Palenzuela2016 Article AssessingCriticalRainfallThres

Nat Hazards (2016) 84:185–212
DOI 10.1007/s11069-016-2416-8
ORIGINAL PAPER
Assessing critical rainfall thresholds for landslide

triggering by generating additional information
from a reduced database: an approach
with examples from the Betic Cordillera (Spain)
José Antonio Palenzuela1 • Jorge David Jiménez-Perálvarez1 •
José Chacón1 • Clemente Irigaray1
Received: 8 December 2015 / Accepted: 7 June 2016 / Published online: 16 June 2016
Springer Science+Business Media Dordrecht 2016
Abstract The denudation of young reliefs prone to landslides can have severe consequences
for society and the environment. However, landslide databases and the additional infor-
mation (landslide type, date and triggering factors) necessary to deal with landslide hazard
assessment and the development of effective and reliable landslide warning systems are
usually scarce or non-existent. In this way, by taking into account the date of landslide events
and by expanding the analysis of cumulative rainfall from these dates to a broader time
period that includes the days or months leading up to a landslide, the corresponding trig-
gering rainfall threshold can be assessed more accurately. In this paper, a methodology based
on a partial duration series analysis applied to rainfall variables allows the possibility to
better understand precipitation patterns. Another advantage of analysing precipitation
variables within a broader time period is the ability to identify greater accuracy rainfall
anomalies such as extreme rainfalls with their return period related to a low number of dated
landslide events (in this case, 20 landslide events). The landslide spatial distribution within a
regional area requires the processing and analysis of data from multiple long-term historical
daily rainfall records from different rainfall gauges, which notably increase the number of
calculations to be dealt with. To overcome this inconvenience, these processes were
streamlined by using macro-automation. Additionally, different rainfall durations can be
interactively identified from graphical outputs that show anomalies on more than one rainfall
variable after applying this methodology. Among these rainfall variables, the antecedent
& Jorge David Jiménez-Perálvarez

jorgejp@ugr.es
José Antonio Palenzuela
jpalbae@ugr.es
José Chacón
jchacon@ugr.es
Clemente Irigaray
clemente@ugr.es
1
Department of Civil Engineering, ETSICCP, University of Granada, Campus Fuentenueva s/n,
18071 Granada, Spain
123
186 Nat Hazards (2016) 84:185–212
accumulated rainfall (A1) was found to be the most suitable to apply the occurrence prob-
ability analysis. When compared to other variables, the return period values of A1 were
determined to be conservative, neither too high nor too low. Using this approach, the return
period curve was shown to be an important graphic object in detecting uncommon rainfalls
that are contemporaneous or previous to landslides. The relevant findings of this research
show a power-law trend with a = 88.005 and b = 0.69 in the correlation of intensity and
duration associated with antecedent cumulative rainfall (A1) anomalies. The mean return
period for these anomalies resulted in 12.4 years, while for 50 % of the landslides, the
recurrence interval was estimated in less than or equal to 3.6 years. In addition, significant
differences were found between catalogued slope-cut failures and natural landslides.
Moreover, differences were also found between simplified types of natural landslides.
Keywords Return period Partial duration series Rainfall threshold Landslide Betic
Cordillera
1 Introduction
Landslides or slope mass movements occur frequently and can lead to the eventual trig-
gering of natural catastrophes. In most cases, landslides occur as a secondary effect of
earthquakes and storms (Varnes 1984; Panizza 1996; Aleotti and Chowdhury 1999;
Guzzetti et al. 1999; Chacón 2003; Lacasse and Nadim 2009). The cumulative loss caused
by landslides in terms of deaths, damage to public infrastructure and private property, as
well as the disruption of social and environmental systems is significant and widespread
compared to other geological hazards (Varnes 1984; Guzzetti et al. 1999; Cardinali et al.
2002). The compilation of the IAEG Commission on Landslides reported 2660 casualties
due to landslides from 1971 to 1975 worldwide, 14 % of the total deaths due to natural
catastrophes (UNESCO 1973–1979). More recently, Petley (2012) compiled a global
spatial–temporal database of fatalities caused by landslides. A total of 2620 landslides and
32322 deaths were recorded by Petley (2012) between 2004 and 2010. The largest number
of human fatalities due to this phenomenon was found to be along the Himalayan Arc and
in China. In Europe, a historical compilation of major landslides occurred from 1998 to
2009 can be found in Spizzichino et al. (2010) and Spizzichino et al. (2013). During this
period, 312 people were killed in addition to infrastructure and private property damage.
The authors conceded that in this study there were major shortcomings in their findings due
to a lack of data, variable resolution and different information levels found within Euro-
pean databases. Schuster (1978) established the annual cost of landslides in the USA to
exceed 1000 million dollars. In Spain, Ayala et al. (1987) estimated the potential risk
related to landslides for the period 1986–2016 which was later updated by Suárez and
Regueiro (1997). This potential risk was then established in 9000 million euros in damages
when considering a medium risk scenario and 10800 million euros in the case of a max-
imum risk situation. In order to prevent or mitigate damages from landslides, the hazard of
a potential event must be assessed. The term hazard is used to express the probability of
occurrence of a natural phenomenon event within a specified period of time and within a
given area (Varnes 1984). Given the importance of landslide databases and the difficulties
involved in the gathering of landslide attributes, researchers have endeavoured to establish
international approaches to improve landslide knowledge and risk management within an
123
Nat Hazards (2016) 84:185–212 187
international framework (Casale et al. 1994; Dikau et al. 1996; Panizza 1996; Chacón et al.
2006a; European Community 2007; Van Den Eeckhaut and Hervás 2012). Nevertheless,
these studies show a significant lack of historical information that is valuable when dealing
with quantitative hazard assessment, in particular, when dating landslide initiation and
reactivation. This is partly due to the well-known difficulties of gathering data. Moreover,
once a landslide has occurred, the focus of official institution data gathering shifts to the
assessment of damage and the total cost to a community rather than the cause–effect
relationship of a phenomenon such as a landslide (Aleotti and Chowdhury 1999; Guzzetti
et al. 1999; Van Den Eeckhaut and Hervás 2012). Special consideration should be given to
those regions where the Alpine orogeny has uplifted younger and higher reliefs. An
example of this can be seen in the Betic Cordillera in south-east Spain, which shows high
rates of mass wasting due to landslides (Chacón 2003; Fernández et al. 2003; Irigaray et al.
2007; El Hamdouni et al. 2008; Jiménez-Perálvarez et al. 2011; Chacón et al. 2012). The
assessment of the direct frequency of landslides is severely hampered by the inexistence of
a detailed catalogue that includes event dates. In these circumstances, indirect frequency
analysis or intermediate hazard assessment (Corominas and Moya 2008; Fell et al. 2008)
via linking rainfall thresholds with landslides events could reveal new and relevant
information that would be useful for land-use planning and risk alert systems.
In rainfall-induced landslides, threshold refers to the rainfall, soil moisture and the
hydrological conditions that, when reached or exceeded, are likely to trigger landslides
(Endo 1969; White et al. 1996; Reichenbach et al. 1998; Corominas 2000; Aleotti 2004;
Wieczorek and Glade 2005; Guzzetti et al. 2007; Guzzetti et al. 2008; Vennari et al. 2014).
The analysis of rainfall thresholds has been applied worldwide (Caine 1980; Innes 1983;
Jibson 1989; Crosta and Frattini 2001; Cannon and Gartner 2005; Guzzetti et al. 2008), by
using physically based numerical (process-based models and/or conceptual models) (De vita
et al. 2013; Papa et al. 2013) and empirical methods (Aleotti 2004; Guzzetti et al. 2007;
Guzzetti et al. 2008; Vennari et al. 2014). Physically based numerical models account
explicitly for dynamic factors such as soil saturation or cohesion due to the presence of root
and/or to partial saturation, which contribute to landslide trigger (Crosta and Frattini 2003).
In order to successfully identify the dynamic factors involved in landslide triggering, more
detailed information is needed. Unfortunately, landslide data documentation is rarely
available, which prevents the successful application of physically based numerical models
(Aleotti 2004). However, some examples can be found in the literature (Borga et al. 1998;
Aleotti et al. 2003; Frattini et al. 2004; Li et al. 2011; Ma et al. 2014). The empirical methods
used within the field of landslide research can be divided into two separate categories. One
subcategory is directly based on rainfall that is likely to trigger a landslide. The second
subcategory combines antecedent conditions such as groundwater level, soil moisture and
the antecedent rainfall as an empirical measurement of soil moisture. The parameter of
antecedent rainfall is used to substitute the measurement of soil moisture when it is not
available (Terlien 1998; Guzzetti et al. 2007; Guzzetti et al. 2008). Examples using ante-
cedent conditions can be found in numerous studies: Wieczorek and Glade (2005), Lumb
(1975), Crozier and Eyles (1980), Crozier (1986), Kim et al. (1991), Terlien (1996), Glade
et al. (2000). These empirical methods can be applied to areas with minimum to null
information on geotechnical and hydrogeological parameters associated with the historical
landslide occurrence. Thus, rainfall variables (i.e. accumulated rainfall, duration, intensity,
and other) are analysed in order to establish critical rainfall threshold (CRT) curves. Once
CRTs are determined, the occurrence probability of these thresholds (i.e. frequency or return
period) is assigned to areas with a potential for landslide event (Corominas and Moya 2008).
A review of CRT curves is summarized in Guzzetti et al. (2007) and Guzzetti et al. (2008).
123
188 Nat Hazards (2016) 84:185–212
Despite the fact that the reliability of the CRTs and the probability of the occurrence of
landslides greatly depend on the quality and the comprehensive nature of the database in
question (Glade et al. 2000), these models are more feasible and widely used (Borga et al.
2002; D’Odorico and Fagherazzi 2003; Guzzetti et al. 2004; Dykes et al. 2008; Hromadka
and Phillips 2010). Most literature related to rainfall-triggered landslides focus on shallow
landslides as they are initiated directly by primary climatic influences such as short and
intense rainstorms, whereas deep-seated landslides depend on the long-term (several years)
variation of annual rainfall (Bonnard and Noverraz 2001; Aleotti 2004; Wieczorek and Glade
2005). A review of methods that can be found in Guzzetti et al. (2007) links 124 empirical
thresholds to landslide events that have taken place throughout the world.
The research presented in this paper focuses on the critical rainfall threshold estimation
linked to a small set of dated landslides (20 landslide events) and their probability of
occurrence. Given the focus on the relationship between the specific set of landslides
researched in this study with CRTs, a new methodology was developed to generate a high
number of rainfall series for a wide range of duration (1–90 days) and 12 rainfall gauges.
The main processes of this methodology, such as the calculation of a large amount of data
related to accumulated rainfalls and return periods, are performed and streamlined by using
VBA macros. When plotting solely rainfall variables (daily or accumulated rainfall),
uncommon peaks linked to landslide dates are not readily distinguishable from the normal
rainfall patterns. Nonetheless, when return periods for every accumulated rainfall–duration
combination are graphically represented together with rainfall variables, the anomalies of
the triggering factor (rainfall) associated with landslides can be identified more accurately.
Consequently, by applying the new data obtained from the return periods and accumulated
rainfalls, a retrospective analysis of rainfall anomalies and their occurrence probability
based on the triggering of past landslides is possible. This methodology has been applied
and discussed in a regional area (2370 km2) at the Betic Cordillera (southern Spain).
2 Setting of the study area
The target area extends from the S-SW slopes of the Sierra Nevada in the Betic Cordillera
near the Mediterranean coastline in southern Spain (Fig. 1). This stretch of land covers a
regional area (2370 km2) with different heights (*5–1450 m). In this geographical area, the
climate is predominantly Mediterranean—Cs—(Köeppen 1936), alternating from semi-arid
in dry summers (May–September) to sub-humid during the rainy season (October–April)
(Ruiz Sinoga and Martinez Murillo 2009; Jiménez-Perálvarez et al. 2011). Temperatures
average 16 C, but they are lower in the highest mountains. Nevertheless, the highest peaks
in temperature are reached during the drought period, often exceeding 40 C. The relative
humidity is usually quite low, not exceeding 30 %. The majority of rainfall occurs between
the months of October and April, when the yearly recorded precipitation averages 650 mm,
although the total precipitation can be unevenly distributed one year to another (Trujillo
1995; Schutt 2005). The maximum rainfall occurs in spring or autumn (Trujillo 1995).
Spatially, in the south and south-east adjacent to the Sierra Nevada, there exists a transition
from the cold steppe climate—BSk—(Köeppen 1936) zone at the base of the mountains
which is followed by a hot steppe climate—BSh—(Köeppen 1936) zone found on the coastal
plain (Schutt 2005).
The main meteorological element controlling the weather over the region is the Azores
anticyclone. In winter the Azores anticyclone high-pressure band is located between 40
123
Nat Hazards (2016) 84:185–212 189
Fig. 1 Global location of the study area and geology settings with projections of landslides events and
pluviometric stations
and 30N. During this time, the Polar Front together with the Iceland Low creates the
atmospheric conditions for the majority of rainfall in the region. In contrast, during the
summer, the Azores anticyclone central axis enters the area between 35 and 45N, directly
affecting the study area and causing a warm-dry environment, which is typical of
Mediterranean summers. Sometimes, this anticyclone moves between 45 and 55N,
exposing the research area to Polar Front and Subtropical Jet cloud systems, which in turn
causes stormy weather. This high relief of the Betic Cordillera causes an altitudinal gra-
dient which creates cold temperatures and a rainfall gradient. The existent relief and its
orientation control the appearance of convective cells and rain shadow areas (Trujillo
1995; Schutt 2005). In general, precipitation in the research area shows an east–west
increasing trend due to significant influence from cloud systems coming from the Atlantic
Ocean. Additionally, the altitude within the study area also contributes to an increase in
precipitation, reaching maximum values (*1500 mm) at the highest peaks. This altitude
effect causes the local climate to change. This is reflected typically in Mediterranean
climate characteristics (high temperatures, erratic rainfall and strong insolation) with a
deepening in continental climate conditions as altitude increases. This gradation that is
influenced by altitude favours a drop, the occurrence of frost and an increase in rainfall. It
is worth highlighting two historical annual rainfall records occurred in the study area. The
first took place in the 1996–1997 winter. The second one occurred during the 2009–2010
123
190 Nat Hazards (2016) 84:185–212
wet season, when the rainfall reached more than double the average cumulative precipi-
tation for a typical hydrological year in Spain (AEMET 2010). These extreme meteoro-
logical conditions triggered numerous floods, landslides and heavy erosion, resulting in
significant damage and negative social consequences (Irigaray et al. 2000; Palenzuela et al.
2013; Palenzuela et al. 2014).
Landslides of variable size and typology are widespread throughout the study area. This
geographical area is characterized by a rugged relief with overexcavated river channels,
which in turn provide the preconditioning factors that have the potential to make slopes
unstable (Thornes and Alcántara-Ayala 1998; El Hamdouni 2001; El Hamdouni et al.
2008; Jiménez-Perálvarez et al. 2011). These topographical features resulted from Qua-
ternary geomorphological changes linked to active tectonic processes (Keller et al. 1996).
The major determinant factors involved in landsliding are hydrology, topography and
lithology (including soil cover and vegetation) together with land use (Castillo et al. 1996;
Thornes and Alcántara-Ayala 1998; Jiménez-Perálvarez et al. 2009, 2011).
The study area is geologically located in the Internal Zones of the Betic Cordillera,
which largely comprises metamorphic rocks at both the Nevado-Filabride and the Alpu-
jarride complexes. The area in question covers the steepest part of S-SW corner of Sierra
Nevada, which is criss-crossed by deep valleys. Post-tectonic Neogene and Quaternary
deposits are located in the topographically lowest part with gentle slopes (Fig. 1), com-
prising sediment such as marls and silts covered by conglomerates. The Nevado-Filabride
Complex lithology is composed of dark schists and feldspathic mica schists, while the
Alpujarride Complex is composed of Triassic calc-schists together with marbles, phyllites
and quartzites (Gómez-Pugnaire et al. 2004). The most superficial layer includes recent
colluvium, which is deposited at the bottom of slopes. Additionally, found in the most
superficial layer is alluvial fill which is transported throughout the watershed.
This lithology, combined with the study area geomorphometry, is associated with dif-
ferent types of landslide (Chacón et al. 2006a, b, 2010). Because of this, knowledge about
the characteristics, distribution and mechanisms related to this phenomenon are of high
interest for the research community and society. Thus, earlier investigations were carried
out for several decades, which were generally focused on spatial landslide inventories
throughout several parts of the study areas (Fernández et al. 2003, 2009; Jiménez et al.
2005; Irigaray et al. 2007; El Hamdouni et al. 2008; Jiménez-Perálvarez et al. 2011). These
landslide inventories were built up by using photointerpretation techniques backed up by
field campaigns. The major features included within the databases are landslide size and
type, its development stage and affected lithology. Research concerned with landslide
susceptibility zoning has been broadly developed by taking into account determinant
factors (lithology, slope gradient, slope aspect and other) and applying the GIS Matrix
Methodology (GMM) (Irigaray et al. 1999, 2007; Fernández et al. 2003), which was
recently optimized (Jiménez-Perálvarez et al. 2011). Landslide susceptibility results and
maps were then made publicly available by official institutions (Diputación de Granada
and the Instituto Geológico Minero de España 2007). By limiting the temporal range,
recent works have dealt with the assessment and monitoring of landslide yearly dis-
placements by applying the TLS (terrestrial laser scanning) technique (Irigaray and
Palenzuela 2013; Palenzuela et al. 2013). Additionally, an ALS (aerial laser scanning)-
based methodology was applied to map the spatial distribution of a seasonal inventory
(Palenzuela et al. 2014). In turn, a DInSAR-based methodology was applied to the
assessment of very low displacements in some zones within the study area (Fernández et al.
2009) during a seven-year period (1993–2000). Moreover, Fernández et al. (2011) analysed
the evolution of a single landslide by developing a photogrammetry-based approach, which
123
Nat Hazards (2016) 84:185–212 191
was later applied to two analogical imagery datasets (1956 and 1992) and one digital
imagery dataset (2008). The sequential application of these approaches is valuable in its
contribution to create more comprehensive and accurate geodatabases, enabling a spatial
yearly average estimation for landslide frequency. Finally, the results of these investiga-
tions were combined and studied in conjunction with recent findings from a dendrogeo-
morphological study, as well as the recurrence of critical rainy periods (i.e. associated with
landslides and flooding resulting in damages within different parts of the study area)
(Jiménez-Perálvarez 2012). The integration of these studies and the results obtained related
to the recurrence of rainy periods triggering landslides, enabled the establishment of
preliminary annual rainfall thresholds. This was carried out by applying a descriptive
statistical analysis based on total annual rainfall. Accordingly, water years could be
classified and associated with the occurrence of shallow and deep landslides. As a result, a
5-year return period was linked to shallow landslides, whereas an 18-year return period
was estimated for the occurrence of both deep and shallow landslides (Jiménez-Perálvarez
2012). Nonetheless, no one of these research works provide accurate information about
landslide history. Specifically, this includes reliable data on the exact landslide occurrence
time or date and hydrological/hydrogeological pattern information beginning with the
antecedent period until the landslide in question takes place.
3 Materials and methods
3.1 Gathering necessary information
To deal with the establishment of CRTs for landslide and their occurrence probability, the
methodology presented in this paper starts with a review of sources where information on
landslide events and rainfall variables can be found.
Because of the absence of a landslide catalogue or database, the present work required first
the search for landslide event information within the study area. Unfortunately, this was
constrained by well-known difficulties (Ibsen and Brunsden 1996). Most notably, there was a
lack of available information about landslides that could be found at official institutions.
Moreover, accessing landslide event information was further complicated by the bureaucratic
procedures needed to extract public data from specific projects and reports that could have
made a valuable contribution to this work. To make matters worse, the information provided
from the selected sources was usually incomplete and typically more concerned with the
damage and social disruption caused by a landslide event. Among the available information
sources that proved to be of value for the research of this study was the regional newspaper
IDEAL which in turn provided the majority of data related to landslides. In order to locate
landslide data in the newspaper, firstly, the pages related to periods of significant precipitation
were found and reviewed. Later, the search was expanded to the remaining database in the
newspaper. After reviewing the data, it was concluded that the greatest number of landslide
events was recorded within important rainfall periods when civil infrastructure and other
entities were damaged by this phenomenon, and in this way aroused enough social interest
and ultimately written about in the newspaper. The first issue of the IDEAL newspaper dates
back to 8 May 1932 (IDEAL 2014). Most of the rainfall measurements of the study area began
to be recorded in the mid-1940s. Thus, a period of about 60 years (1940s to 2010) was
considered in the search of landslides events and their associated rainfall variables. The
search of landslide events was carried out by reviewing newspaper headlines and bylines, as
123
192 Nat Hazards (2016) 84:185–212
well as articles, published on dates occurring close to those during the selected wet seasons.
The research of data related to landslide events was later extended to the remaining database,
taking into account that some landslide processes (infiltration and rise in pore-water pressure)
related to major cause of a landslide can be delayed. Once the information was located in the
documentation, the most important pages were printed (from the publisher scanned archive)
or saved (from the digital newspaper library beginning in January 2006). Subsequently, these
pages were reviewed in depth to isolate landslide events and link different news coverage of
the same event. From these revisions, a preliminary database containing 166 newspaper
articles related to landslides and some other coalescent phenomena (e.g. important flooding)
was compiled and recorded. In addition, other events from earlier research studies carried out
on the S-SW region of the Sierra Nevada (Irigaray et al. 2000) were added to the database.
After compiling the database, only a total of 20 events were kept for use in this study and
analysed as they contained the most accurate information about date and location (Table 1;
Fig. 1). This group of landslides was found within two hydrological periods: 1996–1997 and
2009–2010. Other information such as size, type or variables related to the rainfall that trigged
these landslides was difficult to get or even inexistent. Moreover, the non-technical language
used by witnesses and journalists was frequently found to be more qualitative and charac-
terized by ambiguous terms (e.g. collapse, rockslide, ground movement and ground in
motion). Nevertheless, the types of affected landscapes (slope-cut or man-made slopes, and
natural slopes) and simplified types of landslides (slide, complex or rockfalls) could be
interpreted from the literal description and their attached photographs. When this interpre-
tation was not possible, the landscape or landslide type was assigned to the category of
‘‘unknown’’. In addition, the date could also be recorded in the database when certain
information was located in the newspaper articles (yesterday a ground movement, the past
‘‘date’’ a rockslide […]). The exact location of landslide events within the affected area for
12 cases could accurately be pinpointed from photographs and the geographical explorer
Google Earth. Other locations were approximately determined by applying a radius that take
into account a wider area that coincided with the description or images of the location. The
point coordinates for every landslide were subsequently entered and displayed within a vector
layer of the geographical information system ArcGIS (ESRI 2012), while events affecting a
significant part of a stretch of road were projected as polylines. For the assignment of the
lithology involved in landslides, the geological map in Fig. 1 was created and extended to
display the total study area. This was carried out by integrating and simplifying the digital
layers of 9 original maps at scale 1:5000 in the GIS. In the same way, via ArcGIS, the research
area topography was modelled by using the LiDAR (Light Detection and Ranging or Laser
Imaging Detection and Ranging) databases (5 m 9 5 m cell size) from the National Aerial
Orthophotography Plan (IGN 2008–2012). This digital model enabled the spatial analysis of
the slope gradient added in Table 1.
Several factors are involved in the rainfall of a landslide-affected area: in general,
rainfall gauge–slope movement distance, differences in elevation and/or orientation, and
conditions affecting the gauge position (wind, rain barriers, and other) (Aleotti 2004,
Brunetti et al. 2013). Of all these factors, gauge–landslide distance has a great influence on
the differences in precipitation that can exist between landslide site rainfall and the
recorded values at the rainfall station (Aleotti 2004). In this study, the selection of multiple
rain measurement stations was simplified. This was done by assuming that a measuring
station is more representative of the conditions in the landslide area if both, the station and
the landslide, are located within an area with similar rainfall characteristics. To test this
condition, the average annual rainfall obtained from the database Environmental Infor-
mation Network of the Regional (REDIAM 2016) was classified into 3 classes by the
123
Nat Hazards (2016) 84:185–212 193
Table 1 Landslide event reference or ID, closest station reference and date
ID Pluviometric Event date Slope Mean Lithology Landslide type Height
gauge ID type slope () (m)
31 183 03/02/1996 S 23.53 CR, SC, P RF 779

39 225 31/01/1997 X 24.12 P SL 60
42 225 25/12/1996 S 27.43 SC SL 131
44 100 04/01/1997 N 27.87 SC SL 358
45 153 06/01/1997 N 34.03 MS COM (SL-RF) 881
45 153 30/01/2010 N 34.03 MS COM (SL-RF) 906
49 8 02/01/1997 N 17.31 NQ SL 303
50 22 08/01/1997 X 23.46 SC COM (SL-RF) 973
60 225 27/01/1997 X 27.43 SC COM (SL-RF) 973
72 220 06/02/2009 X 37.18 SC COM (SL-RF) 1215
74 154 30/01/2009 N 31.47 SC SL 33
78 22 26/12/2009 S 35.56 SC RF 577
89 447 07/01/2010 N 28.08 CR, SC RF 955
94 141 05/03/2009 N 29.29 CR COM (SL-RF) 370
101 102 20/12/1996 X 21.38 CR, SC, P RF 335
102 141 20/12/1996 X 25.36 CR, P RF 370
110 392 12/11/1996 X 14.79 NQ COM (SL-RF) 370
114 153 12/11/1996 X 25.19 MC RF 328
153 225 01/01/1997 S 23.82 SC, P, CR COM (SL-RF) 559
155 225 08/01/1997 S 27.43 SC RF 317
Slope type N (natural), S (slope-cut movement), X (several or unknown types)

Lithology CR (calcareous rocks—limestone, dolomite, marble and calc-schist), P (phyllite), SC (schists), MS
(mica schists), NQ (Neogene and Quaternary undifferentiated detritus)
Simplified landslide type: RF (rockfalls), SL (slide), COM (complex)
natural breaks method (Jenks 1967) (Fig. 2). If several stations were found to be within the
same annual rainfall class as that of the corresponding landslide, the station with the
longest landslide historical record was finally selected to carry out a more accurate fre-
quency analysis (Sect. 3.2.2). This step was carried out iteratively by applying a buffer
operation that changes the search radius to those stations closest to each catalogued
landslide event. After every iteration, the number of stations found closest to each cata-
logued landslide event was checked to find out whether they fell within the same plu-
viometric area and had therefore similar rainfall characteristics or not. According to the
spatial distribution of the rainfall–gauge stations, it was found that a 3 km radius enabled
finding at least one station close to each landslide event. Applying this method, only 12
gauge stations series were selected from a total of 64 that were provided by the Regional
Water Agency Environmental Information Network and the National Meteorological
Institute (Agencia de Medio Ambiente y Agua 2014; AEMET 2014) (Table 1).
3.2 Rainfall thresholds and frequency analysis
When dealing with the indirect hazard assessment through the evaluation of CRTs trig-
gering landslides, two major questions appear:
123
194 Nat Hazards (2016) 84:185–212
Fig. 2 Spatial classification of the average annual rainfall classification
1. What is the value or values of some rainfall variable or a combination of variables that
would most likely trigger landslides? (An experimental threshold based on rainfall
variables).
2. And, what is the occurrence probability of those CRTs in the same geological,
geomorphic and climate conditions? (The return period).
To answer the first question, it is essential to analyse the rainfall series; to find out how
much duration (D), cumulative rainfall and intensity (I) are necessary to trigger any
landslide. Accordingly, the cumulative rainfall must be evaluated taking into account a
wide range of durations instead of a short set of duration values. This range should be
wider for deep-seated landslides than for the shallow ones. Later it is possible to review,
analyse and interpret the information generated from different rainfall variables. This
ultimately enables the identification of CRTs as rainfall anomalies with peculiar charac-
teristics, i.e. noticeable larger return periods, higher intensity or accumulated rainfall.
Once the thresholds are found, the second question can be answered by applying a
frequency analysis based on past registered rainfall in which these thresholds had been
reached or exceeded.
To assess the values of rainfall variables and identify the CRTs that would most likely
trigger a landslide, as well as their occurrence probability, the following steps were carried
out via the application of the present methodology:
123
Nat Hazards (2016) 84:185–212 195
3.2.1 Generating rainfall variables
In the analysis of critical rainfall thresholds triggering landslides, common variables are
frequently used. The majority of these variables can be found in Guzzetti et al. (2007) and
Guzzetti et al. (2008).
Taking into account the double action of precipitation, both as conditioning and
triggering factor, first the cumulative event rainfall (E) must be calculated by adding up
the total daily rainfall from the initial date (t0) of a given period of time until the
landslide date (LD). Alternatively, by setting the cumulative rainfall ending date prior to
a landslide date, the preconditioning or antecedent rainfall (A) of the landslide triggering
can be assessed. To assess the observed frequency of the cumulative rainfall, it is nec-
essary to obtain all the possible values of cumulative rainfall leading up to each date
found in the daily rainfall log. In addition, when applying this methodology, it is
essential to consider the different durations (D) ranging from 1 to 90 days when cal-
culating the possible values of cumulative rainfall. As a result, the number of repetitive
calculations (adding daily rainfall data) increases in direct proportion to the number of
rainfall gauges (12), the length of the rainfall record and the range of duration taken into
account (90 days). Thus, generating the new time series that contains the cumulative
rainfall values would have involved a tedious and time-consuming task. Given the
prospect having to deal with numerous calculations, these time series values were
automatically calculated and saved in different worksheets by the creation of a VBA-
macro. After obtaining these time series, the information related to rainfall was expan-
ded. Using the new information as a starting point enabled the assessment of the
observed return periods associated with the millions of cumulative rainfall values, thus
allowing the detection of rainfall anomalies with greater precision.
3.2.2 Frequency analysis and return period
In river-flooding studies, plot positions (observed frequencies) are usually obtained,

and then, the recurrence intervals for specific quartiles are predicted by fitting distri-
bution curves. By using these same curves, the quartiles can also be determined for
specific return periods. Similarly, in hydrological reports, only several values of
monthly or annual peaks are assessed. When considering only landslide dates, the
isolated rainfall peaks or anomalies triggering these events are not directly apparent.
This is due to the fact that landslides can occur after an elapsed period of one or more
days of cumulative rainfall. To uncover these anomalies or periods of atypical rainfall
leading up to or previous to a landslide date, it is necessary to visualize the continuous
spectrum of return periods for each value of cumulative rainfall. To deal with this issue,
the frequency for each cumulative rainfall value, as well as its observed return period,
has to be assessed. This part of the methodology also requires another cumbersome
process which involved coding another macro. After the new time series were com-
pleted, the cumulative rainfall values for the entire range of durations (from 1 to
90 days) could be plotted together with their corresponding return periods (T). In this
way, the anomalies of the cumulative rainfall were better graphically highlighted and
visualized. After these anomalies were identified, the quantities of cumulative rainfall,
as well as the recurrence probability (return period), were extracted from the corre-
sponding time series. The general calculations of the frequency analysis are explained
in the following steps:
123
196 Nat Hazards (2016) 84:185–212
• Asses the experimental cumulative distribution function (CDF) of the cumulative

rainfall for each duration (D) and every rainfall gauge by applying the widely
accepted ranking method of Weibull (Weibull 1939; Gumbel 1968; Barnett 1975;
Helsel and Hirsch 2002; Duggal and Soni 2005; Makkonen and Pajari 2014) (Eq. 1).
This leads to a large number of daily time series (90 durations multiplied by 12
gauges in total) for every cumulative rainfall event, with their corresponding daily
frequency expressed in terms of cumulative probability (FD) or exceedance
probability (FEX). The former stands for the probability of achieving a value lower
or equal to x belonging to a variable X (here the cumulative rainfall), and the latter
represents the probability that the value x be exceeded (i.e. FEX = 1 – FDG). The use
of partial duration series (PDS) (Cunnane 1973) is considered here more suitable for
landsliding hazard assessment. Applying a PDS-based frequency analysis is
appropriate when there may be more than one rainfall peak, or possibly a longer
period of rainfall which enables the necessary infiltration to trigger or reactivate a
failure by increasing the pore-water pressure (Iverson 2000). Thus, a number of
possibilities greater than the series length in years, months or a number of maximum
peaks can be analysed unlike common flood frequency analysis where annual
maximum series (AMS) or maximum annual flows (MAF) are used (Cunnane 1973;
Mkhandi et al. 2005; Mohssen 2009). In this particular case, the PDSs considered will
contain every daily cumulative rainfall values, enabling the automatic search of any
repetition of X B x throughout every gauge record.
i
FD ¼ ð1Þ
N þ1
where i represents the data ranking position and N is the total number of rainfall records
used from rainfall series.
• The previous PDS daily values are converted into annual values of frequencies and
return periods by introducing the occurrence rate, k (Cunnane 1973; Mohssen 2009)
(Eq. 2).
FA ¼ FD k ð2Þ
FA represents the annual cumulative frequency.
The occurrence rate (Eq. 3) converts daily frequency values to annual frequency val-
ues, with N representing the occurring possibilities (‘‘repetitions or realizations of the
experiment’’) and I representing the observed occurrences of a random phenomenon
(Duggal and Soni 2005). Then, by applying the coefficient k to FD, the annual fre-
quency can increase, remain equal or decrease in relation to the daily frequency values,
depending on the resulting rate (k [ 1, k = 1 or k \ 1, respectively). The k coefficient
is directly related to the number of records (N) used in the analysis with respect to the
series length in years—Y—(Eq. 3).
i N 0 days þ 1 N 0 ocurrences of X x N þ 1 i i
FA ¼ k ¼ 0
0
¼ ¼
Nþ1 N years N days þ 1 Y N þ 1 Y ð3Þ
• Finally, the return period (T) together with its confidence limits is evaluated for every
variable value. The return period is inversely proportional to the annual exceedance
frequency (Eq. 4).
123
Nat Hazards (2016) 84:185–212 197
Table 2 Example of the basic tabulated daily PDS frequency analysis results transformed to annual series
values
Data series length Ranked No. Occurrences FD (Eq. 1) FA (Eq. 2) FAEX (Eq. 4) T (y)
x [X B x]
N days Y years mm Rank, i
24,195 66.21 2 1 0.00004 0.0151 365.4531 0.002736

k 5 2 0.00008 0.0302 365.4380 0.002736
365.47 … … … … … …
250 24,170 0.99893 365.0755 0.3927 2.546365
408 24,187 0.99963 365.3323 0.1359 7.356164
520 24,195 0.99996 365.4531 0.0151 66.21
FAEX ¼ k ð1 FD Þ ð4Þ
The confidence limits (TL and TU) represent the error at the 90 % confidence level by
using and implementing the Oosterbaan (1988) method in the worksheets. This method
is based on binomial distribution and was developed for its application in distribution-
fitting software (CumFreq 2014).
In Table 2, an example of a basic frequency analysis calculation is shown. From this
example, the logical and common results of a PDS analysis can be appreciated. The
greater the x-value, the greater the rank and its cumulative frequency. Additionally, it
can be also observed how T increases with the rank, leading to the maximum for the
highest x-value. This occurs when T becomes equal to Y (the series length in years).
Therefore, the lowest x-values (*0 mm of cumulative rainfall) easily reach a high
annual exceedance frequency (FAEX * 365) as they occur almost every day. In con-
trast, the maximum x-value is only observed one time through the entire time series
(66.21 years).
Once all the multiple rainfall variable combination values (E–D and A–D) and their
return periods (T) are obtained, these values can then be found and used in the next
steps.
3.3 Determining rainfall thresholds for landslide triggering through

the selection of uncommon cumulative rainfall periods
At this point, each of the 90 time series corresponding to each duration (from the landslide
date to the 89th day before) contains the rainfall variables and frequency values similar to
those in Table 2. These completed time series enable the automatic search of rainfall
variable values and frequency analysis results linked to every landslide event date. This is
accomplished by running a new macro. This process is repeated for every landslide saving
the results in a separate worksheet. The found values are tabulated and then plotted in a
graph similar to those shown in Figs. 2 and 3. In these graphs, the x-axis represents the
rainfall duration (D) in days (d), which ranges from the landslide date (LD, with D = 1) to
the 89th previous day (D = 90) to the landslide event. The left vertical axis represents the
rainfall (daily and accumulated rainfall) in millimetres (mm), while the right vertical axis
represents the return period (T) in years (y). In addition, the TL and TU limits for the T-
curve are also plotted.
123
198 Nat Hazards (2016) 84:185–212
Fig. 3 Graph results for the landslide event 49 and the pluviometric station 8, showing the three markers
(SL, EA and SA) that divide the graph taking into account the existing anomalies in rainfall
The 20 graphs (for the 20 catalogued landslides) were interactively supervised in order
to divide the graphical rainfall variable information by using scrollbars to set three
markers. It is important to note the visualization of T-changes together with the variation in
accumulated rainfall, and the single peaks in the daily rainfall, makes it easier to identify
anomalies previous or contemporaneous to landslide occurrence. In Fig. 3, the cumulative
rainfall for D, ranging between 17 and 26 days, produces a significant anomaly in
T compared to the general trend. In this process, the start of antecedent rainfall (SA) is set
at the duration where a daily rainfall peak or change in the accumulated rainfall matches
the first significantly differentiated change (slope) or peak in the T-values when reading the
graph curve going towards the origin (0, 0). On the contrary, the end of the antecedent
rainfall (EA) is set at the point from which T reaches the common trend heading towards
the origin. The last storm initiation (SL) is set at the first notably differentiated T-peak, T-
change or daily rainfall within the last 5 days including the landslide date, taking into
account that shallower landslides can be triggered for a simple short duration storm. If not
rainfall is registered during these days, this point is set at LD.
The setting of markers (SA, EA and EA) produces the segments E1, E2 and E3, each
covering a set of days that includes LD. In addition, they also limit the antecedent rainfall
segments A1 and A2 (Fig. 3). After these segments are limited, and the accumulated
rainfall is taken into account for each of these segments, then their intensity (I) and return
periods (T, or mean T in case of A) are obtained by running a last coded macro. Thus,
instead of constraining the establishment of CRTs to a preselected set of durations, more
than one representative segment of recorded rainfall can be analysed and compared.
Given the characteristics of the graph results, three basic types of graphs were found.
The first case is shown in Fig. 3, where an area of uncommon values in the accumulated
rainfall appears between durations 17 to 26, whereas the last daily rainfall peaks include
the landslide date (D = 1 and D = 2). Different cases are shown in Fig. 4a–b. Figure 4a
only shows a peak with higher T-value, while Fig. 4b shows two major anomalies during
123
Nat Hazards (2016) 84:185–212 199
Fig. 4 Different types of rainfall patterns leading up to LD. In a only an unusual storm was detected during
the maximum 90-day period, which coincided with the start of the last storm (SL). In b the two markers SA
and EA have been set to contain two unusual antecedent rainfalls, while SL is set at the start of the last storm
which occurred after those anomalies
the antecedent rainfall, and one last daily rainfall peak (D = 3) occurs close to the land-
slide date.
4 Results
By applying the present methodology, the occurrence probability for critical rainfall
thresholds (CRTs) triggering landslides was assessed. In this approach, the recorded
rainfall was divided into 5 representative segments (Fig. 3) that were related to the
landslide event date. Later, the accumulated rainfall values for these segments, the duration
(D), the intensity (I) and the return periods (T), were tabulated. Then, the tabulated variable
123
200 Nat Hazards (2016) 84:185–212
Table 3 Landslides associated rainfall variables (D, I) related to E1, and T with its upper and lower
confidence limits (at 90 % level)
Landslide LD E1 D I Lower T T Upper T % ** %
event ID (dd/mm/yyyy) (mm) (d) (mm/d) (y) (y) (y) related related
to T* to T*
Tabulated results for E1

31 03/02/1996 57.0 3 19.0 0.2 0.2 0.2 4.5 3.0
39 31/01/1997 0.0 1 0.0 0.0 0.0 0.0 0.0 0.0
42 25/12/1996 2.1 1 2.1 0.0 0.0 0.0 5.6 4.8
44 04/01/1997 55.0 3 18.3 0.1 0.2 0.2 0.5 0.2
45 06/01/1997 118.0 7 16.9 0.1 0.1 0.1 0.6 0.2
45 30/01/2010 37.0 5 7.4 0.0 0.0 0.0 0.3 0.2
49 02/01/1997 105.0 2 52.5 1.3 2.1 2.8 5.1 2.1
50 08/01/1997 16.0 1 16.0 0.1 0.1 0.1 4.5 3.1
60 27/01/1997 29.1 6 4.9 0.0 0.0 0.0 0.1 0.0
72 06/02/2009 76.0 3 25.3 0.3 0.4 0.5 38.2 30.3
74 30/01/2009 3.0 1 3.0 0.0 0.0 0.0 377.6
78 26/12/2009 4.0 1 4.0 0.0 0.0 0.0 0.1 0.0
89 07/01/2010 30.0 2 15.0 0.1 0.1 0.1 1.1 0.4
94 05/03/2009 14.2 1 14.2 0.1 0.1 0.1 260.3
101 20/12/1996 83.4 3 27.8 0.6 0.8 0.9 68.2 52.6
102 20/12/1996 40.0 1 40.0 0.5 0.6 0.7 123.7 106.0
110 12/11/1996 140.0 2 70.0 4.0 13.5 29.7 100.0 45.4
114 12/11/1996 200.0 2 100.0 4.9 13.2 24.5 16,607.4
153 01/01/1997 28.3 1 28.3 0.2 0.3 0.3 28.8 22.7
155 08/01/1997 13.1 2 6.6 0.0 0.0 0.0 1.5 1.1
Mean 52.6 2.4 23.6 0.6 1.6 3.0 881.4 16.0
T quartiles 25 % 50 % 75 %
0.0 0.1 0.5
* Reference T values for comparison in Table 4

** Removing the three larger T values or outliers
values were compared in order to determine which of these segments were the most
adequate to represent the catalogued landslide CRTs. In this way, the rainfall event
characteristics triggering the recorded landslides are represented by these CRT rainfall
variable tabulated values. Similarly, the observed occurrence probability for landslides
occurring within the same hydrological and geomorphological conditions is expressed by
the tabulated CRT T-values. The tabulated results and graphs are shown and interpreted
bellow:
E1 (Table 3). This segment represents the accumulated rainfall from the last storm,
covering a period that includes the landslide date. During this period, it is common to get
null and low values in terms of accumulated rainfall, which are in turn associated with
substantially low T-values, some even lower than a seasonal cycle (1 year). Thus,
thresholds based on E1 could potentially underestimate the recurrence probability and
thereby be unsuitable for landslide hazard assessment. Evidence of this is the resulting T-
mean value of column ‘‘** % related to T*’’ (omitting outliers), which represents 16 % of
123
Nat Hazards (2016) 84:185–212 201
the mean return period for A1 (T* in Table 4). Similarly, E3 does not cover other
anomalies previous to the last storm. Therefore, T also leads to low values although the
accumulated rainfall increases with the higher duration.
• A1 (Table 4). Considering that this variable involves only the values within the
antecedent rainfall anomalies, it is assumed to be a suitable variable to assess CRTs and
their linked T-values (not being underestimated or overestimated values).
• A2 (Table 5). This variable covers a longer duration than A1 and E1. In contrast, A2
contains common daily rainfall values. Consequently, the accumulated rainfall found in
A2 leads to low T-values as in the case of E1.
• E2 (Table 6). This variable includes all the rainfall recorded from the first anomaly
appearing in the antecedent rainfall until the last storm related to the landslide date.
Unlike A2, by adding the last storm (contemporaneous to the landslide date), the higher
accumulated rainfall values result in an overestimation of T-values.
Accordingly, with the above observations, segment A1 is assumed as the more suit-
able to assess CRTs together with their T values, giving a mean T of 12.4 years (Table 4).
However, this value represents the global T for any type of landslide and their charac-
teristics (lithology, slope and other). In addition to this descriptive result, the quartiles
show that up to 25 % of the landslides events exceed a 24-year return period, whereas a
25 % of catalogued landslides were linked to return periods ranging from 3.6 to 24.0 years.
The median T (50 %) was established at 3.6 years.
Fig. 5 a The I-D and T-D plots for A1 and I-D adjusted curve. b Ibid after removing outliers for
T-D and adding its adjusted curve
123
202 Nat Hazards (2016) 84:185–212
Table 4 Landslides associated rainfall variables (D, I) related to A1, and T with its upper and lower
Landslide LD E1 A1 D I Lower Mean Upper
event ID (dd/mm/yyyy) (mm) (mm) (d) (mm/d) mean T T (T*) mean T
(y) (y) (y)
Tabulated results for A1

31 03/02/1996 57.0 200.0 18 11.1 2.3 4.4 6.5
39 31/01/1997 0.0 67.5 2 33.8 9.2 36.6 94.7
42 25/12/1996 2.1 361.0 40 9.0 0.3 0.4 0.5
44 04/01/1997 55.0 175.0 2 87.5 7.8 33.2 88.9
45 06/01/1997 118.0 200.0 2 100.0 5.1 22.0 58.9
45 30/01/2010 37.0 126.5 9 14.1 3.7 9.6 17.6
49 02/01/1997 105.0 180.3 10 18.0 10.9 42.0 104.0
50 08/01/1997 16.0 235.0 36 6.5 1.5 2.7 3.9
60 27/01/1997 29.1 80.5 3 26.8 8.3 30.0 73.3
72 06/02/2009 76.0 55.0 1 55.0 0.7 1.1 1.4
74 30/01/2009 3.0 151.0 64 2.4 0.0 0.0 0.0
78 26/12/2009 4.0 233.0 6 38.8 10.6 40.4 98.8
89 07/01/2010 30.0 110.0 4 27.5 2.0 7.8 19.8
94 05/03/2009 14.2 97.3 4 24.3 0.0 0.0 0.0
101 20/12/1996 83.4 231.0 2 115.5 0.7 1.1 1.4
102 20/12/1996 40.0 368.0 40 9.2 0.4 0.5 0.6
110 12/11/1996 140.0 104.0 1 104.0 4.0 13.5 29.7
114 12/11/1996 200.0 64.0 1 64.0 0.1 0.1 0.1
153 01/01/1997 28.3 361.0 40 9.0 0.6 1.0 1.2
155 08/01/1997 13.1 361.0 40 9.0 1.3 2.4 3.5
Mean 52.6 188.1 16.3 38.3 3.5 12.4 30.2
T quartiles 25 % 50 % 75 %
0.9 3.6 24.0
* Reference T values for comparison
Regarding the only reactivation registered (ID = 45), as can be observed in Table 4, the
accumulated rainfall for A1 shows significantly higher values than the last storm (E1)
occurring in both dates (06/01/1997 and 30/01/2010). The event reactivation with ID = 45
was reported in the 2009–2010 wet season, when the A1 resulted in a significantly lower
value than its magnitude in the 1996–1997 wet season (126.5 mm vs. 200 mm). This fact
could be attributed to the reactivation of pre-existing surface failure with lower strength
parameters. However, these two landslide events can also be interpreted as two events
exceeding a lower threshold.
When representing the critical rainfall thresholds by the I-D curve for A1, a power-law
correlation can be adjusted (see Eq. 5 and Fig. 5a). In the same way, when the outliers are
omitted, the trend depicted by the relationship T-D also fits a power-law function (Eq. 6
and Fig. 5b).
123
Nat Hazards (2016) 84:185–212 203
Table 5 Landslides associated rainfall variables (D, I) related to A2, and T with its upper and lower
Landslide LD E1 A2 D I Lower Mean T Upper %
event ID (dd/mm/yyyy) (mm) (mm) (d) (mm/d) mean T (y) mean T related
(y) (y) to T*
Tabulated results for A2

31 03/02/1996 57.0 532.0 48 11.1 1.2 2.2 3.2 50.4
39 31/01/1997 0.0 607.1 81 7.5 0.8 1.9 3.7 5.3
42 25/12/1996 2.1 398.5 44 9.1 0.3 0.4 0.4 91.8
44 04/01/1997 55.0 633.0 52 12.2 1.0 2.3 4.7 7.0
45 06/01/1997 118.0 832.5 54 15.4 0.9 1.9 3.6 8.7
45 30/01/2010 37.0 696.6 44 15.8 0.8 2.0 3.7 21.3
49 02/01/1997 105.0 474.5 24 19.8 5.1 18.4 44.5 43.7
50 08/01/1997 16.0 506.0 57 8.9 1.1 2.0 2.7 72.2
60 27/01/1997 29.1 606.6 77 7.9 1.1 2.5 4.8 8.4
72 06/02/2009 76.0 71.0 3 23.7 0.4 0.6 0.7 51.6
74 30/01/2009 3.0 148.0 63 2.3 0.0 0.0 0.0 95.6
78 26/12/2009 4.0 282.0 7 40.3 9.2 34.7 84.7 85.9
89 07/01/2010 30.0 471.5 19 24.8 0.5 1.8 4.4 23.4
94 05/03/2009 14.2 139.6 31 4.5 0.0 0.0 0.0 99.5
101 20/12/1996 83.4 394.5 37 10.7 0.3 0.3 0.4 31.1
102 20/12/1996 40.0 328.0 39 8.4 0.4 0.5 0.6 99.4
110 12/11/1996 140.0 0.0 1 0.0 1.5 3.4 5.4 25.0
114 12/11/1996 200.0 70.0 29 2.4 0.3 0.4 0.5 525.1
153 01/01/1997 28.3 452.2 51 8.9 0.5 0.8 1.0 82.0
155 08/01/1997 13.1 553.3 57 9.7 1.0 1.8 2.5 73.5
Mean 52.6 409.8 40.9 12.2 1.3 3.9 8.6 75.0
T quartiles 25 % 50 % 75 %
0.5 1.9 2.3
I of A1 ¼ 88:005 x0:69 ð5Þ
T of A1 ¼ 62:733 x1:25 ð6Þ
By analysing a small set of landslides events, it is difficult to correlate landslides types or

their characteristics with rainfall variables. Nonetheless, some results and differences are
summarized in Table 7 taking into account the antecedent rainfall represented by A1. The
findings show that landslides affecting natural slopes (N) are less probable (higher T-mean)
than man-made slope-cuts failures (S), with a difference between the T-mean for N and the T-
mean for S that exceeds 6.5 years (16.4–9.7 years). It was also found that the A1-intensity
decreases from N to S slopes. Regarding the landslide type, slides (SL) are less frequent
(T * 22 y), while rockfalls (RF) and complex (COM SL-RF) types have similar return
periods (8–10 years). In the case of the A1-mean intensity, it was found that complex
123
204 Nat Hazards (2016) 84:185–212
landslides were triggered by higher intensity rainfall than rockfalls or slides were, while the
higher duration (23.6 days) of slides appeared associated with the highest return period.
5 Discussion and conclusions
By implementing the methodology presented in this paper, landslide trigger factor char-
acteristics were added to the scant information provided by a small set of landslide events
(20 landslides, one of which was a reactivation) occurring within two important hydro-
logical years (1996–1997, 2009–2010). This was made possible by applying a frequency
analysis based on the generation of multiple partial duration series (PDSs) of the cumu-
lative rainfall after considering a wide range of durations (1 to 90 days). PDS has the
advantage of not omitting relative annual maximums values when compared to using an
annual maximum series (AMS). The use of a wide range of rainfall event durations enables
the possibility to detect and select the most suitable threshold instead of limiting the
analysis to specific duration sets. To obtain the starting time or date of a rainfall event
triggering a landslide, it is common for researchers to use heuristic or visual inspection to
recognize the peak or significant changes in rainfall depth (Aleotti 2004; Vennari et al.
2014). These methods may work better with shallow landslides as they are triggered by
heavy and short storms. In turn, in order to deal with every landslide type effectively, it is
necessary to study all the possible results from the cumulative rainfall by taking into
account a large interval of rainfall duration. Thus, rainfall anomalies can be found more
accurately instead of using threshold values that are defined by daily rainfall peaks or
preselected rainfall durations. For these reasons, in this study, a PDS-based frequency
analysis was applied to the entire spectrum of possible cumulative rainfall values to detect
uncommon values. These cumulative rainfall values were calculated for a long duration
range [1, 90]. In this way, the continuous return period curve for each exceedance prob-
ability could be obtained for all the possible cumulative rainfall–duration combinations.
Thus, cumulative rainfall anomalies were better highlighted by their corresponding return
period (T) spectrum anomalies. In addition, another advantage of this approach is that more
than one possible threshold could be analysed and compared after segmenting the resulting
graph curves into intervals which show rainfall variable value anomalies. The general
results provided by the segmentation of the PDS-based frequency analysis rainfall vari-
ables were then interpreted to determine the most suitable segment, A1, to assess CRT
values. Consequently, the rainfall variable values linked to A1, i.e. antecedent cumulative
rainfall, intensity and duration values, as well as their corresponding return periods, were
extracted to enable a descriptive statistical analysis of these variables. Additionally, the
CRT curves could also be plotted from the rainfall variable values obtained from the
segmentation process.
Unlike previous research efforts addressed in the study area, this paper involves a
retrospective analysis with the generation of more detailed information concerning rainfall
behaviour and its role as a major landslide triggering factor. This methodology was applied
to a regional extent to estimate overall thresholds and landslide recurrence. This is
achieved beginning from the individualized frequency analysis for every dated landslide.
In this way, uncommon rainfall events were highlighted and their rainfall variables and
return period were obtained. After the application of this methodology, the major outcomes
can be summarized and discussed below:
123
Nat Hazards (2016) 84:185–212 205
Table 6 Landslides associated rainfall variables (D, I) related to E2, and T with its upper and lower
Landslide LD E1 E2 D I Lower T T Upper T %
event ID (dd/mm/yyyy) (mm) (mm) (d) (mm/d) (y) (y) (y) related
to T*
Tabulated results for E2

31 03/02/1996 57.0 589.0 51 11.5 4.6 11.0 18.6 249.5
39 31/01/1997 0.0 939.8 82 11.5 15.0 65.9 178.0 180.0
42 25/12/1996 2.1 400.6 45 8.9 0.6 0.9 1.1 217.4
44 04/01/1997 55.0 688.0 55 12.5 14.9 65.4 176.7 1971.3
45 06/01/1997 118.0 862.5 57 15.1 9.9 43.4 117.0 197.4
45 30/01/2010 37.0 733.6 49 15.0 5.1 15.2 30.2 158.8
49 02/01/1997 105.0 579.5 26 22.3 9.7 33.1 72.9 78.9
50 08/01/1997 16.0 522.0 58 9.0 4.9 12.3 21.7 453.7
60 27/01/1997 29.1 939.3 78 12.0 15.0 65.9 178.1 219.5
72 06/02/2009 76.0 147.0 6 24.5 0.7 1.1 1.4 100.0
74 30/01/2009 3.0 151.0 64 2.4 0.0 0.0 0.0 278.5
78 26/12/2009 4.0 286.0 8 35.8 14.4 63.2 170.7 156.4
89 07/01/2010 30.0 501.5 21 23.9 2.5 11.0 29.8 141.1
94 05/03/2009 14.2 153.8 32 4.8 0.1 0.1 0.1 154.0
101 20/12/1996 83.4 477.9 40 11.9 1.3 2.0 2.7 183.7
102 20/12/1996 40.0 368.0 40 9.2 0.4 0.6 0.6 111.5
110 12/11/1996 140.0 140.0 3 46.7 4.0 13.5 29.7 100.0
114 12/11/1996 200.0 270.0 31 8.7 0.1 0.1 0.1 100.0
153 01/01/1997 28.3 480.5 52 9.2 2.5 5.0 7.6 514.6
155 08/01/1997 13.1 566.4 59 9.6 2.9 6.2 9.7 251.8
Mean 52.6 489.8 42.9 15.2 5.4 20.8 52.3 202.2
T quartiles 25 % 50 % 75 %
1.1 11.0 35.7
• The corresponding T-values for the cumulative rainfall during the last storm (E1)
become lower than when only more significant antecedent rainfall (A1) anomalies are
dealt with. This could result in excessively low return period values or an
underestimation of landslide hazard. From the tabulated findings, the average T for
E1 results in a 16 % of the average T for A1. On the contrary, the cumulative rainfall
E2 covering the period from the beginning with the detection of the first anomaly and
ending with the landslide date tends to overestimate the T-mean and consequently
overestimate landslide hazard. In this study, the T-mean for E2 doubles the T-mean for
A1 (202 %). Thereby, it is suggested here to select A1 to analyse its characteristics as
well as their expected return period.
• The I-D threshold curve for A1 fits a power law I ¼ a Db , which is a common trend
(Guzzetti et al. 2007). In this specific case, its parameters resulted in a = 88.005 and
b = 0.69.
• Considering the two major segments analysed after dividing the graphical informa-
tion—A1 and E1—a negative correlation was observed that associates higher
123
206 Nat Hazards (2016) 84:185–212
Table 7 A summary of relevant rainfall variables related to landslides characteristics. Global Av. repre-
sents the variable average including every landslide event
Slope type T-mean based Mean intensity Landslide type T-mean Mean Mean
on A1 of A1 based on A1 intensity duration
(y) (mm/d) (y) of A1 of A1
(mm/d) (d)
S 9.7 15.4 SL 22.4 30.1 23.6

X 10.7 39.1 RF 8.1 39.3 15.9
N 16.4 51.9 COM (SL-RF) 10.0 42.5 12
Global Av. 12.4 38.3 Global Av. 12.4 38.3 16.3
antecedent rainfall to lower rainfall magnitudes for the last storm (E1) contempora-
neous to the landslide date. Nevertheless, in some cases, this trend is reversed or both
variables resulted in low or high values. A low reading in both variables may be
attributed to an underestimation of recorded rainfall, that is, the landslide site
precipitation could be significantly greater than the precipitation recorded at its closest
pluviometric gauge. However, other unknown factors should also be considered, such
as a lower strength reactivation conditions, poor stabilization measures or other factors
possibly omitted in the reported data on landslides. As observed in Fig. 5b, even some
of the distribution tail points that fit a power law get nearly null T-values. After
carrying out an additional review of the entire wet seasons related to those points, their
corresponding very low return periods in Table 4 and Fig. 5b were confirmed.
Nevertheless, the total precipitation when adding up the contemporaneous and
antecedent rainfall (A1 and E1), for each of the 5 cases that had lower T-values than
1 year (IDs: 42, 74, 94, 102 and 114), notably ranged from 15 % to 57 % (56, 23, 15,
57 and 31 %, respectively) of the mean annual rainfall (650 mm recorded from October
to April). These high percentages emphasized the importance of the total accumulated
rainfall compared to only the landslide contemporaneous or antecedent rainfall. In
general, these cases are characterized by low intensity and high accumulated rainfall.
Among the 5 cases pointed out, an exception is observed when dealing with ID = 114.
In this case, the contemporaneous rainfall to the landslide event is significantly greater
(200 mm recorded in 2 days) than the antecedent rainfall. Thus, the contemporaneous
rainfall (E1 in Table 3) return period of 13.2 years should be considered here to express
the landslide recurrence probability, instead of using the antecedent rainfall return
period.
• One of the shortcomings derived from the small set of landslide events was the minor
number of different landslide types found, which made it difficult to deal with
statistical inferences. Nonetheless, several conclusions could be established. The mean
natural landslide return period was found to be 6.5 years higher than the mean slope-cut
failure return period (16.4 years vs. 9.7 years). In addition, natural landslides were
triggered by a higher antecedent rainfall (A1) mean intensity (39.1 mm/d vs. 15.4 mm/
d). In the case of landslide type, slides were triggered by lower intensity (30.1 mm/d)
than rockfalls (39.3 mm/d) or complex landslides (42.5 mm/d). Nonetheless, despite
the lower intensity values for slides, they showed longer return periods (22.4 y vs. 8.1
y) associated with a significantly greater mean duration. In addition, 50 % of the
catalogued landslides were linked to return periods of up to 3.6 years, whereas only
25 % of the antecedent rainfall magnitudes (A1) exceeded a return period of 24 years.
123
Nat Hazards (2016) 84:185–212 207
• Finally, just one landslide (ID = 45) was activated or reactivated in the first period
(1996–1997) and later reactivated in the second period (2009–2010) with lower rainfall
variable values. This evolution can be attributed to an increase in slope-failure
weakening, which results in a decrease in the pore-water pressure that enables the shear
strength of the slide zone to be exceeded (Campbell 1975; Wilson 1989; Terlien 1998;
Iverson 2000). An alternative explanation that could also be considered is that a lower
threshold not being exceeded by rainfall variables resulted in the lower values found in
the second landslide event associated with ID = 45.
In general, the methodology developed through the present research work constitutes a
process capable of expanding and completing valuable rainfall information related to the
triggering of past landslides. Nevertheless, the findings obtained here should be treated as
preliminary results, whereas a greater number of registered landslide events would enable
the validation and calibration of threshold curves. Similarly, to better infer the threshold
and occurrence probabilities for different landslide types, climate conditions, and possible
geomorphological features, an updated and larger database is also recommended. Fur-
thermore, adequate planning in fields and in laboratory testing campaigns, as well as the
monitoring of potentially unstable areas, is necessary to apply a physically based
methodology when no previous geotechnical and hydrological data are available. The CRT
values that result from the physically based methodology can then be compared and cross-
validated with the CRT values found through the experimental methodology presented in
this paper.
The information generated throughout the application of this methodology is necessary
and useful for landslide hazard and risk assessment. Moreover, its value lies in its potential
to be used as a warming system input. The need of more comprehensive and reliable
landslide databases in spatial information infrastructures and risk management programs
has been put forward in recent inter-governmental policies and project reports within the
last decade. Nonetheless, it is clear that the findings of the applied methodologies are
severely limited by a lack of available information. Thus, further efforts are necessary for
the continued improvement of any methodology concerned with the assessment of land-
slide risk components, such as susceptibility, hazard, vulnerability and total risk. The
implementation of the following suggestions would greatly contribute to the improvement
of landslide risk assessment:
• The creation and continually updating of landslides databases that follow risk
assessment guides and policies. This activity would involve compiling important data
such as landslide date, size and type; slope gradient, lithology, failure-plane depth, as
well as daily or even hourly measurements that are associated with triggering factor
variables. These data should be gathered every time that a new landslide event is
recognized or reported.
• Make the above data publicly available, avoiding long administrative procedures to
save time in the development and application of risk assessment methodologies. For
example, granting online access to public databases without long waiting periods to
obtain the data.
• In addition, the advantages of new remote sensing technology to gather sequential data
(e.g. inter-wet seasonal acquisitions), which requires only a small investment in time,
should be taken into account. In this way, using the latest technology, these data can be
then analysed to assess the landslide occurrence probability in terms of return period,
which is inversely proportional to the yearly landslide frequency given in either small
or large areas.
123
208 Nat Hazards (2016) 84:185–212
• Finally, to deal more effectively with advanced hazard assessment, the instalment of a
denser rainfall–gauge network in areas more susceptible to landslides is advisable.
With the same aim of gathering more precise information, geotechnical monitoring
should be carried out in areas with higher susceptibility level. This practice will enable
a more accurate modelling of the processes involved in landslides and thus an advance
in the knowledge concerning the relationship between landslide occurrence and their
triggering factors (cause–effect relationship).
Acknowledgments The implementation of some statistical formulae into the worksheets and macros was
possible thanks to the guidelines provided by Ronald. J. Oosterbaan, researcher of International Institute for
Land Reclamation and Improvement (ILRI), Wageningen, The Netherlands. Rainfall records have been
supplied by the National Meteorological Institute of Spain. This research was supported by the project
CGL2008-04854 funded by the Ministry of Science and Education of Spain. It was developed in the RMN-
121 Research Group funded by the Andalusian Research Plan.
References
AEMET (2010) Resumen anual climatológico 2010. Agencia Estatal de Meteorologı́a. URL: http://www.
aemet.es/documentos/es/serviciosclimaticos/vigilancia_clima/resumenes_climat/anuales/res_anual_
clim_2010.pdf (Last accessed: 13/11/2014)
AEMET (2014) Agencia Estatal de Meteorologı́a. URL: http://www.aemet.es/es/portada (Last accessed:
13/11/2014)
Agencia de Medio Ambiente y Agua (2014) Red de información Ambiental de Andalucı́a (REDIAM).
Consejerı́a de Medio Ambiente y Ordenación del Territorio, Junta de Andalucı́a
Aleotti P (2004) A warning system for rainfall-induced shallow failures. Eng Geol 73(3–4):247–265. doi:10.
1016/j.enggeo.2004.01.007
Aleotti P, Chowdhury R (1999) Landslide hazard assessment: summary review and new perspectives. Bull
Eng Geol Environ 58(1):21–44
Aleotti P, Canuti P, Falorni G, Fanti R, Grimaldi G, Guida D, Lombardi G, Pappalardo G, Polloni G (2003)
Assessment of potential debris flow inundation areas on a small alluvial fan in southern Italy. In:
Proceedings of the international conference on fast movements—prediction and prevention, Sorrento,
May, pp 11–13
Ayala FJ, Elizaga E, González de Vallejo LI (1987) Impacto económico y social de los riesgos geológicos
en España. In: Serie Geológica Ambiental. IGME, Madrid, p 134
Barnett V (1975) Probability plotting methods and order statistics. J R Stat Soc: Ser C (Appl Stat)
24(1):95–108. doi:10.2307/2346708
Bonnard C, Noverraz F (2001) Influence of climate change on large landslides: assessment of long term
movements and trends. In: Proceedings of the international conference on landslides, Gluckauf, Essen,
Davos, pp 121–138
Borga M, Dalla Fontana G, Da Ros D, Marchi L (1998) Shallow landslide hazard assessment using a
physically based model and digital elevation data. Environ Geol 35(2–3):81–88. doi:10.1007/
s002540050295
Borga M, Dalla Fontana G, Cazorzi F (2002) Analysis of topographic and climatic control on rainfall-
triggered shallow landsliding using a quasi-dynamic wetness index. J Hydrol 268(1–4):56–71. doi:10.
1016/s0022-1694(02)00118-x
Brunetti MT, Luino F, Vennari C, Peruccacci S, Biddoccu M, Valigi D, Luciani S, Cirio CG, Rossi M,
Nigrelli G, Ardizzone F, Palma M, Guzzetti F (2013) Rainfall thresholds for possible occurrence of
shallow landslides and debris flows in Italy. In: Schneuwly-Bollschweiler M, Stoffel M, Rudolf-Miklau
F (eds) Dating torrential processes on fans and cones: methods and their application for hazard and risk
assessment. Springer, Dordrecht, pp 327–339. doi:10.1007/978-94-007-4336-6_22
Campbell RH (1975) Soil slips, debris flows, and rainstorms in the Santa Monica Mountains and vicinity,
Southern California survey professional paper. US Geological 851:51
Cannon S, Gartner J (2005) Wildfire-related debris flow from a hazards perspective. Debris-flow hazards and
related phenomena, vol Sec. 15. Springer, Berlin, pp 363–385. doi:10.1007/3-540-27129-5_15
123
Nat Hazards (2016) 84:185–212 209
Cardinali M, Reichenbach P, Guzzetti F, Ardizzone F, Antonini G, Galli M, Cacciano M, Castellani M,

Salvati P (2002) A geomorphological approach to the estimation of landslide hazards and risks in
Umbria, Central Italy. Nat Hazards Earth Syst Sci 2(1–2):57–72. doi:10.5194/nhess-2-57-2002
Casale R, Fantecchi R, Flageolet JC (1994) Temporal occurrence and forecasting of landslides in the
European Community. In: Casale R, Fantecchi R, Flageolet JC (eds) Final report. Programme Epoch
(Ct. 90 0025). European Community, p 957
Castillo A, Martı́n-Rosales W, Osorio R (1996) Erosión hı́drica en la cuenca del rı́o Guadalfeo (Granada);
estudio comparativo de las metodologı́as de la U.S.L.E y Fournier. Geogaceta 19:142–145
Chacón J (2003) Geologic and Geomorphologic Risks: Identification, Analysis and Consequence prevention
of Landslides. Áreas: revista Internacional de Ciencias Sociales (Ejemplar dedicado a: Los procesos de
riesgo con origen natural: una constante en la relación entre hombre y medio). Ediciones de la
Universidad de Murcia. Murcia 23:33–64
Chacón J, Irigaray C, Fernandez T, El Hamdouni R (2006a) Engineering geology maps: landslides and
geographical information systems. Bull Eng Geol Environ 65(4):341–411. doi:10.1007/s10064-006-
0064-z
Chacón J, Irigaray Fernández C, Fernández T, El Hamdouni R (2006b) Landslides in the main urban areas of
the Granada province, Andalucia, Spain. IAEG 2006. Nottingham
Chacón J, Irigaray C, El Hamdouni R, Jiménez-Perálvarez J (2010) Diachroneity of landslides. In: Williams
AL, Pinches GM, Chin CY, McMorran TJ and Massey CI (eds.) Geologically Active. Taylor & Francis
Group. CRC Press-Balkema, vol. 1, p. 999-1006
Chacón J, Irigaray C, El Hamdouni R, Valverde-Palacios I, Valverde-Espinosa I, Calvo F, Jiménez-Per-
álvarez J, Chacon E, Fernández P, Garrido J, Lamas F (2012) Engineering and Environmental Geology
of Granada and its Metropolitan Area (Spain). Environ Eng Geosci 18(3):217–260. doi:10.2113/
gseegeosci.18.3.217
Community European (2007) Directive 2007/2/EC of the European Parliament and of the Council of 14
March 2007 establishing an Infrastructure for Spatial Information in the European Community
(INSPIRE). Off J Eur Union L108:14
Corominas J (2000) Landslides and climate. Keynote lecture. In: Bromhead E, Dixon N, Ibsen ML, Cardiff
AA (eds) Proceedings 8th international symposium on landslides. Balkema, vol 4, pp 1–33
Corominas J, Moya J (2008) A review of assessing landslide frequency for hazard zoning purposes. Eng
Geol 102(3–4):193–213. doi:10.1016/j.enggeo.2008.03.018
Crosta GB, Frattini P (2001) Rainfall thresholds for triggering soil slips and debris flow. In: Mugnai A,
Guzzetti F, Roth G (eds) Proceedings 2nd EGS Plinius conference on Mediterranean Storms, Siena,
pp 463–487
Crosta G, Frattini P (2003) Distributed modelling of shallow landslides triggered by intense rainfall. Nat
Hazards Earth Syst Sci 3(1–2):81–93. doi:10.5194/nhess-3-81-2003
Crozier M (1986) Landslides: causes, consequences and environment. Croom Helm, London, p 252
Crozier MJ, Eyles RJ (1980) Assessing the probability of rapid mass movement. In Proceedings of 3rd
Australia-New Zealand Conference on Geomechanics, Wellington, N.Z.: Institution of Professional
Engineers New Zealand, 1980: 2-47-2-51. Proceedings of Technical Groups. Vol. 6, p. 247-251
CumFreq (2014) CumFreq: Cumulative frequency analysis with probability distribution fitting. URL: http://
www.waterlog.info/cumfreq.htm [Last accessed: 05/09/2014]
Cunnane C (1973) A particular comparison of annual maxima and partial duration series methods of flood
frequency prediction. J Hydrol 18(3–4):257–271. doi:10.1016/0022-1694(73)90051-6
D’Odorico P, Fagherazzi S (2003) A probabilistic model of rainfall-triggered shallow landslides in hollows:
A long-term analysis. Water Resources Research 39 (9): ESG61-ESG614
De Vita P, Napolitano E, Godt J, Baum R (2013) Deterministic estimation of hydrological thresholds for
shallow landslide initiation and slope stability models: case study from the Somma-Vesuvius area of
southern Italy. Landslides 10(6):713–728. doi:10.1007/s10346-012-0348-2
Dikau R, Cavallin A, Jäger S (1996) Databases and GIS for landslide research in Europe. Geomorphology
15(3):227–239
Diputación de Granada and the Instituto Geológico Minero de España (2007) Atlas de Riesgos Naturales en
la Provincia de Ganada. Ferrer M (dir). ISBN/ISSN: 978-84-7807-438-9
Duggal KN, Soni JP (2005) Statistics Applied to Hydrology. Elements of water resources engineering. New
Age International Publishers. New Delhi, p. 47. ISBN/ISSN: 81-224-0507-9
Dykes AP, Gunn J, Convery KJ (2008) Landslides in blanket peat on Cuilcagh Mountain, northwest Ireland.
Geomorphology 102(3–4):325–340. doi:10.1016/j.geomorph.2008.04.003
El Hamdouni R (2001) Estudio de Movimientos de Ladera en la Cuenca del Rı́o Ízbor mediante un SIG:
Contribución al Conocimiento de la Relación entre Tectónica Activa e Inestabilidad de Vertientes.
Unpublished PhD Thesis. Department of Civil Engineering, University of Granada, Spain, p 429
123
210 Nat Hazards (2016) 84:185–212
El Hamdouni R, Irigaray C, Fernández T, Chacón J, Keller EA (2008) Assessment of relative active

tectonics, southwest border of the Sierra Nevada (southern Spain). Geomorphology 96(1–2):150–173.
doi:10.1016/j.geomorph.2007.08.004
Endo (1969) Probable distribution of the amount of rainfall causing landslides, Annual Report 1968.
Hokkaico Branch, For. Exp. Stn., Sapporo. Japan, pp 122–136
ESRI (2012) ArGIS Desktop 10.1. Environmental System Research Institute, Inc (ESRI)
Fell R, Corominas J, Bonnard C, Cascini L, Leroi E, Savage WZ (2008) Guidelines for landslide suscep-
tibility, hazard and risk zoning for land use planning. Eng Geol 102(3–4):85–98. doi:10.1016/j.enggeo.
2008.03.022
Fernández T, Irigaray C, El Hamdouni R, Chacón J (2003) Methodology for landslide susceptibility
mapping by means of a GIS. Application to the Contraviesa Area (Granada, Spain). Nat Hazards
30(3):297–308. doi:10.1023/B:NHAZ.0000007092.51910.3f
Fernández P, Irigaray C, Jimenez J, El Hamdouni R, Crosetto M, Monserrat O, Chacon J (2009) First
delimitation of areas affected by ground deformations in the Guadalfeo River Valley and Granada
metropolitan area (Spain) using the DInSAR technique. Eng Geol 105(1–2):84–101. doi:10.1016/j.
enggeo.2008.12.005
Fernández T, Pérez J, Delgado J, Cardenal F, Irigaray C, Chacón J (2011) Evolution of a diachronic
landslide by comparison between different DEMs obtained from Digital Photogrammetry Techniques
in Las Alpujarras (Granada, Southern Spain). In: Conference of geoinformation for disaster man-
agement (GI4DM), Antalya
Frattini P, Crosta GB, Fusi N, Dal Negro P (2004) Shallow landslides in pyroclastic soils: A distributed
modelling approach for hazard assessment. Eng Geol 73(3–4):277–295. doi:10.1016/j.enggeo.2004.01.
009
Glade T, Crozier M, Smith P (2000) Applying probability determination to refine landslide-triggering
rainfall thresholds using an empirical ‘‘Antecedent Daily Rainfall Model’’. Pure appl Geophys
157(6–8):1059–1079. doi:10.1007/s000240050017
Gómez-Pugnaire MT, Galindo-Zaldı́var J, Rubatto D, González-Lodeiro F, López Sánchez-Vizcaı́no V,
Jabaloy A (2004) A reinterpretation of the Nevado-Filábride and Alpujárride Complexes (Betic
Cordillera): Field, petrography and U-Pb ages from orthogneisses (western Sierra Nevada, S Spain).
Schweiz Mineral Petrogr Mitt 84(3):303–322. doi:10.1016/j.lithos.2010.07.002
Gumbel EJ (1968) Statistics of extremes. United States of America, Columbia University Press, New York,
pp 28–34
Guzzetti F, Carrara A, Cardinali M, Reichenbach P (1999) Landslide hazard evaluation: a review of current
techniques and their application in a multi-scale study, Central Italy. Geomorphology
31(1–4):181–216. doi:10.1016/S0169-555X(99)00078-1
Guzzetti F, Cardinali M, Reichenbach P, Cipolla F, Sebastiani C, Galli M, Salvati P (2004) Landslides
triggered by the 23 November 2000 rainfall event in the Imperia Province, Western Liguria, Italy. Eng
Geol 73(3–4):229–245. doi:10.1016/j.enggeo.2004.01.006
Guzzetti F, Peruccacci S, Rossi M, Stark CP (2007) Rainfall thresholds for the initiation of landslides in
central and southern Europe. Meteorol Atmos Phys 98(3–4):239–267. doi:10.1007/s00703-007-0262-7
Guzzetti F, Peruccacci S, Rossi M, Stark C (2008) The rainfall intensity–duration control of shallow
landslides and debris flows: an update. Landslides 5(1):3–17. doi:10.1007/s10346-007-0112-1
Helsel DR, Hirsch RM (2002) Chapter A3 statistical methods in water resources. Hydrologic analysis and
interpretation. Tech Water-Resour Investig US Geol Surv 4:1–8
Hromadka TV, Phillips M (2010) Use of rainfall statistical return periods to determine threshold for mass
wasting events. Environ Eng Geosci 16(4):343–356. doi:10.2113/gseegeosci.16.4.343
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(3–4):241–258. doi:10.1016/0169-555X(95)00073-E
IDEAL (2014) Historia. URL: http://canales.ideal.es/acercaIdeal/historia.html (Last accessed: 15/9/2014)
IGN (2008–2012) Instituto Geográfico Nacional de España. Plan Nacional de Ortofotografı́a Aérea (PNOA).
Ministerio de Fomento. Dirección General del Instituto Geográfico Nacional. URL: http://www.ign.es/
PNOA/presentacion.html (Last accessed: 28/02/2013)
Innes JL (1983) Debris flows. Prog Phys Geog 7:469–501
Irigaray C, Palenzuela JA (2013) Análisis de la actividad de movimientos mediante láser escáner terrestre en
el suroeste de la Cordillera Bética (España). Revista de Geologı́a Aplicada a la Ingenierı́a y al
Ambiente 31:53–67
Irigaray C, Fernández T, El Hamdouni R, Chacón J (1999) Verification of landslide susceptibility mapping:
a case study. Technical report. Earth Surf Proc Land 24:537–544
123
Nat Hazards (2016) 84:185–212 211
Irigaray C, Lamas F, El Hamdouni R, Fernández T, Chacón J (2000) The importance of the precipitation and
the susceptibility of the slopes for the triggering of landslides along the roads. Nat Hazards
21(1):65–81. doi:10.1023/a:1008126113789
Irigaray C, Fernandez T, El Hamdouni R, Chacon J (2007) Evaluation and validation of landslide-sus-
ceptibility maps obtained by a GIS matrix method: examples from the Betic Cordillera (southern
Spain). Nat Hazards 41(1):61–79. doi:10.1007/s11069-006-9027-8
Iverson RM (2000) Landslide triggering by rain infiltration. Water Resour Res 36(7):1897–1910. doi:10.
1029/2000wr900090
Jenks GF (1967) The data model concept in statistical mapping. Int Yearbook Cartogr 7:186–190
Jibson RW (1989) Debris flows in southern Puerto Rico. Geol Soc Am Spe Pap 236:29–56. doi:10.1130/
SPE236-p29
Jiménez J, Irigaray C, El Hamdouni R, Fernandez T, Chacon J (2005) Rasgos geomorfológicos y
movimientos de ladera en la Cuenca Alta del rı́o Guadalfeo, sector Cádiar-Órgiva (Granada). Actas del
VI Simposio Nacional de Taludes y Laderas Inestables 2:891–902
Jiménez-Perálvarez JD (2012) Movimientos de ladera en la vertiente meridional de sierra nevada (Granada,
España): identificación, análisis y cartografı́a de susceptibilidad y peligrosidad mediante SIG.
Unpublished PhD Thesis. Department of Civil Engineering. University of Granada, Spain, p 210.
ISBN: 9788490282892
Jiménez-Perálvarez JD, Irigaray C, El Hamdouni R, Chacón J (2009) Building models for automatic
landslide-susceptibility analysis, mapping and validation in ArcGIS. Nat Hazards 50(3):571–590.
doi:10.1007/s11069-008-9305-8
Jiménez-Perálvarez JD, Irigaray C, El Hamdouni R, Chacón J (2011) Landslide-susceptibility mapping in a
semi-arid mountain environment: an example from the southern slopes of Sierra Nevada (Granada,
Spain). Bull Eng Geol Environ 70(2):265–277
Keller EA, Sanz de Galdeano C, Chacón J (1996) Tectonic geomorphology and earthquake hazard of Sierra
Nevada, Southern Spain. In: Chacón J, Rosua JL (eds) 1a Conferencia Internacional Sierra Nevada.
Granada, pp 201–218
Kim SK, Hong WP, Kim YK (1991) Prediction of rainfall-triggered landslides in Korea. In: Bell C (ed) 6th
international symposium on landslides, vol 2. A.A. Balkema, Rotterdam
Köeppen (1936) Das geographische system der klimate. In: Köeppen W, Geiger R (eds) Handbuch der
klimatologie, 1C. Gebrüder borntraeger, Berlin, p 44
Lacasse S, Nadim F (2009) Landslide risk assessment and mitigation strategy. In: Sassa K, Canuti P (eds)
Landslides—disaster risk reduction. Springer, Berlin, Sec. 3, p 31–61. ISBN/ISSN: 978-3-540-69966-
8. doi:10.1007/978-3-540-69970-5_3
Li C, Ma T, Zhu X, Li W (2011) The power-law relationship between landslide occurrence and rainfall
level. Geomorphology 130(3–4):221–229. doi:10.1016/j.geomorph.2011.03.018
Lumb P (1975) Slope failure in Hong Kong. Q J Eng Geol 8:31–65
Ma T, Li C, Lu Z, Wang B (2014) An effective antecedent precipitation model derived from the power-law
relationship between landslide occurrence and rainfall level. Geomorphology 216:187–192. doi:10.
1016/j.geomorph.2014.03.033
Makkonen L, Pajari M (2014) Defining sample quantiles by the true rank probability. J Prob Stat. Article ID
326579, p 6. doi:10.1155/2014/326579
Mkhandi S, Opere A, Willems P (2005) Comparison between annual maximum and peaks over threshold
models for flood frequency prediction. In International conference of UNESCO Flanders FIT FRIEND/
Nile project–towards a better cooperation, Sharm-El-Sheikh, Egypt, CD-ROM proceedings
Mohssen M (2009) Partial duration series in the annual domain. 18th World IMACS/MODSIM Congress.
Cairns, Australia, pp 2694–2700
Oosterbaan RJ (1988) Frequency predictions and their binomial confidence limits. International Commision
on Irrigation and Drainage, Special Technical Session, Economic Aspects of Flood Control and non
Structural Measures, Dubrovnik
Palenzuela JA, Irigaray C, Jiménez-Perálvarez JD, Chacón J (2013) Application of terrestrial laser scanner
to the assessment of the evolution of diachronic landslides. In: Margottini C, Canuti P, Sassa K (eds)
Landslide science and practice. Springer, Berlin, Sec. 68, pp 517–523. doi: 10.1007/978-3-642-31445-
2_68
Palenzuela JA, Marsella M, Nardinocchi C, Pérez JL, Fernández T, Chacón J, Irigaray C (2014) Landslide
detection and inventory by integrating LiDAR data in a GIS environment. Landslides. doi:10.1007/
s10346-014-0534-5
Panizza M (1996) 3 Geomorphological hazard. In: Mario P (eds) Developments in earth surface processes.
Elsevier, vol 4, pp 75–76. ISBN/ISSN: 0928-2025. doi: 10.1016/S0928-2025(96)80020-4
123
212 Nat Hazards (2016) 84:185–212
Papa MN, Medina V, Ciervo F, Bateman A (2013) Derivation of critical rainfall thresholds for shallow
landslides as a tool for debris flow early warning systems. Hydrol Earth Syst Sci 17(10):4095–4107.
doi:10.5194/hess-17-4095-2013
Petley D (2012) Global patterns of loss of life from landslides. Geology. The Geological Society of
America, vol 40. 10, pp 927–930. ISBN/ISSN: 0091-7613. doi: 10.1130/G33217.1
Red de Información Ambiental de Andalucı́a (REDIAM) (2016) Precipitación media anual en Andalucı́a:
periodo 1971–2000. Consejerı́a de Medio Ambiente, Junta de Andalucı́a. URL: http://www.
juntadeandalucia.es/medioambiente/site/rediam
Reichenbach P, Cardinali M, De Vita P, Guzzetti F (1998) Regional hydrological thresholds for landslides
and floods in the Tiber River Basin (central Italy). Environ Geol 35(2–3):146–159. doi:10.1007/
s002540050301
Ruiz Sinoga JD, Martinez Murillo JF (2009) Effects of soil surface components on soil hydrological
behaviour in a dry Mediterranean environment (Southern Spain). Geomorphology 108(3–4):234–245.
doi:10.1016/j.geomorph.2009.01.012
Schuster RL (1978) Introduction. In: Schuster RL, Krizek RJ (eds) Landslides—analysis and control,
Chapter 1, pp 1–10
Schutt B (2005) Late quaternary environmental change on the Iberian Peninsula. Erde 136(1):3
Spizzichino D, Margottini C, Trigila A, Iadanza C, Linser S (2010) Landslides. In: Ludlow D (ed) Mapping
the impacts of natural hazards and technological accidents in Europe—An overview of the last decade,
chapter 9, pp 81–93. EEA Technical report No 13/2010, 144p. ISBN 978-92-9213-168-5. ISSN
1725-2237. doi: 10.2800/62638
Spizzichino D, Margottini C, Trigila A, Iadanza C (2013) Landslide impacts in Europe: weaknesses and
strengths of databases available at European and National Scale. In: Margottini C, Canuti P, Sassa K
(eds) Landslide science and practice vol 1: landslide inventory and susceptibility and hazard zoning.
Springer, Berlin, pp 73–80. ISBN 978-3-642-31324-0. doi: 10.1007/978-3-642-31325-7_9
Suárez RR, Regueiro M (1997) Guı́a ciudadana de los Riesgos Geológicos. I.C.O.G, Madrid
Terlien MTJ (1996) Modelling spatial and temporal variations in rainfall-triggered landslides. PhD thesis.
ITC. Enschede. Holland, vol 32, p 251. ISBN/ISSN: 9061641152
Terlien MTJ (1998) The determination of statistical and deterministic hydrological landslide-triggering
thresholds. Environ Geol 35(2–3):124–130. doi:10.1007/s002540050299
Thornes JB, Alcántara-Ayala I (1998) Modelling mass failure in a Mediterranean mountain environment:
climatic, geological, topographical and erosional controls. Geomorphology 24(1):87–100. doi:10.1016/
S0169-555X(97)00103-7
Trujillo F (1995) III. Clima e información meteorológica PLAN INFOCA. Un plan de acción al servicio del
monte mediterráneo andaluz. Consejerı́a de Medio Ambiente. Junta de Andalucı́a. ISBN: 84-95785-88-
984-95785-88-9
UNESCO (1973–1979) Annual summaries of information on natural disasters, 1971–1975. UNESCO, Paris
Van Den Eeckhaut M, Hervás J (2012) State of the art of national landslide databases in Europe and their
potential for assessing landslide susceptibility, hazard and risk. Geomorphology 139–140:545–558.
doi:10.1016/j.geomorph.2011.12.006
Varnes DJ (1984) Landslide Hazard Zonation: a Review of Principles and Practice. Commision on Land-
slides of IAEG, UNESCO. United Nations Educational, Scientific and Cultural Organization. Paris.
Nat Hazards 3:61
Vennari C, Gariano SL, Antronico L, Brunetti MT, Iovine G, Peruccacci S, Terranova O, Guzzetti F (2014)
Rainfall thresholds for shallow landslide occurrence in Calabria, southern Italy. Nat Hazards Earth Syst
Sci 14(2):317–330. doi:10.5194/nhess-14-317-2014
Weibull W (1939) A statistical theory of the strength of materials. Ing Velenskaps Akad Handl 151:1–45
White ID, Mottershead DN, Harrison JJ (1996) Environmental systems, 2nd edn. Chapman and Hall,
London, p 616
Wieczorek G, Glade T (2005) Climatic factors influencing occurrence of debris flows. Debris-flow hazards
and related phenomena. Springer, Berlin, Sec. 14, pp 325–362. ISBN/ISSN: 978-3-540-20726-9.
doi:10.1007/3-540-27129-5_14
Wilson RC (1989) Rainstorms, pore pressures, and debris flows: a theoretical framework. In: Morton DM,
Salder PM (eds) Landslides in a semi-arid environment. Publications of the Inland Geological Society,
2: 1. California, vol 2
123

Palenzuela2016 Article AssessingCriticalRainfallThres

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Palenzuela2016 Article AssessingCriticalRainfallThres

Uploaded by

Copyright:

Available Formats

Nat Hazards (2016) 84:185–212

Assessing critical rainfall thresholds for landslide

José Antonio Palenzuela1 • Jorge David Jiménez-Perálvarez1 •

José Chacón1 • Clemente Irigaray1

& Jorge David Jiménez-Perálvarez

2 Setting of the study area

3 Materials and methods

3.1 Gathering necessary information

31 183 03/02/1996 S 23.53 CR, SC, P RF 779

Slope type N (natural), S (slope-cut movement), X (several or unknown types)

3.2 Rainfall thresholds and frequency analysis

Fig. 2 Spatial classification of the average annual rainfall classification

3.2.1 Generating rainfall variables

3.2.2 Frequency analysis and return period

In river-flooding studies, plot positions (observed frequencies) are usually obtained,

• Asses the experimental cumulative distribution function (CDF) of the cumulative

24,195 66.21 2 1 0.00004 0.0151 365.4531 0.002736

3.3 Determining rainfall thresholds for landslide triggering through

Tabulated results for E1

* Reference T values for comparison in Table 4

Tabulated results for A1

* Reference T values for comparison

Tabulated results for A2

* Reference T values for comparison in Table 4

I of A1 ¼ 88:005 x0:69 ð5Þ

T of A1 ¼ 62:733 x1:25 ð6Þ

By analysing a small set of landslides events, it is difficult to correlate landslides types or

5 Discussion and conclusions

Tabulated results for E2

* Reference T values for comparison in Table 4

S 9.7 15.4 SL 22.4 30.1 23.6

Cardinali M, Reichenbach P, Guzzetti F, Ardizzone F, Antonini G, Galli M, Cacciano M, Castellani M,

El Hamdouni R, Irigaray C, Fernández T, Chacón J, Keller EA (2008) Assessment of relative active

You might also like

I of A1 ¼ 88:005 x0:69 ð5Þ

T of A1 ¼ 62:733 x1:25 ð6Þ