You are on page 1of 5

Proceedings of International Sessions in Coastal Engineering, JSCE, Vol.

4, 2012

ESTIMATION OF SEA LEVEL RISE DUE TO


STORM SURGE AND GLOBAL WARMING IN 2050
ALONG THE BANGLADESH COAST USING
EMPIRICAL MODE DECOMPOSITION AND
EXTREME VALUE ANALYSIS
Han Soo Lee1, Hendri2, ABM Sertajur Rahman3, Md. Irfanul Islam4 and Takao
Yamashita5
1Graduate School for International Development and Cooperation, Hiroshima University, hslee@hiroshima-u.ac.jp
2State University of Papua, hendri888@gmail.com
3Roads and Highway Department Bangladesh, sertajrahman@gmail.com
4Bangladesh Railway, irfanul@msn.com
5 Graduate School for International Development and Cooperation, Hiroshima University,
tkoyamashita@hiroshima-u.ac.jp

Extreme sea levels due to storm surge and future sea-level rise (SLR) in the year 2050 are estimated
using ensemble empirical mode decomposition (EEMD) and extreme value analysis (EVA) based on
long-term sea level records from Hiron Point (HP) on the coast of western Bangladesh. EEMD is an
adaptive method that can detrend the non-linear trend and separate the tidal motions from the original sea
level records to reconstruct storm surge levels at HP. The reconstructed storm surge levels are then
applied to EVA to obtain the extreme storm surges in the target return periods at a 95% confidence
interval (CI). The 30-, 50-, and 100-yr return levels at HP obtained by EVA are 1.59 m, 1.66 m and 1.75
m. The SLR trend obtained from EEMD is 4.46 mm/yr over April 1990-March 2009, which is larger than
the recent altimetry-based global rate of 3.3 0.4 mm/yr over the period from 1993-2007. The resulting
SLR in 2050 is estimated as 0.34 m. Therefore, the extreme sea level in 2050 due to SLR and the storm
surge at a 100-yr return level would be 2.09 m (95% CI from 1.91 m - 2.48 m). The SLR depends not
only on changes in the mass and volume of sea water but also on other factors, such as local subsidence,
river discharge, sediment and the effects of vegetation. The residual non-linear trend of SLR obtained
from EEMD can be regarded as an adaptive sea level after considering those factors and their
nonlinearity.
Key Words : Sea level rise, storm surge, extreme sea level, ensemble empirical mode decomposition, extreme value
analysis, Bangladesh

1. INTRODUCTION
Global mean sea levels have increased steadily
over the past century due to an increase in the global
mean atmospheric temperature (Cazenave and
Llovel, 2009; IPCC, 2007). Continued increases in
mean sea levels are predicted to have catastrophic
impacts on coastal environments around the world
in the near future. Coastal hazards due to flood
events are almost invariably associated with extreme
sea levels by tropical cyclone induced storm surges
(Butler et al., 2007). Therefore, the impacts of
global warming on coastal flood risk depend heavily

on the future trend in extreme storm surges. The


future projection of an extreme positive storm surge
is thus a crucial consideration in the study of climate
change impacts and adaptation strategies.
Bangladesh lies on mostly flat, alluvial land at
the mouth of the Ganges-Brahmaputra-Meghna
(GBM) Basins facing the Bay of Bengal to the
south, with approximately 710 km of coastline (Fig.
1). Bangladesh is well known to be one of the most
vulnerable countries to the future sea level rise
(SLR) due to global warming. While it is not
currently experiencing adverse impact, it is
projected to be the most vulnerable country to the

SLR in the future climate. Therefore, this study


estimates the future extreme sea level due to (1)
storm surge and (2) future SLR based on a
data-driven statistical modeling approach using
ensemble empirical mode decomposition (EEMD)
and extreme value analysis (EVA) based on
long-term sea level records from the coast of the
Bangladesh. By combining the EEMD and EVA
methods, we illustrate a possible way of projecting
extreme sea level (extreme positive storm surges +
SLR) to the mid-term future of 2050.

The original data from the Hiron Point (HP) are


referenced with respect to the Public Works Datum
(PWD) of Bangladesh, which is 0.46 m below the
mean sea level, and consists entirely of
observations, while the data from the other 10
stations are reconstructed based on simple tidal
phase shift and amplitude corrections of the HP
data. Therefore, further data analyses and results are
conducted and discussed separately for the first set
of hourly observations from April 1977 to March
1990 and the second set of 30-min observations
from April 1990 to March 2009 only at HP.

2. DATA, METHOD, AND PROCEDURE


(1) Data
Sea level records from 13 stations, mainly on the
western Bangladesh coast were obtained from the
Bangladesh Inland Water Transportation Authority
(BIWTA) and the Flood Forecasting and Warning
Center (FFCW), Bangladesh (Fig. 1). The data used
in this study comprise 32 years, from April 1977 to
March 2009. The observation intervals are 1 hr from
April 1977 to March 1990 and 30 min from April
1990 to March 2009. The total number of original
raw data points is 428,251, with 18,821 intermittent
missing data points (4.21% of the complete data).
The data are reconstructed by filling in the missing
data with supplementary predicted tides generated
using the MIKE21 TIDAL toolbox (DHI, 2007).
Figure 2 illustrates (a) the original raw data and (b)
the reconstructed source data with predicted tides in
red.

(2) Empirical Mode Decomposition


Contrary to almost all of the previous
decomposing methods, EMD is empirical, intuitive,
direct, and adaptive, without requiring any
predetermined basis functions, and thus is
applicable to nonlinear and non-stationary data. The
EMD method can decompose any complicated data
set into a finite number of Intrinsic Mode Functions
(IMFs). The EEMD is the improved method for
obtaining IMFs with more direct physical meaning
and greater uniqueness. The EEMD defines the true
IMF components as the mean of an ensemble of
trials, which dramatically improves the mode
mixing problem due to the subject selection of scale
for scale separation. For more details on the HHT
and its applications to geophysical research, please
refer to Huang and Shen (2005) and Huang and Wu
(2008), respectively.
(3) Statistical significance test
To determine whether a dataset or its
components contain useful information, statistical
significance tests were performed based on the
characteristics of Gaussian white noise with EMD in
the following way: first, decompose the targeted
noisy dataset (normalized) into IMFs; second,
calculate the spread function of the given percentiles
for the IMFs decomposed from Gaussian white
noise; third, select the confidence-limit level (e.g.,
95%) and determine the upper and lower spread
lines; and finally, compare the energy density for
the IMFs from the noisy data with spread functions.
The IMFs with energy located above the upper
bound or below the lower bound should be
considered as containing signal information at that
selected confidence level.

Fig. 1 Locations of sea level observations (black circles) on the


western and central Bangladesh coast, together with detailed
bathymetry and topography from GEBCO 30 arc-sec and
SRTM. Locations of the maximum observed storm surges due
to the 1991 Bangladesh Cyclone (black triangles) are also
presented along the eastern coast.

(3) Extreme value analysis


The distinguishing feature of the widely used
statistical methodology of EVA is the ability to
quantify the stochastic behavior of a process at
unusually large or small levels. In particular, EVA
usually requires the estimation of the probability of

events that are more extreme than any that have ever
been observed (Coles, 2001).
In this study, we use the block maxima to fit the
annual maxima of the reconstructed storm surges to
a general extreme value (GEV) distribution for
estimating the return level of storm surges along the
coast of Bangladesh in the year 2050 to assess the
extreme storm surge in the near future. The
distribution parameters were estimated by the
maximum likelihood method. The confidence
interval for the return level is obtained based on the
delta method, which assumes normality for the
return level. The Extremes Toolkit (extRemes)
(Gilleland and Katz, 2011), operated through the
open-source R software, is used to apply the
extreme value analysis method.
(3) Analysis procedure
Figure 3 depicts the data analysis and statistical
modelling procedure used to estimate the extreme
sea level using EEMD and EVA.
First, the original raw observed records are
supplemented with predicted tides to fill in the
missing data to form the reconstructed source data
as described in the previous section 2(1) (Fig. 2).
Second, the reconstructed source data are
separated into two parts based on the observation
interval, from April 1977 to March 1990 and from
April 1990 to March 2009. Then, each part is
individually subjected to EEMD with 50 ensembles.
The EEMD produces a number of IMFs from high
to low frequencies and the residual from the source
data. The statistical significance test is then
performed to ensure the significance of all IMFs.
The IMFs and significance test result are not shown
here due to limited space.
Third, after ensuring the significance levels of
IMFs, the residual is removed from the EEMD
results to detrend each part of the source data. Then,
the IMFs corresponding to astronomical tidal
components are eliminated from the EEMD results.
Therefore, the storm surge levels can be
reconstructed by combining the remaining non-tidal
IMFs (Fig. 4). The EEMD is used as a detrending
and filtering method in this process. In other words,
by applying the EEMD, we can detrend the source
data, and concurrently and automatically filter out
tidal components for the EVA of the storm surge in
the next process.
Fourth, we calculate the annual maxima from
each part of the reconstructed storm surge levels
obtained in the previous process and merge them
into one dataset. Then, these maxima are applied to
the EVA using block maxima and fit to the GEV
distribution to estimate the storm surge return levels
for the target years with confidence intervals.

Fig. 2 (a) Raw data (the original observed sea level), (b) the
reconstructed source data at Hiron Point (HP) generated by
filling in the missing gaps with predicted tides (red) at HP.

Fig. 3 Procedure for estimating the extreme sea level due to


storm surge and SLR using EEMD and EVA (IMF: intrinsic
mode function)

Fifth, to obtain future projections of sea level


rise, we use the above-mentioned residual from the
EEMD results as the non-linear trend of sea level
change. To project future SLR, we use the current
non-linear trend of the second part of the source
data from April 1990 to March 2009. We tried
various functions and selected a quadratic
Chebyshev approximation, which exactly fits the
non-linear trend of sea level change (r=1.0). Then,
the polynomial function is extended in a
straightforward manner to predict the sea level
change to 2050 (Fig. 5(a)). Therefore, the current
acceleration of the non-linear sea level rise is taken
into account in the mid-term future prediction.
Finally, by combining the estimated return level
and projected SLR for a target year, we can
quantitatively obtain the future SLR due to storm
surge and other factors along with confidence levels
(Fig. 5).

3. RESULTS
(1) Storm surge from EEMD
As described, the observed sea level recorded at

HP is divided into two parts, from April 1977 to


March 1990 and from April 1990 to March 2009.
Then, the EEMD is applied separately to the two
datasets. Based on the result of the significance test
of the IMFs, interestingly, all IMFs are statistically
significant at the 99% confidence level.
Within the resulting IMFs, the IMF2 and IMF3
correspond to the composites of semi-diurnal and
diurnal tides, respectively. Due to the adaptive
nature of the EEMD method, it is possible to
interpret that the semi-diurnal tides such as M2, S2,
N2, and K2 are embedded in the IMF2. Similarly to
the IMF2, the IMF3 can represent the mixed signal
of diurnal tides such as K1, O1, P1, and S1. IMF11
indicates the clear seasonal cycle due to the biennial
tendency of monsoons and their effects on sea level.
The IMFs 2, 3, and 11 exhibit the highest
normalized energy density in the signal based on the
significance test.

Fig. 4 The reconstructed storm surges (blue line) from the


non-tidal IMFs and the non-linear trend (red line) from the
residual at HP for (a) April 1977 - March 1990 and (b)April
1990 - March 2009.

Table 1 Previous (~ March 1990) and current (April 1990 ~)


SLR trend and the predicted SLR for 2050, together with 30-,
50- and 100-yr return levels of the storm surge at Hiron Point.
Station Average trend
SLR
Return levels (m)
(mm/yr)
(m)
April
April
April 30-yr 50-yr 100-yr
1977 1990 2009
March
March
1990
2009
2050
Hiron
8.04
4.46
0.34
1.59
1.66
1.75
Point

As the purpose of this study is to estimate the


extreme storm surge using the EVA in the next step,
we reconstruct the storm surge levels by combining
all non-tidal IMFs only after removing the IMF2
and IMF3 and the residuals, corresponding to
composites of semi-diurnal and diurnal tides and the
non-linear trend, respectively, as indicated above.
Fig. 4 represents the reconstructed storm surge
levels as blue lines together with non-linear trends
in red. Fig. 4(a) exhibits the reconstructed storm
surges and the non-linear trend for the first part of
the data, while Fig. 4(b) is obtained from the second
part of the data.
(2) Future storm surge levels from EVA
From the reconstructed storm surges in Fig. 4,
annual maxima are calculated for each year from
1977 to 2009. Then the annual maxima applied to
the EVA. The distribution parameters determined by
the maximum likelihood estimation are 0.920 (std.
error: 0.049) for location (), 0.258 (std. error:
0.033) for scale (), and -0.170 (std. error: 0.086)
for shape (). The 100-year return level is 1.75 m,
with a 95% confidence interval of 1.57 - 2.14 m.
The detailed return levels are presented in Table 1
and Fig. 5(b).
(3) Sea level rise projection
The EEMD method yields the residual value and
the IMFs. As described earlier, we interpret the
residue as the non-linear trend of sea level records.
The red lines in Fig. 4 indicate the non-linear SLR
trends from the two parts of the dataset. The annual
rates of sea level change are 8.04 mm/yr for the first
part and 4.46 mm/yr for the second part of the data.
The resulting SLR from April 2009 to December
2050 is 0.34 m, with an average annual rate of 7.79
mm/yr (See Table 1 and Fig. 5(a)).

4. DISCUSSIONS
Fig. 5 (a) The current non-linear trend and the SLR projection
by 2050 and (b) the storm surge return levels for 1- to 100-yr
return periods at HP. The extreme sea level prediction has to
consider those two results together.

As is widely known, SLR results from two


mechanisms. First is the steric effect, in which the
volume of sea water changes due to a change in

density in response to variations in temperature


(thermosteric) and salinity (halosteric). Second is
the eustatic effect, in which the mass of water
changes due to the exchange of water with the
over-lying atmosphere and continents, such as from
glaciers and ice sheets, through precipitation,
evaporation, river runoff, and ice melting (Miller
and Douglas, 2004).
In addition to the steric and eustatic effects,
other factors contribute to changes in sea level
records such as local subsidence, tectonic motion,
and river discharge and sediments. According to the
Ali (1996), approximately 10% of any sea level
change in Bangladesh is due to local subsidence.
The non-linear trend of SLR in Fig. 5 exhibits an
acceleration of sea level rise after 2000. The
acceleration of recent sea level change can be
partially explained by two factors. One is the local
subsidence of the coastal zone in Bangladesh as
described earlier, and the other is the Farakka
Barrage construction in the upstream portion of the
Ganges River by India in 1975, which reduced river
discharge and thus reduced sediment deposits (Fig.
1). The western coastal region of Bangladesh is
mainly covered by the worlds largest mangrove
ecosystem, Sundaban Mangroves, which is fed by
the Ganges River. After the construction of the
barrage, the increased salt water intrusion hinders
the growth of the mangrove forest, induces the
retrogression of forest types and reduces the
productivity of forest on the coast of the western
Bangladesh (Islam and Gnauck, 2008). Therefore,
the positive role of mangroves in stabilizing the land
and coastal zone has lessened, contributing to the
acceleration of SLR.

5. CONCLUSIONS
The EEMD method successfully filters out the
tidal variations in the raw sea level records to
reconstruct storm surge levels. The non-linear SLR
trend is identified and obtained from the residue
from EEMD. The annual mean rates of SLR at HP
obtained in this way are 8.04 mm/yr from April
1977 to March 1990 and 4.46 mm/yr from April
1990 to March 2009, larger than the altimetry-based
global average rate of SLR of 3.3 0.4 mm/yr from
1993-2007 (Cazenave and Llovel, 2009). This
analysis suggests an SLR from April 2009 to 2050
of 0.34 m. Then, the reconstructed storm surge
elevations undergo EVA to estimate the extreme
storm surges at HP. The 30-, 50-, and 100-yr return
levels are obtained as 1.59 m, 1.66 m and 1.75 m.
Therefore, the extreme sea level in 2050 due to
future SLR and storm surge with a 100-yr return

level will be 2.09 m with a 95% confidence interval


from 1.91 m - 2.48 m.
A number of studies have been conducted
locally and internationally to assess the impact of
and examine adaptation strategies to future SLR and
storm surges in Bangladesh, although these studies
have used rather simple SLR scenarios with high
levels of uncertainty. To achieve sufficient accuracy
to support adaption strategies to climate change,
however, the extreme sea level and subsequent
scenarios must be estimated based on physical
principles and must be statistically sound.
Moreover, small differences in the SLR scenarios
could induce considerable differences in the final
affected areas and properties. At the regional or
national scale, differences on the order tens of
centimeters would significantly affect local people,
governments and stakeholders. Therefore, we hope
that the novel method by the combination of EEMD
and EVA introduced here will help to improve sea
level assessment and adaptation processes to climate
change in Bangladesh.
ACKNOWLEDGMENT: The study is supported
by GELs Education Program, and a Grant-in-Aid for
Young Scientists (B) from MEXT.
REFERENCES
Ali, A. (1996), Vulnerability of Bangladesh to climate change
and sea level rise through tropical cyclones and storm surges,
Water, Air, & Soil Pollution, Vol. 92, No. 1, 171-179.
Butler, A., J. E. Heffernan, J. A. Tawn, R. A. Flather, and K. J.
Horsburgh (2007), Extreme value analysis of decadal
variations in storm surge elevations, J. Mar. Syst., Vol. 67,
No. 12, 189-200.
Cazenave, A., and W. Llovel (2009), Contemporary Sea Level
Rise, Annu. Rev. Mar. Sci., Vol. 2, No. 1, 145-173.
Coles, S. (2001), An introduction to statistical modeling of
extreme values, 208 pp., Springer, London.
DHI (2007), MIKE21 Tidal Analysis and Prediction Module
Scientific Documentation, 86 pp, DHI.
Gilleland, E., and R. W. Katz (2011), New software to analyze
how extremes change over time, EOS TRANS. AGU, Vol.
92, No. 2.
Huang, N. E., and S. S. P. Shen (2005), Hilbert-Huang
Transform and its Applications, World Sceintific.
Huang, N. E., and Z. Wu (2008), A review on Hilbert-Huang
transform: Method and its applications to geophysical
studies, Rev. Geophys., Vol. 46, No. 2, RG2006.
IPCC (2007), Climate Change 2007: The Physical Science
Basis. Contribution of Working Group I to the Fourth
Assessment Report of the IPCC, edited by S. Solomon, et al.,
Cambridge University Press Cambridge, U.K. and New
York, NY, USA.
Islam, S., and A. Gnauck (2008), Mangrove wetland ecosystems
in Ganges-Brahmaputra delta in Bangladesh, Front. Earth
Sci. China, Vol. 2, No. 4, 439-448.
Miller, L., and B. C. Douglas (2004), Mass and volume
contributions to twentieth-century global sea level rise,
Nature, Vol. 428, No. 6981, 406-409.
(Received June 19, 2013)

You might also like