A Coupled Wave-Surge Modeling Study in Rhode Island Coastal Waters

This thesis is comprised of two manuscripts, both of which involve investigating the sensitivity of storm surge in Rhode Island coastal waters. The first study details the effect of wave-induced enhanced bottom friction on surge for a simple case study. The second study sheds light on the impact of using different hurricane wind models to simulate storm surge and waves. The interaction of waves and circulation (tide and surge) is characterized by the effect of storm surge and currents on waves, and the effect of waves on storm surge and currents. Quantifying this effect for a given area may be important for storm surge prediction purposes. As a result of wave-induced near-bed orbital velocities, the bed roughness will increase for storm surge propagation. Here, a sensitivity analysis was performed for Rhode Island coastal waters. A method developed by Soulsby (2006) was implemented to compute the increased bottom friction (i.e. drag coefficient) due to the effect of waves. Further, the interaction between waves and currents are incorporated in a coupled ADCIRC+SWAN model (which does not have this process). The storm surge was simulated with and without considering the impact of waves on the bottom roughness. Preliminary results indicate that enhanced bottom friction is largest in wavedominant areas, compared with areas where currents are strong. In other words, if the wave induced shear stresses are higher than the current induced bed shear stresses, the bottom friction increases significantly. A case study for Hurricane Irene (2011) shows that although the effect is considerable on increasing the Manning coefficient, storm surge is not that sensitive to enhanced bottom friction. The second study deals with the effect of wind models on storm surge. Storm surge and wave models are routinely used to assess the impact of hurricanes/cyclones for emergency preparedness. While these models are forced by wind fields, generated by meteorological models in hindcast or forecast mode, selecting a wind model which can accurately resolve the wind field, especially near the hurricane/cyclone core, is a challenging task. We use several wind hindcast models to force a coupled wave and storm surge model for selected hurricanes, including Bob (1991), Irene (2011) and Sandy (2012). The resulting simulated storm surge and wave parameters are compared to observations. Thewindmodels include the EuropeanCenter forMedium-RangeWeather Forecasts (ECMWF), the Northeast Coastal Ocean Forecasting System (NECOFS) based on the Weather Research and Forecasting (WRF) model, and parametric wind based on National Hurricane Center (NHC) datasets. Storm surge and waves are best predicted using ECMWF wind for Hurricane Irene, parametric wind for Hurricane Bob, and NECOFS WRF winds for Hurricane Sandy. Our results show that a wind model, which has an error in peak wind speed of less than 20% when compared with observations, could lead to convincing storm surge of wave predictions. The impact of using a poor wind model can result in error as high as 50% in storm surge and wave predictions. There is no unique "best" wind model for all hindcast applications. This choice depends on the nature of the hurricane, in particular, the ability to adequately characterize the spatial structure of the wind forcing field. Therefore, storm track and storm scale should be considered in selecting a wind model.


LIST OF TABLES
Effect of wave-induced enhanced bottom friction on storm surge

Introduction
The interaction of waves and circulation (tide and surge) is characterized by the effect of storm surge and currents on waves, and the effect of waves on storm surge and currents. Fro example, in the presence of currents, the frequency of waves will either decrease (with current), or increase (against current) to a stationary observer, which is known as the Doppler shift. In return, waves induce a force on the circulation in the form of radiation stresses, affecting water levels and currents, especially nearshore (wave set-up). Additionally, several studies have shown that the bottom roughness increases for currents due to the interaction of waves within the bottom boundary layer (see [1,2,3]).
Advances in numerical modeling and unstructured meshes have provided the ability to examine wave-current interactions in areas of complex geometry. The national network of operational forecast systems employed by the National Oceanic and Atmospheric Administration (NOAA; tidesandcurrents.noaa.gov) utilize several three-dimensional circulation and wave models such as the "Regional Ocean Modeling System" (ROMS), the "Finite-Volume Community Ocean Model" (FVCOM), and the "Semi-implicit Eulerian- In recent years, the influence of waves on storm surge and surge on waves has been a topic of discussion in the scientific community. Huang et al. (2010;[6]) investigated wave-surge interactions for a hypothetical hurricane in the Tampa Bay, Florida area using FVCOM for circulation and SWAN for waves. Considering the complex bathymetry in the region, the effect of storm surge on waves was more significant than the influence of wave-induced forces on surge. In the northeast US, Sun et al. (2013;[7]) performed a case study for Hurricane Bob (1991), and noted that the contribution of wave-induced forces improved the surge simulation along the southern shore of Rhode Island by 10-25 cm utilizing the coupled current-wave model FVCOM-SWAVE, where SWAVE is a modified version of SWAN for implementation over an unstructured grid.
The nonlinear interaction between the wave and current boundary layers at the sea floor induces a change in the bed shear stress. In the presence of waves, added turbulence in the wave bottom-boundary layer causes what appears to be an increase in the bottom roughness to the current. The bottom-boundary layers are represented by the currentonly and wave-only bed shear stresses, which are controlled by the near-bed current and wave orbital velocities, respectively. The expression for calculating near-bed wave orbital velocity is of most importance because of its contributions to the wave-induced shear stress in the nonlinear interaction. Soulsby proposed a method for calculating orbital velocity beneath waves in 1987 [8] given wave height, period and water depth, which was further summarized and compared with several additional methods in 2006 [9]. Not included in [9] are studies by Wilberg et al. (2008;[10]), Elfrink et al. (2006;[11]) and You (2009;) in which wave orbital velocities are estimated, derived, and statistically distributed, respectively, in nearshore regions where field measurements were available for comparison.
In addition, the effect of wave-current interactions on bed shear stress has been in-vestigated in recent years. When comparing two different expressions of bottom friction dissipation in a coupled wave-surge simulation, Rosales et al. (2008;[13]) found that the maximum bottom stress was doubled with inclusion of wave-current interactions, concluding the selection of such an expression is significant in determining the shear stress. Bing-chen and Hau-Jin (2007;[14]) introduced wave-enhanced bottom shear stress based on a form of Grant and Madsen [15] into a coupled hydrodynamic-wave model (COHERENS-SWAN), and found that the inclusion of random waves produced a lower bottom shear stress than the case without random waves, highlighting the differences between one way and two way interaction of waves and currents. Both Huang (2010) and Sun (2013) compared the results from a coupled wave-circulation model with those of a current-only and/or wave-only model to assess the role of wave-current interaction in storm surge simulation.
The present study focuses on (1)  In addition, a case study is presented to quantify the effect of the enhanced bottom drag coefficient on storm surge prediction using the coupled ADCIRC+SWAN model.
Formulations proposed by Soulsby (1993;[1]) in determining the enhanced bottom friction are described in Section 1.2, along with a description of the assumptions we made and the model settings we used. Section 1.3 illustrates the relationships between wave climate, near-bed wave orbital velocity, and other location-specific conditions, as well as their contributions to the enhancement of bottom friction. A discussion and summary are provided in Sections 1.4 and 1.5.

Formulation of enhanced bottom friction
As the interaction of the wave and current boundary layers leads to the enhancement of the bed shear stress, the friction force in the momentum equation changes, which can lead to a change in water elevation and storm surge. In terms of the bottom boundary layer, the effects of wave-current interaction have been studied in previous research with respect to sediment transport applications (see [1,2,3,16]). This study focuses on the sensitivity of the bottom friction to such interactions, and its significance to storm surge prediction.
The enhancement of bottom friction in a given area is a result of the nonlinear interaction between the wave and current bottom-boundary layers. The amount of enhancement depends on the current-and wave-induced shear stresses, τ c and τ w , respectively, which can be quantified as a mean bed shear stress, τ m (Eq. 1.1; [2]) in the combined wave-current flow field. The mean bed shear stress was empirically derived from eight different bottom-boundary layer models; details of this formulation can be found in [2].
The mean (τ m ) and pure-current (τ c ) shear stresses are proportional to the square of the depth-averaged current velocity u c , as well as the bottom drag coefficient in the presence and absence of waves, C * D and C D , respectively (Eq. 1.2). Depth-averaged current velocities can be based on measured values from current meters or simulated values from a circulation model. The ratio of the combined wave-current bottom drag coefficient to pure-current drag coefficient, Eq. 1.1 can be rewritten as the measure of enhanced drag in a given area ( ): where λ is the ratio of the wave-induced shear stress to the current-induced shear stress, λ = τ w τ c . Intuitively, we know that in regions of weak currents (u c → 0), the current-induced shear stress is small (τ c → 0), where the same applies to the near-bed orbital velocity and wave-induced shear stress. From Equation 1.3, we can see that is primarily dominated by τ w , and therefore in cases of weak currents and strong waves, approaches a maximum value of 2.2. Thus the calculation of the near-bed root-meansquare (r.m.s.) wave orbital velocity U rms is of the utmost importance to determine the magnitude of the wave-induced bed shear stress: Here, f w is the friction factor, A is the semi-orbital wave excursion, k s is the Nikuradse bottom roughness (k s = 2.5d 50 ) for mean sediment grain size diameter d 50 , and T z is the zero-crossing period (T z = 0.781T p ; [2]), where T p is the peak wave period.
In addition to linear wave theory, the estimation of U rms has been approached in several studies (see [8,9,11,10,12]) in which near-bed orbital velocities can be approximated analytically using other measured wave parameters from observational buoys, theoretical wave parameters from an idealized wave spectra, or simulated surface wave parameters from a spectral wave model. Of these, the exponential approximation discussed by Soulsby (1987 and2006, [8, 9]) describes the near-bed r.m.s. orbital velocity as the variance of the bottom velocity spectrum based on a Joint North Sea Wave Project (JONSWAP) spectrum of a particular sea-state: with significant wave height H s , zero-crossing period T z , water depth h, and gravitational acceleration g. Details of this formulation can be found in [8].
Lastly, the combined wave-current and pure current bottom drag coefficients can be related to other friction factors such as the Manning's n quadratic friction coefficient.
Since C D = gn/h 1/3 , the enhanced Manning coefficient is n * = n √ [17]. It should be noted that the process outlined here can be completed using measured/observed values of H s , T p , d 50 , u c , and U rms from wave buoys, current meters, etc., or by using simulated values from a circulation or wave model.

Sensitivity analysis
From the previous section, we know that the enhanced bottom friction is a function of depth-averaged current velocity, bottom roughness, water depth, near-bed r.m.s. wave orbital velocity, significant wave height, and peak/average wave period. If these govern- Depending on the sea state of a given area, the wave height and wave period can be dependent or independent. In the case of fully developed seas, the equation governing the growth of waves with fetch is described by Eq. 1.8 in the Coastal Engineering Manual [18] for peak period T p and energy-based significant wave height H mo . The one-to-one relation for fully developed seas is close to that derived from linear wave theory for peak period and significant wave height H s ( Fig. 1.3). However, a one-to-one relationship is not always the case. A wave period can also correspond to several wave heights and vice versa, which can be seen by plotting observed surface wave parameters from offshore buoys (see Fig. 1 [18]) and by linear wave theory (LWT) have also been plotted.
Accordingly, a range of wave heights and periods based on the wave climate in our region can be applied to the sensitivity of the enhancement of bottom friction via the near-bed r.m.s. orbital velocity. U rms is a function of significant wave height, zerocrossing period, and water depth (i.e. U rms (H s ,T z ,h)), which can be expressed using Eq.
1.7 given a range of realistic values. In doing so, the expected magnitude of near-bed orbital velocity can be estimated for any known depth, wave height or wave period1. This sensitivity analysis was completed for variable significant wave height (H s = 0 to 10 m), three zero-crossing periods (T z = 6, 10, 14 s), and variable water depth (h = 0 to 50 m).
A simple wave breaking criterion was applied assuming solitary waves, where H s does not propagate over depths less than 0.78h [2].
Lastly, the expected level of enhanced bottom friction (Eq. 1.3) for a given area can be estimated. A sensitivity analysis is performed for a range of feasible bottom 1The ability to estimate U r ms for a set of wave conditions is also important for the protection of submerged objects on the sea floor, such as Acoustic Doppler Current Profilers (ADCPs). roughness k s , current velocity u c , and near-bed r.m.s wave orbital velocity U rms values based on the study region. Sediment grain size diameters representing medium to very coarse sand (d 50 = 0.25, 0.5, 0.75, and 1.0 mm) are selected, which correspond to bottom roughness values of k s = 0.000625 m, 0.00125 m, 0.001875 m, and 0.0025 m, respectively (k s = 2.5d 50 ; [1]). Depth-averaged current velocity is varied from 0.1 m/s to 1.0 m/s, representative of values measured and/or estimated in the literature (see [19,20]). The near-bed r.m.s. wave orbital velocity varied from 0 to 1.5 m/s when based on varying wave climate, which agreed with velocities measured/estimated in [11,12].

Coupled wave-current model ADCIRC
The ADvanced CIRCulation hydrodynamic model solves the equations of motion for a kinetic fluid on a rotating body, formulated using traditional hydrostatic pressure and Boussinesq approximations that have been discretized in space by the Galerkin finiteelement method and in time by the three-level finite-difference method [4]. The surface water elevation is computed by the depth-integrated continuity equation found in the "General Wave Continuity Equation" (GWCE), and depth-integrated current velocity is computed from the non-conservative momentum equations [4]. ADCIRC is forced along the open ocean boundary primarily by water elevation and surface stress (i.e. wind).

SWAN
Simulation WAves Nearshore (SWAN) is a third-generation spectral model that computes wave conditions based on the conservation of the wave action density, which is conserved in the presence of ambient currents. SWAN takes into account wavecurrent interactions via radiation stresses in shallow water, and includes the formulations for wave generation by wind, nonlinear wave-wave interactions, and dissipation due to whitecapping, bottom friction, and depth-induced breaking [21]. Surface waves induced water particle motion through the water column to the sea floor, giving rise to friction in the turbulent bottom boundary layer. This dissipation of wave energy due to bottom friction depends on the near-bed r.m.s. wave orbital velocity U rms , in which SWAN provides its solution as a function of the energy density spectrum, water depth, and wave period (Eq. 1.9; [21]).

Coupled ADCIRC+SWAN model
As mentioned, the enhancement of bottom friction due to waves is not included in the present version of the ADCIRC+SWAN model. Here, we manually modified the spatially variable friction (n * = n √ ) of ADCIRC based on the wave field computed by SWAN. The process can be automated, but here we studied the sensitivity of storm surge to this effect.
The enhancement of bottom shear stress, and thus bottom friction, is a result of the interaction of the wave field with the current bottom boundary layer. Computation of wave-current interactions, including wave set-up and set-down, in coupled models [22] provides more accurate predictions of waves and storm surge, as concluded by Huang et al. (2010;[6]). The ADCIRC+SWAN coupled wave-circulation model is used in this study. The coupling of ADCIRC and SWAN is carried out in parallel on identical sub-meshes using intra-model communication on the same computational core [22], in which SWAN is treated as subroutine. At each node, SWAN is passed wind speeds, water levels, and currents computed by ADCIRC, which are averaged each time step and used to recalculate the water depth and related wave processes such as wave propagation and depth-induced breaking. In return, ADCIRC is partially driven by the radiation stress gradients computed by SWAN, extrapolating them forward in time [22].

Case study Study area
Rhode Island is comprised of several barrier systems along its southern shore, which include dunes, coastal ponds, headlands, and inlets, and are subject to moderate to severe coastal flooding during significant storm events. In order to preserve the domain extent for hurricane hindcast purposes and to provide enough discretization in the study area, the RI mesh was merged with the FVCOM GOM4 mesh, leading to a total 105,560 nodes. Figure 1.5 displays the combined mesh for New England and the higher resolution mesh for RI with corresponding bathymetry of the region.

Sources of data
High resolution bathymetric (30m) and topographic (1m) data for Rhode Island was acquired from the RI Geographical Information System (GIS) (www.rigis.org/) and applied to the ADCIRC domain around RI where the mesh has improved resolution.
Digital Elevation Models (DEMs) of Connecticut and Massachusetts, provided by each state's GIS database, were used to define the outer extents of the RI mesh. The bathymetry and topography of the remaining regions were based on the NECOFS original mesh (GOM4).
Available observed and hindcast data were reviewed in the region (focused in RI;

Hurricane Irene (2011)
Hurricane Irene (2011) formed from a tropical wave that exited the African coast on August 15, 2011, and strengthened over the Atlantic basin, leading to a destructive landfall in North Carolina as a strong Category 1 hurricane [24]. The storm continued to travel northward along the US east coast to the west of Rhode Island with a radius of maximum wind of 100 nautical miles or 185 km. RI experienced severe wind gusts that left much of the state without power for several days, and mild flooding in Narragansett Bay due to storm surge. Hurricane Irene was the first major storm to impact Rhode Island since Hurricane Bob in 1991. This storm was simulated using the "European Center for Medium-range Weather Forecasts" (ECMWF) wind model, a combined general circulation model and data assimilation system. More details about the ECMWF wind model can be found in [25,26]. The simulation took place for 7 days from August 21, 2011 00:00 to August 30, 2011 00:00 GMT. Time series of the simulated wave parameters and current velocity are compared with observations.

ADCIRC+SWAN Model Settings
The coupled ADCIRC+SWAN model is run on a CENT OS 7 Linux cluster with 60 processors and took 6.5 hours for a 6 day simulation. ADCIRC is run in 2-D mode  (2012)); [25].
An initial simulation was completed with Manning's n quadratic friction set to the default value (0.02). Post processing is performed on the wave parameters and current velocity from the initial run in order to calculate the increased Manning's n * coefficient for application in a second simulation. The water elevations are then compared between the initial and enhanced scenarios to assess the sensitivity of storm surge to enhanced bottom friction.

Model validation
In this study, simulated significant wave height and peak period from the wave model SWAN are applied in the formulation of enhanced bottom friction. Since observed surface wave parameters were not used directly, it is necessary to first compare the simulated values with those observed by a local wave buoy in Rhode Island. Figure   1.6b,d displays the time series of simulated H s and T p compared with observations from NDBC 44097 wave buoy for Hurricane Irene. We observe good agreement at the peak of the storm when comparing H s to observations, and moderate agreement between simulated and observed T p . In addition, contours (i.e. a heat map) of a snapshot (i.e. a single point in time) of the simulated mean and maximum H s and T p experienced in the  Torres et al. (2017;[25]) for this historical storm, which is also detailed in the second manuscript of this thesis.
Similarly, the simulated mean depth-averaged current velocity u c,sim for the study region is compared to the observed depth-averaged current velocity u c,ob recorded by Woods Hole Group during Hurricane Irene (WHG; [19] is compared to the U rms computed analytically from the exponential approximation described by Soulsby (Eq. 1.7;[9]) at a single location offshore. The time series of the surface wave parameters computed by SWAN are used in Eq. 1.7. In comparing the two time series at a single point ( Fig. 1.7b), the approximation shows good agreement with the solution from SWAN. From here, the Soulsby (2006) method is concluded to be adequate for calculating the near-bed wave orbital velocity everywhere in the domain.
More validation is needed at other locations, however, to confirm this conclusion.
In addition, a snapshot of the average and maximum U rms is mapped over the domain

Results
The enhancement of the bottom friction is a function of several parameters that describe physical conditions such as the wave climate (H s , T p ), bottom roughness (k s ), water depth (h), wave current velocity (u c ), and near-bed r.m.s. wave orbital velocity (U rms ). The response of these variables to each other and their influence on bottom friction is investigated for various scenarios. In doing so, the expected amount of enhanced bottom friction for any given water depth, wave climate, and current condition can be estimated. In addition, a sensitivity analysis was carried out to see how the increased friction affects the storm surge in the Rhode Island region. This was initially done for the peak of storm (i.e. a snapshot) for Hurricane Irene, and can be extended to unsteady cases in future studies.

Orbital velocity
Recall that a range of wave heights and periods based on the wave climate in our region can be applied to the sensitivity of the enhancement of bottom friction via the near-bed r.m.s. orbital velocity. The corresponding near-bed r.m.s wave orbital velocity for variable wave conditions and water depth (h = 0 to 50 m) is presented in Figure   1.8 based on the Soulsby formulation (Eq. 1.7). Three instances of zero-crossing wave period (T z = 6, 10, 14 s) are explored based on the simulated wave field generated during Hurricane Irene. A simple wave breaking criterion (H ≥ 0.78h) was applied at the shallow water limit [2] for each instance. The near-bed r.m.s. orbital velocity U rms is shown to increase with increasing wave height and period, and decrease with increasing water depth.
The estimated magnitude of U rms extends up to 1.8 m/s for large wave conditions (H s between 8 and 10 m, T z = 14 s) in about 10 m of water depth. While these wave

Enhanced bottom friction
Recall, the major contributors to the magnitude of the enhanced bottom friction are the depth-averaged current velocity u c and the near-bed r.m.s. wave orbital velocity U rms since the corresponding shear stresses are proportional to the square of the respective velocities. In wave-dominated conditions, or similarly in areas of low current, the ratio  It can be seen that for small current velocity (0.25 m/s), quickly approaches its maximum value (2.2) with increasing near-bed r.m.s. wave orbital velocity. For stronger currents (u c = 1.5 m/s), the effect of enhanced bottom friction is less significant (i.e. is smaller), and may not be a first order impact for hurricane-induced currents nearshore.
Inclusion of may be more important for simulating storms that produce large swells (i.e. large orbital velocity). Further, is only equal to 1 when u c is greater than U rms , and surpasses 1 even when u c and U rms are equal. This highlights the increased sensitivity of enhanced bottom friction to wave orbital velocity.
With regards to bottom roughness, enhanced bottom friction is less sensitive at its minimum and maximum (i.e. when approaches 1 or 2.2). Otherwise, tends to be for RI study area.
higher for larger grain size diameters by up to 0.27. Therefore, regions with gravelly sediment (d 50 = 1.0 mm or k s = 0.0025 m) and relatively equal current to wave orbital velocity are subject to greater enhancements in bottom friction than in sandy regions.

Case study: sensitivity of storm surge to enhanced bottom friction
The enhancement of bottom friction is applied over the entire computational domain for the peak of the storm under maximum wave, current, and near-bed r.m.s. wave orbital velocity conditions with a constant bottom roughness. The depth-averaged current velocity u c ( Fig. 1.10a) is shown to reach a maximum value of 1.5 m/s near the coastline and around islands. Other studies have reported the magnitude of tidal current velocity in the study area in recent years (see [20,27,28]). However, the focus of this study is in using hurricane-induced current velocity, where maximum values are expected to surpass those due to tides alone. The enhanced Manning's n * quadratic friction coefficient is applied in the coupled ADCIRC+SWAN model for Hurricane Irene. Relatively no significant change in maximum water elevation is observed between the original simulation ( Fig. 1.12a) and the simulation which is based on the enhanced bottom friction (Fig. 1.12b) in RI. Taking the difference in elevation between 1.12b and 1.12a reveals that the original simulation predicts higher storm surge near the coastline and in Narragansett Bay ( Fig. 1.12c), as expected due to increased friction. In other words, inclusion of this effect leads to slightly less storm surge (i.e. less conservative) near shore.

Discussion
Recall that the estimation of the enhanced bottom friction was based on an analytical expression of the near-bed r.m.s. orbital velocity and empirical formulation of the mean bed shear stress in the presence of combined waves and currents for a single point in time. The current methodology includes the exponential approximation for spectral waves proposed by Soulsby (2006;[9]), which implements the significant wave height and zero-crossing period of a representative JONSWAP spectrum to the bottom velocity spectrum. This spectrum is based on a North sea study which does not have hurricanes.
The influence of waves on local bottom friction can be further assessed by considering other formulations of near-bed r.m.s. orbital velocity, specifically those that consider irregular waves, such as solutions discussed in Wilberg (2008;[10]) and Elfrink (2006;[11]). In practice, this process can be automated in the coupled ADCIRC+SWAN model by changing the source code to extract near-bed orbital velocities and apply the bottom friction formulation at each time step.
The variability of near-bed r.m.s. orbital velocity and current velocity with water depth is of particular interest, as the orbital velocity controls how much the bottom friction is enhanced in a given area. To highlight this relationship in our region, a transect of water depth, orbital velocity, current velocity, and enhanced bottom drag coefficient are taken from the domain during the peak of Hurricane Irene ( Fig. 1.13a,b).
The water depths considered in this study (up to 30 m in Rhode Island) can be considered relatively deep or intermediate water. In this region, small tidal currents are felt near the sea floor (u c ≈ 0.2m/s), but the effect of waves (i.e. orbital velocities) has a greater presence over larger water depths, causing the bottom friction to increase significantly ( Fig. 1.13b) with not much affect on the storm surge ( Fig. 1.12c). In very deep water (e.g. > 150 m), this is not the case because waves do not generally penetrate to near-bed, save for times of long period waves. In shallow water near shore, current velocities are greater than in deep water, which reduces the amount of enhanced bottom friction; however, storm surge is impacted in these regions ( Fig. 1.12c).
The wave-current interaction studies discussed by Huang et al (2010;[6]) and Sun et al. (2013;[7]) incorporate the 3D FVCOM-SWAVE coupled model. In the 3D coupled model, radiation stresses are included in the momentum equations to define wave-driven motions. Also in some 3D models, the bottom-boundary layer could be resolved, which means that enhanced bottom friction could be explicitly included. The use of ADCIRC 3DL (three-dimensional, local) and other 3D models is suggested for assessing nearbed interactions between waves and currents, as well as turbulent mixing of the water column. Accordingly, while using 2D models, the wave-current interaction processes (a) (b) Fig. 1.13. Transect of bathymetry, root-mean-square orbital velocity, current velocity, ratio of bottom drag coefficient , and difference in water elevation between the enhanced and original simulation ∆ξ during the peak of Hurricane Irene; (a) transect line over bathymetry contour and (b) water depth, orbital and current velocities, and as a function of distance from beginning of transect line.
can be parameterized for the sake of computational cost.

Conclusion
Waves interact with the bottom boundary layer and increase the apparent roughness felt by ocean currents. Therefore, they can potentially change the storm surge. on a USGS study [29]. The enhancement of the bottom drag coefficient is calculated based on a formulation proposed by Soulsby (1993), using current-and wave-induced bed shear stresses, and converted to an enhanced Manning's n * quadratic friction coefficient for implementation in the coupled model. The sensitivity of storm surge to the enhanced bottom friction was then assessed in Rhode Island for Hurricane Irene (2011).
As water depth increased, the r.m.s. orbital velocity is seen to decrease. However, the orbital velocity is also seen to increase for longer period waves (e.g. 14 s), more significantly in deeper water (< 20 m). In wave-dominate conditions, or similarly in areas of low current, the ratio of the wave-to current-induced shear stress approaches infinity.
Thus the ratio of enhanced to pure current bottom friction approaches its maximum value (2.2). Therefore, an area with a strong wave climate and weak currents will experience significant increases in friction.
Prior to applying the enhanced bottom friction to the coupled model, an initial simulation of Hurricane Irene is performed with default Manning coefficient (0.02). The simulated depth-average current velocity during the peak of the storm is greatest ( it is expected that the enhanced bottom friction will reduce hurricane-induced storm surge predictions nearshore, leading to less conservative estimates, though this may be more realistic and may lead to better model validations. in excess of 4 meter storm surge in some areas [2], in part due to its large displacement speed and track to the west of the state. Accordingly, even storms with tracks farther away from the coast can lead to significant damage, such as the most recent Hurricane Sandy in 2012, which led to major economic loss in this region [3].

List of References
Climate scientists have been studying the frequency and intensity of hurricanes over time and space, and have developed global and regional climate models to better predict the characteristics of future hurricanes. In parallel, similar efforts have been made, by ocean scientists/engineers, to predict storm surge and waves generated by these storms. The primary model used by the National Weather Service for predicting storm surge due to hurricanes is the "Sea, Lake, and Overland Surges from Hurricanes" (SLOSH) model [4]. On a local scale, SLOSH's curvilinear grid does not resolve complex coastal geometry. A popular tool for numerical simulation of storm surge is the "ADvanced CIRCulation" (ADCIRC) model, which solves the problem over a flexible, unstructured, computational domain [5]. In terms of wave generation during hurricanes, the "Simulating WAves Nearshore" (SWAN) model is a popular spectral wave model that solves the spectral action balance equation, usually coupled with ADCIRC as will be done in this study [6]. With regard to wind forcing of tropical storms, ADCIRC has a wide range of options, including the [7]) parametric wind model to compute wind velocities at each node, or using actual wind field data (wind velocity and surface pressure) over a regular grid, from which ADCIRC can interpolate this forcing onto its domain.
Models for accurately predicting storm surge and waves require reliable wind data for hindcast/forecast purposes. Hurricane wind information is available from the National Hurricane Center "HURricane DATabase" (HURDAT) (www.nhc.noaa.gov), and the "Extended Best Track" (EBT) database (rammb.cira.colostate.edu/) based on HUR-DAT. The HURDAT and EBT databases provide hurricane track, intensity and structure information, and can be converted to a wind field via a parametric wind model such as that of Holland [7]. In practice, the modified "Dynamic Holland Model" (DHM) better captures the surface level winds for developing hurricanes [8]. Alternately, global numerical weather hindcast/forecast models such as "European Center for Medium-Range Weather Forecasts" (ECMWF; www.ecmwf.int) that provide meteorological hindcast/forecast data with temporal resolution of 3 hours and spatial resolution of 1/8 • (∼9 km), can be used for wind forcing. The current ECMWF monthly wind database cannot fully represent the center of a tropical cyclone as it doesn't include a synthetic vortex in the analysis, causing an underestimation of minimum sea level pressure and maximum wind speed in the storm center [9]; on the other hand, the DHM only includes the winds resulting from the hurricane, and not those due to background meteorological conditions. The Northeast Coastal Ocean Forecasting System (NECOFS) is an atmosphere-ocean model covering the northeast U.S. coastal region comprised of meteorological input from the "Weather Research and Forecasting" (WRF) model over an unstructured "Finite-Volume Community Ocean Model" (FVCOM) mesh with hourly forecast fields of surface winds, air pressure, sea level, and wave heights (fvcom.smast.umassd.edu/necofs), and is an alternate source for tropical storm wind forcing. FVCOM incorporates both fluid motion (e.g. water elevation and current) and waves (variation of SWAN; [10]) in 3D, and is primarily used for hindcast/forecasting purposes in New England. NECOFS contains several FVCOM unstructured meshes of varying resolutions from global to regional to local scales.
The sensitivity of storm surge models to wind forcing has been studied in earlier work. Houston et al. (1999;[11]) evaluated the statistical differences between the "Hur- forcing produced a better match to pressure, wind speed, and water level observations. Following Hurricane Sandy (2012), Bennett and Mulligan (2017;[13]) investigated the spatial and temporal distribution of bulk wave parameters simulated using three wind fields (2D Holland, 2D GAHM, and 3D WeatherFlow Regional Atmospheric Modelling System, WRAMS), and determined that a regional atmospheric wind model with the most accurate wind field description is best for hurricane hindcast simulations. Cardone and Cox (2009; [14]) addressed the concern of surface wind measurement practice and explored the surge sensitivity to dynamic, kinematic, and blended wind fields, for Hurricane Katrina (2005) in the Gulf of Mexico; they concluded that real-time wind fields generated from warning center advisories had an uncertainty of up to about 20% in the inner core surface wind speed.
In this work, we investigate the accurate prediction of storm surge and waves from three wind models -EBT DHM, ECMWF and NECOFS WRF -in New England, particularly in the coastal waters of Rhode Island, which have experienced severe coastal flooding during past hurricanes. We first introduce the study region and sources of observational/hindcast data at offshore and near shore locations, as well as the numerical models used. We then discuss details of the wind models and their implementation in the coupled ADCIRC+SWAN modeling system. Finally, we compare the wind, wave, and surge predictions against observed data for the three wind models and several hurricanes.
Conclusions are provided at the end.

Methods 2.2.1 Study area
In the past century, Rhode Island has been impacted by five significant storm events, all hurricanes; Table 2.1 lists these hurricanes which were selected as their extreme water levels surpassed the 10-year exceedance probability from NOAA (tidesandcurrents.noaa.gov). Of these five storms, Hurricane Bob was the only one to make landfall not once, but twice in RI, causing considerable coastal flooding along the southern shore and up Narragansett Bay in Providence, RI. Hurricane Irene (2011) is of particular interest given the fact that wind, wave, and water level gauges and stations had been temporarily deployed nearshore in RI at the time; this is further discussed in the following Data Section (Fig. 2.1). Hurricane Bob was also selected in this study due to its significance in RI history, and Hurricane Sandy because it caused the most recent impacts along the southern shore of RI, leading to significant damage and destruction.
The computational domain used in this study was based on the NECOFS FV-COM model for the Gulf of Maine, Version 4 (GOM4) mesh, developed by Univer- and to provide enough discretization in the study area, the RI mesh was merged with the FVCOM GOM4 mesh, leading to a total 105,560 nodes. Figure 2.2 displays the combined mesh for New England and the higher resolution mesh for RI with corresponding bathymetry of the region.

Details of the Selected Hurricanes
Hurricane Bob (1991) developed from an area of low pressure near the Bahamas on August 16, 1991. The storm's partial track is shown in Figure 2 Hurricane Irene (2011) formed from a tropical wave that exited the African coast on August 15, 2011, and was strengthened by favorable environmental conditions, leading to a destructive landfall in North Carolina as a strong Category 1 hurricane [16]. The storm continued to travel northward along the U.S. east coast to the west of Rhode Island ( Fig. 2.3(b)). RI experienced severe wind gusts that left much of the state without power for several days, and mild flooding in Narragansett Bay due to storm surge. Hurricane Irene was the first major storm to impact Rhode Island since Hurricane Bob in 1991.
Hurricane Sandy (2012)  Cuba as a Category 3 [17]. The storm underwent a complex evolution as it weakened over the Bahamas, growing in size as it traveled northeastward and ended up turning northwestward, making landfall in New Jersey as a post-tropical cyclone ( Fig. 2.3(c)).
The tropical to extratropical cyclone transition was caused by the presence of a shallow low-pressure trough combined with cool temperatures [18]. As a result, significant storm surge along the U.S. east coast, and up to 2 m above mean sea level along the southern shore of RI and lower Narragansett Bay, rivaled the coastal flooding caused by Hurricane Bob. Table 2.2 provides an outline of the historical hurricanes, wind models, and observation/hindcast stations we used. Location and sources of observational data, as well as descriptions of the wind models are provided in the following sections.
The main tidal constituents that dominate the study area are listed in Table 2.3 in Newport and Providence, RI. The ADCIRC+SWAN model was forced with five constituents from the LeProvost tidal database: M2, S2, N2, O1, and K1.

Model
The risk posed by coastal storms results from a combination of wave action and storm surge; therefore it is important to simulate both in a coupled manner. Studies showed that computing wave-surge interactions (wave set-up and set-down) in coupled models [20] results in more accurate predictions of waves and storm surge. The ADCIRC+SWAN coupled model was used in this study. ADCIRC is a two dimensional (2-D) (optional 3D), finite-element, free surface circulation model, which is described in [5] and [20].
SWAN is an open-source third-generation spectral wave model as described by [6]).
In short, SWAN's formulation is based on the conservation of wave action density N = E(σ, θ)/σ, where E(σ, θ) is the directional wave spectrum with σ the relative angular frequency and θ the direction. SWAN takes into account interactions between waves and currents via radiation stresses, and includes parameterizations and equations for wave generation by wind, propagation, nonlinear wave-wave interactions, and dissipation due to whitecapping, bottom friction, and depth-induced breaking. Water elevation and the effect of ambient currents on waves are not explicitly computed in SWAN, but are solved implicitly by coupling it with a hydrodynamic circulation model, such as ADCIRC. The coupling of ADCIRC and SWAN is carried out in parallel on identical sub-meshes using intra-model communication on the same computational core [20]. At each grid point, SWAN is passed wind speeds, water levels, and currents computed by ADCIRC, which are period averaged at each time step and used to recalculate the water depth and related wave processes such as wave propagation/refraction and depth-induced breaking. In turn, ADCIRC is partially driven by radiation stress gradients computed by SWAN, extrapolating them forward in time [20].
ADCIRC was run in 2-D mode with a default Manning's n coefficient of 0.02, and a time step of 0.5 s. The "General Wave Continuity Equation" (GWCE) weighting factor, τ 0 , that weighs the relative contribution of the primitive and wave portions of the GWCE, was adjusted for stability from its default value of 0.03 to 0.02. SWAN was run in non-stationary mode over the unstructured ADCIRC mesh, with 36 directional bins and 40 frequency bins with a low frequency cut-off of 0.031 Hz, and was forced by the same wind field as ADCIRC. The default formulations were applied for breaking and whitecapping (KOMEN), with Manning's n quadratic friction input from ADCIRC. Fleming et al. (2008;[8]) addressed the challenges of acquiring and applying meteorological wind forcing for operational storm surge forecasting, including the uncertainty of hurricane forecasts, the lack of prompt availability of data at high resolution, and the computational expense of using large datasets. The reliability of wind forcing is crucial for accurate storm surge prediction. With regards to the application of hurricane hindcasts for storm surge validation purposes, the same challenges of acquiring accurate wind data and processing large datasets are present. There are several options available for meteorological forcing input in ADCIRC; common inputs include either wind velocity and pressure on a regular grid which are interpolated in space onto the ADCIRC domain, or storm parameters formatted as the "Automated Tropical Cyclone Forecasting" (ATCF)

Wind Forcing
Best Track (BT) file published by the NHC. For this format, wind stress and pressure is calculated by the DHM parametric wind model. The latter approach does include wind forcing outside the immediate area impacted by the storm.
NHC maintains an archive of hurricane hindcasts containing six-hourly storm parameters such as location, maximum sustained wind speed, central pressure, etc. in the ATCF Best Track format. HURDAT does not always contain storm size, which is necessary for parametric wind models. Alternately, the "Risk Prediction Initiative" (RPI) developed the EBT dataset with additional wind structure parameters appended to the post-storm best track files from NHC (rammb.cira.colostate.edu). Maintaining the same format, EBT wind input provides a better source for hurricane hindcast purposes than the standard HURDAT input. Parametric wind models are advantageous in hurricane forecasting and hindcasting due to the relatively small amount of storm input data required and the ability to calculate wind stress and pressure on the fly as a subroutine [8]. The current parametric model used by ADCIRC is the DHM, a modification of the original Holland model [7] by Fleming [8] to address dynamically developing hurricane parameters. The DHM generates the storm vortex following Schloemer's (1954) hyperbolic hurricane pressure profile and the gradient wind equations. From there, the wind is separated into north and east components at each grid point and adjusted for boundary layer (10 m wind velocity) and time (10 min winds).
Global weather hindcast/forecast systems combine several models covering the atmosphere, land, and ocean in order to accurately predict weather conditions across the globe. ECMWF is a combined general circulation model and data assimilation system that includes a set of physical parameterizations to represent processes such as convection, radiation, friction, and diffusion, for real-time, climate analyses [21]. The advantage of a global weather forecast for storm surge modeling is the inclusion of environmental wind speeds outside of the hurricane circumference; however current ECMWF monthly wind datasets do not fully capture the center of tropical cyclones because the ensemble does not include synthetic vortex parameters in its analyses. The technique of inserting of a synthetic vortex from a tropical cyclone of similar location, strength, and motion into the initialization of model simulations is employed by some weather centers such as the U.S. National Meteorological Center [9]; both Aberson (2001; [22]) and Elsberry et al. (2010;[23]) disregarded the ECMWF wind model due to this limitation. However, ECMWF is widely used in the meteorological community, and efforts to improve quality forecasts and reanalysis wind fields are continuously being sought (i.e. Dee et al. (2011); [24]). In this study, the ECMWF wind velocities (m/s) and surface pressure (Pa) are input onto a rectangular grid that completely covers the ADCIRC domain, and interpolated in space onto the ADCIRC mesh. The default wind drag law initially used in this study was Garratt's formula (1967) to calculate wind stress from the input wind velocities.
In addition to the above models, NECOFS utilizes WRF driven by the "North American Meso-scale" (NAM) weather model for meteorological input, with a horizontal resolution up to 3 km, and a two-way nesting method from basin to regional to local scales [15]. Hindcasts are run daily using updated conditions from meteorological observations from buoys, when available. Ocean modeling is completed via coupled FVCOM-SWAVE, where SWAVE is a version of SWAN developed onto the FVCOM framework, in which a flux-corrected transport algorithm is numerically solved with boundary conditions provided by a larger WAVEWATCH-III (WWIII) domain. More details can be found in Qi et al. (2009;[10]). For hindcast simulations, NECOFS incorporates parameters from synthetic storms to better represent the inner structure of tropical cyclones, increasing the accuracy of the peak wind of the storm. NECOFS' outputs are limited to the FVCOM unstructured mesh (GOM4), and are interpolated onto a regular grid before wind speed and pressure can be applied to the high resolution ADCIRC domain covering RI. The output of the NECOFS WRF model for Hurricane Sandy (2012) was provided on a 10 km resolution regular grid directly from NECOFS [25]. The WRF wind model was not available to simulate Hurricanes Bob and Irene.

Results
The performance of ADCIRC in predicting tides was first assessed at the NOAA Newport and Providence water elevation stations. The model was run for 20 days from May 1, 2016 to May 21, 2016 with a one day ramping period, covering a spring-neap cycle. The observed and modeled elevations were processed using T_Tide [26] to compute the amplitude and phases of tidal constituents, and are presented in Table 2    previous research [9,27] found that the center of the cyclone depicted by ECMWF is often represented to be a few degrees away from the position specified by the track data, which can be seen for Hurricanes Bob (Fig. 2.4b) and Sandy (Fig. 2.4f).

Simulation of historical hurricanes
The simulation setup for three historical hurricanes -Hurricane Bob (  For Hurricane Irene, comparisons between simulated and observed wind speed and pressure were made at the BUZM3 station offshore (Fig. 2.5c,d), and meteorological stations in Charlestown and Point Judith, RI nearshore ( Fig. 2.6a,b). The ECMWF wind model better predicted wind speeds (within 15%) at all wind station locations ( Table   2.4) compared to the performance of the parametric wind, which varied significantly for peak wind speed (6.3 -40%). The time series of the parametric wind model was similar across each wind station location. In particular, the difference in peak wind speed between offshore and nearshore stations was minimal due to the inability of the DHM to account for the presence of land. Both the parametric and ECMWF wind models estimated the minimum pressure accurately.
Looking at Hurricane Sandy, the NECOFS WRF wind predicted the maximum wind speed within 18% when compared to wind speed and pressure observations at NDBC C-MAN station BUZM3 (Fig. 2.5e,f). Table 2.4 provides the wind speed comparisons at the other wind station locations. The ECMWF wind model is shown to consistently underestimate the peak wind speed up to 25%, and the parametric wind varied significantly between offshore and nearshore locations (8 -30%). The pressure was in good agreement among the wind models. The largest overestimation of peak wind speed occurred at the New London meteorological station by the WRF and DHM wind models. This can be associated with the proximity of the station to land and the reduced mesh resolution in that area.

Storm Surge
For Hurricane Bob, the time series of water elevation at NOAA tidal stations in Newport and Providence, RI are compared with water levels simulated by each wind model in Figure 2.7. The corresponding RMSE values for surge are presented in Figure   2.9, and are discussed later in the paper. The parametric wind model overestimated the   Fig. 2.6(c,d)). In Figure   2.9, the RMSE is presented for each wind model at each surge station. The EBT DHM parametric wind underestimated the peak storm surge by 50% for a majority of the surge locations (  Figure 2.9. As a result, the WRF wind forcing produced the lowest error during the surge caused by Hurricane Sandy, and the EBT DHM forcing produced the highest errors. Consult the Discussion Section for further evaluation of wind forcing performance. these are listed in Table 2.6. When compared to the WIS 63079 hindcast, the resulting simulated peak wave heights varied among the three wind models, with errors as high as 25% (EBT DHM forcing) and as low as 1.5% (NECOFS WRF forcing).

Discussion
The performance of the three wind models with respect to storm surge prediction was assessed by computing the RMSE during the surge event, and comparing differences between the maximum peak surge of observations and simulations. Figure 2  The formulation of the wind drag coefficient is significant in accurate storm surge prediction. [28]) provided the complex history of the various wind drag coefficient formulations and corresponding wind stress used in storm surge simulations, and pointed out the common technique of capping a linear drag coefficient at a certain threshold. For sensitivity analyses, the wind drag law was adjusted from the ADCIRC default value of Garrett (1977), to Powell (2006) for Hurricane Sandy using the NECOFS WRF wind forcing. As previously described, the Garrett formulation calculates wind drag at each grid point in the domain based on the local wind speed. Powell's method divides the storm into three sectors (right, rear, and left) and applies different drag formulations for each sector. Grid points that do not fall within the three sectors are defaulted to Garrett. Powell's method has been claimed to perform better for tropical storms [29]. However, the sensitivity analysis we performed did not lead to considerable impact on the maximum storm surge or significant wave height. The Powell formulation underestimated the peak storm surge at Skip's Dock and Weekapaug Inlet by 6.9% and 3.7%, respectively, compared to the original underestimate of 6.5% and 3% using the Garrett formulation; additionally, the estimation of peak significant wave height decreased by 2.5% from Garrett (10%) to Powell (12.5%). In response to this, we investigated the storm surge sensitivity to limiting the wind drag coefficient, for values 0.002, 0.0025, and 0.003. Comparisons at surge locations in Newport, Providence, Point Judith, and Westerly, RI revealed little (within 10 cm) to no change in the peak water level during Hurricane Sandy. Therefore, the default wind drag formulation used in ADCIRC was deemed satisfactory for the simulations presented in this study.
The error of simulated storm surge for different wind models was quantified in Rhode Island coastal waters. Results at neighboring states were not considered due to the limited model resolution in other areas, as it was unclear whether the error stemmed from the wind models themselves or from the low resolution of the FVCOM mesh.
Focusing on RI, the parametric wind model provides a good representation of hurricane winds, significant wave height, and storm surge for Hurricane Bob. [11]) reached the same conclusion about the SLOSH parametric model results as compared to the NOAA Hurricane Research Division (HRD) surface winds, and determined that the wind fields observed during and after a hurricane landfall are best simulated using parametric wind forcing. For other hurricanes with tracks farther from the study area, the NECOFS WRF wind model is preferred. Parametric wind models are advantageous in hurricane forecasting and hindcasting due to the relatively small amount of storm input data required such as the hurricane track, intensity and structure information provided in the NHC EBT database. Missing from this database is the definition of environmental/background winds outside of the hurricane center. Global weather hindcast/forecast systems, such as ECMWF, combine several models covering the atmosphere, land, and ocean in a general circulation model, and include environmental wind speeds outside of the hurricane radius. However, without the implementation of a synthetic vortex in the model, the maximum wind speeds in the cyclone center are not fully captured in ECMWF. The combination of a hindcast/forecast system with inserted synthetic vortices (i.e. a blended wind model) is achieved by NECOFS WRF, providing regional wind fields for the U.S. northeast coast.

Conclusions
Access to this wind model was limited and was only available for Hurricane Sandy in this research.
Observation/hindcast stations for wind, surge, and waves were available both offshore and nearshore in Rhode Island coastal waters. Permanent station locations included three tidal gauges operated by NOAA in New London, CT, Newport, RI, and Providence, RI, as well as two NDBC wind and wave buoys located outside of Buzzards Bay, MA (BUZM3) and offshore of Block Island, RI (44097). Several nearshore temporary wind, water level, and wave stations were in operation during Hurricane Irene, including two wind gauges, three water elevation gauges, and two wave gauges along the southern coast of RI. Hindcasts from the USACE WIS program were also available for wind and waves.
Hurricane Bob passed directly over Rhode Island with a radius of maximum wind of 30 nautical miles (56 km). As a result, the EBT DHM parametric wind model estimated the peak wind speed within 10% at the BUZM3 wind station, peak surge within 20% at the tidal gauge in Providence, RI, and peak significant wave height within 53% at the WIS 63079 location. At these same locations, the ECMWF wind model underestimated the maximum wind speed, surge, and wave height more than 50%.
Hurricane Irene traveled west of RI through New York state with a radius of maximum wind of 100 nautical miles (185 km). The resulting ECMWF wind model simulation estimated peak wind speed within 15% when compared with observed wind speeds both offshore and nearshore; differences were as high as 40% for the EBT DHM wind model. Peak storm surge was within 25% and up to 52% for the ECMWF and DHM wind models, respectively. Peak significant wave heights were estimated by the ECMWF wind model within 15%, and varied considerably between offshore and nearshore wave station locations when forced by the DHM wind model.
Hurricane Sandy stayed southwest of RI as it made landfall in New Jersey with a radius of maximum wind of 110 nautical miles (204 km). As a result, the NECOFS WRF wind model simulated peak surge within 11%, peak wind speed within 18%, and peak significant wave height within 10% when compared with offshore and nearshore observations. At the same locations, the ECMWF wind predicted all peaks within 20% and the EBT DHM parametric wind model underestimated peak wind and surge up to 30% and 45%, respectively, and overestimated peak wave height up to 14%.
When modeling storm surge and waves, the selection of a wind model is a crucial step that can affect the results significantly, even more than other parameters such as bottom friction or wind drag. There is no unique "best" wind model for all hindcast applications. This choice depends on the nature of the hurricane, in particular, the size of the storm (i.e. radius of maximum wind) and its storm track relative to the measurement locations. We have quantified that a wind model, which has an error in peak wind speed less than 20% when compared with observations, can successfully be used for storm surge and wave simulations, and the impact of using a poor wind model can result in error as high as 50% in storm surge and wave predictions. The parametric wind model based on the NHC EBT database has significant shortcomings, however it is best used for small storms. In addition, we have proposed the best wind model for our region is the NECOFS WRF blended wind model that addresses background winds as well as the vortex winds of tropical cyclones.