1
The seasonality of convective events in the Labrador Sea 1
Hao Luo1, Annalisa Bracco1*, Fan Zhang1 2
1School of Earth and Atmospherics Sciences, 3
Georgia Institute of Technology, Atlanta, GA, 30332, USA 4
5
6
7
8
9
10
11
12
13
Corresponding author: Dr. Annalisa Bracco 14
School of Earth and Atmospheric Sciences 15
Georgia Institute of Technology 16
Email: [email protected] 17
2
18 Abstract 19
Modeling deep convection is a key challenge for climate science. Here two simulations of the 20
Labrador Sea circulation obtained with the Regional Oceanic Modeling System (ROMS) run at a 21
horizontal resolution of 7.5 km are used to characterize the response of convection to 22
atmospheric forcing, and its seasonal variability over the period 1980-2009. The integrations 23
compare well with the sparse observations available. The modeled convection varies in three key 24
aspects over the 30 years considered. First, its magnitude changes greatly at decadal scales. This 25
aspect is supported by the in-situ observations. Second, the initiation and peak of convection (i.e. 26
initiation and maximum) shifts by two to three weeks between strong and weak convective years. 27
Third, the duration of convection varies by approximately one month between strong and weak 28
years. The last two changes are associated to the variability of the time integrated surface heat 29
fluxes over the Labrador Sea during winter and spring, while the first results from changes in 30
both atmospheric heat fluxes and oceanic conditions through the lateral inflow of warm Irminger 31
Water from the boundary current system to the basin interior. Changes in surface heat fluxes 32
over the convective region are linked to large scale modes of variability, the North Atlantic 33
Oscillation and Arctic Oscillation. Implications for modeling the climate variability of the 34
Labrador basin are discussed. 35
36
37
38
39
40
3
1. Introduction 41
The Meridional Overturning Circulation (MOC) plays a critical role in transporting heat 42
across latitudes in the ocean, and its variability has significant impacts on the global climate 43
system (Bryden et al., 2005). The North Atlantic branch of the MOC is characterized by deep 44
water formation in few specific locations, including the Nordic Seas and the Labrador Sea 45
(Dickson et al. 1996; Marshall et al. 2001). Here in winter and early spring the ocean releases 46
heat to the atmosphere and the surface waters become dense enough to mix by convective 47
instability (Kuhlbrodt et al. 2007). 48
In the Labrador Sea, deep convection takes places in the deeper portion of the basin, 49
seaward of the western continental slope, and forms the Labrador Sea Water (LSW) that often 50
reaches as deep as 2000 m, exits the Labrador Sea and becomes a distinct component of the 51
North Atlantic Deep Water, feeding the MOC (Talley and McCartney 1982; Marshall et al. 1998; 52
Lazier et al. 2002; Rhein et al. 2002; Yashayaev et al. 2003; Yashayaev 2007). The LSW 53
variability therefore influences the MOC, and it has been suggested that an intensification of the 54
convective activity in the Labrador Sea will lead to an intensification of the MOC, with an 55
overall increase in poleward heat transport (Eden and Willebrand, 2001). 56
The formation of LSW displays strong interannual variability that is influenced by the 57
atmospheric fluxes, which are highly variable at the latitudes in consideration, and by the 58
characteristics of the Irminger Current (Myers et al. 2007; Rattan et al. 2010). The Irminger 59
Current flows along the Greenland coast and its warm and salty subsurface water is brought to 60
the Labrador Sea interior through anticyclonic mesoscale eddies, the Irminger Rings (Lilly et al. 61
2003; Katsman et al. 2004; Bracco et al. 2008; Hátún et al. 2007; Chanut et al. 2008; Rykova et 62
4
al. 2009; Gelderloos et al. 2011; Luo et al. 2012). Since the mid-1990s the Irminger Current has 63
experienced continuous warming, which has been linked to the concurrent weakening of the 64
subpolar gyre (Holland et al. 2008). Such weakening, in turn, may be the expression of the 65
subpolar gyre decadal variability (Häkkinen and Rhines 2004; Böning et al. 2006). 66
The Labrador Sea convection has been sampled by highly frequency hydrographic 67
measurements the Ocean Weather Station Bravo (OWSB) from 1945 until 1974 with 24 years of 68
continuous recording (e.g. Lazier, 1980; Sathiyamoorthy and Moore, 2002), by moorings and P-69
ALACE floats in the second half of the 1990s (e.g. Marshall et al., 1998, Lavender et al., 2000; 70
Avsic et al., 2006), by gliders (Frajka-Williams et al. 2014), by hydrographic surveys usually 71
conducted in late spring or summer and at least yearly from 1990 by the Bedford Institute of 72
Oceanography (BIO), and by ARGO profiling floats since June 2002. The surveys primarily 73
sample the AR7W WOCE/CLIVAR line extending from Newfoundland (53.67° N, 55.5° W) to 74
the west coast of Greenland (60.5° N, 48.25° W), and have been extensively analyzed (e.g. 75
Lazier et al. 2002; Yashayaev et al. 2003; Lu et al. 2006; Yashayaev 2007; Yashayaev et al. 2007; 76
Yashayaev and Loder 2009). ARGO floats, on the other hand, provide temperature and salinity 77
measurements of the upper 2000 m of the water column, and represent a continuous in time but 78
irregularly in space data set (Yashayaev and Loder 2009; Luo et al. 2012). 79
Focusing on the last three decades (1980-2009) in-situ data show that the convection in the 80
Labrador Sea intensified in the late 1980s and early 1990s, and weakened since 1995, with a 81
limited recovery in the 2007-2008 winter (Våge et al. 2008; Yashayaev and Loder 2009). 82
Various external causes have been proposed to influence the onset and intensity of convection, 83
including local and remote atmospheric forcing, the Irminger Current conditions, and the state of 84
the subpolar gyre (Marshall and Schott 1999; Lazier et al. 2002; Straneo 2006; Straneo et al. 85
5
2010). The variability of convective events, other than their interannual intensity, however, have 86
been – and can be - investigated only for limited time spans using observations (Straneo, 2006; 87
Gelderloos et al., 2012) due to their sparseness in space and/or time. 88
On the modeling side, until recently the attribution of variability of convective events in the 89
Labrador Sea have been attempted only for decadal modulations (Mizoguchi et al., 2003) due to 90
the generally poor representation of the details of LSW formation (Canuto et al., 2004; Tréguier 91
et al., 2005). Luo et al. (2012, LBYD12 in the following) using a regional ocean model were able 92
to reproduce the interannual component as well, and to quantify the relative importance of the 93
atmospheric forcing and of the Irminger Current conditions. Here we build on two of the 94
numerical simulations described in LBYD12 to zoom on the seasonal scales. We analyze how 95
the time of initiation, peak, and seasonality of convection vary in years of intense versus weak 96
activity, with immediate application to monitoring planning. Additionally, we explore the drivers 97
of those changes, and we discuss the modeling challenges. Both ocean-only and coupled climate 98
models display large biases and divergent behaviors in simulating the formation of deep water 99
masses and their variability (Canuto et al., 2004; MacMartin et al., 2013). By analyzing the 100
interplay of oceanic and atmospheric forcings in the representation of the seasonal cycle and of 101
the interannual modulation of LSW in the North Atlantic in realistic simulations, we hope to 102
provide a base to validate and test the representation of the Atlantic MOC in coupled climate 103
models. 104
2. Model configuration 105
The Regional Ocean Modeling System (ROMS) is a free-surface, primitive equation model 106
based on the Boussinesq approximation and hydrostatic balance (Shchepetkin and McWilliams 107
6
2003, 2005). Here ROMS is configured in the Labrador Sea over the domain 35-65°W, 51-66°N 108
(Fig. 1a), with horizontal resolution of 7.5 km and 30 vertical layers, 8 of which are confined in 109
the upper 300 m. The nonlocal K-Profile Parameterization (KPP) scheme is used to parameterize 110
vertical mixing (Large et al. 1994). The integration period begins in January 1980 and ends in 111
December 2009. 112
The bathymetry is derived from ETOPO2 (Smith and Sandwell 1997). A modified Shapiro 113
smoother (Penven et al. 2008) is applied to the original bathymetric data to avoid pressure 114
gradient errors. The smoother is applied everywhere except for three small regions around Cape 115
Desolation. Retaining the original bathymetric details in these regions does not affect 116
significantly the bottom velocities. However, these details are critical for a correct representation 117
of the eddy-generation along the West Greenland coast, as shown by Bracco et al., (2008). A 118
detailed discussion of this problem is contained in Appendix A of Luo et al. (2011, LBD11 in the 119
following). 120
Boundaries are open to the east, south and north sides of the domain, where the velocity, 121
temperature and salinity fields are nudged to the Simple Ocean Data Assimilation (SODA) ocean 122
reanalysis version 2.1.6 (Carton and Giese 2008). A modified radiation boundary condition is 123
also applied following Marchesiello et al. (2001). SODA has been evaluated against other ocean 124
reanalysis products in LBYD12, and it presents the advantage of reproducing quite realistically 125
the lateral and vertical extent of the Irminger Current at the north-east corner of the domain. As 126
mentioned, the time period investigated in this work extends from 1980 to 2009. Since SODA 127
2.1.6 ends in 2008, boundary conditions of 2008 are repeated for the computation of 2009. 128
Surface fluxes are from NCEP/NCAR (1980-2009). Over the period November 1999 – June 129
2009 the surface wind data consist of monthly averages of the spatial blending of high-resolution 130
7
satellite data (Seawinds instrument on the QuikSCAT satellite - QSCAT) and the NCEP re-131
analysis (Milliff et al., 2004). To avoid long-term drift of SST linked to the errors in the 132
NCEP/NCAR heat fluxes (Josey 2001), QNCEP are corrected by the NOAA extended SST 133
(SSTNOAA) (Smith and Reynolds, 2004) on a monthly timescale and 2°×2° resolution, according 134
to QMod = QNCEP + dQMod/dSSTMod × (SSTMod - SSTNOAA), where the subscript Mod indicates 135
modeled quantities.1 Nudging to the monthly surface salinity climatology provided by the Word 136
Ocean Atlas 2009 (WOA09, Antonov et al., 2009) partially account for the seasonal cycle of the 137
fresh water fluxes associated to sea-ice melting and the Hudson River outflow. Those anomalies 138
dominate the seasonal variability of the surface salinity field along the western coast of the basin. 139
In this study, therefore, we do not account for the interannual variability of the surface fresh 140
water fluxes except for the eastern and southern model boundaries. 141
We consider two simulations forced by the same atmospheric forcings, consisting of 142
monthly varying NCEP/NCAR heat and momentum fluxes, but by different ocean boundary 143
conditions. In the first simulation (CLIMA), we nudge the open boundaries to monthly mean 144
climatological values obtained by averaging the SODA monthly output from 1980 to 2009 145
(where 2009 is identical to 2008). In the second simulation (VARY), we retain the interannual 146
variability in the boundary conditions without any further averaging. Initial conditions for both 147
integrations are derived from a spin-up run, forced by climatological monthly averaged 148
NCEP/NCAR atmospheric fluxes and SODA boundary conditions, that extend for 50 years after 149
a stationary state is reached. CLIMA and VARY are then initiated in 1976 and the first four 150
years of simulations are discarded. 151
1 Differences between QNCEP and QMod are very small and to all effects of analyses here negligible.
8
Additionally, we investigate if using high frequency surface heat fluxes (daily) and winds (6 152
hourly) to force ROMS influences its representation of convective events in the Labrador Sea 153
performing a third run limited to the period November 1999 – December 2004. The set-up is 154
identical to VARY but for the frequency of the atmospheric forcing products. No significant 155
changes were detected in the representation of LSW formation. The outcome of this comparison 156
is summarized in the Appendix. 157
3. Validation of model output and potential temperature variability 158
A detailed validation of the model representation of the surface circulation is provided in 159
LBD11. It is found that ROMS reproduces accurately the Labrador Sea circulation and its 160
variability compared to altimeter data, from seasonal to interannual scales, including the surface 161
EKE along the Greenland coast and in the convective area (Fig. 1a). Independently of the forcing 162
fields used, and although the horizontal resolution of the model is just below the Rossby radius 163
of deformation (~13 km) of the basin, the model captures the observed eddy variability. In 164
particular, most statistics related to the Irminger Ring population are in good agreement with 165
observations, with the exception of the lifespan of large eddies (six to seven months in CLIMA 166
and eight to eleven months in VARY, versus twelve to eighteen months in observations). 167
Consequently, fewer than observed Irminger Rings migrate as far south as 58o S in the model. 168
The interannual variability of potential temperature through the water column over the 169
period 1980-2009 is discussed in LBYD12. Modeled PT values are compared with hydrographic 170
surveys along the AR7W line (Lazier et al. 2002; Yashayaev 2007; Straneo 2006; van Aken et al. 171
2011) and with ARGO data from the summer of 2002 in the so-called Central Labrador Sea 172
(CLS), defined accordingly to Yashayaev and Loder (2009) as the region comprised by the 3250 173
9
m isobaths and within 150 km of the AR7W hydrographic line. The CLS is not the region of 174
greatest convective activity, but is close enough to it to be affected by the spreading of LSW 175
within weeks from the events, and is where we have the most uniform - in time and space - in-176
situ data coverage over the period considered. The model representation matches well the 177
observations from the base of the mixed layer to 2500 m; below such depth ROMS is 178
approximately half a degree too warm compared to in-situ data and less uniformly stratified than 179
observed. This discrepancy is likely due to poor representation of vertical mixing at depth and/or 180
to insufficient model vertical resolution; SODA boundary conditions may further contribute to 181
the model bias. 182
The PT in the first half of the record (1980-1994) is characterized by the alternation of 183
warmer and colder periods, while the second half (1995-2009) is dominated by a warming trend 184
beginning around 1995 in both CLIMA and VARY (Fig. 1b). The modeled interannual 185
variations are consistent with hydrographic observations (LBYD12); for example, the model 186
successfully captures the strong convection events in 1982-1984 and in the early 1990s’, the 187
reduction in the convective activity after 1995 and the partial recovery in 2008. Independent of 188
the integration considered, the intensity of convection is reduced significantly after 1995 in both 189
CLIMA and VARY. LBYD12 showed that the warming in CLS results from the combined 190
changes in local heat fluxes and warming of the Irminger Current, which has been reported 191
continuously since 1995 (Böning et al. 2006; Straneo 2006) and is realistically represented in 192
SODA. Irminger Current water is advected by Irminger eddies into the central portion of the 193
basin, where it facilitates restratification (Katsman et al. 2004; LBD11). 194
Moving to seasonal scales, LBYD12 compared the PT seasonal cycle in ROMS with the 195
observed one by ARGO floats from mid-2002 (see their Fig. 6) and found very good agreement 196
10
between the two. The ARGO period, however, is characterized by weak convective episodes in 197
all years except 2008, and it is not representative of the interannual variability observed during 198
the 30 years under investigation. Therefore, as further validation, in Fig. 2 we present a 199
comparison of the seasonal cycle of heat content and salinity in ROMS and P-ALACE float data 200
covering 1996-2000 in the top 200 m and in a deeper level, extending from 200 to 1300 m, 201
following the analysis in Straneo (2006) for the CLS. This period includes four strong convective 202
events. The general behavior is well represented both at the surface and at depth. The model 203
slightly underestimates the amplitude of the heat content cycle and overestimates the salinity one 204
in the surface layer. At depth the seasonal evolution of ROMS heat content is within the standard 205
deviation of the observations, while the modeled salinity does not agree with the P-ALACE data 206
and does not display evident seasonal variations. While a similar lack of seasonality was noticed 207
by Straneo (2006) using OWSB data over the period 1964-1974 independently of the inclusion 208
of Great Salinity Anomaly years (Dickson et al., 1988), we cannot exclude a model bias. Indeed 209
the representation of salinity at high latitudes poses a challenge to ocean modeling, as previously 210
noticed by Tréguier et al. (2005) analyzing the variability of the North Atlantic subpolar gyre in 211
four high resolution models. Nonetheless, we can conclude that ROMS provides a good 212
representation of seasonal changes in potential temperature also in periods of strong convection. 213
4. Seasonality of convective events in the Labrador Sea 214
4.1 Vertical velocity and convection 215
Ocean deep convection takes place within vertical plumes of O(1) km in radius and vertical 216
extent of about 2km (Marshall and Shott, 1999). Those plumes cause the water column to 217
homogenize within the so-called mixed patch. The model adopted in this investigation, and more 218
generally all ocean models used for climate studies, do not have sufficient horizontal resolution 219
11
to represent directly the convective process. The integral effect of the convective plumes and 220
their impact on vertical mixing, and temperature and salinity distributions, however, can be 221
parameterized (Send and Marshall 1995). In ROMS they are accounted for by the nonlocal KPP 222
scheme. KPP assumes that the vertical turbulent fluxes of momentum, heat and salinity (or any 223
other tracer) can be expressed as the sum of a down-gradient flux and a non-local contribution. 224
Convective events can then be characterized by any quantity directly affected by such fluxes, 225
from the absolute value of the vertical velocity field, to temperature and/or salinity profiles, or 226
mixed layer depth (MLD) and turbulent kinetic energy. Vertical velocities present the advantage 227
over the other tracers in defining convective episodes, and in particular in defining initiation and 228
duration of convection, by allowing a clear-cut threshold. This cannot be achieved as easily using 229
temperature or density profiles because following strong convective events the water mixed at 230
depth during the convective season remains in the central portion of the Labrador basin for long 231
enough to influence the PT or density averages for the following year (see for example Fig. 13 in 232
LBYD12). 233
Fig. 3 shows the mean seasonal cycle of the absolute value of the vertical velocity field, |w|, 234
averaged over depths comprised between 150 and 2000 m as modeled by ROMS during the 30 235
years considered. The top 150 m are excluded as strongly influenced by Ekman pumping and 236
surface momentum forcing (see e.g. Koszalka et al., 2009 for an analysis of relative contributions 237
to w in an idealized wind driven domain). High values of |w| are found at all times along the 238
boundary current system in correspondence of steep gradients of the continental slope, broadly 239
covering the areas comprised between the 1000 and 2000 m isolines. In those areas eddy 240
generation by baroclinic instability contributes to high level of surface EKE, and the eddies 241
extend from the ocean surface to the slope (e.g. LBD11). In the basin interior, on the other hand, 242
12
the largest vertical velocities are found only in late winter and early spring over a region 243
encompassing the deeper portion of the basin east of the western continental slope as observed, 244
among others, by Lavander et al. (2000), Pickart et al. (2002) and Frajka-Williams et al. (2014). 245
This area, indicated as CR for convective region in the following, is approximated by the black 246
box in Fig. 3 with coordinates (56.6 – 52o W, 57 - 59.5o N). We verified that in ROMS the same 247
area identifies also the maximum MLD (Fig. 3e and 3f). Any MLD definition requires the use of 248
an ad-hoc threshold in temperature or density. In the Labrador Sea the density differences 249
between the surface and the base of the mixed layer during convective events can be very small 250
(Fraika-Williams et al. 2014), smaller than 0.01 kg m−3, usually adopted in numerical studies 251
(Lazier et al. 2002). A robust observational constrain is not available, and the ‘correct’ threshold 252
is likely to be one that accounts for seasonal changes and interannual variability, particularly in 253
presence of trends. In this work the MLD is defined as the depth at which density differences 254
with the surface are equal to 0.008 kg m−3. As further test, we compared the modeled convective 255
region from 2003 to 2009 with the one identified by the MLD in the ARGO data. In both 256
datasets, convection is patchy, and approximately located over a smaller area than the CLS or the 257
CR regions, and close to the intersection of the two. Whenever the average over the period is 258
considered, the agreement between model and observations is very good (not shown). 259
The time series of |w| and MLD in CLIMA and VARY and their differences are presented 260
in Fig. 4. Compared to the few observational estimates the modeled MLD appears generally too 261
deep. This bias is directly linked to the deep layers in ROMS being less uniformly stratified than 262
observed. The correlation between |w| and MLD has coefficient R = 0.9. Increasing the density 263
threshold in the MLD criterion to the more common 0.01 kg/m-3 (e.g. Tréguier et al., 2005) does 264
not affect the correlation but further increase the maximum MLD in the last ten years of the 265
13
integration, when a warming trend is apparent. If a threshold of 0.02 kg/m-3 or higher is used 266
instead, the correlation decreases dramatically and the mixed layer reaches the ocean bottom 267
during most events. 268
From the |w| time series we separate years of strong and weak convection using a threshold 269
of 0.85x10-3 m/s. This allows us to group 15 years (1982, 1983, 1984, 1985, 1988, 1989, 1990, 270
1991, 1992, 1993, 1994, 1999, 2000, 2002, 2008) in the strong, and 15 years (1980, 1981, 1986, 271
1987, 1995, 1996, 1997, 1998, 2001, 2003, 2004, 2005, 2006, 2007, 2009) in the weak 272
convection categories, respectively. For six of the thirty years considered a change in the 273
threshold by ± 10% could modify their classification, and we choose to include three of those 274
normal events in each category. While the |w| criterion provides a differentiation similar to the 275
one of MLD exceeding (or not) 1000 m in observational data (e.g. Pickart et al. 2002), the 276
classification in VARY does not match perfectly the observations. For example the winter of 277
1999 is commonly categorized as having weak convection, while 1997 is seen as a strong. This is 278
captured in CLIMA but not in VARY, even if the two runs have similar Irminger Current inflow 279
over the 1995-1999 period, and is associated to the (modeled) internal ocean variability. In our 280
analysis, we retain the same groups in CLIMA to simplify the comparison and have an equal 281
number of events in each group. The vertical velocity time series in Fig. 4a and 4b are highly 282
correlated (correlation coefficient R = 0.95), but according to the chosen w threshold four events 283
classified as weak in VARY would follow under the strong category in CLIMA. We verified that 284
our conclusions do not depend on the details of the classification in CLIMA, or on the |w| 285
threshold in VARY. 286
The interannual variability of convective events that emerges from the vertical velocity field 287
is consistent with the one from PT or potential density over the same region (Fig. 5 and Fig. 6). 288
14
The classification of weak and strong years does not change with the variable used, but the 289
differences between VARY and CLIMA, indicative of the influence of the Irminger Current on 290
convective activity, appear amplified (Fig. 5c). For the first half of the record the potential 291
temperature averaged between 150 and 2000 m depth over the CLS domain is greater in CLIMA 292
by about 0.07 °C on average, and vertical velocities display large differences but of alternating 293
sign, so that their average is close to zero. After 1995, PT (density) in VARY is consistently 294
higher (lower) than in CLIMA, and the absolute vertical velocities are mostly higher in CLIMA. 295
The PT warming trend is consistent with the one found in the incoming Irminger Current at the 296
model boundary (see LBYD12, their Fig. 9). Overall, the boundary current contribution explains 297
about half of the warming recorded in VARY in the CR or CLS regions after 1995. 298
It is important to notice that the Irminger Current warming recorded after 1995 is not 299
unprecedented, at least according to SODA, but is part of the decadal variability of the system. 300
The period from 1958 to 1970 was indeed characterized by an Irminger Current as warm as 301
during the first decade of the XXI century. 302
4.2 Seasonality and strength of convection 303
The seasonal cycle of convective events over the Labrador Sea from 1980 to 2002 cannot be 304
easily quantified in the observations due to their limited time and/or space coverage except for 305
the P-ALACE period (1996-2000) (Straneo 2006; Lavender et al., 2000). Shipboard and 306
hydrographic surveys have sampled late spring and summer, moorings provides continuous time 307
series but at one location, and floats have been deployed only for limited time intervals. Since 308
mid-2002 ARGO data, complemented by the K1 mooring (deployed in 1995 near the former 309
15
BRAVO station at about 25 km distance, e.g. Avsic et al., 2006) and at times by gliders, allowed 310
to measure the seasonal cycle with greater confidence. 311
Here we compare the seasonality of weak and strong convective events using the modeled 312
|w|, and we find that they differ in the timing of their initiation, maxima and shutdown (Fig. 7a). 313
On average, weak events initiate two (CLIMA) to three (VARY) weeks after the strong ones, 314
they reach their peak of intensity two or three weeks later, and terminate approximately two 315
weeks earlier. This behavior is common to all weak events, independent of the decade considered, 316
and is further amplified if the six ‘normal’ years are excluded by the calculation. Consequently, 317
the convective activity in weak years is approximately one month shorter than in strong ones and 318
shifted further into spring. 319
The reduced intensity, shifted seasonality and shortened duration of convection in the CR 320
for weak events in both integrations are associated predominately to reduced heat fluxes and 321
wind intensity in winter (Figs. 7b and 7c). It is noticeable that atmospheric changes in the 322
reanalysis are found only between November and March, and both heat and momentum fluxes 323
do not show any significant difference through the remaining of the year. Also, the atmospheric 324
fluxes display a substantial change in intensity, but almost no shift of seasonality in their annual 325
cycle. In weak years, it takes at least two weeks more for the convection to start because of the 326
reduced heat flux and associated wind intensity. Fig. 7 reveals also that the convection period 327
defined by |w| does not coincide with the period during which the heat flux to the atmosphere 328
forces convective mixing. The offset at initiation is associated with the excess buoyancy 329
accumulated in the upper 150 – 200 m between May and August, when the surface heat fluxes 330
are, on average, positive. Such excess buoyancy needs to be removed (Bailer et al. 2005; 331
Gelderloos et al., 2012; Frajka-Williams et al. 2014) before convection can take place. Thereafter, 332
16
it is enough for the surface heat fluxes to remain negative, even if weakly, for the convection to 333
continue in the top 200 - 300 m of the water column, and for |w| to remain large at those depths. 334
The input of warm, salty Irminger water from the Western Greenland coast to the center of 335
the Labrador basin increases the stratification, and consequently more heat has to be released to 336
the atmosphere for the convective instability to begin, further delaying its initiation. The impact 337
of the boundary current interannual variability on convection is quantified as the difference 338
between VARY and CLIMA and it is more prominent in weak than in strong years, due to the 339
significant warming of Irminger Current experienced since 1995 (Stein 2005; Myers et al. 2007). 340
For instance, the maximum absolute value of the CR-averaged vertical velocities in VARY for 341
weak years is 0.5×10-3 m/s, to be compared 0.7×10-3 m/s in CLIMA, and it is achieved about a 342
week later (Fig. 7a). In contrast, mean |w| in VARY and CLIMA are almost identical in strong 343
convective years. 344
4.3 Strength of convection and heat fluxes 345
Figure 8 shows the relation between the strength of convection, measured by the absolute 346
value of w averaged over the period in which |w| > 0.2 ms-1, and the surface heat fluxes averaged 347
over the CR in different seasons for all, or weak and strong years separately. A high coefficient 348
(R) implies that the modeled convection intensity is regulated by the local surface heat fluxes. 349
The Labrador Sea convection occurs in late winter and spring, and the correlation between 350
vertical velocities in the CR and the atmospheric heat fluxes is the greatest when heat fluxes 351
from December to April are considered. Coefficients are generally higher in CLIMA than in 352
VARY (Figs. 8a and 8b), but are greater than 0.85 in both cases when all years are considered. 353
The wintertime heat flux, particularly in January and February, is the most important (Figs. 8c-354
17
8d) since the conditions for the initiation of the convective events are set. Moreover, the 355
accumulated heat flux in winter is larger than in spring (Fig. 7a). On the other hand, the delayed 356
onset in weak years requires considering early spring, March and April, together with (D)JF to 357
obtain a statistical significant relation for weak events in VARY, and greatly improve the 358
correlation in CLIMA.2 In contrast, the atmospheric forcing averaged over three, six and nine 359
months prior of the inception of convection does not influence the convective activity, in 360
agreement with the analysis by Straneo (2006) that showed that the positive heat fluxes into the 361
ocean in late spring and summer (May to August) are absorbed in the upper 200 m and are then 362
released in autumn and early winter (September to December). 363
The strength of convection depends also on the inflow of Irminger Current waters: the 364
maxima correlations achieved in VARY are all slightly smaller than in CLIMA, and the slopes 365
describing the linear relation between vertical velocities and heat fluxes change more 366
dramatically between strong and weak years. Notwithstanding the modulation at decadal scales 367
associated with changes of the Irminger Current characteristics, the seasonal variability of the 368
convective activity in the Labrador Sea appears controlled to a large extent by the local heat 369
fluxes immediately before and during the convective season. Heat fluxes over the CR result, in 370
part, from large scale anomalies and during the winter season are correlated with both the North 371
Atlantic Oscillation (NAO) (Dickson et al., 1996), and the Arctic Oscillation (AO) (Thompson 372
and Wallace, 1998) (see Table 1). 373
5. Discussion and conclusions 374
2 If only JFM are considered the correlation in VARY decreases to R = 0.8 when all years are considered and R = 0.4 for weak events.
18
In this paper, two regional simulations of the Labrador Sea circulation in the period 1980-375
2009 are used to investigate changes in the seasonal cycle of convective events. The numerical 376
integrations, CLIMA and VARY, differ in their boundary conditions and allow for isolating the 377
impact of atmospheric forcing, through heat and momentum fluxes, and oceanic forcing, through 378
changes in the properties of the incoming currents in the basin, on the interannual and seasonal 379
variability of convective events. Both CLIMA and VARY reproduce well the observed 380
circulation, the surface eddy kinetic energy and its interannual variability. In particular ROMS 381
simulates realistically the population of Irminger Rings, and the surface EKE agrees with 382
observational estimates from altimeter data along the west coast of Greenland, where the Rings 383
are formed, and in the central portion of the basin where convection occurs (Luo et al., 2011). 384
Additionally, the evolution of potential temperature throughout the water column is well 385
represented (Luo et al., 2012) up to 2500 m of depth. As expected, VARY reproduces satellite 386
and hydrographic observations more realistically than CLIMA by accounting for the changes in 387
the Irminger Current. In ROMS the region characterized by convective activity is found seaward 388
of the western continental slope in agreement with observational (Lavander et al., 2000; Pickart 389
et al., 2002) and numerical studies (Mizoguchi et al., 2003). A recent work by Zhu et al. (2014) 390
related the ability of a model to properly simulate the localization of deep convection in the 391
Labrador Sea to its representation of eddy-induced lateral fluxes, associated to the Irminger 392
Rings and to the smaller eddies formed by baroclinic instability along the boundary current 393
system. A number of previous studies have indeed shown limitations in this respect, likely due to 394
coarse resolution (Willebrand et al., 2001), and/or parameterization choices (Canuto et al., 2004). 395
The analysis presented indicate that the annual changes in the strength of convection in the 396
Labrador Sea are predominately determined by local atmospheric forcing, consistent with 397
19
previous studies (Delworth and Greatbatch 2000; Eden and Willebrand 2001; Bentsen et al. 398
2004). Heat fluxes over the CR from December to April correlate highly with the oceanic 399
vertical velocities averaged between 150 and 2000 m depth in both simulations, with the 400
wintertime heat flux having larger impacts in years of strong convection (Dickson et al. 1996), 401
and spring heat fluxes being as important as winter ones during weak convective episodes. In 402
contrast, the atmospheric forcing averaged over three, and six months prior of the inception of 403
convection does not influence the convective activity, in agreement with the analysis by Straneo 404
(2006). The state of the Irminger Current contributes to the modulation of the convective activity 405
through the Irminger Rings that carry warm and salty water to the convective area, increasing the 406
stratification through lateral mixing. Its variability, however, plays only a secondary role in the 407
thirty years considered, as manifested by the difference of vertical velocity between CLIMA and 408
VARY (Fig. 4c). More quantitatively, the comparison between the seasonal cycle of |w| in 409
CLIMA and VARY (Fig. 7a) indicates that the interannual modulation of the boundary current is 410
responsible for a further reduction of about 25% in the maximum of the mixing intensity. 411
The seasonal cycle of Labrador Sea convection has been observed since the deployment of 412
ARGO floats in the basin, but we do not have observations that are continuous in time and space 413
prior to the summer of 2002. Here we show that weak and strong convective events are 414
characterized not only by different intensity, but also by a different seasonal cycle. Both 415
initiation and peak of convection for weak events in VARY are delayed by about three weeks 416
compared to strong ones. Additionally, the duration of convective activity is approximately one 417
month shorter for weak episodes. Those characteristics are linked to the reduced atmospheric 418
cooling observed between December and April – but not in other months – in the climatology of 419
the heat fluxes during weak years. The seasonal cycle of the atmospheric heat and momentum 420
20
fluxes is almost unaltered, but variations in surface cooling rates are translated in changes in 421
intensity, seasonality and duration of ocean convection. Those changes are important for 422
understanding the variability in the oxygen drawdown and carbon sequestration in the basin. 423
Finally, local heat fluxes reflect atmospheric variability at broader scales, and in winter are 424
significantly correlated with NAO/AO. 425
In light of previous work, our analysis suggests that a realistic representation of the 426
interannual variability of LSW formation can be achieved also in a nonhydrostatic model, but at 427
least two ingredients are required. First, the ocean model resolution has to be below the Rossby 428
deformation radius to allow for the direct representation of lateral eddy fluxes. Canuto et al. 429
(2004) have shown that models where the eddies are parameterized display large biases 430
independently of the vertical mixing scheme adopted. The recent analysis of glider data by 431
Frajka-Williams et al. (2014) further pinpointed to the importance of horizontal variations in 432
density over short distances (tens of kilometers) in the convective area for restratifying the region. 433
Variations at those scales (1-2 grid points), and of about 1/3 of the observed value (~ 0.003 kg m-434
3), are present in the 3-day average density fields used in Fig. 6. Model resolution plays also an 435
important role in the representation of the boundary current system and its instabilities. Second, 436
the seasonal cycle and the interannual variability of the surface heat fluxes are both important. 437
The vast majority of coupled climate models, including the ones adopted in the Coupled Model 438
Intercomparison Project Phase 5, have a biased seasonal cycle, with unrealistically large seasonal 439
excursions, and spring (fall) surface atmospheric fields closely resembling the ones in winter 440
(summer). On the positive side, the representation of the time integrated (monthly) effect of 441
winter storms on the surface wind and heat fluxes is sufficient to achieve a realistic 442
21
representation of LSW formation (see Appendix), suggesting that a detailed representation of 443
extreme winter events is not required if their average impacts are accounted for. 444
Given the biases found in the representation of the Atlantic meridional overturning 445
circulation in ocean and coupled climate models, and of the divergent behavior of the latter in 446
future projections (e.g. MacMartin et al., 2013), the high correlation between the intensity of 447
convective activity in the Labrador Sea and both local and large scale atmospheric heat flux may 448
provide an important test for validating climate models, while pointing to a simple way to 449
parameterize and improve the representation of convection when coarser resolution is used. 450
Acknowledgements 451
The simulations used in this work were performed under NSF OCE-0751775. The analysis has 452
been partially supported through the NSF grant OCE-1357373. We thank three anonymous 453
reviewers whose thoughtful comments greatly improved this work. 454
Appendix 455
The results presented in this work are obtained using ROMS forced by monthly atmospheric 456
forcing fields interpolated to the model time step. In the Labrador Sea, convection is driven, 457
predominantly, by the integral of the surface heat flux. This is shown, for example, by 458
Yashayaev and Loder (2009) using NCEP heat fluxes and ARGO data, and has been confirmed 459
by our integrations (see Figure 10 in LBYD12). The prominent role of the heat fluxes is further 460
supported by the investigation of the recovery mechanisms of convection after the shut down due 461
to the Great Salinity Anomaly by Gelderloos et al. (2012), and by the analysis of the 2005 462
convective season using ARGO data and sea-gliders performed by Frajka-Williams et al. (2014). 463
This happens because over most of the ocean, including the convective region in the Labrador 464
22
Sea as defined in the manuscript, wind mixing is limited to the top 200 m of the water column 465
even during strong events. Consequently, the use of higher frequency winds (or/and higher 466
frequency heat fluxes) does not substantially change the representation of deep convection in the 467
center of the basin. The above discussion is supported by a 5-yr long simulation – from Jan 2000 468
to Dec 2004 – identical in set-up to VARY but forced by NCEP/NCAR daily heat fluxes and 469
QuikSCAT – NCEP blended 6-hour winds, VARY-HF in the following. Comparing the mean 470
climatology of potential temperature in VARY-HF and VARY over the common period, it is 471
found that the mean modeled temperatures associated with the high frequency atmospheric 472
fluxes are ~ 0.3 oC warmer immediately below the surface and within the upper 100 m, and 473
slightly cooler (~ 0.1 oC or less) in the deeper layers, in agreement with results by Ezer (2000) 474
over the North Atlantic, and by Cardona and Bracco (2012) in the South China Sea. The latter 475
has shown that the high frequency winds excite near inertial waves as ageostrophic expression of 476
the eddy field, whenever a vigorous one is present, as in the case of the Labrador Sea. Enhanced 477
mixing by near inertial waves causes, in turn, the subsurface warming and the deeper cooling 478
(the frequency spectra analysis reported in Cardona and Bracco has been repeated for this 479
domain confirming their conclusions once the different Coriolis frequency is accounted for). The 480
seasonal cycle of the SST anomalies (Fig. A2) and the interannual variability (Fig. A3) of 481
potential temperature in the CR, however, do not depend on the frequency of the atmospheric 482
forcing used, being the eddy field analogous in the two integrations everywhere but at the tip of 483
Greenland, where the EKE in VARY-HF is twice as higher as in VARY, and in better agreement 484
with observations (Pickart et al., 2003). In conclusion, we exclude a significant role of the 485
atmospheric forcing frequency in determining strength and timing of modeled convective events 486
in the Labrador Sea. 487
23
References 488
Antonov, J. I., D. Seidov, T. P. Boyer, R. A. Locarnini, A. V. Mishonov, H. E. Garcia, O. K. 489
Baranova, M. M. Zweng, and D. R. Johnson, 2010. World Ocean Atlas 2009, Volume 2: 490
Salinity. S. Levitus, Ed. NOAA Atlas NESDIS 69, U.S. Government Printing Office, 491
Washington, D.C., 184 pp. 492
Avsic, T., J. Karstensen, U. Send, and J. Fisher, 2006: Interannual variability of newly formed 493
Labrador Sea Water from 1994 to 2005. Geoph. Res. Lett., 33, L21S02, 494
doi:10.1029/2006GLo26913. 495
Bailer, D.A., P.R. Rhines and S. Häkkinen, 2005: Formation and pathways of North Atlantic 496
Deep Water in a coupled ice-ocean model of the Arctic-North Atlantic Oceans. Clim. Dyn., 497
25, 497-516. 498
Bentsen, M., H. Drange, T. Furevik, and T. Zhou, 2004: Simulated variability of the Atlantic 499
meridional overturning circulation. Clim. Dyn., 22, 701-720. 500
Böning, C. W., M. Scheinert, J. Dengg, A. Biastoch, and A. Funk, 2006: Decadal variability of 501
subpolar gyre transport and its reverberation in the North Atlantic overturning. Geophys. Res. 502
Lett., 33, L21S01. 503
Bracco, A., J. Pedlosky, and R. S. Pickart, 2008: Eddy formation near the west coast of 504
Greenland. J. Phys. Oceanogr., 38, 1992-2002. 505
Bryden, H. L., H. R. Longworth, and S. A. Cunningham, 2005: Slowing of the Atlantic 506
meridional overturning circulation at 25 N. Nature, 438, 655-657. 507
Canuto V.M., A. Howard, P. Hogan, Y. Cheng, M.S. Dubovikov, and L.M. Montenegro, 2004: 508
Modeling ocean deep convection. Ocean Model., 7, 75-95, doi:10.1016/S1463-509
5003(03)00038-6. 510
Cardona, Y., and A. Bracco, 2012: Enhanced vertical mixing within mesoscale eddies due to 511
high frequency winds in the South China Sea. Ocean Modeling, 42, 1-15. 512
Carton, J. A., and B. S. Giese, 2008: A reanalysis of ocean climate using Simple Ocean Data 513
Assimilation (SODA). Mon. Weather Rev., 136, 2999-3017. 514
Chanut, J., B. Barnier, W. Large, L. Debreu, T. Penduff, J. M. Molines, and P. Mathiot, 2008: 515
Mesoscale eddies in the Labrador Sea and their contribution to convection and restratification. 516
J. Phys. Oceanogr., 38, 1617-1643. 517
24
Delworth, T. L., and R. J. Greatbatch, 2000: Multidecadal thermohaline circulation variability 518
driven by atmospheric surface flux forcing. J. Clim., 13, 1481-1495. 519
Dickson, R. R., J. Meincke, S. A. Malmberg, and A. J. Lee, 1988: The “Great Salinity Anomaly” 520
in the northern North Atlantic 1968–1982. Progr. Ocean., 20, 103–151. 521
Dickson, R., J. Lazier, J. Meincke, P. B. Rhines, and J. Swift, 1996: Long-term coordinated 522
changes in the convective activity of the North Atlantic. Prog. Oceanogr., 38, 241-295. 523
Eden, C., and J. Willebrand, 2001: Mechanism of interannual to decadal variability of the North 524
Atlantic circulation. J. Clim., 14, 2266-2280. 525
Ezer, T., 2000: On the seasonal mixed layer simulated by a basin-scale ocean model and the 526
Mellor-Yamada turbulence scheme. J. Geophys. Res., 105, 16843–16855. 527
Frajka-Williams, E., P.B. Rhines, and C.C. Eriksen, 2014: Horizontal Stratification during Deep 528
Convection in the Labrador Sea. J. Phys. Oceanogr., 44, 220-228. 529
Gelderloos, R., C. A. Katsman, and S. S. Drijfhout, 2011: Assessing the roles of three eddy types 530
in restratifying the Labrador Sea after deep convection. J. Phys. Oceanogr., 41, 2102-2119. 531
Gelderloos, R., F. Straneo and C.A. Katsman, 2012: Mechanisms behind the Temporary 532
Shutdown of Deep Convection in the Labrador Sea: Lessons from the Great Salinity 533
Anomaly Years 1968-71. J. Climate, 25, 6743-6755. 534
Häkkinen, S., and P. B. Rhines, 2004: Decline of subpolar North Atlantic circulation during the 535
1990s. Science, 304, 555-559. 536
Holland, D. M., R. H. Thomas, B. De Young, M. H. Ribergaard, and B. Lyberth, 2008: 537
Acceleration of Jakobshavn Isbrae triggered by warm subsurface ocean waters. Nat. Geosci., 538
1, 659-664. 539
Josey, S., 2001: A comparison of ECMWF, NCEP-NCAR, and SOC surface heat fluxes with 540
moored buoy measurements in the subduction region of the northeast Atlantic. J. Clim., 14, 541
1780-1789. 542
Katsman, C. A., M. A. Spall, and R. S. Pickart, 2004: Boundary current eddies and their role in 543
the restratification of the Labrador Sea*. J. Phys. Oceanogr., 34, 1967-1983. 544
Koszalka, I., A. Bracco, J.C. McWilliams, A. Provenzale, 2009: Dynamics of wind-forced 545
coherent anticyclones in the open ocean, J. Geoph. Res., 114, C08011, 546
doi:10.1029/2009JC005388. 547
25
Kuhlbrodt, T., A. Griesel, M. Montoya, A. Levermann, M. Hofmann, and S. Rahmstorf, 2007: 548
On the driving processes of the Atlantic meridional overturning circulation. Rev. Geophys., 549
45, RG2001. 550
Large, W. G., J. C. McWilliams, and S. C. Doney, 1994: Oceanic vertical mixing: A review and 551
a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32, 363-403. 552
Lavender, K. L., R. E. Davis, and W. B. Owens, 2000: Mid-depth recirculation observed in the 553
interior Labrador and Irminger seas by direct velocity measurements. Nature, 407, 66-69. 554
Lazier, J. N. R., 1980: Oceanographic conditions at Ocean Weather Ship Bravo, 1964–1974. 555
Atmos.–Ocean, 18, 227–238. 556
Lazier, J. R. N., R. Hendry, A. Clarke, I. Yashayaev, and P. Rhines, 2002: Convection and 557
restratification in the Labrador Sea, 1990–2000. Deep Sea Res. Part I, 49, 1819-1835. 558
Lilly, J. M., P. B. Rhines, F. Schott, K. Lavender, J. Lazier, U. Send, and E. D’Asaro, 2003: 559
Observations of the Labrador Sea eddy field. Prog. Oceanogr., 59, 75-176. 560
Lu, Y., D. G. Wright, and I. Yashayaev, 2007: Modelling hydrographic changes in the Labrador 561
sea over the past five decades, Prog. Oceanogr., 73, 406-426. 562
Luo, H., A. Bracco, and E. Di Lorenzo, 2011: The interannual variability of the surface eddy 563
kinetic energy in the Labrador Sea. Prog. Oceanogr., 91, 295-311. 564
Luo, H., A. Bracco, I. Yashayaev, and E. Di Lorenzo, 2012: The interannual variability of 565
potential temperature in the central Labrador Sea. J. Geophys. Res., 117, C10016, 566
doi:10.1029/2012JC007988. 567
MacMartin, D.G., E. Tzipperman, and L. Zanna, 2013: Frequency domain multimodel analysis 568
of the response of Atlantic meridional overtuning circulation to surface forcing. J. Climate, 569
26, 8323-8340. 570
Marchesiello, P., J. C. McWilliams, and A. Shchepetkin, 2001: Open boundary conditions for 571
long-term integration of regional oceanic models. Ocean Model., 3, 1-20. 572
Marshall, J., and Coauthors, 1998: The Labrador Sea deep convection experiment. Bull. Am. 573
Meteorol. Soc., 79, 2033-2058. 574
Marshall, J., and F. Schott, 1999: Open-ocean convection: Observations, theory, and models. Rev. 575
Geophys, 37, 1-64. 576
Marshall, J., and Coauthors, 2001: North Atlantic climate variability: phenomena, impacts and 577
mechanisms. Int. J. Climatol., 21, 1863-1898. 578
26
Milliff, R. F., J. Morzel, D. B. Chelton, and M. H. Freilich, 2004: Wind Stress Curl and Wind 579
Stress Divergence Biases from Rain Effects on QSCAT Surface Wind Retrievals. J. Atmos. 580
Ocean Tech., 21, 1216-1231. 581
Mizoguchi, K.-I., S. L. Morey, J. Zavala-Hidalgo, N. Suginohara, S. Häkkinen, and J. J. O’Brien, 582
2003: Convective activity in the Labrador Sea: Preconditioning associated with decadal 583
variability in subsurface ocean stratification, J. Geoph. Res., 108(C10), 3330, 1-28, 584
doi:10.1029/2002JC001735 585
Myers, P. G., N. Kulan, and M. H. Ribergaard, 2007: Irminger water variability in the West 586
Greenland Current. Geophys. Res. Lett., 34, L17601. 587
Penven, P., P. Marchesiello, L. Debreu, and J. Lefèvre, 2008: Software tools for pre-and post-588
processing of oceanic regional simulations. Environ. Model. Software, 23, 660-662. 589
Pickart, R. S., D. J. Torres, and R. A. Clarke, 2002: Hydrography of the Labrador Sea during 590
Active Convection. J. Phys. Ocean., 32, 428-457. 591
Pickart, R. S., M. A. Spall, M. H. Ribergaard, G. K. Moore, and R. F. Milliff, 2003: Deep 592
convection in the Irminger Sea forced by the Greenland tip jet. Nature, 424, 152-156. 593
Rattan, S., P. G. Myers, A.-M. Tréguier, S. Theetten, A. Biastoch, and C. Böning, 2010: Towards 594
an understanding of Labrador Sea salinity drift in eddy-permitting simulations, Ocean Model., 595
35, 77-88, doi:10.1016/j.ocemod.2010.06.007. 596
Rhein, M., and Coauthors, 2002: Labrador Sea Water: Pathways, CFC inventory, and formation 597
rates. J. Phys. Oceanogr., 32, 648-665. 598
Rykova, T., F. Straneo, J. M. Lilly, and I. Yashayaev, 2009: Irminger current anticyclones in the 599
Labrador Sea observed in the hydrographic record, 1990-2004, J. Mar. Res., 67, 361-384. 600
Sathiyamoorthy, S., and G. W. K. Moore, 2002: Buoyancy flux at Ocean Weather Station Bravo. 601
J. Phys. Oceanogr., 32, 458–474 602
Shchepetkin, A. F., and J. C. McWilliams, 2003: A method for computing horizontal pressure-603
gradient force in an oceanic model with a nonaligned vertical coordinate. J. Geophys. Res., 604
108, 3090. 605
——, 2005: The regional oceanic modeling system (ROMS): a split-explicit, free-surface, 606
topography-following-coordinate oceanic model. Ocean Model., 9, 347-404. 607
Smith, T. M., and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854-1997). 608
J. Clim., 17, 2466-2477. 609
27
Smith, W. H., and D. Sandwell, 1997: Marine gravity anomaly from Geosat and ERS 1 satellite 610
altimetry. J. Geophys. Res., 102, 10039-10054. 611
Stein, M., 2005: North Atlantic subpolar gyre warming–impacts on Greenland offshore waters. J. 612
Northw. Atl. Fish. Sci., 36, 43-54. 613
Straneo, F., 2006: Heat and freshwater transport through the central Labrador Sea, J. Phys. 614
Oceanogr., 36, 606-628. 615
Straneo, F., and Coauthors, 2010: Rapid circulation of warm subtropical waters in a major glacial 616
fjord in East Greenland. Nat. Geosci., 3, 182-186. 617
Talley, L. D., and M. S. McCartney, 1982: Distribution and circulation of Labrador Sea water. J. 618
Phys. Oceanogr., 12, 1189-1205. 619
Thompson, D. W., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime 620
geopotential height and temperature fields, Geophys. Res. Lett., 25, 1297-1300. 621
Tréguier, A.-M., S. Theetten, E. Chassignet, T. Penduff, R. Smith, L. Talley, J. O. Beismann, and 622
C. Böning, 2005: The North Atlantic subpolar gyre in four high-resolution models. J. Phys. 623
Oceanogr., 35, 757-774. 624
Våge, K., and Coauthors, 2008: Surprising return of deep convection to the subpolar North 625
Atlantic Ocean in winter 2007–2008. Nat. Geosci., 2, 67-72. 626
van Aken, H. M., M. Femke de Jong, and I. Yashayaev, 2011: Decadal and multi-decadal 627
variability of Labrador Sea Water in the north-western North Atlantic Ocean derived from 628
tracer distributions: Heat budget, ventilation, and advection. Deep Sea Res. Part I, 58, 505-629
523. 630
Yashayaev, I., J. R. N. Lazier, and R. A. Clarke, 2003: Temperature and salinity in the central 631
Labrador Sea, ICES Marine Symposia Series, 219, 32-39. 632
Yashayaev, I., 2007: Hydrographic changes in the Labrador Sea, 1960–2005. Prog. Oceanogr., 633
73, 242-276. 634
Yashayaev, I., H. M. van Aken, N. P. Holliday, and M. Bersch, 2007: Transformation of the 635
Labrador Sea Water in the subpolar North Atlantic, Geophys. Res. Lett., 34, L22605, 636
doi:10.1029/2007GL031812. 637
Yashayaev, I., and J. W. Loder, 2009: Enhanced production of Labrador Sea Water in 2008, 638
Geophys. Res. Lett., 36, L01606, doi: 10.1029/2008GL036162. 639
28
Willebrand, J., and Coauthors, 2001: Circulation characteristics in three eddy-permitting models 640
of the North Atlantic. Progr. Oceanography, 48, 123-161 641
Zhu, J. S., E. Demirov, Y. Zhang, and A. Polomska-Harlick, 2014: Model simulations of 642
mesoscale eddies and deep convection in the Labrador Sea. Adv. Atmos. Sci., doi: 643
10.1007/s00376-013-3107-y,in press. 644
645
646
29
Table 1: Correlation coefficients between seasonal averages (DJF and DJFMA) of the heat 647
fluxes into the ocean over the CR region and the AO/NAO indices for 15 strong convective 648
years, 15 weak episodes, and all 30 years. Correlations using monthly values over the same 649
seasons are statistically indistinguishable. Correlations significant at the 95% confidence 650
level according to a t-test are in bold. 651
652
AO strong
AO weak
AO all
NAO strong
NAO weak
NAO all
Winter (DJF) -0.49 -0.27 -0.47 -0.53 -0.53 -0.55 WinterSpring (DJFMA) -0.40 -0.18 -0.37 -0.48 -0.39 -0.46
653
654
655
656
30
Figure captions 657
Fig. 1 (a) Annual mean distribution of surface eddy kinetic energy cast as a speed !!"! =658
2!"! (in color) and mean surface velocity (arrows). The locations of the AR7W WOCE 659
hydrographic line (black line) and of the Ocean Weather Station BRAVO (red star) are marked. 660
(b) Annual mean of potential temperature in the region comprised by the 3250 m isobaths 661
following Yashayaev (2007). Potential temperature values are averaged over depths comprised 662
between 150 m and 2000 m. 663
Fig. 2 Seasonal variation in heat content (top) and salinity (bottom) in the central Labrador Sea 664
for the top 200 m of the water column (left), and for a lower layer comprised between 200 and 665
1300 m (right) in the model (blue and red lines) and in the observations presented in Straneo 666
(2006) (black and gray lines). The model and the P-ALACE float data cover the period 1996-667
2000. The gray lines represent the standard deviation (std) around the float data. The dashed 668
lines indicate the std of the Ocean Weather Station Bravo data from 1964 to 1974 (also from 669
Straneo, 2006). 670
Fig. 3 Seasonal mean distribution of vertical velocity |w| averaged between 150 and 2000 m 671
depth (a) at the peak of convection (February to April), (b) in late spring and early summer 672
(May to July), (c) in late summer and early fall (August to October), and (d) in fall to early 673
winter (November to January). The seasonal mean distribution of mixed-layer depth in (e) 674
February to April, and (f) May to July, is added for comparison. The black box indicates the 675
convective region (CR) as defined in the paper. 676
Fig. 4 Time series of absolute vertical velocity (in m/s) averaged between 150 and 2000 m depth 677
and of mixed-layer depth (in m) defined using a density criterion of 0.008 kg m-3 over the CR 678
31
from 1980 to 2009. (a) CLIMA, (b) VARY and (c) the difference (CLIMA-VARY). The dashed 679
line in (b) indicates the chosen threshold separating strong and weak convections years. The 680
time-series are constructed using 3-day averages. 681
Fig. 5 Evolution of potential temperature (PT, in oC) in the convective region (CR) from 1980 to 682
2009. (a) CLIMA, (b) VARY and (c) difference between the two integrations (CLIMA-VARY) 683
for time series obtained averaging from 150 to 2000 m depth. The plot is obtained using 3-day 684
averages. 685
Fig. 6 Evolution of potential density (σθ in kg m-3 ) in the convective region (CR) from 1980 to 686
2009. (a) CLIMA, (b) VARY. The plot is obtained using 3-day averages. 687
Fig. 7 (a) Annual cycle of |w| averaged between 150 m and 2000 m depth over the convective 688
region CR. Dashed lines for CLIMA, and solid lines for VARY. The shading indicates the 689
standard deviation around VARY averages, and the vertical lines the time when |w| reaches its 690
maximum in VARY. (b) Annual cycle of downward surface heat flux QMod and (c) surface wind 691
intensity averaged over CR. 692
Fig. 8 Relation between downward surface heat flux as seen by the model (QMod) and absolute 693
value of vertical velocities averaged over the CR region between 150 and 2000 m during 694
convective events. (a) CLIMA, December to April, DJFMA; (b) VARY, DJFMA; (c) CLIMA 695
winter only, DJF; (d) VARY, DJF. Blue, red and black lines show fits using strong, weak, and all 696
years, respectively. The two-digit numbers indicate the year. 697
698
32
Fig. A1 Potential temperature difference between VARY-HF and VARY averaged over the 699
period January 2000 – December 2004 along the AR7W line. 700
Fig. A2 Seasonal cycle of potential temperature anomalies (in oC) in the convective region (CR) 701
in the top 500 m over the period January 2000 – December 2004. (a) VARY, (b) VARY-HF. 702
Fig. A3 Evolution of potential temperature (PT, in oC) in the convective region (CR) from 2000 703
to 2004. (a) VARY, (b) VARY-HF. The plot is obtained using 3-day averages of PT. 704
705
33
706
(a) (b) 707
Fig. 1 (a) Annual mean distribution of surface eddy kinetic energy cast as a speed !!"! =708
2!"! (in color) and mean surface velocity (arrows). The locations of the AR7W WOCE 709
hydrographic line (black line) and of the Ocean Weather Station BRAVO (red star) are marked. 710
(b) Annual mean of potential temperature in the region comprised by the 3250 m isobaths 711
following Yashayaev (2007). Potential temperature values are averaged over depths comprised 712
between 150 m and 2000 m. 713
!"!#
!"!#
!"!#
!"!#
!"!# !"!# !"!#
!"#$%!"#" ! !"#"
!"!!
!"!#
!"$!
!"$#
!"%!
!"%#
!"&!
!"&#
'())*+,*-
.,/!"#$ !"#$ !""# !""# !""" !""#
!"#
!"#
!"#
!"#!!
.0/
!"#$%#&'()*$+,$-'#.-$
!"#$%&'()(*+
!"!#
!"!!
!"!#
34
714
Fig. 2 Seasonal variation in heat content (top) and salinity (bottom) in the central Labrador Sea 715
for the top 200 m of the water column (left), and for a lower layer comprised between 200 and 716
1300 m (right) in the model (blue and red lines) and in the observations presented in Straneo 717
(2006) (black and gray lines). The model and the P-ALACE float data cover the period 1996-718
2000. The gray lines represent the standard deviation (std) around the float data. The dashed 719
lines indicate the std of the Ocean Weather Station Bravo data from 1964 to 1974 (also from 720
Straneo, 2006). 721
722
!"#$!"#$%&$%! !"#$%
&'
%&'(#)" *+,"'%#-$!"#$%
!"# !$#
!%# !&#
35
723
Fig. 3 Seasonal mean distribution of vertical velocity |w| averaged between 150 and 2000 m 724
depth (a) at the peak of convection (February to April), (b) in late spring and early summer 725
(May to July), (c) in late summer and early fall (August to October), and (d) in fall to early 726
winter (November to January). The seasonal mean distribution of mixed-layer depth in (e) 727
February to April, and (f) May to July, is added for comparison. The black box indicates the 728
convective region (CR) as defined in the paper. 729
730
36
731
Fig. 4 Time series of absolute vertical velocity (in m/s) averaged between 150 and 2000 m depth 732
and of mixed-layer depth (in m) defined using a density criterion of 0.008 kg m-3 over the CR 733
from 1980 to 2009. (a) CLIMA, (b) VARY and (c) the difference (CLIMA-VARY). The dashed 734
line in (b) indicates the chosen threshold separating strong and weak convections years. The 735
time-series are constructed using 3-day averages. 736
737
!" !# !$ !% !& !' !( !) !! !* *" *# *$ *% *& *' *( *) *! ** "" "# "$ "% "& "' "( ") "! "*"
"+""#
"+""$
"+""%
"+""&
!" !# !$ !% !& !' !( !) !! !* *" *# *$ *% *& *' *( *) *! ** "" "# "$ "% "& "' "( ") "! "* "
#"""
$"""
%"""
&"""
!" !# !$ !% !& !' !( !) !! !* *" *# *$ *% *& *' *( *) *! ** "" "# "$ "% "& "' "( ") "! "*"
"+""#
"+""$
"+""%
"+""&
!" !# !$ !% !& !' !( !) !! !* *" *# *$ *% *& *' *( *) *! ** "" "# "$ "% "& "' "( ") "! "* "
#"""
$"""
%"""
&"""!"#$%
&'(
&)(
&*(
+%,-
!"#$%./.+%,-
!"#$%&'()$*&+$,-.
!
!"#$%&'()$*&+$,-.
!
!"#$%&'()$*&+$,-.
!
!!'+
37
738
Fig. 5 Evolution of potential temperature (PT, in oC) in the convective region (CR) from 1980 to 739
2009. (a) CLIMA, (b) VARY and (c) difference between the two integrations (CLIMA-VARY) 740
for time series obtained averaging from 150 to 2000 m depth. The plot is obtained using 3-day 741
averages. 742
743
!"#
!$#
!%#
&
'&&
(&&&
('&&
)&&&
)'&&
&
'&&
(&&&
('&&
)&&&
)'&&
*+&
*+,
*+-
*+)
)+.
*+&
*+,
*+-
*+)
)+.
/0123
4356
/01237874356
38
744
Fig. 6 Evolution of potential density (σθ in kg m-3 ) in the convective region (CR) from 1980 to 745
2009. (a) CLIMA, (b) VARY. The plot is obtained using 3-day averages. 746
747
39
748
Fig. 7 (a) Annual cycle of |w| averaged between 150 m and 2000 m depth over the convective 749
region CR. Dashed lines for CLIMA, and solid lines for VARY. The shading indicates the 750
standard deviation around VARY averages, and the vertical lines the time when |w| reaches its 751
maximum in VARY. (b) Annual cycle of downward surface heat flux QMod and (c) surface wind 752
intensity averaged over CR. 753
754
755
Oct NovDec Jan Feb Mar Apr May Jun Jul AugSep0
0.05
0.1
0.15
0.2
0.25
0.3
strongweak
Month
Hea
t flu
xes
Win
d in
tens
ity
(W/m
2)
(N/m
2)
|w| (103 m/s)
0.0
0.5
1.0
1.5
2.0(a)
(b)
(c)
40
756
Fig. 8 Relation between downward surface heat flux as seen by the model (QMod) and absolute 757
value of vertical velocities averaged over the CR region between 150 and 2000 m during 758
convective events. (a) CLIMA, December to April, DJFMA; (b) VARY, DJFMA; (c) CLIMA 759
winter only, DJF; (d) VARY, DJF. Blue, red and black lines show fits using strong, weak, and all 760
years, respectively. The two-digit numbers indicate the year. 761
762
(c)
ww
41
763
Fig. A1 Potential temperature difference between VARY-HF and VARY averaged over the 764
period January 2000 – December 2004 along the AR7W line. 765
766
767
Fig. A2 Seasonal cycle of potential temperature anomalies (in oC) in the convective region (CR) 768
in the top 500 m over the period January 2000 – December 2004. (a) VARY, (b) VARY-HF. 769
770
Top Related