100 1,418,688
5 Eluvial-colluvial deposits 0 0 359,739
6 Scree slope deposits 0.002 0.59 29,353
7 No deposits 0 0 60,258
Deposits thickness (m)
1 0–0.5 0.010 0.98 107,099
2 0.5–3 0.100 20.43 869,346
3 3.0–5.0 0.160 31.82 358,791
4 5.0–10 0.510 100 624,661
5 [10.1 0.250 49.21 79,951
Land use
1 Artificial surfaces 0.0201 6.64 279,602
2 Arable and permanent crops 0.3033 100 1,289,395
3 Pastures 0.133 43.84 6646
4 Heterogeneous agricultural areas 0.301 99.22 307,474
5 Forests 0.246 81.16 135,783
6 Shrubs 0 0 21,254
7 Beaches 0 0 28

cover and use, the areas most prone to landslides were the colluvial deposits (1,418,688 pixels), arable and permanent
arable lands and permanent crops (LI = 0.303). crops (1,289,395 pixels), flat topographic curvature
Concerning the spatial distribution of the conditioning (1,012,150 pixels), clay bedrock unit (877,704 pixels),
factors (Table 4), the most widespread classes were deposit thickness between 0.5 and 3 m (869,346 pixels),

123
 M. Sciarra et al.

Fig. 8 Susceptibility map

Fig. 9 a Success rate curve plot with the relative area under the curve case of the PAI inventory. Colors indicating different landslide
(AUC) value for the case of pre-2012 inventory. b Prediction rate susceptibility levels: red very high susceptibility, orange high
curve plot with the relative AUC value for the case of 2012–2015 susceptibility, yellow medium susceptibility, green low susceptibility
inventory. c Success rate curve with the relative AUC value for the

drainage density between 1.1 and 1.8 km/km2 (822,037 classes using the natural-breaks method of Jenks, defining
pixels), slope angles between 5.1 and 10 (575,260 pixels) the following susceptibility classes:
and 0–3 (566,306 pixels). On the other hand, other
– ‘‘low’’ susceptibility class, representing 21.45 % of the
classes such as beaches (28 pixels), pastures (6646 pixels),
surface basin (10.9 km2);
slope angle [45.1 (295 pixels) are not well represented in
– ‘‘medium’’ susceptibility class, representing 18.52 % of
the Feltrino basin.
the surface basin (9.41 km2);
The output data of the present landslide susceptibility
– ‘‘high’’ susceptibility class, representing 29.82 % of the
assessment were the grid-based (5 m cell-size) susceptibility
surface basin (15.15 km2);
zonation shown in the map of Fig. 8. The susceptibility
– ‘‘very high’’ susceptibility class, representing 30.21 %
values ranged between 0 and 100 in each combination of
of the surface basin (15.35 km2).
classes of the conditioning factors. These values, referring to
mass movements without any distinction of type or state of The landslide susceptibility map showed that areas with
activity, were then reclassified into four main susceptibility ‘‘very high’’ and ‘‘high’’ susceptibility classes were typical

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

of the mid-lower portion of the slopes, especially near the ‘‘low’’ susceptibility class, whereas 31.6 % fell in the
streams confluences. The areal distribution of the ‘‘med- ‘‘high’’ and 62.9 % in the ‘‘very high’’ susceptibility
ium’’ susceptibility class was more widespread along the classes.
scarps that bordered the top of hilly reliefs. Lastly, the
‘‘low’’ susceptibility class covered flat areas (i.e., alluvial
plains and top of the plateaux) and, to a lesser extent, Conclusions
gentle slopes characterizing the easternmost basin.
This study presents a landslide susceptibility analysis for a
Validation test-area of the Adriatic hilly-coastal zone of Central Italy,
and verified the validity and utlity of the derived suscep-
The computed susceptibility map of the Feltrino basin was tibility model. The main outcomes regard the validation
validated through the Success rate and Prediction rate and prediction procedures, proving the effectiveness of the
curves as described before. In all of the cases analyzed, the model in both accuracy and prediction skills. In fact, within
curves had similar shapes: high-sloping followed by a the performance scheme of Guzzetti et al. (2006), the
steady and progressive slope decline (Fig. 9). As we model shows an acceptable percentage of correctly classi-
assumed that the model was correct, we expected that the fied landslides. Moreover, the comparison between the
Success rate curve was better than that of the Prediction obtained susceptibility maps and the multi-temporal
rate. Considering the Success rate curve based on the pre- inventory (2012–2015) highlights that many mapped
2012 landslide inventory, the AUC value was 0.76, indi- landslides were newly formed, proving the reliability and
cating a good model fit with the distribution of known prediction skills of the susceptibility analysis performed.
landslides in the pre-2012 inventory. In detail, the areas Landslides generated by rains, such as those that occurred
portrayed as ‘‘very high’’ and ‘‘high’’ susceptibility classes during the 2012–2015 period and registered by systematic
covered about 98.4 % of the total mapped landslide (of field surveys, offer a good opportunity to validate the
which 68.2 % is ‘‘very high’’), whereas, only 1.6 % of the susceptibility maps drawn beforehand. The comparison of
landslides fell within the ‘‘low’’ susceptibility class. the susceptibility map with the official regional landslides
The following validation procedures, which initially inventory (PAI inventory) provided the opportunity to
considered firstly the 2012–2015 inventory and then the prove that our susceptibility model could also be used also
PAI inventory, focused on reaching two different goals: the in the absence of detailed landslide mapping, by consid-
first aimed at verifying whether the areas having ‘‘high’’ or ering already existing maps. In the next step of our study, it
‘‘very high’’ susceptibility will actually be affected by will be interesting to further validate the potential of the
landslides, demonstrating the prediction power of the model by applying it to other areas with similar geological-
model; the second aimed at extending the application of geomorphological and climatic features. Enlarging the test
our model in other zones with similar geological and area could enhance the spatial coverage of some factors
geomorphological characteristics where detailed (field- classes. This extension, however, will require a deep
based) landslide inventories are lacking, demonstrating the knowledge of the near-surface deposits (both their type and
usability of the model with other existing official landslide thickness). A comprehensive susceptibility analysis is
databases. important not only for understanding the geomorphological
The Prediction rate curve, based on the 2012–2015 dynamics of selected areas, but also for taking the first
inventory, achieved an ACU value of 0.70, suggesting fundamental step in landslide hazard assessment and mit-
fairly good prediction skills of the model, even though the igation in the entire Adriatic hilly-coastal area.
AUC value could be enhanced by including the shallow Regarding the statistical approach, the bivariate tech-
landslide events (contained in the 2012–2015 inventory) in nique method separated each factor considered without
the model. This could better refine the model in predicting providing their relative weights. For this reason, the final
shallow landslides. Few recent landslides (3.7 %) fell map was a simple combination of factors which, although
within the ‘‘low’’ susceptibility class, whereas 80.8 % were recognized as those that are most important, might not have
correctly predicted by the model as they occurred in areas equal incidence. Our results could be further improved by
with ‘‘high’’ (40.6 %) and ‘‘very high’’ (40.2 %) suscepti- the application of more sophisticated methodologies giving
bility classes. a susceptibility analysis (i.e., multivariate techniques) that
The Success rate curve based on the PAI inventory would allow estimation of the relative influence of single
reached an AUC value of 0.72, very close to that of the pre- causative factors. Moreover, in this study, susceptibility
2012 inventory. This proves the effectiveness of the model modeling was performed without distinguishing landslide
also compared with the official landslide inventory. In this type or activity. However, the deep knowledge of land-
case, only 0.6 % of the mass movements matched the slides in the analyzed area (included type and activity state)

123
 M. Sciarra et al.

allowed us to recognize the main conditioning factors and autoritabacini.regione.abruzzo.it/index.php/carta-inventario-pai.

to qualitatively (more than quantitatively) control the Accessed 2 February 2016
Aleotti P, Chowdhury R (1999) Landslide hazard assessment:
validity of the output maps. summary review and new perspectives. Bull Eng Geol Environ
The resulting susceptibility map, on the one hand 58:21–44. doi:10.1007/s100640050066
emphasizes slope, lithology and drainage as main condi- Anbalagan R (1992) Landslide hazard evaluation and zonation
tioning factors, but, on the other hand highlights the key mapping in mountainous terrain. Eng Geol 32(4):269–277.
doi:10.1016/0013-7952(92)90053-2
role played by superficial deposits in the susceptibility Ardizzone F, Cardinali M, Carrara A, Guzzetti F, Reichenbach P
assessment. This latter feature (the type and distribution (2002) Impact of mapping errors on the reliability of landslide
of surficial deposits) generally has not been considered in hazard maps. Nat Hazards Earth Syst Sci 2:3–14. doi:10.5194/
the literature, but plays a key role in the distribution of nhess-2-3-2002
Atkinson PM, Massari R (1998) Generalised linear modelling of
landslides in the analyzed area. A reliable susceptibility susceptibility to landsliding in the Central Appennines, Italy.
model for this area cannot neglect an accurate charac- Comput Geosci 24(4):373–385. doi:10.1016/S0098-
terization of the near surface deposits (colluvial, alluvial, 3004(97)00117-9
slope and landslide deposits) as they are particularly Aucelli PPC, Cavinato GP, Cinque A (1996) Indizi geomorfologica di
tettonica plio-quaternaria sul piedimonte adriatico dell’Appen-
prone to erosion processes. For this reason, fieldwork is of nino abruzzese. Il quaternario 9(1):299–302
paramount importance in the reliable mapping of this Ayalew L, Yamagishi H (2005) The application of GIS-based logistic
particular (but also of the other) conditioning factors of regression for landslide susceptibility mapping in the Kakuda-
landslides. An expert knowledge of the phenomena and Yahiko Mountains, Central Japan. Geomorphology 65:15–31.
doi:10.1016/j.geomorph.2004.06.010
mechanisms in a test-area can allow accurate explanation Ayalew L, Yamagishi H, Ugawa N (2004) Landslide susceptibility
of the results of modeling and an understanding of the mapping using GIS-based weighted linear combination, the case
possible application of statistical models with the same in Tsugawa area of Agano River, Niigata Prefecture, Japan.
dataset in similar physiographic regions. Furthermore, the Landslides 1:73–81. doi:10.1007/s10346-003-0006-9
Bennie J, Hill MO, Baxter R, Huntley B (2006) Influence of slope and
estimation formula of deposit thickness could be aspect on long-term vegetation change in British chalk grass-
improved by a topographic curvature variable, which can lands. J Ecol 94(2):355–368. doi:10.1111/j.1365-2745.2006.
be associated closely with hillslope curvature (Dietrich 01104.x
et al. 1995). Bourenane H, Bouhadad Y, Guettouche MS, Braham M (2015) GIS-
based landslide susceptibility zonation using bivariate statistical
As outlined by Fell et al. (2008), the quality of the input and expert approaches in the city of Constantine (Northeast
data (accuracy of mapping, adequacy of inventory, reso- Algeria). Bull Eng Geol Environ 74:337. doi:10.1007/s10064-
lution of datasets) strongly influences the quality of the 014-0616-6
susceptibility map derived by this statistical approach. In Brabb EE (1984) Innovative approaches to landslide hazard and risk
mapping. In: Proceedings 4th international symposium on
our study, the high resolution of datasets and the accuracy landslides, Canadian Geotechnical Society, Toronto, 1:307–324
of mapping and inventories used in the construction of the Brenning A (2005) Spatial prediction models for landslide hazards:
model allowed us to overcome this issue. review, comparison and evaluation. Nat Hazards Earth Syst Sci
5:853–862
Acknowledgments The authors are grateful to two anonymous Buccolini M, Gentili B, Materazzi M, Aringoli D, Pambianchi G,
reviewers whose suggestions improved the quality of this paper. They Piacentini T (2007) Human impact and slope dynamics evolu-
are also very grateful to Prof. Brent Poe for having revised and tionary trends in the monoclinal relief of the Adriatic central
improved the language of the paper. Italy. Catena 71:96–109
Buccolini M, Coco L, Cappadonia C, Rotigliano E (2012) Relation-
ships between a new slope morphometric index and calanchi
erosion in northen Sicily, Italy. Geomorphology 149–150:41–48.
References doi:10.1016/j.geomorph.2012.01.012
Calcaterra D, Parise M, Palma B, Pelella L (2000) The influence of
Abruzzo Region (2000) 1:25,000 scale Land use maps. L’Aquila: meteoric events in triggering shallow landslides in pyroclastic
Struttura Speciale di Supporto Sistema Informativo Regione deposits of Campania, Italy. In: Bromhead E, Dixon N, Ibsen
Abruzzo. Retrieved from http://opendata.regione.abruzzo.it/ ML (eds) Proceedings of the 8th international symposium on
Abruzzo Region (2007) 1:5,000 scale regional technical maps. landslides, vol 1. Balkema, Cardiff, pp 209–214
L’Aquila: Struttura Speciale di Supporto Sistema Informativo Campbell R (1975) Soil slips, debris flows and rainstorms in the Santa
Regione Abruzzo. Retrieved from http://opendata.regione. Monica Mountains and vicinity, Southern California. USGS
abruzzo.it/ Professional Paper, 851, pp 51
Abruzzo Region (2009) 1:5,000 scale orthophotos color images. Cantalamessa G, Di Celma C (2004) Sequence response to synde-
L’Aquila: Struttura Speciale di Supporto Sistema Informativo positional regional uplift: insights from high-resolution sequence
Regione Abruzzo. Retrieved from http://opendata.regione. stratigraphy of late Early Pleistocene strata, Periadriatic Basin,
abruzzo.it/ central Italy. Sediment Geol 164:283–309
Abruzzo-Sangro Basin Authority (2005) Piano Stralcio di Bacino per Canuti P, Focardi P, Garzonio CA (1985) Correlation between rainfall
l’Assetto Idrogeologico dei Bacini di Rilievo Regionale and landslides. Bull Int Assoc Eng Geol 32:49–54
Abruzzesi e del Bacino del Fiume Sangro. (L.R. 18.05 1989 Canuti P, Casagli N, Ermini L, Fanti R, Farina P (2004) Landslide
n.81 e L. 24.08.2001)– Carta inventario– scala 1:25.000. http:// activity as a geoindicator in Italy: significance and new

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

perspectives from remote sensing. Environ Geol 45(7):907–919. Clerici A, Perego S, Tellini C, Vescovi P (2002) A procedure for
doi:10.1007/s00254-003-0952-5 landslide susceptibility zonation by the conditional analysis
Capolongo D, Pennetta L, Piccarreta M, Fallacara G, Boenzi F (2008) method. Geomorphology 48:349–364. doi:10.1016/S0169-
Spatial and temporal variations in soil erosion and deposition 555X(02)00079-X
due to land-levelling in a semi-arid area of Basilicata (Southern Conforti M, Aucelli PPC, Robustelli G, Scarciglia F (2011)
Italy). Earth Surf Proc Land 33(3):364–379. doi:10.1002/esp. Geomorphology and GIS analysis for mapping gully erosion
1560 susceptibility in the Turbolo catchment (Northern Calabria,
Cappadonia C, Coco L, Buccolini M, Rotigliano E (2016) From slope Italy). Nat Hazards 56:881–898
morphometry to morphogenetic processes: an integrated Conoscenti C, Ciaccio M, Caraballo-Arias N, Gómez-Gutiérrez
approach of field survey, GIS morphometric analysis and Álvaro, Rotigliano E, Agnesi V (2015) Assessment of suscep-
statistics in Italian badlands. Land Degrad Dev 27(3):851–862. tibility to earth-flow landslide using logistic regression and
doi:10.1002/ldr.2449 multivariate adaptive regression splines: a case of the Belice
Carminati E, Doglioni C (2004) Europe–Mediterranean tectonics. In: River basin (western Sicily, Italy). Geomorphology 242:49–64.
Encyclopedia of geology. Elsevier, Dordrecht, pp 135–146 doi:10.1016/j.geomorph.2014.09.020
Carrara A (1989) Landslide hazard mapping by statistical methods: a Corine Land Cover (2012) Land cover map. Environmental European
‘‘black-box’’ model approach. In: Proceedings of the Workshop Agency. Retrieved from http://land.copernicus.eu/
on Natural Disasters in European–Mediterranean Countries, Crescenti U, Milia ML, Rusciadelli G (2004) Stratigraphic and
Perugia, CNR-US NFS, pp 427–445 tectonic evolution of the Pliocene Abruzzi basin (Central
Carrara A, Cardinali M, Guzzetti F, Reichenbach P (1995) GIS Apennines, Italy). Boll Soc Geol It 123:163–174
technology in mapping landslide hazard. In: Carrara A, Guzzetti Crozier M (1986) Climatic triggering of landslide episodes. In:
F (eds) Geographical Information Systems in assessing natural Landslides: causes, consequences and environment. Croom
hazards. Kluwer, Dordrecht, pp 35–176 Helm, London, pp 169–192
Carrara A, Guzzetti F, Cardinali M, Reichenbach P (1999) Use of GIS Cruden DM, Varnes DJ (1996) Landslide types and processes. In: Turner
technology in the prediction and monitoring of landslide hazard. AK, Schuster RL (eds) Landslides investigation and mitigation.
Nat Hazards 20(2–3):117–135 Transportation research board, US National Research Council.
Casadei M, Dietrich WE, Miller NL (2003) Testing a model for Special Report 247, Washington, DC, Chapter 3, pp 36–75
predicting the timing and location of shallow landslide initiation D’Agostino N, Jackson JA, Dramis F, Funiciello R (2001) Interac-
in soil mantled landscapes. Earth Surf Process Landf tions between mantle upwelling, drainage evolution and active
28(9):925–950. doi:10.1002/esp.470 normal faulting: an example from the Central Appennines
Cascini L (2008) Applicability of landslide susceptibility and hazard (Italy). Geophys J Int 147(2):475–497. doi:10.1046/j.1365-246X.
zoning at different scales. Eng Geol 102:164–177. doi:10.1016/j. 2001.00539.x
enggeo.2008.03.016 D’Alessandro L, Genevois R, Paron P, Ricci F (2004) Type and
Catani F, Segoni S, Falorni G (2010) An empirical geomorphology- distribution of rainfall triggered landslides in the Abruzzi region
based approach to the spatial prediction of soil thickness at (central Italy) after the January 2003 meteoric event. 32nd
catchment scale. Water Resour Res 46. doi:10.1029/ International Geological Congress: ‘‘From the Mediterranean
2008wr007450 Area toward a Global Geological Renaissance. Geology, Natural
Cavinato GP, De Celles PG (1999) Extensional basins in the Hazards and Cultural Heritage’’. Firenze 20–28 Aug 2004
tectonically bimodal central Apennines fold-thrust belt, Italy: D’Alessandro L, Miccadei E, Piacentini T (2008) Morphotectonic
response to corner flow above subducting slab in retrograde study of the lower Sangro River valley (Abruzzi, Central Italy).
motion. Geology 27:955–958. doi:10.1130/0091-7613 Geomorphology 102:145–158. doi:10.1016/j.geomorph.2007.06.
Cerdà A, Garcı́a-Fayos P (1997) The influence of slope angle on 019
sediment, water and seed losses on badland landscapes. Dai FC, Lee CF (2002) Landslide characteristics and slope instability
Geomorphology 18:77–90 modeling using GIS, Lantau Island, Hong Kong. Geomorphol-
Cevik E, Topal T (2003) GIS-based landslide susceptibility mapping ogy 42:213–228. doi:10.1016/S0169-555X(01)00087-3
for a problematic segment of the natural gas pipeline, Hendek Dai FC, Lee CF, Li J, Xu ZW (2001) Assessment of landslide
(Turkey). Environ Geol 44:949–962. doi:10.1007/s00254-003- susceptibility on the natural terrain of Lantau Island, Hong
0838-6 Kong. Environ Geol 43:381–391. doi:10.1007/s002540000163
Chiocchini U, Giorelli R (1994) Aspetti geomorfologici e costruttivi De Guidi G, Scudero S (2013) Landslide susceptibility assessment in
delle gallerie Cintioni, San Giovanni e Diavolo della linea the Peloritani Mts. (Sicily, Italy) and clues for tectonic control of
Ancona-Bari tra Ortona e Casalbordino. Geol Appl e Idrog relief processes. Nat Hazards Earth Syst Sci 13:949–963. doi:10.
29:1–17 5194/nhess-13-949-2013
Chiocchini U, Barbieri M, Madonna S, Di Stefano A, Poteti M (2006) DeGraff JV, Romesburg HC (1980) Regional landslide-susceptibility
I depositi del Pleisotcene tra Ortona e la stazione ferroviaria di assessment for wildland management: a matrix approach. In:
Casalboridino (provincia di Chieti). Rend Soc Geol It 2:3–14 Coates DR, Vitek JD (eds) Thresholds in geomorphology, vol
Chung CJF, Fabbri AG (2003) Validation of spatial prediction models 19. Alien & Unwin, Boston, pp 401–414
for landslide hazard mapping. Nat Hazards 30(3):451–472. Dietrich WE, Dunne T (1978) Sediment budget for a small catchment
doi:10.1023/B:NHAZ.0000007172.62651.2b in mountainous terrain. Z Geomorphol Suppl 29:191–206
Chung CJF, Fabbri AG (2008) Predicting landslides for risk Dietrich WE, Reiss R, Hsu ML, Montgomery DR (1995) A process-
analysis—spatial models tested by a cross-validation procedure. based model for colluvial soil depth and shallow landsliding
Geomorphology 94:438–452. doi:10.1016/j.geomorph.2006.12. using digital elevation data. Hydrol Processes 9:383–400
036 Dou J, Yamagishi H, Pourghasemi HR, Yunus AP, Song X, Xu Y,
Chung CJF, Fabbri AG, van Westen CJ (1995) Multivariate Zhongfan Z (2015) An integrated artificial neural network model
regression analysis for landslide hazard zonation. In: Carrara for the landslide susceptibility assessment of Osado Island,
A, Guzzetti F (eds) Geographical information systems in Japan. Nat Hazards 78(3):1749–1776. doi:10.1007/s11069-015-
assessing natural hazards. Kluwer, Dordrecht, 107–134 1799-2

123
 M. Sciarra et al.

Dramis F, Guida D, Cestari A (2011) Nature and aims of geomor- Guzzetti F, Reichenbach P, Ardizzone F, Cardinali M, Galli M (2006)
phological maps. In: Smith M, Paron P, Griffiths J (eds) Estimating the quality of landslide susceptibility models. Geo-
Geomorphological mapping. Elsevier, pp 39–73 morphology 81:166–184. doi:10.1016/j.geomorph.2006.04.007
Ercanoglu M, Gokceoglu C, Van Asch TWJ (2004) Landslide Guzzetti F, Peruccacci S, Rossi M, Stark CP (2007) Rainfall
susceptibility zoning north of Yenice (NW Turkey) by multi- thresholds for the initiation of landslides in central and southern
variate statistical techniques. Nat Hazards 32:1–23. doi:10.1023/ Europe. Meteorol Atmos Phys 98:239–267
B:NHAZ.0000026786.85589.4a Hoek E, Bray JW (1981) Rock slope engineering (revised third
Esu F, Martinetti S (1965) Considerazioni sulle caratteristiche edition). Institute of Mining and Metallurgy, London
tecniche delle argille azzurre dell’Italia sud-orientale in relazione Hopp L, McDonnell JJ (2009) Connectivity at the hillslope scale:
alle diverse condizioni stratigrafiche e tettoniche. Geol Appl e identifying interactions between storm size, bedrock permeabil-
Idrog 7:197–215 ity, slope angle and soil depth. J Hydrol 376:378–391. doi:10.
Evans IS (1972) General geomorphology, derivatives of altitude and 1016/j.jhydrol.2009.07.047
descriptive statistics. Chorley RJ (ed) Spatial analysis in Horton RE (1932) Drainage basin characteristics. Trans Amer
geomorphology. Harper and Row, NY, pp 17–90 Geophys Union 13:350–361
Evans SG (1982) Landslides and surficial deposits in urban areas of Horton RE (1945) Erosional development of streams and their
British-Columbia—a review. Can Geotech J 19(3):269–288 drainage basins: hydrophysical approach to quantitative mor-
Falaschi F, Giacomelli F, Federici PR, Puccinelli A, D’Amato Avanzi phology. Bull Geol Soc Am 56:275–375
G, Pochini A, Ribolini A (2009) Logistic regression versus Irigaray C, Fernandez T, El Hamdouni R, Chacon J (1999)
artificial neural networks: landslide susceptibility evaluation in a Verification of landslide susceptibility mapping. A case study.
sample area of the Serchio River valley, Italy. Nat Hazards Earth Surf Proc Land 24:537–544. doi:10.1002/(SICI)1096-
50:551–569. doi:10.1007/s11069-009-9356-5 9837(199906)24:6
Felicı́simo ÁM, Cuartero A, Remondo J, Quirós E (2013) Mapping Irigaray C, Fernandez T, El Hamdouni R, Chacon J (2007) Evaluation
landslide susceptibility with logistic regression, multiple adap- and validation of landslide-susceptibility maps obtained by a GIS
tive regression splines, classification and regression trees, and matrix method: examples from the Betic Cordillera (southern
maximum entropy methods: a comparative study. Landslides Spain). Nat Hazards 41:61–79. doi:10.1007/s11069-006-9027-8
10(2):175–189. doi:10.1007/s10346-012-0320-1 ISPRA (2007) IFFI, Inventario dei Fenomeni Franosi in Italia. http://
Fell R, Corominas J, Bonnard C, Cascini L, Leroi E, Savage WZ www.apat.gov.it/site/it-IT/Progetti/IFFI_-_Inventario_dei_feno
(2008) Guidelines for landslide susceptibility, hazard and risk meni_franosi_in_Italia/. Accessed 4 June 2011
zoning for land-use planning. Eng Geol 102:85–98. doi:10.1016/ Iverson RM (2000) Landslide triggering by rain infiltration. Water
j.enggeo.2008.03.014 Resour Res 36(7):1897–1910
Fiorillo F (2004) Blue clay slope angle in relation to some geological Kirby MJ (1987) Modelling some influences of soil erosion,
and geotechnical characteristics, Italy. Q J Eng Geol Hydrogeol landslides and valley gradient on drainage density and hollow
37(1):49–59. doi:10.1144/1470-9236/03-027 development. Catena Suppl 10:1–14
Fox DM, Bryan RB, Price AG (1997) The influence of slope angle on Lanni C, Borga M, Rigon R, Tarolli P (2012) Modelling shallow
final infiltration rate for interrill conditions. Geoderma landslide susceptibility by means of a subsurface flow path
80:181–194 connectivity index and estimates of soil depth spatial distribu-
Freer J, McDonnell JJ, Beven KJ, Peters NE, Burns DA, Hooper RP, tion. Hydrol Earth Syst Sci 16:3959–3971. doi:10.5194/hess-16-
Aulenbach B, Kendall C (2002) The role of bedrock topography 3959-2012
on subsurface storm flow. Water Resour Res 38(12):1269. Lanni C, Mcdonnell J, Hopp L, Rigon R (2013) Simulated effect of
doi:10.1029/2001WR000872 soil depth and bedrock topography on near-surface hydrologic
Fressard M, Thiery Y, Maquaire O (2014) Which data for quantitative response and slope stability. Earth Surf Process Landforms
landslide susceptibility mapping at operational scale? Case study 38(2):146–159. doi:10.1002/esp.3267
of the Pays d’Auge plateau hillslopes (Normandy, France). Nat Larsen IJ, Montgomery DR, Korup O (2010) Landslide erosion
Hazards Earth Syst Sci 14:569–588. doi:10.5194/nhess-14-569- controlled by hillslope material. Nat Geosci 3(4):247–251.
2014 doi:10.1038/ngeo776
Gokceoglu C, Aksoy H (1996) Landslide susceptibility mapping of Lee S, Min K (2001) Statistical analysis of landslide susceptibility at
the slopes in the residual soils of the Mengen Region (Turkey) Yongin, Korea. Environ Geol 40:1095–1113. doi:10.1007/
by deterministic stability analyses and image processing tech- s002540100310
nique. Eng Geol 44(1–4):147–161. doi:10.1016/S0013- Lee S, Ryu J, Min K, Won J (2001) Development of two artificial
7952(97)81260-4 neural network methods for landslide susceptibility analysis. Int
Gómez H, Kavzoglu T (2005) Assessment of shallow landslide Geos Rem Sens Symposium 5:2364–2366
susceptibility using artificial neural networks in Jabonosa River Marchetti L, Pellegrini M, Bondesan D, Castaldini GB, Castiglioni C,
Basin, Venezuela. Eng Geol 78(1–2):11–27. doi:10.1016/j. Tellini C (2001) Forme legate a interventi antropici - Forms
enggeo.2004.10.004 connected to anthropic activities. (Comitato glaciologico ital-
Greenway DR (1987) Vegetation and slope stability. In: Anderson iano)–Suppl Geog Fis e Din Quat IV:123–139. ISSN: 1724-475 7
MG, Richards KS (eds) slope stability. Wiley, Chichester, Montgomery DR, Dietrich WE (1994) A physically based model for
pp 187–230 the topographic control on shallow landsliding. Water Resour
Guzzetti F, Cardinali M, Reichenbach F (1996) The Influence of Res 30:1153–1171
Structural Setting and Lithology on Landslide Type and Pattern. Moore ID, Grayson RB, Ladson AR (1991) Digital terrain modelling:
Environ Eng Geosci 2(4):531–555 a review of hydrological, geomorphological, and biological
Guzzetti F, Carrara A, Cardinali M, Reichenbach P (1999) Landslide applications. Hydrol Process 5:3–30
hazard evaluation: a review of current techniques and their Nefeslioglu HA, Gokceoglu C, Sonmez H (2008) An assessment on
application in a multi-scale study, Central Italy. Geomorphology the use of logistic regression and artificial neural networks with
31:181–216. doi:10.1016/S0169-555X(99)00078-1 different sampling strat-egies for the preparation of landslide

123
Assessment and validation of GIS-based landslide susceptibility map: a case study from…

susceptibility maps. Eng Geol 97:171–191. doi:10.1007/s12303- frana: un’applicazione nella Regione Marche (Italia Centrale).
015-0026-1 Italian J Eng Geol Environ 2:63–77
Nilsen TH, Turner BL (1975) Influence of rainfall and ancient Rosso R, Rulli MC, Vannucchi G (2006) A physically based model
landslide deposits on recent landslides (1950–71) in urban areas for the hydrologic control on shallow landsliding. Water Resour
of Contra Costa County, California. US Geological Survey Res 42:W06410. doi:10.1029/2005WR004369
Bulletin 1388, p 18 Ruff M, Czurda K (2008) Landslide susceptibility analysis with a
Oguchi T (1997) Drainage density and relative relief in humid steep heuristic approach in the Eastern Alps (Vorarlberg, Austria).
mountains with frequent slope failure. Earth Surf Process Land- Geomorphology 94:314–324. doi:10.1016/j.geomorph.2006.10.
forms 22:107–120. doi:10.1002/(SICI)1096-9837(199702)22:2 032
Okimura T (1989) Prediction of slope failure using the estimated Saulnier GM, Beven K, Obled C (1997) Including spatially variable
depth of the potential failure layer. J Nat Disast Sci 11(1):67–79 effective soil depths in TOPMODEL. J Hydrol 202:158–172
Oreskes N, Shrader-Frechette K, Belitz K (1994) Verification, Schlunegger F, Zeilinger G, Kounov A, Kober F, Hüsser B (2006)
validation, and confirmation of numerical models in the earth Scale of relief growth in the forearc of the Andes of Northern
sciences. Science 263(5147):641–646 Chile (Arica latitude, 18 S). Terra Nova 18:217–223. doi:10.
Ori GG, Roveri M, Vannoni F (1986) Plio-Pleistocene sedimentation 1111/j.1365-3121.2006.00682.x
in the Apenninic foredeep (Central Adriatic Sea, Italy). In: Allen Sciarra M (2016) Analisi della suscettività da frana nella fascia
PA, Homewood P (eds) Foreland Basins, IAS special publication pedemontana-collinare e costiera abruzzese. PhD thesis in
8. Blackwell, Oxford, pp 183–198 Engineering geology, University ‘‘G. d’Annunzio’’ Chieti-
Parotto M, Praturlon A (2004) The southern Apennine arc. In: Pescara, Italy, p 187
Geology of Italy. Special Volume of the Italian Geological Soeters R, Van Westen CJ (1996) Slope stability: recognition,
Society for the IGC 32 Florence, pp 53–58 analysis and zonation. In: Turner AK, Shuster RL (eds)
Patacca E, Scandone P (2007) Geology of the Southern Apennines. Landslides: investigation and mitigation. Transportation
Boll Soc Geol It 7:75–119 Research Board, Special Report 247, pp 129–177
Petschko H, Brenning A, Bell R, Goetz J, Glade T (2014) Assessing Summerfield MA (1997) Global geomorphology. Longman, New
the quality of landslide susceptibility maps—case study Lower York, p 537
Austria. Nat Hazards Earth Syst Sci 14:95–118. doi:10.5194/ Tromp-van Meerveld HJ, McDonnell JJ (2006) Threshold relations in
nhess-14-95-2014 subsurface stormflow: 2. The fill and spill hypothesis. Water
Piacentini D, Troiani F, Soldati M, Notarnicola C, Savelli D, Resour Res 42:W02411. doi:10.1029/2004WR003800
Schneiderbauer S, Strada C (2012) Statistical analysis for Van Westen CJ (1993) Application of geographic information
assessing shallow-landslide susceptibility in South Tyrol systems to landslide hazard zonation. PhD Dissertation, Tech-
(south-eastern Alps, Italy). Geomorphology 151–152:196–206. nical University Delft, ITC-Publication Number 15, ITC,
doi:10.1016/j.geomorph.2012.02.003 Enschede, The Netherlands
Piacentini T, Sciarra M, Miccadei E, Urbano T (2015) Geomorpho- Van Westen CJ (1994) GIS in landslides hazard zonation: a review
logical mapping of near-surface deposits: hillslope evolution of with examples from the Andes of Colombia. In: Price MF,
the Adriatic piedmont of the Central Apennines (Feltrino Stream Heywood DJ (eds) Mountain environments and geographic
basin and minor coastal basins, Abruzzo, Italy). information systems. Taylor and Francis, London, pp 135–167
J Maps 11(2):299–313. doi:10.1080/17445647.2014.949884 Van Westen CJ, Rengers N, Terlien MTJ, Soeters R (1997) Prediction
Piacentini T, Urbano T, Sciarra M, Schipani I, Miccadei E (2016) of the occurrence of slope instability phenomena through GIS-
Geomorphological mapping of the floodplain at the confluence based hazard zonation. Geol Rundsch 86:404–414
of the Aventino and Sangro rivers (Abruzzo, Central Italy). Van Westen CJ, Rengers N, Soeters R (2003) Use of geomorpho-
J Maps 12(3):443–461. doi:10.1080/17445647.2015.1036139 logical information in indirect landslide susceptibility assess-
Poli S, Sterlacchini S (2007) Landslide representation strategies in ment. Nat Hazards 30:399–419
susceptibility studies using weights-of evidence modeling tech- Van Westen CJ, Van Asch TWJ, Soeters R (2006) Landslide hazard
nique. Nat Resour Res 16:121–134. doi:10.1007/s11053-007- and risk zonation; why is it still so difficult? Bull Eng Geol
9043-8 Environ 65(2):167–184. doi:10.1007/s10064-005-0023-0
Pradhan AMS, Kim YT (2014) Relative effect method of landslide Van Westen CJ, Castellanos E, Kuriakose SL (2008) Spatial data for
susceptibility zonation in weathered granite soil: a case study in landslide susceptibility, hazard, and vulnerability assessment: an
Deokjeok-ri Creek, South Korea. Nat Hazards 72(2):1189–1217. overview. Eng Geol 102(3–4):112–131. doi:10.1016/j.enggeo.
doi:10.1007/s11069-014-1065-z 2008.03.010
Pradhan B, Lee S (2010) Landslide susceptibility assessment and Varnes DJ (1978) Slope movement types and processes. In: Schuster
factor effect analysis: backpropagation artificial neural networks RL, Krizek RJ (eds) Landslides, analysis and control, special
and their comparison with frequency ratio and bivariate logistic report 176: transportation research board. National Academy of
regression modelling. Environ Model Softw 25(6):747–759. Sciences, Washington, DC, pp 11–33
doi:10.1016/j.envsoft.2009.10.01 Varnes JD (1984) IAEG commission on landslides and other mass
Reichenbach P, Busca C, Mondini AC, Rossi M (2014) The influence movements, landslide hazard zonation: a review of principles
of land use change on landslide susceptibility zonation: the Briga and practice. The UNESCO Press, Paris, p 63
catchment test site (Messina, Italy). Environ Manage Von Ruette J, Papritz A, Lehmann P, Rickli C, Or D (2011) Spatial
54(6):1372–1384 statistical modeling of shallow landslides—validating predic-
Remondo J, Gonzalez A, Dıaz de Teran JR, Cendrero A, Fabbri A, tions for different landslide inventories and rainfall events.
Cheng CF (2003) Validation of landslide susceptibility maps: Geomorphology 133:11–22. doi:10.1016/j.geomorph.2011.06.
examples and applications from a case study in Northern Spain. 01
Nat Hazards 30:437–444. doi:10.1023/B:NHAZ.0000007201. Wilson RC (1989) Rainstorms, pore pressures, and debris flows: a
80743.fc theoretical framework. In: Morton DM, Sadler PM (eds)
Romeo RW, Mari M, Floris M, Pappafico G, Gori U (2011) Un Landslides in a semi-arid environment, 2nd edn. Publications
approccio per coniugare la suscettività spaziale e temporale da of the Inland Geological Society, California, pp 101–117

123
 M. Sciarra et al.

Wilson JP, Gallant JC (2000) Digital terrain analysis. In: Wilson JP, Yesilnacar E, Topal T (2005) Landslide susceptibility mapping: a
Gallant JC (eds) Terrain analysis: principles and applications. comparison of logistic regression and neural networks methods
Wiley, New York, pp 1–27 in a medium scale study, Hendek region (Turkey). Eng Geol
Yalcin A, Bulut F (2007) Landslide susceptibility mapping using GIS 79:251–266. doi:10.1016/j.enggeo.2005.02.002
and digital photogrammetric techniques: a case study from Yilmaz I (2010) Comparison of landslide susceptibility mapping
Ardesen (NE-Turkey). Nat Hazards 41:201–226. doi:10.1007/ methodologies for Koyulhisar, Turkey: conditional probability,
s11069-006-9030-0 logistic regression, artificial neural networks and support vector
Yalcin A, Reis S, Aydinoglu AC, Yomralioglu T (2011) A GIS-based machine. Environ Earth Sci 61(4):821–836. doi:10.1007/s12665-
comparative study of frequency ratio, analytical hierarchy 009-0394-9
process, bivariate statistics and logistics regression methods for
landslide susceptibility mapping in Trabzon, NE Turkey. Catena
85:274–287. doi:10.1016/j.catena.2011.01.014

123

View publication stats