Simulation of cyclone-induced storm surges in the low-lying delta of Bangladesh using coupled hydrodynamic and wave model (SWAN + ADCIRC) M. Deb and C.M. Ferreira
Department of Civil, Environmental & Infrastructure Engineering, George Mason University, Fairfax, VA, USA
Correspondence Abstract Mithun Deb, Department of Civil, Environmental & Infrastructure Bangladesh is vulnerable to several natural disasters and cyclone-generated storm Engineering, George Mason University, surges have resulted in the deaths of over 700 000 people since 1960. Advancing Fairfax, VA 22030, USA our capability to model and simulate storm surges using numerical models is Tel: +1 571 265 9815 utmost important to support early warning and emergency response efforts in Emails: [email protected]; the region. This study primarily explored the effectiveness of a hydrodynamic [email protected] model (ADvanced CIRCulation, ADCIRC) coupled with wave model (Simulating WAves Nearshore, SWAN) under a high-performance computing DOI: 10.1111/jfr3.12254 environment to simulate storm surge and inundation in coastal regions of Bangladesh. The modelling framework was validated using data from freely avail- Key words Bangladesh; coastal flooding; cyclone; able historical reports and buoy data. The model-generated storm surge water 2 storm surge; SWAN + ADCIRC. level shows good agreement with the observations with maximum R value of 0.98 and root mean square error of 0.30 m. Ultimately, research findings have highlighted the importance of the coupled wave and hydrodynamic modelling to calculate storm surges in a region with poor observational coverage.
Introduction the storm surge height, causing disastrous floods along the coast (Murty et al., 1986; Dube et al., 1997; Madsen and In Bangladesh, flooding due to tropical cyclones has Jakobsen, 2004). resulted in the deaths of over 700 000 people since 1960 Several numerical models have been previously devel- (Chowdhury and Karim, 1996) and it is still considered as oped to simulate storm surges associated with cyclonic one of the most destructive meteorological phenomena in storms making landfall on the coast of Bangladesh (e.g. the region. Almost one sixth of tropical cyclones that Flather and Khandoker, 1993; Flather, 1994; Roy, 1995; develop in the Bay of Bengal make landfall on the Henry et al., 1997). Other studies have demonstrated lim- Bangladesh coast (Islam et al., 2011). Approximately 5% of itations in storm surge modelling in the Bay of Bengal (e.g. the global tropical cyclones form over the Bay of Bengal; Ali, 1979; Murty et al., 1986; Das, 1994a, b; Dube et al., however, fatalities were observed to be 80% of the global 1997; Chittibabu, 1999). More recently, depth-averaged record (Debsharma, 2007). Numerous historic super two-dimensional (2D) hydrodynamic models have been cyclones, such as the 1970 Bhola cyclone, the 1991 used to solve the shallow-water continuity equations and Bangladesh cyclone (hereafter referred as BOB 1991) and calculate storm surges in the Bay of Bengal (e.g. Dube et al., the Sidr 2007 cyclone had catastrophic effects in the coastal 1994; Madsen and Jakobsen, 2004; Bhaskaran et al., 2013, areas of Bangladesh by taking thousands of lives and caus- 2014; Murty et al., 2014). These hydrodynamic models ing significant property damage. In November 1970, the require wind and pressure fields derived from dynamic most deadly cyclone of the century passed through Bhola storm models (e.g. Jelesnianski and Taylor, 1973; Holland, and killed nearly 500 000 people in Bangladesh (Madsen 1980) using observed cyclone data (track, central pressure and Jakobsen, 2004; Islam et al., 2011). The coastal region and maximum wind speed). It is worthwhile to note that of Bangladesh is particularly vulnerable to storm surge implementation of a coupled modelling system that gener- flooding because of low-lying heavily inhabited areas and ates combined effects of physical processes such as tides, continental shelf with shallow bathymetry, which amplifies storm surges and waves in Bay of Bengal is the subject of
J Flood Risk Management (2016) © 2016 The Chartered Institution of Water and Environmental Management (CIWEM) and John Wiley & Sons Ltd Deb and Ferreira ongoing research. It has been shown that wave radiation we developed an unstructured high-resolution numerical stresses can affect the wave-induced currents and increase mesh for the Bangladesh coast. We examined the effective- water level by 5%–20% in regions with broad continental ness of SWAN + ADCIRC model in assessing waves and shelf and nearly 35% in regions of steep slope (Funakoshi storm surge in a data-poor region and showed the model et al., 2008; Dietrich et al., 2010). A coupled wave and performance considering the limitations with data availabil- hydrodynamic model can provide better understanding of ity and accuracy. nearshore hydrodynamics during extreme events such as cyclones (Dietrich et al., 2010). Although a few recent stud- ies have employed a coupled wave-tide circulation model- Study area ling framework for the South China Sea and Bay of Bengal (e.g. Moon, 2005; Bhaskaran et al., 2013, 2014; Murty et al., The Bay of Bengal is among the most vulnerable and least 2014), further development of this framework is needed for studied areas when compared with other cyclone-prone effective implementation. parts of the world. The coastal regions of Bangladesh, India With the advancement of computational resources and and Myanmar suffer the most in terms of casualties by numerical models, comprehensive validations of coupled storm surges, while the actual occurrence of extreme tropi- hydrodynamic and wave models have been established for cal cyclones is not relatively high (Debsharma, 2007). The the Atlantic Ocean (e.g. Dietrich et al., 2011). For example, model domain introduced here includes the entire the 2D, depth-integrated and finite element hydrodynamic Bangladesh coast and part of the Northern Indian Ocean, circulation model ADCIRC (ADvanced CIRCulation) the east coast of India and part of Myanmar (Burma) (Westerink et al., 1993) has been validated for several hur- (Figure 1). The Meghna estuary along the coastal belt of ricanes in the Atlantic basin (Westerink et al., 2008; Bunya Bangladesh has frequently suffered from impacts of tropical et al., 2010), and it has been used by the US Army Corps of cyclones causing heavy loss of life and property. This Engineers (USACE) to estimate hurricane flooding risk in research has prioritised the low-lying Ganges delta, coastal the United States (e.g. Cialone et al., 2008). Similarly, a rivers and mangrove forests in the coast of Bangladesh for third-generation wind-wave model SWAN (Simulating storm surge vulnerability. Among numerous contributing WAves Nearshore), which describes the evolution of the factors in the coastal stretch of the Bangladesh shallow 2D wave energy spectrum, is being used extensively nowa- water shelf, densely populated small islands and countless days for simulating shallow water waves (Dietrich et al., number of inlets are considered as the most sensitive 2011; Ferreira et al., 2014a). The coupled SWAN + elements for disastrous storm surges. Therefore, a high- ADCIRC model is highly scalable and performs well to resolution unstructured numerical mesh with approxi- compute waves and circulation during their transmission mately 200 000 nodes and 400 000 elements has been gen- fi from deep to shallow water zones (Dietrich et al., 2010). erated by keeping a ne resolution of 300 m in the shallow Recently, the SWAN + ADCIRC model has been validated water zones of the Bangladesh coast and 52 km in the deep for the Atlantic basin, Gulf of Mexico and several coastal ocean boundary. Several numerical experiments were car- areas in the United States with a significant number of ried out using the meteorological forcing representative of National Oceanic and Atmospheric Administration the BOB 1991 and the 2007 Sidr cyclones, which travelled – – (NOAA) water-level recording gauges and National Data across the Ganga Brahmaputra Meghna deltaic system of Buoy Center (NDBC) stations (e.g. Ferreira et al., 2014a). Bangladesh and were among the most hazardous natural While the availability of high-resolution data, such as light calamities of the century. detection and ranging (LiDAR), land cover and extensive water-level monitoring, has supported the rapid advance- Methodology ment of the implementation and validation of storm surge models in some areas of the world, it is as yet difficult to Mesh development implement accurate coupled hydrodynamic models in coastal regions of developing countries like Bangladesh, The precision of coupled wave and hydrodynamic models where high-resolution topographic and bathymetric data depends on the quality of data available (Sebastian et al., sets are not available and the number of water-level record- 2014). Higher-resolution topographic data (e.g. LiDAR) ing gauges is limited for model validation. can support the simulation of storm surge propagation In this study, we have applied a tightly coupled SWAN + accurately (Wang et al., 2014), but is not yet publically ADCIRC model on a single unstructured mesh to explore available for the Bangladesh coastal zones. For this analysis, its performance under various meteorological forcing con- global topographic and bathymetric data used to generate ditions and the subsequent storm surge flooding in the the numerical mesh for SWAN + ADCIRC simulation coastal areas of Bangladesh. During the validation process, were collected from the freely available General
© 2016 The Chartered Institution of Water and Environmental Management (CIWEM) and John Wiley & Sons Ltd J Flood Risk Management (2016) Cyclones and coastal flooding in Bangladesh
Figure 1 Study area (Bay of Bengal: Bangladesh coast).
Bathymetric Chart of the Oceans (GEBCO) database (e.g. of 300 m (Figure 2(a)). This unstructured numerical mesh Bhaskaran et al., 2013). The GEBCO data set consists of a comprises almost 200 000 nodes and 400 000 triangular 30 arc-second (900 m) resolution grid generated by joining elements that can represent many of the tortuous rivers quality-controlled ship depth soundings (Henstock et al., and smaller islands in the estuary (Figure 2(b,c)). This wide 2006) and land data based on the 3 arc-second (90 m) reso- coverage of ocean boundary allows tides to be specified at lution Shuttle Radar Topography Mission (SRTM30) the deeper basin of the Bay and storms to propagate for a (Jarvis et al., 2004). A numerical mesh with high resolution reasonable time inside the domain. Bathymetry in some in shallow water zones is the most critical factor in storm nearshore water bodies, such as rivers and channels, was surge computations (Blain et al., 1994). The numerical extracted from local riverine studies (e.g. Ali et al., 2007), mesh boundary of earlier Bay of Bengal models was as the 900 m GEBCO data set has poor definition of these restricted to the vicinity of the coast only (Flather, 1994; smaller streams. All of the collected data sets were observed Salek, 1998), excluding coastal districts of Bangladesh. Roy to have the same datum, mean sea level (MSL) similar to (1995) has defined a model domain for the Bay of Bengal GEBCO. For example, Lewis and Bates (2013) have pro- of about 1.4 × 1.3 km2 resolution for shallow water zones vided the dredging information of Pussur River channel to consider small islands of the estuary and 22.2 × 21 km2 which kept the bathymetry at constant of 6 m. Additional of coarse resolution for the outer ocean boundary with information about Meghna estuary bathymetry has been 10 612 grid points. In a related study, Mandal et al. (1996) derived from Ali et al. (2007). considered a nested finite element mesh consisting of 5686 grid points, with 46 × 81 points in the fine mesh scheme and 40 × 49 grid points in the coarse mesh scheme. More- ADCIRC + SWAN over, in a recent study, Mashriqui et al. (2006) have estab- We performed the hurricane storm surge simulations using fi lished an unstructured nite element mesh including about the coupled hydrodynamic and wave model SWAN + 363 399 elements and 186 981 nodes for the Bay of Bengal, ADCIRC (Dietrich et al., 2011). ADCIRC is a finite- consisting part of the Northern Indian Ocean, the east element hydrodynamic model that generates water levels coast of India and all of the coasts of Bangladesh. The and current velocities, and is widely used for storm surge model domain introduced in this analysis also contains the modelling in the east coast of United States and Gulf of entire Bangladesh coast and part of the Northern Indian Mexico (e.g. Ferreira et al., 2014b). We used the 2D, depth- Ocean but uniquely focusing on coastal districts and low- integrated version of ADCIRC (Luettich and Westerink, fi fi lying Ganges delta for the rst time with a ne resolution 2004) that solves the vertically integrated generalised wave
J Flood Risk Management (2016) © 2016 The Chartered Institution of Water and Environmental Management (CIWEM) and John Wiley & Sons Ltd Deb and Ferreira