Content uploaded by Mercy Mwaniki
Author content
All content in this area was uploaded by Mercy Mwaniki on Jul 10, 2015
Content may be subject to copyright.
209
Landslide Inventory Using Knowledge Based
Multi-sources Classification Time Series Mapping:
A Case Study of Central Region of Kenya
Mercy Mwaniki1, Matthias Möller2 and Gerhard Schellmann1
1,2Bamberg University; Germany · mercimwaniki@yahoo.com
2Beuth Hochschule für Technik, Berlin/Germany
Full paper double blind review
Abstract
Advances in classification using multispectral remote sensing imagery have gained in-
creasing attention in solving environmental problems, and the management of disasters
such as floods and landslides, due to their wide coverage and enabling ease of access in
times of calamities. Multispectral data has facilitated the mapping of soils, land-cover, and
structural geology, all of which are factors affecting landslide occurrence. The main aim of
this research was to map landslide-affected areas using remote sensing techniques for the
central region of Kenya, where landslide disasters are common occurrences. The study area
has a highly rugged terrain, and rainfall has been the main trigger of recent landslide events.
False colour composite (FCC), Principal Component Analysis (PCA), Independent Com-
ponent Analysis (ICA), spectral indices in the form of Normalised Difference Index (NDI),
and knowledge-based classification formed the methodology. PCA and ICA were per-
formed on Landsat data sets, and the components with the most geologic information after
factor loading analysis were chosen to be used in a colour composite. The blue component
of the colour composite was a spectral index involving bands 7 and 3 for Landsat ETM+, or
bands 7 and 4 for Landsat OLI. The FCC formed the inputs for knowledge based classifica-
tion with the following 13 classes: runoff, extreme erosions, other erosions, landslide areas,
highly erodible, stable, weathering rocks, agriculture, green, new forest regrowth areas, as
well as clear, turbid, and salty water. Validation of the mapped landslide areas with field
GPS locations of landslide affected areas showed that 66% and 62% of the points coincided
well with landslide areas mapped in the years 2000 and 2014, respectively. The classi-
fication maps showed extreme erosions taking place along drainage channels and other
erosions in agricultural areas; with highly eroble zones charchaterised by already weathered
rocks or deposit area, while fluvial deposits mainly characterised runoff areas. Thus,
landuse and rainfall processes play a major role in landslide processes in the study area.
1 Introduction
Advances in classification in remote sensing to applications in environment, disaster moni-
toring, and management, have attracted attention in recent research including landslide
mapping. Detection of landslides through remote sensing intepretation techniques requires
GI_Forum ‒ Journal for Geographic Information Science, 1-2015.
© Herbert Wichmann Verlag, VDE VERLAG GMBH, Berlin/Offenbach. ISBN 978-3-87907-558-4.
© ÖAW Verlag, Wien. ISSN 2308-1708, doi:10.1553/giscience2015s209.
M. Mwaniki, M. Möller and G. Schellmann 210
image processing methods, which enhance recogntion leading to their classification. Land-
slide recognition and mapping is facilitated by the difference in spectral characteristics,
contrast from the sorrounding areas, and their morphological expression (MANTOVANI et al.
1996). Therefore, multispectral remote sensing data have gained more utility in landslide
mapping and monitoring, owing to their better discrimination through image enhancement
techniques. The main objective of this research was to map landslide areas using remote
sensing technologies, in order to create a landslide inventory for rainfall induced landslides
in the central region of Kenya. In the past, the study area has experienced landslides
triggerred by rainfall, but little has been done to map the affected areas, although some
studies (NGECU & ICHANG’I 1998, NGECU et al. 2004) recorded the incidences and inves-
tigated possible causes − among them heavy rainfall triggers and geology. Little has been
done scientifically to link the landslides and the controlling factors in this study area.
Creating a landslide inventory will boost landslide susceptibility studies as a second step in
this research.
The successful implementation of a classification system requires suitable variables be used
in the classification (LU & WENG 2007). Such variables may include vegetation indices,
feature space transformed images, band ratios, topographical data, textural data, multi-
temporal images, multisensor data, and spectral signatures. The ability to incoporate an-
cillary data in a classification system using expert rules has greately reduced spectral con-
fusion among landcover classes in a complex biophysical environment. For example,
FRANKLIN et al. (1994) and MEYER et al. (1993) improved land-cover classification ac-
curacy by incorporating topographic data in mountainous regions, since land cover distri-
bution relates to relief and elevation, while MYINT (2001) incorporated texture analysis in
an urban land use classification. The implementation of such classifications is enabled
through knowledge based classification techniques (AMARSAIKHAN & DOUGLAS 2004),
decision rules, and artificial neural networks.
Research based on improving image classification, such as that conducted by JENSEN
(1996) and LANDGREBE (2003), has empasized the selection of optimal spectral bands for
use in classifications. Multivariate analysis techniques of PCA and ICA work by trans-
forming and reducing multispectral axes into orthogonal components, which preserve as
much of the desired information as possible. The use of tranformed feature space data (e.g.
minimum noise fraction, tassel cap, PCA, ICA) and feature extraction techniques can
greatly boost a classification, by resulting in well differentiated landcover classes. For
Example; DI PRINZIO et al. (2011) incorporated PCA and ICA to improve unsupervised
neural classification (SOM) for a catchment elements classification.
2 Remote Sensing of Disturbed Vegetation
Landslides are localized phenomena, which reflect specific conditions of site stability,
where vegetation cover contributes to root cohesion and thus influence slope stability
(ROERING et al. 2003, SCHMIDT et al. 2001). Applications involving landsliding or risk map-
ping require the identification of disturbed vegetation or loss of vegetation, which are then
related to landslide incidences. For example, MEYER et al. (2001) managed to relate how
fire and storm events acceleralated soil erosion and shallow landsliding in Idaho batholith.
Vegetation indices provide a measure of the protective role of vegetation, in places where
Landslide Inventory Using Knowledge Based Multi-sources Classification 211
vegetation plays a crucial role in preventing soil erosion (DEWITTE et al. 2012). The amount
of fuel moisture influences burning efficiency (CECCATO et al. 2002).
Spectral indicies of vegetation utilise the near infrared and visible (red) bands to distinguish
healthy vegetation (high reflectance in the near infrared region) from diseased, burnt, or
dying vegetation, which have decreased reflectance in the NIR region (JENSEN 2007). This
provides the basis for vegetation monitoring, where the vegetation indices are designed to
assess the contribution of green vegetation content to multispectral observations. BANNARI
et al. (1995) have summarised the commonly used vegetation indices, usually presenting a
complex mixture of vegetation, soil brightness, environmental effects, shadow, soil colour,
and moisture. CECCATO et al. (2002) described a methodology of designing a spectral index
in the solar spectrum domain and the limitations and robustness of such an index. Spectral
vegetation indices often take the form of a NDI, expressed as equation 1 (BRIGHT et al.
2010) or ratio form. Examples are: NDVI (TUCKER 1979), NDWI (GAO 1996), NBR
(EPTING et al. 2005), PRI (BRIGHT et al. 2010).
where is the measured reflectance at bands and
ij
ij
NDI i j
(1)
Landsat imagery has found major applications in developing spectral indices, not only for
burnt vegetation (e.g. PATTERSON & YOOL 1998, ROGAN & YOOL 2001), but also for
mineral mapping (NIELD et al. 2007), soil applications (e.g. BOETTINGER et al. 2008), and
band ratios for geological applications (ALI et al. 2012, KAVAK 2005, SABINS 1999).
VOHORA & DONOGHUE (2004) developed a Normalised Difference Mid infrared spectral
index (NDMIDIR) using Landsat bands 7 and 4 to map areas of disturbed vegetation
(including stressed forests or forest fires). NDMIDIR and NDVI were produced using
equations 2 and 3, where NDMIDIR successfully indentified landslide scars and forest fire
scars compared to NDVI, which only depicted areas with healthy vegetation.
TM4 TM7
NDMIDIR TM4 TM7
(2) TM4 TM3
NDVI TM4 TM3
(3
)
Image colour enhancement techniques, involving processed images or multispectral data
provide a powerful means of visualising landslide features. For example, MONDINI et al.
(2011) used an FCC comprising (δNDVI, IC4, PC4) to visualise landslide areas triggered
by typhoons, while FERNÁNDEZ et al. (2008) used several FCCs comprising bands from the
visible, NIR, and SWIR regions to show landslide scarps and lithological changes related to
landslides. Landslide mapping utilising spectral indices such as NDVI (e.g. TSAI et al.
2010) or NDMIDIR (e.g. ZHANG et al. 2005) had similar enhancement levels as the use of
FCC (FERNÁNDEZ et al. 2008). WHITWORTH et al. (2005) used FCC image enhancement
using airborne photography imagery and investigated the effect of texture enhancement in
indentifying landslides, where the classification successfully differentiated landslide areas
from stable slopes. PCA techniques as an image enhancement method has also been used in
landslide mapping (e.g. PETLEY et al. 2002, RAWASHDEH et al. 2006). PC2 was used to map
burn servereity by (PATTERSON & YOOL 1998, ROGAN & CHEN 2004).
Research presented here has modified NDMIDIR spectral index developed by VOHORA &
DONOGHUE (2004), to NDMIDR for landsat TM/ETM+ and NDMIDIR for Landsat OLI.
These have been used in a false colour composite involving independent and principal
component as the basis of landslide mapping. The modification was informed by usage of
M. Mwaniki, M. Möller and G. Schellmann 212
bands in CAMPBELL (1996), where Landsat TM bands 2 and 4 are highly reflective zones
for vegetation, while band 3 is helpful for discriminating soil from vegetation due to the
high absorbency effect of vegetation. Bands 5 and 7 are best suited for rock and soil stud-
ies, since soil has a high absorption in band 7, and a high reflectance in band 5.
3 Methodology
3.1 Study Area
The study area is located in the central region of Kenya and ranges in longitude from
35°34´00"E to 38°15´00"E, and latitudes 0°53´00"N to 2°10´00"S (Figure 1). It has a high-
ly rugged mountainous terrain, with deep incised river valleys and narrow ridges, and alti-
tude varying from 450m to 5100m above mean sea level. Deep weathering of rocks is attri-
buted to soil formation where NGECU et al. (2004) noted 3 major types of soils: nitosols,
andosols, and cambisol. Land use and land cover are mainly dependent on climatic charac-
teristics, whereby mainly highland to savanna climate is prevalent in the region, thus forest,
agriculture and settlement are the most prevalent land use and land covers. Landslides trig-
gered by rainfall are also a major threat on the south eastern slopes of the Aberdare moun-
tain ranges, as reported in MWANIKI et al. (2011) and NGECU & ICHANG’I (1998).
Fig. 1: Map of the Study Area
Landslide Inventory Using Knowledge Based Multi-sources Classification 213
3.2 Data Description
Landsat 7, ETM+ (year 2000) and Landsat 8, OLI (year 2014), scenes p168r060, p168r061,
and p169r060 nearly free of cloud cover, were downloaded from the USGS web page, and
pre-processed individually for each year by layerstacking, mosaicking, and subsetting.
Atmospheric effects were reduced using the Landsat 5 TM haze reduction method in Erdas
Imagine, since path 169 and path 168 of Landsat imagery were taken about three weeks
apart, but in the same season. For Landsat 8, band 9 was subtracted from all the bands to
improve the image contrast. STRM DEM, 25m resolution for the same scenes was also
downloaded from the USGS site, mosaicked, and subset with the study area outline.
3.3 Image Enhancement and Knowledge Based Classification
Image enhancement involved principal and independent component analysis, formation of a
spectral index, which enhances de-vegetated and landslide areas, and their combination in a
false colour composite. A false colour composite involving the independent and principal
component with the highest contribution from bands 5,7 and 3, and a spectral index given
by equation 6 and 7 for landsat 5/7 and 8 respectively, was the basis for landslide areas
identification.
73
73
TM TM
NDMIDR TM TM
(6) or 74
74
OLI OLI
NDMIDIR OLI OLI
(7)
NDMIDR was compared to NDVI visually, where NDMIDR was found to have enhanced
geological features, while NDVI highlighted only vegetation greenness content. FERNÁN-
DEZ et al. (2008) recommended the combination of textural analysis and digital classifica-
tion in order to identify landslide features or mobilized areas. Therefore, NDMIDR spectral
index provided some textural characteristics and was used in the FCC to visualise landslide
areas. The input variables to the knowledge based classification were: principal component,
independent component, normalised difference index involving bands 7 and 3 for Landsat
5/7, and bands 4 and 7 for Landsat 8, slope and elevation. The FCC combinations for the
years 2000 and 2014 are presented in figure 2.
Fig. 2: FCC combinations involving IC, PC and Spectral Index with bands 7 & 3/4
M. Mwaniki, M. Möller and G. Schellmann 214
The expert classification rules were formed using one of the methods stated by (HODGSON
et al. 2003), where rules could be generated from observed data using cognitive methods.
Each false colour combination was first visually examined, followed by assigning major
colours to a particular class, after which the histogram of each layer was examined, and the
highest and lowest values noted. For each class, the file pixel values range was then
examined, and set in the knowledge based file by inquiring at various points with the same
colour. A trial run classification was performed after which the rules and classes were
refined further until all the values in each layer of the FCC had been assigned a class. The
classification rules for this study are as presented in the tables 1 and 2 below. Elevation and
slope variables were used to restrict crop and water covers to <3200m and <15 degrees,
respectively. A total of 13 classes, namely: runoff, extreme erosions, other erosions, land-
slide areas, highly erodible, stable, weathering rocks, agriculture areas, green, new forest
regrowth, rivers/clear water, turbid water, and salty water were identified and mapped as
shown in figures 3 and 4 in the next section.
Table 1: 2000 classification rules
4 Result and Discussion
The classification results after running the classification rules in tables 1 and 2 are pre-
sented in figures 3 and 4, where landslide areas were mapped amongst other vegetated areas
(green forest, new forest regrowth, agricultural crops), water or non-vegetated covers which
Landslide Inventory Using Knowledge Based Multi-sources Classification 215
were associated with eroded areas (extreme erosions, other erosions,), depositional areas
(runoff, highly erodible, stable), or rocks. The landslide mapping results between the years
2000 versus 2014 are analysed, using table 3 for comparison. The classification may serve
to explain the contribution of land use and rainfall to landsliding, where depositional and
water areas had increased, whereas eroded, exposed rocks, as well as landslide areas and
vegetation had decreased comparing the years 2000 and 2014 (table 3). The observations
are consistent with the heavy El-Niño rains falling in the period October 1997 to May 1998,
and whose devastating effects are recorded by NGECU & MATHU (1999). Consequently, the
classification for the year 2000 had more landslide areas compared to the year 2014. De-
posit areas and water cover are, however, on the increase compared to vegetated areas.
Landsat SWIR bands (5 and 7) are sensitive to canopy moisture content (VOHORA &
DONOGHUE 2004). Therefore, the use of band 7 in the NDMIDR spectral index together
with band 3 captured vegetation moisture properties and differentiated soil reflectance in
vegetated areas, thereby overcoming the challenge of using the NDVI spectral index, which
emphasized only vegetated areas. Also, Landsat band 7, which is predominantly used as a
geology band, contributed to the NDMIR spectral index by incorporating geology aspects,
and provided texture details unlike the NDVI, which is very smooth. The SWIR region is
sensitive to clay bearing minerals, while the VNIR region provides some information about
iron oxides (ABDEEN & HASSAN 2009). Therefore, the NDMIR highlighted clayey areas
and possible deposit areas. A colour image comprising IC, PC, and NDMIR spectral indices
best visualized landslide areas, where the components allowing maximum separation and
containing the highest contribution from bands 5, 7, and 3 were chosen. Principal com-
ponent and independent canonical analysis, which are feature extraction and data reduction
methods (JENSEN 1996), were used to extract pertinent information from Landsat bands 1-5
& 7.
Fig. 3: Erodibility/landslide
classification {Inputs: IC1, PC5,
band ratio index {7-3/7+3}
Fig. 4: {Inputs: IC2, PC3}
M. Mwaniki, M. Möller and G. Schellmann 216
Table 3: Representation of landslide areas in percentage in comparison to other covers
Landslides mapped through classification were verified in comparison with sample GPS
points, previously mapped in a field study over the study area. 66% and 62% of the points
coincided well with landslide areas mapped in the years 2000 and 2014, respectively. This
is because landslide rehabilation takes place as soon as the rain period is over, thereby
making it difficult to map some areas by the remote sensing method described in this study.
Instead, information from the local inhabitants was found useful in verifying the landuse
activities (deforestation, cultivation on very steep slopes, lack of proper water channels),
thereby explaining why some GPS points coincided with eroded or deposit areas in the
classification maps.
5 Conclusions and Recommendations
The colour image comprising (IC, PC, NDMIDR in R, G, B) displayed important surface
characteristics of lithology, soil moisture, and vegetation canopy moisture, which facilitated
the classification mapping, differentiating between areas characterised by landslides, ero-
sion, deposits, stable, runoff, exposed weathering rocks, vegetated areas, and water cover.
Specifically, the use of a spectral index involving bands 7 and 3 for Landsat TM/ETM+,
and bands 7 and 4 for Landsat OLI as a Normalised Difference Index helped to emphasize
moisture content in vegetation and soils, while the use of PC and IC components captured
the lithology components. The classification achieved the objective of mapping landslide-
affected areas in the years 2000 and 2014, by distinguishing, among other covers, mainly
vegetated, non-vegetated or water areas. The classification may serve to explain the contri-
bution of land use and rainfall to landsliding, whereby the classification for the year 2000
captured the devastating effects of El-Niño rains (October 1997 to May 1998), while the
classification for 2014 captured decreasing vegetation and increasing depositional areas.
Landslide Inventory Using Knowledge Based Multi-sources Classification 217
Field observations showed rehabiliation taking place in some affected areas; thus it was not
possible to map such areas as landslide areas in the classification method.
The classification maps showed extreme erosions taking place along drainage channels,
other erosions taking place in agricultural areas, highly eroble zones with already weathered
rocks or deposit area, while runoff was mainly fluvial deposits. The results obtained are
going to form the basis for landslise suceptibility mapping for the study area, considering
other landslide factors such as slope, closeness to drainage and roads, changes in landuse,
geology, rainfall, soil properties, and aspect.
Acknowledgement
We thank Nathan Agutu and all our anonymous reviewers for their contribution to improve
the quality of this paper. We acknowledge support from USGS website for the Landsat
imagery and thank DAAD/NACOSTI for their continued support for post graduate research
study grant no A/12/94131.
References
ABDEEN, M. M. & HASSAN, S. M. (2009), Utilisation of Spectral Signature and PCA, of
TERRA ASTER images for exploring new sites of building sand and gravels, NW Gulf
of Suez, Egypt.
ALI, E. A., EL KHIDIR, S. O., BABIKIR, A. A. & ABDELRAHNAM, E. M. (2012), Landsat
ETM+7 Digital Image Processing Techniques for Lithological and Structural Lineament
Enhancement: Case Study Around Abidiya Area, Sudan. Open Remote Sens. J., 5,
83-89.
AMARSAIKHAN, D. & DOUGLAS, T. (2004), Data fusion and multisource image classifica-
tion. Int. J. Remote Sens., 25, 3529-3539.
BANNARI, A., MORIN, D., BONN, F. & HUETE, A. R. (1995), A review of vegetation indices.
Remote Sens. Rev., 13, 95-120.
BOETTINGER, J. L., RAMSEY, R. D., BODILY, J. M., COLE, N. J., KIENAST-BROWN, S.,
NIELD, S. J., SAUNDERS, A. M., & STUM, A. K. (2008), Landsat Spectral Data for Digital
Soil Mapping. In: HARTEMINK, A. E., McBratney, A. & MENDONÇA-SANTOS, M. DE L.
(Eds.), Digital Soil Mapping with Limited Data. Springer, Netherlands Dordrecht,
193-202.
BRIGHT, A., HAMMELL, R. & AARDT, J. VAN (2010), Identification of spectral Indicators in
Vegetation Disturbance. Chester F. Carlson Centre for Imaging Science, Rochester In-
stitute of Technology.
CAMPBELL, J. B. (1996), Introduction to remote sensing. Guilford Press, New York, NY.
CECCATO, P., GOBRON, N., FLASSE, S., PINTY, B. &TARANTOLA, S. (2002 ), Designing a
spectral index to estimate vegetation water content from RS data: Part 1. Remote Sens.
Environ., 82, 188-197.
DEWITTE, O., JONES, A., ELBELRHITI, H., HORION, S. & MONTANARELLA, L. (2012), Satel-
lite remote sensing for soil mapping in Africa: An overview. Prog. Phys. Geogr., 36,
514-538.
M. Mwaniki, M. Möller and G. Schellmann 218
EPTING, J., VERBYLA, D. & SORBEL, B. (2005), Evaluation of remotely sensed indices for
assessing burn severity in interior Alaska using Landsat TM and ETM+. Remote Sens.
Environ., 96, 328-339.
FERNÁNDEZ, T., JIMÉNEZ, J., FERNÁNDEZ, P., EL HAMDOUNI, R., CARDENAL, F. J., DELGA-
DO, J., IRIGARAY, C. & CHACÓN, J. (2008), Automatic detection of landslide features
with RS techniques in the betic cordilleras (Granada, Spain). Int. Arch. Photogramm.
Remote Sens. Spat. Inf. Sci., 37, 351-356.
FRANKLIN, S. E., CONNERY, D. R. & WILLIAMS, J. A. (1994), Classification of alpine vege-
tation using Landsat Thematic Mapper, SPOT HRV and DEM data. Can. J. Remote
Sens., 20, 49-56.
GAO, B. (1996 ), NDWI ‒ A normalized difference water index for remote sensing of vege-
tation liquid water from space. Remote Sens. Environ., 58, 257-266.
HODGSON, M. E., JENSEN, J. R., TULLIS, J. A., RIORDAN, K. D. & ARCHER, C. M. (2003),
Synergistic Use of Lidar and Color Aerial Photography for Mapping Urban Parcel Im-
perviousness. Photogramm. Eng. Remote Sens., 69, 973-980.
JENSEN, J. R. (1996 ), Introductory digital image processing: a remote sensing perspective.
Prentice Hall, Upper Saddle River, N.J.
JENSEN, J. R. (2007), Remote Sensing of vegetation. In Remote Sensing of the Environ-
ment: An Earth Resource Perspective. Pearson Prentice Hall, Upper Saddle River, NJ.
KAVAK, K. S. (2005), Determination of palaeotectonic and neotectonic features around the
Menderes Massif and the Gediz Graben (West. Turkey) using Landsat TM image. Int. J.
Remote Sens., 26, 59-78.
LANDGREBE, D. A. (2003), Signal theory methods in multispectral remote sensing. Wiley,
Hoboken, New Jersey.
LU, D. & WENG, Q. (2007), A survey of image classification methods and techniques for
improving classification performance. Int. J. Remote Sens., 28, 823-870.
MANTOVANI, F., SOETERS, R. & VAN WESTEN, C. J. (1996), Remote sensing techniques for
landslide studies and hazard zonation in Europe. Geomorphology, 15, 213-225.
MEYER, G. A., PIERCE, J. L., WOOD, S. H. & JULL, A. J. T. (2001 ), Fire, storms, and ero-
sional events in the Idaho batholith. Hydrol. Process., 15, 3025-3038.
MEYER, P., ITTEN, K. I., KELLENBERGER, T., SANDMEIER, S. & SANDMEIER, R. (1993), Ra-
diometric corrections of topographically induced effects on Landsat TM data in alpine
environment. ISPRS Int. J. Photogramm. Remote Sens., 48, 17-28.
MONDINI, A. C., CHANG, K.-T. & YIN, H.-Y. (2011), Combining multiple change detection
indices for mapping landslides triggered by typhoons. Geomorphology, 134, 440-451.
MWANIKI, M. W., NGIGI, T. G. & WAITHAKA, E. H. (2011), Rainfall Induced Landslide
Probability Mapping for Central Province. In Fourth Intern. Summer School and Conf.,
(JKUAT, Kenya: Publications of AGSE Karlsruhe, Germany), 203-213.
MYINT, S. W. (2001), A Robust Texture Analysis and Classification Approach for Urban
Land Use and Land Cover Feature Discrimination. Geocarto Int., 16, 29-40.
NGECU, M. W. & ICHANG’I, D. W. (1998), The environmental impact of landslides on the
population living on the eastern footslopes of the Aberdare ranges in Kenya: a case
study of Maringa Village landslide. Environ. Geol., 38, 259-264.
NGECU, W. M. & MATHU, E. M. (1999), The El-Nino- triggered landslides and their socio-
economic impact on Kenya. Eng. Geol., 38, 277-285.
NGECU, W. M., NYAMAI, C. M. & ERIMA, G. (2004), The extent and significance of mass-
movements in Eastern Africa: case studies of some major landslides in Uganda and
Kenya. Environ. Geol., 46, 1123-1133.
Landslide Inventory Using Knowledge Based Multi-sources Classification 219
NIELD, S. J., BOETTINGER, J. L. & RAMSEY, R. D. (2007), Digitally Mapping Gypsic and
Natric Soil Areas Using Landsat ETM Data. Soil Sci. Soc. Am. J., 71, 245-252.
PATTERSON, M. W. & YOOL, S. R. (1998), Mapping Fire-Induced Vegetation Mortality
Using Landsat Thematic Mapper Data. Remote Sens. Environ., 65, 132-142.
PETLEY, D. N., CRICK, W. D. & HART, A. B. (2002), The use of satellite imagery in land-
slide studies in high mountain area. In: ACRS 2002, Kathmandu, Nepal.
DI PRINZIO, M., CASTELLARIN, A. & TOTH, E. (2011), Data-driven catchment classification:
application to the pub problem. Hydrol. Earth Syst. Sci., 15, 1921-1935.
RAWASHDEH, S. A., SALEH, B. & HAMZAH, M. (2006), The use of Remote Sensing Tech-
nology in geological Investigation and mineral Detection in El Azraq-Jordan. Cybergeo
Eur. J. Geogr. Syst. Model. Geostat.
ROERING, J. J., SCHMIDT, K. M., STOCK, J. D., DIETRICH, W. E. & MONTGOMERY, D. R.
(2003), Shallow landsliding, root reinforcement, and the spatial distribution of trees in
the Oregon Coast Range. Can. Geotech. J., 40, 237-253.
ROGAN, J., & CHEN, D. (2004), Remote sensing technology for mapping and monitoring
land-cover and land-use change. Prog. Plan., 61, 301-325.
ROGAN, J. & YOOL, S. R. (2001), Mapping fire-induced vegetation depletion in the
Peloncillo Mountains, Arizona and New Mexico. Int. J. Remote Sens. 22, 3101-3121.
SABINS, F. F. (1999), Remote sensing for mineral exploration. Ore Geol. Rev., 14, 157-183.
SCHMIDT, K. M., ROERING, J. J., STOCK, J. D., DIETRICH, W. E., MONTGOMERY, D. R. &
SCHAUB, T. (2001), The variability of root cohesion as an influence on shallow land-
slide susceptibility in the Oregon Coast Range. Can. Geotech. J., 38, 995-1024.
TSAI, F., HWANG, J.-H., CHEN, L.-C. & LIN, T.-H. (2010), Post-disaster assessment of land-
slides in southern Taiwan after 2009 Typhoon Morakot using remote sensing and spatial
analysis. Nat. Hazards Earth Syst. Sci., 10, 2179-2190.
TUCKER, C. J. (1979), Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sens. Environ., 8, 127-150.
VOHORA, V. K. & DONOGHUE, S. L. (2004), Application of remote Sensing data to Land-
slide mapping in Hong Kong. Int. Arch. Photogr. Remote Sens. GIS, V, 489-494.
WHITWORTH, M. C. Z., GILES, D. P. & MURPHY, W. (2005), Airborne remote sensing for
landslide hazard assessment: a case study on the Jurassic escarpment slopes of Wor-
cestershire, UK. Q. J. Eng. Geol. Hydrogeol., 38, 285-300.
ZHANG, Z., GONG, H., ZHAO, W. & ZHANG, Y. (2005), Application of remote sensing to
study of landslide. In: IGARSS 2005 Proceesings, IEEE, 1546-1549.
