Nat Hazards DOI 10.1007/s11069-012-0193-6

ORIGINAL PAPER

Simulation of water levels and extent of coastal inundation due to a cyclonic storm along the east coast of

A. D. Rao • P. L. N. Murty • Indu Jain • R. S. Kankara • S. K. Dube • T. S. Murty

Received: 28 September 2011 / Accepted: 6 April 2012 Ó Springer Science+Business Media B.V. 2012

Abstract The devastation due to storm surge flooding caused by extreme wind waves generated by the cyclones is a severe apprehension along the coastal regions of India. In order to coexist with nature’s destructive forces in any vulnerable coastal areas, numerical ocean models are considered today as an essential tool to predict the sea level rise and associated inland extent of flooding that could be generated by a cyclonic storm crossing any coastal stretch. For this purpose, the advanced 2D depth-integrated (ADCIRC-2DDI) circulation model based on finite-element formulation is configured for the simulation of surges and water levels along the east coast of India. The model is integrated using wind stress forcing, representative of 1989, 1996, and 2000 cyclones, which crossed different parts of the east coast of India. Using the long-term inventory of cyclone database, syn- thesized tracks are deduced for vulnerable coastal districts of Tamil Nadu. Return periods are also computed for the intensity and frequency of cyclones for each coastal district. Considering the importance of Kalpakkam region, extreme water levels are computed based on a 50-year return period data, for the generation of storm surges, induced water levels, and extent of inland inundation. Based on experimental evidence, it is advocated that this region could be inundated/affected by a storm with a threshold pressure drop of 66 hpa. Also it is noticed that the horizontal extent of inland inundation ranges between 1 and 1.5 km associated with the peak surge. Another severe cyclonic storm in Tamil Nadu (November 2000 cyclone), which made landfall approximately 20 km south of Cuddalore, has been chosen to simulate surges and water levels. Two severe cyclonic storms that hit Andhra coast during 1989 and 1996, which made landfall near and Kakinada, respectively, are also considered and computed run-up heights and associated water levels. The simulations exhibit a good agreement with available observations from the different

A. D. Rao (&) P. L. N. Murty I. Jain S. K. Dube Centre for Atmospheric Sciences, Indian Institute of Technology, New Delhi 110 016, India e-mail: [email protected]

R. S. Kankara ICMAM-Project Directorate, NIOT Campus, Pallikaranai, Chennai 600 100, India

T. S. Murty Department of Civil Engineering, University of Ottawa, Ottawa, ON, Canada 123 Nat Hazards sources on storm surges and associated inundation caused by these respective storms. It is believed that this study would help the coastal authorities to develop a short- and long-term disaster management, mitigation plan, and emergency response in the event of storm surge flooding.

Keywords ADCIRC model Storm surges Return period Inland inundation Water levels

1 Introduction

Natural disasters such as tropical cyclones heavily impact the Indian coasts. The destruction due to flooding caused by extreme wind waves generated by the cyclones is a serious concern along the coastal regions of India. The rise of sea level would be more if the cyclone crosses the coast during high-tide conditions. Therefore, it is important to know the maximum total water elevation that could possibly occur at a particular coastal location due to the combined effect of surge, tide, and wind waves. As the effect of storm surges is confined to few 100 km along the coast around the landfall location of the cyclone, the prediction will hopefully be improved when high- resolution location-specific models are used. Since modeling of storm surges in complex terrain requires true representation of coastal geometry as well as detailed onshore topography and bathymetry, finite-element models may be more useful for surge prediction and associated inland inundation. When used operationally, the numerical models are useful to provide early warning to low-lying areas and guide evacuation and rescue operations. In India, the models based on finite-difference methods have been developed and used extensively to simulate/predict storm surges for the last two decades. Johns et al. (1982) employed a finite-difference method, which involves a continuously moving lateral boundary. In this model, the coastline representation is based on a conformal mapping procedure that cannot be used in case of sharp curvatures in the boundary. A detailed review of the problem concerning storm surges in the Bay of Bengal is provided by Rao et al. (1994) and Dube et al. (1985). Rao et al. (2009) describes a comprehensive com- parison of surge simulations using finite-difference and finite-element models. The study concludes that resolving the atmospheric forcing of the cyclone is as important as pro- viding the detailed coastline geometry for surge development. Inland intrusion of water associated with storm surges is the main cause for extensive damage due to extreme events along the Indian coastline. Storm surges have inundated large stretches of coastal regions, sometimes penetrating up to 10–15 km inland, particularly, when the cyclone passes through a river deltaic region (SMRC 1998). To minimize the damage, prediction of inland inundation is as important as the prediction of surge heights. Generally, in storm surge prediction models, the rigid lateral boundaries are taken as vertical sidewalls, through which there is no flux of water. In a realistic sense, water level associated with the surge will continuously move onshore. Hence, the assumption of idealized vertical walls will lead to spurious error in simulations of the surge development. In addition, simplified hori- zontal inland extrapolation of the predicted surge height from the fixed sidewall that estimates the distance of inland penetration do not take into account the underlying dynamics of the intrusion processes. The importance of a numerical model comes to light when providing an early warning to a low-lying area, thereby guiding evacuation and rescue operations. Earlier, Jelesnianski and Chen (1979) developed a numerical model that allows flow of water inland from the model coastline. This has been accomplished by prescribing the advancing 123 Nat Hazards waterfront to move discontinuously from one grid point to another according to a preset criterion. However, to obtain acceptable accuracy, this procedure requires extremely fine grid spacing, which may not be otherwise necessary. In order to properly describe the physics of storm surges, a numerical model must resolve coastal features that can affect storm surge generation and propagation. This means the model domain must necessarily incorporate complex coastal geometries (bathymetry and topography), large gradients in bathymetry along the continental shelf that permits reasonable boundary conditions (i.e., tidal elevations). In this context, a finite-element- based model is assumed to be the best choice, as it allows flexibility to represent a larger spatial domain while permitting higher grid resolutions near the landward boundary. Blain et al. (1994) showed in a grid convergence study that near-coastal resolution is the most critical factor in determining the accuracy of storm surge computations. Although the coastal stretch along the east coast of India is vulnerable to storm surges and associated coastal inundation, no attempt seems to have made in detail on modeling the extent of inland inundation. In the present work, we use the advanced circulation (AD- CIRC) model that integrates domain-specific tools, standard grid, and portal tools to provide an integrated environment for forecasting and information dissemination. This framework allows storm surge computations to be run in a distributed grid environment. Here, the ADCIRC model (described in details by Luettich et al. 1992) is configured for maritime states of Andhra and Tamil Nadu along the east coast of India for computation of extreme surges and associated water level. Numerical experiments are carried out to compute the storm surges using the wind stress forcing representative of 1989 Kavali cyclone, November 1996 Andhra cyclone, and 2000 Cuddalore cyclone. The selection of these cyclones in the present study is based on the availability of observations as well as importance of the geographical locations of the landfall of the storms. The model computes water levels from the more accurate onshore topography associated with extreme surges generated by the cyclonic wind field. Winds in the model are calculated by using a dynamic storm model of Jelesnianski and Taylor (1973). Water levels along the open boundary are obtained from FES95.2 database. This database was developed using a global tidal model and has been found to perform very well in deep waters. The peak water level and associated extent of horizontal inundation simulated at the time of cyclone landfall are found to be in good agreement with observations of India Meteorological Department (IMD). It is to be noted that the vertical run-off height can be used as benchmark for the design level criteria of facilities for the particular region. This study would enable the coastal authorities for appropriate planning for disaster reduction in the event of any severe cyclone crosses a particular stretch of the coast.

2 Data sources

The cyclone tracks along with its intensity were collected from various sources viz; IMD reports, SMRC (1998), Unisys Hurricane Database (2006), and several research publications. Using this database, composite synthesized track is derived from observed tracks as well as the theoretical ones based on most favorable direction of the cyclone for Kalpakkam. The intensity of a cyclonic disturbance is measured by the strength of the associated winds. The feasibility study is carried out using parameters of the cyclone based on 50-year return period as an input to the model using the synthesized tracks. This may provide valuable information on extreme water levels that could occur along the Kalpakkam coast. 123 Nat Hazards

3 Model setup

3.1 Description of ADCIRC model

A detailed description of the finite-element-based hydrodynamic model ADCIRC-2DDI is available in Luettich et al. (1992). The governing model equations comprise of the depth- integrated equations for mass and momentum conservation, subject to incompressibility, Boussinesq, and hydrostatic pressure approximations. A set of conservation equations in primitive form expressed in the Cartesian coordinate system is incorporated in the model (Flather 1988; Kolar et al. 1994).

4 Wind forcing module

Two main forces, which generate storm surges on the sea surface, are surface wind stress and pressure gradient force. These act parallel and normal to the sea surface, respectively, and their relative importance depends on the water depth. Thus, the surface wind field associated with a tropical cyclone is the primary requirement for modeling of storm surges. For this purpose, the wind field at the sea surface is derived by using a dynamic storm model (Jelesnianski and Taylor 1973). To obtain a dynamic wind profile, initially, a stationary symmetric model wind profile is taken, and then, correction is applied to approximate the asymmetry due to storm motion. The vector equation that rules the hor- izontal motion of wind flow near the sea surface in the area of storm is

dVg 1 ¼ grad p þ f Vg k þ F ð1Þ dt qa where k is the vertical unit vector, Vg is the wind velocity, qa is the density of air, f is Coriolis parameter, p is the pressure of atmosphere, and F is the horizontal frictional force/ unit mass. Assume that the pressure and the wind field move forward without change during forward motion of the storm. For a stationary symmetric storm, from Eq. 1 we get, at a distance r from the center of the cyclone (Myers and Malkin 1961; Ueno 1981) 9 dp 2 Ks 1 dVg > ¼ qaVg => dr sin u Vg dr ð2Þ V2 > 1 dp g 2 du 2 ;> cos u ¼ f Vg þ cos u Vg sin u þ KnVg qa dr r dr where uðrÞ is the inflow angle of wind, that is, angle toward the storm center. Ks and Kn are empirically determined coefficients representing stress coefficients in the directions opposite and to the right of the wind, respectively. These stresses are given by the coef- ficient times the square of the wind speed. By eliminating pressure from Eq. 2, one can get du K u ¼ s fr K V r ð3Þ dr sin u n g where u ¼ Vgr cos u: Equation 3 can be solved numerically to obtain distribution of uðrÞ or uðrÞ:. On integrating the first part of Eq. 2, we obtain the distribution of p(r), which can be known from the wind profile relation of Jelesnianski and Taylor (1973):

123 Nat Hazards

2R r V r V m 4 gð Þ¼ R 2 2 ð Þ Rm þ r where VR is the value of the maximum wind speed and Rm is the radial distance from the storm center where the maximum wind speed appears. The value of Rm is usually fixed from the synoptic map, while the value of VR is determined by solving the Eqs. 2 and 3 numerically. The cyclonic wind forcing is calculated for every time step using the above formulation by providing the observed values of VR and Rm.

5 Numerical experiments

In the present study, ADCIRC model is applied for two different regions along the east coast of India to compute the storm surges and associated inland inundation. The model domain of the coast varies from 79.5°–87°E to 12.5°–19°N as shown in Fig. 1a. The model domain for Tamil Nadu coast comprises of 79.5°E–81.5°E and 11°N– 14°N as depicted in Fig. 1b. The geometry of the model domain is extracted from digital world map of National Geophysical Data Center (NGDC) and corresponding bathymetry is obtained from the general bathymetric chart of the oceans (GEBCO) 1-min global bathymetric grid. The storm surge model requires the wind stress forcing as the basic input to the model, which is described in the wind forcing module in Sect. 4. Wind speed in terms of surface stress is explicitly prescribed in the ADCIRC model, based on the rela- tionship proposed by Garrett (1977). The value for the bottom friction coefficient is set to 0.0026. Further, the wind model provides the pressure gradient to the ADCIRC model. The conversion of pressure gradient to equivalent water column height is obtained through the transformation P/qwg proposed by Blain et al. (1994). Finally, the wind stress and equivalent water column heights are linearly interpolated at each computational node of the finite-element mesh used in the model. The finite-element mesh was constructed using the software package surface modeling system (SMS) (Westerink et al. 1994). The program generates a grid with a low resolution in the deeper region, and high resolution when approaching near the coast. The node spacing varies between 50 m near the coast and about 3 km in the open ocean. The model mesh showing the variable grid resolution is shown in Fig. 1a, b. The cyclone tracks that are considered for our study are also shown in the figure. The landward boundary of the model is fixed from the coast based on 10 m topo contour, presuming that the surge would never exceed 10 m in this region. An explicit scheme is used in time discretization with a time step of 3 s. Minimum depth of 0.5 m is preset to delineate the wet and dry elements with a horizontal eddy viscosity coefficient of 5 m2 s-1.

6 Results and discussion

The simulations for different cyclone events have been discussed in detail in the following subsections.

6.1 Kavali 1989 cyclone

The 1989 Kavali cyclone crossed south Andhra coast 30 km north of Kavali around 1900 UTC on 8 November. Based on the observations, associated pressure drop and radius of 123 Nat Hazards

(a) November 1989 Kavali cyclone November 1996 Andhra cyclone .Srikakulam .Vijayanagaram . Visakhapatnam Andhra Pradesh

. Kakinada 10m topo

Coast line . Machilipatnam

. Ongole Bay of Bengal

. Kavali .

(b)

10m topo Coast line

Mypadu a.

Sriharikota. Bay of Bengal

Chennai . Tamilnadu Kalpakkam .

.Pondicherry . Cuddalore . Parangipettai

Hypothetical cyclone Cuddalore cyclone, 2000

Fig. 1 a The model grid for the Andhra coast along with the cyclone tracks. b The model grid for Tamil Nadu coast along with the cyclone tracks

123 Nat Hazards maximum winds used in the model are 70 hPa and 30 km, respectively. The model was integrated for 30 h till the cyclone crossed the coast. The wind module used in all the models provide dynamic storm conditions based on pressure drop and radius of maximum winds. The model is used for the computation of sea surface elevations using the 1989 Kavali cyclone data. The maximum wind associated with the cyclone is about 60 ms-1. The peak surge at the time of landfall of the cyclone and associated inland inundation along with water levels are shown in Fig. 2a, b. Here, the surge is shown only in the limited area, which was affected by the cyclone. The maximum surge (run-up height) is computed about 4.0 m. The maximum water level simulated is about 2.6 m, and the associated extent of horizontal inundation varies 1–3 km to the right of the cyclone landfall location. It is evident from the figure that vast coastal stretch of about 30 km is affected by surges[2m on the right-side of the track. The simulated surge heights are in good agreement with observations of about 3–4 m as reported by IMD (Table 1).

6.2 November 1996 Andhra cyclone

A well-marked low over the west central bay on November 4 and this low strengthened into a depression near 17°N and 86°E at 0300 UTC on November 5. It has became a cyclonic storm near 17°N and 85.5°E at 1200 UTC on the same day. It is intensified into a severe cyclonic storm near 17°N and 84.5°E at 0600 UTC on November 6. The model is integrated ahead in time upto 48 h with a pressure drop of 50 hPa and the radius of maximum wind of 30 km. Figure 3a shows the maximum run-up heights due to the

Fig. 2 Simulations of 1989 Kavali cyclone a computed maximum run-up heights from the mean sea level and b water levels

Table 1 Comparison of simulations of surges and extent of inundation with observations S. no. Cyclonic event Observations Simulations

Surge (m) Inundation (km) Surge (m) Inundation (km)

1 November 1989 Kavali cyclone 3–4 1–2 4.0 2 km 2 November 1996 Andhra cyclone 3.4 3.0 3.0 3 km 3 November 2000 Cuddalore cyclone 1.2–1.5 No inundation 1.4 No inundation

123 Nat Hazards cyclone that gives a maximum surge of about 3 m computed near Kakinada. The model- computed peak surge of 3 m is in good agreement with the post-storm survey reports of IMD of about 3.4 m. The extent of horizontal inland inundation is computed about 4 km, while it is reported only 3.0 km. The maximum water levels computed is about 2.6 m as shown in Fig. 3b.

6.3 Cuddalore 2000 cyclone

This simulation experiment comprised of two parts; the first part is to assess the perfor- mance of ADCIRC for a cyclonic event of the November 2000 cyclone. After performing the validation for this case study, the model is further used to simulate for an event close to the Kalpakkam coast. Firstly, the model is used to compute the surges associated with November 2000 cyclone, which crossed approximately 20 km south of Cuddalore. The maximum wind of the cyclone at the landfall time as seen from the Fig. 4a was about 35 ms-1. A peak surge of about 1.4 m (Fig. 4b) is simulated at Cuddalore, and surge heights of about 1 m is simulated in the close vicinity between north of Cuddalore and Pondicherry. The simulated peak surge agrees very well with the range of surge estimates of about 1.2–1.5 m provided by Unisys (2006).

6.4 Application to the Kalpakkam coast

As our primary concern is for the coast of Kalpakkam, we have made the synthesized track (as discussed in the earlier section) that had its landfall in the coastal area of Kanchipuram district. Statistical projections are made based on extreme value analysis (Gumble 1958)of available pressure drop values for this region during the past 100 years. Accordingly, it is found that the pressure drop is 66 hPa for a 50-year return period. The cyclone crossed the coast approximately about 40 km south of Kalpakkam (as shown in Fig. 1b). Numerical experiments have been performed for 50-year return period event using ADCIRC model to

(a) (b)

Kakinada . Kakinada.

. Coringa . Coringa

Pallipalem . Pallipalem .

Katrenikona . Katrenikona .

Fig. 3 Simulations of 1996 Andhra cyclone for zoomed region a computed maximum run-up heights from the mean sea level and b computed water levels 123 Nat Hazards

(a) (b)

Nellore.

. Pondicherry

Thiruvallur .

Chennai .

Kanchipuram. . Cuddalore Kalpakkam .

Pondicherry . Cuddalore.

Fig. 4 a Computed wind field at landfall time for November 2000 cyclone (the track shown by a long arrow). Arrows indicate wind direction and magnitude in color b shown surge height in color and associated currents with arrows in the zoomed region simulate the maximum probable surge amplitudes and associated inland inundation along the coast. In this experiment, the value for radius of maximum winds is considered as 40 km. The model is integrated for about 36 h until the cyclone crossed the coast. The maximum wind associated with the cyclone was about 62 ms-1. The model-simulated surges and associated currents along with the inundation are shown in Fig. 5. In this illustration, the surge is shown only for the limited area (Fig. 5a) that was affected by the cyclone. A maximum surge of 3.2 m was computed about 40 km to the right of the landfall near Kovalam. It is evident from this figure that vast coastal stretch of approximately 30 km is affected by surges [2 m, which prevailed on the right-side of the track. The associated simulated currents are very strong in the order of about 2.0 ms-1 along the affected coast. It is interesting to note that the currents are toward the coast where the maximum surges occurred; elsewhere, it is southward to the right of the landfall. These onshore currents eventually lead to piling up of water near the coast. Figure 5b provides the surge and associated inundation in the zoomed region surrounding Kalpakkam. It is noted that the region is affected by surges of about 2.7 m and associated inland inundation of about 0.5 km. In addition, it is critical to know the details of the water levels associated with the surge development for emergency purpose and also evacuation of people along the vulnerable coast. As the surges are computed from the mean sea level, the water levels are deduced by subtracting local topography from the surges as shown in Fig. 6a. Accordingly, the water level simulated is about 2.6 m near Kovalam, and the associated extent of horizontal inundation is about 1.1 km. Figure 6b depicts the water levels for the enlarged view encompassing the area within Kalpakkam region. It is to be noted that the coastal stretch between Kokilamedu and Kalpakkam is affected by water levels of about 2 m. This suggests that the region is more vulnerable to inland inundation causing significant damage to the local development if the cyclone-generated surges are more than 3.5 m. 123 Nat Hazards

(a) (b)

Run-up height (m) Run-up height (m)

Fig. 5 a Computed storm surges from the mean sea level along with associated currents and inundation based on 50-year return period b same as a except for zoomed portion of the southern region

(a) (b)

Bay of Bengal

Maximum inundation: 1.1 km Maximum water level: 2.6 m

Fig. 6 a Water levels associated with inundation based on 50-year return period b same as a except for zoomed portion of the southern region

7 Conclusions

As the coastline geometry is modeled more accurately, the simulation of extreme water levels and associated inland inundation from ADCIRC looks very realistic. The study reports importance of near-shore grid resolution and representation of complex coastline on computed storm surge and its location resulting from a tropical cyclone. To achieve high precision, grid size with resolution of about 50 m or less is required, which resolves the complexity of coastline leading to realistic surge heights and location of peak surge. The 123 Nat Hazards conclusions from this study have implications toward building long-term and short-term planning for disaster risk reduction along the coastal districts. The incorporation of high- resolution offshore topography and bathymetry, particularly in the continental shelf, may further enhance the capability of the existing ADCIRC model for precise computation of surge amplitude and associated inundation levels. The simulations of water levels along with horizontal extent of inundation are useful to provide early warnings to low-lying areas, guide evacuation of local population, and rescue operations.

Acknowledgments The authors are thankful to Dr. Westerink and Dr. Luettich for providing the ADCIRC model. Thanks are also due to the ICMAM-Project Directorate, Ministry of Earth Sciences, Government of India, Chennai, for granting the financial support to carry out this study.

References

Blain CA, Westerink JJ, Luettich RA (1994) Domain and grid sensitivity studies for hurricane storm surge predictions. In: Peters A et al (eds) Computational methods in water resources X, vol 1. Kluwer Academic, Heidelberg, p 1548 Dube SK, Sinha PC, Rao AD, Rao GS (1985) Numerical modelling of storm surges in the Arabian Sea. Appl Math Model 9:289–294 Flather RA (1988) A numerical model investigation of tides and diurnal-period continental shelf waves along Vancouver Island. J Phys Ocean 18:115–139 Garrett JR (1977) Review of the drag coefficients over oceans and continents. Mon Weather Rev 105:915–929 GEBCO http://www.bodc.ac.uk/products/bodc_products/gebco/ Gumble EJ (1958) Statistics of extremes. Columbia University Press, New York Jelesnianski CP, Chen J (1979) SLOSH (sea lake and overland surges from hurricanes). Report of TDL, NWS, Silver Spring, MD, USA, p 16 Jelesnianski CP, Taylor AD (1973) NOAA technical memorandum. ERL, WMPO-3, p 33 Johns B, Dube SK, Sinha PC, Mohanty UC, Rao AD (1982) The simulation of continuously deforming lateral boundary in problems involving the shallow water equations. Comp Fluids 10:105–116 Kolar RL, Gray WG, Westerink JJ, Luettich RA (1994) Shallow water modeling in spherical coordinates: equation formulation, numerical implementation, and application. J Hydraul Res 32:3–24 Luettich RA Jr, Westerink JJ, Scheffner NW (1992) ADCIRC: an advanced three dimensional circulation model for shelves coasts and estuaries, report 1: theory and methodology of ADCIRC-2DDI and ADCIRC-3DL. Dredging Research Program Technical Report DRP-92-6, U.S. Army Engineers Waterways Experiment Station, Vicksburg, MS, USA, p 137 Myers VA, Malkin W (1961) Some properties of hurricane wind fields as deduced from trajectories. Nat Hurric Res Proj Rep 49:5–9 Rao AD, Dube SK, Chittibabu P (1994) Finite difference techniques applied to the simulation of surges and currents around Sri Lanka and Southern Indian Peninsula. Comput Fluid Dyn 3:71–77 Rao AD, Jain I, Murthy MVR, Murty TS, Dube SK (2009) Impact of cyclonic wind field on interaction of surge-wave computations using finite-element and finite-difference models. Nat Hazards 49:225–239 SMRC (1998) The impact of tropical cyclones on the coastal regions of SAARC countries and their influence in the region, SMRC-no. 1. SAARC Meteorological Research Centre, Dhaka, Bangladesh, October 1998, p 329 Ueno T (1981) Numerical computations of the storm surges in Tosa Bay. J Oceanogr Soc Jpn 37:61–73 Unisys Hurricane Database, Unisys Corporation (2006). http://weather.unisys.com/hurricane/indian_oc/index.html Westerink JJ, Blain CA, Luettich RA, Scheffner NW (1994) ADCIRC: an advanced three dimensional circulation model for shelves coasts and estuaries, report 2: users manual for ADCIRC-2DDI. Dredging Research Program Technical Report DRP-92-6, U.S. Army Engineers Waterways Experi- ment Station, Vicksburg, MS, USA, p 156

123