Estimation of extreme sea levels over the eastern continental shelf of North America

Authors


Abstract

[1] This study presents distributions of extreme sea levels over the eastern continental shelf of North America (ECSNA) associated with storm surges and tidal surface elevations produced by a 2-D ocean circulation model for the period 1979–2010. The 2-D circulation model is driven by atmospheric and tidal forcing. The large-scale atmospheric forcing is the wind stress and sea level atmospheric pressures extracted from NCEP (National Centers for Environmental Prediction) Climate Forecast System Reanalysis (CFSR) fields at 6 h intervals. A parameterized vortex is inserted into the CFSR fields to better represent the atmospheric forcing associated with a tropical storm or hurricane. The tidal forcing includes specification of tides at model open boundaries and tide generating potential at each model grid. The model performance is assessed using observed sea levels over the ECSNA. The simulated surface elevations driven by the atmospheric force are used to estimate the 50 year return level of extreme sea levels associated with storm surges over the ECSNA using an extremal analysis. The potential regions of the ECSNA to be threatened by severe storm surges are presented. The extreme total sea levels due to tides and storm surges are also estimated using the Monte Carlo method from model results. Regions over the ECSNA experiencing severe 50 year extreme total sea levels due to the storm surges and tides are similar to those regions of 50 year extreme surge-induced elevations, but with much higher extreme values.

1. Introduction

[2] Many coastal communities of the world are vulnerable to extreme marine conditions such as extreme sea levels, ocean waves, and currents. In fact, much of losses of human lives and property in the past are caused by extreme sea levels due to hurricane-induced storm surges [Murty et al., 1986; McRobie et al., 2005; Fritz et al., 2007; Lin et al., 2010]. Over the eastern continental shelf of North America (ECSNA, Figure 1), storm surges also caused severe losses in human lives and property damage. For example, Hurricane Katrina in 2005 struck New Orleans and caused 80% of the city flooded due to severe storm surges, with losses of about 1800 lives and property damage of about US$81 billion (2005 USD). Hurricane Sandy in 2012 generated about 4 m storm surges during its landfall, with the property damage of about US$71.4 billion. The other serious issue for coastal communities is sea level rise due to the climate change. The global mean sea level rise was estimated to be in the range between 9 and 58 cm or larger for the 21st century [Rahmstorf, 2007]. As a result, the future risks of coastal flooding due to storm surges will increase.

Figure 1.

Major topographic features over the eastern continental shelf of North America (ECSNA). Colored symbols represent locations of tide gauge stations at which the observed sea levels are used in this study. The names of tide gauge stations and major topography over the regions marked by dotted lines are given in Figure 2. Green stars represent tide gauge stations in Hudson Bay and Labrador Sea. Thick yellow lines represent three open boundaries of the 2-D circulation model.

[3] Reliable estimation of extreme sea levels is needed to mitigate hazardous impacts of extreme marine conditions over coastal areas. Most of previous studies focused on estimations of extreme sea levels based on tide gauge data with long record lengths (30 years or longer) using various statistical methods. For example, Tsimplis and Blackman [1997] used the r-largest order statistic approach to estimate the return periods of extreme sea levels from available tide gauge records in the Aegean and Ionian seas. Letetrel et al. [2010] used the generalized Pareto distribution (GPD) approach to investigate extreme sea levels in Marseille. Bernardara et al. [2011] applied the regional frequency analysis and the GPD approach to estimate the extreme storm surges along the French Atlantic coasts. Others used the generalized extreme value (henceforth the GEV) statistic approach to estimate sea level extremes [Dixon and Tawn, 1999; Woodworth and Blackman, 2002].

[4] Over regions where tide gauge data of long record lengths are not available, estimations of extreme sea levels can be made from sea level fields produced by ocean circulation models. The extreme storm surges of 40 year return periods over the northwest Atlantic coastal waters were estimated by Bernier and Thompson [2006] from storm-surge model results using the GEV approach. The total sea levels due to storm surges and tides with 40 year return periods were estimated for several selected tide gauge data based on the joint probability method [Pugh and Vassie, 1980]. A r-largest statistical method was used by Butler et al. [2007] to analyze decadal variations in storm-surge elevations over the northwest European shelf. The GEV approach was also used in estimating extreme subtidal ocean currents from model results by Oliver et al. [2012].

[5] The main objective of this study is to estimate the extreme sea levels due to storm surges and tides over the ECSNA using 32 year simulations produced by a 2-D circulation model using the GEV approach and the Monte Carlo method (MCM). The structure of this paper is as follows. Section 2 describes the sea level observations, the ocean circulation model, and the external forcing. Section 3 presents the performance assessment of the 2-D circulation model. Section 4 presents the extreme sea levels with the 50 year return period over the ECSNA. Section 5 is a summary and conclusions.

2. Data and Numerical Model

2.1. Sea Level Observations

[6] The observations used in this study include hourly time series of observed sea levels with record lengths greater than 2 years at 90 tide gauge stations over the coastal waters of the ECSNA, and those with relatively shorter record lengths at four tide gauges inside James Bay and adjacent waters (Figures 1 and 2). The tide gauge data along the east coast of Canada were provided by the Marine Environmental Data Service (http://www.meds-sdmm.dfo-mpo.gc.ca). The tide gauge data along the northern coast of the Gulf of Mexico, Puerto Rico, Barbuda, and the east coast of the United States were taken from the Center for Operational Oceanography Products and Services (http://www.co-ops.nos.noaa.gov/). The tide gauge data from the Caribbean Sea and along the east coast of Mexico were provided by the Sea Level Center at the University of Hawaii (http://uhslc.soest.hawaii.edu/). Figure 3 shows the record lengths and gaps of the tide gauge observations used in this study. All the tide gauge data were subject to a careful quality control, including eliminations of any datum shift errors and abnormal values.

Figure 2.

Locations (colored symbols) and indices of tide gauge stations (a) over the east coast of the United States and Canada and (b) along the coast of the Gulf of Mexico and Caribbean Sea. Abbreviations are used for Northumberland Strait (N.S.), the Bay of Fundy (BOF), and the Gulf of Maine (GoM). The legend in each panel describes the colored symbols used to define locations of tide gauge stations. The indices for the tide gauge stations along Greenland, Hudson Bay, and Baffin Island are marked in Figure 1. Depth contours are shown for the 200 m (blue line) and 2000 m (gray line).

Figure 3.

Availability of hourly time series of observed sea levels at 90 tide gauge stations over the ECSNA from 1979 to 2011.

[7] The observed sea levels were decomposed into the tidal and nontidal components on a yearly basis using a tidal analysis package known as T_TIDE [Pawlowicz et al., 2002]. Since the 2-D model used in this study is barotropic and forced mainly by six-hourly winds and pressures and hourly tidal forcing, the model does not simulate the annual (Sa), semiannual (Ssa), and interannual (Zo) variabilities of sea levels. The latter are caused mainly by baroclinic processes in the ocean such as the net heat and fresh water fluxes at the sea surface and influence of freshwater discharge from rivers and coasts. Furthermore, the 2-D model does not generate high-frequency sea level variability, such as wave setup and seiches over coastal waters. As a result, the annual means of nontidal components in the observed sea level were first removed (removal of Zo), and a band-pass filter was then applied to remove variability at periods shorter than 12 h and longer than four weeks. The resultant tidal and surge-generated components of the observed sea levels will be used to validate the model performance in simulating tides and storm surges.

2.2. Model Configuration

[8] A 2-D ocean circulation model has widely been used in simulating tidal and surge-generated sea levels in the ocean [Bernier and Thompson, 2006; Marcos et al., 2009, 2011]. In this study, the external (or depth-averaged) component of the Princeton Ocean Model (POM) [Mellor, 2004] is applied to the ECSNA based on the following momentum and continuity equations:

display math(1)
display math(2)

where math formula represents the depth-averaged horizontal velocity vector, math formula, η is the sea level which includes the inverted barometer effect, math formula is the upward pointing vector of the Coriolis parameter, math formula and math formula are the surface and bottom stress vectors, respectively, A is the horizontal viscosity coefficient, H is the total water depth (H = η + h), h is the mean water depth, Pa is the atmospheric pressure at the sea level, and ρo is the reference (constant) water density.

[9] The model domain covers the region between 7°N and 70°N and between 100°W and 35°W (Figure 1). The model horizontal resolution is (1/16)°, which is about 7 km in latitude and 4.9 km in longitude at 45°N. The bathymetric data used in the model were extracted from ETOPO1, which is a 1 arc-minute global relief data set of Earth's land topography and ocean bathymetry (http://www.ngdc.noaa.gov/mgg/global/). The ECSNA model has three open lateral boundaries in the north, east, and south, respectively (Figure 1).

[10] The model is driven by tidal forcing and atmospheric forcing. For simulations of tides, the tide generating potential that includes the earth tide and the ocean loading tide is specified at each model grid point according to Schwiderski [1980] and Foreman et al. [1993]. In addition, specification of tidal sea levels and depth-averaged tidal currents with eight tidal major constituents of M2, S2, K1, O1, N2, K2, P1, and Q1 is made at the model open boundaries based on the results produced by the TPXO-7.2 global tidal model with (1/4)° resolution [Egbert and Erofeeva, 2002]. A radiation condition [Davies and Flather, 1978] is applied at model open boundaries based on

display math(3)

where Ut is the tidal currents normal to the open boundary and ηt is the tidal elevations. The sign is chosen to ensure that waves approaching the boundary radiate outward.

[11] For simulations of storm surges, the model is driven by wind stress and the atmospheric pressure at the sea level (SLP). The wind stress is calculated from the wind velocity at 10 m above the sea surface using the bulk formula of Large and Pond [1981]. The SLP fields and 10 m wind velocities were taken from the Climate Forecast System Reanalysis (CFSR) fields at 6 h intervals [Saha et al., 2010]. It should be noted that the CFSR fields have horizontal resolutions of 0.3° and 0.5° for the winds and pressure, respectively, which are much finer than the resolutions (about 2°) of previous generations of the reanalysis data set produced by the National Centers for Environmental Prediction (NCEP). Nevertheless, the horizontal resolutions of the CFSR fields are still not fine enough to resolve well the fine structure of the wind and atmospheric pressure fields associated with a hurricane or tropical storm. As a result, a parameterized vortex is inserted to the CFSR fields based on the idealized wind and pressure profiles in hurricanes suggested by Holland [1980]:

display math(4)
display math(5)

where ps(r) and Vs(r) are respectively the atmospheric pressure (in millibars) at the sea level and the gradient wind speeds (in m s−1) at radius r (in km) with respect to the hurricane center at (xc, yc), pn is the ambient pressure, pc is the atmospheric pressure at the storm center, f is the Coriolis parameter, and ρa is the air density which are set to be 1.15 kg m−3. In equations (4) and (5), A and B are scaling parameters, and their relation is determined by the following equations:

display math(6)

where Rm is the radius (in units of km) at which the maximum wind speed occurs. The above parameterized hurricane model has successfully been applied in storm-surge simulations over the western Atlantic [Peng et al., 2006; Rego and Li, 2010; Lin et al., 2010]. It should be noted that A, B, and Rm are key parameters which define the wind and SLP fields associated with a hurricane. Unfortunately, the values of Rm are not available in the best storm track (BST) data set taken from the National Hurricane Center database (HURDAT) (http://www.aoml.noaa.gov/hrd/hurdat/DataByYearandStorm.htm). As a result, only the six-hourly storm center position (xc, yc), the maximum wind speed (Vm), and central atmospheric pressure (pc) taken from HURDAT were used in this study. Values of Rm and B were estimated using the following formula suggested by Vickery and Wadhera [2008]:

display math(7)
display math(8)

where Δp (in millibars) is the difference between the central and ambient atmospheric pressures, Rm is the radius in units of kilometer at which the maximum wind speed occurs, and ϕ is the latitude in degree. These empirical formula were successfully applied in the risk assessment study on the hurricane-induced storm surges for New York City [Lin et al., 2010]. In this study, hurricane wind and pressure profiles are calculated using equations (4)-(8) and inserted into the CFSR atmospheric fields by using spatial interpolation during periods when the best storm track data are available.

2.3. Extremal Analysis Technique and Monte Carlo Method

[12] Several extremal analysis techniques were developed to estimate the return periods of extreme sea levels [Tsimplis and Blackman, 1997; Letetrel et al., 2010; Bernardara et al., 2011; Dixon and Tawn, 1999; Woodworth and Blackman, 2002]. In this study, the GEV approach is used to estimate the extreme sea levels from the annual maxima of hourly sea levels taken from observations or computed from numerical simulations. We define Mn as the maximum sea level over a block of length n:

display math(9)

where η1, …, ηn represent a sequence of independent random variables which have a common distribution function F. In this study, η is the hourly sea levels data in each year, and Mn is the annual maximum value of η. This distribution is conveniently summarized by the GEV distribution defined as [Coles, 2001]:

display math(10)

where Fgev(χ) is the cumulative probability distribution (CDF) of Mn, a is the location parameter, b > 0 is the scale parameter, and ξ is the shape parameter.

[13] For ξ → 0, the GEV distribution converges to the Type I distribution:

display math(11)

[14] Maximum likelihood is used to estimate parameters by fitting equation (11) to extreme annual values. It should be noted that typically 30 or more annual maxima are required for reliable estimations of a and b [Bernier and Thompson, 2006].

[15] For estimating the total sea levels due to the combination of tides and storm surges, we follow Oliver et al. [2012] and use the Monte Carlo approach to generate new realizations of total sea levels by randomly changing the time lag between the tidal and nontidal components of the surface elevations. By assuming that the nonlinear interaction between tides and nontidal components is small to be ignored, the observed or simulated total sea levels (ηt) can be decomposed as

display math(12)

where math formula is the nontidal component and math formula is the tidal component of ηt. In this study, new realizations of total sea levels were generated by repeating the nontidal component M times and each time adding a tidal component with a randomly chosen time lag relative to the nontidal component:

display math(13)

for math formula. Here N is the length of math formula and math formula, and math formula are random integers between 1 and 8760 h (or 8784 h in a lunar year) in a year. In comparison with the simple approach that adds the tidal and storm-surge components to yield the total sea levels, the Monte Carlo approach provides more realistic estimations of the total sea levels and associated extremes.

3. Numerical Experiments and Model Validation

[16] Five numerical experiments were conducted in this study to assess the model performance and also to determine main physical factors affecting extreme sea levels over the ECSNA. These five numerical experiments are as follows:

[17] (a) Barotropic run (exp-BR): The 2-D circulation model based on the POM in this experiment is driven by all of the forcing functions discussed in section 2.2, which include tidal forcing, the CFSR wind forcing and atmospheric pressure fields, and the insertion of a vortex associated with a hurricane (or tropical storm) based on Holland's hurricane model.

[18] (b) No Holland model (exp-NH): The model forcing in this experiment is the same as in exp-BR except for the vortex insertion.

[19] (c) No tides (exp-NT): The model forcing in this experiment does not include the tidal forcing, with all other forcing functions to be the same as in exp-BR.

[20] (d) CFSR only (exp-CFO): The model in this experiment is driven by only the CFSR wind and atmospheric pressure fields.

[21] (e) Tide only (exp-TO): The model is driven by only the tidal forcing which includes specification of tidal elevations and currents at the model open boundaries and the tide generating potential at each model grid point.

[22] The 2-D circulation model was integrated for 32 years from 1979 to 2010 in the above five experiments.

3.1. Tidal Simulation

[23] We first assess the model performance in simulating tidal surface elevations over the ECSNA. Previous studies demonstrated that the semidiurnal principal lunar M2, semidiurnal principal solar S2, diurnal lunisolar K1, and diurnal principal lunar O1 are the four principal tidal constituents in this region [Hill et al., 2011]. To compare simulated tides with observations, the model-calculated tidal harmonic constants of these four tidal components were computed from model results in exp-TO.

[24] Figure 4a demonstrates the simulated M2 tidal waves propagate into the ECSNA from deep waters of the North Atlantic. The simulated amplitudes of M2 reach about ∼5 m at the head of the Bay of Fundy and ∼4 m at Hudson Strait due mainly to the resonant effect [Greenberg, 1979; Arbic et al., 2007]. Over the St. Lawrence Estuary, the M2 amplitudes reach up to 2 m. Over the coastal waters of Hudson Bay, Newfoundland, western Greenland Island, and eastern Florida, the amplitudes of M2 are in the range between 0.5 and 1.5 m.

Figure 4.

(left) Coamplitudes (color image) and cophases (contours) of tidal constituents M2 and K1 computed from model results in exp-TO and (right) results produced by the TPXO-7.2 global tidal model.

[25] The simulated amplitudes and phases of S2 have large-scale distributions very similar to M2 over the ECSNA, but with reduced amplitudes (not shown). The simulated S2 amplitudes reach approximately 0.5 m in regions between Greenland Island and Baffin Island, more than 1.0 m in Hudson Strait, and up to 0.6 m in the St. Lawrence River Estuary and the Bay of Fundy. Over the eastern coastal waters of the United States and the western coastal waters of Florida, the simulated S2 amplitudes are in the range of 0.1–0.25 m, and below 0.1 m in the Gulf of Mexico except for coastal waters of western Florida.

[26] The simulated tidal elevations of K1 (Figure 4c) has a large-scale anticlockwise propagation in the central North Atlantic, with amplitudes increasing from a few centimeters in the deep waters to about 0.1–0.4 m over the coastal waters of the ECSNA (Figure 4c). In comparison with other regions of the ECSNA, the K1 amplitudes are relatively large over coastal waters of western Greenland Island, the western Gulf of St. Lawrence, and the Gulf of Mexico, with maximum amplitudes of about 0.35, 0.25, and 0.18 m, respectively. The simulated O1 amplitudes and phases (not shown) are similar to K1 in the large-scale distribution, except for relatively smaller amplitudes over coastal waters of western Greenland Island.

[27] The simulated amplitudes and phases of the four principal tidal constituents produced by the POM are in good agreement with previous studies of tidal simulations over the study region using the basin-scale [Westerink et al., 1994; Hill et al., 2011] or regional ocean circulation models [Gouillon et al., 2010; Han et al., 2010; Hasegawa et al., 2011]. For comparison, Figures 4b and 4d present the amplitudes and phases of M2 and K1 calculated from results produced by TPXO-7.2 global data-assimilative tidal model with a horizontal resolution of (1/4)° [Egbert and Erofeeva, 2002]. The simulated tidal amplitudes and phases produced by the POM are in good agreement with the tidal elevations extracted from the TPXO-7.2 data set, especially in terms of spatial distributions of the tidal amplitudes and amphidromic points over the ECSNA.

[28] The model performance in simulating tidal elevations is further assessed by comparing the model results with the observed tidal sea levels at 90 tide gauge stations over the coastal waters of the ECSNA. It should be noted that the horizontal resolution of the 2-D model used in this study is (1/16)°, which is not fine enough to resolve the local topography in harbors and bays of which the length scales are comparable to, or less than, the model horizontal resolution. As a result, the 2-D model with the current resolution should perform less well in reproducing the observed tides at locations where the tides are significantly affected by the local topography.

[29] The following three statistical parameters are used to assess the model performance. We follow Thompson and Sheng [1997] and use γ2 to quantify the fit between the observed and simulated hourly time series of sea levels at each tide gauge:

display math(14)

where Var represents the variance operator, and O and M represent the observed and simulated time series of sea levels, respectively. The smaller γ2 is, the better the performance of the model is. In this study, the γ2 values less than 0.3 and 0.6 are set to be criteria for assessing the model performances in simulating tides and surges, respectively. We also quantify the model performance using the root-mean-square error (RMSE, σ) defined as

display math(15)

where N is the number of observations, Mi is the simulation value, and Oi is the observed value. In addition, the average absolute difference (AAD) is used to estimate the mean model error in reconstructing the observations. The AAD is the average error over sample numbers of the absolute deviation between model results and observations [Urrego-Blanco and Sheng, 2012]:

display math(16)

where Oi is the observed values, Mi is the simulated values, and n is the number of observations.

[30] Figure 5 presents time series of observed and simulated tidal sea levels and differences between them during the period from 1 February to 23 March 2010 at four selected tide gauge stations (Cap-aux-Meules, Halifax, Sandy Hook, and Panama City) in four different regions. The 2-D circulation model reproduces reasonably well the tidal components of the observed sea levels at these four locations during this period, with RMSE, AAD, and γ2 values less than 0.1 m, 0.12 m, and 0.25, respectively.

Figure 5.

Hourly time series of observed (red) and simulated (blue) tidal sea levels and their differences (black) at four selected tide gauge stations over the ECSNA.

[31] Table 1 lists the γ2 and RMSE values at 90 tide gauge stations calculated using hourly time series of simulated and observed sea levels during the periods when the observed sea levels are available at each tide gauge station (see Figure 2) for the period 1979–2010. At tide gauges over coastal waters of the Scotian Shelf (stations 12–14), the Gulf of Maine (stations 24–28), and eastern United States (stations 29–44), the γ2 values are less than 0.24 and the RMSE values are less than 30 cm (except for 46 cm at Saint John), indicating that the 2-D circulation model based on the POM performs reasonably well in simulating tidal elevations over these regions. It should be noted that the RMSE at Saint John is relatively large (about 0.46 m), but the γ2 value is small (0.04), which is due to the fact that the maximum tidal amplitude at this position is large (about 4 m). The γ2 values are relatively higher (0.4–1.25) in Hudson Bay (station 6), Gulf of St. Lawrence (stations 7–10, 20–23), embayments in the coast of Gulf of Mexico (stations 49, 51, 52, 55, 57, 62, and 63), and some islands in the Caribbean Sea (stations 78, 86, 89, and 90).

Table 1. Locations of Tide Gauge Stations and Statistics of Sea Level Variabilitya
Station NameIndex math formulaσt math formulaσs
  1. a

    Here math formula and math formula are γ2 used to quantify model performance in simulating tidal surface elevation and storm surges, respectively; σt and σs are the RMSE (m) used to quantify model performance in simulating tidal surface elevation and storm surges, respectively. “–” means no observed data exists during simulation period (1979–2010).

Ilulissat10.310.320.280.06
Sisimiut20.030.170.470.07
Ammassalik30.010.040.460.11
Qaqortoq40.020.090.430.08
Qikiqtarjuaq50.080.080.600.07
Churchill60.680.530.370.10
Cap-aux-Meules70.200.090.430.06
Riviere-au-Renard80.220.200.510.07
Sept-Iles90.310.400.370.07
Rimouski100.560.740.350.06
Saint John110.040.460.490.06
Yarmouth120.010.140.360.06
Halifax130.020.070.390.05
North Sydney140.240.070.380.06
Port Aux Basques150.280.220.320.07
St. Lawrence160.280.220.350.07
Argentia170.270.260.420.06
St Johns180.660.270.310.07
Nain190.310.230.270.06
Charlottetown200.700.540.320.06
Shediac Bay210.510.500.290.08
Lower Escuminac220.450.190.370.07
Belledune230.570.390.410.07
Eastport240.050.240.360.05
Bar Harbor250.020.230.360.05
Portland260.090.300.330.06
Boston270.110.330.320.06
Woods Hole280.240.090.290.05
Montauk290.130.080.270.06
New London300.110.090.270.06
Bridgeport310.030.120.270.07
Kings Point320.050.200.290.08
Sandy Hook330.040.110.250.07
Atlantic City340.030.080.270.07
Lewes350.050.100.290.08
Kiptopeke360.020.040.320.07
Duck370.030.060.350.08
Wrightsville Beach380.040.080.650.09
Springmaid390.040.080.520.09
Charleston400.080.160.460.09
Fort Pulaski410.040.140.440.09
Mayport420.040.090.450.08
Trident Pier430.030.060.570.08
Virginia Key440.170.090.830.07
Key West450.540.120.700.05
Naples460.490.180.460.06
Clearwater Beach470.220.110.330.06
Cedar Key480.250.160.310.08
Apalachicola490.470.120.310.07
Panama City500.250.070.300.06
Pensacola510.700.120.360.06
Dauphin Island520.710.100.380.07
Pilots Station East530.340.050.520.07
Grand Isle540.390.070.520.07
Lawma550.760.130.450.08
Freshwater Canal560.120.070.400.11
Calcasieu Pass570.490.120.460.11
Galveston580.140.060.530.10
Uscg Free Port590.110.100.440.09
Port Isabel600.180.070.770.08
Veracruz610.050.040.640.07
Ciudad del Carmen620.980.140.880.09
Progreso631.010.300.580.10
Belize64
Puerto Cortes65
Cochino Pequeno660.310.040.890.05
Puerto Castilla67
Limon680.310.050.820.05
Cristobal690.260.050.630.04
Cartagena700.210.040.640.05
La Guaria710.110.040.730.05
Siboney720.020.020.410.05
Gibara730.050.040.590.04
Guantanamo Bay740.140.040.910.04
Settlement Point750.020.040.490.04
Nassau76
Exuma770.050.050.860.04
South Caicos781.250.600.600.08
Port Kingston79
Lime Tree800.150.030.660.04
Magueyes Island810.100.020.650.04
Mona Island820.140.030.580.05
Charlotte Amalie830.100.020.600.03
San Juan840.070.030.640.04
Barbuda850.110.020.620.03
Pointe-a-pitre861.120.260.470.05
Le Robert870.550.080.460.04
Bridgetown880.350.110.600.06
Port of Spain891.120.260.560.05
Point Fortin901.010.330.670.04

[32] Figure 6 presents scatterplots of observed and simulated amplitudes and phases of M2 and K1 tides at 90 tide gauge stations. Overall, the 2-D model is capable of reproducing the observed amplitudes of M2 (γ2 ≈ 0.09) but performs less well the observed amplitude of K1 (γ2 ≈ 0.55) at these stations. For the tidal phase, the 2-D model performs better in simulating the observed phases of K1 (γ2 ≈ 0.06) than M2 (γ2 ≈ 0.14). It should be noted that the performance of the 2-D model in simulating tides differs spatially over the coastal waters and shelf seas of the ECSNA. The 2-D model reproduces reasonably well the observed M2 amplitudes at the 90 stations, except for those in the semienclosed seas and coastal waters, such as the Gulf of St. Lawrence (station 6, 9, and 10), Bay of Fundy (stations 25–27), Gulf of Mexico (station 48), along Cuba Island (station 78), and Eastern Caribbean (stations 89 and 90), with absolute differences ranging from 0.21 to 0.66 m. The 2-D model also performs less well in simulating the observed M2 phases at these stations mentioned earlier, with absolute differences ranging between 32° and 67°. The 2-D model overestimates the observed K1 amplitudes by 0.01–0.1 m at most of the 90 stations. Furthermore, the 2-D model reproduces less well the observed K1 phases with relatively larger absolute differences compared with observed values in the Gulf of St. Lawrence (stations 7, 20, and 21), Gulf of Mexico (stations 44, 49, 51, and 52), Mexico coast (stations 63–65 and 67), and Cuba Island (station 78), and over the Southern Caribbean Sea (station 69), with absolute differences ranging from 15.6° to 50.1°. For the S2 and O1 tides (not shown), the 2-D model performs reasonably well in simulating their amplitudes and phases with the γ2 values less than 0.2 and 0.15, respectively. The deficiency of the 2-D model in simulating tidal amplitudes and phases over the semienclosed seas and coastal waters of the ECSNA is, as mentioned earlier, due mostly to the coarse model resolution used in this study, which does not resolve well the local topography around these stations or the channels connecting semienclosed seas with the deep ocean waters. Coastal topography and constricted channels can effectively damp the diurnal and semidiurnal tidal frequencies. Poor representation of local topography over these areas in the model will lead to relatively large errors in simulating the phase speed and transport of the tidal circulation.

Figure 6.

Scatterplots of observed and simulated (left) amplitudes and (right) phases of M2 and K1 at 90 tide gauge stations over coastal waters of the ECSNA. The phases are in degrees relative to the UTC standard time. The colored symbols represent locations of tide gauges marked in Figure 2.

[33] We next compare our tidal simulations using the POM with the results produced by TPXO-7.2 model. Figure 4 compares distributions of tidal amplitudes and phases produced by the TPXO-7.2 model and the POM. To quantify the performance of the two models, the γ2 and RMSE values are calculated for the two models at the 90 tide gauges. Results show that the POM performs slightly better than the TPXO-7.2 model in simulating amplitudes but slightly less well in simulating phases. For the tidal amplitudes, the POM performs similar or slightly better than TPXO-7.2 with similar γ2 and RMSE for the four principle tidal constituents. For the tidal phases, the RMSE values for the POM in simulating phases of the four tidal constituents are 2°–8° greater than TPXO-7.2 and with relatively higher γ2 values (γ2 values for our 2-D model are in the range between 0.05 and 0.15; the γ2 values for TPXO-7.2 model are in the range between 0.02 and 0.13).

3.2. Storm-Surge Simulation

[34] We next examine the performance of the 2-D circulation model in simulating storm surges over the ECSNA. Figure 7 presents hourly time series of observed and simulated storm surges at six selected tide gauge stations in years 1989, 1998, 2003–2005, and 2008. The 2-D model reproduces reasonably well the observed surge-induced sea level variations associated with hurricanes and winter storms at the six stations in selected years, with the γ2 values, RMSE, and AAD less than 0.6, 0.13 m, and 0.1 m, respectively. For James Bay, only short-term observations at four tide gauge stations inside James Bay and adjacent areas are available for validations. The 2-D model performs fairly in simulating storm surges in James Bay and adjacent waters (not shown), with the values of γ2, RMSE, and AAD less than 0.6, 0.18 m, and 0.15 m, respectively.

Figure 7.

Time series of observed (red) and simulated (blue) surge-induced sea levels at Charleston (40), Sandy Hook (33), Duck (37), Charlottetown (20), Cedar Key (48), and Calcasieu Pass (57). The gray shaded vertical bars indicate hurricane events.

[35] The γ2 and RMSE values for storm-surge simulations in exp-BR at 90 tide gauge stations are also listed in Table 1. The RMSE values for simulating storm surges are generally less than 0.l m at the 90 stations. At tide gauges over coastal waters of the Scotian Shelf (stations 12–14), the Gulf of Maine (stations 24–27), and eastern United States (stations 28–37), the γ2 values are less than 0.4. At tide gauges along the northeastern coast of the Gulf of Mexico (stations 47–52), the γ2 values are in the range of 0.30–0.38. These results indicate that the 2-D circulation model performs reasonably well in simulating storm surges over these regions.

[36] It should be noted that the 2-D circulation model performs less well in reproducing the observed storm surges at tide gauges in estuaries, harbors, small bays, and islands. For example, at tide gauges of the St. Lawrence Estuary (stations 7, 8, 23) and the Bay of Fundy (station 11), the γ2 values are from 0.41 to 0.51. The γ2 values are in the range of 0.40–0.83 at stations 38–39, 43–44, and 53–58; and in the range of 0.46–0.86 at tide gauges of the islands in the Gulf of Mexico (station 45) and the Caribbean Sea (stations 75–90). In addition, the γ2 values at tide gauges over coastal waters of central America (stations 61–69), Columbia (station 70), Venezuela (station 71), and Cuba (stations 72–74) are also relatively large (0.41–0.88).

[37] We next examine the model performance in simulating storm-surge events in terms of the maximum surge, the surge duration, and the maximum surge time. The maximum surge is defined as the maximum sea level forced by the atmospheric forcing (i.e., the combination of wind stress and sea level atmospheric pressures) during a storm-surge event. The surge duration is defined as the period (in hours) between the beginning and end of the storm-surge event. The beginning and end of a storm-surge event are defined as the time when the surge-generated sea level reaches 15% of the maximum surge before and after the maximum storm surge, respectively. The maximum surge time is defined as the hour (0–23) of the day when maximum storm surge occurs. The comparison is restricted to the region with latitudes between 24°N and 50°N over the ECSNA for three reasons. First, the sea level observations at tide gauge stations at latitudes lower than 24°N are very sparse, especially over coastal waters of Caribbean Sea and Cuba. Second, existing observations at latitudes lower than 24°N demonstrate that surge-generated sea level variability is not large in this region. Third, regions at latitudes higher than 50°N are mainly influenced by large-scale winter storms, which are represented reasonably well by the CFSR forcing.

[38] Figures 8 and 9 present the scatterplots of observed and simulated maximum surges, maximum surge durations, and maximum surge times at the 53 tide gauge stations from 1979 to 2010 in four numerical experiments. In exp-BR (Figure 8a), the 2-D circulation model has the satisfactory accuracy in simulating the maximum storm surges with the correlation, RMSE, and AAD values of about 0.823, 0.159 m, and 0.124 m, respectively.

Figure 8.

Scatterplots of observed and modeled maximum surges at tide gauge stations with latitudes between 24°N and 50°N, at which sea levels are directly influenced by land-falling tropical storms or hurricanes from 1979 to 2010. The colored symbols represent locations of tide gauges marked in Figure 2.

Figure 9.

Scatterplots of observed and modeled (a) surge durations and (b) maximum surge time at tide gauge stations with latitudes between 24°N and 50°N, at which sea levels are directly influenced by land-falling tropical storms or hurricanes from 1979 to 2010. The colored symbols represent locations of tide gauges marked in Figure 2.

[39] To determine whether the model systematically underestimates or overestimates the observations, a linear slope is calculated from the scatterplot shown in Figure 8 using the least squares fitting. The 2-D model underestimates (overestimates) systematically the observations if the linear slope is much smaller (larger) than 1.0. Figure 8 demonstrates that the linear slope is near the unity and about 0.96 and 0.94 in exp-BR and exp-NT, respectively. In these two experiments, the parameterized vortex is inserted into the newly generated CFSR fields. In comparison, the linear slope is much smaller than the unity and about 0.72 and 0.70 in experiments exp-NH and exp-CFO, respectively. This indicates that the 2-D circulation model does not have the underestimation/overestimation problem in generating the maximum storm surges in exp-BR and exp-NT. By comparison, the 2-D model systematically underestimates the observed maximum storm surges in exp-NH and exp-CFO in which the parameterized vortex is not used. The linear slope, correlation values, RMSE, and AAD are very similar in experiment exp-BR and exp-NT, indicating the inclusion of the nonlinear tide-surge interaction in the model does not improve significantly the model performance due to the coarse model resolution used in this study. Table 2 lists the linear slopes in simulating the maximum surges in exp-BR and exp-NH at stations over seven regions between 24°N and 50°N over the ECSNA. In exp-NH in which the model is forced by the CFSR fields, the linear slope is about 0.83 in the GoM and relatively smaller over other six regions. In exp-BR in which the parameterized vortex is inserted into the CFSR fields, the linear slopes are close to the unity over these seven regions, with the highest value of 0.98 over the east coast of the United States and the smallest of 0.89 over the Gulf of St. Lawrence. Therefore, the regional difference in the model performance is most likely due to the accuracy of the atmospheric forcing, rather than the bottom friction and gridded topography parameterization used in the 2-D model.

Table 2. Linear Slopes in Different Regions Calculated From Data Points Show in Figure 8 for exp-BR and exp-NHa
Regionsexp-BRexp-NH
  1. a

    Abbreviations are used for the Gulf of St. Lawrence (GSL), eastern Canadian shelf (ECS), Gulf of Maine (GoM), Long Island (LI), east coast of the United States (UEC), western Florida shelf (WFS), and northern Gulf of Mexico (NGOM).

GSL0.890.70
ECS0.930.70
GoM0.920.83
LI0.930.71
UEC0.980.78
WFS0.950.70
NGOM0.960.71

[40] Figure 9a presents the scatterplot of simulated and observed storm-surge durations in exp-BR. In this case, the correlation coefficient is about 0.48, the linear slope is about 0.72, and AAD is 11.63 h. The model significantly underpredicts the observed surge durations in several storm events landing in the east coast of the United States, Long Island, western Florida shelf, and northern Gulf of Mexico. Figure 9b presents the scatterplot of simulated and observed maximum surge times in exp-BR. The 2-D model reproduces reasonably well the observed maximum surge times over the ECSNA, with the averaged AAD of about 2.0 h. The 2-D model performs less well in predicting the maximum surge times in some storm events in the northern Gulf of Mexico, western Florida shelf, Long Island, and Gulf of Maine with the averaged AAD of 3–6 h. The model performances in exp-NH, exp-NT, and exp-CFO are very similar to that in exp-BR in reproducing the surge durations and the maximum surge times (not shown). It should be noted that the parameterized vortex represents less well the wind and SLP profiles of some tropical storms and hurricanes, particularly during their landfall processes. As a tropical storm or hurricane approaches the land and makes its landfall, the storm has significant structure deformation with rapid variations in the wind and atmospheric pressure fields. This severe deformation could not be resolved well by the parameterized vortex used in this study. Less realistic representation of the storm structure particularly during the landfall certainly affects the performance of the model in simulating storm surges as expected.

[41] As shown earlier, the insertion of the parameterized vortex into the CFSR fields improves the model accuracy in simulating the maximum storm surges. More accurate simulations of the surge durations and maximum surge times can be made if there is more reliable information about the wind and SLP profiles and translation of hurricanes or tropical storms.

4. Estimation of Extreme Surge-Induced and Total Sea Levels

[42] The following five steps were taken in this study to estimate the extreme surges and extreme total sea levels of the 50 year return period from model results over the ECSNA. First, the surge-induced sea levels were extracted directly (without any filtering) from the simulated total sea levels produced by the 2-D circulation model in exp-BR. Second, the annual maximum surges of 32 years at each model grid were computed from time series of simulated surge-induced sea levels. Third, the location and scale parameters of the Type I distribution at each grid were estimated by using the extremal analysis technique discussed in section 2. Fourth, the 50 year return levels of the extreme storm surges were calculated. Finally, the 50 year return levels of the extreme total sea levels were determined from simulated tidal and surge-induced sea levels using the extremal analysis technique and the MCM.

4.1. Extreme Surge-Induced Sea Levels With the 50 Year Return Period

[43] Figure 10a presents the extreme surge-induced sea levels with the return period of 50 years estimated from the simulated surge-induced sea levels (or storm surges) at 32 tide gauge stations over coastal waters of the ECSNA. At these 32 stations, the record lengths of the sea level observations are longer than 30 years in the period 1979–2010. The modeled extreme surge-induced sea levels of the 50 year return period are relatively large and about 1.0 m or higher at many tide gauge stations over coastal waters of the northwestern Atlantic including stations at Churchill (station 6), Riviere-au-Renard (station 8), Sept-Iles (station 9), Rimouski (station 10), Charlottetown (station 20) Lower Escuminac (station 22), Boston (station 27), Sandy Hook (station 33), Lewes (station 35), Duck (station 37), and Galveston (station 58). The 50 year extreme surge-induced sea levels are relatively small and about ∼0.2 m at San Juan (station 84) and ∼0.3 m at Key West (station 45). Figure 10a also shows the observed 50 year extreme surge-induced sea levels estimated from the filtered and unfiltered sea level measurements using the same procedure discussed earlier. The spatial variability of the modeled extreme surge-induced sea levels are consistent with those estimated from the filtered sea level observations made at these 32 stations (Figure 10a). The RMSE is about 0.118 and the AAD is about 0.086 m.

Figure 10.

(a) The 50 year extreme surge-induced sea levels calculated from model results in exp-BR (light gray bars) and from sea level measurements (red bars for unfiltered observations, blue bars for filtered observations) at tide gauge stations with hourly time series longer than 30 years in the period 1979–2010. AAD and RMSE are statistic between filtered observations and model results. (b) Differences in the 50 year extreme surge-induced sea levels between exp-BR surge and exp-NT.

[44] Large differences ranging from 0.2 to 0.6 m, however, occur between the modeled and observed extreme surge-induced sea levels estimated from the unfiltered observed sea levels at most of the tide gauge stations over the ECSNA (Figure 10a). There are four plausible reasons for these large differences. First, the less accurate BST data set was used (section 3.2). Second, the model horizontal resolution is too coarse (section 3.2). Third, the wave-tide-surge interaction over some coastal waters of the ECSNA may not well be resolved by the 2-D circulation model. Four, the 2-D circulation model do not generate the annual (Sa), semiannual (Ssa), and interannual (Zo) variabilities of sea levels.

[45] Previous studies demonstrated that the nonlinear tide-surge interaction has significant contributions to storm surges over some coastal waters of the ECSNA. Bernier and Thompson [2007] demonstrated that the tide-surge interaction is very strong in the Northumberland Strait, and its amplitude can reach up to 0.2 m during the storm event. Rego and Li [2010] demonstrated that nonlinear tide-surge interaction can reach up to 70% of the tidal amplitude during Hurricane Rita in the Gulf of Mexico. In this study, the differences in the extreme surge-induced sea levels of the 50 year return period between exp-BR and exp-NT at 32 tide gauge stations are used to quantify the role of the tide-surge interaction in the extreme surge-induced sea levels over the coastal waters of the ECSNA (Figure 10b). Our results demonstrate that the tide-surge interaction makes positive contributions to the extreme surge-induced sea levels over the St. Lawrence Estuary, Northumberland Strait, Gulf of Maine, and the Pamlico Sound. By comparison, the tide-surge interaction has negative contributions at Lower Escuminac (station 22), Woods Hole (station 28), Montauk (station 29), and New London (station 30). It should be noted that due to the coarse model resolution used in this study, the nonlinear tide-surge interaction could be significantly underestimated by the model. More accurate simulations of tide-surge interactions over the coastal waters of the ECSNA remain to be done using a high-resolution numerical model.

[46] Figure 11 presents spatial distributions of extreme surge-induced sea levels of the 50 year return period computed from simulated surge-induced sea levels over the ECSNA using the extremal analysis. Over the deep waters of the ECSNA, the extreme surge-induced sea levels are relatively large at latitudes between 40°N and 70°N and relatively small at latitudes lower than 35°N. Over the coastal waters of the ECSNA, by comparison, the extreme surge-induced sea levels are relatively large, particularly over southern Hudson Bay, the western and southwestern Gulf of St. Lawrence, the coastal waters along the east coast of the United States from Long Island to Pamlico Sound, and coastal waters of the northern Gulf of Mexico. It should be noted that the spatial patterns of the extreme surge-induced sea levels in the deep waters of the ECSNA shown in Figure 11 are very similar to the patterns of extreme wave heights and extreme wind speeds [Young et al., 2012].

Figure 11.

Distribution of 50 year extreme surge-induced sea levels (storm surges) over the ECSNA estimated from surge-induced sea level fields produced by the 2-D circulation model in exp-BR using the generalized extreme value statistical approach.

[47] The 50 year extreme surge-induced sea levels reach about 2.2 m over southern James Bay (Figure 11), which are among the largest extreme values of surge-induced sea levels over the ECSNA, but they could not be verified directly by the observations due to the lack of long records of sea level measurements in James Bay. Nevertheless, our estimates of extreme surge-induced sea levels in James Bay are consistent with previous studies by Manning [1951] and Godin [1975]. These two studies demonstrated that strong autumn wind storms can cause large storm surges of greater than 1.2 m over southern James Bay, and sometimes storm surges can extend kilometers inland beyond the normal high water mark. Furthermore, Stewart and Lockhart [2005] suggested that hurricanes and winter storms cross Hudson Bay can cause large surge-induced sea levels (storm surges) in James Bay and Hudson Bay. The shallow water depths (∼40 m) and semienclosed coastline in James Bay can further intensify the storm surges.

[48] The 50 year extreme surge-induced sea levels are in the range of 1.3–2.0 m along the east coast of the United Sates from Long Island Sound to Pamlico Sound, with the maximum values of ∼2.0 m over the western part of Long Island Sound, upper part of Delaware Bay, and Pamlico Sound. The extreme surge-induced sea levels are also large and in the range of 1.3–2.0 m along the northern coast of the Gulf of Mexico (from Galveston to Dauphin Island) and over the northern part of western Florida shelf and the area around Mississippi river mouth.

[49] Over the Gulf of St. Lawrence, the extreme surge-induced sea levels for the 50 year return period increase gradually from east to west and reach maximum values of 1.25 and 1.4 m in the St. Lawrence Estuary and Northumberland Strait, respectively. Our results in simulating the surge-induced sea levels in the northwest Atlantic are similar to results produced by Bernier and Thompson [2006]. It should be noted that the CFSR fields used in the present study include six-hourly sea level pressure fields, which are dynamically more consistent than fields estimated from the atmospheric environment service (AES) data set used by Bernier and Thompson [2006]. Furthermore, the combination of CFSR atmospheric fields and a parameterized vortex used in this study is a better representation of the atmospheric forcing during hurricane events than the forcing used by Bernier and Thompson [2006].

4.2. Extreme Total Sea Levels With the 50 Year Return Period

[50] The MCM was used to estimate the total sea levels associated with tides and storm surges (section 2.3). One important condition for the use of the MCM is that this method is applied only to the regions where the tidal elevations are dominant. We follow Oliver et al. [2012] and use the ratio (α) of the standard deviation of tidal elevations to that of the surge-induced elevations to identify the tidally dominant regions over the ECSNA. The α ratio at each model grid is calculated from the surge-induced and tidal sea levels produced by the 2-D circulation model in exp-NT and exp-TO, respectively.

[51] Figure 12 presents the distribution of the α ratios over the ECSNA. The tidal elevations are important over the most of the ECSNA except for the Caribbean Sea, the tropical North Atlantic at latitudes of about 20°N, deep waters centered at (49°N, 39°W) and central part of Hudson Bay and James Bay. The tidal elevations, in particular, are nearly 5 to 6 times larger than the surge-induced elevations in the Bay of Fundy, Hudson Strait, and northwestern Gulf of St. Lawrence. Over western Greenland, the tidal elevations are 3 to 4 times larger than the surge-induced elevations.

Figure 12.

Ratios (α) of the standard deviation of tidal surface elevations to that of the surge-induced surface elevation. The black contour line denotes the cutoff ratio of 1.1.

[52] The MCM was applied only to the regions where the α ratios are greater than 1.1. This cutoff of the ratio was chosen based on the consideration that tides play a very important role over the region with the ratio greater than this cutoff.

[53] Figure 13 presents the map of the extreme total sea levels with the 50 year return period. Regions over the ECSNA experiencing severe 50 year extreme total sea levels are similar to those regions of 50 year extreme surge-induced elevations or storm surges, but with much higher extreme values. In Hudson Bay, for example, the 50 year extreme total sea levels reach a maximum value of ∼6 m in the Hudson Strait, between 1.7 and 3 m over southern James Bay and between 1.5 and 2.6 m over coastal waters of southern Hudson Bay. In the Gulf of St. Lawrence, the 50 year extreme total sea levels reach a maximum value of ∼4 m in the St. Lawrence Estuary and smaller values of ∼2 m in Shediac Bay and the Northumberland Strait. In the Gulf of Maine, the extreme total sea levels reach a maximum value of ∼6 m in the Bay of Fundy and ∼2 m over the central Gulf region.

Figure 13.

Distribution of 50 year extreme total sea levels (due to the combination of storm surges and tides) over the ECSNA estimated from surge-induced and tidal sea level fields produced by the 2-D circulation model in exp-NT and exp-TO using the generalized extreme value statistical approach and the Monte Carlo method.

[54] The 50 year extreme total sea levels reach a maximum value of nearly 3 m in Long Island Sound and between 1.3 and 2.5 m over other coastal waters along the east coast of the United States (Figure 13). The 50 year extreme total sea levels are in the range of 1.4–1.9 m over the coastal waters of the western Florida shelf and in the range of 1.2–1.5 m over the northern Gulf of Mexico.

[55] To assess the accuracy of the 50 year extreme total sea levels estimated from the model results, we calculate the 50 year extreme total sea levels from the tide gauge observations at 32 stations using the MCM and compare them with their counterparts computed from the model results (Figure 14a). The modeled extreme total sea levels of the 50 year return period are largest at stations over the Bay of Fundy (Saint John (11) and Eastport (24)), with the extreme total sea levels of 4.61 and 3.89 m, respectively. The 50 year extreme total sea levels are nearly 3 m at stations in Hudson Bay, the St. Lawrence Estuary, and the Gulf of Maine, including stations at Churchill (6), Sept-Iles (9), Rimouski (10), Yarmouth (12), Bar Harbor (25), Portland (26), and Boston (27). The modeled extreme total sea levels at stations over the Scotian Shelf, Newfoundland, and along the eastern coast of the United States have relatively lower values ranging from 1.2 to 2 m. The spatial variability of the modeled extreme total sea levels calculated from model results is consistent with those estimated from the total sea level observations made at these 32 stations (Figure 14a). The RMSE is about 0.324 m and the AAD is about 0.245 m. A comparison of the estimated 50 year extreme total sea levels based on model results to these based on the observations without the use of the MCM is shown in Figure 14b. The RMSE (0.371 m) and AAD (0.283 m) for model-data comparisons in Figure 14b are highly comparable for these values in Figure 14a.

Figure 14.

The 50 year extreme total sea levels calculated from model results of (a) exp-NT and exp-TO using MCM (light gray bars in Figure 14a) and (b) exp-BR without the use of MCM (light gray bars in Figure 14b) and from sea level measurements (black bars) at tide gauge stations with hourly time series longer than 30 years in the periods 1979–2010 (a) with or (b) without the use of MCM.

[56] Figure 14a shows that relatively large differences of about 0.2–0.5 m occur between modeled and observed extreme total sea levels at several tide gauge stations over the ECSNA. Besides the model deficiency in storm-surge simulations (section 3.2), less accurate performance of tidal simulations due to coarse resolution of the model at embayments and harbors (section 3.1) also contributes to the deficiency between modeled and observed extreme total sea levels.

5. Summary and Conclusion

[57] The extreme sea levels with the 50 year return period over the ECSNA were estimated from simulated sea level fields produced by a 2-D circulation model using the GEV approach and the MCM. The total sea levels were decomposed into the surge-generated and tidal components. These two components were simulated separately by the 2-D circulation model with the atmospheric and tidal forcing. The atmospheric forcing is the combination of the wind stress and sea level atmospheric pressures (SLP) extracted from the NCEP CFSR fields at 6 h intervals and a parameterized vortex suggested by Holland [1980]. The latter was used to better represent the wind and atmospheric pressure profiles associated with a tropical storm or hurricane.

[58] Efforts were made in this study to assess the model accuracy in simulating surge-induced and tidal components of sea levels using tide gauge observations over the ECSNA and the previous model results. It was found that the 2-D circulation model with a horizontal resolution of (1/16)° has reasonable skill in simulating tides and storm surges over the study region. Using the combination of the parameterized vortex suggested by Holland [1980] and the atmospheric forcing taken from the newly generated CFSR fields, the 2-D circulation model reproduces reasonably well the maximum storm surges but slightly less well the surge durations and the times at which the maximum surges occur. This indicates that more realistic atmospheric forcing than the Holland's parameterized vortex is needed in order to improve the model accuracy in simulating storm surges during a tropical storm or hurricane.

[59] The 50 year extreme surge-induced sea levels over the ECSNA were estimated from the simulated surge-induced sea level (i.e., storm surges) fields in the 32 year period (1979–2010) using the GEV approach. The most potential regions to be threatened by the 50 year extreme storm surges include the northern coast of the Gulf of Mexico and western Florida shelf, coastal waters along the east coast of the United States from Long Island Sound to Pamlico Sound, the southern part of Hudson Bay and James Bay, and the western Gulf of St. Lawrence. The 50 year extreme total sea levels over the ECSNA were estimated from the simulated surge-induced and tidal sea levels produced by the 2-D model using the MCM. Results show that the most potential regions to be threatened by 50 year extreme total sea levels are similar to those regions of 50 year extreme surge-induced sea levels, but with much higher extreme values.

[60] The results presented in this study are very useful in identifying potential regions of the ECSNA to be threatened by extreme storm surge and extreme total sea levels. More studies are needed, however, to improve the accuracy of extreme sea level estimations. First, the 2-D circulation model can be run at a higher resolution with unstructured [Rego and Li, 2010] or nested-grid model [Shan et al., 2011] to better resolve tide-surge interaction and other physical processes. In addition, a 3-D baroclinic circulation model and an ocean wave model [Xie et al., 2008] should be used in the future study. Finally, wind forcing and SLP fields with higher spatial and temporal resolutions in the past and the future are required, not only to better represent the general structure of a tropical storm or hurricane during the landfall but also to assess the potential risk caused by extreme sea levels due to the climate change [Marcos et al., 2011].

Acknowledgments

[61] The authors wish to thank Shiliang Shan, Eric Oliver, Guoqi Han, and Keith Thompson for their contributions. This research is funded by Lloyd's Register Foundation (LRF) and the Natural Sciences and Engineering Research Council of Canada (NSERC). LRF is a UK registered charity and sole shareholder of Lloyd's Register Group Ltd, which invests in science, engineering, and technology for public benefit, worldwide. The authors thank three anonymous reviewers for their constructive suggestions.