Forgot your password?
Document Actions
Impacts from the 2004 Indian Ocean Tsunami: Analysing the Potential Protecting Role of Environmental Features (Chatenoux and Peduzzi, 2007)
Nat Hazards (2007) 40:289–304
DOI 10.1007/s11069-006-0015-9
ORIGINAL PAPER
Impacts from the 2004 Indian Ocean Tsunami: analysing
the potential protecting role of environmental features
B. Chatenoux Æ P. Peduzzi
Received: 23 August 2005 / Accepted: 23 February 2006 /
Published online: 17 October 2006
Ó Springer Science+Business Media B.V. 2006
Abstract The tsunami that deeply impacted the North Indian Ocean shores on 26
December 2004, called for urgent rehabilitation of coastal infrastructures to restore the
livelihood of local populations. A spatial and statistical analysis was performed to identify
what geomorphological and biological configurations (mangroves forests, coral and other
coastal vegetation) are susceptible to decrease or increase coastal vulnerability to tsunami.
The results indicate that the width of flooded land strip was, in vast majority, influenced by
the distance to fault lines as well as inclination and length of proximal slope. Areas covered
by seagrass beds were less impacted, whereas areas behind coral reefs were more affected.
The mangroves forests identified in the study were all located in sheltered areas, thus
preventing to address the potential protecting role of mangroves forests.
Keywords Tsunami Æ Indian ocean Æ Impact assessment Æ GIS Æ Bathymetry Æ
Vulnerability Æ Coral Æ Mangroves forests Æ Seagrass beds Æ Environment
Acronyms
CRED Centre for Research on Epidemiology of Disasters
DEM Digital Elevation Model
FAO Food and Agriculture Organisation
GIS Geographical Information System
IFRC International Federation of the Red Cross
UNEP United Nations Environment Programme
UNEP/GRID United Nation Environment Programme, Global Resource Information
Database
WCMC World Conservation Monitoring Centre
UTM Universal Transverse Mercator
GPS Ground Positioning System
B. Chatenoux Æ P. Peduzzi (&)
´ ˆ
UNEP/GRID-Europe, 11, ch. des Anemones, 1219 Chatelaine-Geneva, Switzerland
e-mail: pascal.peduzzi@grid.unep.ch
123
290 Nat Hazards (2007) 40:289–304
Introduction
Context
The tsunami that brought havoc to the North Indian Ocean coasts on 26 December 2004,
killed more than 226,000 persons, left millions in despair and caused nearly US $8 billions
damage from direct impacts (CRED 2005). In order to minimise risk in the future, United
Nations Environment Programme (UNEP), amongst other organisations, has called for
improved coastal management and rebuilding in safer places as well as in minimising
impacts on the environment. This is particularly relevant since the livelihood of the
population depends on the quality of the environment: tourism, fishing and aquaculture are
all economic activities requiring clean coasts and water. The scope of this study was to
improve understanding of factors leading to higher coastal vulnerability to tsunami, and
more specifically, to test whether environmental features could provide an efficient pro-
tection. To this end, a statistical and spatial analysis was conducted by UNEP/GRID-
Europe for the UNEP Asian Tsunami Disaster Task Force (UNEP 2005a).
Tsunami impacts
Why are some areas less impacted than neighbouring ones? Is it only dependent on
geomorphology, or do environmental features play a protecting role? Whereas the geo-
morphological role in tsunami propagation is well studied (e.g. Kowalik 2003) and the
influence of small-scale submarine topography has already been modelled (Mofjeld et al.
2000), less is known about the potential protective role of environmental features. Sci-
entific studies of the potential protective role of coral and mangroves forests are very
scarce and although several press releases stated that environment components such as
mangroves forests played a major role in reducing the impacts from the tsunami (Khor
2005; Friends of the Earth 2005), other more reliable sources mention the negligible role of
mangroves forests since they are mainly located in estuaries (Jimenez et al. 1985; Lewis
1982; Field 1996). Experiments conducted in in-door basins demonstrated that structures
with properties similar to mangroves forests decreased the height of a solitary wave in a
channel (Harada et al. 2002). Hiraishi and Harada (2003) highlighted the protecting role of
other coastal vegetation such as the Hibiscus tiliaceus, they also confirmed that mangroves
forests do not grow on sandy beaches. In situ observations delineate the protecting role of
other species such as Scaevola sericea and Pemphis acidula (UNEP 2005b).
Objective
The aim of this analysis was to assess the potential protective role of mangroves forests,
coral reefs, seagrass beds and coastal vegetation, apart from the near-shore geomorpho-
logical influence. To this end, data on bathymetry (water depth), orientation of the coast,
length of proximal slope, distance to tectonic features, presence of coral, seagrass beds,
mangroves forests and type of land cover were extracted using GIS technologies. Then, the
width of flooded land strip was evaluated either by interpreting high-resolution satellite
images or from available ground measurements. Finally, multiple regressions were per-
formed to identify the parameters that best explain the width of flooded land strip
(thereafter D) following a method already applied previously (Peduzzi et al. 2002; Dao and
Peduzzi 2004).
123
Nat Hazards (2007) 40:289–304 291
The study was based on global datasets to provide a first cut-off as well as to identify the
key parameters that are linked to higher coastal vulnerability to tsunami.
Data collection
Selection of the study area
The research was initiated in March 2005, and was based on information available at that
time. For instance, there was little material available for Seychelles, Yemen, Somalia, and
none for Burma and Andaman Islands (India). Thus the 62 sites selected are located in
Indonesia, Thailand, continental India, Sri Lanka and Maldives (Fig. 1). They cover a wide
range of different configurations (distance from tectonic event, bathymetry, as well as
environmental parameters).
Data on width of flooded land strip
The tsunami impact was determined using the maximal D in a given area. This information
was derived using several types of data.
Fig. 1 Study area and selected site distribution taken perpendicularly to the coastline
123
292 Nat Hazards (2007) 40:289–304
The first data type consists of interpreted satellite images (sources in Table 1) that show
the extent of the area flooded by the tsunami, based on an overlay and comparison of pre-
and post-tsunami images. Such images are the most accurate type of data. However,
coverage remains limited as it required good quality images and a long and methodical
processing.
In order to increase the number of test sites, post-tsunami satellite images were visually
interpreted. This was easy for large D values (several hundreds of metres), but difficult, if
not impossible, for width smaller than hundred metres.
The information on flooded land was completed using field surveys from the Research
Centre for Disaster Reduction Systems (DRS) and the Disaster Prevention Research
Institute (DPRI) of Kyoto University. The aim of those surveys being an evaluation of the
tsunami run up from clear landmark, locations available do not specifically correspond to
the maximal extent of the flooded area; it only confirms that the water reached at least this
distance.
Data for extracting potential parameters related to width of flooded land strip
Two sets of parameters were collected as being prone to have an influence on the width of
flooded land. The first set is related on the role of near coast geomorphology. The
hypothesis is based on studies such as Kowalik’s (2003), which modelled tsunami prop-
agation in presence of an escarpment and another study on interaction of tsunami waves
with small-scale submarine topography revealed that ‘‘the most important factor (...) is the
depth of a feature compared with the depth of the surrounding region’’ (Mofjeld et al.
2000). Consequently parameters were chosen in order to characterise near-shore
bathymetry changes.
The second set of parameters is related to the role of environmental features. This is less
documented and more empirical, hence a large panel of geographical and environmental
descriptors having a potential an effect on tsunami propagation were collected.
These parameters were extracted from a wide range of sources (Table 2): location of
epicentres coordinates; fault lines; elevation level; information on coastlines; land cover;
distribution of coral; mangroves forests and seagrass beds. The geomorphologic parameters
were obtained by computation and transformation of bathymetry or elevation (from
respectively GEBCO and SRTM), thus providing information on slope and depth.
Table 1 Interpreted and other satellite images data sources
Provider Data source
DLR – Centre for Satellite http://www.zki.caf.dlr.de/applications/2004/indian_ocean/indian_
Based Crisis Information (ZKI) ocean_2004_en.html
Global Land Cover Facility http://glcfapp.umiacs.umd.edu:8080/esdi/index.jsp
(GLCF) – ESDI
Service Regional de traitement http://sertit.u-strasbg.fr/documents/asie/asia_en.html
d’image et de teledetection (SERTIT)
UNEP/DEWA/GRID-Europe http://www.grid.unep.ch/activities/assessment/indianocean_
crisis/index.php
UNEP–WCMC imaps viewer http://tsunami.unep-wcmc.org/imaps/tsunami/viewer.htm
UNOSAT http://unosat.web.cern.ch/unosat/asp/charter.asp?id=55
USGS tsunami disaster website Restricted area
123
Table 2 Data sources
Nat Hazards (2007) 40:289–304
Data Providers Data source
Earthquakes epicentres and replicas Northern California Earthquake Data http://quake.geo.berkeley.edu/cnss/catalog-search.html
Centre and related contributors
Subduction fault UNEP/DEWA/GRID-Europe Digitised from USGS tectonic map
Digital Elevation Model (DEM) USGS, SRTM (90 m) http://srtm.csi.cgiar.org
Bathymetry General Bathymetric Chart of the Oceans (GEBCO) http://www.bodc.ac.uk/projects/gebco/index.html
Vector country border NIMA Vmap level 0, UN Cartographic Section http://www.mapability.com/info/vmap0_intro.html
Islands coastlines Christian DEPRAETERE, Institut de Recherche pour le Data no yet public
´ ´
Developpement (IRD) – Laboratoire d’etude des Transferts
en Hydrologie et Environnement (LTHE)
Global Land Cover 2000 EU/Joint Research Centre and related collaborators http://www-gvm.jrc.it/glc2000
Coral distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
Mangroves forests distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
Seagrass beds distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
123
293
294 Nat Hazards (2007) 40:289–304
Methodology
Theoretical model of tsunami for parameters’ selection
To explain the role of environmental parameters, an a priori estimation and standardisation
of the other parameters is needed. This can be achieved by modelling the effect of geo-
morphology.
During a tsunami, bathymetry has a direct link with wave height and velocity, a well
known process. When the water depth decreases, the wave slows down and the wavelength
decreases accordingly. This compresses the wave, which then builds up in height. The
wave breaks when water depth goes down to 1.3 times the wave height (Fox 2004).
Several others parameters were extracted, assuming the bathymetry could be reduced to
a model as shown in Fig. 2. Shore elevation, length and slope of the proximal and distal
slope, and depth at given distances from coast were acquired for each test site using GIS
techniques.
In order to also take into account the origin of the tsunami, distance from the fault line
as well as the angle of the waves with the coastline were included in the dataset. Finally,
the environmental parameters were integrated by estimating the percentage of coastline
behind coral reef, mangroves forests and seagrass beds. The coastal vegetation was
classified following an ordinal ranking in five classes of resistance (Table 3).
The table in the Appendix presents a synthesis of the variables used in the statistical
analysis.
Methodology of statistical analysis
To describe the GIS processing details is beyond the scope of this paper. This chapter will
summarise the statistical techniques that were applied.
In order to test the validity of the hypothesis (distance of impact dependant on
bathymetry and presence/absence of natural features), all the parameters for each site were
transformed so that multiple regression analysis could be applied. These transformations
are necessary to obtain a normal distribution as well as a standardisation to compare
different types of measures, such as percentage, angles or distances.
The variables were transformed by taking the natural logarithm (LN) of scalar or, in
some cases, the LN of transformed values. Transformations already proved to be efficient
in previous studies (Peduzzi et al. 2002, Dao and Peduzzi 2004). For variables ranging
D Lengprox Lengdist
d
lan
ed v Slprox
co
d
loo Ldg
ax f
of m n g
it
Ma Sea Sldist
Lim
10 m
al
Orient Cor
Ldto10m Pcav1km
1k m
Dff
Dfeq
Fig. 2 Bathymetric model (not to scale); for the parameters abbreviations see Table A1
123
Nat Hazards (2007) 40:289–304 295
Table 3 Land cover resistance-roughness ordinal index
Legend Resistance index
Herbaceous Cover, Bare Areas, Water Bodies 1
Shrub Cover, evergreen, deciduous, sparse shrub cover 2
Regularly flooded shrub and/or herbaceous cover, cultivated and managed areas 2
Mosaic: Cropland/Shrub and/or grass cover 2
Mosaic: Cropland/Tree Cover/Other natural vegetation 3
Mosaic: Tree Cover/Other natural vegetation, Tree cover, regularly flooded 4
Tree cover, broad-leaved, evergreen 5
between 0 and 1 (e.g. percentage) Equation 1 was applied, for transformation of angle and
orientation Equation 2 was used:
Equation 1. Transformation of variables ranging between 0 and 1 (e.g. percentage)
1
Vi ¼ LN
1 À Vi
where Vi is the variable to be transformed and Vi is the transformed value
Equation 2. Transformation of variables expressing angle/orientation
cos a
Va ¼ LN
1 À cos a
where a is the angle or orientation to be transformed and Va is the transformed value.
A correlation matrix was computed between all the variables and was used to separate
variables that are too correlated to be taken together in regression analysis. Groups of
independent variables were generated, each one corresponding to a specific hypothesis,
which was tested by running multiple regression analysis. The selection of the most rel-
evant hypothesis was based on relevance (p-value < 0.05) and maximisation of R2. This
process allows the identification of combinations of parameters that best explain the LN of
D and thus confirms or rejects the hypothesis on the role of the different environmental and
geomorphological features.
The choice of logarithmic regression was made to reflect the interactivity between the
different parameters, i.e. that they have a multiplicative effect on each other (an addition of
LN being a multiplication of the exponents). This is believed to be pertinent, given the
complexity of sites where one factor can mitigate or enhance another.
Results and discussion
Statistical results
The regression analysis identified correlations between combinations of parameters and the
width of the flooded land strip (D) (Table 4). Several combinations were relevant, the best
one consisting of the five following variables, namely: the distance from the tectonic origin
(distance from subduction fault line); the near-shore geomorphology (through average
depth at 10 km and length of proximal slope); but also with environmental features
(percentage of coraland percentage of seagrass beds).
123
296 Nat Hazards (2007) 40:289–304
Table 4 Best parameters combination with weights (B) and respective p-levels
p-levela
Variables B
)0.828
LnDFF 0.000014
)0.312
LnAV10KM 0.007119
LnLENGPROX 0.644 0.002405
)0.133
LnSEAG 0.000107
LnCORAL 0.158 0.000392
Intercpt 8.698 0.000000
R = 0.809, R2 = 0.655, N = 56 sites
where:
LnDFF = Ln (Distance from subduction fault line); LnAv10 Km = Ln (Average depth at 10 km);
LnLengprox = Ln (length of proximal slope); LnSeag = Ln (%age of seagrass beds); LnCoral = Ln (%age of
coral)
a
In broad terms, a p-value smaller than 0.05, shows the significance of the selected indicator, however this
should not be used blindly
The analysis was performed on 56 sites. A correlation coefficient of 0.81 was obtained
between the D and the parameters listed in Table 4, i.e., an R2 equal to 0.655, indicating
that about 65.5% of the variance is explained by the model. The very low p-values of the
variables (much smaller than 0.05) attest the significance of the selection.
From Table 4 values, an equation can be derived for evaluating the theoretical D in
other areas based on the values of the five selected variables (Equation 3).
Equation 3 Model for Width of flooded land strip
h
Dm ¼ exp 0:16ÁLnCoral À 0:13 Á LnSeag þ 0:64 Á LnLengprox À 0:31 Á LnAv10Km
i
À 0:83 Á LnDFF þ 8:70
where: Dm = modelled width of flooded land strip
To assess the validity of the model, the Dm plotted against the observed values D
(Fig. 3). In the scatter plot, the sites distribution shows several gaps, reflecting the geo-
graphical distribution of test sites. As countries are located at different range of distance
from the fault lines (Indonesia and Andaman being the closest, Maldives the further away),
there is not a continuum in the width of flooded land strip distribution, the countries closer
to the tectonic event origin being more impacted than those located farther away.
The model identified six outliers (white circles in Fig. 3) located in Maldives (1),
Thailand (1) Indonesia (1) and Sri Lanka (3). These are believed to be the result of
particular geomorphologic conditions that do not fit the model constrains, as well as the
different methods used to assess the maximal flooded distance (namely remote sensing
versus on-ground measurements).
The use of logarithmic regression can lead to large errors if used directly to model the
expected distance on a linear scale. In order to decrease this effect, only categories of
magnitude should be derived from such analysis.
A cluster analysis was run on the test sites using a classificatory tool (Dao 2004) to
minimise intra-class distances and minimise inter class distances. The following thresholds
of D were identified and adapted in order to gain in understanding (Table 5).
123
Nat Hazards (2007) 40:289–304 297
D observed
10 km
1 km
100 m
10 m
10 m 100 m 1 km 10 km
D modelled
Selected sites
Outliers
Fig. 3 Width of flooded land strip: predicted versus observed
In order to assess the confidence of the model, observed and modelled D were classified
using the rounded thresholds from Table 5. Then a comparison between the classes was
carried out to count when the modelled classes where similar to the observed classes and if
not, what was the number of classes differences. The matrix in Table 6 provides the results
with the differences of classes between observed and modelled on a region by region basis.
At the regional level, 42% of the sites correctly classify; 22% and 25% of the sites
being respectively within plus or minus one class. If these percentages are added, the
confidence of the model can be estimated at 89% to fall within plus or minus one class.
Since there are only five classes, this might seem to be a somewhat disappointing result;
however, given the fact that global data sets were used, this result was found to be
surprisingly good. Seen at a national level, the model underestimates the impact, especially
Table 5 Width of flooded land strip classes
Categories (impact) Width of flooded land strip (m) Rounded ranges (m)
1 (low) Less than 32.65 Lower than 30
2 (moderate) 32.65–107.35 30–100
3 (medium) 107.35–321.40 100–300
4 (high) 321.40–956.68 300–1000
5 (very high) Longer than 956.68 1000 and up
123
298 Nat Hazards (2007) 40:289–304
Table 6 Differences between modelled and observed classes (with of flooded land)
%
Cat. Error Nicobar India Indonesia Maldives Sri Lanka Thailand Total
)4 0
)3 0
)2 1 3 4 7
)1 4 2 1 7 1 15 25
0 2 2 11 3 3 4 25 42
1 3 2 2 3 3 13 22
2 1 1 2
3 1 1 2
4 1 1 2
Total 10 6 14 3 16 11 60
for Sri-Lanka. In Thailand, however, the impact is overestimated. This is believed to be
due to the geomorphological complexity of the proximal slope which is not taken into
account in the model, and which probably strongly dampens the energy of the tsunami in
reality.
Discussion
The five factors identified as having an influence on D, fall in three categories, namely:
distance from the fault line; width; geomorphology and environmental parameters, de-
scribed hereafter.
The negative sign before the coefficient means that the closer from the fault line, the
larger the value of D. This is consistent with description found in the literature ‘‘Tsunamis
typically cause the most severe damage and casualties very near their source. There the
waves are highest because they have not yet lost much energy to friction or spreading.’’
(NOAA 2004b).
Geomorphology of near-shore
The average depth at 10 km is related to the average slope of the sea floor. A steep slope is
known to block the energy of a tsunami, whereas a flatter slope is more dangerous as it
helps build up a higher wave. A greater depth for the same distance means a steeper slope,
hence less dangerous, a smaller depth being related to a flatter slope, more dangerous. The
negative sign before the coefficient is consistent with the theory.
The positive sign before the coefficient relative to the length of the proximal slope
means that a longer proximal slope is leading to a larger width of flooded land strip. This is
also related to the slope; the longer the length the lower the angle. Together with the
average depth, the two parameters indicate a higher risk configuration when a long shallow
area precedes the coast.
Environmental parameters
Seagrass beds (or seagrass substrate) seems to have a positive role in absorbing the energy
of tidal wave, the negative sign indicating that the higher the percentage of seagrass beds,
the shorter the D values. From such statistical analysis, it is impossible to differentiate if
123
Nat Hazards (2007) 40:289–304 299
the presence of seagrass beds has a mechanical influence that absorbs the energy of the
waves, or if the area that seagrass usually colonise is already protected from the wave. The
result, however, is that behind areas covered by seagrass, the distance of impact was in
majority shorter than in other areas having similar geomorphology.
Among the results of the statistical analysis, the surprise came from coral. A positive
sign preceding the coefficient suggest that the higher the percentage of coral, the larger the
D behind. This was unexpected, as one would imagine water behind a coral reef to be
somewhat sheltered.
A visual confirmation of this phenomenon was gained, using satellite images, which
confirmed larger D behind corals. In Fig. 4, despite a double barrier of coral reef, the area
on top of the map was more impacted than the area without reef. However, in this case, the
land elevation, facing the gap of coral reef, is steeper. The first hypothesis to explain this
positive correlation is that coral is mostly located in shallow areas, with a gentle slope
continuing inland, hence the low-lying areas are easily flooded. Conversely areas without
coral could be steeper hence would block the tsunami wave on a shorter distance in-land.
This apparently logical explanation was contradicted by the statistical verification, which
shows no correlation between presence/absence of coral and in-land slope, at least not with
the 90 m resolution data used.
Fig. 4 Example of coral influence in Lho’Nga, Sumatra (Indonesia)
123
300 Nat Hazards (2007) 40:289–304
Another explanation for this phenomenon comes from the length of tsunami waves,
which are about 1000 times longer than that of usual waves. If coral is offering protection
for usual waves, a more significant one might not be stopped but would continue to build
up on such shallow area.
This surprising result was backed up by UNEP ground assessments in Maldives and
Seychelles, where the following observations were made: ‘‘Fringing reef crests serve a
protective role against normal waves. However, in the case of the tsunami, major ter-
restrial and coastline damage was located in areas sheltered by fringing reefs. At these
locations, damage was focused near deeper channel that allowed the waves to break closer
onshore.’’ (UNEP 2005b, p. 19).
To better understand how coral, (or coral location slope) influences D, mathematical
modelling or in situ observations should be performed. Pending further investigations,
the results tend to indicate that it would not be wise to rebuild on coasts behind coral
reefs.
The case of mangroves forests
Mangroves forests are said to help reducing the impacts from tsunamis (Khor 2005; Friend
of the Earth 2005). If by common sense we can conceive that a barrier of vegetation with a
complex root system can indeed offer protection, during the present study it was impos-
sible to find patches of mangroves forests located on coast directly facing open sea.
Identification of mangroves forests was made by looking at both WCMC dataset and
satellite imagery. Mangroves forests were only present in estuaries, areas sheltered by
stretch of coastline or in protected bay (example in Fig. 5).
This was confirmed by literature: mangroves forests do not survive in area where wave
are too active (Jimenez et al. 1985; Lewis 1982, Field 1996, Hiraishi and Harada 2003). An
extract of an article from DIPE (2002) states that ‘‘mangrove establishment requires
protection from strong winds and wind generated waves, as wave action prevents seedling
establishment. As a consequence, mangrove communities tend to be located within shel-
tered coastal areas, surrounding highly indented estuaries, embayment and offshore
islands protected by reefs and shoals’’. In such case it is suspected that areas covered by
mangroves forests were less impacted by tsunami just because mangroves forests com-
munities tend to be located within sheltered coastal areas.
This is not to say that mangroves forests cannot protect coastlines, apart from their role
in filtering land run-off (Thom 1967) and reducing coastal erosion (Davis 1940). In the
case of tropical cyclones (one of the most devastating natural hazard in India and Ban-
gladesh), the role of mangroves forests could be important in reducing the impact from this
type of hazard (Saenger and Siddique 1993 in Kairo et al. 2003). In Vietnam, replanting
mangroves forests has helped reduce the cost of dyke maintenance by $7.3 m per year for
an investment of US $1.1 m (IFRC 2002). However, replanting mangroves forests can only
be done in areas suitable for them.
Conclusion
The applied method proved to successfully link contextual parameters with recorded D.
The model could be extrapolated to the whole area to provide five classes of exposure to
tsunami, thus easing the prioritisation of collection of data during the coastal management
123
Nat Hazards (2007) 40:289–304 301
Fig. 5 Example of mangroves forests location in Phangnga province (Thailand)
rebuilding procedures. This method was applied to identify vulnerability of exposed coast
depending on presence of seagrass beds, presence of coral, distance from event, length of
proximal slope as well as water depth at 10 km.
Whereas the geomorphological parameters role follow theoretical knowledge, the
environmental parameters were more surprising. Caution should be kept with the findings,
as the study was conduced using global datasets. The coarse resolution of bathymetry data
might not always capture the complexity of coastline at detailed scale.
To the question ‘‘are biological features a protection from tsunamis impacts?’’, the
answer varies with the type of environmental features. Remaining mangroves forests being
only identified in sheltered area in the observed cases, it is, therefore, difficult to distin-
guish whether the areas covered by mangroves forests suffered less impact because of their
intrinsic nature, or because they were sheltered by coastline or other physical protection.
Literature confirms that mangroves forests request calm water, but that they play a role in
preventing soil erosion (Saenger and Siddique 1993 in Kairo et al. 2003) and protecting
dikes (IFRC 2002).
If open sea and sandy beaches are not suitable for growing mangroves forests, other
types of vegetation can be used; such as waru trees (Hibiscus tiliaceus). In Maldives,
123
302 Nat Hazards (2007) 40:289–304
Magoo (Scaevola sericea) and Kuredhi (Pemphis acidula) shrubs have been reported as
having dissipated much of the tsunami force (UNEP 2005b). Similar vegetation might be
considered for protecting shores. To address the issue on the protecting role of mangroves
forests, areas where they were removed could be compared with areas still covered by
mangroves forests.
Areas where coral is growing are generally in shallow waters with small slopes, two
conditions leading to higher waves. Although there are little doubts in the positive pro-
tecting role of coral from usual waves, caution should be kept while rebuilding facilities in
the shore zone; the geomorphology of areas where coral usually grow might be not a safe
place for tsunami protection. This was confirmed by ground observation (UNEP 2005b).
Further research should be made with more detailed data as the coarse resolution of the
bathymetry and coral location prevent more precise conclusions. A mathematical model-
ling of wave, or in situ measure could be a good solution to study the behaviour of tsunami
waves in area covered by coral.
The statistical model show lower impacts in area behind seagrass beds. This could
be for two reasons, the mechanical role of seagrass beds acting as a damping filter that
help reducing the energy of the wave or because seagrass beds are located in areas
where the configuration of bathymetry is not favourable for building high wave. The
true reason remains unexplained, but the correlation with lower distance of impact is
significant.
It is important to note that all this analysis was made on a single event, the tsunami of 26
December 2004. Different magnitude and origin of a tsunami could result in drastically
different wavelengths thereby might induce different effects.
The identification of area most exposed to tsunami following our method can be used as
a first cutoff for choosing where more detailed data should be collected, modelling the
vulnerability of coast being the next logical step.
Acknowledgements The present publication is a part of a scientific Report for the UNEP Asian Tsu-
nami Disaster Task Force.This research would not have been possible without the close collaboration of
Alain Retiere and Olivier Senegas at UNOSAT, who provided free access to their collection of satellite
images, Phillip Fox and Kaveh Zahedi at WCMC who supplied the data on coral, mangroves forests and
seagrass beds, Christian Depraetere who provided the detailed coast lines for islands. We would like to
thank Jean-Michel Jaquet for his precious support as well as for reviewing this article, and Arthur Dahl
for his advices.
Appendix
Table A1 List of variables computed or extracted
Abbreviation Description Units
AV10 KM Average slope until 10 km Degrees
AV1 KM Average slope until 1 km Degrees
AV2_5 KM Average slope until 2.5 km Degrees
AV20 KM Average slope until 20 km Degrees
AV25 KM Average slope until 25 km Degrees
AV30 KM Average slope until 30 km Degrees
AV50 KM Average slope until 50 km Degrees
AV5 KM Average slope until 5 km Degrees
%age
CORAL Percentage of protection from coral preceding the site
COSORIEN Cosinus of orientation Scalar
DFEQ Distance from main earthquake Kilometres
123
Nat Hazards (2007) 40:289–304 303
Table A1 Continued
Abbreviation Description Units
DFF Distance from subduction fault line Kilometres
DFS Distance from source Kilometres
D Width of flooded land strip Metres
LDCOV Land cover resistance index Cardinal values 1 to 6
LDTO10M Average slope until an inland height of 10 m Degree
LDTO30M Average slope until an inland height of 30 m Degree
LENGDIST Length of distal slope Kilometres
LENGPROX Length of proximal slope Metres
%age
MANG Percentage of protection from mangroves
preceding the site
ORIENT Orientation between the tsunami energy Degrees
and a perpendicular to the coast
%age
PCAV10KM Average slope until 10 km
%age
PCAV1KM Average slope until 1 km
%age
PCAV2_5 K Average slope until 2.5 km
%age
PCAV20KM Average slope until 20 km
%age
PCAV25KM Average slope until 25 km
%age
PCAV30KM Average slope until 30 km
%age
PCAV40KM Average slope until 40 km
%age
PCAV500M Average slope until 5 km
%age
PCAV50KM Average slope until 50 km
%age
PCAV5KM Average slope until 5 km
%age
PCLDTO10 Average slope until an inland height of 10 m
%age
PCLDTO30 Average slope until an inland height of 30 m
%age
PCSLDIST Angle of Distal slope
%age
PCSLPROX Angle of Proximal slope
%age
SEAG Percentage of protection from Seagrass
beds preceding the site
SLDIST Angle of Distal slope Degree
SLPROX Angle of Proximal slope Degree
References
Articles and books
Dao H, Peduzzi P (2004) Global evaluation of human risk and vulnerability to natural hazards, In: Proc.
EnviroInfo 2004 conference, vol 1, Editions du Tricorne, Geneva, Switzerland, pp 435–446
Davis JH Jr (1940) The ecology and geological role of mangroves. In: Florida Carnegie Inst. Wash., vol 32,
Papers of the Tortugas Laboratory, pp 303–412
Field C (1996) Restoration of mangrove Ecosystems. International Tropical Timber Organization and
International Society for Mangrove Ecosystems, Okinawa, Japan
Harada K, Latief H, Imamura F (2002) Effect on reducing tsunami by the green belt and coastal permeable
structure, In: Proc. International Workshop on Tsunami Risk and Its reduction in the Asia-Pacific
Region, Bandung, Indonesia, Session 4
Hiraishi T and Harada K (2003) Greenbelt tsunami prevention in South-Pacific Region. Report Port Airport
Res Inst 42(2):1–23
IFRC (2002) Mangrove planting saves lives and money in Vietnam, World Disaster Report 2002, Inter-
national Federation of Red Cross and Red Crescent Societies (IFRC)
Jimenez JA, Martinez R and Encarnocion L (1985) Massive tree mortality in Puerto Rican mangroves
forests. Caribbian J Sci 21:75–78
Kairo JG, Dahdouh-Guebas F, Bosire J and Koedam N (2001) Restoration and management of mangrove
systems – a lesson for and from the East African region. South African J Bot 67:383–389
123
304 Nat Hazards (2007) 40:289–304
Kowalik Z (2003) Basic relations between tsunamis calculation and their physics – II. Sci Tsunami Hazards
21(3):152–173
Lewis RR (1982) Mangrove forests. In: Lewis RR (ed) Creation and restoration of coastal plant commu-
nities. CRC Press, pp153–173
Mofjeld HO, Titov VV, Gonzalez FI and Newman JC (2000) Analytic theory of tsunami wave scattering in
the open ocean with application to the North Pacific. NOAA Tech Memo OAR PMEL-116:1–38
Peduzzi P, Dao H and Herold C (2002) Global Risk And Vulnerability Index Trends per Year (GRAVITY),
Phase II: Development, analysis and results, scientific report. UNEP, Geneva Switzerland
Thom BG (1967) Mangrove ecology and deltaic geomorphology, Tabasco, Mexico. J Ecol 55:301–334
UNEP (2005a) After the tsumani – rapid environmental assessment report. UNEP, Nairobi, pp 1–140
UNEP (2005b) Maldives post-tsumami environmental assessment. UNEP, Nairobi, Kenya, p 19
References on the Internet
CRED (2005), EM-Dat, disasters list, http://www.em-dat.net
Dao H (2004), Classificator, Dpt of Geography, Geneva University.
http://geomap.unige.ch/sescat/stats/classify_func_example.html
DIPE (2002) Mangrove management in the Northern Territory, p. 5. Department of infrastructure planning
and environment, Northern Territory Government, Darwin NT, Australia.
http://www.lpe.nt.gov.au/dlpe/advis/land/mangrove/mangrovemgt/
DLR-ZKI (2005) ZKI satellite mapping supports international humanitarian relief teams in Southeast Asia,
http://www.zki.caf.dlr.de/applications/2004/indian_ocean/indian_ocean_2004_en.html
DRS-DPRI (2005) The December 26, 2004 earthquake tsunami disaster of Indian Ocean, Kyoto University,
http://www-drs.dpri.kyoto-u.ac.jp/sumatra/index-e.html
Fox R (2004) Water in motion, Department of Biology, Lander University,
http://www.lander.edu/rsfox/415hydrodyn-1Lec.html
Friends of the Earth (2005) Coral reefs and mangroves may have helped saved lives
http://www.foei.org/tsunami/malaysia.html
Khor M (2005), Save the Mangroves to fight tsunamis, Third World Network,
http://www.twnside.org.sg/title2/gtrends39.htm
Nelson SA (2005) What is a Tsunami?, http://www.tulane.edu/~sanelson/geol204/tsunami.htm
NOAA (2004a) International tsunami information center, Tsunami Glossary,
http://www.prh.noaa.gov/itic/library/pubs/glossary/tsu_glossary_html/tsunami_glossary11.html
NOAA (2004b) NOAA Backgrounder, NOAA and Tsunami,
http://www.prh.noaa.gov/itic/tsunami_events/media/factsheets/backgrounder.pdf
123
DOI 10.1007/s11069-006-0015-9
ORIGINAL PAPER
Impacts from the 2004 Indian Ocean Tsunami: analysing
the potential protecting role of environmental features
B. Chatenoux Æ P. Peduzzi
Received: 23 August 2005 / Accepted: 23 February 2006 /
Published online: 17 October 2006
Ó Springer Science+Business Media B.V. 2006
Abstract The tsunami that deeply impacted the North Indian Ocean shores on 26
December 2004, called for urgent rehabilitation of coastal infrastructures to restore the
livelihood of local populations. A spatial and statistical analysis was performed to identify
what geomorphological and biological configurations (mangroves forests, coral and other
coastal vegetation) are susceptible to decrease or increase coastal vulnerability to tsunami.
The results indicate that the width of flooded land strip was, in vast majority, influenced by
the distance to fault lines as well as inclination and length of proximal slope. Areas covered
by seagrass beds were less impacted, whereas areas behind coral reefs were more affected.
The mangroves forests identified in the study were all located in sheltered areas, thus
preventing to address the potential protecting role of mangroves forests.
Keywords Tsunami Æ Indian ocean Æ Impact assessment Æ GIS Æ Bathymetry Æ
Vulnerability Æ Coral Æ Mangroves forests Æ Seagrass beds Æ Environment
Acronyms
CRED Centre for Research on Epidemiology of Disasters
DEM Digital Elevation Model
FAO Food and Agriculture Organisation
GIS Geographical Information System
IFRC International Federation of the Red Cross
UNEP United Nations Environment Programme
UNEP/GRID United Nation Environment Programme, Global Resource Information
Database
WCMC World Conservation Monitoring Centre
UTM Universal Transverse Mercator
GPS Ground Positioning System
B. Chatenoux Æ P. Peduzzi (&)
´ ˆ
UNEP/GRID-Europe, 11, ch. des Anemones, 1219 Chatelaine-Geneva, Switzerland
e-mail: pascal.peduzzi@grid.unep.ch
123
290 Nat Hazards (2007) 40:289–304
Introduction
Context
The tsunami that brought havoc to the North Indian Ocean coasts on 26 December 2004,
killed more than 226,000 persons, left millions in despair and caused nearly US $8 billions
damage from direct impacts (CRED 2005). In order to minimise risk in the future, United
Nations Environment Programme (UNEP), amongst other organisations, has called for
improved coastal management and rebuilding in safer places as well as in minimising
impacts on the environment. This is particularly relevant since the livelihood of the
population depends on the quality of the environment: tourism, fishing and aquaculture are
all economic activities requiring clean coasts and water. The scope of this study was to
improve understanding of factors leading to higher coastal vulnerability to tsunami, and
more specifically, to test whether environmental features could provide an efficient pro-
tection. To this end, a statistical and spatial analysis was conducted by UNEP/GRID-
Europe for the UNEP Asian Tsunami Disaster Task Force (UNEP 2005a).
Tsunami impacts
Why are some areas less impacted than neighbouring ones? Is it only dependent on
geomorphology, or do environmental features play a protecting role? Whereas the geo-
morphological role in tsunami propagation is well studied (e.g. Kowalik 2003) and the
influence of small-scale submarine topography has already been modelled (Mofjeld et al.
2000), less is known about the potential protective role of environmental features. Sci-
entific studies of the potential protective role of coral and mangroves forests are very
scarce and although several press releases stated that environment components such as
mangroves forests played a major role in reducing the impacts from the tsunami (Khor
2005; Friends of the Earth 2005), other more reliable sources mention the negligible role of
mangroves forests since they are mainly located in estuaries (Jimenez et al. 1985; Lewis
1982; Field 1996). Experiments conducted in in-door basins demonstrated that structures
with properties similar to mangroves forests decreased the height of a solitary wave in a
channel (Harada et al. 2002). Hiraishi and Harada (2003) highlighted the protecting role of
other coastal vegetation such as the Hibiscus tiliaceus, they also confirmed that mangroves
forests do not grow on sandy beaches. In situ observations delineate the protecting role of
other species such as Scaevola sericea and Pemphis acidula (UNEP 2005b).
Objective
The aim of this analysis was to assess the potential protective role of mangroves forests,
coral reefs, seagrass beds and coastal vegetation, apart from the near-shore geomorpho-
logical influence. To this end, data on bathymetry (water depth), orientation of the coast,
length of proximal slope, distance to tectonic features, presence of coral, seagrass beds,
mangroves forests and type of land cover were extracted using GIS technologies. Then, the
width of flooded land strip was evaluated either by interpreting high-resolution satellite
images or from available ground measurements. Finally, multiple regressions were per-
formed to identify the parameters that best explain the width of flooded land strip
(thereafter D) following a method already applied previously (Peduzzi et al. 2002; Dao and
Peduzzi 2004).
123
Nat Hazards (2007) 40:289–304 291
The study was based on global datasets to provide a first cut-off as well as to identify the
key parameters that are linked to higher coastal vulnerability to tsunami.
Data collection
Selection of the study area
The research was initiated in March 2005, and was based on information available at that
time. For instance, there was little material available for Seychelles, Yemen, Somalia, and
none for Burma and Andaman Islands (India). Thus the 62 sites selected are located in
Indonesia, Thailand, continental India, Sri Lanka and Maldives (Fig. 1). They cover a wide
range of different configurations (distance from tectonic event, bathymetry, as well as
environmental parameters).
Data on width of flooded land strip
The tsunami impact was determined using the maximal D in a given area. This information
was derived using several types of data.
Fig. 1 Study area and selected site distribution taken perpendicularly to the coastline
123
292 Nat Hazards (2007) 40:289–304
The first data type consists of interpreted satellite images (sources in Table 1) that show
the extent of the area flooded by the tsunami, based on an overlay and comparison of pre-
and post-tsunami images. Such images are the most accurate type of data. However,
coverage remains limited as it required good quality images and a long and methodical
processing.
In order to increase the number of test sites, post-tsunami satellite images were visually
interpreted. This was easy for large D values (several hundreds of metres), but difficult, if
not impossible, for width smaller than hundred metres.
The information on flooded land was completed using field surveys from the Research
Centre for Disaster Reduction Systems (DRS) and the Disaster Prevention Research
Institute (DPRI) of Kyoto University. The aim of those surveys being an evaluation of the
tsunami run up from clear landmark, locations available do not specifically correspond to
the maximal extent of the flooded area; it only confirms that the water reached at least this
distance.
Data for extracting potential parameters related to width of flooded land strip
Two sets of parameters were collected as being prone to have an influence on the width of
flooded land. The first set is related on the role of near coast geomorphology. The
hypothesis is based on studies such as Kowalik’s (2003), which modelled tsunami prop-
agation in presence of an escarpment and another study on interaction of tsunami waves
with small-scale submarine topography revealed that ‘‘the most important factor (...) is the
depth of a feature compared with the depth of the surrounding region’’ (Mofjeld et al.
2000). Consequently parameters were chosen in order to characterise near-shore
bathymetry changes.
The second set of parameters is related to the role of environmental features. This is less
documented and more empirical, hence a large panel of geographical and environmental
descriptors having a potential an effect on tsunami propagation were collected.
These parameters were extracted from a wide range of sources (Table 2): location of
epicentres coordinates; fault lines; elevation level; information on coastlines; land cover;
distribution of coral; mangroves forests and seagrass beds. The geomorphologic parameters
were obtained by computation and transformation of bathymetry or elevation (from
respectively GEBCO and SRTM), thus providing information on slope and depth.
Table 1 Interpreted and other satellite images data sources
Provider Data source
DLR – Centre for Satellite http://www.zki.caf.dlr.de/applications/2004/indian_ocean/indian_
Based Crisis Information (ZKI) ocean_2004_en.html
Global Land Cover Facility http://glcfapp.umiacs.umd.edu:8080/esdi/index.jsp
(GLCF) – ESDI
Service Regional de traitement http://sertit.u-strasbg.fr/documents/asie/asia_en.html
d’image et de teledetection (SERTIT)
UNEP/DEWA/GRID-Europe http://www.grid.unep.ch/activities/assessment/indianocean_
crisis/index.php
UNEP–WCMC imaps viewer http://tsunami.unep-wcmc.org/imaps/tsunami/viewer.htm
UNOSAT http://unosat.web.cern.ch/unosat/asp/charter.asp?id=55
USGS tsunami disaster website Restricted area
123
Table 2 Data sources
Nat Hazards (2007) 40:289–304
Data Providers Data source
Earthquakes epicentres and replicas Northern California Earthquake Data http://quake.geo.berkeley.edu/cnss/catalog-search.html
Centre and related contributors
Subduction fault UNEP/DEWA/GRID-Europe Digitised from USGS tectonic map
Digital Elevation Model (DEM) USGS, SRTM (90 m) http://srtm.csi.cgiar.org
Bathymetry General Bathymetric Chart of the Oceans (GEBCO) http://www.bodc.ac.uk/projects/gebco/index.html
Vector country border NIMA Vmap level 0, UN Cartographic Section http://www.mapability.com/info/vmap0_intro.html
Islands coastlines Christian DEPRAETERE, Institut de Recherche pour le Data no yet public
´ ´
Developpement (IRD) – Laboratoire d’etude des Transferts
en Hydrologie et Environnement (LTHE)
Global Land Cover 2000 EU/Joint Research Centre and related collaborators http://www-gvm.jrc.it/glc2000
Coral distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
Mangroves forests distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
Seagrass beds distribution UNEP World Conservation Monitoring Centre (WCMC) http://www.unep-wcmc.org
123
293
294 Nat Hazards (2007) 40:289–304
Methodology
Theoretical model of tsunami for parameters’ selection
To explain the role of environmental parameters, an a priori estimation and standardisation
of the other parameters is needed. This can be achieved by modelling the effect of geo-
morphology.
During a tsunami, bathymetry has a direct link with wave height and velocity, a well
known process. When the water depth decreases, the wave slows down and the wavelength
decreases accordingly. This compresses the wave, which then builds up in height. The
wave breaks when water depth goes down to 1.3 times the wave height (Fox 2004).
Several others parameters were extracted, assuming the bathymetry could be reduced to
a model as shown in Fig. 2. Shore elevation, length and slope of the proximal and distal
slope, and depth at given distances from coast were acquired for each test site using GIS
techniques.
In order to also take into account the origin of the tsunami, distance from the fault line
as well as the angle of the waves with the coastline were included in the dataset. Finally,
the environmental parameters were integrated by estimating the percentage of coastline
behind coral reef, mangroves forests and seagrass beds. The coastal vegetation was
classified following an ordinal ranking in five classes of resistance (Table 3).
The table in the Appendix presents a synthesis of the variables used in the statistical
analysis.
Methodology of statistical analysis
To describe the GIS processing details is beyond the scope of this paper. This chapter will
summarise the statistical techniques that were applied.
In order to test the validity of the hypothesis (distance of impact dependant on
bathymetry and presence/absence of natural features), all the parameters for each site were
transformed so that multiple regression analysis could be applied. These transformations
are necessary to obtain a normal distribution as well as a standardisation to compare
different types of measures, such as percentage, angles or distances.
The variables were transformed by taking the natural logarithm (LN) of scalar or, in
some cases, the LN of transformed values. Transformations already proved to be efficient
in previous studies (Peduzzi et al. 2002, Dao and Peduzzi 2004). For variables ranging
D Lengprox Lengdist
d
lan
ed v Slprox
co
d
loo Ldg
ax f
of m n g
it
Ma Sea Sldist
Lim
10 m
al
Orient Cor
Ldto10m Pcav1km
1k m
Dff
Dfeq
Fig. 2 Bathymetric model (not to scale); for the parameters abbreviations see Table A1
123
Nat Hazards (2007) 40:289–304 295
Table 3 Land cover resistance-roughness ordinal index
Legend Resistance index
Herbaceous Cover, Bare Areas, Water Bodies 1
Shrub Cover, evergreen, deciduous, sparse shrub cover 2
Regularly flooded shrub and/or herbaceous cover, cultivated and managed areas 2
Mosaic: Cropland/Shrub and/or grass cover 2
Mosaic: Cropland/Tree Cover/Other natural vegetation 3
Mosaic: Tree Cover/Other natural vegetation, Tree cover, regularly flooded 4
Tree cover, broad-leaved, evergreen 5
between 0 and 1 (e.g. percentage) Equation 1 was applied, for transformation of angle and
orientation Equation 2 was used:
Equation 1. Transformation of variables ranging between 0 and 1 (e.g. percentage)
1
Vi ¼ LN
1 À Vi
where Vi is the variable to be transformed and Vi is the transformed value
Equation 2. Transformation of variables expressing angle/orientation
cos a
Va ¼ LN
1 À cos a
where a is the angle or orientation to be transformed and Va is the transformed value.
A correlation matrix was computed between all the variables and was used to separate
variables that are too correlated to be taken together in regression analysis. Groups of
independent variables were generated, each one corresponding to a specific hypothesis,
which was tested by running multiple regression analysis. The selection of the most rel-
evant hypothesis was based on relevance (p-value < 0.05) and maximisation of R2. This
process allows the identification of combinations of parameters that best explain the LN of
D and thus confirms or rejects the hypothesis on the role of the different environmental and
geomorphological features.
The choice of logarithmic regression was made to reflect the interactivity between the
different parameters, i.e. that they have a multiplicative effect on each other (an addition of
LN being a multiplication of the exponents). This is believed to be pertinent, given the
complexity of sites where one factor can mitigate or enhance another.
Results and discussion
Statistical results
The regression analysis identified correlations between combinations of parameters and the
width of the flooded land strip (D) (Table 4). Several combinations were relevant, the best
one consisting of the five following variables, namely: the distance from the tectonic origin
(distance from subduction fault line); the near-shore geomorphology (through average
depth at 10 km and length of proximal slope); but also with environmental features
(percentage of coraland percentage of seagrass beds).
123
296 Nat Hazards (2007) 40:289–304
Table 4 Best parameters combination with weights (B) and respective p-levels
p-levela
Variables B
)0.828
LnDFF 0.000014
)0.312
LnAV10KM 0.007119
LnLENGPROX 0.644 0.002405
)0.133
LnSEAG 0.000107
LnCORAL 0.158 0.000392
Intercpt 8.698 0.000000
R = 0.809, R2 = 0.655, N = 56 sites
where:
LnDFF = Ln (Distance from subduction fault line); LnAv10 Km = Ln (Average depth at 10 km);
LnLengprox = Ln (length of proximal slope); LnSeag = Ln (%age of seagrass beds); LnCoral = Ln (%age of
coral)
a
In broad terms, a p-value smaller than 0.05, shows the significance of the selected indicator, however this
should not be used blindly
The analysis was performed on 56 sites. A correlation coefficient of 0.81 was obtained
between the D and the parameters listed in Table 4, i.e., an R2 equal to 0.655, indicating
that about 65.5% of the variance is explained by the model. The very low p-values of the
variables (much smaller than 0.05) attest the significance of the selection.
From Table 4 values, an equation can be derived for evaluating the theoretical D in
other areas based on the values of the five selected variables (Equation 3).
Equation 3 Model for Width of flooded land strip
h
Dm ¼ exp 0:16ÁLnCoral À 0:13 Á LnSeag þ 0:64 Á LnLengprox À 0:31 Á LnAv10Km
i
À 0:83 Á LnDFF þ 8:70
where: Dm = modelled width of flooded land strip
To assess the validity of the model, the Dm plotted against the observed values D
(Fig. 3). In the scatter plot, the sites distribution shows several gaps, reflecting the geo-
graphical distribution of test sites. As countries are located at different range of distance
from the fault lines (Indonesia and Andaman being the closest, Maldives the further away),
there is not a continuum in the width of flooded land strip distribution, the countries closer
to the tectonic event origin being more impacted than those located farther away.
The model identified six outliers (white circles in Fig. 3) located in Maldives (1),
Thailand (1) Indonesia (1) and Sri Lanka (3). These are believed to be the result of
particular geomorphologic conditions that do not fit the model constrains, as well as the
different methods used to assess the maximal flooded distance (namely remote sensing
versus on-ground measurements).
The use of logarithmic regression can lead to large errors if used directly to model the
expected distance on a linear scale. In order to decrease this effect, only categories of
magnitude should be derived from such analysis.
A cluster analysis was run on the test sites using a classificatory tool (Dao 2004) to
minimise intra-class distances and minimise inter class distances. The following thresholds
of D were identified and adapted in order to gain in understanding (Table 5).
123
Nat Hazards (2007) 40:289–304 297
D observed
10 km
1 km
100 m
10 m
10 m 100 m 1 km 10 km
D modelled
Selected sites
Outliers
Fig. 3 Width of flooded land strip: predicted versus observed
In order to assess the confidence of the model, observed and modelled D were classified
using the rounded thresholds from Table 5. Then a comparison between the classes was
carried out to count when the modelled classes where similar to the observed classes and if
not, what was the number of classes differences. The matrix in Table 6 provides the results
with the differences of classes between observed and modelled on a region by region basis.
At the regional level, 42% of the sites correctly classify; 22% and 25% of the sites
being respectively within plus or minus one class. If these percentages are added, the
confidence of the model can be estimated at 89% to fall within plus or minus one class.
Since there are only five classes, this might seem to be a somewhat disappointing result;
however, given the fact that global data sets were used, this result was found to be
surprisingly good. Seen at a national level, the model underestimates the impact, especially
Table 5 Width of flooded land strip classes
Categories (impact) Width of flooded land strip (m) Rounded ranges (m)
1 (low) Less than 32.65 Lower than 30
2 (moderate) 32.65–107.35 30–100
3 (medium) 107.35–321.40 100–300
4 (high) 321.40–956.68 300–1000
5 (very high) Longer than 956.68 1000 and up
123
298 Nat Hazards (2007) 40:289–304
Table 6 Differences between modelled and observed classes (with of flooded land)
%
Cat. Error Nicobar India Indonesia Maldives Sri Lanka Thailand Total
)4 0
)3 0
)2 1 3 4 7
)1 4 2 1 7 1 15 25
0 2 2 11 3 3 4 25 42
1 3 2 2 3 3 13 22
2 1 1 2
3 1 1 2
4 1 1 2
Total 10 6 14 3 16 11 60
for Sri-Lanka. In Thailand, however, the impact is overestimated. This is believed to be
due to the geomorphological complexity of the proximal slope which is not taken into
account in the model, and which probably strongly dampens the energy of the tsunami in
reality.
Discussion
The five factors identified as having an influence on D, fall in three categories, namely:
distance from the fault line; width; geomorphology and environmental parameters, de-
scribed hereafter.
The negative sign before the coefficient means that the closer from the fault line, the
larger the value of D. This is consistent with description found in the literature ‘‘Tsunamis
typically cause the most severe damage and casualties very near their source. There the
waves are highest because they have not yet lost much energy to friction or spreading.’’
(NOAA 2004b).
Geomorphology of near-shore
The average depth at 10 km is related to the average slope of the sea floor. A steep slope is
known to block the energy of a tsunami, whereas a flatter slope is more dangerous as it
helps build up a higher wave. A greater depth for the same distance means a steeper slope,
hence less dangerous, a smaller depth being related to a flatter slope, more dangerous. The
negative sign before the coefficient is consistent with the theory.
The positive sign before the coefficient relative to the length of the proximal slope
means that a longer proximal slope is leading to a larger width of flooded land strip. This is
also related to the slope; the longer the length the lower the angle. Together with the
average depth, the two parameters indicate a higher risk configuration when a long shallow
area precedes the coast.
Environmental parameters
Seagrass beds (or seagrass substrate) seems to have a positive role in absorbing the energy
of tidal wave, the negative sign indicating that the higher the percentage of seagrass beds,
the shorter the D values. From such statistical analysis, it is impossible to differentiate if
123
Nat Hazards (2007) 40:289–304 299
the presence of seagrass beds has a mechanical influence that absorbs the energy of the
waves, or if the area that seagrass usually colonise is already protected from the wave. The
result, however, is that behind areas covered by seagrass, the distance of impact was in
majority shorter than in other areas having similar geomorphology.
Among the results of the statistical analysis, the surprise came from coral. A positive
sign preceding the coefficient suggest that the higher the percentage of coral, the larger the
D behind. This was unexpected, as one would imagine water behind a coral reef to be
somewhat sheltered.
A visual confirmation of this phenomenon was gained, using satellite images, which
confirmed larger D behind corals. In Fig. 4, despite a double barrier of coral reef, the area
on top of the map was more impacted than the area without reef. However, in this case, the
land elevation, facing the gap of coral reef, is steeper. The first hypothesis to explain this
positive correlation is that coral is mostly located in shallow areas, with a gentle slope
continuing inland, hence the low-lying areas are easily flooded. Conversely areas without
coral could be steeper hence would block the tsunami wave on a shorter distance in-land.
This apparently logical explanation was contradicted by the statistical verification, which
shows no correlation between presence/absence of coral and in-land slope, at least not with
the 90 m resolution data used.
Fig. 4 Example of coral influence in Lho’Nga, Sumatra (Indonesia)
123
300 Nat Hazards (2007) 40:289–304
Another explanation for this phenomenon comes from the length of tsunami waves,
which are about 1000 times longer than that of usual waves. If coral is offering protection
for usual waves, a more significant one might not be stopped but would continue to build
up on such shallow area.
This surprising result was backed up by UNEP ground assessments in Maldives and
Seychelles, where the following observations were made: ‘‘Fringing reef crests serve a
protective role against normal waves. However, in the case of the tsunami, major ter-
restrial and coastline damage was located in areas sheltered by fringing reefs. At these
locations, damage was focused near deeper channel that allowed the waves to break closer
onshore.’’ (UNEP 2005b, p. 19).
To better understand how coral, (or coral location slope) influences D, mathematical
modelling or in situ observations should be performed. Pending further investigations,
the results tend to indicate that it would not be wise to rebuild on coasts behind coral
reefs.
The case of mangroves forests
Mangroves forests are said to help reducing the impacts from tsunamis (Khor 2005; Friend
of the Earth 2005). If by common sense we can conceive that a barrier of vegetation with a
complex root system can indeed offer protection, during the present study it was impos-
sible to find patches of mangroves forests located on coast directly facing open sea.
Identification of mangroves forests was made by looking at both WCMC dataset and
satellite imagery. Mangroves forests were only present in estuaries, areas sheltered by
stretch of coastline or in protected bay (example in Fig. 5).
This was confirmed by literature: mangroves forests do not survive in area where wave
are too active (Jimenez et al. 1985; Lewis 1982, Field 1996, Hiraishi and Harada 2003). An
extract of an article from DIPE (2002) states that ‘‘mangrove establishment requires
protection from strong winds and wind generated waves, as wave action prevents seedling
establishment. As a consequence, mangrove communities tend to be located within shel-
tered coastal areas, surrounding highly indented estuaries, embayment and offshore
islands protected by reefs and shoals’’. In such case it is suspected that areas covered by
mangroves forests were less impacted by tsunami just because mangroves forests com-
munities tend to be located within sheltered coastal areas.
This is not to say that mangroves forests cannot protect coastlines, apart from their role
in filtering land run-off (Thom 1967) and reducing coastal erosion (Davis 1940). In the
case of tropical cyclones (one of the most devastating natural hazard in India and Ban-
gladesh), the role of mangroves forests could be important in reducing the impact from this
type of hazard (Saenger and Siddique 1993 in Kairo et al. 2003). In Vietnam, replanting
mangroves forests has helped reduce the cost of dyke maintenance by $7.3 m per year for
an investment of US $1.1 m (IFRC 2002). However, replanting mangroves forests can only
be done in areas suitable for them.
Conclusion
The applied method proved to successfully link contextual parameters with recorded D.
The model could be extrapolated to the whole area to provide five classes of exposure to
tsunami, thus easing the prioritisation of collection of data during the coastal management
123
Nat Hazards (2007) 40:289–304 301
Fig. 5 Example of mangroves forests location in Phangnga province (Thailand)
rebuilding procedures. This method was applied to identify vulnerability of exposed coast
depending on presence of seagrass beds, presence of coral, distance from event, length of
proximal slope as well as water depth at 10 km.
Whereas the geomorphological parameters role follow theoretical knowledge, the
environmental parameters were more surprising. Caution should be kept with the findings,
as the study was conduced using global datasets. The coarse resolution of bathymetry data
might not always capture the complexity of coastline at detailed scale.
To the question ‘‘are biological features a protection from tsunamis impacts?’’, the
answer varies with the type of environmental features. Remaining mangroves forests being
only identified in sheltered area in the observed cases, it is, therefore, difficult to distin-
guish whether the areas covered by mangroves forests suffered less impact because of their
intrinsic nature, or because they were sheltered by coastline or other physical protection.
Literature confirms that mangroves forests request calm water, but that they play a role in
preventing soil erosion (Saenger and Siddique 1993 in Kairo et al. 2003) and protecting
dikes (IFRC 2002).
If open sea and sandy beaches are not suitable for growing mangroves forests, other
types of vegetation can be used; such as waru trees (Hibiscus tiliaceus). In Maldives,
123
302 Nat Hazards (2007) 40:289–304
Magoo (Scaevola sericea) and Kuredhi (Pemphis acidula) shrubs have been reported as
having dissipated much of the tsunami force (UNEP 2005b). Similar vegetation might be
considered for protecting shores. To address the issue on the protecting role of mangroves
forests, areas where they were removed could be compared with areas still covered by
mangroves forests.
Areas where coral is growing are generally in shallow waters with small slopes, two
conditions leading to higher waves. Although there are little doubts in the positive pro-
tecting role of coral from usual waves, caution should be kept while rebuilding facilities in
the shore zone; the geomorphology of areas where coral usually grow might be not a safe
place for tsunami protection. This was confirmed by ground observation (UNEP 2005b).
Further research should be made with more detailed data as the coarse resolution of the
bathymetry and coral location prevent more precise conclusions. A mathematical model-
ling of wave, or in situ measure could be a good solution to study the behaviour of tsunami
waves in area covered by coral.
The statistical model show lower impacts in area behind seagrass beds. This could
be for two reasons, the mechanical role of seagrass beds acting as a damping filter that
help reducing the energy of the wave or because seagrass beds are located in areas
where the configuration of bathymetry is not favourable for building high wave. The
true reason remains unexplained, but the correlation with lower distance of impact is
significant.
It is important to note that all this analysis was made on a single event, the tsunami of 26
December 2004. Different magnitude and origin of a tsunami could result in drastically
different wavelengths thereby might induce different effects.
The identification of area most exposed to tsunami following our method can be used as
a first cutoff for choosing where more detailed data should be collected, modelling the
vulnerability of coast being the next logical step.
Acknowledgements The present publication is a part of a scientific Report for the UNEP Asian Tsu-
nami Disaster Task Force.This research would not have been possible without the close collaboration of
Alain Retiere and Olivier Senegas at UNOSAT, who provided free access to their collection of satellite
images, Phillip Fox and Kaveh Zahedi at WCMC who supplied the data on coral, mangroves forests and
seagrass beds, Christian Depraetere who provided the detailed coast lines for islands. We would like to
thank Jean-Michel Jaquet for his precious support as well as for reviewing this article, and Arthur Dahl
for his advices.
Appendix
Table A1 List of variables computed or extracted
Abbreviation Description Units
AV10 KM Average slope until 10 km Degrees
AV1 KM Average slope until 1 km Degrees
AV2_5 KM Average slope until 2.5 km Degrees
AV20 KM Average slope until 20 km Degrees
AV25 KM Average slope until 25 km Degrees
AV30 KM Average slope until 30 km Degrees
AV50 KM Average slope until 50 km Degrees
AV5 KM Average slope until 5 km Degrees
%age
CORAL Percentage of protection from coral preceding the site
COSORIEN Cosinus of orientation Scalar
DFEQ Distance from main earthquake Kilometres
123
Nat Hazards (2007) 40:289–304 303
Table A1 Continued
Abbreviation Description Units
DFF Distance from subduction fault line Kilometres
DFS Distance from source Kilometres
D Width of flooded land strip Metres
LDCOV Land cover resistance index Cardinal values 1 to 6
LDTO10M Average slope until an inland height of 10 m Degree
LDTO30M Average slope until an inland height of 30 m Degree
LENGDIST Length of distal slope Kilometres
LENGPROX Length of proximal slope Metres
%age
MANG Percentage of protection from mangroves
preceding the site
ORIENT Orientation between the tsunami energy Degrees
and a perpendicular to the coast
%age
PCAV10KM Average slope until 10 km
%age
PCAV1KM Average slope until 1 km
%age
PCAV2_5 K Average slope until 2.5 km
%age
PCAV20KM Average slope until 20 km
%age
PCAV25KM Average slope until 25 km
%age
PCAV30KM Average slope until 30 km
%age
PCAV40KM Average slope until 40 km
%age
PCAV500M Average slope until 5 km
%age
PCAV50KM Average slope until 50 km
%age
PCAV5KM Average slope until 5 km
%age
PCLDTO10 Average slope until an inland height of 10 m
%age
PCLDTO30 Average slope until an inland height of 30 m
%age
PCSLDIST Angle of Distal slope
%age
PCSLPROX Angle of Proximal slope
%age
SEAG Percentage of protection from Seagrass
beds preceding the site
SLDIST Angle of Distal slope Degree
SLPROX Angle of Proximal slope Degree
References
Articles and books
Dao H, Peduzzi P (2004) Global evaluation of human risk and vulnerability to natural hazards, In: Proc.
EnviroInfo 2004 conference, vol 1, Editions du Tricorne, Geneva, Switzerland, pp 435–446
Davis JH Jr (1940) The ecology and geological role of mangroves. In: Florida Carnegie Inst. Wash., vol 32,
Papers of the Tortugas Laboratory, pp 303–412
Field C (1996) Restoration of mangrove Ecosystems. International Tropical Timber Organization and
International Society for Mangrove Ecosystems, Okinawa, Japan
Harada K, Latief H, Imamura F (2002) Effect on reducing tsunami by the green belt and coastal permeable
structure, In: Proc. International Workshop on Tsunami Risk and Its reduction in the Asia-Pacific
Region, Bandung, Indonesia, Session 4
Hiraishi T and Harada K (2003) Greenbelt tsunami prevention in South-Pacific Region. Report Port Airport
Res Inst 42(2):1–23
IFRC (2002) Mangrove planting saves lives and money in Vietnam, World Disaster Report 2002, Inter-
national Federation of Red Cross and Red Crescent Societies (IFRC)
Jimenez JA, Martinez R and Encarnocion L (1985) Massive tree mortality in Puerto Rican mangroves
forests. Caribbian J Sci 21:75–78
Kairo JG, Dahdouh-Guebas F, Bosire J and Koedam N (2001) Restoration and management of mangrove
systems – a lesson for and from the East African region. South African J Bot 67:383–389
123
304 Nat Hazards (2007) 40:289–304
Kowalik Z (2003) Basic relations between tsunamis calculation and their physics – II. Sci Tsunami Hazards
21(3):152–173
Lewis RR (1982) Mangrove forests. In: Lewis RR (ed) Creation and restoration of coastal plant commu-
nities. CRC Press, pp153–173
Mofjeld HO, Titov VV, Gonzalez FI and Newman JC (2000) Analytic theory of tsunami wave scattering in
the open ocean with application to the North Pacific. NOAA Tech Memo OAR PMEL-116:1–38
Peduzzi P, Dao H and Herold C (2002) Global Risk And Vulnerability Index Trends per Year (GRAVITY),
Phase II: Development, analysis and results, scientific report. UNEP, Geneva Switzerland
Thom BG (1967) Mangrove ecology and deltaic geomorphology, Tabasco, Mexico. J Ecol 55:301–334
UNEP (2005a) After the tsumani – rapid environmental assessment report. UNEP, Nairobi, pp 1–140
UNEP (2005b) Maldives post-tsumami environmental assessment. UNEP, Nairobi, Kenya, p 19
References on the Internet
CRED (2005), EM-Dat, disasters list, http://www.em-dat.net
Dao H (2004), Classificator, Dpt of Geography, Geneva University.
http://geomap.unige.ch/sescat/stats/classify_func_example.html
DIPE (2002) Mangrove management in the Northern Territory, p. 5. Department of infrastructure planning
and environment, Northern Territory Government, Darwin NT, Australia.
http://www.lpe.nt.gov.au/dlpe/advis/land/mangrove/mangrovemgt/
DLR-ZKI (2005) ZKI satellite mapping supports international humanitarian relief teams in Southeast Asia,
http://www.zki.caf.dlr.de/applications/2004/indian_ocean/indian_ocean_2004_en.html
DRS-DPRI (2005) The December 26, 2004 earthquake tsunami disaster of Indian Ocean, Kyoto University,
http://www-drs.dpri.kyoto-u.ac.jp/sumatra/index-e.html
Fox R (2004) Water in motion, Department of Biology, Lander University,
http://www.lander.edu/rsfox/415hydrodyn-1Lec.html
Friends of the Earth (2005) Coral reefs and mangroves may have helped saved lives
http://www.foei.org/tsunami/malaysia.html
Khor M (2005), Save the Mangroves to fight tsunamis, Third World Network,
http://www.twnside.org.sg/title2/gtrends39.htm
Nelson SA (2005) What is a Tsunami?, http://www.tulane.edu/~sanelson/geol204/tsunami.htm
NOAA (2004a) International tsunami information center, Tsunami Glossary,
http://www.prh.noaa.gov/itic/library/pubs/glossary/tsu_glossary_html/tsunami_glossary11.html
NOAA (2004b) NOAA Backgrounder, NOAA and Tsunami,
http://www.prh.noaa.gov/itic/tsunami_events/media/factsheets/backgrounder.pdf
123