Observing and Modeling the Response of Placentia Bay to an Extratropical Cyclone

An extratropical cyclone reported to have the largest wind speed in Newfoundland in morethan a decade landed on the island of Newfoundland on 11 March 2017. The oceanic responses inPlacentia Bay on the southeast coast of Newfoundland to the winter storm were examined usingobserved data and the Finite-Volume Community Ocean Model (FVCOM). The peak non-tidal waterlevel increase, i.e., storm surge, reached 0.85mat St. Lawrence and 0.77mat Argentia on Placentia Bay.Sea surface temperature slightly decreased after the storm passage according to buoy and satellitemeasurements. Root mean square dierences (RMSD) of the magnitude of storm surge between modelresults and observations are 0.15 m. The model sea surface temperature showed a small decrease,consistent with observations, with RMSDs from 0.19 to 0.64 C at buoy stations. The simulatedsurface current changes agree with buoy observations, with model-observation velocity dierenceratios (VDR) of 0.75–0.88. It was found that, at Argentia (St. Lawrence), the peak storm surge inPlacentia Bay was dominantly (moderately) associated with the inverse barometric eect, and thesubsequent negative surge was mainly due to the wind eect at both stations. The sea surface coolingwas associated with oceanic heat loss. In the momentum balance, the Coriolis, pressure gradient,and advection terms were all important during the storm, while the first two terms were predominantbefore and after the storm.


Introduction
Tropical cyclones (such as hurricanes) and extratropical cyclones can generate strong winds and high coastal storm surges. The combination of storm surge and wind waves can produce significant inundation and cause severe damage in coastal zones [1,2].
Placentia Bay is located on the southeast coast of Newfoundland in the northwestern Atlantic Ocean, bordered by the Avalon Peninsula to the east and the Burin Peninsula to the west (see Figure 1). It supports essential fisheries such as Atlantic cod [3] and emerging aquaculture. It is also an important marine transport route, especially for crude oil tankers. Therefore, it is important to Occasionally, Atlantic tropical storms pass over Placentia Bay in summer and fall. Using observational data and modeling results, Han et al. [4] and Ma et al. [5] studied storm surges and other oceanic responses (temperature, currents) to Hurricane Igor on September 21, 2010 off Newfoundland. Han et al. [6] used satellite thermal imagery and ocean color data to examine the impacts of Hurricane Igor on the sea surface temperature (SST) and phytoplankton over the Grand Banks of Newfoundland. Ma et al. [7] investigated the oceanic responses of Placentia Bay to Hurricanes Igor on September 21, 2010 and Leslie on September 11, 2012 using a coastal ocean model and in situ observations.
In winter, the Newfoundland area is frequently battered by extratropical cyclones. Extratropical cyclones are a dominant source of mid-latitude weather, and their various formation and evolution mechanisms have been studied using a wide range of observational, theoretical, and modeling approaches [8]. Extratropical cyclones are also the principal cause of storm surge events [9]. Currently, many studies are focused on increasing the prediction precision of extratropical cyclone tracks and assessing the links between the extratropical cyclones and extreme events [10][11][12][13]. Using coupled air-sea model solutions, Nelson and He [14] found that air-sea interactions during winter extratropical cyclone outbreaks enhanced ocean heat loss over the Gulf Stream and supported rapid extratropical cyclone intensification. Nevertheless, studies on oceanic responses of Newfoundland waters to extratropical cyclones are limited.
According to online news [15], an exatratropical cyclone passed over Newfoundland on March 11, 2017 (Figure 1), with peak wind speeds up to 44 m s −1 at St. John's International Airport. Meteorologists described this storm as the strongest in more than a decade because the peak wind speed broke the record in some regions of Newfoundland. Three-quarters of Newfoundland was affected by the extratropical cyclone, resulting in widespread power outage.
To understand the oceanic responses (such as storm surge, ocean temperature evolution, and current change) in Placentia Bay to the strong extratropical cyclone, we analyze in situ observations and remote sensing data, and apply a coastal ocean circulation model in Placentia Bay and adjacent shelf waters. Furthermore, we aim to discuss the differences between the oceanic responses of Placentia Bay to winter and summer storms, in reference to Ma et al. [7].
This manuscript is arranged as follows: In Section 2, we describe the data used and the model setup. Section 3 presents the oceanic responses to the extratropical cyclone using observations and model results. Section 4 discusses the storm surge mechanisms and the momentum balance during Occasionally, Atlantic tropical storms pass over Placentia Bay in summer and fall. Using observational data and modeling results, Han et al. [4] and Ma et al. [5] studied storm surges and other oceanic responses (temperature, currents) to Hurricane Igor on 21 September 2010 off Newfoundland. Han et al. [6] used satellite thermal imagery and ocean color data to examine the impacts of Hurricane Igor on the sea surface temperature (SST) and phytoplankton over the Grand Banks of Newfoundland. Ma et al. [7] investigated the oceanic responses of Placentia Bay to Hurricanes Igor on 21 September 2010 and Leslie on 11 September 2012 using a coastal ocean model and in situ observations. In winter, the Newfoundland area is frequently battered by extratropical cyclones. Extratropical cyclones are a dominant source of mid-latitude weather, and their various formation and evolution mechanisms have been studied using a wide range of observational, theoretical, and modeling approaches [8]. Extratropical cyclones are also the principal cause of storm surge events [9]. Currently, many studies are focused on increasing the prediction precision of extratropical cyclone tracks and assessing the links between the extratropical cyclones and extreme events [10][11][12][13]. Using coupled air-sea model solutions, Nelson and He [14] found that air-sea interactions during winter extratropical cyclone outbreaks enhanced ocean heat loss over the Gulf Stream and supported rapid extratropical cyclone intensification. Nevertheless, studies on oceanic responses of Newfoundland waters to extratropical cyclones are limited.
According to online news [15], an exatratropical cyclone passed over Newfoundland on 11 March 2017 (Figure 1), with peak wind speeds up to 44 m s −1 at St. John's International Airport. Meteorologists described this storm as the strongest in more than a decade because the peak wind speed broke the record in some regions of Newfoundland. Three-quarters of Newfoundland was affected by the extratropical cyclone, resulting in widespread power outage.
To understand the oceanic responses (such as storm surge, ocean temperature evolution, and current change) in Placentia Bay to the strong extratropical cyclone, we analyze in situ observations and remote sensing data, and apply a coastal ocean circulation model in Placentia Bay and adjacent shelf waters. Furthermore, we aim to discuss the differences between the oceanic responses of Placentia Bay to winter and summer storms, in reference to Ma et al. [7]. This manuscript is arranged as follows: In Section 2, we describe the data used and the model setup. Section 3 presents the oceanic responses to the extratropical cyclone using observations and model results. Section 4 discusses the storm surge mechanisms and the momentum balance during the storm. Finally, Section 5 gives the summary and conclusion.  [16]) and downloaded from the French Archiving, Validation, and Interpretation of Satellite Oceanographic data (AVISO, www.aviso.altimetry.fr). Each track repeats exactly every 9.9156 days. The 20-Hz data have an along-track spatial resolution of about 300 m. An ascending track (Pass 039) of Jason-3 passed over the domain during the extratropical cyclone event. We make corrections for Jason-3 altimetric range data in the Ku band, including ionospheric delay and sea state bias provided by CNES, radiometer wet tropospheric delay provided by the Jet Propulsion Laboratory (JPL), and dry tropospheric delay provided by the European Center for Medium-range Weather Forecasting (ECMWF). We also correct the altimetry data for geocentric ocean tide using Goddard Ocean Tide 4.8 (GOT4.8 [17]), solid earth tide [18], and pole tide [19]. Finally, we calculate the temporal sea surface height anomalies relative to the sea level averaged from 18 February 2016 to 18 July 2017.

Data and Model
Multiscale Ultrahigh Resolution (MUR) Level 4 SST data [20,21] with spatial resolution of 0.01 • × 0.01 • are obtained from the Group for High Resolution Sea Surface Temperature (GHRSST, https://www.ghrsst.org/). The data are produced at the JPL Physical Oceanography Data Distributed Active Archive Center (PO.DAAC), based on in situ data, the Advanced Very High Resolution Radiometer (AVHRR) data, and the Advanced Microwave Scanning Radiometer-EOS (AMSR-E) data.  Figure 1).
Meteorological data are obtained from the Canadian Weather, Climate and Hazard Service (https: //www.canada.ca/en/services/environment/weather.html). Hourly wind speed and wind direction at SL were included in this study.

Numerical Model Data
This study makes use of a high-resolution global reanalysis dataset, which include hourly 10 m wind, heat flux, mean sea level pressure and sea surface height from National Centers for Environmental Prediction (NCEP) Climate Forecast System Version 2 (CFSv2, https://rda.ucar.edu/, [22]). The horizontal resolution for wind, heat flux, and mean sea level pressure is 0.2 • , while that for sea surface height is 0.5 • .
Hourly temperature and salinity data are obtained from the HYbrid Coordinate Ocean Model (HYCOM [23]). Computations of HYCOM are carried out on a Mercator grid between 78 • S and 76 • N with horizontal resolution of 1/12 • (about 7 km in our study area) and 32 vertical layers ranging from 0 to 5500 m.

FVCOM (3.2) Ocean Circulation Model
With the combined advantages of horizontal grid flexibility and computational efficiency, the FVCOM model [24] is suitable for simulations in area with complex coast lines. FVCOM has been widely used in simulating geophysical dynamic processes in coastal waters [5,7,25,26]. For the present application, we run the model with the three-dimensional setup and in the prognostic mode. A barometric pressure term is added in the momentum equations to account for the effect of atmospheric pressure change on sea level. The k-ε turbulence model from the General Ocean Turbulence Model (GOTM, http://www.gotm.net/, [27]) is used to calculate the vertical mixing.

Model Domain and Grid
The model domain includes Placentia Bay and adjacent shelf waters (see Figure 1b). The geometry of the domain includes several banks and channels.
The triangular grid ( Figure 2) is designed using the Surface Water Model System (SMS, www. aquaveo.com/software/sms-learning). The grid resolution is lowest along the open boundary over the continental shelf (3-5 km) and highest along the coastlines in the inner bay (50 m). The domain grid consists of 31 unequally spaced sigma levels in the vertical, as well as 25,414 nodes and 47,629 elements in the horizontal. The model bathymetry is derived from General Bathymetric Chart of the Oceans (GEBCO, http://www.gebco.net/), which is in a one arc-minute grid and updated in 2008. To minimize pressure gradient errors [28], the bathymetry is smoothed using the same method as in Ma et al. [29]. The integration time step is set as 0.6 s for the external mode and 6 s for the internal mode. With the combined advantages of horizontal grid flexibility and computational efficiency, the FVCOM model [24] is suitable for simulations in area with complex coast lines. FVCOM has been widely used in simulating geophysical dynamic processes in coastal waters [5,7,25,26]. For the present application, we run the model with the three-dimensional setup and in the prognostic mode. A barometric pressure term is added in the momentum equations to account for the effect of atmospheric pressure change on sea level. The k-ε turbulence model from the General Ocean Turbulence Model (GOTM, http://www.gotm.net/, [27]) is used to calculate the vertical mixing.

Model Domain and Grid
The model domain includes Placentia Bay and adjacent shelf waters (see Figure 1b). The geometry of the domain includes several banks and channels.
The triangular grid ( Figure 2) is designed using the Surface Water Model System (SMS, www.aquaveo.com/software/sms-learning). The grid resolution is lowest along the open boundary over the continental shelf (3-5 km) and highest along the coastlines in the inner bay (50 m). The domain grid consists of 31 unequally spaced sigma levels in the vertical, as well as 25,414 nodes and 47,629 elements in the horizontal. The model bathymetry is derived from General Bathymetric Chart of the Oceans (GEBCO, http://www.gebco.net/), which is in a one arc-minute grid and updated in 2008. To minimize pressure gradient errors [28], the bathymetry is smoothed using the same method as in Ma et al. [29]. The integration time step is set as 0.6 s for the external mode and 6 s for the internal mode.

Model Forcing, Open Boundary Conditions and Initial Conditions
The model is forced by spatially and temporally varying winds, air pressure and heat fluxes at the sea surface. Tidal and non-tidal sea levels, temperature and salinity are specified at the open boundary. Hourly, 0.2° wind speeds at 10 m above sea surface and barometric pressure field at the sea surface obtained from CFSv2 are interpolated onto the elements and nodes, respectively. Eight dominant tidal constituents (semi-diurnal: M2, S2, N2 and K2; and diurnal: K1, O1, P1 and Q1) are obtained from OTIS (OSU Tidal Inversion Software [30]) and specified along the open boundaries.

Model Forcing, Open Boundary Conditions and Initial Conditions
The model is forced by spatially and temporally varying winds, air pressure and heat fluxes at the sea surface. Tidal and non-tidal sea levels, temperature and salinity are specified at the open boundary. Hourly, 0.2 • wind speeds at 10 m above sea surface and barometric pressure field at the sea surface obtained from CFSv2 are interpolated onto the elements and nodes, respectively. Eight dominant tidal constituents (semi-diurnal: M2, S2, N2 and K2; and diurnal: K1, O1, P1 and Q1) are obtained from OTIS (OSU Tidal Inversion Software [30]) and specified along the open boundaries. Non-tidal sea level elevations along the open boundaries are extracted from CFSv2. A difference between non-tidal sea level elevations from CFSv2 and tide-gauge data at stations AG and SL is noted while processing the data. The difference of 0.58 cm averaged for the two stations and over the model period is added to every open boundary node. River runoff is not considered since it is very low in winter. Initial sea level and velocity of the model are set to zero. The temperature and salinity conditions are initialized from the HYCOM monthly mean output of February 2017.
The model is run from 14 February to 31 March in 2017. The model reaches an approximate dynamic equilibrium after running for 15 days. Hourly results from March 1 to March 31 are analyzed to examine the influence of the extratropical cyclone on sea level, temperature, salinity, and currents in Placentia Bay.

Model Validation Metrics
The model results are compared with observational data and evaluated qualitatively and quantitatively by examining their correlation coefficient (cor) and the root mean square differences (RMSD) [7,31]: where A m and A o are the model output and observational data, respectively. In addition, the velocity difference ratio (VDR) and the speed difference ratio (SDR) are included to validate the currents. VDR is defined as the ratio of the sum of the squared magnitudes of the vector velocity differences to the sum of the squared magnitudes of the observed velocities, that is: where V m and V o are the model velocities and the observational velocities, respectively. VDR = 0 means exact agreement. VDR < 1 means some agreement; the closer to 0 the better the agreement. VDR ≥ 1 means little or no agreement. Similarly, SDR is defined as the ratio of the sum of squared speed differences to the sum of the squared magnitudes of the observations, that is: VDR is usually larger than SDR, since the former considers not only speed but also direction.

Extratropical Storm Detection
Extratropical storms are identified from CFSv2, using an automated storm detection and tracking methodology [32,33]. Storms are detected according to the minima of air pressure at sea level. We consider a storm for further study based on the following criteria: (i) there is a closed low pressure contour with central air pressure less than 1005 hPa, which is generated in the extratropical region from 25 • N to 55 • N; (ii) the low-pressure system lasts for at least 24 h; and (iii) the storm track is from southwest to northeast. Figure 3, a low-pressure system formed over the northwest Atlantic on 10 March 2017, moving northeastward and gradually intensifying. In the morning of 11 March 2017, the extratropical cyclone arrived on the southwestern Newfoundland Shelf and headed north toward Placentia Bay. It made landfall at about 13:30 (UTC, similarly hereinafter) and crossed over Newfoundland in the afternoon. By the evening of 11 March, the storm center was located on the northeastern Newfoundland Shelf. It stayed in the area for about two days, gradually weakening and eventually dissipating.

Wind
The predominant wind over the study domain during the winter is eastward [34]. When the extratropical cyclone approached Newfoundland, the dominant wind was southwestward over Placentia Bay ( Figure 4). As the storm center moved over Placentia Bay, the observed wind shifted northwestward. The wind direction finally turned northeastward after the storm center left the bay. The maximum hourly wind speed was recorded at 26.6 m s −1 over Placentia Bay during the period of March 10-16, 2017. Figure 5 shows the comparison between CFSv2 and observed 10-m winds at four locations in the model domain.

Wind
The predominant wind over the study domain during the winter is eastward [34]. When the extratropical cyclone approached Newfoundland, the dominant wind was southwestward over Placentia Bay ( Figure 4). As the storm center moved over Placentia Bay, the observed wind shifted northwestward. The wind direction finally turned northeastward after the storm center left the bay. The maximum hourly wind speed was recorded at 26.6 m s −1 over Placentia Bay during the period of 10-16 March 2017. Figure 5 shows the comparison between CFSv2 and observed 10-m winds at four locations in the model domain.

Storm Surge
When the storm reaches the bay in the morning of 11 March 2017, the water level increases. Peak surges of 0.85 m and 0.77 m are reached at SL (Figure 6a) and AG (Figure 6b), respectively, a few hours after the storm makes landfall. Afterwards a sharp decrease in sea level is seen at SL and AG. Negative surges with magnitudes similar to the preceding positive surges occur in the evening. There is indication of high-frequency oscillations, which may be associated with seiches [7].

Storm Surge
When the storm reaches the bay in the morning of March 11, 2017, the water level increases. Peak surges of 0.85 m and 0.77 m are reached at SL (Figure 6a) and AG (Figure 6b), respectively, a few hours after the storm makes landfall. Afterwards a sharp decrease in sea level is seen at SL and AG. Negative surges with magnitudes similar to the preceding positive surges occur in the evening. There is indication of high-frequency oscillations, which may be associated with seiches [7]. We compare the non-tidal sea level anomalies (defined as non-tidal sea levels minus their mean over the period from March 1 to 31, 2017) from the model solutions with the tide-gauge observations at AG and SL. Tide-gauge data are first detided by using the T-Tide toolbox [35].  (Figure 6a). While the causes for the phase error at this magnitude is complicated and difficult to identify, likely major factors are wind forcing and non-tidal sea levels specified at the open boundary. The Jason-3 underestimation may be due to errors in altimetric measurements and environmental and geophysical corrections [36]. The sea surface height anomalies decrease offshore (Figure 7b). The We compare the non-tidal sea level anomalies (defined as non-tidal sea levels minus their mean over the period from 1 to 31 March 2017) from the model solutions with the tide-gauge observations at AG and SL. Tide-gauge data are first detided by using the T-Tide toolbox [35]. Eight tidal constituents (M2, S2, N2, K2, K1, O1, P1, and Q1) are removed from the model results by using the harmonic analysis. The temporal evolutions of hourly non-tidal sea level anomalies from 10 to 16  and geophysical corrections [36]. The sea surface height anomalies decrease offshore (Figure 7b). The RMSD of sea surface height anomalies between model and low-pass filtered altimeter observations is 0.12 m.
Atmosphere 2019, 10, x FOR PEER REVIEW 9 of 20 RMSD of sea surface height anomalies between model and low-pass filtered altimeter observations is 0.12 m.    Next, we examine satellite observations. A slight drop of SST appears in satellite data in Placentia Bay and surrounding waters after the passage of the extratropical cyclone (Figure 9a,b). The biggest drop in the spatially averaged SST for Placentia Bay is 0.89 °C from March 10 to March 14 ( Figure 9f). The discrepancies in the SST changes between satellite and buoy data may be ascribed to the different measurement depths used: the sea surface for the former and 0.5 m below it for the latter. The sea surface cooling from satellite observations (Figure 9a,b) is more prominent in the inner and middle bay than in the outer bay, since the outer bay can be strongly influenced by relatively warm shelf water through advection and lateral mixing. The distribution of SST from the FVCOM solutions (Figure 9c,d) presents a similar pattern overall. Nevertheless, there are some notable spatial differences between observed and model SST changes, for which we are unable to pinpoint exact causes. The maximum SST drop from March 10 to March 13 is 0.54 °C (Figure 9f). Next, we examine satellite observations. A slight drop of SST appears in satellite data in Placentia Bay and surrounding waters after the passage of the extratropical cyclone (Figure 9a,b). The biggest drop in the spatially averaged SST for Placentia Bay is 0.89 • C from 10 March to 14 March (Figure 9f). The discrepancies in the SST changes between satellite and buoy data may be ascribed to the different measurement depths used: the sea surface for the former and 0.5 m below it for the latter. The sea surface cooling from satellite observations (Figure 9a,b) is more prominent in the inner and middle bay than in the outer bay, since the outer bay can be strongly influenced by relatively warm shelf water through advection and lateral mixing. The distribution of SST from the FVCOM solutions (Figure 9c,d) presents a similar pattern overall. Nevertheless, there are some notable spatial differences between observed and model SST changes, for which we are unable to pinpoint exact causes. The maximum SST drop from 10 March to 13 March is 0.54 • C (Figure 9f).

Temperature
The SST averaged for the three buoy stations drops by 0.42 • C from 10 March to 13 March, which represents the maximal SST change during the storm (Figure 9f). Figure 9e shows the time series of daily air temperatures observed from buoys at RI and NB and from weather station at SL. The air temperature is about 0 • C before the storm, and drops about 7 • C during the storm (from 9 March to 12 March). Atmosphere 2019, 10, x FOR PEER REVIEW 11 of 20 The SST averaged for the three buoy stations drops by 0.42 °C from March 10 to March 13, which represents the maximal SST change during the storm (Figure 9f). Figure 9e shows the time series of daily air temperatures observed from buoys at RI and NB and from weather station at SL. The air temperature is about 0 °C before the storm, and drops about 7 °C during the storm (from March 9 to March 12).

Sea Surface Current
Modeled surface currents are compared with measurements taken at buoys HP, RI, and MP ( Figure 10). The buoy measurements show that the storm-induced surface current changes are mainly

Sea Surface Current
Modeled surface currents are compared with measurements taken at buoys HP, RI, and MP ( Figure 10). The buoy measurements show that the storm-induced surface current changes are mainly associated with the zonal component, which is consistent with the dominant wind changes in the zonal direction ( Figure 5). As the storm approaching, surface currents at HP (Figure 10a,b) change gradually from eastward to westward and then drastically from westward back to eastward soon after landfall. The surface currents at RI are relatively weak even during the extratropical cyclone (Figure 10c,d) due to its location. At MP, the surface currents turn westward before the extratropical cyclone moves over the region (Figure 10e,f). After the storm center passes, the surface currents at MP become eastward under the wind forcing. associated with the zonal component, which is consistent with the dominant wind changes in the zonal direction ( Figure 5). As the storm approaching, surface currents at HP (Figure 10a,b) change gradually from eastward to westward and then drastically from westward back to eastward soon after landfall. The surface currents at RI are relatively weak even during the extratropical cyclone (Figure 10c,d) due to its location. At MP, the surface currents turn westward before the extratropical cyclone moves over the region (Figure 10e,f). After the storm center passes, the surface currents at MP become eastward under the wind forcing. There is no indication of near-inertial oscillation in neither observed nor modeled currents, since the ocean had little stratification in the vertical. Previous studies [7] showed strong inertial oscillation after the passage of hurricanes at MP in summer when the ocean was well stratified in the vertical. Vertical stratification favors the generation of the near-inertial oscillation during a storm [7].

Storm Surge Mechanisms
There are several processes generating storm surge, including winds, atmospheric pressure, tide-surge interaction [38]. Ma et al. [7] investigated the storm surges in Placentia Bay during Hurricanes Igor on September 21, 2010 and Leslie on September 11, 2012. They found the peak storm surge is significantly influenced by local wind-driven and inverse barometric effects during Leslie at Argentia and St. Lawrence, but predominately due to remote forcing during Igor, depending on the storm track location.
To identify the dominant processes for the storm surge during this winter storm, we carried out a series of model experiments with three different forcing configurations, namely, atmospheric There is no indication of near-inertial oscillation in neither observed nor modeled currents, since the ocean had little stratification in the vertical. Previous studies [7] showed strong inertial oscillation after the passage of hurricanes at MP in summer when the ocean was well stratified in the vertical. Vertical stratification favors the generation of the near-inertial oscillation during a storm [7].

Storm Surge Mechanisms
There are several processes generating storm surge, including winds, atmospheric pressure, tide-surge interaction [38]. Ma et al. [7] investigated the storm surges in Placentia Bay during Hurricanes Igor on 21 September 2010 and Leslie on 11 September 2012. They found the peak storm surge is significantly influenced by local wind-driven and inverse barometric effects during Leslie at Argentia and St. Lawrence, but predominately due to remote forcing during Igor, depending on the storm track location.
To identify the dominant processes for the storm surge during this winter storm, we carried out a series of model experiments with three different forcing configurations, namely, atmospheric forcing only, tide forcing only, and atmospheric and tide forcing combined. The corresponding non-tidal sea levels are denoted as η NT , η TO , and η AT , respectively. Then the tide-surge interaction η TSI for the storm follows η TSI = η AT − η TO − η NT . The inverse barometric response is defined as the sea surface height variation caused by an atmospheric pressure fluctuation p atm : where η is the sea surface height variation, g is the gravitational acceleration, and ρ 0 is the reference water density. The wind effect can then be determined by subtracting η from η NT . The barometric pressure declines by 54 hPa from 00:00 to 15:00 on 11 March, when the extratropical cyclone moves toward Placentia Bay. The lowest barometric pressure is 954 hPa at AG, inducing a sea surface height increase of 0.58 m that accounts for 75% of the observed storm surge (0.77 m) (Figure 11a). At SL, the sea surface height increase caused by the inverse barometric effect reached 0.46 m, about 56 percent of the observed storm surge (Figure 11b). Therefore, the peak storm surge is mainly associated with the inverse barometric response at AG, and to a lesser degree at SL. forcing only, tide forcing only, and atmospheric and tide forcing combined. The corresponding nontidal sea levels are denoted as , , and , respectively. Then the tide-surge interaction η TSI for the storm follows = − − . The inverse barometric response is defined as the sea surface height variation caused by an atmospheric pressure fluctuation : where is the sea surface height variation, is the gravitational acceleration, and 0 is the reference water density. The wind effect can then be determined by subtracting from . The barometric pressure declines by 54 hPa from 00:00 to 15:00 on March 11, when the extratropical cyclone moves toward Placentia Bay. The lowest barometric pressure is 954 hPa at AG, inducing a sea surface height increase of 0.58 m that accounts for 75% of the observed storm surge (0.77 m) ( Figure  11a). At SL, the sea surface height increase caused by the inverse barometric effect reached 0.46 m, about 56 percent of the observed storm surge (Figure 11b). Therefore, the peak storm surge is mainly associated with the inverse barometric response at AG, and to a lesser degree at SL. It is well known that Ekman transport induced by winds plays an important role in generating coastal storm surge. In this extratropical cyclone event, the wind effect is much weaker than the inverse barometric effect for the peak surge at AG (Figure 11a)，due to quite low atmospheric pressure and relatively weak winds before and during landfall. Strong eastward wind with speed more than 15 m s −1 (Figure 12a) lasts for several days after landfall. The eastward wind induces Ekman transport which pushes water out of the bay, causing the water level in the bay to drop by about 1.3 Figure 11. Time series of the non-tidal sea level anomalies (m) from observations (red) and model results (blue), the inverse barometric effect (purple), the tide-surge interaction (green), and the wind effect (black) at AG (a) and SL (b).
It is well known that Ekman transport induced by winds plays an important role in generating coastal storm surge. In this extratropical cyclone event, the wind effect is much weaker than the inverse barometric effect for the peak surge at AG (Figure 11a), due to quite low atmospheric pressure and relatively weak winds before and during landfall. Strong eastward wind with speed more than 15 m s −1 (Figure 12a) lasts for several days after landfall. The eastward wind induces Ekman transport which pushes water out of the bay, causing the water level in the bay to drop by about 1.3 m from 14:00 to 21:00 on 11 March. Therefore, the large negative surge is mainly due to the wind effect. m from 14:00 to 21:00 on March 11. Therefore, the large negative surge is mainly due to the wind effect.
Mecking et al. [39] used a barotropic shallow-water model to simulate coastal sea level change around the southeastern Newfoundland in response to Tropical Storm Helene (2000) passing over the Grand Banks. It was found that a combination of wind stress and atmospheric pressure forcing caused a meteorological tsunami. The dynamic response to the wind forcing and atmospheric pressure forcing resulted in sea level increase and decrease at Argentia, respectively. Note that the dynamic pressure response here is different from the inverse barometric effect. At SL, the observed wind speed and directions are quite different from those of CFSv2 from 12:00 to 13:00 on March 11 (Figure 12b). The northwestward winds would induce northeastward Ekman transport, causing the water level to increase. However, the southeastward winds from CFSv2 would induce offshore Ekman transport, causing the water level to decrease in the model results. As a result, the model would underestimate the contribution of winds to the peak storm surge at SL. Following Ma et al. [7], we reconstruct wind fields by blending wind observations at the weather stations and buoys with CFSv2 and use the reconstructed wind fields to force the model. As shown in Figure 13, the model results forced by the reconstructed winds are similar to those forced by the CFSv2 winds, though the reconstructed winds agree well with the observed winds in both magnitude and direction. Therefore, the underestimation of the peak storm surge may be related to something else. While we are not able to pinpoint exact problems, the quality of the open boundary condition may be one of them, since SL is close to the western open boundary. Mecking et al. [39] used a barotropic shallow-water model to simulate coastal sea level change around the southeastern Newfoundland in response to Tropical Storm Helene (2000) passing over the Grand Banks. It was found that a combination of wind stress and atmospheric pressure forcing caused a meteorological tsunami. The dynamic response to the wind forcing and atmospheric pressure forcing resulted in sea level increase and decrease at Argentia, respectively. Note that the dynamic pressure response here is different from the inverse barometric effect.
At SL, the observed wind speed and directions are quite different from those of CFSv2 from 12:00 to 13:00 on 11 March (Figure 12b). The northwestward winds would induce northeastward Ekman transport, causing the water level to increase. However, the southeastward winds from CFSv2 would induce offshore Ekman transport, causing the water level to decrease in the model results. As a result, the model would underestimate the contribution of winds to the peak storm surge at SL. Following Ma et al. [7], we reconstruct wind fields by blending wind observations at the weather stations and buoys with CFSv2 and use the reconstructed wind fields to force the model. As shown in Figure 13, the model results forced by the reconstructed winds are similar to those forced by the CFSv2 winds, though the reconstructed winds agree well with the observed winds in both magnitude and direction. Therefore, the underestimation of the peak storm surge may be related to something else. While we are not able to pinpoint exact problems, the quality of the open boundary condition may be one of them, since SL is close to the western open boundary.

Momentum Balance
The momentum balance is analyzed at MP. The momentum equations and the continuity equation used in FVCOM [40] are: where x, y, and z are the zonal, meridional, and vertical coordinates, respectively; u, v and w are the x, y, and z velocity components, respectively; f is the Coriolis parameter; g is the gravitational acceleration; ρ 0 is the reference water density; p is the pressure; K m is the vertical eddy viscosity; and is the horizontal eddy viscosity. Based on the model results, four dominant terms are considered: Coriolis, pressure gradient, advection, and horizontal diffusion terms. The other terms are smaller by a few orders of magnitude. Figure 14 shows the anomalies of the main terms in the xand y-directions at 5 m below the surface at MP. The anomalies are relative to the means over the period from March 8 to March 14. Before the extratropical cyclone, the variations of the Coriolis term and pressure gradient term were large and thus played dominant roles in the momentum balance in both directions, while those of the advection

Momentum Balance
The momentum balance is analyzed at MP. The momentum equations and the continuity equation used in FVCOM [40] where x, y, and z are the zonal, meridional, and vertical coordinates, respectively; u, v and w are the x, y, and z velocity components, respectively; f is the Coriolis parameter; g is the gravitational acceleration; ρ 0 is the reference water density; p is the pressure; K m is the vertical eddy viscosity; and A is the horizontal eddy viscosity. Based on the model results, four dominant terms are considered: Coriolis, pressure gradient, advection, and horizontal diffusion terms. The other terms are smaller by a few orders of magnitude. Figure 14 shows the anomalies of the main terms in the x-and y-directions at 5 m below the surface at MP. The anomalies are relative to the means over the period from 8 March to 14 March. Before the extratropical cyclone, the variations of the Coriolis term and pressure gradient term were large and thus played dominant roles in the momentum balance in both directions, while those of the advection term and horizontal diffusion term were relatively small. As the storm was approaching, magnitudes of the pressure gradient term and the Coriolis term changed from less than 1 × 10 −5 m s −2 to more than 2 × 10 −5 m s −2 . The variability of the advection term increased by about 1 × 10 −5 m s −2 as well. After the storm, the Coriolis term and the pressure gradient term became dominant again.
Atmosphere 2019, 10, x FOR PEER REVIEW 16 of 20 term and horizontal diffusion term were relatively small. As the storm was approaching, magnitudes of the pressure gradient term and the Coriolis term changed from less than 1 × 10 −5 m s −2 to more than 2 × 10 −5 m s −2 . The variability of the advection term increased by about 1 × 10 −5 m s −2 as well. After the storm, the Coriolis term and the pressure gradient term became dominant again. Figure 14. Anomalies of Coriolis term (blue), advection term (red), pressure gradient term (green) and horizontal diffusion term (purple) in the x-direction (a) and the y-direction (b).
The root mean square (RMS) values of the Coriolis and pressure gradient terms are about four times of other terms in the x and y directions during the pre-and post-storm periods ( Table 1), suggesting that the momentum balance is predominated by the Coriolis and pressure gradient terms. During the storm event (from 12:00 on March 10 to 12:00 on March 12), the RMS values of all terms increased. The Coriolis and pressure gradient terms were about one to two times greater than the other terms.  The root mean square (RMS) values of the Coriolis and pressure gradient terms are about four times of other terms in the x and y directions during the pre-and post-storm periods ( Table 1), suggesting that the momentum balance is predominated by the Coriolis and pressure gradient terms. During the storm event (from 12:00 on 10 March to 12:00 on 12 March), the RMS values of all terms increased. The Coriolis and pressure gradient terms were about one to two times greater than the other terms.

Differences in Responses to Winter and Summer Storms
Using satellite observations, Han et al. [6] found that Hurricane Igor in September 2010 made SST decrease by about 6 • C for the area shallower than 200 m southeast of Newfoundland. The SST cooling in Placentia Bay during Hurricane Igor was about 3 • C and that during Hurricane Leslie was about 2 • C [7]. During these tropical storms, colder subsurface water was brought upward to the near surface layer through vertical mixing, leading to sea surface cooling.
In contrast, the present study reveals a different mechanism of sea surface cooling in response to a winter storm. Figure 15 shows the temporal evolution of temperature at MP during 9-13 March 2017. The temperature is nearly uniform in the vertical before the storm. The storm brings much colder air ( Figure 9e) and enhances oceanic heat loss at sea surface. As a result, the ocean temperature in the upper 100 m decreases by about 0.3 • C. Note that the vertical distribution of modelled oceanic temperature needs to be validated against observations which however are unavailable at the present. Using a regional coupled air-sea model, Nelson and He [14] also found that a winter extratropical cyclone enhanced ocean heat loss over the Gulf Stream.

Differences in Responses to Winter and Summer Storms
Using satellite observations, Han et al. [6] found that Hurricane Igor in September 2010 made SST decrease by about 6 °C for the area shallower than 200 m southeast of Newfoundland. The SST cooling in Placentia Bay during Hurricane Igor was about 3 °C and that during Hurricane Leslie was about 2 °C [7]. During these tropical storms, colder subsurface water was brought upward to the near surface layer through vertical mixing, leading to sea surface cooling.
In contrast, the present study reveals a different mechanism of sea surface cooling in response to a winter storm. Figure 15 shows the temporal evolution of temperature at MP during March 9-13, 2017. The temperature is nearly uniform in the vertical before the storm. The storm brings much colder air (Figure 9e) and enhances oceanic heat loss at sea surface. As a result, the ocean temperature in the upper 100 m decreases by about 0.3 °C . Note that the vertical distribution of modelled oceanic temperature needs to be validated against observations which however are unavailable at the present. Using a regional coupled air-sea model, Nelson and He [14] also found that a winter extratropical cyclone enhanced ocean heat loss over the Gulf Stream. In terms of ocean currents, Ma et al. [7] showed strong inertial oscillation after the passage of hurricanes at MP in summer when the ocean was well stratified in the vertical. Vertical stratification favors the generation of the near-inertial oscillation during a storm [7]. In contrast, there is no indication of near-inertial oscillation in neither observed nor modeled currents in response to the winter storm (Figure 10), since the ocean had little vertical stratification. In terms of ocean currents, Ma et al. [7] showed strong inertial oscillation after the passage of hurricanes at MP in summer when the ocean was well stratified in the vertical. Vertical stratification favors the generation of the near-inertial oscillation during a storm [7]. In contrast, there is no indication of near-inertial oscillation in neither observed nor modeled currents in response to the winter storm ( Figure 10), since the ocean had little vertical stratification.

Summary and Conclusions
Using satellite measurements, in situ data and model results, we investigate the response of Placentia Bay to the extratropical cyclone of 11 March 2017, which is reported to have the largest wind speed in Newfoundland in more than a decade. The peak wind speed observed by the buoys in Placentia Bay reaches 26.6 m s −1 during the extratropical cyclone event. The storm surge induced by the extratropical cyclone is detected by tide-gauge stations and the Jason-3 altimeter. The maximum storm surges at St. Lawrence (SL) and Argentia (AG) were 0.85 m and 0.77 m, respectively, during the storm. Both buoy and satellite observations show a slight SST drop, 0.42 • C from the former and 0.65 • C from the latter.
The Finite-Volume Community Ocean Model (FVCOM) is employed to examine the oceanic response to the storm in Placentia Bay. The magnitude of storm surge from the model generally agrees with observations, with RMSDs of about 0.15 m. However, the model is noted to underestimate the peak surges at SL. The model captures the slight SST drop during the storm. The root-mean-square differences (RMSDs) of SST between the model and buoy observations are less than 0.7 • C. The model surface current changes agree fairly with observations, with SDRs of 0.23-0.34 and VDRs of 0.75-0.88 • C.
The model storm surge results indicate that the inverse barometric response plays an important role in generating the peak surge at AG and SL due to quite low air pressure and relatively weak winds before and during landfall. The subsequent negative surge is mainly due to the wind effect associated with relatively strong eastward winds at both stations.
The model ocean temperature change reveals that the sea surface cooling is associated with the oceanic heat loss during the storm that brings much colder air. There is no indication of near-inertial oscillation in response to the winter storm, due to little oceanic stratification in the vertical.
The variations of major terms in the governing equation at the 5 m depth at MP indicate that the storm perturbs the pre-storm momentum balance which is dominated by the Coriolis term and the pressure gradient term. As a result, the advection term and horizontal diffusion term become important as well during the storm. The Coriolis term and the pressure gradient term become dominant again after the storm passes.
In future work, we plan to improve the open boundary conditions by using a regional ocean model output, which may help improve the storm surge simulation. Including precipitation and evaporation in the atmospheric forcing can be another aspect for enhancement. Furthermore, the FVCOM model can be coupled with an atmospheric model and a surface wave model to better simulate atmosphere-ocean-wave interactions during winter storms. In addition, more observations, such as vertical profiles data, should be collected and applied to validate the model results.