Modeling the formation and circulation processes of water masses and sea ice in the Gulf of St. Lawrence, Canada

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. C8, 3269, doi: /2000jc000686, 2003 Modeling the formation and circulation processes of water masses and sea ice in the Gulf of St. Lawrence, Canada François J. Saucier, François Roy, and Denis Gilbert Fisheries and Oceans, Maurice Lamontagne Institute, Mont-Joli, Québec, Canada Pierre Pellerin and Harold Ritchie Meteorological Service of Canada, Recherche en Prévision Numérique, Dorval, Québec, Canada Received 26 October 2000; revised 5 February 2002; accepted 28 April 2002; published 21 August [1] The seasonal cycle of water masses and sea ice in the Gulf of St. Lawrence is examined using a three-dimensional coastal ice-ocean model with realistic tidal, atmospheric, hydrologic, and oceanic forcing. The model includes a level 2.5 turbulent kinetic energy equation. A model simulation over is verified against available data on sea ice, temperature, and salinity. The results demonstrate a consistent seasonal cycle in atmosphere-ocean exchanges and the formation and circulation of water masses and sea ice. The accuracy of radiative, momentum, and sensible heat exchanges at the sea surface, and the production of turbulent kinetic energy from winds and tides, are critical to the accuracy of the modeled circulation. The analysis of the mean error on near-surface temperature and salinity in the late summer and fall using standard bulk exchange coefficients and radiation (about 1 C too cold and 1 salinity unit too fresh) shows the tradeoff between tidal mixing at the head of the Laurentian Channel, and winddriven circulation and mixing in the surface waters. The results suggest year-long stratification in the estuary and northwestern Gulf, with little mixing except near the head region, where relatively deep warmer waters are mixed to the surface during winter, and cold intermediate waters are efficiently withdrawn during summer. The results suggest that the summer cold waters found at intermediate depths in the estuary and northwestern Gulf are not formed in situ. A significant fraction of these waters enters through the Strait of Belle Isle in wintertime, eventually reaching the estuary within about 6 months. INDEX TERMS: 4207 Oceanography: General: Arctic and Antarctic oceanography; 4243 Oceanography: General: Marginal and semienclosed seas; 4235 Oceanography: General: Estuarine processes; 4540 Oceanography: Physical: Ice mechanics and air/sea/ice exchange processes; KEYWORDS: St. Lawrence, air-sea interactions, circulation, tide, sea ice, numerical modeling, seasonal cycle Citation: Saucier, F. J., F. Roy, D. Gilbert, P. Pellerin, and H. Ritchie, Modeling the formation and circulation processes of water masses and sea ice in the Gulf of St. Lawrence, Canada, J. Geophys. Res., 108(C8), 3269, doi: /2000jc000686, Introduction [2] The Gulf of St. Lawrence (GSL, Figure 1) is a semienclosed sea with an area of about km 2 opened to the Atlantic Ocean through Cabot Strait and the Strait of Belle Isle. On timescales ranging from hours to seasons and years, the circulation is controlled by tides, exchanges with the atmosphere, runoff from land, the seasonal ice cover, and inflow through the bounding straits (e.g., see reviews by Dickie and Trites [1983] and Koutitonsky and Bugden [1991]). The main channels allow Atlantic and Labrador shelf waters to intrude at depth and circulate toward the head region of the lower St. Lawrence Estuary. Continental runoff freshwaters enter the GSL mainly from the estuary and north shore rivers. They partly mix with salt waters, flow through the general cyclonic circulation, and exit Copyright 2003 by the American Geophysical Union /03/2000JC through Cabot Strait. The heat, salt, and momentum distributions of the surface 200 m exhibit a strong seasonal cycle wherein the heat and salt contents of the water column remain everywhere strongly influenced by horizontal currents and stratification. During winter, sea ice of the order of half-meter thickness is produced. The surface mixed layer reaches about 100 m depth, and significant inflows of relatively cold and salty Labrador shelf waters take place at intermediate depths. This leaves a cold intermediate layer (CIL), with temperature near 0 C and salinity between 32 and 33, persisting between 30 and 150 m depth beneath the new spring surface layer [e.g., Lauzier and Graham, 1958; Banks, 1966]. The circulation and fate of the CIL over seasons remain elusive. The waters below the CIL are warmer (2 to 6 C) and saltier (33 to 35). They are slowly advected in the Laurentian Channel (LC) from the continental shelf break toward the estuary [e.g., Lauzier and Bailey, 1957; Lauzier and Trites, 1958; Bugden, 1991]. The sea ice cover, the water mass properties, and the circulation 25-1

2 25-2 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 1. Mercator depth chart of the estuary and Gulf of St. Lawrence. A small sub-area of the 5-km grid is shown along its boundary near Cabot Strait. Montreal (not shown) is at the western end of the onedimensional river channel model, coupled with the 3-D model on the section shown across Île d Orléans at the head of the upper estuary. in the GSL also exhibit large interannual variability [e.g., Bugden, 1991; Petrie et al., 1996; Gilbert and Pettigrew, 1997; Drinkwater et al., 1999a] in response to changes in the atmospheric conditions, the freshwater supply, or shelf water properties outside the GSL. However, the processes linking the forcing and the responses are still largely not quantified. [3] One improves our understanding of the circulation and our capability to model its seasonal to interannual variations for a number of reasons, for example, to understand changes that have occurred in the ecosystem or to downscale climate change scenarios. A realistic numerical model must reproduce the main known oceanographic features like coastal currents (e.g., the Gaspé Current), upwelling and downwelling events, tidal propagation and mixing, the sea ice cover formation and dynamics, the production and persistence of the summer cold intermediate layer, the inflow of shelf waters, and other observations. Such a model must cope with the turbulent motions that transfer momentum and water properties in the vertical, controlling the estuarine circulation, the exchanges with the atmosphere, and the persistent properties of the deeper waters over seasons to years. [4] The main goal of this study is to develop and evaluate a three-dimensional coupled ice-ocean model of the main processes that govern the seasonal cycle of water masses and sea ice circulation in the GSL. The model accounts for the evolution of turbulent kinetic energy, water levels, currents, temperature, salinity, and the sea ice concentration and thickness. It is driven by the tides propagating from the Atlantic Ocean, temperature and salinity at the open boundaries, river runoff, and 6-hourly winds, humidity, air temperature, radiation, and precipitation. [5] The following section presents the physical model and the experimental setting for a hindcast simulation from November 1996 to April 1998, a period covering two consecutive winters and allowing the study of the seasonal cycle. In section 3, the model results are compared with about 650 temperature-salinity profiles, observed sea ice conditions, 18 surface temperature records, and tide gauges. The results show that the heat, freshwater, and sea ice hourly to seasonal variability is rather well resolved. Residual problems are discussed in section 4 regarding mixing, unresolved processes in the estuary, and the accuracy of the forcing fields. The paper ends with a summary and an outlook for future work. 2. Experimental Setting [6] The detailed description of the ice-ocean model is given in Appendix A. Briefly, a hydrostatic ocean solution to the mass, momentum, heat, and salt conservation equations in the Boussinesq and shallow water approximations is initially reached from Backhaus [1983, 1985] and Stronach et al. [1993]. It is modified herein to make use of a fluxcorrected transport scheme [Zalesak, 1979], and a turbulent energy model implemented from Mellor and Yamada [1974, 1982], Galperin et al. [1988], Kantha and Clayson [1994], and Simpson et al. [1996]. Analyses suggest that the level 2.2 closure (with vertical diffusion, but neglecting the advection term of turbulent kinetic energy in equation (A6)), with a diagnostic master length scale limited by the Ozmidov scale, can cope rather well with tidal, wind, and density-driven mixing of the stratified flow over seasons using no restoring conditions. The level 2 model (left member of equation (A6) set to zero) strongly underestimated the mixed-layer depth [e.g., see also Simpson et al., 1996; Meier, 2000]. The ocean model is coupled to a dynamic [Flato, 1993] and thermodynamic [Semtner, 1976] sea ice model. The exchanges between the ocean, sea ice, and atmosphere are governed by bulk aerodynamic exchange formulas [e.g., Parkinson and Washington, 1979].

3 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE 25-3 Figure 2. Domain-averaged forcing over the December 1, 1996, to March 31, 1998, simulation period. Six-hourly and monthly mean (a) air temperature; (b) wind intensity; (c) dew point; (d) cloud cover; (e) P-E, and daily (f ) runoff (dashed line: St. Lawrence River; dash-dotted line: sum of North Shore rivers, solid line: total), and (g) area-averaged accumulated (upper curve) and compacted (lower curve) snow depth over the ice. [7] The model domain extends between open boundaries near Cabot Strait and the Strait of Belle Isle, and the upper limits of tidal influence near Montreal and at the head of the Saguenay Fjord (Figure 1). The horizontal grid resolution is 5 km for the region between the outer straits and I le d Orle ans (landward limit of salt water), and a one-dimensional model is used for the momentum transfer from Que bec City to Montre al [Dronkers, 1969]. The ocean is layered in the vertical with a uniform resolution of 5 m from the surface to 300 m depth, and 10 m below, except for the surface and bottom layers that adjust to the local water level and depth, respectively. [8] The main submodel components described above were found necessary to obtain a periodic seasonal cycle wherein the heat and salt distributions evolve through the internal processes and boundary forcing. The inaccuracies of the forcing at the open ocean boundaries, which is scarcely observed, and the systematic errors in the atmospheric model (found significant in the winds, temperature and cloud cover), strongly limit the precision and accuracy of the quantities that are computed. In the following section, we describe one model realization associated with the exchange coefficients and other parameters that are described in Appendix A. In general, small changes in the model parameters lead to small changes in the solution, and some of these changes may improve the comparisons with the data. Herein we examine robust features of modeled tidal to seasonal variability, and use prior analyses, and the year 1997 as a typical year, to validate some of these features. The model has also been run over four other years and shows qualitatively the same skills from one year to the next. [9] Figure 2 shows the domain-averaged time series of atmospheric and hydrologic forcing over the simulation period. The atmospheric forcing was produced by the Canadian operational regional finite element model prior to February 24, 1997 [Mailhot et al., 1997], and the Global Environmental Multiscale model thereafter [Co te et al.,

4 25-4 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 3. Modeled tidal co-amplitude (A) and co-phase (j) charts for the main semidiurnal M 2 and diurnal K 1 water level harmonic constituents. 1997a, 1997b]. The atmospheric model was run in regional mode with a resolution of 35 km over North America. It provided sequential 12-hour forecasts of 6-hourly fields using uninterrupted series of global analyses for initialization using the variational method in place at the Canadian Meteorological Center, Montreal. The atmospheric model is not coupled to the ice-ocean model. It makes use of global daily analyses of satellite-derived SST and ice concentration. The comparisons of observed and predicted winds at 10 m heights show that the modeled winds are underestimated over the estuary. This error is in part attributed to the unresolved St. Lawrence valley effects and the relatively low near surface vertical resolution. Hourly station data were used to correct the wind stress according to observations on a small island (Bicquette) located in the estuary. [10] The runoff data are prescribed as boundary conditions on momentum and salinity. They are interpolated in time from observations in the 28 most important tributaries (Hydat database, Department of the Environment, Canada), normalized to represent the input from the neighboring drainage basins. The St. Lawrence River runoff is taken from the monthly reanalysis of Bourgault and Koutitonsky [1999]. Figure 2f shows the daily total freshwater input rates. The temperature of the St. Lawrence River is prescribed from continuous observations at Québec City. [11] The initial conditions for temperature and salinity were interpolated for each layer separately using profiles acquired over the GSL in November 1996, and over the estuary in previous years for the lack of data in At the open boundary near Cabot Strait, the temperature and salinity of inflowing waters at depths less than 200 m were specified from the climatology described by Petrie et al. [1996], for the lack of data over the simulation period. Below 200 m depth, T-S profiles acquired at three occasions over the hindcast period were prescribed. In the Strait of Belle Isle, historical T-S observations including those of Petrie et al. [1988] were merged together in order to establish mean monthly values. Not retaining T and S data shallower than 30 m, the following monthly boundary values were prescribed from January to December: T =( 1.55, 1.65, 1.65, 1.5, 1.4, 0.5, 1.8, 2.0, 2.5, 2.0, 1.0, 0.0) C; S = (32.75, 33.2, 33.0, 33.0, 32.9, 32.0, 32.0, 32.0, 32.0, 32.0, 32.0, 32.0). The error on these monthly values is estimated to be within ±0.5 for S, and ±2 C for T. Sea ice concentration and thickness are prescribed in the Strait of Belle Isle from daily charts produced by the Canadian Ice Service. Generally, a significant fraction of sea ice exits the GSL and melts beyond Cabot Strait. Herein sea ice is allowed to leave the computational domain at the open boundaries. [12] The water level is specified at each time step along the open boundaries from 27 water level constituents observed on the shores of each strait. Figure 3 shows the amplitude and phase for the main diurnal and semidiurnal tidal components derived from the modeled hourly water levels. These charts compare well to observations by Farquharson [1962] and Godin [1979]. The present model results are similar to those of Lu et al. [2001] for both the tidal ellipses and tidal mixing (as shown below). Shelf waves produced both in the ocean and in the GSL produce synoptic sea level changes of the order of 10 1 m over periods of 2 to 5 days [Bobanovic and Thompson, 2001]. On seasonal timescales, the propagation of these waves clearly represents a minor source of kinetic energy compared to tides. Herein offshore-generated events are not accounted for. 3. Results 3.1. Gulf-Averaged Seasonal Cycle [13] Figure 4a shows the modeled 6-hourly and monthly mean times series of the ocean-atmosphere heat flux (1-A)Q AO + AQ IO averaged over the GSL. The heat flux exhibits a monthly averaged seasonal cycle of 180 W m 2 amplitude, characterized by large sensible heat loss in the

5 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE 25-5 Figure 4. Six-hourly laterally integrated net oceanic heat and salt fluxes over the simulation period (positive upward). (a) Heat flux Q O =(1-AI ) Q AO + AQ IO ; (b) salt flux. fall and winter and large radiative heating during the spring and summer. The monthly averaged differences in the heat flux among regions of the GSL, controlled by the local stratification and sources of mixing, are generally as large as the seasonal variations. In late fall and early winter of both years, only a few strong wind events control the seasonal mean heat loss before the ice cover forms. For example, two strong wind events with surface stress near 60 Pa occurred during the last week of December 1996 (Figure 2b). The first one is associated with rather normal air temperature and a sensible heat flux maximum of 380 W m 2, while the second one, lasting 4 days, is associated with cold air and a maximum sensible heat flux of 650 W m 2. This single event extracted 10% of the seasonal heat stored in the GSL (about J). The frequency and intensity of such events during the fall and early winter determine the stratification and heat content of the upper 100 m at the onset of sea ice, those having a strong effect on sea ice growth in the months to follow. This includes the heat and freshwater contents of the surface waters exiting through Cabot Strait. The monthly mean heat flux is maximum in January (177 W m 2 in 1997 and 166 W m 2 in 1998). Thereafter the ice cover and stratification at depth greatly reduce the heat flux, which becomes 37 W m 2 on average in March. Surface warming begins in April and is maximum in June with an average monthly intake of 180 W m 2 dominated by incident short wave radiation. The cloud cover carries large uncertainties in the atmospheric model, and sensitivity studies show a change of 20% during the April July period could change the surface mixed-layer temperature by 1 to 2 C until the fall. The sign of the heat flux changes in mid-september as the air becomes colder than the SST and incident radiation decreases. The oceanatmosphere salt flux (Figure 4b) is controlled by ice growth/ melt rates during winter and spring (shown below), and P-E v during the ice-free season. [14] Figure 5 shows the mean seasonal cycle of temperature, salinity, and turbulent quantity profiles averaged over the GSL. The horizontal variability is generally as important as these temporal and depth changes, and will be examined next. The heat and salt contents obey a wellresolved seasonal cycle in the upper 100 to 200 m depth. Significant turbulent energy associated with strong wind events partly mixes the water column to about 100 m depth already during fall. The maximum mixed layer depth reaches about 50 to 150 m in March, against a thermohalocline above the relatively warm and salty bottom Atlantic waters. Salt rejected from sea ice formation decreases stratification and produces positive contributions to the turbulent kinetic energy through upward density gradients. Salt rejection is a relatively small direct contribution to the erosion of the halocline, however, as the vertical salinity gradient there may exceed 10 1 m 1. The production of 1 m of sea ice, for instance, would increase the mixed-layer salinity by about Sea ice growth rates and oceanic fluxes directly depend upon strong wind energy capable of eroding the winter halocline except in three regions: (1) the southern GSL, where the mixed-layer reaches the bottom, (2) the estuary and northwestern GSL where the stratification remains relatively high, and (3) the northeastern GSL where relatively dense intruding Labrador shelf waters may define the halocline. [15] Modeled waters with T <0 C and S > 32 fill 18% of the GSL at the end of the winter of 1997 ( m 3 ), fill less than 5% in midsummer, and nearly vanish in November. For comparison, Forrester [1964] founds a volume of water with T <0 C and S >32of m 3 during late winters of 1956, 1957, and Lauzier and Bailey [1957] found that waters with T <0 C occupy a maximum of about 26% of the GSL at the end of winter, 12% in summer, and 4% in the late autumn. The modeled waters with T <1.5 C and S > 32, represent about 32% of the GSL volume at the end of the spring, and persist to represent about 13% of the volume at the end of November [16] The modeled fall and winter inflow of Labrador shelf waters through the Strait of Belle-Isle is 0.4 Sv on average and contributes 44% of the volume of waters with T <0 C found at the end of March. This important contribution carries large uncertainties due to the lack of data near the boundary, but is in rough agreement with previous estimates of about 26 to 50% by Banks [1966] and Petrie et al. [1988]. [17] A new highly stratified surface layer is produced in the spring as the sea ice melts and runoff increases from continental snowmelt. The surface salinity decreases by 2 units and is minimum at the end of July, after the freshwater disperses throughout the GSL. The surface temperature

6 25-6 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 5. Domain-averaged hourly time series of modeled S, T, and turbulent quantities as functions of depth (in meters). Time series over the year 1997 are shown on the left and yearly averaged profiles on the right. (a) Salinity; (b) temperature; (c) logarithmic turbulent kinetic energy log10(e/m 2 s 2 ); (d) same as Figure 5c but for the lower estuary only; (e) turbulent length scale l; (f ) vertical eddy diffusivity Kvs. Note that higher values near 100, 200, and 300 m depths are related to the relatively high weighting of these depth contours in the interpolation of the grid topography. increases by over 14 C between ice melt and early August. The averaged mixed layer depth, defined as the depth of the maximum vertical density change, is approximately 10 m in the spring and reaches 30 m at the end of August. This layer partly insulates the cold winter waters from the atmosphere throughout the summer. [18] Figures 5c and 5d show the time series and annual mean laterally averaged profiles of the logarithmic turbulent

7 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE 25-7 kinetic energy, log 10 (E/m 2 s 2 ), over the GSL and the estuary, respectively. The turbulent master length scale, l, and vertical eddy diffusion coefficient, Kvs, also averaged over the GSL, are shown in Figures 5e and 5f, respectively. The maximum values of E correspond to strong winds over open water associated with low winter surface stratification. For instance, the large extratropical storm event moving through between March 13 to 15, 1997, is not associated with a large heat flux (warm air) but with high momentum flux producing significant sea ice displacement and mixing down to the bottom of the GSL (as for the late 1996 storm). The neap-to-spring tidal cycle is clearly visible in bottom waters of the lower estuary (Figure 5d), while the nearly permanent stratification there inhibits the downward diffusion of wind-induced turbulent energy below approximately 50 m depth during winter. The limiting turbulence length scale (Figure 5e) is generally the Ozmidov scale l u, as opposed to the parabolic function l d (see Appendix A), except near the surface and bottom (note that the bottom of the LC is not shown on Figure 5). The master length scale is strongly influenced by the seasonal stratification, reaching around 10 m at depths near 30 m in the deeper waters during winter (not shown), but reduced to less than 1 m in the summer surface layer. In the next section, we examine the modeled horizontal variations in the surface temperature and salinity controlling the surface stratification and the ocean-atmosphere heat fluxes Sea Surface Temperature and Salinity [19] Figure 6 shows the modeled seasonal mean SST and SSS over the year The T-S solution for 1997 is generally within the 1s range of observed historical values analyzed by Petrie et al. [1996]. During winter (Jan Mar), the SST is near freezing everywhere except in the Cabot Strait area and at the head of the LC. At the latter location, the surface water temperature is 0.5 to 1 C on average. This water is in contact with cold air and is surrounded by near-freezing surface waters, and its salinity is 1 unit higher than the surrounding waters. This is a result of tidal mixing of the relatively warm salty deep waters with the surface waters, and shows little sensitivity to actual air temperature. This was first predicted by the model and then observed from in situ surface T-S observations acquired with a temperature-conductivity sensor sampling at 8 m depth from onboard a ship transiting in the estuary every week (not shown). Small increases in SST also generally appear at tidal and synoptic scales in regions of upwelling and upward diffusion of the bottom waters. SSS is maximum everywhere at the end of winter because the runoff is lowest, the mixed layer deepens everywhere, sea ice is produced, and inflows take place through the Strait of Belle Isle. [20] Starting in early April, meltwaters from sea ice and continental snow begin covering the estuary, the northwestern GSL, and the lower north shore near major tributaries (Figure 6b). Freshwater exiting the estuary partly recirculates in the northwestern GSL and moves via the Gaspé Current toward the southern GSL. The heating rate of the near-surface waters is maximum in the estuary and the Gaspé Current waters, where stratification is maximum, and in the shallow regions of the southern GSL. During summer, the Gaspé Current waters spread in the southern GSL producing a lagged minimum SSS in late summer there [e.g., Sutcliffe et al., 1976]. At this time, freshwater from the estuary begins exiting the GSL through western Cabot Strait. [21] SST further increases to a summer maximum over 20 C in the stratified and shallow waters of the southern GSL and about 10 C or less in the estuary and northern GSL. While tidal mixing at the head of the LC produces a positive SST anomaly during winter, it clearly does the opposite during summer as the CIL is mixed with the surface waters [e.g., Kelly, 1832, 1837; Gratton et al., 1988; Saucier and Chassé, 2000], producing cold surface conditions in the lower estuary. The summer SST is further controlled by synoptic upwelling events along the north shore of the GSL and south shore of Anticosti Island. The modeled separation of the Gaspé Current from the coast, as evidenced from the modeled SSS and SST during summer, and its subsequent partial recirculation over the northwestern GSL, reproduce the known baroclinic instability of that current [e.g., Dawson, 1913a, 1913b; Mertz et al., 1988; Benoit et al., 1985; Sheng, 2001]. The results for the SSS distribution are in general agreement with the observations. However, SSS is underestimated in the region of influence of the freshwater moving through the estuary, as examined in section 3.4. [22] During fall, both the sensible heat and momentum fluxes increase. SST decreases and SSS increases due to buoyancy loss and the thickening of the mixed layer. The winds destroy stratification before winter not only through turbulent energy production and buoyancy loss at the surface, but also directly through the forced evacuation of the surface waters from the stratified regions of the estuary and the southern GSL Regional Distribution of the Oceanic Heat Fluxes and Turbulent Energy [23] Figure 7 shows the seasonal mean net oceanic heat flux distribution through 1997 (complementing the domain average shown in Figure 4a), illustrating the combined effects of depth, stratification, winds, and tides on the exchanges with the atmosphere. The wintertime average oceanic heat flux is between 150 and 200 W m 2 where northwesterly winds produce upwelling (along the north shore of the estuary and GSL, and on the south shore of Anticosti Island), at the heads of the LC and the Jacques Cartier Strait because of tidal mixing, and in the Cabot Strait area where relatively warm waters enter the GSL. The heat flux is minimum in the southern GSL, because of shallow depths and a nearly complete ice cover, and along the western shore of Newfoundland due to the presence of thick ice. During spring, the heat gain (negative heat flux) is a function of cloudiness (increasing eastward). During spring and summer, the large heat intake in the estuary is associated with low SST maintained through the mixing of CIL waters with surface waters at the head, and upwelling on the north shore. Stratification and depth limit heat intake in the southern GSL. During the fall, the maximum heat loss is about 150 W m 2, found along the north shore and in the southeastern GSL, where the warm summer surface waters are converging to exit. These results generally agree with Doyon and Ingram [2000] except that here the model

8 25-8 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 6. Seasonally averaged sea surface conditions. (a) SST ( C); (b) SSS. suggests higher sensible heat intake during midsummer in the northern GSL. [24] In order to complement Figure 5, Figure 8 shows the depth and seasonally averaged logarithmic turbulent kinetic energy distribution. The strong seasonal cycle, with about 10 times more energy in the winter and fall than during the spring and summer, is a function of wind strength and stratification. Tidally produced turbulent energy is significant at the head of the LC, in the Jacques-Cartier Strait, in the Strait of Belle Isle, in the Northumberland Strait, and in the shallow areas near the coast in the southern GSL. These results complement Pingree and Griffiths [1980] and Lu et al. [2001] in modeling tidal mixing in the GSL. During fall and early winter, winds produce high energy levels in the southern GSL, along the lower north shore, and over the bank near eastern Anticosti Island. The mean depth-averaged turbulent energy production is strongly inhibited by stratification in the lower estuary and northwestern GSL yearlong Comparisons With Temperature and Salinity Observations [25] Figure 9 shows the positions of 18 nearshore thermometers, and 675 temperature and 649 salinity profiles, available for Figure 10 shows the observed and modeled nearshore temperature. The overall mean hourly

9 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE 25-9 Figure 7. Seasonally averaged net oceanic heat flux (1-A) Q AO + AQ IO over the simulation period (W m 2 ). Figure 8. Seasonally and vertically averaged logarithmic turbulent kinetic energy log 10 (E/m 2 s 2 ).

10 25-10 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 9. Positions of coastal temperature observations and CTD profiles acquired during the simulation period. error is 0.3 C with a standard deviation of 2 C. The model resolves the seasonal cycle well and is seen to pick up tides, and some synoptic variability at timescales of 1 to a few days associated with wind-induced coastal upwelling [e.g., Koutitonsky and Bugden, 1991] or with the horizontal circulation. This is particularly evident along the western shore of Newfoundland (instruments 1 and 2), along the Anticosti south shore (instrument 8) and along the lower north shore (instrument 4). Spring warming and fall cooling near the Magdalen Islands (instruments 9 and 10) and at the northern mouth of Baie des Chaleurs (instruments 13 and 14) are well reproduced to within about 1 C. Note the strong neap to spring variability with an amplitude of about 5 C at instrument 18 near the head of the LC. [26] Figure 11 shows samples of observed and modeled T-S profiles in winter, summer and fall, as well as the mean error profiles for each season. Individual comparisons are relatively good, with some small-scale vertical gradients being well resolved. In places the differences are quite large, for example, where the model overestimates the depth of mixing in the LC during winter (last comparison in first row of Figure 11). The largest errors in temperature occur near the pycnocline (±2 C), as small errors in its modeled depth lead to large differences during the high stratification state. The error distribution shows that the model underestimates mixed layer temperature and salinity in the northwestern and southern GSL. This is only marginally related to mixing depth as evidenced in the individual comparisons in Figure 11. Sensitivity experiments point to different controls on this error. Two mechanisms could eliminate the bias in SSS, namely increased wind drag on the surface or enhanced mixing in the upper estuary. The analyses indeed suggest that the wind strength is still being underestimated in the northwestern GSL, particularly during winter By artificially enhancing mixing in the upper estuary and head region of the LC (e.g., doubling the eddy diffusion coefficient), the bias in SSS could also be greatly reduced. The source of error in SST could be explained from the biases in the short wave radiation (e.g., cloud cover) or sensible heat exchanges (e.g., wind strength). Changes in the radiation have the most direct and isolated effect on SST. In general, the errors in the atmospheric fields were found sufficient to explain the model errors from the synoptic to seasonal timescales, although no further attempt was made to correct the atmospheric fields from observations Sea Ice Conditions [27] Figure 12 shows the high-frequency and monthly averages of the domain-averaged sea ice volume, ice growth rate, and ocean-ice heat fluxes. The modeled ice volume is compared in Figure 12a with the composite estimates from satellites, aircraft, and ship reconnaissance (Canadian Ice Service, Ottawa). When estimating the volume from these charts, the expected midthickness for each ice category was used. This generally leads to an error of over 35% on the observed ice volume, especially during the periods of rapid changes in thickness. The ice data were not available toward the end of the season. During the peak of the season in early March, the ice volume reaches about 75 km 3 in 1997 and 35 km 3 in 1998, in general agreement with the observations. Figure 12a also shows the cumulative volume of ice exported through Cabot Strait. About 30 km 3 of sea ice is exported during 1997 (Figure 6a), and very little in Reports confirm that sea ice exited in 1997 and not in 1998 [Drinkwater et al., 1999b]. It is known that ice conditions in Cabot Strait are extremely variable from one year to the next [e.g., Dickie and Trites, 1983; Côté, 1989; Drinkwater et al., 1999a]. Figures 12b and 12c show the sea ice growth rates. The extreme values are associated with instantaneous (hourly) gulf-averaged ice growth/melt events near 2 cm day 1 (1 cm day 1 = ms 1 ). The mean monthly open water growth rate is maximum in February 1997 (0.25 cm day 1 ). The ice growth rate is sensitive to heat loss and stratification in the fall and early winter. Sensitivity experiments show that reduced wind

11 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 10. Comparison with temperature records acquired in 1997 at the positions shown on Figure 9. The observations are in blue and the model results in red (within the corresponding model layer). The station number is displayed along with the depth of the instrument. The two numbers below the station number represent the hourly mean and standard deviation errors. mixing during the fall leads to significantly higher growth rates in January, but smaller ones in February and March. On the other hand, increased mixing (reduced stratification) during early winter is associated with smaller but more persistent growth leading to a more extensive and thicker ice cover. [28] Figure 13 shows the observed and modeled monthly ice conditions for February and March In Figures 13a and 13b, the monthly averaged ice concentration and growth rate are shown for February and March Figure 13c shows the corresponding observed monthly composite ice concentration. These observations are known to be biased toward higher values, first because they neglect the rapid deformation of the ice cover (e.g., wind-driven leads on the north shore) and second for navigation safety. Figure 13d shows the modeled monthly averaged ice velocities. The results generally agree with historical information on ice cover, growth, and drift [e.g., Lauzier and Graham, 1958; Forrester and Vandall, 1968; Murty and Smith, 1973; Côté, 1989; Drinkwater et al., 1999a]. The comparisons with ice charts (Figure 13c) show that the model reproduces the observed patterns but with lesser concentrations. Regions of minimum and maximum concentrations generally agree. In January 1997, sea ice is initially produced in the estuary, the northwestern GSL, along the north shore, and along the southeastern GSL coast. In February, sea ice production becomes important throughout the western half and northeastern GSL. Sea ice formed in the western GSL flows through the Laurentian Channel to partly melt eastward. The sea ice produced in the Jacques Cartier Strait is moved by winds to cover the eastern GSL and accumulate against the Newfoundland coast by mid-february, while the ice cover increases further in the southern GSL and begins exiting through Cabot Strait. The comparison confirms that the head of the LC is ice free most of the winter. Wind-driven leads are formed throughout the winter along the north shore and the southern Anticosti shore. Light ice conditions also occur in the LC southeast of Anticosti Island. We also note that

12 25-12 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 11. Differences between observed and modeled T-S profiles. The data are split into three 4-month periods in 1997 from top to bottom. Sampled comparisons from the months of March, June, and September are displayed to the left (positions shown in decimal N, W; red: temperature; blue: salinity; thin line: observed; thick line: modeled). The panels on the right show mean and standard deviation errors versus depth for n records of temperature and salinity available in each period. melting occurs throughout the winter at the head of the LC and in the region extending from Cabot Strait to the Laurentian deep and Esquiman deep (dark blue in Figure 13b). Mean monthly ice velocities (Figure 13d) show that the sea ice produced in the northwestern GSL and estuary flows at about 0.5 m s 1 toward Anticosti Island and then veers southeastward toward the Gaspé peninsula. This movement also exposes waters that are more or less prone to freeze, and adds to the other forcing to produce highly variable ice growth rates and surface fluxes [e.g., Lauzier and Graham, 1958] Circulation [29] The model results show a mean annual inflow of 0.45 Sv through eastern Cabot Strait and a 0.81 Sv outflow through western Cabot Strait, in rough agreement with past studies of currents there [MacGregor, 1956; Trites, 1972]. The synoptic variability is associated with instantaneous transports of about 2 Sv over each section at Cabot Strait (near 5 Sv during the March storm event), and up to 1 Sv through the Strait of Belle Isle. These events weigh significantly in the variability of the seasonal transports. In the present simulation, the mean transport is minimum during winter and maximum during the fall. The modeled inward transport through the Strait of Belle Isle is 0.33 Sv on the average in 1997, highly variable on synoptic timescales, with a larger mean inflow during the fall and winter months (mean 0.42 Sv) and minimum inflow during spring (0.15 Sv), in rough agreement with Petrie et al. [1988] (who derived 0.3 Sv for winter and 0.13 Sv for summer using measurements acquired in earlier years). [30] Figure 14 shows the mean seasonal surface (0 30 m) circulation, and the yearly mean circulation at depths between 50 to 100 m (Figure 14b), and below 200 m (Figure 14c). As shown in Figure 14a, the surface circulation patterns are similar to those from El-Sabh [1976] and other observed mean currents [Trites, 1972; Gregory et al., 1989]. The yearly mean and seasonal surface currents obey the rather well-defined patterns of density changes and wind stresses. The St. Lawrence runoff and its dispersion in the northwestern and southern GSL control the circulation there. The 1997 fall circulation shows intense wind-driven surface transport in the southern GSL. Model simulations suggest that the interannual variability is greater than about 10 1 ms 1. For example, during winter 1997 the model

13 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 12. Laterally integrated modeled ice properties and related fluxes over the simulation; (a) modeled (thick line) and observed (thin line) daily ice volume inside the domain. Note that observations were not available toward the end of the ice season. The ramping function shows the cumulative ice volume exported through Cabot Strait. Also shown are both the 6-hourly and monthly averaged results for (b) open water ice growth rate; (c) ice growth rate at the ice-ocean interface; and (d) sensible heat flux at the ice-ocean interface. produced mean reversed circulation along the Gaspe coast and a northern exit at the mouth of the estuary (see Figure13d). Such model predictions cannot be verified with data. In contrast, the winter surface circulation modeled for 1998 displays a pattern closer to the annual mean in the NW GSL, showing the strong influence of the ice cover there [e.g., Dawson, 1897]. [31] The mean circulation pattern along the Gaspe peninsula is indicative of two modes of circulation along the coast, one in which the Gaspe Current remains attached to the coast, and the other one in which it is unstable and separates to partly recirculate in the northwestern GSL [e.g., Bayfield, 1860; Dawson, 1913a, 1913b; Mertz et al., 1991; Sheng, 2001]. The transports through the surface 50 m in the northwestern GSL agree rather well with those obtained by Bugden [1981] for March to December. In this same period, the modeled transport out of the estuary is m3 s 1, the same as derived by Bugden [1981], while the modeled winter value is 25% higher. The net upward transport through the northwestern GSL at 50 m depth, gives 0.18 Sv, while Bugden [1981] computed 0.17 Sv. As shown by Mertz et al. [1991], the maximum transport through Honguedo Strait occurs during the fall and winter (0.27 Sv and 0.3 Sv) and is minimum in the spring and summer (0.13 Sv and 0.22 Sv, respectively). Finally, the model roughly agrees with Bugden s estimate of transport from North to South east of Anticosti Island. The model shows that the 1997 transport in the 0 50 m depth range is 0.19 Sv, while Bugden [1981] found 0.13 Sv. [32] The mean 1997 circulation is shown for intermediate and bottom depths in Figures 14b and 14c. The circulation in the CIL between 50 and 100 m depth (Figure 14b) shows a general cyclonic circulation of the order of 0.1 m s 1. Relatively dense waters entering through the Strait of Belle Isle, the withdrawal of intermediate waters at the head of the LC, and the surface circulation govern the circulation at intermediate depths. To illustrate the relationship between the circulation and the inflow of waters through the Strait of Belle Isle, a tracer was added following the same form as equation (A4). The initial field was zeroed, and a unit value was prescribed at the open boundary. Figure 15 shows the mean tracer value between 50 and 100 m depth on October 23, This may be interpreted as the fraction of the water column between 50 and 100 m depth that is made of Labrador shelf waters. The synoptic to seasonal variability of the inflow and the circulation is large, but Figure 15 shows some of the more common features. The waters from the Strait of Belle Isle move along the northwestern shore and flank of the Anticosti Channel, before splitting into a branch flowing through the Jacques Cartier Strait, and a second one circulating around Anticosti Island in the LC. The transport is far more important in the first branch. The LC branch is unstable in the Honguedo Strait, separating from the northern LC to move under the Gaspe Current. It seems to take part of the second, this time anticyclonic, gyre structure seen in the surface currents and associated with the instability of the Gaspe Current (see Figure 14a). Both branches meet in the northwestern GSL and begin to leak into the Anticosti Gyre about 4 months after entering the GSL. The maximum

14 25-14 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE

15 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 14. Modeled horizontal currents. (a) Seasonally and depth averaged (0 30 m) near-surface currents in 1997; (b) depth-averaged ( m) annual mean currents in 1997; (c) depth-averaged ( m) annual mean currents in transport into the gyre occurs in August and September Once these waters enter the northwestern GSL, they are brought toward the mouth of the estuary within about 1 month. The flow splits into a branch moving into the estuary toward the head, and a second one that either recirculates into the northwestern GSL, or flows along the southern LC and Gaspé Peninsula toward Cabot Strait, leaking into the southern GSL. The estimated modeled residence time of the Labrador shelf waters that transit through the estuary and the northwestern GSL is about 1 year. [33] The circulation deeper than 200 m (Figure 14c) is driven from the relatively dense waters inflowing through Cabot Strait and their mixing and withdrawal at the head of the LC. The mean velocity is about 1 cm s 1, in rough agreement with Bugden [1991]. The circulation from Cabot Strait follows the LC and branches out into Esquiman Channel and in the Anticosti Channel. The double-gyre structure of the northwestern GSL is also a dominant feature at these depths. There is recirculation in the LC, in the Esquiman Channel, and in the Anticosti Channel. The circulation shows that a maximum inflow of about 2 cm s 1 occurs during winter. Since the properties of these waters have small seasonal variations, their study shall be the object of further work on the no less important interannual variability of the circulation processes in the GSL. 4. Discussion [34] The error analyses suggest that mixing is not sufficiently high in the upper estuary. This is expected given the model resolution provides merely a few grid points to represent the complex and energetic topographic effects on tides there [e.g., Saucier and Chassé, 2000]. Other small-scale processes are not taken into account well, given that the 5-km grid does not properly resolve the internal Rossby radius of deformation O(10 km). It is also not yet clear how nonhydrostatic effects need to be considered during winter density inversions associated with surface buoyancy loss. [35] Mixing will indeed remain a more general and difficult problem given that turbulence is highly variable and difficult to measure. Herein turbulence modeling using an energy equation that allows the diffusion of windproduced energy downward was found rather adequate to produce a seasonal cycle without restoring conditions to observations. This suggests that the processes governing the heat cycle in the surface 200 m were reasonably well reproduced, but more work needs be done, especially using situ measurements of turbulent quantities and improving turbulent models of stratified flows. [36] Corrections in the surface momentum and heat fluxes are required to reduce the mean bias toward lower SST and SSS found in the model for the northwestern and southern GSL. It was surprising how some choice of parameters, namely a so-called high-flux state wherein the bulk exchange coefficients for sensible heat and momentum exchanges are set to unreasonably high values (i.e., twice higher), along with a super equilibrium level 2 energy equation, could significantly reduce the errors in the ice cover concentration, SST, and SSS. Figure 13. (opposite) Monthly averaged ice conditions for February and March (a) Modeled concentration; (b) modeled growth rate ( 10 7 ms 1 ); (c) observed ice concentration from averaged daily ice charts in February and March 1997; (d) modeled mean monthly ice velocity.

16 25-16 SAUCIER ET AL.: SEASONAL CYCLE IN THE GULF OF ST. LAWRENCE Figure 15. Fraction of waters between 50 and 100 m depth that have intruded through the Strait of Belle Isle between December 1, 1996, and October 23, [37] The complex effect of the dynamic ice cover and oceanic heat fluxes on the lower layer of the atmosphere limits our capability to establish accurate forcing fields [e.g., Gustafsson et al., 1998]. For example, during wintertime cold events, the ice surface may become cold, but open waters remain with near the freezing temperature. This leads to highly variable stability of the lower atmosphere over relatively small spatial scales. Indeed, when comparing the coastal temperature on Anticosti Island and the Magdalen Islands with the atmospheric model temperature, it is found that the model carries larger errors during winter, especially during cold events in January One may not expect to accurately establish the atmospheric forcing without a major step in model precision, that is, to develop a set of equations for a consistent coupling of momentum, heat and mass transfers with a high-resolution atmospheric model. This work is under way with the Global Environmental Multiscale model GEM [e.g., Roy et al., 1999], and with the Canadian Regional Climate Model [Laprise et al., 1998]. [38] The results show that the CIL-type waters entering the GSL through the Strait of Belle Isle spread throughout and partly renew the intermediate waters of the estuary within 6 to 10 months. The inflow of waters through the Strait of Belle Isle is clearly very important in the heat, salt, and momentum budget of the GSL [e.g., see also Bugden, 1981; Petrie et al., 1988]. The lack of data on water properties and sea level in the Strait of Belle Isle is thus very limiting until we can properly monitor the conditions there. 5. Summary [39] The first model of the seasonal cycle of the GSL iceocean conditions was successfully implemented. It was validated against available temperature and salinity data through a 16-month simulation. The model reproduces instantaneous water temperatures with a mean error of about 1 to 2 C, salinity with a mean error of 0.5, long-term transports within about 20% of published estimates for other periods, and the monthly sea ice volume within 20%. The winter in situ cold water mass formation and the inflow of waters through the Strait of Belle Isle are in relative agreement with the values reported for other periods, given the scarcity of observations near the open boundaries and the inaccuracies of the atmospheric forcing. The results suggest that we may begin to model the interannual variability to some extent given the available boundary forcing. It also suggests that sensitivity experiments related to climate variability and change (e.g., changes in runoff from the Great Lakes, warmer air temperature, increased intensity or frequency of extreme events, changes in Labrador shelf waters) could be carried out. [40] The model shows the relationship between turbulent mixing, stratification, and circulation over the tidal to seasonal timescales. In particular, we mention two features of the circulation derived from the model: (1) The head of the LC is a typical wintertime sensible heat polynya whereby the deep warmer Atlantic waters are tidally uplifted and partly mixed into the surface waters to maintain relatively warm (approximately 0.5 to 1 C) surface conditions, preventing and even melting sea ice. However, one notes that surface current divergence also prevents sea ice to accumulate there. This was confirmed with observations. (2) The sharp CIL structure found during summertime in the estuary is not formed in situ, but is rather advected there from the Jacques Cartier Strait and the LC. In November 1997, the model result suggest that 30% to 40% of the waters between 50 m and 100 m in the estuary were brought there from the Labrador Shelf during the previous period of 6 months to a year. [41] The analyses of the errors show that the resolution and accuracy of the atmospheric forcing, and the resolution of the tidal currents in the upper estuary and at the head of the LC, are important limiting factors. Both the atmospheric and oceanic models will increase in resolution as the computing power becomes available, and the results suggest that this will improve the solutions. Preliminary work on fully coupled atmosphere and ocean components show significant effects in near surface conditions both in the atmosphere and the ocean. The sensitivity of the wind fields to ice conditions, for example, suggests that more consistent conditions will be required for more accurate atmosphere-ocean solutions. Appendix A: Model Formulation [42] The equations for the momentum, heat, salt, and turbulent kinetic energy can be written as (the comma subscript denotes partial derivative) Du Dt fv þ r 1 P ; x A H u ; x ; x A Hu ; y ; y K VMu ; z ¼ 0; ða1þ ; z Dv Dt þ fu þ r 1 P ; y A H v ; x ; x A Hv ; y ; y K VMv ; z ¼ 0; ða2þ ; z rðu; v; wþ ¼ 0; ða3þ DT Dt A HT ; x ; x A HT ; y ; y K VsT ; z ; z ¼ 0; ða4þ DS Dt A HS ; x ; x A HS ; y ; y K VsS ; z ¼ 0; ða5þ ; z DE Dt K VM E ; z ; z ¼ K VM u 2 ; z þ g v2 ; z þ K Vs r r ; z q3 B 1 l ; ða6þ