Tyoda Landslide 2013

LANDSLIDE SUSCEPTIBILITY MAPPING: REMOTE SENSING AND GIS
APPROACH
by
Zipho Tyoda
Thesis presented in fulfilment of the requirements for the degree of Masters in Geography and Environmental
Studies, in the faculty of Science at Stellenbosch University
Supervisor: Dr Jaco Kemp

Co-Supervisor: Ms Jeanine Engelbrecht
March 2013
Stellenbosch University http://scholar.sun.ac.za
Declaration
By submitting this thesis electronically, I declare that the entirety of the work contained therein is
my own, original work, that I am the sole author thereof (save to the extent explicitly otherwise
stated), that reproduction and publication thereof by Stellenbosch University will not infringe any
third party rights and that I have not previously in its entirety or in part submitted it for obtaining
any qualification.
March 2013
Copyright © 2013 Stellenbosch University
All rights reserved

ABSTRACT
Landslide susceptibility maps are important for development planning and disaster management. The current
synthesis of landslide susceptibility maps largely applies GIS and remote sensing techniques. One of the most
critical stages on landslide susceptibility mapping is the selection of landslide causative factors and weighting of
the selected causative factors, in accordance to their influence to slope instability. GIS is ideal when deriving
static factors i.e. slope and aspect and most importantly in the synthesis of landslide susceptibility maps. The
integration of landslide causative thematic maps requires the selection of the weighting method; in order to
weight the causative thematic maps in accordance to their influence to slope instability. Landslide susceptibility
mapping is based on the assumption that future landslides will occur under similar circumstances as historic
landslides. The weight of evidence method is ideal for landslide susceptibility mapping, as it calculates the
weights of the causative thematic maps using known landslides points. This method was applied in an area
within the Western Cape province of South Africa, the area is known to be highly susceptible to landslide
occurrences. A prediction rate of 80.37% was achieved. The map combination approach was also applied and
achieved a prediction rate of 50.98%.
Satellite remote sensing techniques can be used to derive the thematic information needed to synthesize landslide
susceptibility maps and to monitor the variable parameters influencing landslide susceptibility. Satellite remote
sensing techniques can contribute to landslide investigation at three distinct phases namely: (1) detection and
classification of landslides (2) monitoring landslide movement and identification of conditions leading up to an
event (3) analysis and prediction of slope failures. Various sources of remote sensing data can contribute to
these phases. Although the detection and classification of landslides through the remote sensing techniques is
important to define landslide controlling parameters, the ideal is to use remote sensing data for monitoring of
areas susceptible to landslide occurrence in an effort to provide an early warning. In this regard, optical remote
sensing data was used successfully to monitor the variable conditions (vegetation health and productivity) that
make an area susceptible to landslide occurrence.
ii
CONTENTS
Landslide susceptibility mapping: Remote sensing and GIS approach ..................................................... i
Abstract ..................................................................................................................................................... ii
Contents ................................................................................................................................................... iii
Figures ...................................................................................................................................................... vi
Tables ..................................................................................................................................................... xiv
ACKNOWLEDGEMENTS ................................................................................................................... xvi
Chapter 1 Background to the study ..................................................................................................... 1
1.1 Introduction to Landslides and landslide susceptibility ............................................................. 1
1.2 Project aims and goals ................................................................................................................ 2
1.3 The research questions and objectives ....................................................................................... 3
1.4 Landslides in South Africa and Description of the study area ................................................... 5
Chapter 2 Literature review and theory ............................................................................................... 9
2.1 Landslide Classification ............................................................................................................. 9
2.2 Landslide controlling parameters ............................................................................................. 11
2.2.1 Static factors ...................................................................................................................... 13
2.2.2 Variable factors ................................................................................................................. 15
2.2.3 Triggering mechanisms ..................................................................................................... 17
2.3 Remote sensing and GIS for Landslide susceptibility mapping .............................................. 17
2.4 GIS modelling for landslide susceptibility mapping ................................................................ 18
2.4.1 Remote sensing on landslide susceptibility mapping ....................................................... 22
2.5 Final remarks ............................................................................................................................ 25
iii
Chapter 3 GIS-Based landslide susceptibility mapping: Materials, results and accuracy assessment
26
3.1 Input data .................................................................................................................................. 26
3.1.1 Geology ............................................................................................................................. 26
3.1.2 Geomorphology ................................................................................................................ 30
3.1.3 Landcover.......................................................................................................................... 33
Description ...................................................................................................................................... 39
3.1.4 Anthropogenic influences ................................................................................................. 41
3.2 Landslide susceptibility mapping – the weights of evidence APPROACH............................. 42
3.2.1 Weight of evidence landslide susceptibility map .............................................................. 44
3.2.2 Accuracy Assessment for the Weight of Evidence method .............................................. 49
3.3 Landslide susceptibility mapping – the Map combination Approach ...................................... 53
3.3.1 Accuracy Assessment for the Map Combination approach .............................................. 59
Chapter 4 Remote monitoring of variable conditions and identification of triggering mechanisms. 62
4.1 Extraction of NDVI and NDWI – MODIS data ....................................................................... 65
4.2 Extraction of NDVI and NDWI and phenology data – Landsat data ....................................... 73
4.2.1 Image pre-processing ........................................................................................................ 74
4.2.2 Derivation of information on vegetation health and productivity and moisture conditions
77
4.3 Triggering mechanisms ............................................................................................................ 82
Chapter 5 Discussion, recommendations and conclusions ................................................................ 87
5.1 Discussion ................................................................................................................................ 87
5.2 Intergration of landslide susceptibility mapping and development planning........................... 90
iv
5.3 Recommendations .................................................................................................................... 91
5.4 Conclusions .............................................................................................................................. 91
References ............................................................................................................................................... 92
Special references ................................................................................................................................. 103
v
LIST OF ACRONYMS
ARC Agricultural Research Council

ANN Artificial Neural Network
CSIR Council for Scientific and Industrial Research
DST Department of Science and Technology
DEM Digital Elevation Model
DTM Digital Terrane Model
DN Digital Numbers
GCP Ground Control Points
LSI Landslide Susceptibility Index
LIDAR Light Detection And Ranging
GIS Geographic Information system
MODIS Moderate Resolution Imaging Spectroradiometer
NGO Non-governmental organization
NDVI Normalized Difference Vegetation Index
NDWI Normalized Difference Water Index
NIR Near-infra-red
NGI National Geo-spatial Information
NIR Near-Infrared
RMSE Root Mean Square Error
SWIR Short-Wave-Infrared
SAR Synthetic Aperture Radar
SANSA South African National Space Agency
vi
FIGURES
Figure 1: The schematic diagram illustrating the standard procedure that is used when modelling landslide susceptibility maps.
....................................................................................................................................................................4
Figure 2: The extent of the study area in the Western Cape Province of South Africa and a selection of field-verified historical
landslide positions. ....................................................................................................................................6
Figure 3: Geological map of the study area. .............................................................................................8
Figure 4: A classification of mass movement processes on slope (Carson & Kirby 1972).....................11
Figure 5: Effect of water content on cohesive strength of clay (Zhou 2006).The x-axis shows the water content and the y-axis it shows
the cohesive strength. ...............................................................................................................................16
Figure 6: The taxonomy of the different weighting approaches when conducting landslide susceptibility modelling (Source: Kanungo
et al. 2009 pp 11). ....................................................................................................................................19
Figure 7: The spectral reflectance curve of green and dry vegetation and soil along with the spectral wavelength (Clarck et al.1999).
..................................................................................................................................................................24
Figure 8: The geological parameter of the study area, the lithological units are shown on the legend. This geological layer was one of
the causative thematic layers used to modell a landslide susceptibility map of the study are using the weight of evidence method.
..................................................................................................................................................................27
vii
Figure 9: The 250 meter buffered lithological contacts and faults layer for the study area. This thematic layer was one of the causative
thematic layers used to model a landslide susceptibility map of the study area. The localities for the landslides are also shown. 28
Figure 10: The graph shows the number of landslides per lithology. On the y-axis is the stratigraphic units and on the x-axis is the
number of landslides The alluvium deposits recorded the highest number of landslides and the Skurweberg, Rietvlei and Peninsula
stratigraphic units recorded the second highest, and the third highest number of landslides, in decreasing order.29
Figure 11: Slope layer of the study area. The classes that were used in the weight of evidence method are also shown on the legend.
..................................................................................................................................................................30
Figure 12: The number of landslides per slope class. Roughly 90 % of the landslides fall on the 0-20° and 20-40° class. 31
Figure 13: The aspect layer for the study area. The classes are shown in the legend. The aspect thematic layer was one of the
causative layers used to model a landslide susceptibility map of the study area, using the weight of evidence method. The black dots on
the map are the localities for the known historic landslides within the study area. ................................32
Figure 14: The number of landslides per aspect class. The south facing slope recorded the highest number of landslides and the east
facing slopes recorded the least number of landslides. ...........................................................................33
Figure 15: The land cover layer used when modelling the landslide susceptibility map of the study area. The landcover classes are
shown in the legend. .................................................................................................................................34
Figure 16: The number of landslides per landcover class. A large number of landslides fall on the Shrubland and low fynbos class.
The other landcover classes recorded a very low number of landslides. ................................................35
viii
Figure 17: The soil depth layer of the study areas. The depth classes are also shown on the legend. This thematic layer was one of the
causative parameters used to model a landslide susceptibility map of the study area. ..........................36
Figure 18: The number of landslides per soil depth class. More than 90 % of the landslides fall on the soil depth less 300 mm, 300-
600mm and 600-900 mm; in decreasing order. Soil depth less than 300mm recorded the highest number of landslides and the soil
depth between 900-1200mm recorded the least number of landslides. ...................................................37
Figure 19: The soil type layer for the study area, which was used to model the landslide susceptibility map using the weight of
evidence method. The description for the codes used in the legend is shown on the Table 3. ................38
Figure 20: The number of landslides per soil type class. Class Lb (Rock outcrops comprise >60% of land type) recorded the highest
number of landslides and class Gb (Podzols occur (comprise >10% of land type); dominantly shallow) recorded the least number of
landslides. ................................................................................................................................................41
Figure 21: The layer containing the roads and rails within the study area. A buffer of 50 meter was used. The localities of known
historic landslides are represented by the black dots on the map. ..........................................................42
Figure 22: Classification of the landslide susceptibility map using the natural break method. .............45
Figure 23: The unclassified weight of evidence landslide susceptibility map. The legend shows the increasing susceptibility of the map,
with the higher values representing higher susceptibility and lower values representing lower susceptibility. 46
Figure 24: The landslide susceptibility map of the study area based on the weight of evidence model. The classes are shown on the
legend. The red and orange areas are very high and high susceptibility, the yellow areas are moderately susceptible areas and the
green areas are the very low and low susceptible areas. ........................................................................47
ix
Figure 25: The number of landslides per susceptibility class. These landslides points were used to run the weight of evidence model. A
large number of landslides fall on the very high and high class (96.8 %). .............................................48
Figure 26: The number of landslides per susceptibility class. These landslide points were not used when the susceptibility map was
modelled; they were only used to test the efficiency of the model. A large number of landslides fall on the high and very high class (90
%). ............................................................................................................................................................49
Figure 27: Classification of the landslide susceptibility map using equal interval method. ...................50
Figure 28: The success (blue) and prediction (red) rate curves for the weight of evidence susceptibility model. 51
Figure 29: The overall success percentage for the weight of evidence method, the blue highlighted area is the overall percentage for
the training data set. ................................................................................................................................52
Figure 30: The overall success percentage for the weight of evidence method, the red highlighted area is the overall percentage for
the reference data set ...............................................................................................................................53
Figure 31: The landslide susceptibility map of the study area. The map was modelled using a map combination approach. The
susceptibility values are shown in the legend, the higher the values, the higher the susceptibility and the lower the value the lower the
susceptibility. ...........................................................................................................................................56
Figure 32: The landslide susceptibility map modelled using the map combination approach. The susceptibility classes are shown on
the legend. The red and orange areas represent very high and high susceptibility areas, the yellow areas depict moderate
susceptibility areas and the green areas are low susceptible areas. .......................................................57
Figure 33: The graph shows the number of landslides per susceptibility class. .....................................58
x
Figure 34: The graph shows the number of landslides per susceptibility class. These landslide localities where attained from the
Council for Geosciences. .........................................................................................................................59
Figure 35: The success rate curve for the map combination approach. The red line is the success rate curve for the reference data set
and the blue trend is the success rate curve for the training data set. ....................................................60
Figure 36: The cumulative success percentage for the map combination approach. The blue highlighted area is the success
percentage for the training data. .............................................................................................................60
Figure 37: The cumulative success percentage for the map combination approach. The red highlighted area is the success percentage
for the reference data. ..............................................................................................................................61
Figure 38: The geographic location of the 2005, 2007 and 2008 landslide events. The landslides are located on the far north western
and south western part of the study area. ................................................................................................63
Figure 39: The geographic location of several landslide events. The landslides are located close to the towns of Stanford and
Hermanus. The landslide localities are represented by black dots on the map. .....................................64
Figure 40: Google Earth images indicating the landscape before landslide occurrence (29-09-2004) and the landscape after landslide
occurrence (02-11-2006). ........................................................................................................................64
Figure 41: The NDVI time-series profile, of the selected area which is known to have landslide occurrence. The profile is from the
year 2003 to the year 2006. The red line indicates the period at which the landslide is thought to have occurred. The plot shows that
the minimum NDVI values for the year at which the landslides occurred are lower than the previous years. 66
xi
Figure 42: The NDVI time-series profile of the area known to have landslide occurrence. The plot is from the year 2006 to the year
2009. The red line indicates the period at which the landslide is thought to have occurred. The plot shows relatively low minimum
NDVI values for the year prior to landslide occurrence. ........................................................................68
Figure 43: The NDVI time series profile for the year 2003 to 2008. The red line on the graph indicates the period at which the
landslide is thought to have occurred. The minimum NDVI values are slightly lower for the year prior to the occurrences of landslide
event. ........................................................................................................................................................69
Figure 44: The NDVI time-series profile of an area with numerous landslide scars. The area is close to the town of Hermanus and
Stanford. The landslides are estimated to have occurred between the year 2004 and 2006.The red line on the graph indicates the time
at which the landslide could have occurred, based on the previous observations that landslide in the study area are associated with
low minimum NDVI values for the year prior to landslide occurrence. ..................................................71
Figure 45: The NDVI time-series profile for the area known to have landslide occurrence. The red line on the graph indicates the
period at which the landslide is thought to have occurred. The minimum NDVI values are lower for the year prior to landslide
occurrence (2006). ...................................................................................................................................72
Figure 46: The NDVI time-series profile for the year 2005 to 2009. The red line on the graph indicates the period at which the
landslide is thought to have occurred. The minimum NDVI values are slightly lower for the year prior to landslide occurrence. 73
Figure 47: Steps followed when performing atmospheric corrections. ...................................................77
Figure 48: The 2005-02-27scene for the area close to the town of Stanford and Hermanus. There are no visible landslide scars on this
image, except one feature close the small round water body. .................................................................78
xii
Figure 49: The 2005-09-27scene for the area close to the town of Stanford and Hermanus. The image shows several landslide scars
(the bright feature on the south facing slope on the mountain. ...............................................................79
Figure 50: NDVI change detection computed from the landsat scenes between 2004-09-20 and 2005-02-27. The light red areas on the
map are areas where there has been a decrease in NDVI and the strong red areas are those that had more than 15 % NDVI decrease.
The light green and bright green areas depict those areas that experienced some NDVI increase and greater than 15 % NDVI
increase, respectively. The localities of the landslides are represented by the black stars on the map. 80
Figure 51: NDVI change detection computed from the images taken from 2005-09-07 and 2006-04-19. The light red areas on the map
are areas where there has been a decrease in NDVI and the strong red areas are those that had more than 15 % NDVI decrease. The
light green and strong green areas depict those areas that experienced some NDVI increase and greater than 15 % NDVI increase,
respectively. The exact localities of the landslides are represented by the black stars on the map. .......81
Figure 52: NDVI change detection computed from the images taken from 2006-08-09 to 2007-02-17. The light red areas on the map
are areas where there has been a decrease in NDVI and the strong red areas are those that had more 15 % NDVI decrease. The light
green and strong green areas depict those areas the experienced some NDVI increase and greater than 15 % NDVI increase,
respectively. The exact localities of the landslides scars are represented by the black stars on the map.82
Figure 53: Annual rainfall for the year 2004, 2005 and 2006. The year of 2005 recorded the highest annual rainfall, slightly higher
than 800 mm. ............................................................................................................................................84
Figure 54: The plot for the monthly rainfall for the year 2005. The months of April and June show monthly rainfall greater than 100
mm, with the month of April recording close to 250mm. .........................................................................86

xiii
TABLES
Table 1: The schematic landslide classification system adapted from Varnes (1978). ............................................................................ 10
Table 2: The comprehensive review of the different GIS techniques that have been used for landslide susceptibility modelling. ......... 19
Table 3: The description for the soil type classes (source: ARC institute for Soil, Climate and Water). ................................................ 39
Table 4: The weights and ranks for the causative thematic layers used to model an expert based landslide susceptibility map............ 54
Table 5: Shows the number of landslides and the date of occurrence for each area. .............................................................................. 62
Table 6: The annual average, minimum and maximum values of NDVI. The values are calculated from the beginning of the year to the
end of the year. .......................................................................................................................................................................................... 67
Table 7: The yearly average, minimum and maximum values of NDVI. The values are calculated from the beginning of the year, to the
end of the year. .......................................................................................................................................................................................... 70
Table 8: Landsat scenes that will be used in this study. ........................................................................................................................... 74
Table 9: Table representing the scene date RMS error and the number of GCP's collected when geometric correction was performed.
................................................................................................................................................................................................................... 75
Table 10: The annual rainfall for the weather station in Hermanus (-34.417: 19.237). The monthly average and the total annual
rainfall are also indicated. The rainfall is in millimetres (mm). ............................................................................................................. 83
Table 11: The annual rainfall data (in millimeters) for the weather station close to the town of Hermanus. The blank areas in the
table indicate that no rain fell on that day, *** indicates that the data is missing or not yet available in the current month, C next to
xiv
the value indicates that the rainfall was accumulated over a number of days, = indicates that the total for the month is unreliable due
to missing daily values, and A or B indicates that any rainfall that did occur is included in the accumulation. .................................... 85
xv
ACKNOWLEDGEMENTS
 I would like to express my appreciation to a number of people who contributed in one way or the
other to the completion of this thesis.
 My deepest thanks goes to the Council for Geoscience and Department of Science and
Technology ( DST) for funding this project
 I am grateful to my supervisor Dr Jaco Kemp and Co-supervisor Ms Jeanine Engelbrecht for their
tireless guidance and advice.
 I am thankful for the support I obtained from my colleagues at the Council for Geoscience.
Special thanks go to Dr Chiedza Musekiwa and Dr Stapelberg.
 I would like to extend my thanks and appreciation to my family for their never ending support and
encouragement.
 My honest thank also go to everyone who has not been mentioned but contributed and supported
me, thank you.
xvi
CHAPTER 1 BACKGROUND TO THE STUDY
1.1 INTRODUCTION TO LANDSLIDES AND LANDSLIDE SUSCEPTIBILITY
Landslides are defined as mass movement processes that involve down-slope movement of slope material along
discrete shear surfaces under the influence of gravity (Cruden & Varnes 1996). Landslides play a significant role
in the evolution of the hill-slope and long-term landscape evolution. The abrupt nature and the catastrophic
forces of the process can have undesirable socio-economic impacts. The hazardous nature of landslides can
result in substantial economic losses, fatalities, geomorphologic disturbances, ecosystem disturbances and
infrastructure disturbances. Landslides can be triggered by earthquakes/seismicity, human activities (i.e. road-
cuts and vegetation removal) but in mountainous landscapes, landslides are more frequently triggered by heavy
rainfalls (Brunettii, Peruccacci, Rossi, Luciani, Valigi & Guzzetti 2010). Landslide susceptibility mapping is a
vital tool for disaster management and planning development activities in mountainous terrains of tropical and
subtropical environments (Dahal, Hasegawa, Nonomura, Yamanaka, Masuda & Nishino 2007).
Steep terrain, considerable topographic variation, high relief, diverse geology, humid climate and seismicity
make some parts of South Africa susceptible to landslide activity. Landslides are often associated with severe,
high intensity rainfall events (Singh 2009). In 1989 the estimated annual costs of landslide associated expenses
in Southern Africa, were estimated at approximately US$ 20 million (Paige-Green 1989). Based on an annual
standard inflation rate of 10%, the current suggested amount means that annual landslide associated expenses
would cost Southern Africa ~US$ 163 million (Singh, Forbes, Diop, Musekiwa & Claasen 2011).
Landslide susceptibility mapping has had little improvements in principle and the difficulty in landslide
prediction is the result of different factors controlling landslide occurrences (Kanungo, Arora, Sarkar & Gupta.
2009). Nevertheless the evolution of remote sensing, GIS and field work techniques has produced reliable
landslide susceptibility maps, which has been successfully used during development planning by governments
and NGOs in different regions around the world (e.g. Sarkar & Kanungo 2004, Hung, Batelaan, San & Van
2005). Landslide prediction methodologies are based on the assumption that future landslides will occur under
circumstances similar to the ones of past landslides (Chung & Shaw 2000). Consequently, previous research has
devoted significant amount of time on developing techniques for studying the spatial distribution of the landslide
controlling parameters. Historically, landslide susceptibility assessment and mapping were considered to be
laborious and time consuming. However, significant developments in remote sensing, computer application and
1
geographic information systems have facilitated the process significantly (Dahal et al. 2007) and played a
significant role in landslide forecasting and modelling (Temesgen , Mohammed & Korme, 2001; Sarkar &
Kanungo 2004; Hung et al 2005, Singh 2009).
The research presented here aimed to synthesize a landslide susceptibility map of a selected area in the Western
Cape Province of South Africa. Additionally, standard techniques and methodologies for landslide susceptibility
modelling were introduced that can be used for landslide susceptibility and early warning investigations in South
Africa.
1.2 PROJECT AIMS AND GOALS
The likelihood that an area will be affected by landslides is dependent on several factors. These factors are static
factors (such as the slope of the terrain and the underlying geology) as well as variable factors (such as the health
and productivity of vegetation in the area and the soil water content). If a critical combination of static and
variable conditions is met, the area would have a high likelihood to be affected by a landslide event. The
presence of a triggering mechanism (such as a high intensity rainfall event or an earthquake) would then lead to
slope failure and landslide occurrence.
This study aims to synthesize a landslide susceptibility map by considering the static variables and how they
influence landslide susceptibility. Secondly, the variable factors influencing landslide occurrence will be
investigated using satellite remote sensing techniques and a selection of historical landslide events that affected
parts of the Western Cape Province. Finally, the triggering mechanisms that lead to the landslide occurrence will
be investigated.
The combination of these activities will lead to techniques that can be used for landslide early warning systems
by identifying priority areas and monitoring conditions that can lead up to a landslide event if a triggering event
takes place.
2
1.3 THE RESEARCH QUESTIONS AND OBJECTIVES
The research project aimed to address the following research questions:
1. Can a combination of remote sensing and ancillary data be used to identify conditions leading up to
historical landslide events and their triggering mechanisms?
2. Can a combined GIS and remote sensing approach be used to create a landslide susceptibility map for
selected regions of the Western Cape?
Since investigations on landslide occurrence are based on the assumption that future landslides will occur under
similar conditions as historical landslides (Chung & Shaw 2000) the ultimate objective of the research is to
identify the static and variable conditions leading to historical landslides. This will include the identification of
the triggering mechanisms of those landslides. The results of this phase of the research will then be incorporated
with GIS modelling to create a landslide susceptibility map for the area of interest.The approach that will be
used when deriving landslide susceptibility maps is presented in Figure 1. Information on the landslide
controlling parameters is derived from a combination of existing maps, information derived from remote sensing
data as well as field-based measurements. These maps define thematic data layers which are used as input for
modeling landslide susceptibility maps. The data processing phases consists of using either expert knowledge or
computer algorithms or both to weight the relative importance of each of the controlling factors to landslide
occurrence.
3
Figure 1: The schematic diagram illustrating the standard procedure that is used when modelling landslide susceptibility
maps.
The specific objectives for the research are:
I. Identify landslide causative factors and further investigate their individual influence to slope instability
in the study areas.
II. Apply the weight of evidence method on landslide susceptibility mapping.
III. Apply the map combination approach on landslide susceptibility mapping, in order to compare the
success rate of these two models.
IV. Investigate the applicability of remote sensing as a monitoring system
V. Investigate the triggering mechanism for the landslides within the study area.
4
1.4 LANDSLIDES IN SOUTH AFRICA AND DESCRIPTION OF THE STUDY AREA
South African landslides tend to occur in mountainous regions experiencing high rainfall frequency (Singh et al.
2011). Areas that are highly susceptible to landslides are the Western Cape Mountains, eastern coastal regions
and the mountainous areas of the KwaZulu Natal Drakensberg (Paige-Green 1989). Several studies have
examined the occurrence of landslides in Southern Africa (e.g. Paige-Green 1989; Garland & Olivier 1993).
Durban frequently suffers from landslides and it has been found that housing developments and construction
work have contributed towards most slope failures (Garland & Olivier 1993). In South Africa debris flow occurs
in the KwaZulu Natal Drakensberg and in the Eastern Cape and Western Cape mountains (Lewis 1996,
Boelhouwers, Duiker,van Duffelen 1998). A debris flow deposit has been described by Hanvey, Lewis & Lewis
(1986) near Rhodes in the Eastern Cape and it was suggested that this debris deposit was related to the existence
of a former snow body and occurred under the Quaternary periglacial conditions. In the Eastern Cape debris flow
are extensive even at low altitudes (Lewis 1996). The large section of a road that slide away on the N2 between
Port Elizabeth and Grahamstown on the 21 October 2003 is another example. One example of a large paleo-
landslide is Lake Fududzi in the Limpopo province. This 2km long lake is located in the Soutpansberg Range,
and is an inland freshwater lake formed by a huge palaeo-landslide which blocked the course of the Mutale River
(Janisch 1931).The area most known for rockfalls in the Western Cape is the Chapman’s Peak drive along the
Cape Peninsula Atlantic coastline, prompting extensive structural improvements and removal of loose rocks
from the steep slopes (Singh 2009). Boelhouwers et al. (1998) investigated the morphology and sedimentology
of recent debris flow in the Western Cape Mountains. A debris flow deposit in the Cederberg Mountain of the
Western Cape has also been described by Boelhouwers et al. (1998). Further work from the Cape Province
describes debris flow studied in the Bushmans River Valley, which have been attributed to the heavy rainfall
events under contemporary climatic conditions (Lewis & Illgner 1998). Landslide distributions have been
observed in Du Toit’s Kloof area in the Western Cape, with 78 % of landslides investigated occurring in the
south-facing slopes, which is attributed to slope asymmetry by Boelhouwers et al. (1998).
To test the accuracy of landslide susceptibility modelling and the possibility to define a methodology for
landslide early warning systems, a study area in the Western Cape Province of South Africa was selected. The
area in question was known to be the subject of historical landslide events and the landslide locations and dates
of occurrence have been verified by field observations (Stapelberg pers com. 2011). The location of the study
area and some of the known landslide locations are presented in Figure 2.
5
Figure 2: The extent of the study area in the Western Cape Province of South Africa and a selection of field-verified
historical landslide positions.
The area of interest extends from 0 meters to 2298 meters above sea level with slopes ranging between 0 and
88⁰. The area receives a total annual rainfall in excess of ±822mm (2005 annual rainfall for the weather station
in Hermanus, South African Weather Services) with the majority of precipitation occurring during winter to
spring (May to September). Geologically the study area is situated in the Cape Fold belt which consists of the
Cape Supergroup, Karoo Supergroup and younger tertiary sediments capping the basement of pegmatites and
granitic intrusions of the Namaqua Natal belt. The predominant structural features are the large and small scale
folds, and faulting events associated with the Cape Fold Belt oregeny. Lithologically the Cape Supergroup
consists of the sandstone and shale sequence of the Table Mountain Group, the marine shales and sandstones of
the Bokkeveld Group and the sequence of sandstones and shales of the Witterberg Group. The Karoo Super
6
Group consists of the glacial deposit of the Dwyka Formation, the fluvial sandstones and mudstones of the
Beaufort Formation and the sequence of marine turbidites and shales of the Ecca Group (Johnson, Annhauser &
Thomas 2006). The different lithologies, the location of the lithological boundaries, dolerite contact zones and
the presence of faults and other structural features could have a significant impact on landslide susceptibility.
The impact of the various lithologies and structural features will be discussed in Section 2.2. The geological map
of the study area is presented in Figure 3. The sedimentary successions in the area support mostly Fynbos
vegetation types, which experience their growing season in spring and summer months.
7
Figure 3: Geological map of the study area.
8
CHAPTER 2 LITERATURE REVIEW AND THEORY
2.1 LANDSLIDE CLASSIFICATION

The term "landslide" is a part of a broad spectrum of mass movements involving the movement of surface
materials down a slope (Cruden & Varnes 1978). Standard classification schemes for different kinds of
mass movement do not exist although various researchers have suggested classification schemes based on
different criteria (Terzgaghi 1950; Carson & Kirkby 1972, Varnes, 1978, Cruden & Varnes in 1996).
For several years, the type of classification system used was primarily based on the type of movement
(e.g. Varnes 1978 & 1984). The most accepted landslide classification systems are based on different
factors such as:
• The material being transported (the terms rock, earth and debris are the terms generally used to
distinguish the materials involved in the landslide process. If less than 20% of the material is
greater than 2 millimetres in size, the material will be defined as earth. (Otherwise it will be
termed debris),
• The type of movement ( the main movement types are falls, slides and flows but usually lateral
spread, topples and complex movement are added to these),
• Movement velocity and
• Its current activity (this system is good particularly when evaluating future landslides and
currently active landslide).
Table 1 shows the schematic landslide classification system adapted from Varnes (1978) and modified by
Cruden & Varnes, in 1996.
9
Table 1: The schematic landslide classification system adapted from Varnes (1978).
Type of movement Type of material
Bedrock Engineering soils
Predominantly fine Predominantly
coarse
Falls Rock fall Earth fall Debris fall
Topples Rock topple Earth topple Debris topple
Slides Rotational Rock slump Earth slump Debris slump
Translational Few Rock block Earth block slide Debris block slide
units slide
Debris slide
Many Rock slide Earth slide
units
Literal spread Rock spread Earth spread Debris spread

Flows Rock flow Earth flow Debris flow
Rock avalanche Debris avalanche

Deep creep Soil creep
Complex and compound Combination in and/or in space of two or more principle types
of movement
Another classification scheme is based on the speed at which the material is transported in addition to the
moisture content of the material and the type of material being transported (Carson & Kirkby 1972).
10
Flow
er
Riv
et
W
Mudflow
Earthflow
Solifluction
Landslide
y
Dr
Rockslide Seasonal
Talus creep Soil creep
Slide Heave
Slow
Fast
Figure 4: A classification of mass movement processes on slope (Carson & Kirby 1972).
Irrespective of classification scheme used, the probability of having downslope movement of surface
materials are affected by specific controlling parameters. These parameters or factors are related to the
physical characteristics of the surface in question. The following sections aim to identify specific
landslide controlling parameters for the purpose of highlighting those parameters that will lead to an area
being susceptible to landslide occurrence.
2.2 LANDSLIDE CONTROLLING PARAMETERS
The identification of historic landslides and the analysis of the conditions leading to those landslide
events is critical when attempting to identify landslides controlling parameters (Campbell 1975; Clerici,
Perego, Tellini, Vescovi 2002; Morton, Alvarez, Glade 2005). The parameters affecting landslide
occurrences can be broadly grouped into two categories (1) preparatory factors, which make the area
susceptible to slope failure and (2) triggering factors, which sets off the movement (Crozier & Michael
1986). The parameters that affect an area’s susceptibility to landslide include (1) geology, (2)
geomorphology (3) human activities (4) and landcover (Pearce & O'Loughlin1995; Wu & Siddle 1995;
Atkinson & Massari 1998, Sidle, Dai, Lee, Li & Xu 1991, Sarkar & Kanungo 2004, Dahal et al. 2007,
11
Singh 2009). Hence, in landslide hazard assessment practice, the term “landslide susceptibility mapping”
is addressed without considering the variable factors in determining the probability of occurrence of a
landslide event (Dai et al. 2001). The investigation of triggering mechanisms such as earthquakes and
rainfall are critical but determining the magnitude and temporal behaviour of these parameters and how it
relates to landslide susceptibility has proved to be challenging (Sarkar & Kanungo 2004). The following
sections describe some of the controlling parameters affecting landslide development. These factors have
been subdivided into three categories, each contributing to a separate category of landslide causative
factors (preparatory parameter or triggering mechanism). They are:
1. Static factors – These factors are those that are unlikely to change within a short period of time
like geology, the geomorphology, the soil type and depth and the vegetation type – these define
the landslide preparatory factors.
2. Variable factors – These are the highly variable factors that can vary seasonally to daily including
vegetation health and productivity and soil water contents – these contribute to both preparatory
factors and triggering mechanisms.
3. Triggering mechanisms – These are the mechanisms that, when both static and variable
conditions are favourable for landslide occurrence, will cause a landslide. Potential triggering
mechanisms include high intensity rainfall events and/or seismic activity.
The premise behind the subdivision lies in the fact that the static factors will define the area's
susceptibility to landslide occurrence (Dahal et al. 2007). For instance, at specific geomorphology and
landcover classes, a specific area may be highly susceptible to landslide occurrence. The variable factors
will then define the likelihood of a landslide occurring in the near future. For instance, a dry spell may
cause the health and productivity of vegetation in a susceptible area to decline rapidly, increasing the
likelihood of landslide occurrence. This then creates a scenario where a triggering event will cause a
landslide to occur.
The following sections investigate the static and variable factors affecting landslide occurrence and
describes the potential triggering mechanisms associated with landslide activity.
12
2.2.1 Static factors
2.2.1.1 Geology
The geology of an area is a critical parameter controlling the occurrence of landslides and various studies
have used geology as a parameter when modeling landslide susceptibility maps (e.g. Dahal et al. 2007,
Singh 2009, Chauhan, Sharma, Arora, Gupta 2010; Temesgen et al. 2010; Singh et al. 2011). Different
lithologies have different chemical and physical properties leading to different susceptibility to mass
movement. For example, different rock types have different hydrological properties i.e. transmisivity,
hydraulic conductivity and permeability (Varnes 1984). These properties play a significant role on slope
instabilities during rainfall events. Hence shales and siltstones are considered to be more susceptible to
slope instability, while sandstones and conglomerates are regarded to have moderate to low
susceptibilities to landslide occurrence (Stapelberg pers com. 2011). Singh et al. (2011) has emphasized
the influence of a dip of the strata, abrupt changes in lithological characteristics, geological structure and
bedding planes, on slope instability. The sequence of the stratigraphy can also determine the stability of
the area. One such example is a sequence that consists of an impermeable layer on the bottom, which is
overlain by a permeable layer. Such a sequence would have higher potential to saturate with water during
rainfall events, resulting in a higher susceptibility to landslide occurrence. Additionally, the presence of
dykes and sills are of importance since they could have weakening effects on the lithologies (Singh 2009).
The structural features on the area of interest may also influence landslide occurrence. In this regard,
properties including the dip of the strata and the presence of faults and lithological boundaries may
signify zones of weakness along which slope failures may occur (Dahal et al. 2007). The combination of
rock types and structures in an area will dictate the resistance to weathering and erosion processes and
ultimately, landslide susceptibility (Singh et al. 2011).
2.2.1.2 Geomorphology
The geomorphology of the area has been found to be the most important controlling parameter by several
authors (Sarkar & Kanungo 2004). Information on the geomorphology, including slope, aspect and
profiles can be derived from digital elevation models of the area of interest using GIS techniques. Slope is
the most substantial parameter influencing landslide development. On a slope of uniform isotropic
material, increased slope correlates positively with increased likelihood of failure (Chauhan et al. 2010).
In order to assess the contribution of various slope gradients to the development of landslides, it is
necessary to know the spatial distribution of the slope categories, which can be obtained from a DEM
(Dai & Lee 2002). The other important geomorphologic parameter is relative relief. Landslides generally
13
occur in high relative relief areas. The relief of the area is defined as the difference between maximum
and minimum elevation values within the area. This parameter can be computed using DEM (Chauhan et
al. 2010). Aspect is one of the most important parameter as it directly and indirectly influences the area’s
susceptibility to slope failure. South facing slopes are generally less vegetated in Southern Africa, as there
is limited amount of sunlight reaching the south facing slopes. It is well known that sunlight is vital for
vegetation health, and slopes with healthy vegetation are generally less susceptible to slope failure
(detailed explanation in 2.2.2.1). Secondly South facing slopes receive limited amount of sunlight in the
Southern hemisphere, therefore they are wetter, and more susceptible to landslide occurrence (Stapelberg
pers com. 2011). ). Landslide distribution has been observed in Du Toit’s Kloof area in the Western Cape,
with 78 % of landslides investigated occurring in the south-facing slopes (Boelhouwers et al. 1998),
which signifies the importance of aspect on landslide investigations in South Africa.
2.2.1.3 Landcover
While landcover is not strictly “static” it is regarded to be relatively stable over the course of few months.
It does not change daily just like rainfall and vegetation. Landcover can be defined as the observed
physical and biological cover on the earth's surface. Glade (2002) concurs that vegetation cover is an
important factor influencing the rate of surface runoff, which enhance chances of landslide occurrence.
For instance barren slopes are more likely to have landslide occurrence. In contrast vegetative areas tend
to reduce the action of rainfall thereby preventing the erosion due to the anchorage provided by the tree
roots (Gray & Leiser 1982; Greenway 1987, Styczen & Morgan 1995). In general, sparsely vegetated
areas are associated with higher runoff during rainy seasons when compared to densely vegetated areas.
Similarly, the type of vegetation would have an impact on slope stability (i.e. forested areas are expected
to be more stable than grassland).
Different soil types have different properties such as grain size, porosity, transmisivity and hydrolic
conductivity, therefore different soils have diverse influence on susceptibility to slope failure. Clay rich
deep soils are considered to be more susceptible to landslide, in comparison with sandy shallow soils
(Stapelberg pers com. 2011). An increase in absorbed moisture is a major factor in the decrease in
strength of cohesive soils (Zhou 2006). Dahal et al. (2007) have also emphasized the importance of soil
type as a parameter when modeling landslide susceptibility maps using the weight of evidence method. It
has also been noticed that soil depth between 0.5-2 meters have maximum susceptibility to landslide
(Dahal et al. 2007). It is therefore important to input soil depth as a static parameter when modeling a
landslide susceptibility map.
14
2.2.1.4 Anthropogenic influences

Human-induced changes can affect an area’s susceptibility to landslides and must be understood when
assessing landslide potential of an area. Examples of such activities are road-cuts, deforestation, mining
artificial vibrations, and cutting of slope toe during construction. One of the controlling factors for the
stability of slopes is road construction activity (Dahal et al. 2007). Road cuts in mountainous areas can
make the area susceptible to slope instability. One such example is the Chapman’s Peaks drive in the
Western Cape Province. Unsuitable construction on mountainous areas can also cause slope failures - it
has been documented in Durban, KwaZulu Natal, that urban construction has caused some areas to be
susceptible to slope failure (Garland & Olivier 1993). The increase in moisture content in the soil or
changing the form of a slope can increase the area’s susceptibility to landslide (Garland & Olivier 1993).
Development activities such as cutting and filling along roads and the removal of forest vegetation are
also capable of greatly altering slope form and ground water conditions and therefore increasing the
susceptibility to landslide occurrence (Swanson & Dyrness 1975). These altered conditions may
significantly increase the degree of landslide hazard present (Sidle, Pearce & O’Loughlin 1985). Trees act
as natural anchors during rainy seasons, and therefore reduce the effect of rainfall on erosion (Gray &
Leiser 1982). Deforestations therefore can cause an area to be more susceptible to slope failure, during
rainy seasons. The positive influences of vegetation on slope failure have been discussed on section
2.2.1.3. In South Africa, the positive influence of vegetation to slope failure has also been emphasized by
Stapelberg (pers com, 2011).
2.2.2 Variable factors
2.2.2.1 Vegetation
As mentioned previously, the vegetation in an area has a significant impact on slope instability and
various studies have emphasized the significance of vegetation on slope failure (Gray & Leiser 1982;
Greenway 1987, Styczen & Morgan 1995). However, it is not only the vegetation type that governs
landslide susceptibility, but also the health and productivity of vegetation at a specific time. The effect of
vegetation on slope stability appears to be complex in that, depending on local conditions of soil depth,
soil type, slope and vegetation, a vegetation cover in some ways definitely promotes stability and in other
ways it may not. In a review of behaviour of vegetation on slope stability, Prandini et al. (1977) makes the
following points regarding the beneficial effects of forest cover: as a whole forest cover reduces the action
of climatic agents on natural mass, in a manner favourable to slope stability by:
(1) Intercepting and protecting the mass from the action of sunshine, winds and rains,
15
(2) Retaining a considerable amount of rain water by wetting the large surface made up of
leaves, branches, and trunks and eliminating the water as a vapour,
(3) Eliminating, as a vapour, a large amount of water from the ground by means of
evapotranspiration, and
(4) Vegetal debris on the forest floor immobilizes a large amount of water and cuts down on
runoff and erosion.
When identifying conditions leading up to landslide events, the identification of the vegetative conditions
prior to landslide occurrence can be performed. In this regard, landslides may occur preferentially in areas
with little vegetation or in areas where vegetation is stressed due to drought or disease.
2.2.2.2 Soil water content

In addition to the vegetative conditions, the wetness of the soil as an indication of soil moisture is known
to play a role on slope stability (Ray, Jacobs & de Alba 2009). Saturated soils are believed to be more
prone to instabilities and would therefore have a higher probability of landslide occurrence. Certain clay
minerals react to the presence of water and cause volume changes of the clay mass. The relationship
between an increase in absorbed moisture and the decrease in the cohesive strength of soils is shown
schematically in Figure 5. Water absorbed by clay minerals causes increased water contents that decrease
the cohesion of clayey soils (Zhou 2006). These effects are augmented if the clay mineral happens to be
expansive, e.g., montmorillonite (Zhou 2006). Groundwater and soil moisture therefore play a critical role
in triggering slope failure (Ray et al. 2009).
Figure 5: Effect of water content on cohesive strength of clay (Zhou 2006).The x-axis shows the water content and
the y-axis it shows the cohesive strength.
16
2.2.3 Triggering mechanisms
2.2.3.1 Seismicity
Natural and human-induced seismicity could trigger landslides and other mass-movement events
(Borcherdt 1970, Harp 1991, Griggs 1998). Earth quakes can also trigger landslides in some areas
(Spudich, Hellweg & Lee 1996). In this regard, a magnitude 6 earthquake that struck the town of Ceres in
1969 was associated with rock-falls and other mass movement events (Singh et al. 2011). Although
seismically triggered landslides can be disastrous, South Africa is generally regarded to be seismically
inactive (Singh et al. 2011). Earth tremors in South Africa are generally associated with either naturally
occurring earthquakes or earth tremors associated with mining activities. Seismically triggered landslides
are widespread phenomena within tectonically active mountain ranges.
2.2.3.2 Rainfall
Rainfall is a trigger for several landslides around the globe (Iverson 2000, Cardinali 2005) and in
mountainous areas of South Africa (Singh et al. 2011). Water is recognized to be a factor almost
important as gravity in slope instability (Varnes 1984). Landslides triggered by rainfall are caused by the
buildup of water pressure into the ground (Cambell 1975; Wilson 1989). Iverson (2000) has also linked
slope failure and landslide motion to groundwater pressure heads that change in response to rainfall. van
Schlkwyk & Thomas (1991) have argued that prolonged precipitation events associated with high
intensity rainfall are often the trigger for landslides in South Africa i.e. heavy rainfall of September 1987
and February 1988 occurring in KwaZulu Natal. High intensity and short rainfall duration can trigger
mostly shallow landslides and debris flows in relatively high permeability soils (Corominas & Moya
1999; Corominas 2000). Whereas long rainfall periods characterized by low to moderate average rainfall
intensity can initiate shallow and deep-seated landslides in low permeability soils and rocks (Cardinali,
Galli, Guzzetti, Ardizzone, Reichenbach, Bartoccini 2005).
2.3 REMOTE SENSING AND GIS FOR LANDSLIDE SUSCEPTIBILITY MAPPING
Techniques for landslide mapping have changed little, in principle, over the past few decades even when
newer data sources become available (Sarkar & Kanungo 2004).Landslides are most often detected and
mapped by a combination of interpretation of air photos or multispectral digital imagery and selected
ground verification information (Roering & McKean 2004), and is often based on “professional
judgment” (Wieckzorek 1984). There has been a drastic increase in magnitude and frequency of natural
disasters around the globe but at the same time there has been improvements in the technical capabilities
17
to mitigate them. The increased efficiency of computers has created opportunities for detailed rapid
analysis of natural hazards. The acquisition of information through remote sensing and spatial data
analysis using GIS has improved the capabilities of geo-informatics in the field of disaster management
(Dahal et al. 2007). The following section describes some of the GIS techniques and remote sensing tools
that have been used for landslide susceptibility mapping and early warning.
2.4 GIS MODELLING FOR LANDSLIDE SUSCEPTIBILITY MAPPING
Landslide hazard is normally depicted on maps which show spatial distribution of hazard classes. The
development of these maps requires knowledge of the processes active in the area being studied
(geological, hydrological, land-cover, and morphological factors), as well as triggering mechanisms
leading to the occurrence of landslides (e.g. rainfall and seismicity) (Kanungo et al. 2009). Landslide
hazard maps typically aims to predict where failures are likely to occur without any clear indication of
when they are likely to occur. However, the focus on time-based modelling techniques have proved to be
useful for providing landslide hazard information needed for planning and protection purposes (e.g.
Brunettii et al. 2010 ).
Geographic information systems and the selection of parameters that are deemed to influence landslide
occurrence in a certain area and the consequent preparation of corresponding thematic data layers are
crucial components of models for landslide susceptibility mapping (Sarkar & Kanungo 2004). The
parameters that are generally deemed to govern instabilities include geology, geomorphology, land use,
climatic conditions, hydrology, vegetation and geohydrology (Dahal et al 2007). These factors can vary
both locally and/or regionally. The derivation of landslide susceptibility maps involves the combination
and integration of spatial information on these factors to provide an indication of the areas where the
combination of factors is such that they create an environment conducive to landslide occurrence.
Different approaches have been used to weight landslide controlling parameters and to model landslide
susceptibility maps. The choice of the appropriate technique strongly depends on the nature of the
problem, the observation scale and data availability (Temesgen et al. 2001, Lee, Choi. & Min 2004,
Sarkar & Kunongo 2004). Landslide susceptibility mapping approaches can be grouped into two broad
categories; qualitative and quantitative (Glade & Crozier 2005). In the qualitative approach, a lot of
subjectivity is introduced in preparation of various thematic data layers contributing for landslide
occurrences, which are integrated in a GIS to create a landslide susceptibility map of the area (Kanungo et
al. 2009). The quantitative approach focuses on developing the ways of quantifying the relative
18
importance of various causative factors (Kanungo et al. 2009). A classification of the different approaches
for landslide susceptibility mapping is given in Figure 6 and a summary of different techniques is
provided in Table 2.
Figure 6: The taxonomy of the different weighting approaches when conducting landslide susceptibility modelling
(Source: Kanungo et al. 2009 pp 11).
Table 2: The comprehensive review of the different GIS techniques that have been used for landslide susceptibility
modelling.
Method Description
Qualitative Approach
Distribution This method is also known as landslide inventory and provides a spatial distribution
Analysis of existing landslides represented on a map either as the affected area (polygon) or as
point events (Wieczoreck 1984 & 1987).
Disadvantage: it does not relate landslides to their causative factors
Advantage: it is economic and can cover a large area
19
Map The map combination approach is a simple procedure that combines different
Combination thematic maps based on the knowledge of the expert. This approach involves The
following steps (Soeters & van Westen 1996):
1. The Selection and mapping of landslide controlling parameters.
2. Thematic data layer preparation with relevant categories of the parameters.
3. Assignments of weights and rankings to parameters and their categories

respectively.
4. Integration of thematic data layers.
5. Preparation of landslide susceptibility map showing different zones.
Disadvantage: It strongly depends on expert knowledge and therefore can inherit

human error and bias judgment.
Advantage: It is simple as compared to the other methods, which normally use
complex equations.
Quantitative Approach
Probabilistic This approach compares the spatial distribution of landslides in relation to different
Approach causative factors. It is based on the Bayesian probability. Some models based on this
approach include conditional probability model, Weight of evidence method,
certainty factor method under favourability mapping model, etc.
Disadvantages: It requires known landslide points as an input data set and can over
estimates if the number of known landslide points is too much. Therefore random
selection of the landslide point that would be used is crucial.
Advantages: The fact that it uses known landslides points makes it the most
suitable model for landslide susceptibility mapping, as landslide studies are based on
the assumption that future landslides will occur under similar circumstances as
historic landslide.
Artificial Neural Amongst others Gomez & Kavzglu (2005) used Artificial Neural Networks (ANN)
Network Based black box approach for landslide susceptibility mapping. In this process, multilayer
Approach perceptron with back propagation learning algorithm are used. The approach uses a
wide range of causative factors and the existing landslide distribution layer derived
from DEM, remote sensing imagery and field data for neural network training and
20
testing. Afterwards, existing landslides are considered to validate the landslide

susceptibility map. Recently Chauhan et al. (2010) modified the ANN by creating a
rating system that depicts the influence of each category on a parameter, on landslide
occurrence. Satisfactory results were obtained.
Disadvantages: There are no known disadvantages for this model as it has not been
applied extensively on landslide susceptibility mapping.
Advantages: It uses both known landslide localities and areas known not to have
landslide occurrences, therefore it is less likely to over predict.
Fuzzy Set Based This model was proposed by Elias & Bandis (2000) for landslide susceptibility
Approach mapping. Fuzzy linguistic rules are used to assign fuzzy membership values to
different categories of thematic data layers. The fuzzy membership values are used to
provide data to the input neurons for neural network model. A single output neuron
with values from 0 to 1 is considered to represent the degree of landslide
susceptibility based on actual landslide data. The back error propagation neural
network is used for training and a landslide susceptibility map is prepared.
Bivariate In bivariate statistical analysis, each individual thematic data layer is compared to the
Statistical existing landslide distribution layer. The weighting value of each category of the
analysis controlling parameter is assigned based on landslide density. It is based on this

equation:
 ρLandslides / parameterclass 
LSI = ∑ log e  
 ρLandslide  {1}
Advantage: it provides a good combination between expert-derived parameter
choices and quantitative spatial analysis-It renders quantitative and objective
measure on landslide susceptibility.
Disadvantage: it assumes complete independence of input parameter.
Multivariate approaches consider relative contribution of each of the thematic data

Multivariate layer to the total susceptibility within a defined area. The procedure involves several
Statistical important steps ( Aleotti & Chowdhury 1999):
Analysis 1) Identification of percentage of landslide affected areas in each pixel and their
classification into stable and unstable zones.
21
2) Preparation of an absent/present matrix of given category of a given thematic

layer.
3) Multivariate statistical analysis and reclassification of the area based on the
results and their classification into susceptibility classes.
2.4.1 Remote sensing on landslide susceptibility mapping

Satellite remote sensing techniques can be used to derive the thematic information needed to synthesize
landslide susceptibility maps and to monitor the variable parameter influencing landslide susceptibility
(e.g. Lee, Choi & Min 2004b). The advancements in digital image processing has provided additional
tools such as data fusion or data merging, enhancement, classification and accuracy assessment
techniques. To put these technical advancements into good use the interpretation of remote sensing data
should focus on extracting information related to the following features:
• Distinctive features associated with slope movement
• Morphological expression of landslides
• Landslide characteristics including size, shape and contrast to surrounding areas
In this regard, the interpretability of remote sensing data is strongly influenced by the contrast that results
from the spectral differences between landslides and its surroundings.
Satellite remote sensing techniques can contribute to landslide investigations at three distinct phases
namely: (a) detection and classification of landslides (b) monitoring landslide movement and
identification of conditions leading up to an event, and (c) analysis and prediction of slope failures
(Morton et al. 2003). Various sources of remote sensing data can contribute to these phases including
medium and high resolution optical data, synthetic aperture radar (SAR) data and LiDAR data (Joyce,
Samsonov, Levick 2011).
Synthetic Aperture Radar (SAR) data can be employed for the detection and classification of landslides
through the analysis of radar backscatter or as early warning by detecting slow-moving landslides through
differential interferometry techniques (Joyce et al. 2011). LiDAR data through the derivation of very high
resolution digital terrain models (DTMs) is useful for the delineation of landslide morphological features
(Joyce et al. 2011). Furthermore, the high resolution DTM data will provide high quality
geomorphological information for landslide susceptibility mapping. In addition to SAR and LiDAR data,
high resolution optical data including aerial photographs have been commonly used for the detection and
classification of landslides (Mantovani, Soeters & van Western 2000). After detection of landslides, the
22
movement of the landslide can be monitored. This involves a comparison of conditions associated with
landslides over time including the aerial extent of the landslide, the speed of movement and the changes
in surface topography. Here optical, SAR and LiDAR data can be used in combination with several
change detection algorithms (Joyce et al. 2011).
Although the detection and classification of landslides through the techniques described above is
important to define landslide controlling parameters, the ideal is to use remote sensing data for monitoring
of areas susceptible to landslide occurrence in an effort to provide an early warning. In this regard, optical
remote sensing data can be used to monitor the variable condition (vegetation health and productivity and
soil water content) that makes an area susceptible to landslide occurrence.
To monitor the health and productivity of the vegetation in an area, optical remote sensing data have
frequently been used. Using remote sensing data through normalized difference vegetation indices
(NDVIs) and tasseled cap greenness components, the vegetative conditions prior to landslide occurrence
can be identified. In addition to the vegetative conditions, the wetness of the soil as an indication of soil
moisture is known to play a role on slope stability. Saturated soils are believed to be more prone to
instabilities and would therefore have a higher probability of landslide occurrence. In this regard, satellite
remote sensing data could assist with the identification of the moisture content of soils through tasseled
cap analysis and consequent analysis of the derived wetness component. Although not perfect yet,
research is ongoing on the use of SAR backscatter for soil moisture retrieval (Wagner & Pathe 2008)
2.4.1.1 Normalized difference vegetation index (NDVI)

The Normalized Difference Vegetation index (NDVI) is the simple equation that has been used for
several years to calculate vegetation health. It is an index of plant greenness or photosynthetic activity,
and is one of the most commonly used vegetation indices (Anderson, Hanson, Haas 1993.). The
Normalized Difference Vegetation Index (NDVI) is related to the proportion of photosynthetically
absorbed radiation. Many natural surfaces are about equally as bright in the visible red and near-infrared
part of the spectrum with the notable exception of green vegetation. Red light is strongly absorbed by
photosynthetic pigments (such as chlorophyll) found in green leaves, while near-infrared light either
passes through or is reflected by live leaf tissues, regardless of their colour (Stoner & Baumgardner
1980). This means that areas of bare soil having little or no green plant material will appear similar in
both the red and near-infrared wavelengths, while areas with much green vegetation will be very bright in
the near-infrared and very dark in the red part of the spectrum. Put another way, for healthy living
23
vegetation, this ratio will be high due to the inverse relationship between vegetation brightness in the red
and infrared regions of the spectrum (Paruelo, Epstein, Lauenroth & Burke 1997). The Normalized
Difference Vegetation Index (NDVI) is a simple numerical indicator that has been used to analyse remote
sensing measurements, typically but not necessarily from a space platform (Perry & Laulenschlager,
McFeeters 1996, Pats & Charez 1996). This technique can also be used to gather information about the
land cover of the study area (Xiao, Zhang, Yan, Wu , Biradar, 2009, McCloy 2010).The NDVI algorithm
subtracts the red reflectance values from the near-infrared and divides it by the sum of near-infrared and
red bands.
NDVI = (NIR-RED)/ (NIR+RED) {2}
The spectral reflectance curve of green and dry vegetation and soil is presented on Figure 7.
Figure 7: The spectral reflectance curve of green and dry vegetation and soil along with the spectral wavelength
(Clarck et al.1999).
2.4.1.2 Normalise difference water index
The normalised difference water index (NDWI) is derived using similar principles to the Normalised
Difference Vegetation Index (NDVI). In an NDVI (the comparison of differences of two bands, red and
near-infra-red (NIR), the presence of terrestrial vegetation and soil features is enhanced while the
24
presence of open water features is suppressed because of the different ways in which these features reflect
these wavelengths (McFeeters 1996). The NDWI is derived from the Near-Infrared (NIR) and Short
Wave Infrared (SWIR). The SWIR reflectance reflects changes in both the vegetation water content and
the spongy mesophyll structure in vegetation canopies, while the NIR reflectance is affected by leaf
internal structure and leaf dry matter content but not by water content (Rahman 2011). The combination
of the NIR with the SWIR removes variations induced by leaf internal structure and leaf dry matter
content, improving the accuracy in retrieving the vegetation water content. The NDWI is calculated
according to the following equation (Gao 1996):
NDWI = (SWIR – NIR) / (SWIR + NIR) {3}
2.5 FINAL REMARKS
Landslides have been extensively studied, dating back to 1970s. There have been only minor changes in
the principles of landslide susceptibility mapping over the years but the evolution of remote sensing and
GIS has contributed significantly to the efficiency and accuracy of landslide susceptibility maps.
Landslide susceptibility mapping strongly depends on the identification of landslide causative factors and
weighting of the causative factors in accordance with their significance to slope instability. The landslide
causative factors can be categorized into three classes (1) static factors (i.e. slope, aspect, geology,
infrastructure, landcover, soil type, and soil depth), (2) variable factors (i.e. vegetation and soil moisture)
and (3) triggering factors (i.e. rainfall, seismicity and human causes). The static factors are the only
parameter used to model a landslide susceptibility map and the variable factors can be used to monitor the
changes that make an area susceptible to landslide and can therefore be used as an early warning system.
25
CHAPTER 3 GIS-BASED LANDSLIDE SUSCEPTIBILITY MAPPING:

MATERIALS, RESULTS AND ACCURACY ASSESSMENT
Landslide susceptibility mapping is the technique that is used to zone those areas that are likely to have
landslides, using different data sets. GIS and remote sensing have been used extensively when deriving
landslide causative thematic maps. A geographic information system is a major tool that is used to
integrate different landslide causative thematic maps when modeling landslide susceptibility maps. The
landslide causative factors that are used when modeling a landslide susceptibility map are discussed on
Section 2.2. Landslide susceptibility mapping methodologies are based on the assumption that future
landslides will occur under similar conditions as historic landslides. For this reason the weight of
evidence method was used to model a landslide susceptibility map of the study area. Moreover, the
weight of evidence method was chosen because of its proven success in landslide susceptibility mapping
(Athanasopoulos, Pelekis, Leonidou 1995). The map combination approach is also evaluated in this study,
mainly because it does not use landslide inventory data as one of the inputs when modeling a landslide
susceptibility map, unlike the weight of evidence. The following section describes the input data available
for susceptibility mapping by considering the static controlling parameters discussed in Section 2.2 of
Chapter 2. This is followed by the description of the landslide susceptibility mapping techniques
employed and presents the results and accuracy of the landslide susceptibility maps.
3.1 INPUT DATA
The two models require a variety of input data set. The Cape datum was used. These inputs are discussed
below.
3.1.1 Geology
The 1:250 000 geological map was attained from the Council for Geoscience. The main lithological types
that were encountered within the study area were: shales, sandstone, siltstone, conglomerate, glacial
deposits (Dwyka) and unconsolidated alluvium deposits. The granites of the Cape Granites suit were also
prominent. The geological layer is presented in Figure 8. The structural features such as faults, dykes,
lineaments and lithological contacts will affect the area’s susceptibility to landslides, as discussed in
Section 2.2.1.1. Since geological mapping took place at 1: 250 000 scale, a 250 m buffer was applied to
the faults and lithological contacts data.
26
Figure 8: The geological parameter of the study area, the lithological units are shown on the legend. This
geological layer was one of the causative thematic layers used to modell a landslide susceptibility map of the study
are using the weight of evidence method.
The faults and lithological contacts layer is presented on Figure 9, where a buffer of 250 metre was
applied.
27
Figure 9: The 250 meter buffered lithological contacts and faults layer for the study area. This thematic layer was
one of the causative thematic layers used to model a landslide susceptibility map of the study area. The localities for
the landslides are also shown.
The numbers of landslides per stratigraphic unit is presented on Figure 10, a large number of landslides
fall on the Quaternary Alluvium deposits, Skurweberg, Rietvlei and Peninsula stratigraphic units, in
decreasing order. It is observed that a large number of landslides fall on the unconsolidated alluvium
deposits. The Peninsula, Rietvlei and Skurweberg stratigraphic units are dominated by sandstones and
quadzites lithologies, which are generally considered to be moderately susceptible to landslide
occurrences. These stratigraphic units recorded a large number of landslides. It is also observed that 53
landslides out of 93 landslide localities used to run the weight of evidence method fall within the 250
meter faults and lithological contacts buffer. This indicates that areas with faults and lithological contacts
are generally more susceptible to landslide occurrence.
28
Figure 10: The graph shows the number of landslides per lithology. On the y-axis is the stratigraphic units and on
the x-axis is the number of landslides The alluvium deposits recorded the highest number of landslides and the
Skurweberg, Rietvlei and Peninsula stratigraphic units recorded the second highest, and the third highest number of
landslides, in decreasing order.
29
3.1.2 Geomorphology
The geomorphological factors affecting landslide susceptibility (including slope and aspect) were derived
from the 20 meter resolution digital elevation model (DEM). The digital elevation model was interpolated
from a 20 meter contour map, attained from National Geo-spatial Information (NGI). The slope and
aspect layers are presented on Figure 11 and Figure 13, respectively; the known landslide localities are
also shown on these layers.
Figure 11: Slope layer of the study area. The classes that were used in the weight of evidence method are also
shown on the legend.
Figure 12 shows the number of landslides per slope class. Approximately 90 percent of the landslides fall
on the 0-20° and 20-40°classes. This is no surprise as slopes greater than 18° in South Africa are
considered to be susceptible to landslide occurrence. There is only one landslide on slopes greater than
30
40°, it is more likely that the landslide is a rockfall which are normally caused by road cuts on
mountainous areas.
60
50
Number of Landslides
40
30
20
10
0
0-20 degrees 20-40 degrees 0 degrees 40-60 degrees >60 degrees
Slope class
Figure 12: The number of landslides per slope class. Roughly 90 % of the landslides fall on the 0-20° and 20-40°
class.
31
Figure 13: The aspect layer for the study area. The classes are shown in the legend. The aspect thematic layer was
one of the causative layers used to model a landslide susceptibility map of the study area, using the weight of
evidence method. The black dots on the map are the localities for the known historic landslides within the study
area.
The number of landslides per aspect class is presented on Figure 14, interestingly all the aspect classes
recorded at least one landslide. A large number of landslides fall on the South and South Western slopes.
This can be explained by the fact that south facing slopes are generally wetter as they receive a limited
amount of sunlight as compared to the north facing slopes. Moreover the south facing slopes are less
vegetated when compared to the north facing slopes. This makes them more susceptible to landslide
occurrence. The north facing slopes have a smaller number of landslides, this is no surprise as the north
facing slopes receive a large amount of sunlight and therefore generally dry in comparison to the south
facing slopes. The east facing slope have only one landslide points.
32
Number of landslides per Aspect class

N
35
30
NW 25 NE
20
15
10
5
W 0 E Number of landslides
per Aspect class
SW SE
S
Figure 14: The number of landslides per aspect class. The south facing slope recorded the highest number of
landslides and the east facing slopes recorded the least number of landslides.
3.1.3 Landcover
Landcover parameters that affect the area’s susceptibility to slope instability include (biological and
physical landcover, soil depth and soil type). The significance of these parameters on slope instability has
been discussed on Section 2.2.1.3 of Chapter 2. The biological and physical landcover layer of the study
area was acquired from the CSIR (this is the National Landcover 2000 (NLC2000) product). The soil type
and soil depth layer of the study area was acquired from the ARC Institute of Soil Climate and Water
(ICSW). The landcover layer (Figure 15), soil depth layer (Figure 17) and the soil type map (Figure 19)
were some of the input parameters that were in the weight of evidence method and map combination
approaches.
33
Figure 15: The land cover layer used when modelling the landslide susceptibility map of the study area. The
landcover classes are shown in the legend.
The number of landslides per landcover class is shown in Figure 16. It is observed that more than 80% of
the landslides fall on the shrubland and low fynbos. The shrubland and low fynos is the most dominant
class (in area), in the study area.
34
Figure 16: The number of landslides per landcover class. A large number of landslides fall on the Shrubland and
low fynbos class. The other landcover classes recorded a very low number of landslides.
Soil depth is considered to be one of the most important causative factors when modelling landslide
susceptibility maps. The soil depth map of the study area is shown on Figure 17. The localities for known
landslide points are represented by the black dots on the layer.
35
Figure 17: The soil depth layer of the study areas. The depth classes are also shown on the legend. This thematic
layer was one of the causative parameters used to model a landslide susceptibility map of the study area.
The number of landslides per soil depth class is presented on Figure 18. A large number of landslide
points fall on soil depth less than 300 mm, 300 mm- 900 mm, and 600-900 mm; in decreasing order.
Looking at the soil depth map on Figure 17, it is visible that these three classes are the most dominant soil
depth classes in the study area. The soil depth greater than 1200 mm makes far less than 1 percent of the
study area and the 900-1200mm class makes roughly 1% of the study area.
36
45
40
35
Number of landslides
30
25
20
15
10
0
<300mm 300-600mm 600-900mm 900-1200mm >1200mm
Depth Class
Figure 18: The number of landslides per soil depth class. More than 90 % of the landslides fall on the soil depth less
300 mm, 300-600mm and 600-900 mm; in decreasing order. Soil depth less than 300mm recorded the highest
number of landslides and the soil depth between 900-1200mm recorded the least number of landslides.
The soil type thematic map was also used as the input parameter when modelling the landslide
susceptibility map using the weight of evidence method. The soil type thematic layer is presented on
Figure 19.
37
Figure 19: The soil type layer for the study area, which was used to model the landslide susceptibility map using the
weight of evidence method. The description for the codes used in the legend is shown on the Table 3.
The numbers of landslides per soil type class are presented on Figure 20, a large number of landslides fall
on class Lb, Lc, Fb and Hb in decreasing order. The Lb and Lc classes are dominated by rocky outcrops,
as explained on Table 3. Such a high number of landslides on these classes were unexpected. The Fb class
is dominated by shallow lime rich soils, and shallow soils are generally low to moderately susceptible to
landslide occurrence. The only class that is known to be more susceptible to landslide occurrence is Hb.
This class consists of deep grey sands, deep soils are considered to be more susceptible to landslide
occurrence according to some landslide researchers.
38
Table 3: The description for the soil type classes (source: ARC institute for Soil, Climate and Water).
Broad
soil
pattern Description
code
Aa Freely drained, red and yellow apedal soils with humic topsoils comprise >40% of the land type
Ab Freely drained, red and yellow, dystrophic/mesotrophic, apedal soils comprise >40% of the land
type (yellow soils <10%)
Ac Freely drained, red and yellow, dystrophic/mesotrophic, apedal soils comprise >40% of the
land type (red and yellow soils each >10%)
Ad Freely drained, red and yellow, dystrophic/mesotrophic, apedal soils comprise >40% of the land
type (red soils comprise <10%)
Ae Freely drained, red, eutrophic, apedal soils comprise >40% of the land type (yellow soils
comprise <10%)
Af Freely drained, red, eutrophic, apedal soils comprise >40% of the land type (yellow soils
comprise <10%); with dunes
Ag Freely drained, shallow (<300 mm deep), red, eutrophic, apedal soils comprise >40% of the
land type (yellow soils comprise <10%)
Ah Freely drained, red and yellow, eutrophic, apedal soils comprise >40% of the land type (red and
yellow soils each comprise >10%)
Ai Freely drained, yellow, eutrophic, apedal soils comprise >40% of the land type (red soils
comprise <10%)
Ba Red and yellow, dystrophic/mesotrophic, apedal soils with plinthic subsoils (plinthic soils
comprise >10% of land type, red soils comprise >33% of land type)
Bb Red and yellow, dystrophic/mesotrophic, apedal soils with plinthic subsoils (plinthic soils
comprise >10% of land type, red soils comprise <33% of land type)
Bc Red and yellow, eutrophic, apedal soils with plinthic subsoils (plinthic soils comprise >10% of
land type, red soils comprise >33% of land type)
Bd Red and yellow, eutrophic, apedal soils with plinthic subsoils (plinthic soils comprise>10% of
land type, red soils comprise <33% of land type)
Ca Land type qualifies as Ba-Bd, but >10% occupied by upland duplex/margalitic soils
Da Duplex soils (sandier topsoil abruptly overlying more clayey subsoil) comprise >50% of land
type; >50% of duplex soils have red B horizons
Db Duplex soils (sandier topsoil abruptly overlying more clayey subsoil) comprise >50% of land
39
type; <50% of duplex soils have non-red B horizons

Dc Either red or non-red duplex soils (sandier topsoil abruptly overlying more clayey subsoil)
comprise >50% of land type; plus >10% occupied by black or red clays
Ea Black or red clays comprise >50% of land type
Fa Shallow soils (Mispah & Glenrosa forms) predominate; little or no lime in landscape
Fb Shallow soils (Mispah & Glenrosa forms) predominate; usually lime in some of the
bottomlands in landscape
Fc Shallow soils (Mispah & Glenrosa forms) predominate; usually lime throughout much of
landscape
Ga Podzols occur (comprise >10% of land type); dominantly deep
Gb Podzols occur (comprise >10% of land type); dominantly shallow
Ha Deep grey sands dominant (comprise >80% of land type)
Hb Deep grey sands sub dominant (comprise >20% of land type)
Ia Deep alluvial soils comprise >60% of land type
Lb Rock outcrops comprise >60% of land type
Ic Rock outcrops comprise >80% of land type
40
Figure 20: The number of landslides per soil type class. Class Lb (Rock outcrops comprise >60% of land type)
recorded the highest number of landslides and class Gb (Podzols occur (comprise >10% of land type); dominantly
shallow) recorded the least number of landslides.
3.1.4 Anthropogenic influences

Anthropogenic influence on slope instability has been discussed on Section 2.2.1.4 of Chapter 2. The road
and rail cut layer of the study area was mapped at 1:50 000 scale, therefore a buffer of 50 metres was used
for the roads and rail cuts. The road and rail cuts data was acquired from National Geo-spatial
Information (NGI).
41
Figure 21: The layer containing the roads and rails within the study area. A buffer of 50 meter was used. The
localities of known historic landslides are represented by the black dots on the map.
3.2 LANDSLIDE SUSCEPTIBILITY MAPPING – THE WEIGHTS OF EVIDENCE

APPROACH
The weight of evidence approach was selected for the creation of a landslide susceptibility map. This
method was first described by Bonham-Carter, Agterberg & Wright 1989). It uses the Bayesian
probability model and was originally developed for mineral potential assessment (Bonham-Cater,
Agterberg & Wright 1988 & 1989). A GIS framework was used to calculate the spatial relationship
between landslide occurrence and specific terrain features that are known to have influence on landslide
occurrences. These terrain features were selected on the basis that there is a known relationship between
the feature and landslide occurrence and that the feature data were available for the entire part of the study
42
area. Additionally, the location of a selection of known landslides was used as input. The weights of the
landslide causative thematic layers (input data) are calculated based on this landslide inventory map. Each
feature layer were prepared as, or converted to, raster images and the spatial data modeler was used in
order to obtain the statistical relationships between terrain features and landslides, using GIS.
The weight of evidence approach combines the spatial relationships between terrain features from a
variety of different thematic layers, each considered to have some relationship to landslide occurrence.
For each thematic layer weights are calculated for each individual factor (e.g. slope between 0-200, 20-
400, 40-600 and greater than 600) and a weight map is developed from that theme. All thematic weight
layers are then added together to develop a landslide susceptibility map. The positive weight (W+) and
negative weight (W-) are calculated as the algorithm of the ratio of spatial probabilities, as follows:
P{B | D}
W + = log e
P{B | D*} {4}
P{B* | D}
W − = log e
P{B* | D*} {5}
Where P is the probability of having landslide, B is the presence of landslide controlling parameter, B* is
the absence of potential landslide controlling parameter, D is the presence of landslide and D* is the
absence of landslide. A positive (W+) and a negative (W-) indicate a positive and negative correlation
between the presence of the predictive variables and the landslide respectively. Positive and negative
weights are calculated for each factor in each theme. The positive weight is an indication of the likelihood
of encountering a landslide when the specific factor is present. The weight contrast (C) is then calculated
and its magnitude reflects the spatial association between the predictable variables and the landslide.
Weights and rankings for all parameters are developed based on the analysis of historical landslides.
A positive weight that is greater than zero implies that landslides are more likely than average to occur in
the presence of the specific landslide causative factor. A positive weight equal to zero suggests landslide
are no more and no less likely than average to be present, and a positive weight less than zero suggests
landslides are less likely, than average to be present, given the presence of the specific landslide causative
factor. The weights that are strongly negative and strongly positive are therefore more useful in
determining landslide susceptibility, as they provide the strongest indication of likely presence or absence
of landslides. The weights close to zero give little insights on the likelihood of landslide occurrence.
43
The algorithm calculates weights for each thematic layer, where each specific factor within that theme is
assigned a numeric value equal to its calculated weight. The weights for each weighted layer were then
added to yield a combined map.
3.2.1 Weight of evidence landslide susceptibility map

The result of the weight of evidence method is presented on Figure 23. The thematic layer (discussed
above) where used to run the weight of evidence method, using spatial data modeler. In addition the
verified landslide localities were part of the input layers. The weights were calculated using the known
landslide localities. The susceptibility values were between zero and one, with one being the highest
susceptibility and zero being the lowest susceptibility. The classified landslide susceptibility map is
presented on Figure 24. The classes were created using natural break classification of ArcGIS, as shown
in Figure 22. The natural break method was used because it minimizes the variance between the classes.
Five classes were selected when applying the natural break classification method. In the map these classes
are shown as Very high, High, Moderate, Low and Very Low. The red and orange areas on the map are
the Very High and High susceptible areas, respectively, the yellow areas are the moderately susceptible
and the light green and green areas are the Low and Very Low susceptible areas, respectively.
44
Figure 22: Classification of the landslide susceptibility map using the natural break method.
45
Figure 23: The unclassified weight of evidence landslide susceptibility map. The legend shows the increasing
susceptibility of the map, with the higher values representing higher susceptibility and lower values representing
lower susceptibility.
46
Figure 24: The landslide susceptibility map of the study area based on the weight of evidence model. The classes are
shown on the legend. The red and orange areas are very high and high susceptibility, the yellow areas are
moderately susceptible areas and the green areas are the very low and low susceptible areas.
Figure 25 shows the number of landslides per susceptibility class. Ideally the Very Low class should not
have any landslide point and a large percentage of landslides should fall on the Very High and High
Class. Only two landslides plot on the Very Low class, one landslide plots on the Low Class, four on the
Moderate Class, 14 on the High class and 72 on the Very High class. Percentage wise, 96.8 % of the
landslides plot on the high and very high class and 4%, 2% and 1 % of the landslides plot on the
moderate, low and very low class; respectively.
47
Number of landslides per class

80
70
60
50
40
30
20
10
0
Very High High Moderate Low Very Low
Figure 25: The number of landslides per susceptibility class. These landslides points were used to run the weight of
evidence model. A large number of landslides fall on the very high and high class (96.8 %).
Two sets of landslide locality data were obtained from the Council for Geosceince. One set was field
verified or confirmed landslides and the other set were not verified. The field verified data set was the
only set used to run the weight of evidence method. The values were extracted from the susceptibility
map for all the landslides that fall within the study area. Figure 26 shows the number of landslide per
susceptibility class, for the unverified data set. Interestingly 730 landslides where on the very high class,
152 on the high class, 68 on the moderate class, 20 on the low class and 2 on the very low class ( as
shown in Figure 26). Roughly 90 % of the landslides were on the high and very high class
48
Number of Landslides per class

800
730
700
600
500
400
Number of Landslides
300
200 152
100 68
20 2
0
Very High High Moderate Low Very Low
Figure 26: The number of landslides per susceptibility class. These landslide points were not used when the
susceptibility map was modelled; they were only used to test the efficiency of the model. A large number of
landslides fall on the high and very high class (90 %).
3.2.2 Accuracy Assessment for the Weight of Evidence method

A more sensible and holistic evaluation of the weight of evidence model performance can be seen by
constructing the success rate curve. The following steps were used to plot the success rate curve:
49
1. The susceptibility map was classified into 10 classes using equal interval method (Figure 27).
Figure 27: Classification of the landslide susceptibility map using equal interval method.
2. Cumulative percentage susceptibility was then calculated for each class

3. The curve of percentage landslides versus cumulative percentage susceptibility was plotted.
4. The overall accuracy was calculated (basically the area under the curve).
Figure 28 is a curve displaying the cumulative percentage of total landslides, which fall within increasing
susceptibility map percentages. The ideal curve on such a plot would be the one that contains 100% of
landslides within a small percentage of the susceptibility map, indicating that the model does not over
predict. Figure 28 displays both the success rate curve (based on the training data “landslide points that
were used to run the model) and the prediction rate curve (based on the reference data “landslide points
that were not used to run the model”). The steeper the prediction rate curve, the higher the capacity the
model has to predict landslides.
50
100
90
80
Percentage of landslides
70
60
50
40
30
Training
20
Reference
10
0
0 10 20 30 40 50 60 70 80 90 100
Percentage of susceptability map
Figure 28: The success (blue) and prediction (red) rate curves for the weight of evidence susceptibility model.
Based on the success rate curve overall accuracy of the weight of evidence susceptibility map is 82.71%
for the training data and a prediction rate of 80.37% for the reference data. This can be visualized in
Figure 29 and Figure 30 for the training and reference data set; respectively.
51
100
90
80
70
60
50
40
30
Training
20
Reference
10
0
0 20 40 60 80 100
Figure 29: The overall success percentage for the weight of evidence method, the blue highlighted area is the
overall percentage for the training data set.
52
100
90
80
70
60
50
40
30
Training
20
Reference
10
0
0 20 40 60 80 100
Figure 30: The overall success percentage for the weight of evidence method, the red highlighted area is the overall
percentage for the reference data set
3.3 LANDSLIDE SUSCEPTIBILITY MAPPING – THE MAP COMBINATION APPROACH
A second landslide susceptibility map was created using the Map Combination Approach. The method
involved the weighting and ranking of the causative factors affecting landslide susceptibility. This
weighting and ranking was done based on expert knowledge (Stapelberg, pers com 2011). Each thematic
layer was given a value based on its significance on landslide occurrence. Those thematic layers that were
deemed to be more important were given a higher value e.g. slope was the most important parameter out
of the eight thematic layers that were used, and it was given a value/weight of eight. Then each thematic
layer was classified based on the expert knowledge e.g. the slope was classed into four classes (0-6⁰, 6-
12⁰, 12-18⁰, and greater than 18⁰). The classes were assigned a value/ranked between one and five, with
classes assigned a value of one being the least influential ones and the most influential classes were
assigned a value of five. In the case of slope, 0-6⁰ was assigned a value of one, 6-12⁰ was assigned a
value of two, 12-18⁰ was assigned a value of 3, and slopes greater than 18⁰ were assigned a value of 5.
The weight of the parameters (layers) was then multiplied with the rank of the classes in that parameter,
and then summed for all eight the causative parameters (layers) to produce the susceptibility index:
53
LSI= (SLw × SLr) + (SDw ×SDr) + (Rw × Rr) + (LCw × LCr) + (Fw × Fr) + (Gw × Gr) +
(STw ×STr) + (ASw ×ASr) {6}
Where LSI is the landslide susceptibility index, SL is slope, SD is soil depth, R is road and rails, LC is
landcover, F is faults and geological contacts, G is geology, ST is soil type and AS is aspect.
• W is the weight of the causative thematic layers
• R is the rank for the classes in the causative thematic layers
Table 4 shows the weight and the rank for all the causative thematic maps that were used to model an
expert based susceptibility map.
Table 4: The weights and ranks for the causative thematic layers used to model an expert based landslide
susceptibility map.
Parameter Classes Rank Weight

Slope 0-6⁰ 1 8
6-12⁰ 2
12-18⁰ 3
>18⁰ 5
Soil Depth <300mm 1 7
300-600mm 2
600-900mm 3
900-1200mm 4
>1200mm 5
Roads and Rails Within 50 meter buffer 5 6
Outside the buffer 0
National Landcover Built up areas 5 5
Grasslands 2
Commercial farmlands 1
Dongas 4
Forests 1
Faults and Lithological
contacts Within the 250 meter buffer 5 4
Outside the buffer 0
Geology Sandstone 1 3
Shales & siltstone 5
Conglomerate 2
Granites 5
Alluvium deposits 5
54
Soil type Clay 5 2

Sandy 2
Deep clay soil 5
Deep mixed soil 5
Shallow soil 1
Rocky 1
Aspect Flat 1 1
North 2
North East 2
East 1
South East 4
South 5
South West 4
West 3
North West 2
The landslide susceptibility map modelled using the expert knowledge approach (Map Combination
approach) is presented in Figure 31.
55
Figure 31: The landslide susceptibility map of the study area. The map was modelled using a map combination
approach. The susceptibility values are shown in the legend, the higher the values, the higher the susceptibility and
the lower the value the lower the susceptibility.
The classified susceptibility map, modelled using the Map Combination approach is presented on Figure
32. The classes were created using the natural break method, as described in Section 3.2.1
56
Figure 32: The landslide susceptibility map modelled using the map combination approach. The susceptibility
classes are shown on the legend. The red and orange areas represent very high and high susceptibility areas, the
yellow areas depict moderate susceptibility areas and the green areas are low susceptible areas.
Figure 33 shows the number of landslides per susceptibility class (only the selected field verified
landslide (these landslides were also used to model the weight of evidence based susceptibility map) and
Figure 34 shows the number of landslide per susceptibility class for the second data set (unverified
landslide localities). The Very High class recorded the least number of landslides, and 42 landslides fall
on the Low and Very Low class. Such a large number of landslides on the Very Low and Low class
signify the inefficiency of the map combination approach. The Very High and High class recorded 42%
of the landslides, while the Very Low and Low class recorded 44 percent of the landslides, slightly higher
than the Very High and High class.
57

30 28
25 24
20 18
Number of
15 14
landslides per class
12
10
0
Very High High Moderate Low Very low
Classes
Figure 33: The graph shows the number of landslides per susceptibility class.
The number of landslides per susceptibility class for the unverified landslide points is presented in Figure
34, the Very High class recorded 126 landslides, the High class recorded 512 landslides, moderate class
recorded 114 landslides, the Low class recorded 188 landslides and the Very Low class recorded 32
landslides. Percentage wise the Very High and High class recorded 66 percent of the landslides, while the
Very Low and Low class recorded 23 percent of the landslides.
58

600
512
500
400
300
Number of
landslides per class
188
200
126 114
100
32
0
Very High High Moderate Low Very low
Classes
Figure 34: The graph shows the number of landslides per susceptibility class. These landslide localities where
attained from the Council for Geosciences.
3.3.1 Accuracy Assessment for the Map Combination approach

The success rate curve was plotted using the same procedure that is explained in Section 3.2.2. The
success rate curve for the map combination approach is presented on Figure 35. The ideal success rate
curve is the one that reaches 100 % within the lowest percentage of susceptibility map.
59
100
90
80
70
60
50 Training Data
40 Reference Data
30
20
10
0
10 20 30 40 50 60 70 80 90 100
Figure 35: The success rate curve for the map combination approach. The red line is the success rate curve for the
reference data set and the blue trend is the success rate curve for the training data set.
The cumulative success percentage for the map combination approach is presented on Figure 36 and
Figure 37 for training data and reference data, respectively. A cumulative success percentage of 45.98 %
was achieved for the training data set and a cumulative prediction rate percentage of 50. 98 % was
achieved for the reference data set.
100
90
80
70
60
50 Training Data
40 Reference Data
30
20
10
0
10 20 30 40 50 60 70 80 90 100
Figure 36: The cumulative success percentage for the map combination approach. The blue highlighted area is the
success percentage for the training data.
60
100
90
80
70
60
50 Training Data
40 Reference Data
30
20
10
0
10 20 30 40 50 60 70 80 90 100
Figure 37: The cumulative success percentage for the map combination approach. The red highlighted area is the
success percentage for the reference data.
The following chapter investigates the application of remote sensing techniques in monitoring the
changes in variable factors and also attempts to correlate the changes with landslide occurrence, in
susceptible areas.
61
CHAPTER 4 REMOTE MONITORING OF VARIABLE CONDITIONS

AND IDENTIFICATION OF TRIGGERING MECHANISMS
After the generation of the landslide susceptibility map outlined in Chapter 3 considering the static factors
affecting landslide susceptibility, the variable factors and triggering mechanisms for landslides can be
considered. The variable factors affecting landslide susceptibility include the health and productivity of
the vegetation as well as the moisture content of the vegetation and the soils. As described in Chapter 2,
satellite remote sensing sensors and analysis approaches are ideal for the monitoring of vegetation
through the use of vegetation indexes.
To determine how the health and productivity of vegetation affect landslide susceptibility, several
historical landslides were considered. The localities for the different landslides are presented in Figure 38.
The period of occurrence and the number of landslides per area are shown in Table 5
Table 5: Shows the number of landslides and the date of occurrence for each area.
Area Number of Landslides Period of occurrence
1 2 1. Winter of 2005
2. Winter of 2008
2 11 Year 2005
3 1 2007
4 1 2007
62
Figure 38: The geographic location of the 2005, 2007 and 2008 landslide events. The landslides are located on the
far north western and south western part of the study area.
Area 2, close to the town of Hermanus had landslide scares that are highly visible even in recent images,
unlike the other localities whereby the landslide have been cover by vegetation. These landslide fall in the
highly susceptible area in the susceptibility map modeled using the weight of evidence method (Figure
39). The Google earth images for this area are presented in Figure 40.
63
Figure 39: The geographic location of several landslide events. The landslides are located close to the towns of
Stanford and Hermanus. The landslide localities are represented by black dots on the map.
29-09-2004 02-11-2006
Figure 40: Google Earth images indicating the landscape before landslide occurrence (29-09-2004) and the
landscape after landslide occurrence (02-11-2006).
64
The following sections describe the results of the analysis of low spatial resolution – high temporal
resolution MODIS data to determine how the vegetative conditions may have influenced the occurrence
of these historical landslides. Finally, the triggering mechanisms of these landslides were determined.
4.1 EXTRACTION OF NDVI AND NDWI – MODIS DATA
The high temporal resolution (daily for the most of the parts of the world) and low spatial resolution (250
m to 1 km) of MODIS data makes it ideal for the study of global change in disciplines including
oceanography, biology, and atmospheric science. For the identification and mapping of landslides after
they have occurred, the use of MODIS data is not ideal due to its low spatial resolution. However, the
high revisit time makes MODIS data a very useful tool to investigate landslide causative factors such as
vegetation productivity and soil wetness conditions. The MODIS derived vegetation index (NDVI) is
designed to provide improved monitoring and understanding of regional and global dynamics and
processes. The temporal trends of the NDVI indices are used for detecting the changes in biophysical
and/or biochemical characteristics of the vegetation and the identification of the phonological stage.
In an effort to identify the long-term status of the health and productivity of the vegetation in the area
where landslides are known to have occurred, the MODIS time series viewer, (available at
http://afis.meraka.org.za/wamis/time-series-viewer) were used to extract information on the phenology of
vegetation in the study area. The viewer, developed by South Africa’s CSIR Meraka Institute, provides
access to NDVI data derived from a long time-series of MODIS data. The long term trends of NDVI were
investigated in an effort to identify potential changes in landcover conditions that may have affected the
occurrence of historical landslides.
For this purpose, areas where landslides are known to have occurred were investigated. The dates of
landslide occurrence were not exact. The following sections describe the results of the analysis of the
long-term trends in NDVI and their potential effect on landslide susceptibility.
AREA 1
Area 1 is located on the northern side of the study area, near the town of Tulbagh. In order to investigate
the vegetative conditions during the occurrence of the landslide using the MODIS data set, the date at
which the landslide occurred must be known. Since the exact date of occurrence is unknown, the exact
65
NDVI values immediately prior to landslide occurrence could not be pinpointed. However, when the
trend of NDVI values is considered for two seasons prior and during 2005 landslide occurrence, some
interesting patterns emerge. The NDVI values during this period is presented in Figure 41
2003-2006 NDVI profile
0.7
0.65
0.6
0.55
NDVI
0.5 NDVI
0.45
0.4
0.35
0.3
01 01 01 01 01 01 01 01 01 01 01 01
Ja Ap Ju Oc Ja Ap Ju Oc Ja Ap Ju Oc
nu ril ly to nu ril ly to nu ril ly t
ar 20 20 be ar 20 20 be ar 20 20 obe
y 03 r2 y 04 r2 y 05 r2
20 03 00 200 4
0
00 200 5
0
00
03 3 4 4 5 5
Data
Figure 41: The NDVI time-series profile, of the selected area which is known to have landslide occurrence. The
profile is from the year 2003 to the year 2006. The red line indicates the period at which the landslide is thought to
have occurred. The plot shows that the minimum NDVI values for the year at which the landslides occurred are
lower than the previous years.
The landslide event was known to have occurred in the winter season of 2005 (between the beginning of
May and the end of July 2005). The time series of NDVI values during this time suggests that, for the
season preceding the landslide occurrence, low NDVI values were recorded. Table 6 indicates the annual
average, minimum and maximum NDVI values recorded for the years 2003 to 2008. It is observed that
the average NDVI values for the year 2004 (prior to landslide occurrence), was low at 0.461 when
compared to later years during which average NDVI values range between 0.485 and 0.510. Additionally,
the maximum NDVI value recorded during 2004 was also lower at 0.637 when compared to the years
2005-2007 where a maximum NDVI of between 0.665 and 0.684 where recorded. An exception is
observed for the year 2008 for which the lowest maximum NDVI of 0.611 was recorded.
66
Table 6: The annual average, minimum and maximum values of NDVI. The values are calculated from the beginning
of the year to the end of the year.
Average Minimum Maximum

Year NDVI NDVI NDVI
2003 0.388 0.379 0.629

2004 0.461 0.342 0.637
2005 0.485 0.331 0.680
2006 0.473 0.355 0.665

2007 0.510 0.364 0.684
2008 0.476 0.370 0.611
2009 0.502 0.370 0.664
The 2008 landslide event occurred in the same area as the 2005 landslide, and also in the winter season
(between May and July of 2008). It is likely that this landside was the reactivation of the 2005 landslide
event. The NDVI time series profile is presented on Figure 42, the NDVI values for the year preceding
landslide occurrence are high. The yearly average, minimum and maximum NDVI values are presented
on Table 6. The annual average for the year prior to landslide occurrence (2007) is high at 0.510. The year
prior to landslide occurrence also records the maximum NDVI value of 0.684. These high NDVI values
for the year preceding landslide occurrence are the opposite of what has been observed for the 2005
landslide event. Nevertheless the lower minimum NDVI values for the year preceding landslide
occurrence that has been the prominent feature for the previous NDVI profile, is observed.
67

0.75
0.7
0.65
0.6
NDVI
0.55
NDVI
0.5
0.45
0.4
0.35
0.3
01
01 o be
01 o be
01 ua
01 il 2 06
01
01
01 il 2 07
01
01
01 il 2 08
01
Ja
Ja
Ja
Ap ry 2
Ju 0 06
Ap y 2
Ju 0 07
Ap y 2
Ju
O
ct
ct
ct
n
nu
nu
ly
ly
ly
r
ob
a r 06
a r 07
20
20
20
er
0
06
07
08
r2
r2
08
20
0
0
0
08
Date
Figure 42: The NDVI time-series profile of the area known to have landslide occurrence. The plot is from the year
2006 to the year 2009. The red line indicates the period at which the landslide is thought to have occurred. The plot
shows relatively low minimum NDVI values for the year prior to landslide occurrence.
AREA 2
Figure 43 is the NDVI time-series profile for the area known to have landslide occurrence. The landslide
is thought to have occurred in the year 2005. There is no exact date of occurrence for this landslide event.
From the graph it is observed that the minimum NDVI values for the year preceding landslide occurrence
were slightly lower than normal.
68

0.75
0.7
0.65
NDVI
0.6 NDVI
0.55
0.5
0.45
01 01 01 0 01 0 01 0 01 0 01 0
Ja Ju Ja 1 Ju Ja 1 Ju Ja 1 Ju Ja 1 Ju Ja 1 Ju
nu yl nu ly nu ly nu ly nu ly nu ly
ar 20 ar 20 ar 20 ar 20 ar 20 ar 20
y 0 y 0 y 0 y 0 y 0 y
20 3 20 4 20 5 20 6 20 7 20 08
03 04 05 06 07 08
Date
Figure 43: The NDVI time series profile for the year 2003 to 2008. The red line on the graph indicates the period at
which the landslide is thought to have occurred. The minimum NDVI values are slightly lower for the year prior to
the occurrences of landslide event.
Several landslide scars were observed at the area close to the town of Hermanus and Stanford, in the
Western Cape province of South Africa. The exact date of occurrence for these landslides could not be
accurately identified, but they are estimated to have occurred between 2004-09-29 and 2006-11-02, based
on the interpretation of the Google earth images.
The MODIS NDVI time-series profile is presented in Figure 44. The yearly average, minimum and
maximum NDVI values are presented in Table 7. There are no major variations on the yearly average,
maximum and minimum NDVI, as from the year 2002 to the year 2006. An exception is the year 2002,
which records the highest NDVI value of 0.724. From the previous NDVI time-series profiles, it has been
observed that landslide occurrence is associated with lower minimum NDVI values, for the year prior to
landslide occurrence.
69
Table 7: The yearly average, minimum and maximum values of NDVI. The values are calculated from the beginning
of the year, to the end of the year.
Average Minimum Maximum
year NDVI NDVI NDVI
2002 0.646 0.576 0.724

2003 0.631 0.568 0.702
2004 0.642 0.584 0.692
2005 0.639 0.567 0.694
2006 0.631 0.582 0.699
70
0.75
0.73
0.71
0.69
0.67
NDVI
0.65 NDVI
0.63
0.61
0.59
0.57
0.55
01 y 20 02
01 y 20 03
01 y 20 04
01 y 20 05
01 y 20 06
01
01 ua
01 tem
01
01 tem
01
01 tem
01
01 pte
01 ua r 20
Se 0 2
Se 0 3
Se 0 4
Se 0 5
Se 0 6
Ja
M ry 2
Ja
M ry 2 02
Ja
M ry 2 03
Ja
M ry 2 04
Ja mb
M ry 2 05
a
a
n
nu er
nu er
nu er
n
p
pt
em
a
a
b
be
e
0
r2
20
20
20
00
Date
6
Figure 44: The NDVI time-series profile of an area with numerous landslide scars. The area is close to the town of Hermanus and Stanford. The landslides are
estimated to have occurred between the year 2004 and 2006.The red line on the graph indicates the time at which the landslide could have occurred, based on
the previous observations that landslide in the study area are associated with low minimum NDVI values for the year prior to landslide occurrence.
71
AREA 3
The NDVI time series profile for the landslide event that occurred during the heavy rains of the year 2007
is presented on Figure 45. The graph shows lower minimum NDVI values for the year preceding
landslide occurrence (2006). These lower than normal minimum NDVI values seems to be the prominent
feature for most of the NDVI time series profile, for areas with landslide occurrence, within the study
area.

0.75
0.7
0.65
NDVI
0.6
NDVI
0.55
0.5
0.45
01
01
01 il 2 0 05
01 ly 2
01 to b
01 ua 0 05
01 ril 2 0 06
01 ly 2
01 to b
01 ua 0 06
01 ril 2 0 07
01 ly 2
01 to b
01 ua 0 07
01 ril 2 0 08
01 ly 2
Ja
A ry
Ju 00
O 00
Ja er
A ry
Ju
O 00
Ja er
A ry 2
Ju 00
O 00
Ja er
A ry
Ju 00
O 00
pr
p
c
ct
nu
ob
a
er
006
5
8
2
20
5
08
Date
Figure 45: The NDVI time-series profile for the area known to have landslide occurrence. The red line on the graph
indicates the period at which the landslide is thought to have occurred. The minimum NDVI values are lower for the
year prior to landslide occurrence (2006).
AREA 4
The NDVI time series profile for the landslide event that occurred on the year 2007 is presented on Figure
46. The only trend that is observed on the NDVI profile is the lower minimum NDVI values for the year
preceding landslide occurrence.
72

0.6
0.55
0.5
NDWI
NDVI
0.45
0.4
0.35
01
01 nua
01 y 2
01 pte
01 nua r 2
01
01 pte
01 nua r 2
01
01 pte
01 nua r 2
01
01 pte
01 nua r 2
01
Ja
M ry
Se 005
Ja mb
Se 006
Ja mb
M ry
Se 007
Ja mb
Se 008
Ja mb
M ry
Se 009
a
ay 20 5
ay 20 6
ay 20 7
ay 20 8
pt
em
2
2
r
r
y
be
20
r2
05
06
07
08
09
00
00
00
00
00
9
Date
Figure 46: The NDVI time-series profile for the year 2005 to 2009. The red line on the graph indicates the period at
which the landslide is thought to have occurred. The minimum NDVI values are slightly lower for the year prior to
landslide occurrence.
The interpretation of the NDVI time series profiles for areas with known landslide occurrences, within the
study area shows the following:
1. The minimum NDVI values for the year preceding landslide occurrence are relatively low, in
most cases.
2. In some cases the maximum, minimum and average NDVI values are relatively low for the year
preceding landslide occurrence.
4.2 EXTRACTION OF NDVI AND NDWI AND PHENOLOGY DATA – LANDSAT DATA
The low resolution MODIS data is captured at 250m resolution. The low resolution of the data may imply
that localised effects are not accurately observed. For this reason, 30m resolution Landsat data was
analysed in an effort to see if localised trends of NDVI can be observed. The problem is that the 30m
resolution data is not available at such frequent intervals as the 250 m resolution MODIS data. The
Normalized Difference Vegetation Index (NDVI) is very important when estimating vegetation health.
The information about the NDVI was computed from the 30m resolution Landsat 5 TM and Landsat 7
ETM+ images. For the purpose of this study, six Landsat 5 and two Landsat 7 images were requested
from South African National Space Agency (SANSA), in order to investigate the possible dates of
occurrence for the landslide events close to the town of Stanford and Hermanus, and also investigate the
vegetation conditions associated with landslide occurrences.
73
Table 8: Landsat scenes that will be used in this study.

Scene No. Satellite Sensor Path/Row Image Date
1 Landsat 7 ETM+ 175/84 2002/03/15
2 Landsat 7 ETM+ 175/84 2002/07/21
3 Landsat 5 TM 175/84 2004/09/20
4 Landsat 5 TM 175/84 2006/02/14

5 Landsat 5 TM 175/84 2006/08/09
6 Landsat 5 TM 175/84 2007/02/17
4.2.1 Image pre-processing

The use of satellite image data for a spatial database requires several pre-processing procedures. These
procedures include geometric correction and image enhancement. The purpose of digital image
processing is to increase both accuracy and the interpretability of the digital data during the image
processing phase. The normalization of satellite imagery takes into account the following: the combined
measurable reflectances of the atmosphere, aerosol scattering and absorption, and the earth’s surface
(Kim & Elman 1990). The goal of atmospheric correction as stated by Hall, Stebel, Nickesen & Goetz
(1991) should be that after image pre-processing, all images should appear as if they were acquired from
the same sensor. Geometric errors are caused by factors such as the earth’s curvature and rotation as well
as variations in the velocity, altitude and attitude of the sensor platform. Geometric rectification of the
imagery changes the pixel grid to fit that of a map projection or another reference image. This becomes
important when scene to scene comparisons of individual pixels in applications such as change detection
are being sought (ERDAS 1999).
4.2.1.1 Geometric correction

The major purpose of geometric corrections is to correct the inaccuracy between the location coordinates
of the picture elements in the image data, and the actual location coordinates on the ground. Geometric
correction of the imagery changes the pixel grid to fit that of a map projection or another reference image.
The ERDAS IMAGINE 2011 image geometric correction module was used for the geometric correction
of the images. The Landsat geometric model was chosen, which allows for the orthorectification of
Landsat data. A 20 meter resolution digital elevation model (DEM) of the study area was used as the
elevation file. Twenty five ground control points were collected and the maximum root mean square
error (RMSE) of 1.6 was achieved (refer to Table 9).The nearest neighbour re-sampling method was
74
applied. This method uses the value of the closest pixel to assign to the output pixel value and thus
transfers original data values without averaging them as other methods do, therefore, the extremes and
subtleties of the data values are not lost (ERDAS 1999). Nearest neighbour re-sampling was selected
because it preserves pixel values perfectly during resampling. Table 9 shows the RMSE and the number
of ground control points (GCP) for each Landsat scene that was geometrically corrected.
Table 9: Table representing the scene date RMS error and the number of GCP's collected when geometric
correction was performed.
SCENE DATE RMS error NUMBER GCP
2004-09-20 0.5526 25
2005-02-27 1.2596 25
2005-09-07 0.9788 25
2006-08-09 1.1970 25
2006-04-19 1.5570 25
2007-02-17 1.5262 25
4.2.1.2 Radiometric calibration

In order to standardize the impact of illumination geometry on the 8-bit digital number (DN) imagery, it
first had to be converted to at-satellite radiance, and then to at satellite reflectance, using information
extracted from the image header files.
Calculation of at-sensor spectral radiance is the fundamental step in converting image data from multiple
sensors and platforms into a physically meaningful common radiometric scale. The following equation is
used to perform the digital number to spectral radiance conversion for Level 1 products:
L λ = ((LMAX λ - LMIN λ )/ (QCALMAX-QCALMIN)) * (QCAL-QCALMIN) + LMIN λ {7}
Where
Lλ = Spectral radiance at the sensor’s aperture [W/ (m2 srμm)]
Qcal = Quantized calibrated pixel value [DN]
Qcalmin = Minimum quantized calibrated pixel value corresponding to LMINλ [DN]
Qcalmax = Maximum quantized calibrated pixel value corresponding to LMAXλ [DN]
LMINλ = Spectral at-sensor radiance that is scaled to Qcalmin [W/ (m2 srμm)]
LMAXλ = Spectral at-sensor radiance that is scaled to Qcalmax [W/ (m2 srμm)]
Grescale = Band-specific rescaling gain factor [(W/ (m2 srμm))/ DN]
Brescale = Band-specific rescaling bias factor [W/ (msrμm)]
75
Conversion to TOA reflectance (Lλ-to- ρP)
A reduction in scene-to-scene variability can be achieved by converting the at-sensor spectral radiance to
exoatmospheric TOA reflectance, also known as in-band planetary albedo. When comparing images from
different sensors, there are three advantages to using TOA reflectance instead of at-sensor spectral
radiance. First, it removes the cosine effect of different solar zenith angles due to the time difference
between data acquisitions. Second, TOA reflectance compensates for different values of the
exoatmospheric solar irradiance arising from spectral band differences. Third, the TOA reflectance
corrects for the variation in the Earth-Sun distance between different data acquisition dates. These
variations can be significant geographically and temporally. The TOA reflectance of the Earth is
computed according to the equation:
π∙Lλ ∙d2
ρλ = {8}
ESUNλ ∙cosθs
ρλ = Planetary TOA reflectance [unitless]

π = Mathematical constant approximately equal to 3.14159 [unitless]
Lλ =Spectral radiance at the sensor’s aperture [W/ (m2 srμm)]
d = Earth-Sun distance [astronomical units]
ESUNλ = Mean exoatmospheric solar irradiance [W/ (m2 μm)]
Θ =solar elevation angle (from imagery header files)
4.2.1.3 Atmospheric correction

The objective of a radiometric atmospheric correction is to convert satellite generated digital counts to
ground reflectance (absolute reflectance). Various complex models have been developed to correct
atmospheric effect (Chavez 1996). However, these models need some data, about atmospheric conditions
for the date and time that remote sensing data was acquired, which are not usually available. The ENVI’s
fast line-of-sight atmospheric analysis of spectral hypercubes (FLAASH) was used. FlAASH is the
MODRANT-4 based atmospheric correction software package, which consist of term to account for the
adjacency effect and also performs spectral polishing.
FLAASH is a first-principles atmospheric correction tool that corrects wavelengths in the visible through
near- infrared and shortwave infrared regions, up to 3 μm. Unlike many other atmospheric correction
programs that interpolate radiation transfer properties from a pre-calculated database of modelling results,
FLAASH incorporates the MODTRAN-4 radiation transfer code. You can choose any of the standard
MODTRAN model atmospheres and aerosol types to represent the scene; a unique MODTRAN solution
is computed for each image.
76
FLAASH also includes the following features:

 Correction for the adjacency effect (pixel mixing due to scattering of surface-reflected radiance).
 An option to compute a scene-average visibility (aerosol/haze amount). FLAASH uses the most
advanced techniques for handling particularly stressing atmospheric conditions, such as the
presence of clouds.
The steps shown on Figure 47 were used when performing atmospheric correction using ERDAS
2011.
Figure 47: Steps followed when performing atmospheric corrections.
4.2.2 Derivation of information on vegetation health and productivity and moisture

conditions
The Landsat series of satellites provide the longest continuous record of satellite-based observations. As
such, Landsat is an invaluable resource of monitoring regional and global changes. The 30 meter
resolution Landsat data is a useful tool on the identification of small and large landslides. The 2005-02-27
77
Landsat scene (see Figure 48); shows no landslide scars in the Hermanus area; while the 2005-09-07
scene shows several landslides scars (see Figure 49). Therefore the landslide events in the area occurred
between the 27th February and the 7th September 2005.
Figure 48: The 2005-02-27scene for the area close to the town of Stanford and Hermanus. There are no visible
landslide scars on this image, except one feature close the small round water body.
78
Figure 49: The 2005-09-27scene for the area close to the town of Stanford and Hermanus. The image shows several
landslide scars (the bright feature on the south facing slope on the mountain.
NDVI change detection
The primary goal of using the high spectral and spatial resolution images is to investigate the vegetation
conditions associated with landslide occurrences. For this purpose, the Landsat image scenes were
analysed for the NDVI, using ERDAS IMAGINE 2011. Change detection is an important application of
remote sensing technology. It is a technology ascertaining the changes of specific features within a certain
time interval. Remote sensing change detection technique can be classified as either pre or post-
classification change method. The computerized NDVI change detection (pre-classification) was used to
analyse the amount of change in NDVI. The NDVI change detection between two dates can be
summarised by the following equation:
NDVI change detection = ((NIR-Red)/ (NIR+Red)) t2- ((NIR-Red)/ (NIR+Red)) t1 {9}
Where t1 and t2 represent the two dates of images captured. The changes in NDVI values from 2004-09-
20 to 2005-02-27 are presented on Figure 50, the light red and bright red colours on the map depicts areas
that have experienced less than 15% decrease in NDVI and more than 15 % decrease in NDVI,
respectively. The light green and the bright green colours on the map depict areas that have less than 15%
NDVI increase and more than 15% increase, respectively. The exact localities for the landslide scars that
79
have been observed close to the town of Stanford and Hermanus are represented by the black stars on the
map. It is observed from the NDVI change detection map that all the landslide points plot on the areas
that have experience some decrease and a 15 % decrease in NDVI values, from the 20th of September
2004 to the 7th of February 2005. This means that the landslide occurrences in this area coincide with a
decrease in vegetation conditions for the season prior to landslide occurrence. This does not necessarily
mean that a decrease in NDVI values will cause landslide occurrences, but it simply means that the
decrease in NDVI values makes the area more likely to have landslide when the triggering mechanism
befalls.
Figure 50: NDVI change detection computed from the landsat scenes between 2004-09-20 and 2005-02-27. The
light red areas on the map are areas where there has been a decrease in NDVI and the strong red areas are those
that had more than 15 % NDVI decrease. The light green and bright green areas depict those areas that
experienced some NDVI increase and greater than 15 % NDVI increase, respectively. The localities of the
landslides are represented by the black stars on the map.
Figure 51 shows the changes in NDVI as from 2005-09-07 to 2006-04-19. The light green and green areas
depict areas that have experienced less than 15% increase and more than 15% increase in NDVI,
respectively. The light red to red areas represent areas that have experienced less than 15% decrease and
more than 15% increase in NDVI values, respectively. The large parts of the area shows a decrease in
NDVI values, but there were no new landslide scares or expansion of the existing landslides, which has
been observed from the Landsat images. This does not rule out the influence of the decrease in NDVI on
landslide susceptibility, but it could simply mean that there were no heavy rainfalls to trigger landslide
80
occurrence in the area, during the time when the vegetation conditions were more suitable for landslide
occurrence.
Figure 51: NDVI change detection computed from the images taken from 2005-09-07 and 2006-04-19. The light red
areas on the map are areas where there has been a decrease in NDVI and the strong red areas are those that had
more than 15 % NDVI decrease. The light green and strong green areas depict those areas that experienced some
NDVI increase and greater than 15 % NDVI increase, respectively. The exact localities of the landslides are
represented by the black stars on the map.
The changes in NDVI values 2006-08-09 to 2007-02-17 are presented on Figure 52. The light green and
bright green areas on the map depict areas that have experienced less than 15% increase in NDVI and
more than 15% increase in NDVI values, respectively. The light red and red areas depict those areas that
have experienced less than 15% decrease in NDVI values and more than 15% increase in NDVI,
respectively. It can be seen that large parts of the area shows an increase in NDVI values, therefore based
on the previous observations the vegetation conditions do not favour landslide occurrence. Nevertheless
that does not mean that the landslide will not occur in the area when the triggering mechanism crops up,
as vegetation is not the sole parameter that controls landslide occurrence.
81
Figure 52: NDVI change detection computed from the images taken from 2006-08-09 to 2007-02-17. The light red
areas on the map are areas where there has been a decrease in NDVI and the strong red areas are those that had
more 15 % NDVI decrease. The light green and strong green areas depict those areas the experienced some NDVI
increase and greater than 15 % NDVI increase, respectively. The exact localities of the landslides scars are
represented by the black stars on the map.
4.3 TRIGGERING MECHANISMS

Rainfall data from the South African Weather Services was used to investigate if rainfall was the
triggering mechanism. For those landslides that have known dates of occurrences, the intensity of rainfall
during the occurrence of those landslides was investigated. It is crucial to know the rainfall conditions
during the occurrence of landslides in order to know if the rainfall was the triggering mechanism, and also
to have an insight of the rainfall threshold that triggers landslides.
In order to investigate the triggering mechanism for the landslide events close to the town of Stanford and
Hermanus the seismic and the rainfall record of the area was investigated. Investigation of the seismic
data that was obtained from the seismology unit of the Council for Geosciences, showed no seismic
activity for the year 2005, which is the year at which the landslides occurred. Nevertheless the rainfall
record acquired from the South African Weather Services (SAWS) for the weather station located in the
town of Hermanus showed some interesting results. The three year annual rainfall record in this area
82
showed that the year of 2005 had the highest rainfall at 822.5mm as compared to 619.5 mm of the year
2004 and 641 mm of the year 2006. The monthly average for the year 2005 was also at the highest at
68.54 mm as compared to 51.63mm and 53.42 mm, for the year 2004 and 2006, respectively. Based on
Landsat image interpretation, the landslides were discovered to have occurred on the year 2005 (after the
27th of February 2005), the high rainfall for the year 2005 supports the longstanding idea that most
landslides in mountainous areas of South Africa are triggered by heavy rainfalls, but nothing much has
been done to identify the rainfall threshold that triggers landslides.
Table 10: The annual rainfall for the weather station in Hermanus (-34.417: 19.237). The monthly average and the
total annual rainfall are also indicated. The rainfall is in millimetres (mm).
Month 2004 Monthly rainfall 2005 Monthly Rainfall 2006 Monthly rainfall
January 15.5 52.5 14.5
February 32 31.5 33
March 33 12.5 16
April 33 246.5 56
May 14.5 101.5 93.5
June 67 178 33
July 86 13 156.5
August 48.5 88.5 60
September 52 38.5 30.5
October 161 24 81.5
November 32.5 26 39.5
December 44.5 10 27
Monthly Average 51.63 68.54 53.42
TOTAL 619.5 822.5 641
83
Annual Rainfall
900
800
700
600
Rainfall ( mm)
500
400 Rainfall
300
200
100
0
2004 2005 2006
Year
Figure 53: Annual rainfall for the year 2004, 2005 and 2006. The year of 2005 recorded the highest annual rainfall,
slightly higher than 800 mm.
Rainfall is the major triggering mechanisms for landslides in many parts of the world, and also in South
Africa. Heavy rainfall for a number of days can trigger landslides in areas that are susceptible to landslide
occurrence. Landsat interpretation indicates that the landslides occurred between the 27th of February
2005 and 7th of September 2005, this also gives insight that the winter rainfalls are likely to have
triggered the landslides in the area. The month of April and June recorded the highest and the second
highest monthly rainfall for the year 2005, respectively (referrer to Figure 54). The daily rainfall record
for the Month of April in the year of 2005 shows significant rainfall in the space of three days. It is worth
mentioning that 82% of the rainfall in the month of April occurred between the 9th and 11th of April
2005 (see Table 11), based on these observations the landslides are more likely to have occurred between
the 9th and the 11th of April 2005.
84
Table 11: The annual rainfall data (in millimeters) for the weather station close to the town of Hermanus. The blank areas in the table indicate that no rain fell on that day,
*** indicates that the data is missing or not yet available in the current month, C next to the value indicates that the rainfall was accumulated over a number of days, =
indicates that the total for the month is unreliable due to missing daily values, and A or B indicates that any rainfall that did occur is included in the accumulation.
Hermanus Station: Latitude and longitude: -34.417:19.237 Year: 2005

Day JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
1 1.0 C 5 0.5
2 18.5 2 2.5
3 0.5 A A 7 4
4 3.5 32.0 C 16.0 C 2.5
5 0.5 6 3.5
6 2.5
7 3.5 3.5 36 A A
8 30 6.0 C A
9 9 A A
10 164.0 C A
11 1.5 42 A 9.0 C A
12 18.5 7.5 A 15.5 C
13 16 6 6 A
14 9.5 16.5 18.0 C
15
16 2
17 22.5 1 18.5
18 1.5 3.5 1.5 1
19 3.5 19 14 3 7.5 1 8
20 1 9.5 6.5 1.5 3
21 3.5
22 3.5 A 11 3 0.5
23 13.5 C 9
24 0.5 7.5 A 2
25 A 24 7.5 C
26 2 22.0 C 2 7 4.5
27 4.5 A
28 8.5 8.5 C
29 ***
30 *** A 4.5 1
31 *** *** 17.5 *** 0.5 *** ***
TOTAL 52.5 31.5 12.5 246.5= 101.5= 178.0= 13 88.5= 38.5= 24.0= 26.0= 10
85
2005 Montly rainfall
300
250
200
Rainfal(mm)
Monthly Rainfall
150
100
50
0
Ja
Fe
Ju
Au
Se
D
M
Ap
Ju
ov
ec er
ct
ar y
ay
n
ne
ly
br
gu
pt
ril
ua
ob r
em
em
ch
ua
em
st
ry
er
r
be
be
r
Date
Figure 54: The plot for the monthly rainfall for the year 2005. The months of April and June show monthly rainfall
greater than 100 mm, with the month of April recording close to 250mm.
86
CHAPTER 5 DISCUSSION, RECOMMENDATIONS AND CONCLUSIONS
5.1 DISCUSSION
The aims of the study was to investigate landslide causative factors, synthesize a landslide susceptibility map
using geographic information system and also monitor changes that affect the areas susceptibility using remote
sensing techniques. Selecting the landslide causative parameters is the primary and the most critical step when
modeling a landslide susceptibility map. The static factors are the only causative factors that are typically used
when modeling a landslide susceptibility map, mainly because they do not change over a short space of time.
Different static factors were investigated in this study and their influence on slope instability was investigated.
It has been observed that slope is the most important parameter on landslide susceptibility and the high and very
high susceptibility classes on the weight of evidence modeled susceptibility map coincide with steeply sloped
areas. About 90 % of field verified landslides fell on slopes between 0-40⁰, with slopes between 0 and 20⁰
recording slightly more than 50 % of those landslides. This is no surprise in South Africa, as slopes greater than
18° are considered to be susceptible to landslides, and therefore not suitable for construction. South facing slopes
have long been known to be more susceptible to landslides, Boelhouwers et al (1998) was one of the authors to
make this findings. This was further confinmed in this study when the South facing slopes recorded the highest
number of landslides (see Figure 14). The south facing slopes in South Africa, receive less sunlight as compared
to the north facing slopes. The increase water content in the soil decreases the cohesive strength of clay minerals
and also pore fluid pressure can have negative impact on slope stability. Sunlight plays a significant role in
decreasing the water content in the soil, in two ways (1) releasing of water from the ground through evaporation,
and (2) releasing of water from the ground through the process of transpiration. The south facing slopes are
generally wetter than the north facing slope, resulting in an increased susceptibility to landslides (Stapelberg pers
com, 2011). It was also discovered by Boelhouwers et al. (1998) that 78 % of the landslides fell on the South
facing slopes, in an area within the Western Cape Province of South Africa. The north facing slopes recorded a
very small number of landslides (see Figure 14), this can be explained by the fact that the north facing slope
receive a significant amount of sunlight, in South Africa. Therefore the negative impact of increased water
content is eliminated, as water is released from the ground through the process of evapotranspiration. The high
sunlight content can have negative impact on landslide susceptibility, if the dominant soil type in the area in
question is expansive clay i.e montmorillonite. The clay would expand during rainy days and contract during
sunny days, resulting in the weakening of the bonds between clay particles. Expansive clays are known not to be
suitable for building construction. Stapelberg (pers com, 2011) also highlighted increased susceptibility in clayey
rich soil.
87
Vegetation has been proven by Varnes (1985) to have positive effects on slope stability i.e. the natural anchorage
that limits the effects of rainfall on erosion, and the release of water from the ground, as a vapour during
evepotransiration. In general the north facing slopes are normally dry and therefore less vegetated and the south
facing slopes are generally much more wetter and therefore more vegetated, in South Africa. It would be
expected that the South facing slopes would be less susceptible to landslides, as they are more vegetated. This is
not the case, as this study has shown that the was a large number of landslides in the south facing slopes than the
north facing slopes, which also supports Boelhouwers et al. (1998) observations. Although vegetation plays a
positive role on decreasing the negative effect of water on slope stability, its impact can be inconsequential if
there is a limited amount of sunlight, as the rate of transpiration will be much lower and also the leafs can cover
the ground counteracting the process of evaporation.
The soil type is also a major parameter on slope instability as different soils have different hydrological
properties. Clay rich soils are more susceptible to landslide occurrence, but surprisingly a large number of
landslides occurred on the rocky outcrops. But also a considerable number of landslide occurred on deep soils.
Soil depth has been found to be the most important parameter when modeling landslide susceptibility maps. Soil
depth less than 300 mm and between 600 and 900 mm recorded the highest and second highest number of
landslides, respectively. Deepest soil class had the least number of landslides but this was mainly because this
class covered mealy less than 1% of the study areas. It has been noted by Dahal et al. (2007) that deep soils are
more susceptible to landslides, but this was not the case in the study area, mainly because the deeper soils
comprised less than 1 percent of the study areas.
The geology of the area plays a significant role on landslide susceptibility, as different rock types have different
hydrological properties and different strength. Geological structures can have weakening effects on rocks. A
large number of landslides plotted on unconsolidated alluvium deposits. These deposits have a higher hydraulic
transmisivity and less cohesive strength, the effect of increased water content on cohesive strength of clayey
minerals is shown on Figure 4. Moreover, pore fluid pressure would have negative impact on slope stability.
Most of the landslides in the study area could be proven that they were triggered by rainfall event, hence the
hydrological properties of rocks is of importance when investigating landslide susceptibility. The sandstone and
quartzite units also recorded a significant number of landslides. Sandstone has a very high hydraulic
transimisvity and quartzites have slightly lower hydraulic conductivity. Higher transimisivity means easy
passage of water deep into the ground; increasing negative effects of water (i.e. decrease of cohesive strength of
clay minerals and increased pore fluid pressure (Warric, Mullen & Nielsen 1977, Wilson 1980). If the top rock
unit is composed of highly transimisive sandstone units and the lower unit is made of less permeable and
expansive clay units; such an area would be highly susceptible. However this needs to be specifically
88
investigated in future studies. Slighly more than 50 % of landslides fell within the 250 m buffer of faults and
geological contacts. This also shows that the faulted areas are much more susceptible to landslides. Faults and
lithological contacts are known to be zones of weakness; hence they are much more likely to be susceptible to
landslides than unfaulted areas.
Understanding and selecting the landslide causative factors is important, if the causative parameters are properly
understood, higher level of accuracy can be achieved on landslide susceptibility modelling. The accuracy of the
landslide susceptibility map strongly depends on the weighting of the causative parameters in accordance to their
influence to slope instability; therefore the selection of a suitable weighting system was critical in this study. In
this study the weight of evidence and the map combination approach were used. The weight of evidence method
produced exceptional results in comparison with the map combination approach. This can be seen from the
success rate curve and the cumulative percentage of these models, presented in Chapter 3. It is worth mentioning
that similar approach, as the map combination approach (DRASTIC) have been applied successfully to model a
ground water vulnerability maps, by the Council for Geosciences. The unsatisfactory results of the map
combination approach on landslide susceptibility mapping are largely attributed by the fact that there are several
causative parameters that govern slope instability. The map combination approach can be more useful when
modelling a certain type of mass movement i.e. rock falls, debris flow etc. The very high susceptibility class on
the susceptibility map modelled using the map combination approach coincides with the road cuts, even in low
lying areas. This is mainly because the expert classified road cuts as the major parameter when weighting the
causative factors. If the focus of the study was to model a rock fall susceptibility map, it is likely that the map
combination approach could have yielded acceptable results.
One of the primary goals of this study was to investigate if remote sensing techniques can be used to monitor the
variable parameters that influence landslide susceptibility, and use them as a warning system for potential slope
failure. The remote sensing techniques are ideal for monitoring changes that can increase the area’s susceptibility
to landslides, mainly for variable factors i.e. vegetation health. Lower than normal minimum NDVI values were
observed on the MODIS derived NDVI time series profile, and in some cases the minimum, maximum and
average NDVI values were low for the year prior to landslide occurrence. The Landsat derived NDVI change
detection also showed a decrease in NDVI for the season prior to landslide occurrence. These findings imply that
the vegetation health was poor for the season prior to landslide occurrences. Based on these findings, a decrease
in NDVI values (vegetation health) increases the area’s susceptibility to landslides and therefore remote sensing
can be used to monitor changes in variable factors, which favours landslide occurrence. There might be a
seasonal component to the variations, and the conclusions drawn from this data might not be entirely conclusive.
89
The triggering mechanism for the landslides within the study area was rainfall. Heavy rainfall for a number of
days is a major trigger for landslides in mountainous areas of the Western Cape. Due to the uncertainty in the
timing of landslide occurrences, the rainfall threshold that triggers landslide could not be properly investigated in
this study. The rainfall threshold that triggers landslide has not been properly investigated in this study. However
a 200 mm rainfall in the space of three days, in April 2005 triggered several shallow landslides (debris flow) in
the mountainous areas within the study area. This indicates that heavy rainfalls in a short space of time can
trigger landslides, in susceptible areas.
5.2 INTERGRATION OF LANDSLIDE SUSCEPTIBILITY MAPPING AND DEVELOPMENT

PLANNING
The susceptibility of the area to landslide occurrence is crucial for development planning. Landslide activities
can adversely affect human activities and also have socio-economic disturbances. In this regard, information
about the susceptibility of the area to landslides is crucial. Although landslide susceptibility maps are only
capable of depicting areas that are likely to have landslides, not the period at which the landslide can occur. The
planner can use these estimates to certain decisions regarding the site suitability, the type of development, and
appropriate mitigation measures. By so doing the planner has determined the acceptable or unacceptable risk for
development programs. Decisions can then be made regarding avoidance, prevention or mitigation of existing
and future landslide hazard in the development program.
Determining whether the is a need for landslide susceptibility information is the first step in ensuring that
landslide risk does not exceed an acceptable level in planning for future land use. Landslide risk can be defined
as the degree of loss due to a particular landslide. The objectives of landslide information are to determine which
landslide susceptible areas are best suited for what type of development activities. For example, assessing
landslide susceptibility would not be crucial when determining areas that could be national parks or game
reserves; on the other hand, landslide susceptibility would be important in development of building infrastructure
in mountainous, tropical terrain. The type of development in an area is deterministic for the amount of landslide
information needed for the area in question. The failure to consider the potential effect of landslides on a project
can bring an increased risk; and some other projects like road cuts can affect the susceptibility of an area to
landslide, therefore landslide information is crucial for development planning
(http://www.oas.org/dsd/publications/unit/oea66e/ch10.htm).
90
5.3 RECOMMENDATIONS
1. Detailed landslide inventory and classification of the landslides prior to the modeling of landslide
susceptibility map needs to be done. This would allow the researcher to develop a rock fall susceptibility
map, debris flow susceptibility map and a deep seated landslide susceptibility map. This is crucial as
these class of landslides are influence by different causative factors e.g. roads and rail cuts are more
influential on rockfalls,
2. Accurate dating (i.e. dendrochronological dating) of landslides would allow the correlation of landslides
with rainfall events and help on investigating the rainfall threshold that triggers landslide in the study
area and Western Cape as a whole. A lot of work still needs to be done on the investigation of the
rainfall threshold that triggers landslides.
3. Future studies should put more focus on performing sensitivity analysis, as this can help in
understanding the influence of different landslide causative parameters, which govern landslide
susceptibility.
4. Although very high accuracies of landslide susceptibility was obtained using the weight of evidence
method, in theory, for landslide susceptibility even higher accuracies would be ideal. Additional expert-
based information and additional parameters affecting landslide susceptibility should be investigated.
5. It would be interesting if future studies investigate the performance of the map combination approach,
on a specific class of mass movement, and see if it yields acceptable results.
5.4 CONCLUSIONS
• The weight of evidence method is ideal when modelling landslide susceptibility maps. The results
attained when modelling a landslide susceptibility map using this method where highly acceptable.
• The map combination approach is not the best method to use when modelling landslide susceptibility
maps.
• Remote monitoring of variable factors i.e. vegetation can be achieved using remote sensing techniques
and this can be used as the warning system.
91
REFERENCES
Aleotti P. & Chowdhury R 1999. Landslide Hazard Assessment: Summary, Review and New
Perspectives, Bulletin of Engineering Geology & Environment, vol. 58, pp.21-44.
Anderson GL, Hanson JD & Haas RH 1993. Evaluating landsat thematic mapper derived
vegetation indices for estimating above-ground biomass on semiarid rangelands. Remote Sensing
of the Environment. vol. 45(2), pp.165-175.
Athanasopoulos GA., Pelekis, PC.& Leonidou E A 1995. Effects of surface topography on seismic
ground response in the Egion (Greece) 15 June 1995 earthquake, Soil Dyn. Earthqu. Eng., vol. 18,
pp.135–149.
Atkinson PM & Massari R 1998. Generalized linear modelling of landslide susceptibility in the
Central Apennines, Italy. Computer Geoscience, vol.24 (4), pp. 373–385.
Boelhouwers J, Duiker JMC & van Duffelen EA 1998. Spatial, morphological and
sedimentological aspects of recent debris flows in Du Toit's Kloof, Western Cape. South African
Journal of Geology. vol. 101, pp.73–89.
Bonham-Carter GF, Agterberg FP & Wright DF 1988. Integration of geological datasets for gold
exploration in Nova Scotia. Photogram. Remote Sens., vol. 54, no. 11, pp. 1585–1592.
92
Bonham-Carter GF, Agterberg, FP & Wright DF 1989. Weights of evidence modelling: a new
approach to mapping mineral potential, in Agterberg. Statistical Applications in the Earth
Sciences: Geological Survey Canada paper 9-9, pp. 171–183.
Borcherdt RD 1970. Effects of local geology on ground motion near San Francisco Bay, Bull. Seis.
Soc. Am. vol.60, pp. 29–61.
Brunettii MT, Peruccacci S, Rossi M., Luciani S, Valigi D & Guzzetti F 2010. Rainfall thresholds
for the possible occurrence of landslide in Italy. Natural Hazards and Earth System Sciences,
vol.10, pp. 447-458.
Campbell RH 1975. Soil slips, debris flows, and rainstorms in the Santa Monica Mountains and
vicinity, southern California. In: US Geological Survey Professional Paper 851. Washington DC:
U.S. Government Printing Office, pp. 51.
Cardinali M, Galli M, Guzzetti F, Ardizzone F., Reichenbach P & Bartoccini P 2005. Rainfall
induced landslides in December 2004 in South-Western Umbria, Central Italy. Nat Hazard Earth
Sys Sci 6. pp. 237–260.
Carson MA & Kirkby MJ 1972. Hillslope form and process. Cambridge University Press,
Cambridge University Press, Cambridge, pp. 475.
93
Chauhan S, Sharma M, Arora MK & Gupta NK 2010.Landslide susceptibility zonation through
ratings derived from artificial neural network. International journal of applied earth observation
and geoinformation, vol. 12, pp. 340-350.
Chung CJF & Shaw JM 2000. Quantitative Prediction Models for Landslide Hazard Mapping.
http://www.nrcan.gc.ca/gsc/mrd/sdalweb/sdi_cd/index.html
Clark RN 1999. Spectroscopy of Rocks and Minerals, and Principles of Spectroscopy. In A. N.
Rencz (Ed.). Manual of Remote Sensing. Remote Sensing for the Earth Sciences, New York: John
Willey and Sons, vol. 3, pp. 3-58.
Clerici A, Perego S, Tellini C & Vescovi P 2002. A procedure for landslide susceptibility zonation
by the conditional analysis method. Geomorphology, vol. 48 (4), pp. 349-364.
Corominas J 2000. Landslides and climate. Keynote lecture- In Proceedings 8th International
Symposium on Landslides, (Bromhead E, Dixon N, Ibsen ML, eds). Cardiff: A.A. Balkema, vol. 4,
pp.1–33.
Corominas J & Moya J, 1999. Reconstructing recent landslide activity in relation to rainfall in the
Llobregat River basin, Eastern Pyrenees, Spain. Geomorphology. vol. 30, pp. 79–93.
Crozier & Michael J 1986. Landslide, consequences & environment.
94
Cruden DM & Varnes DJ 1996. Landslide types and Processes. In Landslides investigation and
Mitigation. Transportation Research Board, US National Research Council. Special Report, vol.
247, pp. 36-75.
Chavez PS, Jr 1996. Image-based atmospheric corrections-revised and improved. Photogranometric
Engineering and Remote Sensing, vol. 62 (9), pp.1025-1036.
Dahal RK, Hasegawa S, Nonomura A, Yamanaka M, Masuda T& Nishino K 2007. GIS-based
weight of evidence modelling of rainfall-induced landslides in small catchments for landslide
susceptibility mapping. Environ Geol.
Dai FC, Lee CF, Li J & Xu ZW 2001. Assessment of landslide susceptibility on the natural terrain
of Lantau Island, Hong Kong. Environ Geol. vol.40, pp.381–391.
Dai FC & Lee CF 2002. Landslide characteristics and slope instability modelling using GIS,
Lantau Island, Hong Kong. Geomorphology, vol.42: pp. 213-228.
Elias PB & Bandis SC 2000. Neurofuzzy Systems in Landslide Hazard Assessment, In: Proceedings
of 4th International Symposium on Spatial Accuracy Assessment in Natural Resources and
Environmental Sciences, pp. 199-202.
Gao B 1996. NDWI – A normalized difference water index for remote sensing of vegetation liquid
water from space, Remote Sensing of Environment, vol. 58(3), pp.257–266.
95
Garland G & Olivier MJ 1993. Predicting landslides from rainfall in a humid, sub-tropical region,
Geomorphology, vol. 8, pp. 165-173.
Glade T & Crozier M 2005b. A review of scale dependency in landslide hazard and risk analysis.
Glade T 2005. Linking debris-flow hazard assessments with geomorphology, Geomorphology, vol
66, pp. 189-213.
Gomez H & Kavzoglu T 2005. Assessment of Shallow Landslide Susceptibility using Artificial
Neural Networks in Jabonosa River Basin, Venezuela, Engineering Geology, vol 78(1-2), pp.11-27.
Gray DH & Leiser, AT 1982. Biotechnical slope protection and erosion control. Van Nostrand
Reinhold, New York.
Greenway DR 1987. Vegetation and slope stability. In: Anderson MG, Richards KS (eds) Slope
stability. Wiley, New York, pp.187–230.
Griggs GB & Plant N 1998. Coastal-bluff failures in northern Monterey Bay induced by the
earthquake, in: The Loma Prieta, California, earthquake of 17 October 1989. Landslides, (Ed)
Keefer, D., USGS Prof. Paper 1551-C, C33–C50.
96
Hall FG, Strebel DE, Nickeson JE & Goetz SJ 1991. Radiometric rectification. Toward a common
radiometric response among multidate, multisensor images, Remote Sensing of Environment, vol
35, pp.11–27.
Hanvey PM., Lewis CA. & Lewis GE 1986. Periglacial slope deposits in Carlisle's Hoek, near Rhodes,
Eastern Cape Province. South African Geographical Journal, vol. 68, pp. 164–174.
Harp EL, Wilson RC & Wieczorec GF 1981. Landslides from the 4 February 1976, Guatemala
earthquake, The Guatemala earthquake of 4 February 1976, Geology. Survey. Prof. Paper 1204-A,
pp.1–35.
Hung LQ, Batelaan O, San DN & Van TT 2005. GIS –Remote sensing application of landslide
hazard mapping-Case study Thua-Thien-Hue Province, Vietnam. New Strategies for European
remote sensing.
Iverson MR 2000. Landslide triggering by rain infiltration. Water resource research, vol.36 (7),
pp.1897-1910.
Janisch EP 1931. Notes on the central part of the Soutspanberge Range and on the origin of Lake
Funduzi. Transactions of the Geological Society of South Africa, vol. 34, pp. 151–162.
Johnson MR, Annhauser CR & Thomas RJ 2006. The geology of South Africa.
Johannesburg/Pretoria. Geological Society of South Africa/Council for Geoscience. pp 691.
97
Joyce KE, Samsonov S & Levick SR 2011. The current status of remote sensing for mapping and
monitoring geological hazards.
Kanungo DP, Arora MK., Sarkar S & Gupta RP 2009. Landslide susceptibility zonation (LSZ)
mapping-A review. Journal of South Asia Disaster Studies, vol.2.
Kim HH & Elman GC 1990. Normalization of satellite imagery. International Journal of Remote
Sensing, 11(8), pp.1331-1347.
Lee S, Choi J & Min K 2004b. Probabilistic Landslide Hazard Mapping using GIS and Remote
Sensing Data at Boun, Korea. International Journal of Remote Sensing. vol. 25, pp. 2037-2052.
Lewis CA. & Illgner PM 1998. Fluvial conditions during the Holocene as evidenced by alluvial
sediments from above Howison's Poort, near Grahamstown, South Africa. Transactions of the Royal
Society of South Africa, vol.53, pp. 53–67.
Lewis CA 1996. Periglacial features. In The Geomorphology of the Eastern Cape, South Africa,
Grocott & Sherry, Grahamstown. pp. 103–119.
Mantovani F, Soeters F & Van Western CJ 2000. Remote sensing techniques for landslide studies
and hazard zonation in Europe. Geomorphology, vol. 15, pp. 213-225.
Mc Cloy KR 2010. Development and evaluation of phenology changes indices derived from time
series of image data. Remote Sensing, vol. 12, pp.2442-2473.
98
Mc Feeter SK 1996. The normalised difference water index (NDWI) in the delineation of open
water features. International Journal of Remote Sensing, vol. 17, pp. 1425-1432.
Morton M, Alvarez, RM & Campbell RH 2003. Preliminary soil-slip susceptibility maps, south
western California. USGS Open-File Report OF 03-17.
Paige-Green P 1989. Landslides: extent and economic significance in southern Africa. In: Brabb,
Paruelo JM, Epstein HE, Lauenroth WK & Burke IC., 1997. ANPP estimates from NDVI for the
central grassland region of United State of America. Ecology, vol. 78(3), pp.953-958.
Pats S & Chavez JR 1996. Image-Based atmospheric corrections-Revised and improved.
Photogrammetric Engineering and Remote Sensing, vol. 62 (9), pp.1025-1036.
Perry CR & Lautenschlager LF 1984. Functional equivalence of spectral vegetation indices.
Remote Sensing of Environment, vol. 14, pp.169-182.
Prandini L, Guidicini G, Buttura JA, Pancano WL & Santos AR 1977. Behaviour of the vegetation
in slope stability: A critical review. Int.Ass. Engineering Geology, vol.16, pp.5-51.
Rahman KM 2011. Remote sensing-based determination of deciduous and understory phenology
over boreal forest. PhD thesis. University of Gagary Department of geoinformatics Engineering.
99
Ray RL, Jacobs JM & de Alba P 2009. Impact of vadose zone soil moisture and groundwater on
slope instability, Journal of Geotechnical and Geoenvironmental Engineering.
Roering J & McKean J 2004. Objective landslide detection and surface morphology mapping using
high resolution airborne laser altimetry. Geomorphology, vol 57, pp. 331-351.
Sarkar S & Kanungo DP 2004. An integrated approach for landslide susceptibility mapping using
remote sensing and GIS. Photogrammetric Engineering & Remote Sensing, vol. 70 (5), pp. 617-
625.
Siddle HJ, Jones DB & Payne HR 1991. Development of a methodology for landslip potential
mapping in the Rhondda Valley In: Chandler RJ (ed) Slope stability engineering. Thomas Telford,
London, pp.137–142.
Sidle RC, Pearce AJ & O'Loughlin, CL 1985. Hillslope Stability and Land Use, Water Resources
Monograph (Washington, D.C.: American Geophysical Union, 1985), Series No. 11.
Singh R, Forbes C, Diop S, Musekiwa C & Claasen D 2011. Report on landslide geohazards, their
socio-economic impacts, mitigation and remediation measures as well as landslide susceptibility
mapping of South Africa. Earth Observation & Geological Hazard Assessment: Towards creation of
the Geological Atlas of South Africa.
100
Singh RG 2009. Landslide classification, characterization and susceptibility modelling in
KwaZulu-Natal. Master of Science Thesis, Witwatersrand University.
Soeters R & van Westen CJ 1996. Slope Instability Recognition, Analysis and Zonation. In: Turner.
A. K.& Schuster R. L. (eds.) Landslides, Investigation and Mitigation, Transportation Research
Board, National Research Council, National Academy Press, Washington, DC, U.S.A., Special
Report 247, pp. 129-177.
Spudich P, Hellweg M, & Lee WHK 1996. Directional topographic site response at Tarzana
observed in aftershocks of the Northridge, California, earthquake: implications for mainshock
motions, Bull. Seis. Soc. Am, vol.86, pp.193–208.
Stoner ER & Baumgardner MF 1980. Physiochemical site, and bidirectional reflectance factor
characteristics of uniformly moist soils. Tech Report 11679, LARS/Purdue University of West
Lafayette.
Styczen ME & Morgan RPC 1995. Engineering properties of vegetation. In: Morgan RPC, Rickson
RJ (eds) Slope stabilisation and erosion control: a bioengineering approach, London, pp 5–58.
Swanson FJ & Dyrness CT 1975. Impact of Clearcutting and Road Construction on Soil Erosion by
Landslides in the Western Cascade Range, Oregon, Geology, vol. 3, pp. 393-396.
101
Temesgen B, Mohammed MU & Korme T 2001. Natural hazard assessment using GIS and remote
sensing methods, with particular reference to the landslide in the wondogenet area, Ethiopia.
Phys.Chem.Earth ©, vol.26 (9), pp. 665-675.
Terzaghi K 1950. Mechanism of landslides, in Application of Geology to Engineering Practice.
Van Schalkwyk A & Thomas MA 1991. Slope failures associated with the floods of September
1987 and February 1988 in Natal and Kwa-Zulu, Republic of South Africa. Geotechnics in the
African Environment, Blight et al. (Eds), pp. 57-63.
Varnes DJ 1978. Slope movement types and processes. In: Schuster R.L and Krizek R. J., ED.,
Landslides, analysis and control. Transportation Research Board Special report, vol. 176, pp 11-
33.
Varnes DJ 1984. Landslide hazard zonation: a review of principles and practice. IAEG
Publication. Paris, Unesco.
Warrick AW, Mullen GJ & Nielsen DR 1977. Scaling field measured soil hydraulic properties
using a similar media concept. Water Resource. Res. vol. 13 (2), pp. 355-362.
Wagner H & Pathe C 2008. Has SAR failed in soil moisture retrieval? ENVISAT & ERS
Symposium, Salzburg, Austria, 6-10 September 2004, ESA SP-572, pp. 1745-1751.
102
Wieczorek GF 1984. Preparing a Detailed landslide-Inventory map for Hazard Evaluation and
Reduction, Bulletin of International Association of Engineering Geologists, vol. 21, pp.337-342.
Wieczorek GF 1987. Effect of rainfall intensity and duration on debris flows in central Santa Cruz
Mountains. In: Debris flow/avalanches: process, recognition, and mitigation (Costa JE, Wieczorek
GF, eds), Geological Society of America, Reviews in Engineering Geology, vol.7: pp. 93–104.
Wilson RC 1989. Rainstorms, pore pressures, and debris flows: a theoretical framework. In:
Landslides in a semi-arid environment (Morton DM, Sadler PM, eds). California: Publications of
the Inland Geological Society, vol.2, pp. 101–117.
Wu W & Siddle RC 1995. A distributed slope stability model for steep forested basins. Water
Resource Res, vol. 31:pp.2097–2110.
Xiao X., Zhang J., Yan H., Wu W & Biradar C 2009. Land surface phenology: Convergence of
Satellite and Co2 Eddy Flux observations. A. Noormets (Ed.), Phenology of ecosystem processes,
New York: Springer Siences & Business Media, pp. 247-270.
Zhou Y 2006. Slope stability. Geotechnical Engineering. Publication No. FHWA NHI-06-088.
SPECIAL REFERENCES
Stapelberg 2011. Engeneering geologist at Council for Geosciences. Bellville. Interviewed on the 9
June 2011 abouth the parameters that affect the area’s susxceptibility to landlside.
103
Landslide hazard zonation (Online). Available from

http://www.oas.org/dsd/publications/unit/oea66e/ch10.htm (Accessed 30 October 2012).
104

Tyoda Landslide 2013

Uploaded by

Copyright:

Available Formats

Tyoda Landslide 2013

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tyoda Landslide 2013

Uploaded by

Copyright:

Available Formats

LANDSLIDE SUSCEPTIBILITY MAPPING: REMOTE SENSING AND GIS

Supervisor: Dr Jaco Kemp

Copyright © 2013 Stellenbosch University

All rights reserved

Landslide susceptibility mapping: Remote sensing and GIS approach ..................................................... i

Contents ................................................................................................................................................... iii

Tables ..................................................................................................................................................... xiv

ACKNOWLEDGEMENTS ................................................................................................................... xvi

Chapter 1 Background to the study ..................................................................................................... 1

1.1 Introduction to Landslides and landslide susceptibility ............................................................. 1

1.2 Project aims and goals ................................................................................................................ 2

1.3 The research questions and objectives ....................................................................................... 3

Chapter 2 Literature review and theory ............................................................................................... 9

2.1 Landslide Classification ............................................................................................................. 9

2.2 Landslide controlling parameters ............................................................................................. 11

2.2.1 Static factors ...................................................................................................................... 13

2.2.2 Variable factors ................................................................................................................. 15

2.2.3 Triggering mechanisms ..................................................................................................... 17

2.4 GIS modelling for landslide susceptibility mapping ................................................................ 18

2.4.1 Remote sensing on landslide susceptibility mapping ....................................................... 22

2.5 Final remarks ............................................................................................................................ 25

3.1 Input data .................................................................................................................................. 26

3.1.1 Geology ............................................................................................................................. 26

3.1.2 Geomorphology ................................................................................................................ 30

3.1.4 Anthropogenic influences ................................................................................................. 41

3.2 Landslide susceptibility mapping – the weights of evidence APPROACH............................. 42

3.2.1 Weight of evidence landslide susceptibility map .............................................................. 44

3.2.2 Accuracy Assessment for the Weight of Evidence method .............................................. 49

3.3 Landslide susceptibility mapping – the Map combination Approach ...................................... 53

3.3.1 Accuracy Assessment for the Map Combination approach .............................................. 59

Chapter 4 Remote monitoring of variable conditions and identification of triggering mechanisms. 62

4.1 Extraction of NDVI and NDWI – MODIS data ....................................................................... 65

4.2.1 Image pre-processing ........................................................................................................ 74

4.3 Triggering mechanisms ............................................................................................................ 82

Chapter 5 Discussion, recommendations and conclusions ................................................................ 87

5.1 Discussion ................................................................................................................................ 87

5.2 Intergration of landslide susceptibility mapping and development planning........................... 90

5.3 Recommendations .................................................................................................................... 91

5.4 Conclusions .............................................................................................................................. 91

Special references ................................................................................................................................. 103

ARC Agricultural Research Council

landslide positions. ....................................................................................................................................6

Figure 3: Geological map of the study area. .............................................................................................8

the cohesive strength. ...............................................................................................................................16

et al. 2009 pp 11). ....................................................................................................................................19

facing slopes recorded the least number of landslides. ...........................................................................33

shown in the legend. .................................................................................................................................34

depth between 900-1200mm recorded the least number of landslides. ...................................................37

the training data set. ................................................................................................................................52

the reference data set ...............................................................................................................................53

Council for Geosciences. .........................................................................................................................59

percentage for the training data. .............................................................................................................60

for the reference data. ..............................................................................................................................61

and south western part of the study area. ................................................................................................63

occurrence (02-11-2006). ........................................................................................................................64

NDVI values for the year prior to landslide occurrence. ........................................................................68

occurrence (2006). ...................................................................................................................................72

Figure 47: Steps followed when performing atmospheric corrections. ...................................................77

than 800 mm. ............................................................................................................................................84

mm, with the month of April recording close to 250mm. .........................................................................86