Cross-Border Risk Assessment of
Earthquake-induced Landslides in Central Asia
vorgelegt von
Master of Science
Annamaria Saponaro
geb. in Bari
von der Fakultät VI - Planen Bauen Umwelt
der Technischen Universität Berlin
zur Erlangung des akademischen Grades
Doktorin der Naturwissenschaften
- Dr.rer.nat. -
genehmigte Dissertation
Promotionsausschuss:
Vorsitzender: Prof. Dr.-Ing. Yuri Petryna
Gutachter: Prof. Dr.-Ing. Frank Rackwitz
Gutachter: Dr. Fausto Guzzetti
Gutachter: Prof. Dr. Stefano Parolai
Tag der wissenschaftlichen Aussprache: 11. September 2017
Berlin 2018
In memory of my father
ACKNOWLEDGEMENTS
This research could not have been undertaken without the generous contribution of
some people I would like to acknowledge.
First of all, I would like to express my deepest gratitude to my supervisor Stefano
Parolai for his trust and patient guidance throughout these years. His continuous
encouragement and motivation is highly appreciated. Special thanks also to my
colleagues Marco, Marc, Max, Dino, Kevin, Shahid for inspiring fruitful discussions
and for making this work possible.
I would like to thank the colleagues of the Center of Early Warning and of the Section
2.1 for their friendship and the nice time spent together. Particularly thanks to Camilla
for each shared scientific (and not) coffee break, to Kevin for the English language
proof reading, to Dorina and Susanne for their constant and prompt help.
I am also grateful to people in CAIAG, for their warm hospitality during my trips in
Central Asia. Special thanks go to Bolot Moldobekov, who inspired interesting
landslide discussions.
I would also like to thank the reviewers for taking the time to read the dissertation.
Finally, my profound thanks go to my sister and my mum for their unconditional love
and support. Words cannot express how thankful I am to you.
AUTHOR’S DECLARATION
I hereby declare that I have produced this thesis without the prohibited assistance of
third parties and without making use of aids other than those specified; notions taken
over directly or indirectly from other sources have been identified as such. This thesis
has not previously been presented in identical or similar form to any other German or
foreign examination board.
Potsdam, 08.09.2015 Annamaria Saponaro
ABSTRACT
Central Asia is one of the most challenging places in the world where various natural
hazards can heavily injury populations and resources. Among these hazards, landslides
pose a serious threat to human life and human facilities. The large variability of local
geological materials, together with the difficulties in forecasting heavy precipitation
locally and in quantifying the level of ground shaking, call for harmonized procedures
to better quantify the hazard and the negative impact of slope failures across the Central
Asian countries. Furthermore, the increase of population and the expansion of urban
settlements towards landslide-prone slopes, exacerbate negative consequences
associated to slope failures. Especially in less developed countries which are
particularly suffering for appropriate resources, there is the urgent need to address
landslide research in order to support disaster management and planning activities at the
regional level.
Under conditions of data scarcity as well as of geographic remoteness – which are of
particular concern to Central Asian countries - a sound statistical approach is presented
that is able to quantify landslide hazard and risk, which also allows for the inference of
information about landslide potential and expected damage for areas where no data
coverage is available.
With these premises, the main objective of the work is to provide a cross-border risk
map of earthquake-induced landslides in Central Asia. To this scope, the main
components of risk are evaluated and two main questions are addressed: 1) where are
earthquake-induced landslides more likely to occur in future, 2) how is it possible to
quantify damages to exposed assets.
A first essential step of any landslide hazard and risk assessment is the preparation of a
landslide susceptibility map with the objective to identify areas where the potential for
landslide activation exists. For the purposes of this research, a landslide susceptibility
analysis is performed by exploiting new advances in Geographic Information System
(GIS) technology, together with concepts from Bayesian statistics, and promoting the
use of open-source tools. Specifically, a range of conditioning factors and their potential
impact on landslide occurrence are quantitatively assessed on the basis of the spatial
distribution of landslides by applying weights-of-evidence modelling based on (1) a
landslide inventory of past events, (2) terrain-derived variables of slope, aspect and
curvature, (3) a geological map, (4) a distance from faults map, and (5) a seismic
intensity map. A spatial validation of the proposed method is performed, indicating
sufficient measures of significance to predicted landslide susceptibility results, which
present an overall level of accuracy greater than 70%.
Secondarily, the evaluation of expected damage to exposed population is achieved at a
transnational level. A physically-based procedure, applicable over the entire Central
Asian region, is adopted 1) to identify landslide source areas, 2) to define downhill
trajectories of mass movements, and 3) to finally retrieve their impact velocities. By
incorporating such dynamic concepts, a landslide hazard map is prepared allowing for a
better quantification of the natural phenomenon. For the finally landslide risk task, the
expected destructiveness due to landslide activation is calculated by integrating the
distribution of population density with the hazard map. Results show that a relatively
high level of risk is expected for large cities, i.e., those located in the Fergana Valley, in
the proximity of Tashkent in Uzbekistan and Jalal-Abad in Kyrgyzstan, being highly
dense populated areas. On the other hand, there is a general medium level of risk
extensively expected over the entire Central Asia region, primarily due to the presence
of small settlements (having a population density in the order of 1000 persons/km2)
exposed to relatively high landslide hazard.
ZUSAMMENFASSUNG
Zentralasien ist einer der Orte der Welt, an dem verschiedenste Naturgefahren die
Bevölkerung und die vorhandene Infrastruktur schwer treffen können. Unter diesen
Gefahren stellen Hangrutsche eine bedeutende Bedrohung sowohl für menschliches
Leben als auch die Wirtschaft und die zivile Infrastruktur dar. Starke räumliche
Variationen unterschiedlicher geologischer Einheiten, zusammen mit nur schwer
vorherzusagenden lokalen Starkregenereignissen und einer ebenso schwer
abzuschätzenden Intensität der Bodenbewegung während eines Erdbebens, fordern eine
harmonisierte Herangehensweise, um die Gefährdung und die daraus entstehenden
negativen Folgen von Hangrutschen in Zentralasien in einem harmonisierten Ansatz zu
quantifizieren. Neben der ohnehin schon hohen Gefährdung erhöhen ein starkes
Bevölkerungswachstum und die daraus resultierende Ausdehnung von urbanen
Siedlungen hin zu den von Hangrutschen bedrohten Talrändern die negativen
Konsequenzen, die mit einem solchen Geländesturz einhergehen können. Insbesondere
in weniger entwickelten Ländern, die besonders an einem Mangel an Ressourcen leiden,
besteht der dringende Bedarf, die Vorsorge- und Nachsorgeaktivitäten auf regionaler
Ebene für den Katastrophenfall zu verstärken.
Unter den nachteiligen Bedingungen eines vorherrschenden Datenmangels als auch von
geographischer Abgelegenheit - welche besonders auf die zentralasiatischen Länder
zutreffen - wird in dieser Arbeit ein fundierter statistischer Ansatz vorgestellt, der es
ermöglicht, die Hangrutschungsgefährdung als auch das -risiko zu quantifizieren;
darüberhinaus erlaubt es das vorgestellte Verfahren auch, Informationen über
Hangrutschungspotential und erwartete Schäden für Gebiete ohne verfügbare
Datenabdeckung abzuschätzen.
Unter diesen Prämissen besteht das Hauptziele dieser Arbeit in der Erstellung einer
harmonisierten und grenzüberschreitenden Risikokarte für erdbeben-induzierte
Hangrutsche in Zentralasien. Zu diesem Zweck werden die Hauptkomponenten des
Risikos ausgewertet, und es werden zwei zentrale Fragen untersucht: 1) Wo befinden
sich Gebiete, in denen eine hohe Wahrscheinlichkeit für das Auftreten zukünftiger
durch Erdbeben induzierter Hangrutsche besteht? 2) Inwieweit ist es möglich, Schäden
an exponierten Vermögensgütern zu quantifizieren?
Für eine angemessene Hangrutschungsgefährdungs- und Risikoanalyse wurde in einem
ersten Schritt die Hangrutschsuszeptibiliät kartiert mit dem Ziel, insbesondere jene
Regionen zu identifizieren, die ein erhöhtes Potential für Hangrutsche aufweisen.
Hierzu macht sich die vorgelegte Arbeit die Fortschritte bei den geographischen
Informationssystemen (GIS) und Konzepte der Bayesschen Statistik zunutze. Es
werden eine Vielzahl an Bedingungsfaktoren und deren potentieller Einfluss für das
Auftreten von Hangrutschen quantitativ analysiert. Auf Basis der räumlichen Verteilung
von Hangrutschen wird eine Weight-of-Evidence Modellierung durchgeführt mittels (1)
einer Inventarisierung vergangener Hangrutschungsereignisse, (2) von der Topographie
abgeleiteter Variablen der Hangneigung, Exposition und Krümmung, (3) einer
geologischen Kartierung, (4) einer Karte mit der Entfernung zur nächstliegenden
tektonischen Verwerfung, und (5) einer probabilistischen seismischen
Gefährdungsanalyse in Einheiten der makroseismischen Intensität. Im Anschluss wird
eine räumliche Validierung der vorgeschlagenen Methode durchgeführt, die insgesamt
mit einem Genauigkeitsniveau von mehr als 70% ein ausreichendes Maß an Signifikanz
zur Vorhersage der Hangrutschsuszeptibilität aufweist.
Darauf aufbauende werden in einem zweiten Schritt die erwarteten Verluste unter der
gefährdeten Bevölkerung auf grenzüberschreitender Skala abgeschätzt. Dazu wird ein
auf die gesamte zentralasiatische Region anwendbarer, physikalisch basierter Ansatz
eingeführt, der 1) Ursprungsort von Hangrutschen, 2) Trajektorien der hangabwärts
gerichteten Massenbewegung und 3) schließlich die Auftreffgeschwindigkeit der
abwärts gleitenden Masse abschätzt. Durch die Überarbeitung der
Hangrutschungsgefährdungskarte unter Berücksichtigung dieser dynamischen Konzepte
ist es möglich, das natürliche Auftreten der Hangrutsche genauer zu quantifizieren. Für
die letztliche Risikoabschätzung, d.h. die zu erwartende Zerstörungskraft aufgrund
ausgelöster Hangrutsche, wird die Verteilung der Bevölkerungsdichte in die
Gefährdungskarte integriert.
Die Ergebnisse zeigen, dass für große Städte in der Umgebung von Taschkent
(Usbekistan) und Jalalabad (Kirgisistan) sowie für die dicht besiedelten Gebiete im
Ferganatal ein relativ hohes Risiko gegenüber Hangrutschen besteht; die gesamte
zentralasiatische Region weist ein mittleres Risikoniveau auf. Dafür ist hauptsächlich
die große Anzahl kleiner Siedlungen (mit einer Bevölkerungsdichte in der
Größenordnung von 1000 Personen/km²) verantwortlich, die einer relativ hohen
Hangrutschungsgefährdung ausgesetzt sind.
CONTENTS
ACKNOWLEDGEMENTS ................................................................................................. IV
AUTHOR’S DECLARATION ..................................................................................... V
ABSTRACT .................................................................................................................... VI
ZUSAMMENFASSUNG .................................................................................................. VIII
CONTENTS ..................................................................................................................... X
LIST OF FIGURES .......................................................................................................... XII
1 INTRODUCTION ....................................................................................................... 1
1.1 LANDSLIDE PHENOMENA IN CENTRAL ASIA ............................................................. 4
1.2 STATE OF THE ART ................................................................................................... 9
1.3 RELEVANCE OF THE WORK ..................................................................................... 12
1.4 OUTLINE OF THE THESIS ........................................................................................ 13
2 CONCEPTUAL LANDSLIDE RISK FRAMEWORK ......................................... 15
2.1 INTRODUCTION ...................................................................................................... 15
2.2 LANDSLIDE SUSCEPTIBILITY .................................................................................. 17
2.3 LANDSLIDE HAZARD .............................................................................................. 20
2.4 LANDSLIDE EXPOSURE ........................................................................................... 22
2.5 LANDSLIDE VULNERABILITY .................................................................................. 25
2.6 FINAL REMARKS ON LANDSLIDE RISK ..................................................................... 28
3 METHODOLOGIES ................................................................................................ 32
3.1 WEIGHTS OF EVIDENCE THEORY ........................................................................... 32
3.2 DYNAMIC SLOPE-STABILITY ANALYSIS .................................................................. 36
3.3 ADVANTAGES AND INNOVATIVE METHODOLOGICAL ASPECTS ............................... 39
4 DATA COLLECTION AND SPATIAL DATABASE ........................................... 41
4.1 INTRODUCTION ON DATA COLLECTION ................................................................... 41
4.2 LANDSLIDE LOCATIONS ......................................................................................... 45
4.3 TOPOGRAPHIC FACTORS: SLOPE GRADIENT, SLOPE ASPECT, PROFILE CURVATURE . 46
4.4 GEO-TECTONIC FACTORS: GEOLOGY, DISTANCE FROM FAULTS .............................. 49
4.5 POPULATION DENSITY ............................................................................................ 51
4.6 TRIGGER MECHANISM: SEISMIC GROUND MOTION .................................................. 52
5 APPLICATION OF WEIGHT-OF-EVIDENCE METHOD ................................ 55
5.1 TEST FOR CONDITIONAL INDEPENDENCY OF LANDSLIDE FACTORS ........................ 55
5.2 WEIGHTS’ CALCULATION ....................................................................................... 56
5.3 LANDSLIDE SUSCEPTIBILITY MODEL ...................................................................... 58
6 REGIONAL SLOPE-STABILITY ANALYSIS ..................................................... 59
6.1 IDENTIFICATION OF LANDSLIDE SOURCE AREAS: SEED-POINTS GENERATION.......... 59
6.2 COMPUTATION OF DOWNHILL FLOW LINES ............................................................. 62
6.3 CALCULATION OF IMPACT VELOCITY ..................................................................... 63
6.4 CREATION OF THE LANDSLIDE HAZARD INDEX (LHI) ........................................... 65
7 RESULTS AND VALIDATION .............................................................................. 68
7.1 LANDSLIDE SUSCEPTIBILITY RESULTS AND VALIDATION ........................................ 68
7.2 LANDSLIDE SUSCEPTIBILITY MAP FOR CENTRAL ASIA ........................................... 75
7.3 LANDSLIDE HAZARD INDEX (LHI) MAP FOR CENTRAL ASIA .................................. 76
7.4 LANDSLIDE RISK MAP FOR CENTRAL ASIA ............................................................. 77
8 DISCUSSION ............................................................................................................. 81
9 CONCLUSIONS ........................................................................................................ 88
10 REFERENCES ........................................................................................................ 92
11 LIST OF PUBLICATIONS .................................................................................. 105
LIST OF FIGURES
FIGURE 1: MAJOR TYPES OF LANDSLIDE MOVEMENTS AFTER VARNES (1978) (SOURCE: CROZIER, 2013). 2
FIGURE 2: DISTRIBUTION OF LANDSLIDE-INDUCED FATALITIES WORLDWIDE BETWEEN 1915 AND 2014
(CRED, 2015). ...................................................................................................................................... 3
FIGURE 3: LOCATION MAP OF THE STUDY AREA AND SOME OF THE STRONGEST PAST EARTHQUAKES
AND SEISMICALLY-INDUCED LANDSLIDES. IN PARTICULAR, THE 1911 SAREZ (TAJIKISTAN), THE 1911
KEMIN (KYRGYZSTAN), THE 1946 CHATKAL (KYRGYZSTAN), THE 1949 KHAIT (TAJIKISTAN), THE 1989
GISSAR (TAJIKISTAN), AND THE 1992 SUUSAMYR (KYRGYZSTAN) EARTHQUAKES ARE KNOWN TO
HAVE TRIGGERED THE 1911 USOI (TAJIKISTAN), THE 1911 ANANIEVO AND KAINDY (KYRGYZSTAN),
THE 1946 CHATKAL (KYRZGYSTAN), THE 1949 KHAIT (TAJIKISTAN), THE 1989 SHARORA AND OKULI-
BOLO (TAJIKISTAN), THE 1992 BELALDY (KYRGYZSTAN) LANDSLIDES, RESPECTIVELY. ....................... 5
FIGURE 4: EXAMPLES OF SEISMICALLY-TRIGGERED LARGE SLOPE-FAILURES IN CENTRAL ASIA. TOP LEFT:
THE ANANEVO ROCKSLIDE (AFTER KEMIN EARTHQUAKE, 1911); TOP RIGHT: USOI ROCKSLIDE
(AFTER SAREZ EARTHQUAKE, 1911); BOTTOM: KHAIT ROCK AVALANCHE (AFTER KHAIT
EARTHQUAKE, 1949) (SOURCE: HAVENITH & BOURDEAU, 2010.) ..................................................... 7
FIGURE 5: TECTONIC MAP OF CENTRAL ASIA SHOWING THE ASIA-INDIA CONTINENTAL COLLISION ZONE,
AFTER MOLNAR AND TAPPONNIER, (1975) ( SOURCE: THOMAS ET AL., 2002). ................................ 8
FIGURE 6: OVERVIEW OF THE COMMON APPROACH APPLIED FOR LANDSLIDE HAZARD AND RISK
EVALUATION (SOURCE: NADIM ET AL., 2006). ................................................................................. 16
FIGURE 7: CONCEPTUAL FRAMEWORK TO CARRY OUT LANDSLIDE SUSCEPTIBILITY ANALYSIS. IN
PARTICULAR, A NUMBER OF GEO-ENVIRONMENTAL PARAMETERS (SLOPE GRADIENT, SLOPE
ASPECT, PROFILE CURVATURE, GEOLOGY, DISTANCE FROM FAULTS, AND SEISMIC INTENSITY) ARE
COMBINED WITH LANDSLIDE OCCURRENCES TO MAP LANDSLIDE SUSCEPTIBILITY. DETAILS ON THE
METHOD WILL BE OUTLINED IN CHAPTER THREE. ........................................................................... 19
FIGURE 8: OVERVIEW OF THE STEPS FOLLOWED TO CARRY OUT THE LANDSLIDE HAZARD ANALYSIS. ON
THE BASIS OF PRIOR-KNOWN LANDSLIDE SUSCEPTIBILITY, A PHYSICAL MODELING OF DOWNHILL
SLOPE FAILURES IS PERFORMED TO RETRIEVE LANDSLIDE HAZARD INDEX. .................................... 22
FIGURE 9: EARTHQUAKE-INDUCED LANDSLIDES, SICHUAN PROVINCE, CHINA, 12 MAY 2008 (SOURCE:
USGS). ............................................................................................................................................... 24
FIGURE 10: FORCE DIAGRAM OF A LANDSLIDE IN DRY, COHESIONLESS SOIL THAT HAS A PLANAR SLIP
SURFACE. W IS THE WEIGHT PER UNIT LENGTH OF THE LANDSLIDE, K IS THE PSEUDOSTATIC
COEFFICIENT, S IS THE SHEAR RESISTANCE ALONG THE SLIP SURFACE, AND Α IS THE ANGLE OF
INCLINATION OF THE SLIP SURFACE (JIBSON, 2011). ........................................................................ 37
FIGURE 11: DIGITIZATION OF GEOLOGICAL FEATURES FOR THE TERRITORY OF KYRGYZSTAN. IN THE TOP,
THE ORIGINAL GEOLOGICAL MAP IS SHOWN; IN THE BOTTOM, DIGITIZED STRATIGRAPHIC UNITS
ARE SHOWN: QUATERNARY (Q), NEOGENE-QUATERNARY (NQ), NEOGENE (N), PALEOGENE-
NEOGENE (EN), PALEOGENE (E), CRETACEOUS (K), CRETACEOUS-PALEOGENE (KE), JURASSIC-
CRETACEOUS (JK), JURASSIC (J), TRIASSIC-JURASSIC (TJ), TRIASSIC (T), PERMIAN-TRIASSIC (PT),
PERMIAN (P), CARBONIFEROUS-PERMIAN (CP), CARBONIFEROUS (C), DEVONIAN-CARBONIFEROUS
(DC), DEVONIAN (D), SILURIAN-DEVONIAN (SD), SILURIAN (S), ORDOVICIAN-SILURIAN (OS),
ORDOVICIAN (O), CAMBRIAN-ORDOVICIAN (CAO), CAMBRIAN (CA), CAMBRIAN-PROTEROZOIC
(PRCA), PROTEROZOIC (PR), ARCHEAN (AR), IGNEOUS ROCKS (IR). ................................................. 43
FIGURE 12: FREQUENCY HISTOGRAMS RELATIVE TO CLASSIFIED LANDSLIDE POTENTIAL FACTORS IN
JALAL-ABAD PROVINCE (LEFT) AND OVER ALL KYRGYZSTAN (RIGHT). IN PARTICULAR, THE
DISTRIBUTION OF CLASSIFIED VALUES FOR SLOPE GRADIENT, SLOPE ASPECT, PROFILE CURVATURE,
GEOLOGY, DISTANCE FROM FAULTS, AND SEISMIC INTENSITY IS SHOWN. ..................................... 44
FIGURE 13: LOCATIONS OF PAST LANDSLIDES FOR THE TERRITORY OF KYRGYZSTAN (SOURCE:
KALMETIEVA, ET AL., 2009). THE JALAL-ABAD STUDY AREA IS SHOWN IN BLUE. ............................ 45
FIGURE 14: SAMPLE OF LANDSLIDE LOCATIONS FOR THE JALAL-ABAD DISTRICT, SUBDIVIDED IN
TRAINING (YELLOW POINTS) AND TEST (GREEN POINTS) DATASETS ............................................... 46
FIGURE 15: DISTRIBUTION OF SLOPE GRADIENT FOR THE TERRITORIES OF KYRGYZSTAN, TAJIKISTAN AND
UZBEKISTAN, RANGING FROM 0° TO 89° AND DIVIDED INTO FOUR BINS (QUANTILE
CLASSIFICATION), 0°-6.6°, 6.6°-16.6°, 16.6°-27.5°, >27.5°. ............................................................... 47
FIGURE 16: DISTRIBUTION OF SLOPE ASPECT FOR THE TERRITORIES OF KYRGYZSTAN, TAJIKISTAN AND
UZBEKISTAN, CLASSIFIED ACCORDING TO AZIMUTH AND CORRESPONDINGLY DIVIDED INTO EIGHT
BINS, NORTH, NORTH-EAST, EAST, SOUTH-EST, SOUTH, SOUTH-WEST, WEST, NORTH-WEST. ....... 48
FIGURE 17: DISTRIBUTION OF PROFILE CURVATURE FOR THE TERRITORIES OF KYRGYZSTAN, TAJIKISTAN
AND UZBEKISTAN, CLASSIFIED (QUANTILE CLASSIFICATION) INTO FOUR BINS, -0.02507 -0.00101, -
0.00101 -0.00005, -0.00005 0.00095, 0.00095 -0.01891. ............................................................... 49
FIGURE 18: GEOLOGY MAP FOR THE TERRITORIES OF KYRGYZSTAN, TAJIKISTAN AND UZBEKISTAN,
BASED ON THE CLASSIFICATION OF STRATIGRAPHIC UNITS INTO CENOZOIC, MESOZOIC, AND
PALEOZOIC ERAS. .............................................................................................................................. 50
FIGURE 19: DISTANCE FROM FAULTS MAP FOR THE TERRITORIES OF KYRGYZSTAN, TAJIKISTAN, AND
UZBEKISTAN, PRESENTED THROUGH FOUR-BUFFER ZONE MAPS (< 1KM, 1 - 5KM, 5 - 10KM, >
10KM ). .............................................................................................................................................. 51
FIGURE 20: DISTRIBUTION OF POPULATION DENSITY FOR THE COUNTRIES OF KYRGYZSTAN, TAJIKISTAN
AND UZBEKISTAN. (SOURCE: LANDSCAN, 2012).THE MAP IS CLASSIFIED INTO FOUR BINS
(QUANTILE CLASSIFICATION), 0-1059, 1059-2137, 2137-3846, > 3846 (PEOPLE/KM2). ................... 52
FIGURE 21: DISTRIBUTION OF SEISMIC INTENSITY VALUES FOR THE COUNTRIES OF KYRGYZSTAN,
TAJIKISTAN, AND UZBEKISTAN, EXPRESSED THROUGH THE OBSERVED MACRO-SEISMIC INTENSITY
(MSK 64), AND CLASSIFIED INTO THREE CLASSES: VII, VIII, IX. .......................................................... 54
FIGURE 22: OVERVIEW OF TASKS AND RELATED TOOLS USED TO CARRY OUT THE LANDSLIDE HAZARD
ANALYSIS. IN BLUE, TOOLS WHICH WERE MADE AVAILABLE FROM THE SENSUM PROJECT, IN
YELLOW SCRIPTING TOOLS WHICH WERE DEVELOPED EX-NOVO, IN RED SCRIPTING TOOLS WHICH
WERE PREPARED TO ADAPT AVAILABLE QGIS TOOLS AND TO INTEGRATE R AND QGIS TOOLS,
RESPECTIVELY. .................................................................................................................................. 60
FIGURE 23: THE QGIS PROCESSING TOOLBOX, SHOWING SEVERAL AVAILABLE SCRIPTS. IN PARTICULAR,
THE SENSUM SET OF TOOLS WHICH ARE USED TO GENERATE SEED-POINTS FOR LANDSLIDE
HAZARD ANALYSIS IS SHOWN. .......................................................................................................... 61
FIGURE 24: DISTRIBUTION OF SOURCE-LOCATION POINTS FOR THE COUNTRIES OF KYRGYZSTAN,
TAJIKISTAN, AND UZBEKISTAN. DARK AREAS INDICATE HIGH DENSITY OF POINTS, IN AGREEMENT
WITH HIGH LANDSLIDE SUSCEPTIBLE AREAS. ON THE CONTRARY, LIGHT AREAS REPRESENT LOW
DENSITY OF POINTS, IN AGREEMENT WITH LOW LANDSLIDE SUSCEPTIBILITY LEVELS. ................... 62
FIGURE 25: EXEMPLIFICATION OF PUNCTUAL MODELING OF DOWNHILL VELOCITY. DETAILS ON THE
VALUES ASSUMED BY THE VARIABLES ARE PROVIDED IN THE TEXT. ............................................... 64
FIGURE 26: EXTRACT OF R OBJECT DATA FRAME (FIRST 10 OBSERVATIONS). “CAT” ATTRIBUTE
(STANDING FOR CATEGORY) IDENTIFIES THE POINTS BELONGING TO THE SAME FLOW LINE; “X”
AND “Y” COLUMNS CORRESPOND TO LONGITUDE AND LATITUDE, RESPECTIVELY; THE “SLOPE”
ATTRIBUTE IS EXPRESSED IN RADIANS; “ACC” AND “VEL” ARE IN M/S2 AND M/S, RESPECTIVELY;
“DS” AND “DT” REPRESENT THE DISTANCE AND TRAVEL TIME BETWEEN CONSECUTIVE POINTS OF
THE SAME FLOW LINE, IN METERS AND SECONDS, RESPECTIVELY. ................................................. 65
FIGURE 27: DISTRIBUTION OF IMPACT VELOCITY VALUES FOR KYRGYZSTAN, TAJIKISTAN, AND
UZBEKISTAN, AFTER THE INTERPOLATION AND THE APPLICATION OF THE SLOPE THRESHOLD. ..... 67
FIGURE 28: LANDSLIDE SUSCEPTIBILITY INDEX (LSI) MAPS FOR THE JALAL-ABAD STUDY AREA,
KYRGYZSTAN, BASED ON THE COMBINATIONS OF CONDITIONAL INDEPENDENT FACTORS (MODEL
A, B, C, D), AND A COMBINATION OF ALL FACTORS (MODEL E, AS OUTLINED IN
TABLE5).SPECIFICALLY, MODEL A IS DERIVED FROM THE COMBINATION OF SLOPE, ASPECT,
PROFILE CURVATURE, GEOLOGY, AND DISTANCE FROM FAULTS FACTORS, WHILE MODEL B IS
FROM THE COMBINATION OF ASPECT, PROFILE CURVATURE, GEOLOGY, AND DISTANCE FROM
FAULTS FACTORS. NORMALIZED SUSCEPTIBILITY VALUES ARE SHOWN. THE YELLOW CIRCLES
INDICATE PREVIOUS LANDSLIDE LOCATIONS (TRAINING DATASET IN FIGURE 14). ......................... 69
FIGURE 29: ACCURACY ASSESSMENT OF LANDSLIDE SUSCEPTIBILITY MODELS FOR TRAINING (A) AND
TEST (B) DATABASES, RESPECTIVELY. RECEIVING OPERATING CHARACTERISTIC CURVES (ROC) ARE
USED TO CHECK THE VALIDITY AND ACCURACY OF LANDSLIDE SUSCEPTIBILITY MODELS. ............. 74
FIGURE 30: LANDSLIDE SUSCEPTIBILITY INDEX (LSI) MAP FOR KYRGYZSTAN, TAJIKISTAN AND UZBEKISTAN
CALCULATED WITH RESPECT TO (MODEL E, TABLE 5) SLOPE GRADIENT, SLOPE ASPECT, PROFILE
CURVATURE, GEOLOGY, DISTANCE FROM FAULTS, AND SEISMIC INTENSITY FACTORS.
NORMALIZED SUSCEPTIBILITY VALUES ARE SHOWN. ....................................................................... 75
FIGURE 31: LANDSLIDE HAZARD INDEX (LHI) MAP FOR THE COUNTRIES OF KYRGYZSTAN, TAJIKISTAN,
AND UZBEKISTAN. THE MAP SHOWS HAZARD LEVEL DUE TO THE IMPACT VELOCITY OF SLOPE
FAILURES ACROSS THE REGION. NORMALIZED VALUES ARE SHOWN. ............................................. 77
FIGURE 32: POPULATION DENSITY MAP (TOP) AND LANDSLIDE HAZARD INDEX MAP (BOTTOM) FOR
KYRGYZSTAN, TAJIKISTAN AND UZBEKISTAN. A QUANTILE CLASSIFICATION SCHEME HAS BEEN
CHOSEN TO CATEGORIZE VALUES INTO 4 BINS, BEING 0 – 1059, 1059 – 2137, 2137 – 3846, > 3846
(PEOPLE/KM2), FOR POPULATION DENSITY, AND 0 – 18.39, 18.39 – 22.11, 22.11 – 24.24, 24.24 –
39.13 (M/S), FOR LANDSLIDE HAZARD INDEX MAP. ......................................................................... 78
FIGURE 33: CLASS-BY-CLASS MULTIPLICATIVE APPROACH WHICH HAS BEEN APPLIED TO PREPARE THE
LANDSLIDE RISK MAP. FIRST, EACH CLASS OF LANDSLIDE HAZARD MAP IS MULTIPLIED BY EACH
CLASS OF DENSITY POPULATION MAP (RIGHT); AFTERWARDS, VALUES ARE CLASSIFIED INTO’LOW’,
MEDIUM’, ‘HIGH’, AND ‘VERY HIGH’ LEVEL (RIGHT). ........................................................................ 79
FIGURE 34: RISK MAP OF EARTHQUAKE-INDUCED LANDSLIDES FOR KYRGYZSTAN, TAJIKISTAN AND
UZBEKISTAN. THE MAP SHOWS THE EXPECTED LEVEL OF DAMAGE DUE TO THE OCCURRENCE OF
LANDSLIDES HAVING A CERTAIN IMPACT VELOCITY ACROSS THE REGION. SPECIFICALLY, 4 LEVELS
OF RISK ARE SHOWN: LOW, MEDIUM, HIGH, AND VERY HIGH. ....................................................... 80
FIGURE 35: KHANDIZA BLOCK SLIDE SITE (UZBEKISTAN), OCCURRED IN LOESS AND PROBABLY CAUSED BY
AN EARTHQUAKE IN THE PAMIR-HINDU KUSH REGION (APRIL, 2008) ( SOURCE: NIYAZOV &
NURTAEV, 2013). .............................................................................................................................. 83
Chapter 1: Introduction
Annamaria Saponaro - January 2018 1
1 INTRODUCTION
Landslides are mass movements occurring along slopes under the influence of gravity
(Varnes, 1978). They are complex natural systems, involving different types of
materials as well as diverse types of movements. Consequently, following the
classification proposed by Varnes in 1978, but still widely accepted within the scientific
community - landslides are described by using the criteria of the type of material and
the type of movement. Specifically, the main divisions of materials are rock, debris and
earth, while movements are divided into five types: falls, flows, slides, spreads and
topples. An overview of the main types of mass movements based on both material and
movement is shown in Figure 1. The shape and size of slope movements vary because
of the combination of several factors, such as dissolution, deformation and rupture
induced by a static or dynamic load. These factors are mainly controlled by the
topography (inclination and shape of the slope), the lithology (physical and
geomechanical properties of the geological materials), the geological structure (dip,
faulting, and discontinuity of layers), the hillslope hydrology (pore pressures, water
contents) or a combination of all these factors.
Although landslides can have several predisposing causes, including geological,
morphological, physical and anthropic, they are characterized by a unique trigger
mechanism (Varnes, 1978). The trigger mechanism is defined as the external input, such
as earthquake shaking, intense or prolonged rainfall, or rapid stream erosion that causes
a near-immediate response in the form of a dislocation by rapidly increasing the stresses
or by reducing the strength of materials forming the slope surface (Wieczorek, 1996).
Chapter 1: Introduction
Annamaria Saponaro - January 2018 2
The consequence for the slopes is a reduction in the inter-particle forces and the
associated friction throughout the length of the rupture surfaces.
Throughout this dissertation, notions of landslide susceptibility, hazard and risk will be
constantly recalled: they deal with the landslide potential, the likelihood of occurrence,
and associated impact on the socio-environmental domain.
Figure 1: Major types of landslide movements after Varnes (1978) (Source: Crozier, 2013).
Landslides are well known for their devastating impact on human life, economy and
environment. According to the Centre for Research on the Epidemiology of Disasters
Chapter 1: Introduction
Annamaria Saponaro - January 2018 3
(CRED) database (Guha-Sapir et al., 2015), landslides are responsible for the loss on
average of 600 lives per year, considering the time span of the last 100 years (Figure 2),
as well as associated enormous economic damages.
Figure 2: Distribution of landslide-induced fatalities worldwide between 1915 and 2014 (CRED, 2015).
In addition, it has been shown that economic losses associated with landslides have
been rising over the past decades (Rosenfeld, 1994), mainly due to increased
development and investment in landslide-prone areas. Particularly in developing
countries, which have experienced rapid economic growth and population increase, the
exposure of people and assets to natural hazards is growing at a faster rate than risk-
reducing capacities are being strengthened, leading to increasing disaster risk (UNISDR,
2015).
Furthermore, it should be noted that the magnitude of human losses from landslides is
poorly quantified and generally underestimated (Petley, 2012). This is due to the fact
that most information related to landslide occurrences is systematically missing. In fact,
for a natural phenomenon having an impact over inaccessible terrain, collecting data is
not a trivial task; moreover, post-event databases usually classify information by trigger
0
5000
10000
15000
20000
Africa Americas Asia Europe Oceania
Number of Fatalities
time period: 1915-2014
Chapter 1: Introduction
Annamaria Saponaro - January 2018 4
mechanisms rather than cause of death. Thus, many landslide fatalities and damages are
categorized as being the result of their primary trigger event (i.e., earthquake), resulting
in the underestimation of human and economic losses associated with landslides. An
example is the devastating Wenchuan earthquake (May 12, 2008), with around 70,000
fatalities, which in turn, triggered tens of thousands of landslides (Li et al., 2013). These
mass movements obstructed the road system, and heavily impeded rescue actions by the
Chinese Government, resulting in enormous numbers of losses.
Consequently, there is an overall lack of quantification and thus appreciation of the true
impact of landslides, resulting in the poor prioritization of global-scale landslide
research and mitigation.
1.1 Landslide phenomena in Central Asia
Within the framework of landslide risk research, this dissertation specifically deals with
risk assessment of earthquake-triggered landslides in Central Asia. Central Asia is a
large geographic region including the territory of Kyrgyzstan, Tajikistan, Uzbekistan,
Turkmenistan, and Kazakhstan. However, for the purposes of this research, Central Asia
is hereby referred to Kyrgyzstan, Tajikistan, and Uzbekistan.
Seismic-triggered landslides and their peculiarities have been described and analyzed by
a number of researchers. In his pioneering work, Keefer (1984) showed that the number
of landslides and their dimensions are strongly dependent upon the magnitude of the
earthquake. Based on his investigations, the smallest earthquake magnitude that can
cause landslides is about a magnitude of four. His work also revealed that, even in areas
with abundant susceptible slopes, earthquakes with a magnitude of five typically
produce only a few landslides, whereas events with a magnitude of 7.5 will produce
thousands or tens of thousands of landslides.
It is well recognized that Central Asian countries (i.e., Kyrgyzstan, Tajikistan,
Uzbekistan) constitute a worldwide hotspot in terms of natural hazard with a specific
link between earthquakes and landslides (Nadim, 2006). Compelling evidence of the
destructive power of of secondary-triggered slope failures in Central Asia is readily
Chapter 1: Introduction
Annamaria Saponaro - January 2018 5
available (Figure 3), with landslides, mudslides and debris flows causing an extensive
number of casualties (Table 1) during, e.g., the 1911 M=8.2 Kemin earthquake in the
Kazakh/Kyrgyz border region, the 1949 M=7.4 Khait and the 1989 M=5.5 Gissar
earthquakes in Tajikistan, and the 1946 M=7.5 Chatkal and the 1992 M=7.3 Suusamyr
earthquakes in Kyrgyzstan. In addition, the 1911 M=7.6 Sarez earthquake in Tajikistan
triggered a massive landslide, blocking the Murgab river and forming the tallest
landslide river dam in the world.
Figure 3: Location map of the study area and some of the strongest past earthquakes and seismically-induced
landslides. In particular, the 1911 Sarez (Tajikistan), the 1911 Kemin (Kyrgyzstan), the 1946 Chatkal (Kyrgyzstan),
the 1949 Khait (Tajikistan), the 1989 Gissar (Tajikistan), and the 1992 Suusamyr (Kyrgyzstan) earthquakes are
known to have triggered the 1911 Usoi (Tajikistan), the 1911 Ananievo and Kaindy (Kyrgyzstan), the 1946 Chatkal
(Kyrzgystan), the 1949 Khait (Tajikistan), the 1989 Sharora and Okuli-bolo (Tajikistan), the 1992 Belaldy
(Kyrgyzstan) landslides, respectively.
According to recent surveys for Kyrgyzstan and referring to the time period between
1988 and 2007 (CAC DRMI, 2009), 18 and 27% of yearly reported disasters in
Kyrgyzstan are due to earthquakes and landslides, respectively. In particular, more than
Chapter 1: Introduction
Annamaria Saponaro - January 2018 6
300 large landslides occurred between 1993 and 2010, resulting in an average of 256
deaths per year (Torgoev et al., 2012) with substantial associated economic losses (an
average of 2.5 million USD per year).
Table 1 : List of the largest historical earthquake-induced landslides in Central Asia.
Name
Location
Date
Volume
Trigger
Number of
fatalities
Usoi
landslide
Tajikistan
1911
2.2 bilion m3
Sarez earthquake
54 people
Kaindy,
Ananevo
rock
avalanches
Kyrgyzstan
1911
15 million m3
both
Kemin earthquake
38, 0 people
Chatkal
Kyrgyzstan
1946
15 million m3
Chatkal
Earthquake
NA
Khait rock
avalance
Tajikistan
1949
75 million m3
Khait earthquake
7200 people
Sharora
landslide,
Okuli-bolo
mudslide
Takijistan
1989
5 million m3,
40 milion m3
Gissar earthquake
200, 70
people
Belaldy
rockslide
Kyrgyzstan
1992
40 million m3
Suusamyr
earthquake
35 people
Chapter 1: Introduction
Annamaria Saponaro - January 2018 7
Figure 4: Examples of seismically-triggered large slope-failures in Central Asia. Top left: the Ananevo rockslide
(after Kemin earthquake, 1911); top right: Usoi rockslide (after Sarez earthquake, 1911); bottom: Khait rock
avalanche (after Khait earthquake, 1949) (Source: Havenith & Bourdeau, 2010.)
The root of the strong geological hazard component in the mountainous areas of Central
Asia is related to a number of reasons. The region is located in the Asia-India
continental collision zone (Figure 5), where the northward-moving Indian Plate indents
the Eurasian Plate (Molnar & Tapponnier, 1975; Trifonov et al., 2002). The ongoing
collision has resulted in high mountain topography which is subject to active
deformation, contemporary faulting and frequent strong earthquakes (Gubin, 1962;
Burtman & Molnar, 1993; Pavlis et al., 1997; Sidorova, 1997; Perov & Budarina, 2000),
that combine to give rise to widespread landslide phenomena including massive rock
slope failures in both the Pamir and Tien Shan Mountains (e.g., Gaziev, 1984; Havenith
et al., 1999, 2006; Strom & Korup, 2006; Abdrakhmatov & Strom, 2006; Havenith &
Bourdeau, 2010).
Chapter 1: Introduction
Annamaria Saponaro - January 2018 8
Figure 5: Tectonic map of central Asia showing the Asia-India continental collision zone, after Molnar and
Tapponnier, (1975) ( Source: Thomas et al., 2002).
In addition, much of the topography of the region is mantled by weakly consolidated
sediments, being particularly prone to flowslides triggered by seismic shaking and/or
heavy rainfall (Gubin, 1962; Ishihara, 1989, 2012).
The described strong natural landslide hazard component would not be matter of
concern without considering the ubiquitous ability of landslide processes to impact upon
exposed assets. In this context, it has to be noted that within the last 60 years, Central
Asia has seen a growth in its population from 18 million in 1951 to more than 53
million in 2010 (Lutz, 2010). The increase of population, in combination with the
expansion of urban settlements towards landslide-prone slopes, further contributes to
the destructive impact of landslides due to relatively little investment in understanding
the hazards and risks and an added lack of appropriate resources (Rosenfeld, 1994;
Petley, 2012). The number of fatalities and damages related to landslides is hence
Chapter 1: Introduction
Annamaria Saponaro - January 2018 9
expected to increase over time, calling for the urgent need for risk assessment and
mitigation initiatives.
1.2 State of the art
Actions with the aim of understanding and controlling slope instability phenomena are
therefore necessary to implement appropriate landslide risk mitigation measures.
Authorities and decision makers who are responsible for regional land use planning are
in the constant need for maps that show areas that may be endangered by landslides. In
order to be properly managed, landslide risk must first be quantified in an objective
manner, so that results are comparable between regions.
In this context, over the last few decades, increasing attention has been paid by the
international community to the topic of landslide risk, in order to find viable solutions to
protecting hazard-prone targets (mainly population, buildings and infrastructures)
against harmful slope failures phenomena. In particular, concepts of landslide
susceptibility, hazard, and risk (which will be explained in detail in Chapter Two) have
been addressed through a wide variety of methods, according to the scale of analysis
and the aim of the investigation. Recently, several organizations and scientific
institutions have proposed guidelines for the preparation of landslide hazard and risk
maps (AGS, 2007; Fell et al., 2008; Corominas et al., 2014). In particular, a unified
terminology together with identification of the fundamental data needed to prepare the
necessary maps and guide practitioners in their analyses have been provided.
The foundations for landslide susceptibility and hazard analyses were laid by Varnes
(1984). In his work, he clarified how it is possible to identify areas where a potential for
landsliding exists by exploiting the uniformitarian principle, which states that ‘‘the past
and the present are the keys for the future’’: that is, slope failures in the future are more
likely to happen under the same conditions that led to past and current instability.
General overviews of research on the topic of landslide susceptibility can be found in
the works of Soeters and van Westen (1996), Aleotti & Chowdhury (1999), Carrara et
Chapter 1: Introduction
Annamaria Saponaro - January 2018 10
al. (1999), Guzzetti et al. (1999), Dai et al. (2002), van Westen et al. (2006), and Fell et
al. (2008).
The reliability of landslide susceptibility and hazard maps depends on the amount and
the quality of input data. Geographical Information Systems (GIS) show great promise
for meaningful and time-efficient landslide hazard and risk estimation over various
scales and data quality, also allowing the coverage of large geographical areas.
To date, numerous studies have already shown that the spatial distribution of landslides
at regional scales can be better understood through GIS-based assessments, and
successful examples of regional scale susceptibility, hazard and risk analyses can be
found in the works of Chung and Fabbri (2003), van Westen et al. (2003), Neuhäuser &
Terhorst (2007), Oh & Lee (2010), Schicker & Moon (2012), Van den Eeckhaut et al.,
(2012), Holec et al. (2013), and Dahal et al. (2013). Furthermore, many works have
already been carried out at a local scale in order to highlight the factors that control
landslide activation as well as the expected damage to exposed assets. Some examples
are Bell & Glade (2004) for Iceland, Remondo et al. (2005) for Spain, Sterlacchini et al.
(2007) for Italy, Zezere et al. (2007) for Portugal, Cassidy et al. (2008) for Norway,
Crovelli et al. (2009) for California, Jaiswal et al. (2011) for India, Mousavi et al.
(2011) for Iran, Nefeslioglu et al. (2011) and Akgun et al. (2012) for Turkey,
Andersson-Sköld et al. (2014) for Sweden. Due to difficulties in performing risk
investigations for larger areas, the number of existent landslide risk studies at a
regional-scale level is definitely lower than those achieved at local scales.
Focusing on the Central Asian region, several programs were developed during the
Soviet Union times to deal with the high level of landslide activity. Already as early as
1924, the Soviet government set up a special commission to direct landslide control
measures along the south coast of the Crimea (Sheko, 1983). Over the next 40 years,
landslide observation stations were established in many parts of the Soviet Union.
Regular monitoring of endangered areas was conducted from the 1960s until the
collapse of the Soviet Union at the beginning of the 1990s. These activities also
included extensive field-based mapping of single landslides, as well as detailed
engineering-geological investigations and their relationships to ground water conditions
and precipitation, resulting in a good understanding of local slope instabilities. The
Chapter 1: Introduction
Annamaria Saponaro - January 2018 11
main goal of these investigations had been the timely warning of the population and, if
necessary, their evacuation and resettlement. Based on a 1978 decree of the Soviet
Council of Ministers and coordinated by the State Committee on science and
technology, homogeneous landslide hazard maps have been published for those areas of
high landslide risk which had been of greatest importance for the Soviet economy.
Unfortunately, most of this previous research has been published only in Russian, hence
limiting the possibility to make it publicly available throughout the global scientific
community. Furthermore, for state security reasons, most of the figures have been
published without any related geographic location.
However, after the fall of the Soviet Union in 1991, the possibilities for landslide
investigations and monitoring were drastically reduced. Even in Central Asian countries
such as Kazakhstan, which have greater resources to devote to hazard analysis, landslide
surveys remain underfunded and adequate observation posts are lacking. In addition,
significant parts of the already existing data (e.g., maps and reports) are no longer
available or their use is limited because of the loss of accompanying information related
to methodology and data sources.
More recently, in Kyrgyzstan, landslide processes and their impacts have been again
investigated, although at the local scale. Examples can be found for the Suusamyr
(Havenith, 2006), the Mailuu-Suu (Schlögel et al., 2011; Torgoev & Havenith, 2013),
and Toktogul (Khampilang & Whitworth, 2013) regions. On the other hand, for the
Tajik territory, some relevant research has been conducted for Southeastern (Schuster
& Alford, 2004 ) and Central Tajikistan (Evans et al., 2009).
Moreover, despite the systematic cataloguing activity that was initiated in the 1990s by
the Ministries of Emergency Situations of Kyrgyzstan and Tajikistan (Havenith et al.,
2015), the available landslide hazard and risk analyses across Central Asian countries
are outdated and there is a serious for updated analyses. In addition, a sound statistical
analysis of country-wide and transboundary landslide susceptibility and hazard was not
yet been achieved.
Chapter 1: Introduction
Annamaria Saponaro - January 2018 12
With these premises, the work presented in this thesis sets out to help fill this research
gap by providing a harmonized and cross-border analysis with respect to landslide
potential and related expected damage to exposed assets.
1.3 Relevance of the work
Based on the highlighted research gap which has been described in the previous
paragraph, the principal aim of this dissertation is to develop state-of-the-art techniques
to carry out statistically-based and harmonized analyses of landslide hazard and risk for
Central Asian countries. In particular, new insights are provided regarding the following
themes:
1. By harmonizing topographic, geological, tectonic and seismic data, this work
provides a novel product, which highlights landslide potential at a cross-border
level;
2. Under conditions of data scarcity as well as of geographic remoteness – which
are of particular concern to Central Asian countries - an innovative approach is
presented that is able to run a quantitative analysis of landslide hazard and risk,
which also allows for the inference of information about landslide potential and
expected damage for areas where no data coverage is available; moreover, the
presented approach promotes the implementation of open source software
(QGIS, GRASS, R)
1
, and takes advantage of their ease of distribution, an aspect
1
Quantum GIS Development Team. Quantum GIS Geographic Information System. Open Source
Geospatial Foundation Project. http://qgis.osgeo.org.
GRASS Development Team. Geographic Resources Analysis Support System (GRASS) Software. Open
Source Geospatial Foundation Project. http://grass.osgeo.org.
R Development Core Team. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org.
Chapter 1: Introduction
Annamaria Saponaro - January 2018 13
that is particularly desirable in developing countries such as those in Central
Asia;
3. Considering that damage resulting from seismic-induced landslides is sometimes
greater than damage due to the ground motion and rupture of the earthquake
itself (Hasegawa et al., 2009), this dissertation contributes to the topic of
landslide exposure and presents a statistically-based tool to estimate the
expected destructiveness of landslide occurrences to exposed assets;
4. As will be explained in Chapter Two, interest within the scientific community to
properly address concepts related to landslide risk has recently grown in order to
enhance a constructive interaction with land-use planners and governmental
bodies. For this specific reason, this work tests the applicability of integrating
known standard statistical tools (e.g., Bayesian statistics) with a physically-
based approach to carry out landslide hazard and risk assessments;
5. Finally, considering the overall lack of appreciation of the true impact of
landslides upon society, the research hereby presented aims at filling this gap by
providing natural authorities with landslide hazard and risk products, allowing
for a better quantification of the global disaster burden associated with
landslides.
1.4 Outline of the Thesis
The thesis is organized in the following way. After the first Chapter introducing the
topic, Chapters Two and Three are dedicated to presenting the necessary theoretical
background in order to understand the conceptual and methodological framework,
respectively, in which the work is undertaken. Chapter Four provides an overview of the
data which are used to conduct the analysis. Specifically, landslide locations,
predisposing factors and related trigger mechanism, as well as exposed elements are
described. In Chapters Five and Six, the application of the statistical and physically-
based methodological frameworks in Central Asia is described. In Chapters Seven and
Eight, landslide susceptibility, landslide hazard and landslide risk maps are presented
Chapter 1: Introduction
Annamaria Saponaro - January 2018 14
and discussed, including a detailed explanation of the applied validation procedure, as
well as a discussion about the advantages and limitations related to this research work.
Finally, in Chapter Nine, the conclusions and elements related to the outlook for future
research are drawn.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 15
2 CONCEPTUAL LANDSLIDE RISK FRAMEWORK
2.1 Introduction
The research presented in this thesis was developed within the framework of landslide
risk assessment. From a general perspective, risk assessments represent a crucial basis
for decision making and mitigation purposes. When applied to the context of landslide
analysis, they are particularly supportive across a wide range of sectors, for example
construction, land-use and urban planning.
The topic of landslide risk analysis calls for a specific methodological framework where
a number of well-defined components are addressed. The objective of this Chapter is to
summarize the most relevant landslide risk components, by providing definitions,
explaining their conceptual development over the past decades, and by explaining the
existing challenges related to those concepts, as well as to provide an appropriate
context for those highlighted concepts.
Following the terminology of landslide risk assessment provided by Varnes & the
International Association of Engineering Geology Commision (1984), a landslide risk
evaluation aims to determine the “expected degree of loss due to a landslide (namely
referred to as specific risk) and the expected number of live lost, people injured, damage
to property and disruption of economic activity (referred to as total risk)”. Specifically,
the landslide risk (R) framework is composed of hazard (H), exposure (E) and
vulnerability (V) components, which are mathematically combined as follows:
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 16
,VEHR
1)
where
indicates multiplication. Based on this, a more inclusive overview highlighting
elements which are singularly addressed for quantifying landslide risk is presented in
Figure 6.
Figure 6: Overview of the common approach applied for landslide hazard and risk evaluation (Source: Nadim et al.,
2006).
As can be seen from the scheme (Figure 6), from one side it is possible to recognize a
component of landslide risk (hereby referred to as the natural component), which
mainly relates to the concepts of landslide susceptibility and triggers as a basis for the
hazard. The natural component of landslide risk is specifically linked to the spatial and
temporal probability of landslide occurrence. In the most common cases, landslide
susceptibility and, subsequentially, hazard analyses, are carried out by including
information about topography, geology, vegetation cover and soil, as well as
considering climate-related parameters. Also, information on trigger mechanisms – e.g.,
rainfall, earthquakes – is required to allow the characterization of rainfall intensity, or of
expected seismic shaking. On the other hand, the occurrence of a landslide would not be
a matter of concern if there were not a number of assets (environment, society,
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 17
infrastructures), being potentially affected and threatened by its impact. This latter
component (hereby referred to as the impact component) is typically addressed by
means of landslide exposure and vulnerability concepts, and refers to the analysis of the
exposed assets together with their propensity to suffer due to the occurrence of the
natural hazardous phenomenon of concern, in our case landslides.
2.2 Landslide susceptibility
Landslide susceptibility is defined as the probability of the spatial occurrence of known
slope failures, given a set of geo-environmental conditions (Guzzetti et al., 2005).
Through their pioneering work concerning landslide susceptibility and hazard analyses,
Varnes and the International Association of Engineering Geology Commission (1984)
demonstrated that the potential for mass movements has not a random root, but slope
failures in the future are more likely to happen under the same conditions that led to
past and current instability, in line with the uniformitarian principle. Many landslide
susceptibility analyses have been conducted worldwide. Of particular relevance for their
comprehensive nature in addressing topics of landslide susceptibility and hazard are the
works of Soeters and van Westen (1996), Aleotti & Chowdhury (1999), Carrara et al.
(1999), Guzzetti et al. (1999), Dai et al. (2002), van Westen et al. (2006), and Fell et al.
(2008). These works explain how deterministic, heuristic and statistical approaches can
be adopted to assess the potential of mass movement activation. Deterministic
approaches are based on detailed slope stability analyses to assess landslide probability
at large scales. For medium and small scale analysis, heuristic and statistical approaches
are mainly applied by considering expert opinions for estimating landslide potential
from data on preparatory variables and by developing statistical analyses of the
relationships among variables that led to slope instability in the past.
A number of challenges exist when addressing landslide susceptibility. As already
stated in the beginning of the Chapter, landslide susceptibility concerns the potential of
spatial probability of slope failures, based on the analysis of topographic, geological,
and environmental factors. The availability of information related to these specific
factors, together with knowledge about past landslide occurrences, is therefore crucial.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 18
However, complete data sets concerning geological, tectonic, land use and cover are not
easily obtainable, especially when needed for large areas. In most cases, detailed
investigations are initiated for single sites, where more insight is necessary for
engineering purposes, allowing for geological and geotechnical data to become
available at a later stage. Additionally, compiling a landslide database represents a
difficult task, requiring slope failures to be mapped and described one by one, and
distinguishing each different characteristic. Even in situations where landslide
inventories exist, their maintenance is not always attainable, due to the limited nature of
scientific research projects. Moreover, public works agencies or infrastructure
departments will take care of mapping in detail only those slope failures which have
affected, for instance, a specific segment of a road network. As an additional difficulty,
historical archives often record information concerning only major slope failures,
causing, therefore, a significant underestimation of the number of observations. Due to
these reasons, a landslide database most likely will be incomplete.
It has to be noted that, in conditions of geographical remoteness and data limitation, like
those existing for the region here targeted, the challenge of data acquisition is even
more exacerbated. Therefore, much care should be taken in order to properly collect and
prepare data sets and making a landslide susceptibility analysis realistic.
Another aspect to be considered in landslide susceptibility analyses is the fact that an
area might be subjected to more than one type of slope failure, e.g., rock falls and debris
flows. For these specific situations, a statistical analysis which develops distinctive
susceptibility models, for different landslide types, should be carried out (Fell et al.,
2008). In this context, including information concerning the triggering mechanism of
different landslide types into a landslide susceptibility analysis through a probabilistic-
based approach is a very challenging task.
Based on highlighted significant challenges in the topic of landslide susceptibility, a
procedure is hereby presented to cope with these difficulties. In particular, this research
conveys a statistically-based landslide susceptibility assessment for the investigated
region. For this purpose, a comprehensive database of past landslide locations is
compiled in order to identify areas affected by past slope instability. A GIS archive is
established in order to harmonize data regarding landslide locations, topographic
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 19
attributes, geological and tectonic factors, as well as the seismic input responsible for
landslide activation (Figure 7). Specifically, a number of geo-environmental factors
(slope gradient, slope aspect, profile curvature, geology, distance from faults), and
information concerning seismic intensity, are combined with landslide occurrences to
identify areas having a potential for landslide activation. One specific landslide type is
considered for the purposes of the analysis, allowing a distinctive susceptibility model
to be carried out. Details concerning landslide factors, landslide locations and the
applied statistical method will be outlined in Chapters Three and Four.
Figure 7: Conceptual framework to carry out landslide susceptibility analysis. In particular, a number of geo-
environmental parameters (slope gradient, slope aspect, profile curvature, geology, distance from faults, and seismic
intensity) are combined with landslide occurrences to map landslide susceptibility. Details on the method will be
outlined in Chapter Three.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 20
2.3 Landslide hazard
Landslide hazard represents the probability of occurrence over
an area within a specified period of time of a potentially damaging phenomenon
(Varnes & IAEG, 1984). It could be regarded as a
temporal extension of susceptibility. Landslide hazard is sometimes confused with
landslide susceptibility, although the temporal dimension makes a clear distinction. In
other words, landslide susceptibility might be thought as a special case of landslide
hazard, having one single temporal perspective instead of a time series (Lee & Jones,
2004).
The definition of landslide hazard has undergone some modifications over time.
Guzzetti et al. (1999) revised the definition provided by Varnes and the IAEG
Commission to include the magnitude of the landslide, with a specific link to the area,
volume and velocity of the expected slope failure.
Nowadays, this definition of landslide hazard is widely accepted throughout the
scientific community. In spite of this, there are a number of challenges related to
landslide hazard analyses. First of all, it should be remarked that compared to other
natural hazards, like earthquakes, landslides have different characteristics which make
the hazard evaluation more complex. One of the major problems in landslide hazard
assessment is the incompleteness of landslide observations. Also, establishing a clear
magnitude-frequency relation for a given landslide location is normally hampered.
Indeed, after a landslide occurs, topographic conditions are changed, and the occurrence
of the same slope failure in the same location is not likely to happen. An additional
difficulty is that, unlike for other natural hazards, for landslides no unique measure of
landslide magnitude is available (Hungr et al., 2005). For example, while for
earthquakes, magnitude is a measure of the energy released during an event, in the case
of landslides, a measure of the energy released during the failure is difficult to obtain.
Despite these difficulties, landslide hazard analyses are widely performed (in Iceland:
Bell & Glade, 2004; in Japan: Uchida et al., 2006; in Turkey: Akgun et al., 2012), and
statistically-based methods for local landslide hazard assessments, for modeling
seismically triggered shallow landslides and related run-out (travel distance)
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 21
calculations, are presented. On the other hand, landslide hazard analyses at the regional
scale are quite rare. Run-out calculations are time consuming: they require huge data
collection and integration efforts, as well as high computation times.
With these premises, this work tries to overcome the difficulty of achieving a landslide
hazard analysis over a large region, when only a limited number of landslide
observations are available. By properly integrating GIS and programming tools, slope
failures are simulated and a slope stability analysis is performed for the entire Central
Asian region.
Downhill movements of masses belong to a family of phenomena whose behavior is too
complex to accurately predict. For this reason, the use of a limited number of downhill
movements is not recommended for a reliable estimation of the overall slope stability
conditions. Instead, running a large number of simulations provides information about
the average most-likely physically possible paths along slopes. In this work, this
limitation has been overcome with the adoption of an R package, which includes a
function allowing for the generation of a high number of sampling points, based on an
inhomogeneous Poisson Point Process (e.g., Pittore, 2014).
Finally, it should be noted that unlike susceptibility, landslide hazard intrinsically
contains more elements, allowing for the incorporation of dynamic concepts directly
linked to the loss of life, injury, property damage, loss of livelihoods and services,
social and economic disruption, or environmental damage (UNISDR, 2015). Therefore,
in line with this, among the objectives of this work is the preparation of landslide hazard
maps on the basis of previously computed susceptibility maps, in order to retrieve
impact velocities of simulated slope failures, hence, deriving the hazard for exposed
assets (Figure 8). Additional details concerning the method and data used for are
provided in Chapters Three and Four, respectively.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 22
Figure 8: Overview of the steps followed to carry out the landslide hazard analysis. On the basis of prior-known
landslide susceptibility, a physical modeling of downhill slope failures is performed to retrieve landslide hazard
index.
2.4 Landslide exposure
Landslide exposure refers to the locations and characteristics of assets having a social or
economic value, such as people, buildings, engineering structures, infrastructure areas
and life lines, public service utilities and economic activities, that may be threatened by
landslide occurrences.
The purpose of exposure analysis is to identify elements at risk in areas that could
potentially be affected by natural hazard events. In other words, if a natural hazardous
event occurs in an area with no exposure, there is no risk. Exposure analysis plays a
critical role in risk assessment, also considering that the greatest influence on the output
of loss estimates from risk models arises from exposure data (UNISDR, 2015).
Practically speaking, evaluating the exposure of elements at risk means evaluating the
proportion of the assets that are located in the potential hazardous area. Although
exposure analysis is widely carried out for other natural hazard, especially earthquakes
and floods, very few studies have been devoted so far to the development of approaches
for landslide exposure.
Following the classification provided by Lee & Jones (2004), the elements at risk are
divided into these major groups:
1. Populations. This identifies the number of people which are located in the area
likely to suffer a negative impact. More detailed analysis could include age-sex
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 23
distributions, as well as an indication of the state-of-health, as these aspects
influence death rates and the nature of injuries within a population (see Section
Vulnerability in this Chapter);
2. Buildings, structures, services and infrastructures. The value of these physical
assets is usually determined from local authority tax bands;
3. Property. This includes the contents of houses, businesses and retailers,
machinery, vehicles, and personal property. Information related to the value of
these may be obtained from trade organizations and the insurance industry;
4. Activities. This group includes all activities having both a financial or social
basis, such as those linked to business, commerce, retailing, entertainment,
transportation, agriculture, manufacturing and industry, minerals, cultural,
social and recreation. Losses related to landslide occurrences are mainly linked
to the temporary or permanent disruption of these activities and, expressed in
terms of loss of revenue.
One difficulty related to exposure analysis is that the total value of the elements at risk
does not remain constant over time, and inflation, developmental growth, decline and
depreciation have to be taken into consideration (Lee & Jones, 2004). Vulnerability
itself also changes over time, due to increased initiatives for improvements in health and
safety, standards of living, housing quality, technological development.
Estimates of exposure have to take into account the following two distinct components:
1. Permanent, where fixed assets, such as buildings or pipelines, could be
damaged, irrespective of the timing of the landslide event. Assuming that these
assets are always present in the zone of impact, the adverse consequences for a
particular magnitude of event can be approximated to remain constant;
2. Temporary, where the degree of risk can vary with the timing of the event, being
it night or day, week-day or weekend, tourist season or off-season. In these
specific circumstances, the consequences will reflect the chance of the event
occurring at a time when mobile assets are either within the zone of impact or at
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 24
a relatively high level of concentration (i.e., occupancy rates for housing are
higher at night).
The effects of landslides can be very significant and vary according to geographic
location. For example, landslides in Europe can cause significant economic losses,
while in Asia, they can cause significant loss of life (Alexander, 2005).
Addressing the exposure of people when they represent a mobile asset is very
challenging. The principal difficulties are related to the assessment of the temporal and
spatial distribution of people in an urban center, together with the ethical dilemma in
quantifying the economic value of human injuries or deaths.
Figure 9: Earthquake-induced landslides, Sichuan Province, China, 12 May 2008 (Source: USGS).
Despite these difficulties, fortunately, a catastrophic disaster is not the inevitable
consequence of hazardous event, and much can be done to reduce the exposure of
populations living in areas where natural hazards occur, irrespective of frequency.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 25
For the purpose of this work, a procedure to assess the exposure for people at the
regional scale is presented, although no temporal information on past slope failures is
available for the study-area. However, the quantification of the value of the elements at
risk is not achieved in monetary terms, given the lack of this specific information.
Nevertheless, a quantitative analysis of population exposure is achieved by expressing
the proportion of persons in the zone of impact. By combining landslide hazard maps
with the distribution of population over the entire Central Asian region, it is possible to
retrieve population exposure, and subsequently identify high-exposed areas, which may
require more attention and, if specifically needed, detailed risk analyses.
2.5 Landslide vulnerability
Landslide vulnerability identifies the reaction of the assets when exposed to the
spatially variable forces produced by the hazardous event (Lee & Jones, 2004). It can be
mathematically expressed as (Einstein, 1988):
)10(,0 LLL DLDPV
2)
where
L
D
is the assessed or expected damage to an element given the occurrence of a
hazardous landslide (
L
). It is expressed on a scale from 0 (no loss) to 1 (total loss), and
is a function of the intensity of the phenomenon.
Measuring vulnerability to hazards of natural origins has been addressed in the past over
different scales and for different purposes (Birkmann, 2007). Compared to other
components of risk assessment (e.g., susceptibility, hazard), a literature review reveals
that a relatively low number of available methodologies which have been developed for
the analytical analysis of vulnerability.
Of particular concern is vulnerability analysis of exposed assets to slope instabilities, for
which a more profound research gap exists. The reason for this is linked to the intrinsic
complexity of landslide vulnerability, which depends on the landslide typology and
mechanism, as well as on the intensity of the landslide movement. Additionally, based
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 26
on a literature review, it is evident that there is no common approach used for the
assessment of vulnerability for communities prone to landslide. In particular, in
situations where landslide vulnerability has to be determined over large areas, there is
the problem of a lack of accepted standards among different investigators that can be
used for effective emergency and disaster management (Galli & Guzzetti, 2007).
Landslide vulnerability can be expressed using economic (monetary, quantitative) or
heuristic (qualitative) scales (Alexander, 2005). When using economic measurements,
vulnerability is most commonly expressed in terms of the element value, which can be
expressed by its monetary value, or by its intrinsic value, or by its utilitarian value. The
intrinsic value of human life is incalculable; however, several measures are used in
actuarial work to put a monetary value on death or injury, including the “private value”
of a statistical life (Alexander, 2005).
Most of the work on landslide vulnerability has focused on buildings, structures and
infrastructures. Some examples are the works of Hollenstein (2005), Galli & Guzzetti
(2007), Kaynia et al. (2008), Uzielli et al. (2008), Papathoma-Köhle et al. (2011),
Mavrouli et al. (2014).
When addressed heuristically, landslide vulnerability describes in qualitative terms
expected or definite damage to an element at risk. Landslides can cause not only the loss
of human lives, but they can also cause damage and temporary or permanent
malfunctioning of economic and productive activities. With reference to instability
phenomena, vulnerability typically expresses the level of loss produced in a given
element or group of elements exposed to risk resulting from the occurrence of the
natural phenomenon of a given intensity.
Damage to structures and infrastructures can be classified as (Lee & Jones, 2004):
Superficial (minor damage), where the functionality of buildings and
infrastructures is not compromised, and the damage can be repaired, quickly and
at a relatively low cost;
Functional (medium damage), where the functionality of structures or
infrastructures is compromised, with repairs taking time and significant effort;
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 27
Structural (total damage), where buildings, life lines and transportation routes
are severely damaged or destroyed, and extensive costly repairs or demolition
and reconstruction operations are needed.
While in the 1970s and early 1980s vulnerability was solely linked to physical fragility
(e.g., the likelihood of a building collapsing due to the impact of an earthquake),
nowadays the concepts of vulnerability describe the conditions of a society or element
at risk that also determine the potential or revealed hazard’s impact in terms of losses
and disruption (Birkmann, 2007).
Recently, the concept of vulnerability has been broadened towards a more
comprehensive approach including susceptibility, exposure, as well as different
thematic areas, such as physical, social, economic, and environmental vulnerability (Li
et al., 2010; Papathoma-Köhle et al., 2011; Yang et al., 2015).
Pascale et al. (2010) address landslide vulnerability of territorial systems through a
novel conceptual approach. At the first place, the following subcategories are defined:
physical, functional and systemic vulnerability. Physical vulnerability represents the
extent to which an element suffers damage from a natural phenomenon of a given
intensity; functional vulnerability represents the tendency of an element to suffer
impaired functioning due to external pressure; finally, systemic vulnerability considers
the system or territory as a whole, that is considering together people, infrastructures,
industrial plant, natural elements, etc., and their interconnections. In particular, the
concept of systemic vulnerability measures the tendency of a territorial system to suffer
damage (usually functional) due to its interconnections with other elements of the same
system. Unlike physical and functional vulnerability, the effects of systemic
vulnerability on a territorial system are not linked to the particular disaster typology in
question, but they are related to the level of interconnections between the various
elements in the system (Pascale et al., 2010). In this way, areas with greater
vulnerability within the urban fabric are identifiable, and programming strategies
against landslide risk can be better defined.
As an additional consideration, it should be noted that similarly to landslide exposure,
assessing landslide vulnerability involves a certain degree of temporal variation. In fact,
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 28
landslide vulnerability is expected to change over time, due to increased initiatives for
improvements in health and safety, standards of living, housing quality, technological
development, or the reverse due to, for example, economic problems.
In general, the risk to human life is considerably greater in developing countries,
especially in Central Asia and South America. In both regions, high levels of tectonic
activity, steep and unstable slopes, and populations concentrated in deep valleys where
rockfalls, debris slides and rock avalanches can occur suddenly and with great
devastation, emphasize the dangers involved. An additional distinction concerning
landslide vulnerability should be made with respect to fast and slow landslides.
Extremely rapid landslides may threaten life, as there is little time to react to them; in
contrast, slow and extremely slow landslide rarely threaten life, although they can
destroy structures leading to exorbitant economic costs (Alexander, 2005).
Based on the highlighted challenges and difficulties, this work tries to carry out a
vulnerability analysis of exposed assets to landslide phenomena. Considering the high
impact on population which results from mass movements in developing countries like
those in Central Asia, an approach aimed at identifying the expected level of social
damage is defined, including a link to the expected impact velocity of slope failures.
Specifically, the concept of territorial vulnerability is adapted to the study area, and
proportional dependency between population density and expected human vulnerability
is assumed. Thus, landslide vulnerability analysis is tailored to landslide occurrence, by
considering the proximity of a population to the most landslide-prone geographic areas.
2.6 Final remarks on landslide risk
Landslide risk assessments represent the first step of the landslide risk management
chain, from risk identification to risk reduction and preparedness activities. With the
objective of defining actions to reduce the negative consequences of slope failures,
landslide risk has to be at the first place identified, in a quantitative and objective
manner.
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 29
As previously explained in this Chapter, landslide hazard and risk analysis are
intrinsically complex. Like for any other natural hazard, for a comprehensive
understanding of landslide-related disasters and risk, the natural environment, together
with the social, political and economic environments, must be considered.
Although the landslide risk formula – as defined in Eq. 1 – appears relatively simple,
applying this formulation to practical situations (for example, a specific exposed
element or a given trigger mechanism), gives rise to a number of complications. In
particular, specific characteristics of the natural phenomenon causing damages and
fatalities – i.e., the spatial probability, the temporal probability, the frequency and the
intensity of landslide occurrences - have to be defined. In situations where the incidence
of more than one natural hazard is expected, a potentially substantial increase in the
system complexity also arises. In these circumstances, the definition of characteristics
related to each single natural hazard is fundamental.
In past decades, assessments of landslide risk have generally relied on the judgement
and skills of experienced geologists, engineers, and geomorphologists. Based on their
knowledge, a range of topics have been addressed, covering the recognition of the
hazard, mapping of areas having a potential for slope instability, the creation of a terrain
model and the development of approaches for the assessment of the main contributing
factors and principal causal-mechanism of failure. However, their works are mainly
related to particular areas or specific sites, and therefore a certain degree of subjectivity
in addressing landslide risk is implied. Furthermore, a consensus among different
perspectives is difficult to achieve. There is, hence, the need to introduce more rigorous
and systematic procedures to better formalize the evaluation process towards a more
‘opened’, ‘objective’ and ‘consistent’ assessment of landslide risk.
Furthermore, a certain degree of complexity arises also in relation to the definition of
the exposed assets and their vulnerability. Especially within contexts where an impact to
multiple-assets is expected, the proper characterization of each exposed element is
necessary for a reliable and unbiased risk assessment. To this respect, one should take
care of properly defining the individual elements at risk, which are typically linked to
exposed buildings, people and business infrastructures. In parallel, for these exposed
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 30
assets, the evaluation of physical, social and economic vulnerability has to be
undertaken.
An additional complexity is that risk zoning depends on the elements at risk, and hence
on their temporal-spatial probability and vulnerability. In regions where a future urban
development is expected, it should be clarified that the associated risk will change over
time.
Therefore, considering these premises, necessary criteria and issues related to the
general complexity and associated gaps in knowledge, which have been highlighted in
previous sections, the principal task of this dissertation is to accomplish a quantitative
risk assessment of earthquake-induced landslides and its components for Central Asia.
In particular, given the scope of performing a trans-border analysis, the number of
landslide risk indicators is kept relatively low and the whole system relatively simple. In
this way, it is possible to achieve a quantitatively-based analysis of landslide hazard and
risk over a large area, where only the most relevant parameters involved in landslide
processes are taken into consideration, without diminishing the significance of the
results. In addition, it has to be remarked that an assumption has been made concerning
the adopted landslide triggering mechanism. Being among the worldwide hotspots in
seismicity, the wide threat of earthquakes is recognized over the entire Central Asian
region; on the contrary, the influence of heavy or prolonged precipitations has only a
local effect. Given the final objective of providing a cross-border product, the seismic
input as the principal mechanism of landslide activation is adopted.
A quantitative transnational analysis may be a very challenging and time-consuming
task. This research work takes advantage of procedures for data sets harmonization –
with a specific link to seismic and geo-technical data - which have been conducted
during the period covering these research activities.
In the following Chapters, methodologies and data which have been adopted to
undertake the analysis for the countries of Kyrgyzstan, Tajikistan, and Uzbekistan, will
be outlined in detail. Specifically, the theoretical background to the Weights-of-
Evidence method, and of dynamic slope-stability analysis, combined with an overview
Chapter 2: Conceptual landslide risk framework
Annamaria Saponaro - January 2018 31
of landslide factors, triggering mechanisms and the distribution of population, will be
presented.
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 32
3 METHODOLOGIES
3.1 Weights of Evidence Theory
In this study, the Weights-of-Evidence method is used for generating a landslide
susceptibility map for the entire Central Asian region. The Weights-of-Evidence method
is a data–driven quantitative method used to combine evidences in support of a
hypothesis (Bonham-Carter, 1994). The method was originally developed for medical
studies and subsequently has been applied to other disciplines, e.g., identifying mineral
deposit potential (Bonham-Carter et al., 1989).
With respect to other methods, the Weights-of-Evidence method has been successfully
used in previous landslide susceptibility studies for examining the distribution and
spatial relationships of particular features. The method offers a flexible way of testing
the importance of various input factors to the potential of slope failure, providing a
simple statistical and straightforward tool that allows the calculated weights to be
interpreted. Although already carried out in large scale contexts, the method has not
been previously tested in data-scarce regions.
Within the context of landslide susceptibility analysis, the influence of landslide
potential factors (evidence) on the occurrence of landslides themselves (hypothesis) is
assessed. Weights for each landslide causative factor are calculated based on the
presence or absence of landslides within the study area.
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 33
Considering a given number of cells affected by landslide phenomena (
}{LN
), then the
prior probability of landslide occurrence
}{LP
within the studied area T is expressed as
(Bonham-Carter, 1994):
}{
}{
}{ TN
LN
LP
,
3)
where
}{TN
is the total number of cells in the studied area. This initial estimate can be
increased or decreased based on the relationships between landslide potential factors
and the occurrence of landslides. In particular, the probability of finding a landslide
potential factor within the studied area is given by:
}{/}{}{ TNFNFP
, where
}{FN
is
the number of cells in which the landslide factor is present, and
}{TN
is the total
number of cells in the studied area.
Suppose that a landslide potential factor occurs in the studied area, and that a number of
known landslides occur preferentially within the factor, it is possible to indicate the
probability of finding a landslide given the presence (
F
) or the absence (
F
) of a factor,
through the definition of conditional probabilities:
}{
}|{
}{
}{
}{
}|{ FP
LFP
LP
FP
FLP
FLP
4)
,
}{
}|{
}{
}{
}{
}|{ FP
LFP
LP
FP
FLP
FLP
5)
where
}|{ LFP
and
}|{ LFP
are the conditional probabilities of being and not being
within the factor, given the presence of a landslide.
Equations 3) and 4) can be expressed in an odds-type formulation, where the odds,
O
,
are defined as:
)1/( PPO
.
The weights for a landslide potential factor are, hence, defined as:
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 34
}|{
}|{
ln LFP
LFP
W
5)
,
}|{
}|{
ln LFP
LFP
W
6)
being
W
and
W
the positive and the negative weight, respectively.
In the Weights-of-Evidence method, the natural logarithms of both sides of the
equations are taken, resulting in:
}{ln}|{ln LO
W
FLO
7)
}{ln}|{ln LO
W
FLO
.
8)
In case the influence of several factors on the distribution of landslides is taken into
consideration, the summation of the weights of each factor is then used, provided that
these factors are mutually conditional independent. The general expression for
combining
ni ...3,2,1
landslide factors is therefore:
}.{ln}|{ln
1
321 LO
W
FFFF
LO
n
i
n
9)
If the ith factor is absent, then
W
becomes
.
W
The difference between positive and
negative weights is known as the weight contrast (
WW
C
) and provides a useful
measure of the overall spatial correlation between a certain class of factor and the
occurrence of landslides (Bonham-Carter, 1994). A positive
C
indicates that the
causative factor is present at the landslide location, and its magnitude is a measure of
the positive correlation between the presence of the causative factor and landslides. On
the other hand, a negative
C
is used to assess the importance of the absence of the
factor in landslide occurrence. Factors with contrast values around 0 have no significant
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 35
connection with the occurrence of landslides. The statistical significance of the weights
can be verified by calculating their variances (
S2
) together with the studentized
contrasts (
)(/ CSC
) by means of the following equations:
}]{/1}{/1[)(
2LFNLFN
WS
10)
}]{/1}{/1[)(
2LFNLFN
WS
11)
)()()( 222 WSWSCS
12)
A script code in R has been prepared to perform the necessary calculations.
In Weights-of-Evidence modelling, it is typically assumed that factors, which are
outlined in Chapter Four, are conditionally independent with respect to landslide
occurrences.
It can be shown that equation 9) is equivalent to:
}{
}
2
{}
1
{
}
21
{LN
L
F
NL
F
N
L
FF
N
13)
The left side of equation 13) indicates the observed number of occurrences in the
overlap zone where both
F1
and
F2
are present, while the right side represents the
expected number of occurrences in this overlap zone.
}{LN
is the number of cells
affected by landslides. A contingency table can be prepared based on this relationship
for testing the conditional independency of two factors (Table 2). Outcomes in the table
which are greater than a reference value (for this specific case, the 99% significance
level is adopted), suggest that the tested pair is not significantly different, and should
not be used together in the analysis. Details concerning the test for conditional
independence applied to each pair of landslide factors are provided in Chapter Five.
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 36
Table 2 Contingency table for testing conditional independence between Factor 1 (F1) and Factor 2 (F2)
Factor 1Present
Factor 1 Absent
Totals
Factor 2 Present
L}
FF
N{ 21
L}
F
F
N{ 2
1
L}
F
N{
2
Factor 2 Absent
L}
F
F
N{ 2
1
L}
FF
N{ 21
L}
F
N{
2
L}
F
N{
1
L}
F
N{
1
N{L}
3.2 Dynamic slope-stability analysis
This research work, far from providing a detailed slope characterization analysis, aims
at achieving a regional scale assessment of the overall stability conditions over Central
Asia slopes, part of the landslide hazard analysis. In particular, trajectories
corresponding to downhill movements over geospatial environments are modeled, and
expected run-out extensions of downhill sliding movements of predicted slope failures
are then quantified.
The first attempts at modeling the effects of seismic shaking on slopes were developed
in the early 20th century (Bell, 1900). Later formalized by Terzaghi in 1950, these
efforts mainly led to the definition of three different approach: 1) Pseudostatic analysis,
2) Stress-deformation analysis, and 3) Permanent displacement analysis.
Pseudostatic analysis typically assumes seismic shaking as a permanent body force,
which is added to a conventional static limit-equilibrium analysis, for a dry,
cohesionless slope material (Figure 10). In particular, only the horizontal component of
seismic acceleration is considered, assuming null the vertical one, to determine the
factor of safety (FS), defined as the ratio between the shear strength of the soil and the
shear stress induced on the potential surface. The pseudostatic factor-of-safety equation
is given by:
)kWcos + ]/(Wsin)tankWsin - [(Wcos = FS
14)
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 37
where FS is the pseudostatic factor of safety, W is the weight per unit length of slope, α
is the slope angle, φ is the friction angle of the slope material, and k is the pseudostatic
coefficient.
Due to its simplicity and easy of use, this method has been widely applied (Stewart et
al., 2003; Bray & Travasarou, 2009), although the assumption that the force due to
seismic shaking being constant and acting only in one direction, promoting instability,
may lead to conservative estimates (Kramer, 1996).
Figure 10: Force diagram of a landslide in dry, cohesionless soil that has a planar slip surface. W is the weight per
unit length of the landslide, k is the pseudostatic coefficient, s is the shear resistance along the slip surface, and α is
the angle of inclination of the slip surface (Jibson, 2011).
The objective of Stress-deformation analysis is to model the static and dynamic
deformation of soil, providing the most accurate picture of what actually happens in the
slope during the occurrence of an earthquake (Kramer, 1996). Over time, it has been
applied to solve site-specific problems (Clough & Chopra, 1966), although not over
regional scales.
Finally, permanent displacement methods allow for the determination of displacement
related to the sliding of a rigid block on an inclined plane, when a critical value of
acceleration is exceeded. This family of methods – which is based on the work of
Newmark (1965) – has been broadly applied in the past (Newmark, 1965; Jibson,
2011), provided that slope geometry, soil properties and earthquake strong motion
records are available. Since the effect of dynamic pore pressure is neglected, permanent
displacement methods are considered valid in case of compacted or overconsolidated
materials (Kramer, 1996).
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 38
The application of one of the above described methodologies within the context of the
landslide research in this thesis was not attainable, for the following two main reasons.
Given the focus of determining the main conditions for slope instability over a large
geographic region, the inclusion of several parameters in the modeling procedure is not
recommended. In addition, information concerning the expected seismic acceleration
involved in the stability analysis was not available, nor was that of soil properties,
leading to the necessity of adopting a simplified model procedure.
Considering that seismic stability of a slope is highly influenced by its static stability
(Terzaghi, 1950), the limit equilibrium model is assumed to approximate slope stability
conditions of earthquake-induced landslides across Central Asia. In particular, based on
the type of slope materials that can be approached at best by the model, the choice was
for disrupted and unconsolidated Quaternary and Mesozoic materials, which are typical
of the foothill areas of the study region. These slope failures are relatively small in size,
and can be thought to be characterized by conditions of dried friction as well as null
cohesion.
The limit equilibrium model (Kramer, 1996) allows the analysis of total forces acting on
a mass of rigid soil, above a potential failure surface. The main assumptions are that 1)
the soil above the failure surface is rigid, and 2) the Factor of Safety (FS) is constant
along the failure surface.
For the dynamic stability analysis, it is considered a system consisting of a sliding
punctual mass, under the action of Fn - the normal force, Fs - the sliding force, being α
the slope angle, g the gravity acceleration, and μ the coefficient of friction.
It is assumed that the rapid downhill motion of a relatively small mass is governed by
the laws of dry friction (Scheidegger, 1973; Scheidegger, 1975). Friction and slope
control the acceleration of the sliding mass and this is physically described by the
following equation:
cossin mgmgma
15)
Consequently, observed mass acceleration take on values of:
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 39
)cos(sin
ga
16)
With these premises, the downhill movement of punctual masses is modeled. At first, a
number of sources (seed points) are generated, and some flow lines corresponding to the
expected traces of each sliding mass are subsequently retrieved. The farthest points
reached by the sliding masses, identify the run-out zones.
Based on the type of studied slope failures, a proper value for the angle of friction is
chosen, in relation to common adopted values. Previous geotechnical studies (i.e.,
Hungr et al., 2005; Geotechdata.info, 2013) suggest a value of 30 degrees for a loos
sand soil, hence this value for angle of friction is selected for this work.
It has to be underlined that the approach which has been described above is developed
under a number of assumptions. Those assumptions were necessary to favor the
application of a relatively simple approach over a large area. With regard to the type of
slope failure which has been modeled, a particular case of disrupted soil slides is hereby
addressed, given their known connection with seismic triggering (Keefer, 1984). Based
on data availability, only landslides occurring in soft materials are addressed in this
work, which are a particular case of disrupted soil slides.
3.3 Advantages and innovative methodological aspects
Associated with the above methods, GIS are very powerful known tools for time-
efficient processing of multi-source data covering large areas. This work hereby takes
advantage of GIS tools, by performing a spatial analysis of landslide hazard and risk. In
particular, the presented approach promotes the implementation of open source software
(QGIS, GRASS, R), and take advantage of their ease of distribution, an aspect that is
particularly desirable in developing countries such as those in Central Asia.
The Weight-of-Evidence method offers a uniquely flexible way of testing the
importance of various input factors to the potential of landslides, providing a simple
statistical tool to interpret weights for each questioned landslide factor. On the other
Chapter 3: Methodologies
Annamaria Saponaro - January 2018 40
hand, dynamic slope-stability analysis allows for identifying areas subjected to landslide
movement on the basis of a physically modeled parameter (i.e., impact velocity). In this
way, a cross-border physically-based analysis of slope-failure conditions is achievable,
which in turn allows for identifying exposed assets to landslide runout.
Although landslide processes and their impacts have already been investigated in
Central Asia at local scales, a robust statistical analysis of country-wide landslide
hazard and risk has been achieved yet. Additionally, the Weight-of-Evidence method
has not been applied before over large areas: data which are usually adopted for regional
scale analyses are typically coarsely defined, an aspect which prevents the application
of statistically-based procedures. By properly collecting and processing geospatial data,
this work hence succeeds in applying a statistically-based framework for landslide
susceptibility analysis across Central Asia.
While achieving a statistically-based analysis at the regional scale is a very challenging
task, however, performing a physically-based analysis might be even harder. The most
innovative part of this research is the integration of statically-based analyses with a
physically-based approach, with the objective of identifying the hazard and risk
associated with the occurrence of earthquake-induced landslides across Central Asian
countries.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 41
4 DATA COLLECTION AND SPATIAL DATABASE
4.1 Introduction on data collection
Among the primary steps in any kind of landslide susceptibility or hazard assessment
are data collection and construction of a spatial database. Usually, the identification of
factors correlated with slope instability is based on the choice of physically-based
indicators (Guzzetti et al., 1999). For the present research work, data relevant to a
number of known parameters are collected, which will be described in detail throughout
this Chapter: slope gradient, slope aspect, profile curvature, geology, distance from
faults and seismic intensity are selected to perform the landslide susceptibility analysis,
while population density is subsequently used for the exposure and risk assessment
(Table 3).
An inventory of landslides in the region is compiled and used as a reference dataset. The
NASA released Shuttle Radar Topographic Mission digital elevation model, with a
spatial resolution of around 90m (SRTM, 2004), has been used to derive topographic
attributes over the entire Central Asian territory. Specifically, slope gradient, slope
aspect, and profile curvature are calculated through the terrain analysis tool in QGIS.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 42
Table 3: Overview of data sets and corresponding sources, which are used for the landslide analysis.
Data type
Source
Location of past landslides
Atlas of earthquakes in Kyrgyzstan (Kalmetieva et al., 2009)
Slope gradient
STRM Digital Elevation Model (2004)
Slope aspect
STRM Digital Elevation Model (2004)
Profile curvature
STRM Digital Elevation Model (2004)
Geology
Geological Map of Central Asian and Adjacent Areas” (Tingdong et
al., 2008)
Distance from faults
Geological Map of Central Asian and Adjacent Areas” (Tingdong et
al., 2008)
Population density
LandScan population density (Bright et al., 2012)
Seismic intensity
Probabilistic seismic hazard analysis (Bindi et al., 2012)
With respect to geology and distance from faults, features of interest have been
extracted from the “Geological Map of Central Asian and Adjacent Areas” (Tingdong et
al., 2008), by means of digitization (Figure 11).
In particular, the Jalad-Abad province, in Kyrgyzstan, has been selected as a study area,
given the wide variability of landslide factors in the area. Additionally, as can be seen
from the frequency histograms (Figure 12), the distribution of landslide factors in Jalal-
Abad area is comparable to the one concerning the entire Kyrgyz territory. For this
reason, this area is considered to be representative of the existing relationships among
landslide factors in Kyrgyzstan.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 43
Figure 11: Digitization of geological features for the territory of Kyrgyzstan. In the top, the original geological map
is shown; in the bottom, digitized stratigraphic units are shown: Quaternary (Q), Neogene-Quaternary (NQ),
Neogene (N), Paleogene-Neogene (EN), Paleogene (E), Cretaceous (K), Cretaceous-Paleogene (KE), Jurassic-
Cretaceous (JK), Jurassic (J), Triassic-Jurassic (TJ), Triassic (T), Permian-Triassic (PT), Permian (P),
Carboniferous-Permian (CP), Carboniferous (C), Devonian-Carboniferous (DC), Devonian (D), Silurian-Devonian
(SD), Silurian (S), Ordovician-Silurian (OS), Ordovician (O), Cambrian-Ordovician (CaO), Cambrian (Ca),
Cambrian-Proterozoic (PrCa), Proterozoic (Pr), Archean (Ar), Igneous Rocks (IR).
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 44
Figure 12: Frequency histograms relative to classified landslide potential factors in Jalal-Abad province (left) and
over all Kyrgyzstan (right). In particular, the distribution of classified values for slope gradient, slope aspect, profile
curvature, geology, distance from faults, and seismic intensity is shown.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 45
4.2 Landslide locations
A landslide inventory represents an essential ingredient in order to carry out landslide
hazard analysis at the regional scale (Guzzetti et al., 1999). This helps to identify the
locations of previous landslides in order to be able to predict future slope failures.
There is no agreement within the scientific community on the best technique for the
preparation of landslide inventory maps; researchers usually adopt different inventory
maps where landslides are shown as points, scarps, and seed cells (Yilmaz, 2010).
Small-scale maps may only show landslide locations (point strategy), as due to the scale
of the map, it is not possible to outline the landslide’s extent. On the other hand, large-
scale maps may distinguish between source and deposit areas (Yilmaz, 2010).
As mentioned, for the purposes of this study, we considered only one type of mass
movement, i.e., landslides occurring in soft materials, which are a particular case of
disrupted soil slide type. A selection of landslide sites is defined on the basis of
published information (Kalmetieva et al., 2009), and their distribution is mapped as
point locations (Figure 13).
Figure 13: Locations of past landslides for the territory of Kyrgyzstan (Source: Kalmetieva, et al., 2009). The Jalal-
Abad study area is shown in blue.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 46
The landslide susceptibility analysis was initially conducted for the Jalal-Abad region
over a landslide sample consisting of 1,347 landslide locations (Figure 14). Specifically,
50% of the total number of locations were randomly selected, and then used as the
“training dataset”. The remaining 50% were then used as the “test dataset” for
validating results.
Figure 14: Sample of landslide locations for the Jalal-Abad district, subdivided in training (yellow points) and test
(green points) datasets
4.3 Topographic factors: slope gradient, slope aspect, profile curvature
The stability of a slope is known to be highly dependent upon the slope gradient (angle)
and its material properties (Terzaghi & Peck, 1967). The slope (Figure 15) is presented
in degrees ranging from 0° to 89° and is divided into four bins with approximately the
same number of features (quantile classification), 0°-6.6°, 6.6°-16.6°, 16.6°-27.5°,
>27.5°.
Defined as the direction of maximum slope of the terrain surface, slope aspect is
typically taken into consideration, although in some cases its importance has been
questioned (Guzzetti et al., 1999). For the selected areas, a classification based on
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 47
azimuth being divided into eight bins, North, North-East, East, South-Est, South, South-
West, West, North-West, has been carried out (Figure 16).
Curvature represents one of the topographic attributes which is also commonly included
in landslide susceptibility analysis (Ayalew et al., 2004). In particular, the profile
curvature - defined as the second derivative of the slope with respect to the maximum
steepness direction – may help in understanding patterns of the flows’ acceleration and
deceleration, and therefore, erosion and deposition. For the selected area, profile
curvature values have been classified (quantile classification) into four bins, -0.02507 -
0.00101, -0.00101 -0.00005, -0.00005 0.00095, 0.00095 -0.01891 (Figure 17).
Figure 15: Distribution of slope gradient for the territories of Kyrgyzstan, Tajikistan and Uzbekistan, ranging from
0° to 89° and divided into four bins (quantile classification), 0°-6.6°, 6.6°-16.6°, 16.6°-27.5°, >27.5°.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 48
Figure 16: Distribution of slope aspect for the territories of Kyrgyzstan, Tajikistan and Uzbekistan, classified
according to azimuth and correspondingly divided into eight bins, North, North-East, East, South-Est, South, South-
West, West, North-West.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 49
Figure 17: Distribution of profile curvature for the territories of Kyrgyzstan, Tajikistan and Uzbekistan, classified
(quantile classification) into four bins, -0.02507 -0.00101, -0.00101 -0.00005, -0.00005 0.00095, 0.00095 -0.01891.
4.4 Geo-tectonic factors: geology, distance from faults
Lithology plays an important role in landslide susceptibility studies because different
geological units have different slope failure behaviours. For example, landslides in loess
materials have occurred in Uzbekistan, while landslide-prone slopes in Cretaceous rocks
may be found in Kyrgyzstan. For our study, geological information is obtained from
“The Geological Map of Central Asia and Adjacent Areas” (Tingdong et al., 2008),
scaled 1:2,500,000. Overall, the region is covered by a range of different sedimentary
formations, mostly dated to the Quaternary, Neogene, Paleogene, Cretaceous, Jurassic,
and Triassic. Igneous rocks related to the Palaeozoic epoch are also present.
Based on this information, stratigraphic units have been digitised from Tingdong et al.
(2008) and accordingly classified into Paleozoic, Mesozoic and Cenozoic units (Figure
18).
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 50
Figure 18: Geology map for the territories of Kyrgyzstan, Tajikistan and Uzbekistan, based on the classification of
stratigraphic units into Cenozoic, Mesozoic, and Paleozoic eras.
The presence of major lineaments is among the important factors governing the stability
of slopes (Varnes & IAEG, 1984). Tectonic structures form zones of weakness in rocks
and might accelerate the process of slope failures. In landslide susceptibility studies,
distance from lineament features (i.e., faults) is typically used to investigate any cause-
effect relationships between lineaments and landslide occurrence (Gemitzi et al., 2011;
Pradhan et al., 2010). Central Asia is covered by a large number of active faults (e.g.,
the Talas-Fergana fault). For the actual analysis, fault lines were derived from the 1:
2,500,000 scale geology map and four-buffer zone maps (< 1km, 1 - 5km, 5 - 10km, >
10km ) were prepared in GIS (Figure 19).
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 51
Figure 19: Distance from faults map for the territories of Kyrgyzstan, Tajikistan, and Uzbekistan, presented through
four-buffer zone maps (< 1km, 1 - 5km, 5 - 10km, > 10km ).
4.5 Population density
Population density is known to represent an appropriate measure to assess the exposure
and vulnerability related to earthquakes occurrences. For landslide tailored studies, the
analysis of exposure and vulnerability in small cities as well as rural communities –
which by definition have a lower population density – is particularly relevant. While
large cities often have considerable resources for dealing with natural disaster, smaller
settlements are usually more vulnerable (Cross, 2001).
In this work, the LandScan population density dataset (Bright et al., 2012) is adopted,
given that it provides the finest resolution global population distribution data available
(approximately 1 km resolution) and represents an ambient population (average over 24
hours).
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 52
Figure 20: Distribution of population density for the countries of Kyrgyzstan, Tajikistan and Uzbekistan. (Source:
LANDSCAN, 2012).The map is classified into four bins (quantile classification), 0-1059, 1059-2137, 2137-3846, >
3846 (people/km2).
From this data set, population density map for the territories of Kyrgyzstan, Tajikistan,
and Uzbekistan was prepared (Figure 20). The map is presented in values ranging from
0 to 13172 persons/km2, and divided into four bins with approximately the same
number of features (quantile classification), 0-1059, 1059-2137, 2137-3846, > 3846
(persons/km2).
4.6 Trigger mechanism: seismic ground motion
As the study area is widely and strongly affected by earthquakes, it is necessary to take
seismic ground shaking, expressed through the observed macro-seismic intensity (MSK
64), into account as a triggering factor for landslides. Within the Earthquake Model
Central Asia (EMCA) project, Bindi et al. (2012) carried out an uniform assessment of
the seismic hazard in Central Asia, mainly guided by the observed seismic histories
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 53
without any a-priori assumption on seismic zonation or model of time recurrence. The
application of such an approach to cross-border catalogues and considering intensity
prediction equations developed for the investigated area (Bindi et al., 2011), allowed a
systematic and homogeneous evaluation of the hazard to be obtained, as well as the
evaluation of the probability of exceedance of any given intensity value over a fixed
exposure time (50 years) over the entire region of interest.
For the territory of Kyrgyzstan and Tajikistan, these studies returned intensities of IX to
be expected in the future (over 50 years), while for Uzbekistan, lower intensities of VII
and VIII are expected. The main advantage of the approach followed is that it serves as
a step towards a homogenized and updated seismic input.
After importing values to GIS, a raster map was created and the resulting intensities
categorized into 3 classes: I = VII, I = VIII, and I = IX (Figure 21). Although intensity
values have not been provided at each individual landslide location, such a map
provides an overall suggestion of future intensities, and hence the potential for landslide
triggering.
Chapter 4: Data collection and spatial database
Annamaria Saponaro - January 2018 54
Figure 21: Distribution of seismic intensity values for the countries of Kyrgyzstan, Tajikistan, and Uzbekistan,
expressed through the observed macro-seismic intensity (MSK 64), and classified into three classes: VII, VIII, IX.
Chapter 5: Application of Weight-of-Evidence method
Annamaria Saponaro - January 2018 55
5 APPLICATION OF WEIGHT-OF-EVIDENCE METHOD
5.1 Test for Conditional Independency of landslide factors
As already explained in Chapter Three, the application of the Weight-of-Evidence
method requires landslide factors to be conditional independent of each other. Hence, a
chi-square test for checking the conditional independency for each possible pair of
landslide factor is performed. In particular, equation 13) is adopted to calculate the
number of cells following criteria expressed in Table 2. The chi-square values are
calculated at the 99% significance level and 1 degree of freedom, and compared with
tabulated values, where calculated chi-square values greater than 6.64 suggest that the
pairs are not significantly different. In our study, conditional dependency exists between
distance from faults and seismic intensity (
9.579
2
), while geology and distance
from faults are conditional independent of each other (
0.320
2
). This means, for
example, that the pair distance from faults-seismic intensity should not be used together
to map landslide susceptibility. On the contrary, the pair geology-distance from faults
could be combined. The whole overview of chi-square outcomes in relation to each pair
of landslide factor is provided in Table 4.
Chapter 5: Application of Weight-of-Evidence method
Annamaria Saponaro - January 2018 56
Table 4: Chi-square values for testing pair-wise conditional independency of all factors (99% significance level).
Those pairs where conditional dependency is found are highlighted in bold.
Slope
Aspect
Prof. Curv.
Geology
Dist. Faults
Seismic
Intensity
Slope
-
2.943
0.064
6.310
0.600
12.756
Aspect
-
3.115
1.008
1.801
0.741
Prof. Curv.
-
1.466
0.003
0.817
Geology
-
0.320
176.823
Dist. Faults
-
9.579
Seismic
Intensity
-
5.2 Weights’ calculation
Following the statistical approach described in Chapter Three, the landslide
susceptibility analysis is conducted for the Jalal-Abad district in Kyrgyzstan. The choice
of this test area is primarily due to the need to find a region where the distribution of
landslide factor values is representative of existing relationships for the entire country,
while preserving enough variability in these values.
The landslide susceptibility analysis is carried out over a landslide sample consisting of
1,347 landslide locations. Specifically, 50% of the total number of locations is randomly
selected from the sample, and then used as the “training dataset”. The remaining 50% is
used as the “test dataset” for validating results.
The “training dataset” of landslides is overlaid with each landslide potential factor to
calculate weights and the statistical parameters representative of existent spatial
relationships (Table 5) by applying equations 5, 6, 9, 10, 11 and 12. As can be seen, the
most noteworthy classes of parameters with a positive impact on slope instability are:
slope gradient 6° - 16.6°, north-facing slope aspect, profile curvature -0.00101(1/m) -
0.00005 (1/m), Mesozoic-aged lithologies, distance from faults greater than 10km, and
seismic intensity values (MSK 64) equal to IX. Furthermore, the highest contrasts (
)(/ CSC
) values are found for the geology factor, while lowest ones are for profile
curvature.
Chapter 5: Application of Weight-of-Evidence method
Annamaria Saponaro - January 2018 57
Table 5: Class, computed weights, variances and contrast values obtained from the application of the Weights-of-
Evidence method to the Jalal-Abad study area, Kyrgyzstan.
Factor / Class
total cells
landslide
cells
free from
landslides
cells
W
)(
2WS
W
)(
2WS
C
)(/ CSC
Slope gradient (°)
0-6.6
560,923
36
560,887
-0.970
0.028
0.098
0.002
-1.069
-6.235
6.6-16.6
1,144,613
356
1,144,257
0.608
0.003
-0.420
0.003
1.028
13.240
16.6-27.5
1,193,560
250
1,193,310
0.213
0.004
-0.109
0.002
0.321
4.017
> 27.5
1,025,605
25
1,025,580
-1.938
0.040
0.263
0.002
-2.202
-10.801
Slope aspect (°)
N (337.5 – 22.5)
353,555
84
353,471
0.339
0.012
-0.041
0.002
0.379
3.249
NE (22.5 – 67.5)
364,992
57
364,935
-0.081
0.018
0.008
0.002
-0.089
-0.642
E (67-5 – 112.5)
523,164
84
523,080
-0.053
0.012
0.008
0.002
-0.061
-0.524
SE (112.5 – 157.5)
535,069
84
534,985
-0.076
0.012
0.011
0.002
-0.087
-0.747
S (157.5 – 202.5)
561,148
67
561,081
-0.350
0.015
0.048
0.002
-0.397
-3.085
SW (202.5 – 247.5)
552,571
91
552,480
-0.028
0.011
0.004
0.002
-0.032
-0.287
W (247.5 – 292.5)
602,707
112
602,595
0.093
0.009
-0.018
0.002
0.111
1.068
NW (292.5 – 337.5)
426,817
88
426,729
0.197
0.011
-0.027
0.002
0.224
1.954
Profile curvature (1/m)
-0.02507 -0.00101
1,028,255
189
1,028,066
0.082
0.005
-0.031
0.002
0.113
1.310
-0.00101 -0.00005
934,311
176
934,135
0.107
0.006
-0.036
0.002
0.142
1.617
-0.00005 0.00095
947,914
130
947,784
-0.211
0.008
0.059
0.002
-0.269
-2.757
0.00095 0.01891
1,028,232
172
1,028,060
-0.012
0.006
0.004
0.002
-0.017
-0.187
Geology(era)
Cenozoic
1,161,722
203
1,161,519
0.031
0.005
-0.013
0.002
0.045
0.532
Mesozoic
391,100
351
390,749
1.668
0.003
-0.643
0.003
2.311
29.793
Paleozoic
2,382,883
113
2,382,770
-1.273
0.009
0.743
0.002
-2.016
-19.534
Distance from faults(km)
< 1
522,534
31
522,503
-1.049
0.032
0.095
0.002
-1.144
-6.218
1 – 5
1,586,752
200
1,586,552
-0.295
0.005
0.159
0.002
-0.455
-5.378
5 – 10
970,330
205
970,125
0.221
0.005
-0.084
0.002
0.306
3.642
> 10
834,883
230
834,653
0.487
0.004
-0.185
0.002
0.671
8.241
Seismic Intensity (I)
I = VII
1,145,958
55
1,145,903
-1.261
0.018
0.258
0.002
-1.519
-10.789
I = VIII
973,285
84
973,201
-0.674
0.012
0.149
0.002
-0.823
-7.055
I = IX
1,816,462
528
1,815,934
0.540
0.002
-0.950
0.007
1.490
15.633
Chapter 5: Application of Weight-of-Evidence method
Annamaria Saponaro - January 2018 58
5.3 Landslide susceptibility model
After the weights and statistical parameters of relevance are calculated, landslide factors
maps are re-classified according to their positive or negative correlation with landslide
locations. A landslide susceptibility zonation map is, hence, obtained by combining
previously calculated contrast values with re-classified factors maps based on:
,
1
n
j
Cij
LSI
17)
where
LSI
indicates a Landslide Susceptibility Index and
Cij
represents the contrast
for the i-th bin of the j-th factor.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 59
6 REGIONAL SLOPE-STABILITY ANALYSIS
6.1 Identification of landslide source areas: seed-points generation
The procedure for the analysis of slope-stability for the entire Central Asian region has
the ultimate objective to create an index denoting expected earthquake-induced
landslides’ destructiveness across the region. Hence, a number of statistical and spatial
analysis tools available in the R programming language, in combination with Quantum
GIS (QGIS) and GRASS tools, are adopted in order to develop a computational routine
and, subsequentially, create the landslide hazard index for the whole region (Figure 23).
The first step for the slope-stability analysis is the identification of potential source
areas, meaning those areas where slope failures are more likely to originate. This
operation is conducted on the basis of previously computed landslide susceptibility
values. Specifically, for each investigated country, the landslide susceptibility map is
used as the input for the automatized generation of source-locations (points) samples.
The generation of seed points is achieved thanks to the adoption of a collection of QGIS
scripts and models developed within the FP-7 SENSUM Project to implement
algorithms for generating Focus Maps (SENSUM project, 2014). A focus map is a
representation of the spatial “relevancy” with respect to the set of available information,
and constraints (SENSUM project, 2014). Based on spatial landslide susceptibility, it is
possible to obtain a map defining the spatial density of probability of sampling a
location given the level of susceptibility.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 60
Figure 22: Overview of tasks and related tools used to carry out the landslide hazard analysis. In blue, tools which
were made available from the SENSUM project, in yellow scripting tools which were developed ex-novo, in red
scripting tools which were prepared to adapt available QGIS tools and to integrate R and QGIS tools, respectively.
In QGIS, the Processing Toolbox environment is used to call native and third-party
algorithms by means of a simple graphical interface. In such a geoprocessing
environment (Figure 23), tools are subdivided into scripts and models, each referring to
a specific scripting type, which are automatically loaded when starting QGIS. In this
case, the body of the script is composed of R code. A detailed overview about the
installation and use of Focus Maps utilities is provided throughout the SENSUM
document Deliverable 3.4.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 61
Figure 23: The QGIS Processing Toolbox, showing several available scripts. In particular, the SENSUM set of tools
which are used to generate seed-points for landslide hazard analysis is shown.
In particular, the raster map corresponding to the landslide susceptibility map is used as
the spatial density distribution layer, and a scaling coefficient is adopted to control the
final number of seed points. Due to computational constraints, the scaling factor is set
equal to 0.1, allowing for the generation of a total of 1675156, 879899, and 3713918
points for Kyrgyzstan, Tajikistan and Uzbekistan, respectively (Figure 24). In the end, a
specific function calling for an inhomogeneous Poisson Point Process, allows for the
generation of points, based on input prior landslide susceptibility.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 62
Figure 24: Distribution of source-location points for the countries of Kyrgyzstan, Tajikistan, and Uzbekistan. Dark
areas indicate high density of points, in agreement with high landslide susceptible areas. On the contrary, light areas
represent low density of points, in agreement with low landslide susceptibility levels.
6.2 Computation of downhill flow lines
Next, source points are used to calculate surface trajectories of hypothetical punctual
masses during their downhill movements across mountain slopes. This task is achieved
by means of the r.drain GRASS plug-in combined with R coding. By taking an
elevation model as the input raster layer, r.drain is iteratively used to calculate flows
corresponding to least-cost paths at user-provided locations (a coordinate parameter is
specified). Each path is computed by choosing the steeper slope between adjacent cells.
It should be noted that r.drain currently finds only the lowest point in the input file that
can be reached through directly adjacent cells that are less than or equal in value to the
cell reached immediately prior to it; therefore, it will not necessarily reach the lowest
point in the input file.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 63
Although multiple starting points can be provided, a maximum of 1024 starting points
can be used as input to r.drain. Therefore, for each country, the large seed-points
sample is initially split into smaller subsamples, each having 900 points. Flow lines are
subsequently obtained by means of an automatized procedure through GRASS
command line scripting. Specifically, the procedure includes the following steps:
- Importing the elevation model (raster format) together with sub-samples of
starting points (vector format) as the needed input to r.drain;
- Saving r.drain output flow lines, including the assignment of an identifier (ID)
to each output;
- Extracting nodes from computed flow lines (being the row format of “line”
type);
- Adding geometric attributes to each retrieved node, mantaining the consistency
with the original projected reference system of elevation model.
6.3 Calculation of impact velocity
After having retrieved coordinates of points delineating downhill surface trajectories,
the associated impact velocity is determined, starting from observed mass acceleration
(eq. 15). It is assumed a friction angle (φ) equal to 30°, a value which is commonly
found for loose sand materials, and friction coefficient (μ) equal to tan (φ). For the
calculation of mass acceleration, α represents the slope angle and g the acceleration due
to the gravity. For each trajectory, the distance between adjacent points is computed
(ds), based on the corresponding cell sizes defined from the projected reference system:
72.921m, 77.668m, 74.574m, for Kyrgyzstan, Tajikistan, and Uzbekistan, respectively.
Let’s consider the case of a 4-point trajectory (Figure 25), calculated in Kyrgyzstan. At
point[𝑥1,𝑦1], mass acceleration is equal to 𝑎1= 𝑔(𝑠𝑖𝑛𝛼1 − 𝜇𝑐𝑜𝑠𝛼1), while its initial
velocity is null (𝑣1= 0). Once downhill movement of the point mass starts, its mass
acceleration and velocity values will correspondingly change, following Newton's laws
of motion. At point[𝑥2,𝑦2], the point mass will hence have acceleration 𝑎2=
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 64
𝑔(𝑠𝑖𝑛𝛼2− 𝜇𝑐𝑜𝑠𝛼2), and velocity 𝑣2= 𝑣1 + 𝑎1𝑑𝑡1 (given that 𝑣 = 𝑣0 +𝑎𝑡), 𝑑𝑡1
being the time to cover the distance 𝑑𝑠1(known from the Digital Elevation Model‘s
resolution: ~103m). Since 𝑑𝑠1= 𝑣1 𝑑𝑡1+ 0.5𝑎1(𝑑𝑡1)2 (given that 𝑑𝑠1= 𝑣1 𝑑𝑡1+
0.5𝑎1(𝑑𝑡1)2), it turns out that 𝑑𝑡1= √2𝑑𝑠1𝑎1
⁄ . Similarly, in order to reach the point
[𝑥3,𝑦3], the point mass will cover 𝑑𝑠2 (known from the Digital Elevation Model‘s
resolution: ~73m), with a velocity 𝑣3= 𝑣2+ 𝑎2𝑑𝑡2, and a travel time 𝑑𝑡2= (−𝑣2+
√(𝑣22)+ 2𝑑𝑠2𝑎2) 𝑎2
⁄ .
Figure 25: Exemplification of punctual modeling of downhill velocity. Details on the values assumed by the variables
are provided in the text.
Under this physical framework, the impact velocity of sliding point masses is then
calculated for each single point of the downhill retrieved flow lines. R coding has
shown a better time computation efficiency, the reason why it is used to perform all
necessary calculations. Specifically, for the entire territory of Kyrgyzstan, Tajikistan,
and Uzbekistan, a data frame spatial object including attributes for each point is
constructed, with a total number of 275876, 258779, and 933810 points, for Kyrgyzstan,
Tajikistan, and Uzbekistan, respectively. Each velocity value is calculated between
consecutive points belonging to the same trajectory (hence having, the same identifier
field).
Figure 26 shows an example of the calculations for the country of Kyrgyzstan. In this
figure, each observation refers to a prior modeled flow line identified by means of “cat”
column name, where the “cat” label stands for category. “x” and “y” attributes indicate
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 65
the longitude and latitude for each observation, while the “slope” attribute is converted
from degrees to radians, in order to make practicable the calculations in R; mass
acceleration (“acc”) of rapid moving materials is sub-sequentially calculated as a
function of slope angle, gravity acceleration, and coefficient of friction values.
Distances (“ds”) and expected travel times (“dt”) between consecutive points, are then
used to retrieve velocities values (“vel”). “ds” attribute field can take on a value either
equal to 72.921 m (vertical or horizontal path) or to 103.126 m (diagonal path), in
agreement with the Digital Elevation Model’s resolution in Kyrgyzstan. It should be
noted that, at the source-location point corresponding to the first point in flow line, the
impact velocity is null.
Figure 26: Extract of R object data frame (first 10 observations). “cat” attribute (standing for category) identifies
the points belonging to the same flow line; “x” and “y” columns correspond to longitude and latitude, respectively;
the “slope” attribute is expressed in radians; “acc” and “vel” are in m/s2 and m/s, respectively; “ds” and “dt”
represent the distance and travel time between consecutive points of the same flow line, in meters and seconds,
respectively.
6.4 Creation of the Landslide Hazard Index (LHI)
The creation of the Landslide Hazard Index (LHI) map is accomplished by means of the
following steps:
- Interpolation of velocity values over the entire region;
- Application of a slope threshold to results;
- Quantile classification of velocity values.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 66
It has to be remarked that, due to computation constraints, information concerning the
impact velocities of potential slope failures relate only to 10% of maximum achievable
sampling points (see section 6.1 for more details). For this reason, a point-by-point
physical modeling of the expected impact velocities is not obtained. However, seed-
point samples adopted for the physical modeling are properly scaled to the national
landslide susceptibility input information, and are hence considered representative of
the overall relationships between landslide potential and geometrical slope
configuration.
In order to retrieve the distribution of the impact velocities of the modeled landslides
over a regular grid of points adjusted to the Digital Elevation Model’s resolution, a
spatial interpolation over the entire region covering Kyrgyzstan, Tajikistan and
Uzbekistan is carried out. The Interpolation plugin in QGIS is used to generate an
Inverse Distance Weighted (IDW) interpolation of the impact velocity vector layer,
together with a raster layer preserving spatial extent and resolution of the map. The
computation is performed by averaging the values of data points in the neighborhood of
each processing cell, being a 10km-sized cell. The outcomes of the interpolation in
raster format show an overall distribution of impact velocities, with values ranging
between 11.692. 39.128 m/s, for the entire Central Asian region.
Afterwards, a slope threshold is applied over the entire region in order to limit velocity
values to areas where slope is greater than 6 degrees, under the assumption that slope
failures do not occur for slope gradient values lower than 6 degrees. For this specific
step, the Raster Calculation tool in GIS is adopted to multiply the output raster map of
interpolation velocity values by a user-defined raster map accounting for the slope
threshold. The distribution of impact velocity values can be visualized in Figure 27.
Finally, impact velocity values are classified into 4 sub-classes, based on the quantile
scheme.
Chapter 6: Regional slope-stability Analysis
Annamaria Saponaro - January 2018 67
Figure 27: Distribution of impact velocity values for Kyrgyzstan, Tajikistan, and Uzbekistan, after the interpolation
and the application of the slope threshold.
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 68
7 RESULTS AND VALIDATION
7.1 Landslide susceptibility results and validation
Following the approach explained in Chapter Five, landslide susceptibility maps are
created by combining weighted landslide factors maps. Specifically, maps are prepared
by combining those landslide factors which are found to exhibit conditional
independency of each other (Table 6). In this way, the influence of choosing a certain
combination of landslide factors to map susceptibility can be investigated. Additionally,
the susceptibility model obtained by considering all factors together is considered
(Table 6).
Table 6 : Four possible landslide susceptibility models of independent factors, based on the outcomes of the chi-
square test, which has been described in Chapter 5. Additionally, the model resulting by the combination of all
factors is considered.
Model A
Model B
Model C
Model D
Model E
Slope
Aspect
Prof. curvature
Aspect
Slope
Aspect
Prof. curvature
Geology
Prof. curvature
Aspect
Prof. curvature
Geology
Distance from
faults
Seismic Intensity
Prof. curvature
Geology
Distance from
faults
Geology
Distance from
faults
Distance from
Faults
Seismic Intensity
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 69
Figure 28: Landslide Susceptibility Index (LSI) maps for the Jalal-Abad study area, Kyrgyzstan, based on the
combinations of conditional independent factors (Model A, B, C, D), and a combination of all factors (Model E, as
outlined in Table5).Specifically, Model A is derived from the combination of slope, aspect, profile curvature, geology,
and distance from faults factors, while Model B is from the combination of aspect, profile curvature, geology, and
distance from faults factors. Normalized susceptibility values are shown. The yellow circles indicate previous
landslide locations (training dataset in Figure 14).
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 71
Figure 28 continued. Model E is derived from the combination of all the factors, being slope, aspect, profile
curvature, geology, distance from faults, and Seismic Intensity factors.
The result of the summation is a continuous interval of values ranging from -6.720 to
4.531 (Model A), from -3.771 to 3.503 (Model B), from -3.438 to 3.124 (Model C),
from -2.848 to 2.011 (Model D), and from -9.062 to 6.021 (Model E), indicative of
various degrees of landslide susceptibility. Normalized values are calculated by dividing
the difference between a value and the minimum result by the maximum minus the
minimum, allowing them to take on values between 0 and 1: from 0.033 to 0.955
(Model A), from 0.060 to 0.945 (Model B), from 0.087 to 0.932 (Model C), from 0.103
to 0.903 (Model D), from 0.016 to 0.971 (Model E). Afterwards, susceptibility values
are classified into 10 equal-sized sectors corresponding to different levels of
susceptibility (Figure 28).
In particular, it can be observed (Figure 28) that high LSI levels are present in the
southern area, precisely along the eastern border of Fergana valley, where slope values
mostly range from 0.0º to 16.6º, and the majority of past landslides are also distributed.
In addition, high landslide susceptibility are recognized across the Jalal-Abad province
region, though the lack of landslide observations. This latter result might indicate the
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 72
potential for landslide activation and, therefore, serve as input when estimating
landslide risk.
To check the predictive capabilities of any model, an essential requirement is to carry
out a validation of the results. Without some kind of validation, such results are useless
since they lack knowledge of the degree of confidence in the model, a crucial element
for transferring results to end users and stakeholders (Chung & Fabbri, 2003).
Cross-validation is commonly used for assessing the capability of results from a
statistical analysis to be generalized to an independent data set. The procedure consists
of partitioning a sample of data into complementary subsets. The analysis is, then,
performed on one subset – named the “training set”, and the validation carried out on
the other subset – named the “test set”.
In the field of landslide hazard assessment, the cross-validation of the results is
commonly carried out by partitioning the data in time or in space (Chung & Fabbri,
2003). When the temporal approach is chosen, landslide occurrences are subdivided into
two subsets referring to different time periods, typically named “past” and “future”
landslides. This approach is meant to construct the prediction model based on “past”
occurrences and then to validate the results with respect to “future” ones. When
temporal information is missing, spatial robustness validation is commonly applied.
The validity and accuracy of landslide susceptibility maps are typically ascertained with
the help of success- and prediction-rate curves in combination with the area under the
curves. The curves provide information about the relationship between the proportion of
area identified as being landslide susceptible and the actual landslide occurrences. In
particular, success-rate curves show how good the susceptibility model is in fitting the
already occurred landslides, while prediction-rate curves provide quantitative
information about landslides that might occur in the future. The area under the curve
provides a measure of the total accuracy based on the rate curves, where a total area
equal to one indicates perfect accuracy. The common procedure consists of sorting in
descending order the calculated index values that refer to the total number of cells in the
study area. Landslide susceptibility results are hence cross-tabulated with landslide
locations and presented as a cumulative frequency diagram.
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 73
When comparing the landslide training dataset with landslide susceptibility maps,
77.211%, 60.420%, 57.721%, 79.160%, and 68.966% of landslides are found in the
20% of highest susceptibility classes of model A, B, C, D, and E, respectively. Related
accuracy values are equal to 0.794, 0.734, 0.723, 0.686, 0.805 (Figure 29a). On the
other hand, comparing the landslide test dataset with the five landslide susceptibility
models, 75.537%, 60.299%, 58.060%, 79.104%, and 66.119% of landslides are found in
the 20% of highest susceptibility classes of model A, B, C, D, and E, respectively.
Correspondent accuracy values are equal to 0.788, 0.733, 0.715, 0.691, 0.801 (Figure
29b).
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 74
Figure 29: Accuracy assessment of landslide susceptibility models for training (a) and test (b) databases,
respectively. Receiving Operating Characteristic Curves (ROC) are used to check the validity and accuracy of
landslide susceptibility models.
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 75
7.2 Landslide susceptibility map for Central Asia
Based on the accuracy assessment, the landslide susceptibility model calibrated for
Jalal-Abad region can be considered reliable as the input landslide factors are good
indicators of existing variability conditions. In particular, Model E (Figure 29) shows
the highest accuracy (AUC = 0.800), and is hence used as the basis for extension to the
territories of Kyrgyzstan, Tajikistan and Uzbekistan. For this scope, landslide factors
are re-classified and weighted, following the above described procedure. In particular,
values are normalized and classified into 10 equal-interval classes, ranging from 0.016
to 0.971.
Figure 30: Landslide Susceptibility Index (LSI) map for Kyrgyzstan, Tajikistan and Uzbekistan calculated with
respect to (Model E, Table 5) slope gradient, slope aspect, profile curvature, geology, distance from faults, and
seismic intensity factors. Normalized susceptibility values are shown.
The resulting cross-border landslide susceptibility map (Figure 30) shows where
landslides are preferentially triggered by earthquakes within seismically active Central
Asia mountain belts. The map emphasizes the relatively high potential for landslides
over the entire territory of Kyrgyzstan, specifically along the eastern boarder of the
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 76
Fergana valley, in the South of Talas province, and in the Issyk-kul district. It has to be
noted that high levels of susceptibility can be found in the Naryn province where the
conditions for slope failures exist though the scarce occurrence of past landslides. A
general low level of landslide susceptibility can be observed for almost the entire
territory of Uzbekistan. This outcome is plausible given the prominence of flat and
desert areas. Exceptions are represented by the Tashkent and Buhkara provinces, which
are characterized by high and medium levels of landslide susceptibility, respectively
(Figure 30). With regard to the Tajik territory, high levels of landslide susceptibility are
expected in the central districts of Tojikobod and Nurobod (Figure 30), in the proximity
to the devastating 1949 Khait (Evans et al., 2009).
7.3 Landslide hazard index (LHI) map for Central Asia
This section presents results relative to landslide hazard analysis, obtained by following
the procedure described in Chapter Six.
The Landslide Hazard Index (LHI) map (Figure 31) indicates the expected level of
hazard due to the activation of slope failures across the Central Asian region. The map
shows a relatively high level of hazard where the slope gradient is also relatively high.
In particular, highest hazard values are mainly shown along the Alay Range and the
Pamir, in Kyrgyzstan and in Tajikistan; relatively high hazard values can also be
observed along the North and Central Tien Shan in Kyrgyzstan, in the Pskem
Mountains in Uzbekistan.
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 77
Figure 31: Landslide Hazard Index (LHI) map for the countries of Kyrgyzstan, Tajikistan, and Uzbekistan. The map
shows hazard level due to the impact velocity of slope failures across the region. Normalized values are shown.
7.4 Landslide risk map for Central Asia
The risk analysis is achieved by combining the population density map with the
landslide hazard index map. As far as the vulnerability component is concerned, a
binary function is considered so that vulnerability takes on a value equal to ‘0’, in
absence of exposure, and a value equal to ‘1’, in presence of exposure. Both the
population density and landslide hazard index maps are initially classified into 4
subcategories, following a quantile classification scheme (Figure 32).
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 78
Figure 32: Population density map (top) and landslide hazard index map (bottom) for Kyrgyzstan, Tajikistan and
Uzbekistan. A quantile classification scheme has been chosen to categorize values into 4 bins, being 0 – 1059, 1059 –
2137, 2137 – 3846, > 3846 (people/km2), for population density, and 0 – 18.39, 18.39 – 22.11, 22.11 – 24.24, 24.24 –
39.13 (m/s), for landslide hazard index map.
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 79
Afterwards, a class-by-class multiplicative method is used to obtain the final landslide
risk map. Hence, class 1 of landslide hazard index is multiplied by class 1 of population
density returning a value equal to 1, class 1 of landslide hazard index is multiplied by
class 2 of population density returning a value equal to 2, and so on. Final values are
visualized in form of a matrix, where values ranging between 1 and 2 (shown in yellow)
indicate ‘low level’, values ranging between 3 and 6 (shown in light orange) indicate
‘medium level’, values ranging between 8 and 9 (shown in dark orange) indicate ‘high
level’, and finally values ranging between 12 and 16 (shown in red) indicate ‘very high
level’ (Figure 33). As part of the classification, proportional dependency between
population density and expected people vulnerability is assumed, which in turn results
in assigning a high level of risk to highly populated areas.
Figure 33: Class-by-class multiplicative approach which has been applied to prepare the landslide risk map. First,
each class of Landslide Hazard Map is multiplied by each class of Density Population Map (right); afterwards,
values are classified into’low’, medium’, ‘high’, and ‘very high’ level (right).
The final outcome is, hence, a map showing 4 different levels of risk due to landslide
occurrences across the territory of Kyrgyzstan, Tajikistan and Uzbekistan (Figure 34).
As can be seen from the map, a very high level of risk is locally expected in the Jalal-
Abad and Osh provinces, Kyrgyzstan, in the Gorno-Badakhshan region near the urban
areas of Khorugh and Murghab, Tajikistan, and in the Kashkadarva region, Uzbekistan.
Moreover, a high level of risk can be observed for several areas, i.e., those located in the
Chapter 7: Results and validation
Annamaria Saponaro - January 2018 80
proximity of Fergana Valley and Jalal-Abad areas in Kyrgyzstan, being highly densely
populated areas. A medium level of risk can be identified in the Tashkent province in
Uzbekistan. Although the urban area of the city is not at risk, a certain level of risk is
expected in the surrounding area proximal to the Pskem Mountains, due to the
contribution of vulnerability in the final risk calculation. On the other hand, there is a
low-medium level of risk extensively distributed across Kyrgyzstan, Tajikistan, and
Uzbekistan, primarily due to the presence of small settlements (having a population
density in the order of 1000 persons/km2) exposed to significant landslide hazard.
Besides, areas in white correspond to bodies of water and unpopulated areas, mainly the
mountain regions of the North and South Tien Shan, the Pamir, and the Kyzylkum
desert, for which the risk of having landslide-induced damage is null.
Figure 34: Risk map of earthquake-induced landslides for Kyrgyzstan, Tajikistan and Uzbekistan. The map shows the
expected level of damage due to the occurrence of landslides having a certain impact velocity across the region.
Specifically, 4 levels of risk are shown: low, medium, high, and very high.
Chapter 8: Discussion
Annamaria Saponaro - January 2018 81
8 DISCUSSION
The analysis of landslide susceptibility results reveals that the most influential factors to
slope instability, sorted according to their values of contrast, are: the class of Mesozoic
materials, the class IX of seismic intensity, the class 6.6° – 16.6° of slope gradient,
distance from faults greater than 10km, the class 292.5° – 22.5° of slope aspect, and the
class -0.00101 -0.00005 of profile curvature. In addition, the classes linked to the
highest slope stability probabilities are: the class of Paleozoic materials, the class of
slope gradient greater than 27.5º, the class VII of seismic intensity, distance from faults
less than 1km, the class 157.5º – 202.5º of slope aspect, and the class -0.00005
0.00095 of profile curvature.
Precisely, relatively high slope values do not imply landslide occurrence, in agreement
with the existing evidence (Havenith et al., 2006) that most landslides in Kyrgyzstan
occur on relatively low slope angles (< 20°). Clearly, a relatively low slope angle is
associated with the presence of soft materials due to their mechanical properties.
With regard to slope aspect, small contrast values clearly indicate the partial
contribution of this factor in the landslide susceptibility analysis. Nevertheless, the
presence of southwest monsoon winds, might make south-facing slopes relatively wet
and undisturbed, while leaving north-facing slopes drier, less vegetated, and,
consequently more exposed to landslide phenomena. In agreement with this statement,
similar evidences have already been provided for Central Asia (Strom, 2013).
Chapter 8: Discussion
Annamaria Saponaro - January 2018 82
Most of past landslides are linked to negative values of profile curvature, which
correspond to convex-shaped slopes. Although having small contrast values, the control
of convex morphologies to slope instability can be explained with local seismic
amplification phenomena occurring in topographic ridges.
The influence of geology is also clear, since there is an increase in the contrast values
from the older geological units to the more recent ones. This agrees with the fact that,
especially in the northern Kyrgyzstan (Kalmetieva et al., 2009), most landslides are
found on Quaternary materials. A relatively high degree of rock fracturing might be
responsible for the occurrence of slope failures in Mesozoic rocks, as revealed by the
high contrast values. On the other hand, it is clear that Palaeozoic materials have no
connection with landslide initiation. It has to be remarked that the distribution of
geological materials is not uniform throughout Central Asian countries. For example, a
relatively lower level of landslide susceptibility is found in Tajikistan with respect to
Kyrgyzstan. In fact, the hereby proposed susceptibility model entitles Mesozoic rocks as
the most prone to landslides. However, in Tajikistan the majority of rocks are Paleozoic
materials (in particular fractured igneous rocks) that, based on the calibrated model, are
classified as “not influent”.
Additionally, the widespread presence of loess lithology both in Tajikistan and
Uzbekistan should not be underestimated, considering connections to previous slope
failures (Evans et al., 2009, Niyazov & Nurtaev, 2013). The fact that both fractured
igneous and loess materials are not explicitly addressed in the analysis might reflect the
relatively low susceptibility level found in these countries.
At a relatively large distance from the faults (> 10km), an increase of landslide
occurrences can be observed. The presence of relatively deep hypocentres might be the
reason for a considerable surficial distance between fault lines and landslides, even
beyond 10km. This evidence, in line with what has been previously shown (Gemitzi et
al., 2011), confirms the influence of neotectonic lineaments and fault density on
landsliding phenomena. In addition, a connection between landslide occurrences and
focal mechanisms of earthquakes might be speculated. While a relatively high number
of landslides are expected in case of strike-slip mechanisms (being rocks materials more
Chapter 8: Discussion
Annamaria Saponaro - January 2018 83
fractured in the proximity of the epicenter), the presence of slope failures at large
distances from faults might be connected to reverse focal mechanisms.
Figure 35: Khandiza block slide site (Uzbekistan), occurred in loess and probably caused by an earthquake in the
Pamir-Hindu Kush region (April, 2008) ( Source: Niyazov & Nurtaev, 2013).
As far as the seismic input is concerned, earthquakes of magnitude around 7 are
expected in order to approximate the identified seismic intensity values of VIII and IX.
According to observed seismicity, it is possible to find evidence of past earthquakes
with magnitude of 6.8 - 7 having an impact on the investigated areas. It has to be
underlined that, despite alternative seismic input parameters are possible, in this case,
the intensity assigned at the site accounts for all the possible combination of magnitude
and distance determining ground shaking at the site. Therefore, for the purpose of the
actual landslide susceptibility analysis, seismic intensity is for sure more reliable than
other energy-related parameters like PGA.
Chapter 8: Discussion
Annamaria Saponaro - January 2018 84
In reference to landslide hazard results, it must be noted that impact velocities derived
from physical modelling turn out to be relatively higher that those typically derived for
the same materials (Varnes, 1984). At the first place, this result may be related to the
assumption of rapid downhill motion of relatively small masses (Scheidegger, 1973),
which is at the root of the physical modeling framework. In addition, the contribution of
sudden loading which is typically occurring during an earthquake might be the reason
behind extremely rapid velocities, even greater than 5 m/s (Hungr et al., 2005). Such
slope failures usually involve loose granular materials overlying stable substrate, and
often originate during heavy rain, when the perched saturation condition of the loose
layer is reached.
One limitation of the landslide susceptibility model hereby presented is that rockslides
and large slope failures are not addressed. This is linked to the choice of developing a
susceptibility model by using only landslides occurred in soft materials, due to their
majority.
Unfortunately the available information prevents us to clearly distinguish between the
trigger mechanisms causing those landslides used to develop the susceptibility model.
However, considering the local influence of monsoons, slope failures occurring in the
Fergana Area might have been triggered by heavy or prolonged precipitations.
Even though conventional landslide susceptibility analyses do not incorporate triggering
information (Fell et al., 2008), the present study considers also the inclusion of the
seismic input, in line with susceptibilities analyses previously presented by Schicker &
Moon (2012), who included rainfall, and Holec et al. (2013), who included both
seismicity and rainfall. Besides, given our focus on investigating the potential for
landslide activation over a large territory, only seismicity has been considered as the
triggering mechanism. While having an impact on slope instability, the effect due to
precipitations might only have a local influence and, therefore, is out of the scope of this
study.
Based on the analysis of accuracy values for susceptibility models (Figure 29), it can be
seen that most of susceptibility models show AUC values greater than 0.70 and can,
therefore, be accepted as significant. Model E has finally been chosen as applicable to
Chapter 8: Discussion
Annamaria Saponaro - January 2018 85
the entire country, being the most accurate one. Given that the condition of total
independency among factors in never completely verified in nature (Bonham-Carter,
1994), we have not excluded the combination of those factors known as the most
relevant to landslide initiation in Kyrgyzstan, similarly to what has been carried out by
Dahal et al. (2008) and Pradhan et al. (2010) in Nepal and Malaysia, respectively.
Moreover, for regional scale analyses of landslide susceptibility, overestimating the
number of predicted landslides is a common problem. In order to tackle this problem,
the adoption of one single landslide point per unit area is considered in the present study
(Neuhäuser & Terhorst, 2007).
An important issue in landslide susceptibility and hazard studies is represented by the
influence on the final susceptibility values of transforming continuous variables into
discrete variables. In this respect, Remondo et al. (2003) demonstrated how the
predictive capability of validation curves obtained from input data, which were
classified into only a few intervals, and the outcome from almost continuous variables,
is quite similar. Based on this consideration, landslide potential factors have been
classified in such a way as not to have many classes in the susceptibility analysis. As a
general observation, it can be stated that increasing the number of classes in landslide
factors leads to unstable results.
Overall, the operation of classifying susceptibility and hazard values is an ongoing topic
of debate within the scientific community, given that there are no reference rules on
categorizing data (Ayalew et al., 2004). In this study, susceptibility and hazard values
are at first normalised and afterwards classified, following the quantile classification
scheme among a number of possible alternatives (i.e., natural breaks, equal size). The
reason behind this choice is to have each class approximately equally represented on the
final map. Moreover, quantiles are very useful for ordinal data, since the class
assignment of quantiles is based on ranked data. Finally, by adopting the same number
of classes, a comparison among landslide susceptibility and hazard maps is achievable,
together with a statistically reproducible framework.
Moreover, landslide hazard results are clearly linked to slope geometry as well as to
landslide susceptibility. In fact, most of expected hazardous areas correspond to areas
Chapter 8: Discussion
Annamaria Saponaro - January 2018 86
either having high slope or high susceptibility. For this reason, landslide hazard index,
far from being of site-specific nature, has to be considered as a broad indicator of the
expected impact velocity due to mass movements at a national level. Unfortunately, the
lack of historical information concerning landslide-related damages over a large area
prevents to carry out a validation of both landslide hazard and risk results. Although
only sparse, a number of available local landslide datasets allow for a validity check of
results, and confirm the potential for landslide activation throughout the region.
Besides, historical observations of strongest seismically-induced landslides occurred in
Central Asia (Figure 3), are found inside identified most landslide-prone areas. It has to
be remarked that, these datasets were not used in the statistical analysis because of their
limited nature.
Regarding landslide risk results, it has to be remarked that the physical damage to life
lines induced by slope failures is not specifically addressed, and therefore, the total level
of risk might be underestimated.
As has already been pointed out (Das, 2011), the validation of risk results is a very
difficult task to be achieved, given high uncertain in the estimates. The work hereby
presented, by addressing one single landslide type together with expected impact only to
exposed population, provides a relatively simple framework allowing results’ uncertain
to be relatively low. The choice of population density as element at risk has a particular
meaning in Central Asian countries, where fatalities due to landslides are more
significant than economic losses.
An essential part of any landslide hazard and risk assessment is the prediction of the
character of failure and a quantitative estimate of post-failure motion (“runout”)
including travel distance and velocity (Hungr et al., 2005). In line with this, the
computation framework presented in this work allows not only the computation of
impact velocity but also to infer information about runout distances of simulated slope
failures.
Although various methods to carry out quantitative landslide hazard and risk analyses
are available, applications are still rare and mainly dependent on the occurrence of
disasters. In order to support this research direction, the hereby proposed method
Chapter 8: Discussion
Annamaria Saponaro - January 2018 87
demonstrate the applicability of a Bayesian-based procedure for detecting landslide-
prone areas at the regional scale, and for identifying the most important factors inducing
slope failures. Moreover, a quantitative-based landslide hazard and risk framework
allowing a prior assessment of the expected destructiveness due to landslide activation
is provided. In this way, landslide-prone areas are identified in advance, and the
occurrence of disasters might hence be limited. Reinforcing buildings located within the
landslide susceptible areas would be high costly, and therefore impracticable on an
economic perspective. It is then of vital importance that land planners are provided with
appropriate information and tools, which allow the location of people, buildings and
main infrastructures being more vulnerable to slope failures.
Chapter 9: Conclusions
Annamaria Saponaro - January 2018 88
9 CONCLUSIONS
Central Asia is one of the most challenging places in the world where various natural
hazards can heavily injury populations and resources. Among these hazards, landslides
pose a serious threat to human life and human facilities. Focusing on landslide-related
disasters after they occur is essential from a humanitarian point of view, but
unfortunately not sufficient for reducing their tragic consequences to people,
infrastructures and the environment. Furthermore, considering remote conditions
characterizing Central Asian countries, data are not easily accessible, or limited in
nature, or may be affected by different sources of uncertainty. For this reason, collecting
of resources and knowledge for identifying areas for future landslide activation and
quantifying vulnerability and risk are crucial tasks for long-term risk mitigation.
In order to mitigate landslide risk, this work evaluates its components over the entire
Central Asian region. For the first time, a cross-border risk map of earthquake-induced
landslides is produced at a transnational level in Central Asia. To this scope, an
approach to evaluate the potential of seismically-induced landslides using statistical
relationships between past landslides and the most significant seismo-tectonic,
geological and morphological factors in Central Asia countries is presented. The
Bayesian-based method is initially calibrated and cross-validated with an independent
dataset in Kyrgyzstan, providing with a landslide susceptibility model having an
accuracy level greater than 70%, which allows considering the model sufficiently
reliable for urban planning purposes. Afterwards, due to uniformity in
geomorphological and tectonic factors characterizing all Central Asian countries, an
Chapter 9: Conclusions
Annamaria Saponaro - January 2018 89
extension of the model to the territories of Uzbekistan and Tajikistan is carried out. At a
second stage, the susceptibility analysis is used as prior information for a quantitative
hazard analysis and to evaluate expected damage to people due to the sliding process
induced by earthquakes, over the entire Central Asian region.
It can be concluded that geology plays a critical role in guarantying slope stability.
Mesozoic materials are found to be the most responsible for landslide initiation. The
huge variability of these materials (soft and semi-hard rocks, principally deposits of
clays, argillites, sandstones, limestones - often covered by Quaternary loess), prevents
us to discriminate the influence of specific rock types, which would add more value to
the analysis. The contribution of rock structure to instability is also clear given the fact
that most of landslides occurred in places with relatively moderate slope gradient values
(6-16°). Additionally, due to the presence of neotectonic lineaments all over the
country, the strength of rock materials is reduced and slopes are made unstable. It has
been observed that, at a certain distance from faults, slope failure phenomena are quite
sever. The resulting cross-border landslide susceptibility map emphasizes the relatively
high potential for landslides over the entire country of Kyrgyzstan, specifically along
the eastern boarder of Fergana valley, in the South of Talas province, and in Issyk-kul
district; besides, high levels of susceptibility can be found in the Naryn province where
the conditions for slope failures exist though the scarce occurrence of past landslides. A
general low level of landslide susceptibility can be observed for almost the entire
territory of Uzbekistan, with exceptions of the Tashkent and the Buhkara provinces
characterized by high and medium landslide susceptibility, respectively. With regard to
the Tajik territory, high levels of landslide susceptibility are expected in central districts
of Tojikobod and Nurobod.
Besides, it turns out that impact velocity of earthquake-induced mass movements can be
particularly significant along the North and Central Tien Shan, the Pskem Mountains,
the Alay Range, and the Pamir. In addition, landslide risk results show that a very high
level of risk is locally expected in the Jalal-Abad and Osh provinces, Kyrgyzstan, in the
Gorno-Badakhshan region near the urban areas of Khorugh and Murghab, Tajikistan,
and in the Kashkadarva region, Uzbekistan. Moreover, a high level of risk can be
observed for several areas, i.e., those located in the proximity of Fergana Valley and
Chapter 9: Conclusions
Annamaria Saponaro - January 2018 90
Jalal-Abad areas in Kyrgyzstan, being highly densely populated areas. On the other
hand, there is a low-medium level of risk extensively distributed across Kyrgyzstan,
Tajikistan, and Uzbekistan, primarily due to the presence of small settlements (having a
population density in the order of 1000 persons/km2) exposed to significant landslide
hazard.
By taking into account limitations and assumptions of the described approach, results
hereby presented can be effectively used to model landslide susceptibility, hazard and
risk also in neighboring regions as well as in other data-scarse regions of the world,
being characterized by the same combination of landslide influential parameters as
Central Asia countries.
As part of future work, a more detailed investigation of existing relationships between
quaternary rocks (in particular loess) and slope failures is recommended in order to
provide specific insight for landslide susceptibility in Tajikistan and Uzbekistan. For
this reason, a more detailed inventory of past landslides is under construction, including
geological information at the landslide site. Furthermore, a more detailed study of the
connections between faults and earthquake source mechanisms would provide a deeper
characterization of the seismic ground shaking as input to landsliding phenomena.
Finally, collecting information on damages caused by landslides, including not only the
number of fatalities but also the number of people affected and injured, would definitely
increase the understanding of human vulnerability to landslides. Moreover, the analysis
of physical vulnerability to slope failures in Central Asia might be undertaken, provided
that information about induced-damages to buildings and life-lines are made available.
To this scope, clear dependencies between the impact area of a landslide and the amount
of damage to built-up areas and infrastructures can be established.
Although landslide studies have been already started in Central Asia, a sound statistical
methodology for susceptibility of earthquake-induced landslides and risk mapping
applicable over the entire region at transnational level has never been carried out before.
By identifying areas with a potential for future seismically-induced slope instability
together with expected impact on population, the present work offers the unique value
of presenting first attempts of a cross-border and harmonized landslide analysis, and
Chapter 9: Conclusions
Annamaria Saponaro - January 2018 91
provides national authorities with a regional map which serves as a prelude for landslide
risk mitigation activities.
Chapter 10: References
92
10 REFERENCES
Abdrakhmatov, K., & Strom, A. (2006). Dissected rockslide and rock avalanche
deposits; Tien Shan, Kyrgyzstan. In Evans, Mugnozza, Strom, & Hermanns,
Landslides from massive rock slope failure , 551-570. Netherlands: Springer.
AGS. (2007). Guideline for Landslide Susceptibility, Hazard and Risk Zoning. In
Australian Geomechanics (42, 1), 13-36.
Akgun, A., Kincal, C., & Pradhan, B. (2012). Application of remote sensing data and
GIS for landslide risk assessment as an environmental threat to Izmir city (west
Turkey). Environ. Monit. Assess. (184), 5453-5470.
Aleotti, P., & Chowdhury, R. (1999). Landslide hazard assessment: summary review
and new perspectives. Bull. Eng. Geol. Environ. (58), 21-44.
Alexander, D. (2005). Vulnerability to Landslides. In Glade, Anderson, & Crozier,
Landslide Hazard and Risk ,175-198.
Andersson-Sköld, Y., Falemo, S., & Tremblay, M. (2014). Development of
methodology for quantitative landslide risk assessment - Example Göta river
valley. Natural Science (6,3), p. 14.
Chapter 10: References
93
Ayalew, L., Yamagishi, H., & Ugawa, N. (2004). Landslide susceptibility mapping
using GIS-based weighted linear combination the case of Tsugawa area of
Agano River. Niigata Prefecture Japan. Landslides (1), 73-81.
Bell, J. (1900). General slope stability analysis . J Soil Mech Found Div (92 SM5).
Bell, R., & Glade, T. (2004). Quantitative risk analysis for landslides - Examples from
from Bíldudalur, NW-Iceland. Nat. Hazards Earth Syst. Sci.(4), 117-131.
Bindi, D., Abdrakhmaov, K., Parolai, S., Mucciarelli, M., Grüntal, G., Ischuk, A., . . .
Zschau, J. (2012). Seismic hazard assessment in Central Asia: outcomes from a
site approach. Soil. Dyn. Earthq. Eng. (37), 84-91.
Bindi, D., Parolai, S., Oth, A., Abdrakhmatov, K., Muraliev, A., & Zschau, J. (2011).
Intensity prediction equations for Central Asia. Geophys. J. Int. (187), 327-337.
Birkmann, J. (2007). Risk and vulnerability indicators at different scales: Applicability,
usefulness and policy implications. Environmental Hazards (7), 20-31.
Bonham-Carter, G. F. (1994). Geographic Information System for Geoscientists.
Ottawa: Pergamon Press.
Bonham-Carter, G., Agterberg, F., & Wright, D. (1989). Weights of evidence
modelling: A new approach to mapping mineral potential. Statistical
applications in the Earth Sciences, geological survey of Canada, 171-183.
Bray, J., & Travasarou, T. (2009). Pseudostatic coefficient for use in simplified seismic
slope stability evaluation. J. Geotech. Geoenviron. Eng.(135), 1336-1340.
Bright, E. A., Coleman, P. R., Rose, A. N., & Laboratory, O. R. (2012). LandScan
Global Population Database. Oak Ridge National Laboratory, UT-Battelle,
LLC.
Burtman, V., & Molnar, P. (1993). Geological and geophysical evidence for deep
subduction of continental crust beneath the Pamir. Geol. Soc. Am. (281), 1-76.
Chapter 10: References
94
CAC-DRMI. (2009). Central Asia and Caucasus Disaster Risk Management Initiative.
Desk Study Review.
Carrara, A., Guzzetti, F., Cardinali, A., & Reichenbach, P. (1999). Use of GIS
technology in the prediction and monitoring of landslide hazards. Nat. Hazards
(20), 117-135.
Cassidy, M., Uzielli, M., & Lacasse, S. (2008). Probability risk assessment of
landslides: A case study at Finneidfjord. Can. Geotech. J.(45), 1250-1267.
Chung, C., & Fabbri, A. (2003). Validation of Spatial Prediction Models for Landslide
Hazard Mapping. Nat. Hazards (30), 451-472.
Clough, R., & Chopra, A. (1966). Earthquake stress analysis in earth dams. J. Eng.
Mech-ASCE (92), 197-211.
Corominas, J., van Westen, C., Frattini, P., Cascini, L., Malet, J., Fotopoulou, S., . . .
Smith, J. (2014). Recommendations for the quantitative analysis of landslide
risk. Bull. Eng. Geol. Environ. (73), 209-263.
Cross, J. (2001). Megacities and small towns: different perspectives on hazard
vulnerability . Environmental Hazards (3), 63-80.
Crovelli, R., & Coe, J. (2009). Probabilistic estimation of numbers and costs of future
landslides in the San Francisco Bay region. Georisk (3), 206-223.
Crozier, M. (2013). 'Landslides - Gravity always wins', Te Ara - the Encyclopedia of
New Zealand. Retrieved from
http://www.TeAra.govt.nz/en/diagram/8782/landslide-types
Dahal, R. K. (2013). Regional-scale landslide activity and landslide susceptibility
zonation in the Nepal Himalaya. Environ Earth Sci.
Dahal, R., Hasegawa, S., Nonoumra, A., Yamanaka, M., Dhakal, S., & Paudyal, P.
(2008). Predictive modelling of rainfall-induced landslide hazard in the Lesser
Chapter 10: References
95
Himalaya of Nepal based on weights-of-evidence. Geomorphology (102), 496-
510.
Dai, F., Lee, C., & Ngai, Y. (2002). Landslide risk assessment and management: an
overview. Eng. Geol. (64), 65-87.
Das, I., Kumar, G., & Stein, A. (2011). Stochastic landslide vulnerability modeling in
space and time in a part of the northern Himalayas, India. Environ. Monit.
Assess. (178), 25-37.
Einstein, H. (1988). Landlisde risk assessment procedure. Proc. Int. Symp. on
Landslides, (pp. 1075-1090). Lausanne.
Evans, S. G., Roberts, N. J., Ischuk, A., & Delaney, K. B. (109 (2009)). Landslides
triggered by the 1949 Khait earthquake, Tajikistan, and associated loss of life.
Eng. Geol., 195–212.
Fell, R., Corominas, J., Bonnard, C., Cascini, L., & Leroi, E. S. (2008). Guidelines for
landslide susceptibility, hazard and risk zoning for land-use planning. Eng. Geol.
(102), 99-111.
Galli, M., & Guzzetti, F. (2007). Landslide Vulnarability Criteria: A Case Study from
Umbria, Central Italy. Environ. Manage. (40), 649-664.
Gaziev, E. (1984). Study of the Usoi landslide in Pamir. Proc. 4th Int. Symp. on
Landslides (1), 511-514.
Gemitzi, A., Falalakis, G., Eskioglou, P., & Petalas, C. (2011). Evaluating landslide
susceptibility using environmental factors, fuzzy membership functions and
GIS. Global NEST Journal 13, 28-40.
Geotechdata.info. (2013). Angle of Friction. Retrieved (as of September 14.12.2013),
from http://geotechdata.info/parameter/angle-of-friction.html
Gubin, I. (1962). Seismicity and geological structures of Central Asia. Studia
Geophysica et Geodaetica (6), 410-412.
Chapter 10: References
96
Guha-Sapir, D., Below, R., & Hoyois, P. (2015). EM-DAT: International Disaster
Database – Université Catholique de Louvain – Brussels – Belgium. Retrieved
from www.emdat.be
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), 181–216.
Guzzetti, F., Reichenbach, P., Cardinali, M., Galli, M., & Ardizzone, F. (2005).
Probabilistic landslide hazard assessment at the basin scale. Geomorphology
(72), 272-299.
Hasegawa, S., Dahal, R., Nishimura, T., Nonomura, A., & Yamanaka, M. (2009).
DEM-Based Analysis of Earthquake-Induced Shallow Landslide Susceptibility.
Geotech. Geol. Eng. (27), 419-430.
Havenith, H., & Bourdeau, C. (2010). Earthquake-induced hazards in mountain regions:
a review of case histories from Central Asia - an inaugural lecture to the society.
Geologica Belgica (13), 137-152.
Havenith, H., Strom, A., Abdrakhmatov, A., Delvaux, D., & Tréfois, P. (1999). Seismic
triggering of landslides, Part A: Field evidence from the Northern Tien Shan.
Nat. Haz. Earth Sys. Sc. (3), 135-149.
Havenith, H., Strom, A., Caceres, F., & Pirard, E. (2006). Analysis of landslide
susceptibility in the Suusamyr region, Tien Shan: statistical and geotechnical
approach. Landslides (3), 39-50.
Havenith, H., Strom, A., Torgoev, I., Lamair, L., Ischuk, A., & Abdrakhmatov, K.
(online 2015). Tien Shan geohazards database: Earthquakes and landslides.
Geomorphology, doi:10.1016/j.geomorph.2015.01.037.
Holec, J., Bednarik, M., Šabo, M., Minár, J., Yilmaz, I., & Marschalko, M. (2013). A
small-scale landslide susceptibility assessment for the territory of Western
Carpathians. Nat. Hazards (69), 1081-1107.
Chapter 10: References
97
Hollenstein, K. (2005). Reconsidering the risk assessment concept: Standardizing the
impact desciption as a building block for vulnerability assessment. Nat. Hazards
Earth Syst. Sci. (5), 301-307.
Hungr, O., Corominas, J., & Eherhardt, E. (2005). Estimating landslide motion
mechanism, travel distance and velocity. In Hungr, Fell, Couture, & Eberhardt,
Landslide risk management (pp. 99-128). London: Taylor & Francis Group.
Ishihara, K. (1989). Liquefaction-induced landslide and debris-flow in Tajikistan,
USSR. Landslide News (3), 6-7.
Ishihara, K. (2012). Performances of Rockfill Dams and Deep-Seated Landslides
During Earthquakes. In Sakr, & Ansal, Special Topics in Earthquake
Geotechnical Engineering (pp. 273-314). Netherlands: Springer.
Jaiswal, P., van Westen, C., & Jetten, V. (2011). Quantitative estimation of landslide
risk from rapid debris slides on natural slopes in the Nilgiri hills, India. Nat.
Hazards Earth Syst. Sci. (11), 1723-1743.
Jibson, R. (2011). Methods for assessing the stability of slopes during earthquakes - A
retrospective. Eng. Geology, 43-50.
Kalmetieva, Z., Mikolaichuk, A., Moldobekov, B., Meleshko, A., Janaev, M., &
Zubovich, A. (2009). Atlas of earthquakes in Kyrgyzstan. Bishkek: Central
Asian Institute for Applie Geosciences and United Nations (UNISRD), Springer.
Kaynia, A., Papathoma-Köhle, M., Neuhäuser, B., Ratzinger, K., Wenzel, H., &
Medina-Cetina, Z. (2008). Probabilistic assessment of vulnerability to landslide:
Application to the village of Lichtenstein, Baden-Württemberg, Germany. Eng.
Geol. (101), 33-48.
Keefer, D. (1984). Landslides caused by earthquakes. Bull. Geol. Soc. Am. (95), 406-
421.
Chapter 10: References
98
Khampilang, N., & Whitworth, M. (2013). Creation of Landslide Inventory Map for the
Toktogul Region of Kyrgyzstan, Central Asia. In Margottini, Canuti, & Sassa,
Landslide Science and Practice (1) (pp. 197-202). Berlin Heidelberg: Springer-
Verlag.
Kramer, S. L. (1996). Geotechnical Earthquake Engineering. Prentice-Hall
International Series in Civil Engineering and Engineering Mechanics.
Lee, E., & Jones, D. (2004). Landslide Risk Assessment. London: Thomas Telford
Publishing.
Li, W., Huang, R., Tang, C., XU, Q., & van Westen, C. (2013). Co-seismic Landslide
Inventory and Susceptibility Mapping in the 2008 Wenchuan Earthquake
Disaster Area, China. J. Mt. Sci. (10,3), 339-354.
Li, Z., Nadim, F., Huang, H., Uzielli, M., & Lacasse, S. (2010). Quantitative
vulnerability estimation for scenario-based landslide hazards. Landslides (7),
125-134.
Lutz, W. (2010). Emerging population issues in Eastern Europe and Central Asia.
Reseach Gaps on demographic Trends, Human Capital and Climate Change.
UNFPA, New York, USA.
Mavrouli, O., Fotopoulou, S., Pitilakis, K., Zuccaro, G., Corominas, J., Santo, A., . . .
Ulrich, T. (2014). Vulnerability assessment for reiforced concrete buildings
exposed to landslides. Bull. Eng. Geol. Environ. (73), 265-289.
Molnar, P., & Tapponier, P. (1975). Cenozoic tectonics of Asia Effects of a continental
collision. Science (189), 419-426.
Mousavi, S., Omidvar, B., Ghazban, F., & Feyzi, R. (2011). Quantitative risk analysis
for earthquake-induced landslides - Emamzadeh Ali, Iran. Engineering Geology
(122), 191-203.
Chapter 10: References
99
Nadim, F., Kjekstad, O., Peduzzi, P., Herold, C., & Jaedicke, C. (2006). Global
landslide and avalanche hotspots. Landslides (3), 159-173.
Nefeslioglu, H., & Gokceoglu, C. (2011). Probabilistic risk assessment in medium scale
for rainfall induced earthflows: Catakli catchment area (Cayeli, Rize, Turkey).
Math. Probl. Eng., p. 21.
Neuhäuser, B., & Terhorst, B. (2007). Landslide susceptibility assessment using
"weights-of-evidence" applied to a study area at the Jurassic escarpment (SW-
Germany). Geomorphology (86), 12-24.
Newmark, N. (1965). Effects of earthquakes on dams and embarkments. Geotechnique
(15), 139-159.
Niyazov, R., & Nurtaev, B. (2013). Evaluation of Landslides in Uzbekistan Caused by
the Joint Impact of Precipitation and Deep-Focus Pamir-Hindu Kush
Earthquakes. In Sassa, Rouhban, Briceno, McSaveney, & He, Landslides:
Global Risk Preparedness (pp. 255-267). Berlin Heidelberg : Springer-Verlag.
Niyazov, R., & Nurtaev, B. (2013). Modern Seismogenic Landslides Caused by the
Pamir-Hindu Kush Earthquakes and Their Consequences in Central Asia. In
Margottini, Canuti, & Sassa, Landslide Science and Practice , 343-348. Berlin
Heidelberg: Springer-Verlag.
Oh, H., & Lee, S. (2010). Landslide susceptibility mapping on Panaon Island,
Philippines using a geographic information system. Environ. Earth Sci. (62),
935-951.
Papathoma-Köhle, M., Kappes, M., Keiler, M., & Glade, T. (2011). Physical
vulnerability assessment for alpine hazards: state of the art and future needs.
Nat. Hazards (58), 645-680.
Pascale, S., Sdao, F., & Sole, A. (2010). A model for assessing the systemic
vulnerability in landslide prone areas. Nat. Hazards Earth Syst. Sci. (10), 1575-
1590.
Chapter 10: References
100
Pavlis, T., Hamburger, M., & Pavlis, G. (1997). Erosional processes as a control on the
structural evolution of an actively deforming fold and thrust belt: An example
from the Pamir Tien Shan region, Central Asia. Tectonics (16), 810-822.
Perov, V., & Budarina, O. (2000). Mudflow hazard assessment for the Russian
Federation. In Wieczorek, & Naeser, Debris-flow hazards mitigation:
mechanics, prediction, and assessment, 489-494. Rotterdam, Netherlands:
Balkema.
Petley, D. (2012). Global Patterns of loss of life from landslides. Geology 40(10), 927-
930.
Pittore, M. GFZ- Potsdam. (2014). Focusmapr: computation of Focus Maps from a set
of input raster or vector layers. R package version 1.0.
Pradhan, B., Oh, H.-J., & Buchroithner, M. (2010). Weights-of-evidence model applied
to landslide susceptibility mapping in a tropical hilly area. Geomatics Nat.
Hazards Risk 1(3), 199-223.
Quantum GIS Development Team. Quantum GIS Geographic Information System. Open
Source Geospatial Foundation Project. (2015). Retrieved from
http://qgis.osgeo.org
Remondo, J., Bonachea, J., & Cendrero, A. (2005). A statistical approach to landslide
risk modelling at basin scale: from landslide susceptibility to quantitative risk
assessment. Landslides (2), 321-328.
Remondo, J., Gonzales, A., De Teran, J., Cendrero, A., Fabbri, A., & Chung, C. (2003).
Validation of Landslide Susceptibility Maps: Examples and Applications from a
Case Study in Northern Spain. Nat. Hazards (30), 437-449.
Rosenfeld, C. (1994). The geomorphological dimensions of natural disasters.
Geomorphology (10), 27-36.
Chapter 10: References
101
Scheidegger, A. E. (1973). On the Prediction of the Reach and Velocity of Catastrophic
Landslides. In Rock Mechanics (5) (pp. 231-236). Springer-Verlag.
Scheidegger, A. E. (1975). Physical Aspects of Natural Catastrophes. Elsevier
Scientific Publishing Company.
Schicker, R., & Moon, V. (2012). Comparison of bivariate and multivariate statistical
approaches in landslide susceptibility mapping at the regional scale.
Geomorphology (161), 40-57.
Schlögel, R., Torgoev, I., De Marneffe, C., & Havenith, H. (2011). Evidence of a
changing size–frequency distribution of landslides in the Kyrgyz Tien Shan,
Central Asia. Earth Surf. Process. Landforms (36), 1658–1669.
Schuster, R. L., & Alford, D. (May 2004). Usoi Landslide Dam and Lake Sarez, Pamir
Mountains, Tajikistan. Environ. Eng. Geosci. (10,2), 151–168.
SENSUM. (2014). Framework to integrate Space-based and in-situ sensing for
dynamic vulnerability and recovery monitoring. FP7-SPACE-2012-1.
Deliverable 3.4.
Sheko, A. (1983). Landslides and Mudflows. Reports of the Alma-Ata International
Seminar, Moscow, Centre of International Projects, GKNT.
Sidorova, T. (1997). Potential changes of mudflow phenomena due to global warming.
In Rickemann, & Chen, Debris-flow hazards mitigation: mechanics, prediction,
and assessment , 540-549. Rotterdam, Netherlands: Balkema.
Soeters, R., & van Westen, C. (1996). Slope stability: recognition, analysis and
zonation. In Turner, & Shuster, Landslides: investigation and mitigation (pp.
129-177). Transportation Research Board, Special report 247.
SRTM Shuttle Radar Topography Mission. SRTM digital topographic data US
Geological Survey's EROS Data Center. (2004). Retrieved from
ftp://e0mss21uecsnasagov/srtm/
Chapter 10: References
102
Sterlacchini, S., Frigerio, P., & Giacomelli, M. (2007). Landslide risk analysis: a multi-
disciplinary methodological approach. Nat. Hazards Earth Syst. Sci.(7), 657-
675.
Stewart, J., Blake, T., & Hollingsworth, R. (2003). A screen analysis procedure for
seismic slope stability. Earthq. Spectra (19), 697-712.
Strom, A. (2013). Geological Prerequisites for Landslides' Dams' Disaster Assessment
and Mitigation in Central Asia. In Wang, Miyaijma, Li, Shan, & Fathani,
Progress of Disaster Mitigation Technology in Asia , 17-53.
Strom, A., & Korup, O. (2006). Extremely large rockslides and rock avalanches in the
Tien Shan Mountains Kyrgyzstan. Landslides (3), 125-136.
Terzaghi, K. (1950). Mechanism of landslides. In Paige, Application of Geology to
Engineering Practice (Berkey Volume) (pp. 83-123). New York: Geological
Sosciety of America.
Terzaghi, L., & Pech, R. (1967). Soil mechanics in engineering practice. New York:
Wiley, p. 752.
Thomas, J., Lanza, R., Kazanski, A., Zykin, V., Semakov, N., Mitrokhin, D., &
Delvaux, D. (2002). Paleomagnetic study of Cenozoic sediments from the
Zaisan basin (SE Kazakhstan) and the Chuya depression (Siberian Altai):
tectonic implications for central Asia. Tecnonophysics (351), 119-137.
Tingdong, L., Ujkenov, B., Kim, B., Tomurtogoo, O., Petrov, O., & Strelnikov, S.
(2008). Geological map of Central Asia and Adjacent Areas edt by Geological
Publishing House. Beijing China.
Torgoev, A., & Havenith, H.-B. (2013). Landslide Susceptibility, Hazard and Risk
Mapping in Mailuu-Suu, Kyrgyzstan. In Margottini, Canuti, & Sassa, Landslide
Science and Practice (1) , 505-510. Berlin Heidelberg: Springer-Verlag.
Chapter 10: References
103
Torgoev, I., Alioshin, Y. G., & Torgoev, A. (2012). Monitoring landslides in
Kyrgyzstan. Freiberg Online Geology (33), 130-139.
Trifonov, V., Soboleva, O., Trifonov, R., & Vostrikov, G. (2002). Recent geodynamics
of the Alpine-Himalayan collision belt. Transaction of the Geological Institute
RAS 541, p. 224 (in Russian).
Uchida, T., Osanai, N., Onoda, S., Takayama, T., & Tomura, K. (2006). A Simple
Method for Producing Probabilistic Seismic Shallow Landslide Hazard Maps. In
Disaster Mitigation of Debris Flows, Slope Failures and Landslides , 529-534.
Tokyo, Japan: Universal Academy Press, Inc.
UNISDR. (2015). Making Development Sustainable: The Future of Disaster Risk
Management. Global Assessment Report on Disaster Risk Reduction. Geneva,
Switzerland: United Nations Office for Disaster Risk Reduction (UNISDR).
Uzielli, M., Nadim, F., Lacasse, S., & Kaynia, A. (2008). A conceptual framework for
quantitative estimation of physical vulnerability to landslides. Eng. Geol.(102),
251-256.
Van Den Eeckhaut, M., Hervás, J., Jaedicke, C., Malet, J., Montanarella, L., & Nadim,
F. (2012). Statistical modelling of Europe-wide landslide susceptibility using
limited landslide inventory data. Landslides (9), 357-369.
Van Westen, C., Rengers, N., & Soeters, R. (2003). Use of geomorphological
information in indirect landslide susceptibility assessment. Nat. Hazards (30),
399-419.
Van Westen, C., Van Asch, T., & Soeters, R. (2006). Landslide hazard and risk
zonation - why is it still so difficult? Bull. Eng. Geol. Environ., 297-306.
Varnes, D. (1978). Slope Movement Types and Processes.
Chapter 10: References
104
Varnes, J., & IAEG. (1984). IAEG International Association of Engineering Geology
Commission on Landslides and Other Mass Movements, Landslide hazard
zonation: a review of principles and practice. Paris, 63p: The UNESCO Press.
Wieczorek, G. F. (1996). Landslide triggering mechanisms. In Turner, & Schuster,
Landslide investigation and mitigation. Special report 247 (p. 673). Washington,
D. C.: National Academic Press.
Yang, S., He, S., Du, J., & Sun, X. (2015). Screening of social vulnerability to natural
hazards. Nat. Hazards, 1-18.
Yilmaz, I. (2010). The effect of the sampling strategies on the landslide susceptibility
mapping by conditional probability and artificial neural networks. Environ.
Earth Sci. (60), 505-519.
Zezere, J., Oliveira, S., Garcia, R., & Reis, E. (2007). Landslide risk analysis in the area
North of Lisbon (Portugal): evaluation of direct and indirect ocsts resulting from
a motorway disruption by slope movements. Landslides (4), 123-136.
Chapter 11: List of Publications
105
11 LIST OF PUBLICATIONS
Journals and report
Pilz, M., Roessner, S., Janssen, C., Behling, R., Parolai, S., Saponaro, A., Schäbitz, M.
(2013). Massenbewegungen in Zentralasien: Lawinen aus Boden und Gestein. System
Erde 3, 2. doi: 10.2312/GFZ.syserde.03.02.
Pilz, M., Parolai, S., Bindi, D., Saponaro, A., Abdybachaev, U. (2014): Combining
seismic noise techniques for landslide characterization. Pure and Applied Geophysics
171, 1729-1745, doi: 10.1007/s00024-013-0733-3.
Saponaro, A., Pilz, M., Wieland, M., Bindi, D., Moldobekov, B., & Parolai, S. (online
2014). Landslide susceptibility analysis in data-scarce regions: the case of Kyrgyzstan.
Bull. Eng. Geol. Environ., doi: 10.1007/s10064-014-0709-2.
Saponaro, A., Pilz, M., Bindi, D., & Parolai, S. (2015). The contribution of EMCA to
landslide susceptibility mapping in Central Asia. Annals of Geophysics (58, 1), doi:
10.4401/ag-6668.
Chapter 11: List of Publications
106
Conferences and presentations
Saponaro, A. (2013). An innovative tool for landslide susceptibility mapping in
Kyrgyzstan, Central Asia. 8th PhD Day of the GeoForchungsZentrum, Potsdam, March
2013
Saponaro, A. (2013). An innovative tool for landslide susceptibility mapping in
Kyrgyzstan, Central Asia. Poster presentation at the General Assembly of the European
Geosciences Union (EGU), Vienna, Austria, April 2013
Saponaro, A. (2013). Earthquake-induced Landslides and Site-effects. Workshop in the
frame of TIPTIMON capacity-building activities in Central Asian countries. Bishkek,
Kyrgyzstan, 18th – 22nd November 2013
Fleming, K.M., Saponaro, A. (2014). Use of TRMM rainfall products and other remote
sensing missions for hazard and vulnerability assessment in Central Asia. Symposium
on earthquake and landslide risk in Central Asia and Caucasus: exploiting remote
sensing and geo-spatial information management Bishkek, Kyrgyzstan, 29th – 30th
January 2014
Saponaro, A., Bindi, D. and Parolai S. (2014). A statistical approach for earthquake-
induced landslide susceptibility mapping in Kyrgyzstan. 9th PhD Day of the
GeoForchungsZentrum, Potsdam, March 2014
Saponaro, A., Pilz, M., Parolai, S. (2014). A statistical approach for landslide
susceptibility mapping in Kyrgyzstan, Central Asia. TIPTIMON Midterm Meeting in
the framework of joint activities between CAIAG and GFZ, Potsdam, April 2014
Saponaro, A. (2014). Use of GIS for Landslide Susceptibility Mapping. Talk presented
at the GIS DAY at GFZ 2014, Potsdam, November 2014
Saponaro, A., Pilz, M., Wieland, M., Pittore, M., Bindi, D., Parolai, S. (2014). Towards
Cross-Border Landslide Hazard and Risk Assessment in Central Asia. Oral presentation
at the American Geophysical Union (AGU)’s 47th annual Fall Meeting, San Francisco,
US, December 2014
Chapter 11: List of Publications
107
Saponaro, A., Pilz, M., Wieland, M., Pittore, M., Bindi, D., Parolai (2015). Towards
cross-border landslide hazard and risk assessment in Central Asia, 10th PhD Day of the
GeoForchungsZentrum, Potsdam, March 2015
