A Multi-Scenario Analysis of the Storm Surge Hazard for Sri Lanka

Wijetunge J. J., A multi-scenario analysis of the storm surge hazard for Sri Lanka, in Proc. 35th IAHR World Congress, Chengdu, China, 2013. Proceedings of 2013 IAHR World Congress A Multi-Scenario Analysis of the Storm Surge Hazard for Sri Lanka Janaka J. Wijetunge Senior Lecturer, Dept. of Civil Engineering, University of Peradeniya, Peradeniya 20400, Sri Lanka. Email: janakaw@pdn.ac.lk ABSTRACT: Sri Lanka is vulnerable to storm surges due to cyclones generated mostly in the southern part of Bay of Bengal, and to a lesser extent, to those in the southeast of Arabian Sea. Accordingly, a statistical analysis of the historical events of tropical cyclones in the Bay of Bengal and in the Arabian Sea was carried out to identify cyclone scenarios with appropriate recurrence intervals representing short-, medium-, and long-term timescales. The time- and space-varying wind field was estimated by using the parametric cyclone model of Holland. Furtherr, a numerical model based on depth-averaged shallow water equations has been employed to simulate storm surges corresponding to each scenario. The model set-up was calibrated and verified by comparing the computed surge heights with those observed corresponding to the severe cyclones of 1964 and 1978 that made landfall in the east coast of Sri Lanka. As there is considerable uncertainty on the probable cyclone tracks including the landfall location, the storm surge simulations for each scenario was carried out for a large set of synthetic tracks that are in statistical agreement with the historical database. The computed maximum surge heights at every grid point corresponding to each of the above hazard scenarios was collated to form composite maps of peak surges over the entire model domain. The spatial distribution of the computed surge heights indicates that the level of the storm surge hazard is the highest for the northern province of Sri Lanka and the lowest for the southern province. KEY WORDS: Tropical cyclones, Hazard assessment, Surge height, Storm tide, Numerical modeling. 1 INTRODUCTION Associated with extreme winds, heavy rainfall and storm surge, land-falling tropical cyclones that are formed in the North Indian Ocean (NIO) region, often cause immense death and destruction in vulnerable coasts. Usually, much of the death toll and damage to property in coastal areas is as a result of cyclone-induced storm surge causing inundation of low-lying onshore lands (Dube, 2003; Lin et al., 2010). The height of the storm surge depends on cyclone dynamics such as the wind speed, the translation speed, the angle of attack at landfall, the pressure drop and also on coastal and shelf morphological factors such as the bathymetry and the shape of the coastline (Jelesnianski, et al. 1992). Further, the severity and the extent of onshore inundation depend primarily upon the surge height and the prevailing tide as well as the elevation, the slope and the surface roughness of the terrain. Sri Lanka, an island nation located off the southern tip of India, is vulnerable to cyclones generated mostly in southern part of Bay of Bengal, and to a lesser extent, those in southeast of Arabian Sea (Figure 1). The cyclones which form over the southernmost part of Bay of Bengal at low latitudes, mainly move west or west-northwestwards into the Gulf of Mannar across the coast of Sri Lanka. These cyclones generally form during the later part of the post-monsoon season or early part of the winter and pre-monsoon season. On the other hand, a very few cyclones form over Arabian Sea and Laccadives

Sea which either make landfall at western coast of Sri Lanka or they move in a westerly direction over the Arabian Sea (SAARC, 1998). However, owing to atmospheric dynamics associated with cyclones and the relative proximity of Sri Lanka to the equator, a large proportion of cyclones generated in Bay of Bengal and Arabian Sea, fortunately, do not make landfall in Sri Lanka. Yet, sixteen cyclonic or severe cyclonic storms have made landfall in Sri Lanka during the last century according to the Department of Meteorology, Sri Lanka (DMSL, 2012); some of these, the severe cyclonic storms in particular, have resulted in loss of lives of the order of several hundred as well as considerable damage to housing and other infrastructure due to both the surge and the high winds. In comparison, the death toll in Sri Lanka due to the tsunami inundation in 2004 was of the order of several tens of thousands, although an event of such magnitude is expected only once in several centuries. Consequently, tropical cyclone induced storm surges appear to pose a more frequent, albeit comparatively less severe, hazard to most parts of the coastline of Sri Lanka than the tsunami hazard. In the present analysis, nevertheless, we show that the severity of the storm surge hazard could be greater even compared to the tsunami hazard for certain parts of the coastline of Sri Lanka. However, unfortunately, no detailed analysis and assessment of the storm surge hazard has been carried out for the coastline of Sri Lanka. Clearly, information gathered from such an assessment would provide the basis for disaster risk mitigation policy planning and decision making in regard to the cyclonic storm surge hazard for Sri Lanka. Accordingly, this paper describes a deterministic cyclone-induced storm surge hazard assessment carried out for the entire coastline of Sri Lanka. The hazard assessment, which is based on a statistical analysis of the past cyclone events in Bay of Bengal and Arabian Sea during the last century, utilizes a numerical model based on depth-averaged, non-linear shallow-water equations to simulate the storm surges due to several probable cyclone scenarios. India Sri Lanka Figure 1 Tracks of cyclones in North Indian Ocean region during 1970-2005. 2 A BRIEF REVIEW OF PAST CYCLONE EVENTS A brief review of notable past cyclones that made landfall in Sri Lanka during the past 130 years is given in the following. According to the classification of revolving tropical systems adopted in Sri Lanka (Table 1), maximum sustained wind speeds of 62-88 km/h and 89-118 km/h are termed cyclonic storms and severe cyclonic storms, respectively (DMSL, 2012). Table 1 Classification of tropical cyclonic systems in Sri Lanka (DMSL, 2012). Weather system Maximum wind speed (km/h) Low pressure area < 31 Depression 31-51 2

Deep depression 52-61 Cyclonic storm 62-88 Severe cyclonic storm 89-118 Very severe cyclonic storm 119-221 Super cyclonic storm 222 Sixteen cyclones, of which 5 are severe cyclonic storms, are known to have crossed Sri Lanka during the past 130 years (Table 2). Table 2 also gives the general area of formation of these cyclones as well as the coastal region of Sri Lanka where each cyclone made landfall. We see that, of the 16 cyclones that crossed Sri Lanka during the past 130 years, only two have been formed in Arabian Sea and made landfall in either western or northwestern coastlines of Sri Lanka; the rest have all been formed in Bay of Bengal and made landfall on the north and east coasts. Table 2 Past cyclones that made landfall in Sri Lanka during 1881-2011 (DMSL, 2012). Year/Month Classification Formed in Landfall 1906 Jan Cyclonic Storm Bay of Bengal North 1907 Mar Severe Cyclonic Storm Bay of Bengal East 1908 Dec Cyclonic Storm Bay of Bengal North 1912 Dec Cyclonic Storm Bay of Bengal South 1913 Dec Cyclonic Storm Bay of Bengal South East 1919 Dec Cyclonic Storm Bay of Bengal North 1922 Nov Severe Cyclonic Storm Bay of Bengal East 1925 Mar Cyclonic Storm Arabian Sea North West 1931 Dec Severe Cyclonic Storm Bay of Bengal North 1964 Dec Severe Cyclonic Storm Bay of Bengal East 1966 Nov Cyclonic Storm Bay of Bengal East 1967 Dec Cyclonic Storm Arabian Sea West 1978 Nov Severe Cyclonic Storm Bay of Bengal East 1980 Dec Cyclonic Storm Bay of Bengal East 1992 Dec Severe Cyclonic Storm Bay of Bengal South East 2000 Dec Severe Cyclonic Storm Bay of Bengal East Of the 5 severe cyclonic storms (Table 2), the systems that developed in December 1964 and November 1978 appear to have caused the most notable death and destruction; only little information is available on the impact in terms of number of casualties and damage to property due to the severe cyclonic storms that crossed Sri Lanka in 1907, 1922 and 1931. The severe cyclonic storm that made landfall near Trincomalee on 23 December 1964 with reported maximum wind speeds of 216 km/h (Chittibabu et al., 2002, about 220-240 km/h according to Kulshrestha, 1966 and Dinamina newspaper of 25 December 1964) resulted in a death toll of 842 and 100 missing according to Longshore (1998) (about 650 deaths and 400 missing according to Dinamina newspaper of 25 December 1964) in the northern province including the city of Mannar in the northwest where a storm surge of height up to about 4.5 m (15 feet, Ceylon Daily Mirror issued on 29 December 1964) inundated and destroyed low-lying coastal settlements. A detailed account of the impact of the storm surge is given in Ceylon Daily Mirror newspaper issued on 29 December 1964: A fantastic tidal wave, rising fifteen feet over the land, whipped over the island of Mannar at the very height of the 3

awesome cyclone. Today, six days after the event, only a sheet of water, subsiding ever so slowly, marks the place where once a city stood. Moreover, the 1964 cyclone is quite remarkable because it not only formed at a low latitude of 5 ⁰ N but also attained the intensity of a severe cyclonic storm at about the same latitude (Kulshrestha, 1966). The severe cyclonic storm that made landfall near Batticaloa on 24 November 1978 with satellite estimated maximum wind speeds of 222 km/h (DMSL, 2012) caused loss of lives as well as extensive damage to housing and buildings and generated a storm surge of height 1-2 m near Batticaloa with inland penetration of inundation reaching 1.5 km in Kalkudah (DMSL, 2012). The death toll attributed to this cyclone varies from 150 (Longshore, 1998) to 323 (Murty et al., 1986) and 915 (DMSL, 2012); moreover, 185,000 housing units were severely damaged with another 150,000 partially damaged (DMSL, 2012). The death toll due to the severe cyclonic storm that made landfall on 12 November 1992 was only 4, with over 29,000 housing units damaged. The severe cyclonic storm that crossed Sri Lanka on 26 December 2000 also resulted in 8 deaths and considerable property damage. (DMSL, 2012) It must be added that the casualty and damage figures given above are usually due to the combined influence of one or more of the following cyclone related effects: storm surge, high winds, and heavy rainfall. 3 STUDY METHODOLOGY 3.1 Statistical Analysis A database of historical tropical cyclone events was compiled for the North Indian Ocean (NIO) region for the period 1900-todate using best-track data from several sources including Joint Typhoon Warning Centre (JTWC) of the US Navy and SAARC Meteorological Research Center (SMRC). An observation window or a scan-box bounded by 4-11 o N & 78-93 o E covering probable cyclone generation and feeder regions in southern portions of both Bay of Bengal and Arabian Sea was demarcated and all cyclones that had either formed or crossed the scan box during the above period were considered to have the potential to make landfall in or in the vicinity of Sri Lanka provided that necessary atmospheric forcing satisfied the requirements for the same. Of the subset of 201 independent cyclone events found to be falling within the scan-box mentioned above, the portion of the data prior to satellite observations (i.e., 1945), as well as those events for which it was not possible to assign reliable maximum wind speeds were excluded. Accordingly, the peak annual wind speeds corresponding to the remaining 59 independent cyclonic events were then statistically analysed using Gumbel s (1958) method, following Rupp and Lander (1996) for tropical cyclones in Guam, and several others. The fact that, of the 201 cyclonic events in the database since 1900, only about 8% have made landfall in Sri Lanka was also incorporated into the probabilistic analysis by employing the multiplication rule. Figure 2 shows the resulting plot of wind speed against the reduced variate: the intercept and the slope of the linear regression line give the mode (u = 32.26) and slope (a = 11.68) of the fitted Type-I extreme value distribution. The recurrence interval for different wind speeds could thus be deduced, and accordingly, the following scenarios were selected for the storm surge hazard assessment: Wind speed of 270 km/h with an estimated recurrence interval of 300 years (Scenario-1); a wind speed of 215 km/h with an estimated recurrence interval of 90 years (Scenario-2); a wind speed of 160 km/h with an estimated recurrence interval of 30 years (Scenario-3); and a wind speed of 110 km/h with an estimated recurrence interval of 10 years (Scenario-4). Scenario-1 may be termed as the worst-case scenario whilst scenario-2 a long-term event, scenario-3 a medium-term event, and scenario-4 a short-term event. 4

Figure 2 Analysis of annual maxima of cyclonic wind speeds using the Gumbel method. 3.2 Model Set-up and Formulation The computational domain for the present study was selected based on consideration of past studies of storm surges in the NIO region covering Sri Lanka, for example, those of Henry et al 20. and Chittibabu et al. 8. Accordingly, a rectangular region extending from 77 ⁰ E - 85 ⁰ E and 4 ⁰ N - 12 ⁰ N was selected. The bathymetry for the computational grid of 2 km spatial resolution was at first interpolated from GEBCO data and was then updated with data from navigation charts. These navigation charts typically covered depths down to about 3000 4000 m at scales 1:150,000 or 1:300,000. Supplementary high resolution bathymetric data at scales of 1:10,000 and 1:15,000 were also used to further refine several nearshore localities off port cities such as Galle, Colombo and Trincomalee. The depths in navigation charts were reduced from Chart Datum (i.e., Lowest Astronomical Tide) to Mean Sea Level (MSL). A hydrodynamic model based on the quadratic wind friction formulation and depth-averaged, non-linear equations of conservation of mass and momentum was employed to compute the water surface elevation due to cyclone induced forcing of space- and time-varying wind and pressure fields. The wind and pressure distributions due to the cyclone was computed using an axisymmetric parametric model, i.e., Holland s (1980). 3.2 Sensitivity analysis An analysis was performed at the outset to examine the sensitivity of parameters such as the grid spacing and the time step on the computed surge elevations. Accordingly, a series of hydrodynamic simulations were carried out for a range of grid-sizes: x = 300 m, 400 m, 800 m, 1200 m and 2000 m, and then for a range of time steps for the selected grid-spacing. The computed maximum sea surface elevations corresponding to each grid-size and time step were extracted along the east coast of Sri Lanka to examine their sensitivity to computed surge elevations. 3.3 Model calibration and verification Two past cyclone events that resulted in storm surges in some parts of the coastline of Sri Lanka have been utilized to calibrate and verify the numerical model, namely, the severe cyclonic storms of 1978 and 1964, respectively. The model verification run with 1964 cyclone as the forcing was carried out with the same values of model parameters such as wind friction factor and Manning s coefficient as in the simulation for 1978 cyclone. The computed maximum surge heights were then compared with available records of observed surge heights due to the cyclones of 1978 and 1964. 3.4 Simulation of hazard scenarios Four cyclone scenarios together with respective recurrence intervals have been derived for storm surge hazard assessment based on the statistical analysis carried out in Section 3.1. Scenario-1 may be 5

termed as the worst-case scenario whilst scenario-2 a long-term event, scenario-3 a medium-term event, and scenario-4 a short-term event. As there is considerable uncertainty on the probable cyclone tracks including the landfall location, the storm surge simulations for each scenario was carried out for a large set of synthetic tracks that are in statistical agreement with the historical database (e.g., as in Emanuel and Jagger, 2010 and Hallegatte 2007). Accordingly, for each scenario, the landfall location on the coastline was varied at 0.182 o (~20 km) intervals along the latitude and an array of separate model simulations was carried out for each hypothetical track. The models for cyclone scenarios-1 and -2 were integrated with a maximum pressure drop of 80 hpa and a radius of maximum wind of 40 km. The model for cyclone scenario-3 was integrated with a maximum pressure drop of 70 hpa and a radius of maximum wind of 35 km whilst these parameters for scenario-4 were 60 hpa and 35 km, respectively. The computed maximum surge heights at every grid point corresponding to each of the above hazard scenarios was collated to form composite maps of peak surges over the entire model domain. 4 RESULTS AND DISCUSSION 4.1 Sensitivity analysis The correlation among the computed surge elevations with different grid sizes, i.e., x = 300 m, 400 m, 800 m, 1200 m and 2000 m are given in Table x. It can be seen in Table 5.1 that the computed surge elevations with even the maximum grid size, i.e., x = 2000 m shows good correlation (r 2 = 0.9685) with the minimum grid-size of x = 300 m. Accordingly, a grid spacing of 2000 m was selected in order to minimize computational time without compromising accuracy and also so as to satisfy the numerical stability criterion. It is also required to minimize the output file size, and accordingly, a time step of 60 s was selected for the present study. Table 3 Correlation coefficients for computed surge elevationswith different grid sizes. Grid spacing ( x) Correlation coefficient (r 2 ) 300 m and 400 m 0.9914 300 m and 800 m 0.9817 300 m and 1200 m 0.9671 300 m and 2000 m 0.9685 4.2 Model calibration and verification Table 4 compares the maximum values of the simulated surge levels and the corresponding observed maximum storm tide levels at several locations in Sri Lanka and in South India. The locations of observed storm tides have been identified only by city/village name (general area), and therefore, in the absence of exact coordinates of these locations of observed surge levels, we give a range of simulated maximum surge heights in the vicinity of the general area of each location; the source of information regarding observed maximum storm tide levels is also given. We see in Table 4 that the range of computed storm surge levels are, on the whole, in reasonable agreement with the observed storm tide levels. It must be noted that the simulated surge levels do not include the effects of the tide whereas the observed surge levels include the storm surge and the effects of the tide at the time. Clearly, the highest surge levels in Sri Lanka with regard to 1964 cyclone have occurred near Mannar; the shallow bathymetry as well as the geometry of the coastline in the confined Palk Straits have mainly contributed to this significant amplification of the surge. Table 4 Comparison of simulated and observed maximum surge heights. Cyclone Location Observed storm tide level Simulated maximum storm surge level 6

1964-Cyclone 1978-Cyclone Rameswaram and Madanpan Pamban and Nagapattinam coastal strech Tondi 3.0 4.2 m (Ramesh, 2004) 3.0 5.0 m (Ramesh, 2004) 3.0 6.0 m (Murty, 2005) Dhanushkodi 3.0 6.0 (SMRC, 1998) Mannar 4.8-5.2 m (CDM, 1964) Batticaloa 1.0 2.0 m Chittibabu (2002) Coastal area between Tondi and Devipattinam 3.0 5.0 m Ramesh (2004) 3.0 4.0 m 3.0 3.7 m 3.2 5.8 m 2.8 3.4 m 4.6 m 0.8 1.6 m 2.4 3.3 m 4.3 Computed peak surge heights corresponding to hazard scenarios Figure 3 shows the composite maps of peak surge heights corresponding to four cyclone scenarios: (a) scenario-1 (max. wind speed = 270 km/h), (b) scenario-2 (215 km/h), (c) scenario-3 (160 km/h), and (d) scenario-4 (110 km/h). Note that the surge levels shown in Figure 3 are exclusive of tidal effects. For a conservative, worst-case estimate of storm tide, an additional 0.3 m may be added to the surge levels shown in Figure 3. Clearly, the northwestern and northern coastal areas of Sri Lanka are likely to experience more severe flooding caused by storm surges than the south and the west. The shallow bathymetry and the wider continental shelf fronting the north and northwest are primarily responsible for the higher level of surge heights. On the other hand, the southwestern coastal areas are exposed to a lower level of storm surge hazard. Note that, although there is a higher chance of cyclones making landfall on the eastern and northeastern coasts than the west, cyclones induce surges not only at landfall but whilst exiting a coastline as well. Several severe cyclones have hit Sri Lanka during the past century, with those in 1964 and 1978 being the worst, resulting in loss of lives of the order of several hundred as well as considerable damage to housing and other infrastructure due to both the surge and the high winds. In comparison, the death toll in Sri Lanka due to the tsunami inundation in December 2004 was of the order of several tens of thousands, however, a tsunami event of such magnitude is extremely rare, expected may be only once every several centuries. So, tropical cyclone induced storm surges appear to pose a more frequent, albeit comparatively less severe, threat of flooding in most parts of the coastline of Sri Lanka than tsunami. Nevertheless, it must be noted that the severity of the storm surge hazard could be greater even compared to the tsunami for certain parts of the coastline of Sri Lanka. For example, the city of Mannar and the nearby localities are probably more at risk of coastal flooding due to storm surges than tsunami. So, whilst improving our tsunami preparedness in coastal areas, we must also pay due attention to the potential threat from other hazards such as coastal flooding due to cyclone induced storm surges as well. The present assessment of the potential impact of cyclone induced storm surges for the entire seaboard of Sri Lanka will be useful in the formulation of disaster risk management policies, development of preparedness plans, allocation of resources for disaster risk reduction, and in education and awareness activities. 7

(a) (b) (c) (d) Figure 3 Composite maps of peak surge heights corresponding to four cyclone scenarios: (a) scenario-1, (b) scenario-2, (c) scenario-3, and (d) scenario-4. 4.4 Limitations One limitation is that the resolution of the modeling is no greater or more accurate than the bathymetric data used. Moreover, the tide has been linearly superimposed on the computed storm surge levels on a conservative basis although the tide-surge interaction is non-linear. It must also be added that the set-up due to wave breaking has not been incorporated in the present model simulations. 8

It must also be added that depth-averaged models assume a uniform velocity profile across the flow depth and neglect vertical accelerations. Moreover, the mathematical formulation employed in the present model does not explicitly account for all means of energy dissipation. For instance, although energy dissipation due to bottom friction is included in the present model, dissipation due to turbulence is not explicitly formulated. 5 CONCLUSIONS The spatial distribution of the computed surge heights indicates that the level of the storm surge hazard is the highest for the northern province of Sri Lanka and the lowest for the southern province. The city of Mannar in the northwestern coast of Sri Lanka appears to be the most vulnerable to the storm surge hazard. The shallow bathymetry and the wider continental shelf fronting the northern and northwestern coastline are primarily responsible for the higher level of surge heights. ACKNOWLEDGEMENT The author wishes to acknowledge the support received from the National Science Foundation (Grant No. RG/2011/ESA/01) and The Ministry of Disaster Management of the Government of Sri Lanka. He also wishes to thank Research Assistants Mr M.G..S. C. Preemal and Ms C. K. Marasinghe for their assistance in carrying out some of the numerical model simulations. References CDM, Ceylon Daily Mirror (1964). Ceylon Daily Mirror Newspaper of 29 December, Times of Ceylon, Colombo. Chittibabu, P., Dube, S. K., Sinha, P. C., Rao, A. D. and Murty, T. S., 2002. Numerical Simulation of Extreme Sea Levels for the Tamil Nadu (India) and Sri Lankan Coasts, Marine Geodesy, 25(3): 235-244. DMSL, Department of Meteorology, Sri Lanka, 2012. Tropical cyclones; Cyclone events 1900-2000, Department of Meteorology, Government of Sri Lanka, http://www.meteo.gov.lk/, accessed on 15-2-2012. Dube, S.K., 2003. Storm surge forecasting in the Bay of Bengal and Arabian Sea. In Antarctic Geoscience, Ocean-Atmosphere Interaction and Paleoclimatology (Eds. S. Rajan and P.C. Pandey), National Centre for Antarctic & Ocean Research, Goa, India. Emanuel, K. and Jagger, T., 2010. On estimating hurricane return periods. J. Appl. Meteorology and Climatalogy, 49: 837-844. Gumbel, E.J., 1958. Statistics of Extremes. Columbia University Press, New York. Holland, G. J., 1980. An analytic model of the wind and pressure profiles in hurricanes. Monthly Weather Review, 108, 1212 1218. Hallegatte, S., 2007. The use of synthetic hurricane tracks in risk analysis and climate change damage assessment. Center for Environmental Sciences and Policy, Stanford University, USA. Henry R.F. and T.S. Murty, 1992. Storm surges and tides around Sri Lanka, Recent advance in Marine science and technology, 92:205-218. Jelesnianski, C.P., Chen, J., and Shaffer, W.A., 1992. SLOSH: Sea, Land and Overland Surges from Hurricanes, NOAA Technical Report, NWS 48, Silver Springs, Maryland. Kulshrestha, S.M. and Gupta, M.G., 1966. Satellite study of the Rameswaram cyclonic storm of 20-23 December 1964. Journal Appl. Meteorology, 5(3): 373-376. Lin, N., K. A. Emanuel, J. A. Smith, and E. Vanmarcke, 2010. Risk assessment of hurricane storm surge for New York City. Journal Geophys. Res., 115, D18121. Longshore, D., 2008. Encyclopedia of Hurricanes, Typhoons, and Cyclones, Facts on File, New York. Murty, T.S., Flather, R.A., and Henry, R.F., 1986. The storm surge problem in the Bay of Bengal. Prog. Oceanog., 16: 195-233. Ramesh, R., 2004. Sethusamudram Shipping Canal Project and the unconsidered high risk factors: Can it withstand them? Doctors for Safer Environment, Coimbatore, Tamil Nadu. Rupp, J. A. and Lander, M. A.,1996. A technique for estimating recurrence interval of tropical cyclone-related high winds in the tropics: Results from Guam. J. Ame. Met. Soc., May 1996: 627-637. SAARC, 1998. The impact of tropical cyclones on the coastal regions of SAARC countries and their influence in the region. Publication No. 1, SAARC Meteorological Research Centre (SMRC), Bangladesh. 9