3.1 Quantification of deep convection

We characterized the convection in the SPNA in the winters of 2015–2018 by estimating the mixed layer depths (MLDs) for all Argo profiles collected in the SPNA north of 55^∘ N from 1 January to 30 April of each year (Fig. 1). The MLD was estimated as the shallowest of the three MLD estimates obtained by applying the threshold method of de Boyer Montégut et al. (2004) to θ, S and ρ profiles separately. The threshold method computes the MLD as the depth at which the difference between the surface (30 m) and deeper levels in a given property is equal to a given threshold. In the case that visual inspection of the winter profiles showed a thin stratified layer at the surface, a slightly deeper level (< 150 m) was considered the surface reference level. Following Piron et al. (2017), this threshold was taken as equal to 0.01 kg m⁻³ for ρ. For θ and S, we selected thresholds of 0.1^∘ C and 0.012, respectively, because they correspond to the threshold of 0.01 kg m⁻³ in ρ. The latter was previously shown to perform well in the subpolar gyre on density profiles (Piron et al., 2016). The criteria for temperature and salinity were chosen to perform well when temperature and salinity anomalies within the density-defined mixed layer are density-compensated. Our MLD estimates are comparable to those obtained using MLD determination based on the Pickart et al. (2002) methods (see Sect. S1 and Figs. S1 and S2 in the Supplement).

In this paper, deep convection is characterized by profiles with an MLD deeper than 700 m (colored points in Fig. 1) because it is the minimum depth that should be reached for renewing Labrador Sea Water (LSW) (Yashayaev et al., 2007; Piron et al., 2016). The winter MLD and the associated θ, S and ρ properties were examined for the Labrador Sea and the SECF region by considering the profiles inside the cyan and pink boxes in Fig. 1, respectively. Those two boxes were defined to include all Argo profiles with an MLD deeper than 700 m during 2016–2018 and the minimum of the monthly ADT for either the SECF region or the Labrador Sea. No deep MLD was recorded in the northernmost part of the Irminger Sea during this period. We computed the maximum MLD and the MLD third quartile (Q₃) from profiles with an MLD greater than 700 m in each of the two boxes separately. Q₃ is the MLD value that is exceeded by 25 % of the profiles and is equivalent to the aggregate maximum depth of convection defined by Yashayaev and Loder (2016). Hereafter, we refer to Q₃ as the aggregate maximum depth of convection. The properties (ρ, θ and S) of the mixed layers were defined for each winter as the vertical mean from 200 m to the MLD of all profiles with an MLD deeper than 700 m. For further use, we define the deep convection period as follows. For a given winter, the deep convection period begins the day when the first profile with a deep (> 700 m) mixed layer is detected and ends the day of the last detection of a deep mixed layer.

3.2 Time series of atmospheric forcing

The air–sea buoyancy flux (B_surf) was calculated as the sum of the contributions of Q and FWF (Gill, 1982; Billheimer and Talley, 2013):

where α and β are the coefficients of thermal and saline expansions, respectively, estimated from surface T and S. The gravitational acceleration g is equal to 9.8 m s⁻², the reference density of sea water ρ₀ is equal to 1026 kg m⁻³ and the heat capacity of sea water C_p is equal to 3990 J kg ^−1 ∘C⁻¹. SSS is the sea surface salinity (Q: W m⁻², FWF: m s⁻¹).

For easy comparison with previous results, which only considered the heat component of the buoyancy air–sea flux (e.g., Yashayaev and Loder, 2017; Piron et al., 2017; Rhein et al., 2017), B_surf (m²  s⁻³) was converted (W m⁻²) following Eq. (2) and noted $B_{\mathrm{surf}}^{\ast }$:

The FWF was also converted (W m⁻²) using

We also computed the horizontal Ekman buoyancy flux (BF_ek), which can be decomposed into the horizontal Ekman heat flux (HF_ek) and salt flux (SF_ek).

BF_ek=SF_ek – HF_ek. U_e and V_e are the eastward and northward components of the Ekman horizontal transport estimated from the wind stress meridional and zonal components. SSD, SST and SSS are ρ, T and S at the surface of the ocean (BF_ek, HF_ek and SF_e: J s⁻¹ m⁻²). Because ERA-Interim does not supply SSD or SSS, they were estimated from EN4 as follows. The monthly T and S data at 5 m of depth from EN4 were interpolated on the same time and space grid as the air–sea fluxes from ERA-Interim (12 h and 0.75^∘, respectively). SSD was estimated from those interpolated EN4 data (SST and SSS). Properties at 5 m of depth were considered to be representative of the Ekman layer. Data at locations where the ocean bottom was shallower than 1000 m were excluded from the analysis to avoid regions covered by sea ice.

Following Piron et al. (2016), the time series of atmospheric forcing were estimated for the SECF region and the Labrador Sea as follows. First, the gridded air–sea flux data and the horizontal Ekman fluxes were averaged over the pink (SECF region) and cyan (Labrador Sea) boxes (Fig. 1). Second, we estimated the accumulated fluxes from 1 September to 31 August the year after. Finally, we computed the time series of the anomalies of the accumulated fluxes from 1 September to 31 August with respect to the 1993–2016 mean.

Finally, in order to quantify the net intensity of the atmospheric forcing over the winter, we computed estimates of $B_{\mathrm{surf}}^{*}+{\mathrm{BF}}_{\mathrm{ek}}$ fluxes accumulated from 1 September to 31 March the year after. Following Piron et al. (2017), the associated errors were calculated by a Monte Carlo simulation using 50 random perturbations of Q, FWF and B_surf. The error amounted to 0.05, 0.04 and 0.03 J m⁻² for $B_{\mathrm{surf}}^{*}$, Q and FWF^*, respectively. The error of the horizontal Ekman buoyancy transport was also estimated by a Monte Carlo simulation and amounted to 0.04 J m⁻².

3.3 Preconditioning of the water column

The preconditioning of the water column was evaluated as the buoyancy that has to be removed (B(zi)) from the late summer density profile to homogenize it down to a depth zi:

where σ₀(z) is the vertical profile of potential density anomaly estimated from the profiles of T and S measured by Argo floats in September in the given region (pink or cyan box in Fig. 1).

Following Schmidt and Send (2007), we split B into a temperature (B_θ) and salinity (B_S) term:

In order to compare the preconditioning with the heat to be removed and/or air–sea heat fluxes, B, B_θ and B_S are reported in Joules per square meter. B, B_θ and B_S were estimated for a given year from the mean of all September profiles of B, B_θ and B_S. The associated errors were estimated as $\mathrm{std}\left(B\right)/\surd n$, where n is the number of profiles used to compute the September mean values.

Table 1Properties of the deep convection SECF and in the Labrador Sea in winters 2015–2018. We show: the maximal MLD observed, the aggregate maximum depth of convection, the σ₀, S and θ of the winter mixed layer formed during the convection event and n, which is the number of Argo profiles indicating deep convection. The uncertainties given with σ₀, S and θ are the standard deviation of the n values considered to estimate the mean values.

Download Print Version | Download XLSX

4.1 Intensity of deep convection and properties of newly formed LSW

We examine the time evolution of the winter mixed layer SECF since the exceptional convection event of winter 2015 (W2015 hereinafter) (Table 1 and Figs. 1–3). In W2015, we recorded a maximum MLD of 1710 m south of Cape Farewell (Fig. 1a), in line with Piron et al. (2017). The maximum MLD of 1575 m observed for W2016 (Fig. 1b) is compatible with the active mixed layer > 1500 m observed in a mooring array in the central Irminger Sea by de Jong et al. (2018). For W2015 and W2016, the aggregate maximum depth of convection was 1205 and 1471 m, respectively (Table 1). In W2017, deep convection was observed from three Argo profiles (Figs. 1c and 2a–c). The maximum MLD of 1400 m was observed on 16 March 2017 at 56.65^∘ N–42.30^∘ W. In W2018, the maximum MLD of 1300 m was observed on 24 February at 58.12^∘ N, 41.84^∘ W (Figs. 1d, 2d–f). Float 5903102 measured an MLD of 1100 m south of Cape Farewell (Fig. 1d), but the estimated MLDs coincided with the deepest levels of measurement of the float so that these estimates, possibly biased low (see Fig. 2d–f), were discarded from our analysis. These results show that convection deeper than 1300 m occurred during four consecutive winters SECF.

Figure 2Vertical distribution of σ₀, S and θ of Argo profiles showing an MLD deeper than 700 m SECF in winter 2017 (a, b, c) and in winter 2018 (d, e, f). The black points indicate the MLD. The triangles in (d) are the MLD that coincided with the maximal profiling pressure reached by the float. In the legend, the float and cycle of each profile are indicated.

Download

Although the number of floats showing deep convection in W2017 and W2018 was small (three and two floats, respectively), it represented a significant percentage of the floats operating in the SECF box at that time. The percentage of floats showing deep convection in the SECF region was computed for the deep convection periods defined from 15 January to 21 April 2015, 22 February to 21 March 2016, 16 March to 4 April 2017 and 24 February to 26 March 2018. The longest period of deep convection occurred in W2015 and the shortest in 2017. The percentages of floats showing deep convection during the deep convection period are 73 %, 50 %, 33 % and 50 % for the winters of 2015, 2016, 2017 and 2018, respectively. The lowest percentage is found for W2017, but it is still substantial. It might reflect the fact that for this specific year floats showing a deep MLD were found in the southwestern corner of the SECF box only, suggesting that convection did not occur over the full box.

Figure 3TS diagrams in the mixed layer for profiles with an MLD deeper than 700 m during the winters of 2015, 2016, 2017 and 2018 for (a) the Labrador Sea and (b) SECF. The properties of the mixed layers were estimated as the vertical means between 200 m and the MLD.

Download

The properties (σ₀, S and θ) of the end-of-winter mixed layer were estimated for the four winters (Table 1 and Fig. 3). We observed that, between W2015 and W2018, the water mass formed by deep convection significantly densified and cooled by 0.019 kg m⁻³ and 0.215 ^∘C, respectively (see Table 1 and Fig. 3).

In the Labrador Sea, the aggregate maximum depth of convection increased from 2015 to 2018 (see Table 1). Deep convection observed in the Labrador Sea in W2018 was the most intense since the beginning of the Argo era (see Fig. 2c in Yashayaev and Loder, 2016). From W2015 to W2018, newly formed LSW cooled, became saltier, and densified by 0.134 ^∘C, 0.013 and 0.023 kg m⁻³, respectively (Table 1).

The water mass formed SECF is warmer and saltier than that formed in the Labrador Sea (Fig. 3). The deep convection SECF is always shallower than in the Labrador Sea. This result is discussed later in Sect. 5.

4.2 Analysis of the atmospheric forcing southeast of Cape Farewell

The seasonal cycles of $B_{\mathrm{surf}}^{*}$ and Q are in phase and of the same order of magnitude, while FWF^*, which is positive and 1 order of magnitude lower than Q, does not present a seasonal cycle (Fig. S3). The means (1993–2018) of the cumulative sums from 1 September to 31 March of Q, FWF^* and $B_{\mathrm{surf}}^{*}$ estimated over the SECF box (Fig. 1) are $-\mathrm{2.46}\pm \mathrm{0.43}\times {\mathrm{10}}^{\mathrm{9}}$, $\mathrm{0.28}\pm \mathrm{0.10}\times {\mathrm{10}}^{\mathrm{9}}$ and $-\mathrm{2.22}\pm \mathrm{0.49}\times {\mathrm{10}}^{\mathrm{9}}$ J m⁻², respectively. $B_{\mathrm{surf}}^{*}$ is 10 % lower on average than Q because of the buoyancy addition by FWF^*. Considering the Ekman transports, the 1993–2018 means of the accumulated BF_ek, HF_ek and SF_ek from 1 September to 31 March amount to $\mathrm{0.37}\pm \mathrm{1.15}\times {\mathrm{10}}^{\mathrm{8}}$, $-\mathrm{0.35}\pm \mathrm{1.36}\times {\mathrm{10}}^{\mathrm{8}}$ and $\mathrm{0.02}\pm \mathrm{2.04}\times {\mathrm{10}}^{\mathrm{8}}\times {\mathrm{10}}^{\mathrm{9}}$ J m⁻², respectively. The horizontal Ekman heat flux is negative, while the Ekman buoyancy flux is positive. This buoyancy gain indicates a southeastward transport of surface freshwater caused by dominant winds from the southwest. It is noteworthy that BF_ek is 1 order of magnitude smaller than the $B_{\mathrm{surf}}^{*}$.

Figure 4Time series of anomalies of accumulated (a) $B_{\mathrm{surf}}^{*}$, (b) Q, (c) FWF^* (d) BF_ek, (e) HF_ek and (f) SF_ek averaged in the SECF region. They are anomalies with respect to 1993–2016. The accumulation was from 1 September to 31 August the following year. The winter North Atlantic Oscillation (NAO) index (Hurrell et al., 2018) is also represented in (g). The gray line in (a) is the sum of the anomalies of accumulated $B_{\mathrm{surf}}^{*}$ and BF_ek. Note that the range of values in the y axis is not the same in all the plots.

Download

The total atmospheric forcing SECF was quantified as the sum of $B_{\mathrm{surf}}^{*}$ and BF_ek. The anomalies of accumulated fluxes from 1 September to 31 August the year after, with respect to the mean 1993–2016, are displayed in Fig. 4 for the SECF box. The gray line in Fig. 4a is the total atmospheric forcing anomaly ($B_{\mathrm{surf}}^{*}$ plus BF_ek). We identify years with very negative buoyancy loss in the SECF region, e.g., 1994, 1999, 2008, 2012 and 2015. The very negative anomalies of atmospheric forcing in 1999 and 2015 were caused by the very negative anomalies in both $B_{\mathrm{surf}}^{*}$ (Fig. 4a) and BF_ek (Fig. 4d). This correlation was not observed for all the years presenting a negative anomaly of atmospheric forcing. It is noteworthy that during W2016, W2017 and W2018, the anomaly of atmospheric forcing was close to zero.

Contrary to the very negative anomaly in atmospheric fluxes over the SECF region observed for W2015, the atmospheric fluxes were close to the mean during W2016, W2017 and W2018.

Figure 5Time evolutions of vertical profiles measured from Argo floats in the SECF region: (a) θ, (b) S, (c) σ₁ and (d) the thickness of 0.01 kg m⁻³ thick σ₁ layers. The white horizontal bars in plots (a), (b) and (c) indicate the maximal convection depth observed in the Irminger Sea or SECF when deep convection occurred. The white line in plot (a) indicates the depth of the isotherm 3.6 ^∘C. The black vertical ticks on the x axes of plot (b) indicate times of Argo measurements. These figures were created from all Argo profiles reaching deeper than 1000 m in the SECF region (56.5–59.3^∘ N, 45–38^∘ W; pink box in Fig. 1). The yearly numbers of Argo profiles used in this figure are shown in Fig. S5.

Download

4.3 Analysis of the preconditioning of the water column southeast of Cape Farewell

Our hypothesis is that the exceptional deep convection that happened in W2015 in the SECF region favorably preconditioned the water column for deep convection the following winters. The time evolutions of θ, S, σ₁ and Δσ₁=0.01 kg m⁻³ layer thicknesses (Fig. 5) show a marked change in the hydrographic properties of the SECF region at the beginning of 2015 caused by the exceptional deep convection that occurred during W2015 (see Piron et al., 2017). The intermediate waters (500–1000 m) became colder than the years before, and despite a slight decrease in salinity, the cooling caused the density to increase (Fig. 5c). Figure 5d shows Δσ₁=0.01 kg m⁻³ layer thicknesses larger than 600 m appearing at the end of W2015 for the first time since 2002. In the density range 32.36–32.39 kg m⁻³, these layers remained thicker than ∼450 m during W2016 to W2018. This indicates low stratification at intermediate depths and a favorable preconditioning of intermediate waters for deep convection initiated by W2015 deep convection. The denser density of the core of the thick layers in 2017–2018 compared with 2015–2016 agrees with the densification of the mixed layer SECF shown in Table 1 and Fig. 3.

Figure 6Vertical profile of (a) the buoyancy to be removed (B), (b) the thermal component (B_θ) and (c) the salinity component (B_S). They were calculated from all Argo data measured in the SECF box (see Fig. 1) in September before the winter indicated in the legend. For W2015 and W2018, we considered data from 15 August to 30 September 2017 because not enough data were available in September. The number of Argo profiles taken into account to estimate the B profiles was more than 10 for all the winters.

Download

B(zi) is our estimate of the preconditioning of the water column before winter (see the Methods section). Figure 6a shows that, deeper than 100 m, B for W2016, W2017 and W2018 was smaller than B for W2015 or B for the mean W2008–W2014. Furthermore, for W2016, W2017 and 2018, B remained nearly constant with depth between 600 and 1300 m, which means that once the water column has been homogenized down to 600 m, little additional buoyancy loss results in the homogenization of the water column down to 1300 m. Both conditions, (i) less buoyancy to be removed and (ii) the absence of a gradient in the B profile down to 1300 m, indicate a more favorable preconditioning of the water column for W2016, W2017 and W2018 than during W2008–W2015.

To understand the relative contributions of θ and S to the preconditioning, we computed the thermal (B_θ) and haline (B_S) components of B ($B=B_{\mathit{\theta }}+B_{\mathrm{S}}$). In general, B_θ (B_S) increases with depth when θ decreases (S increases) with depth. On the contrary, a negative slope in a B_θ (B_S) profile corresponds to θ increasing (S decreasing) with depth and is indicative of a destabilizing effect. The negative slopes in B_θ and B_S profiles are not observed simultaneously because density profiles are stable.

We describe the relative contributions of B_θ and B_S to B by looking first at the mean 2008–2014 profiles (discontinuous blue lines in Fig. 6). B_θ accounts for most of the increase in B from the surface to 800 m and below 1400 m (see Fig. 6a and b). The negative slope in the B_S profile between 800 and 1000 m (Fig. 6c) slightly reduces B (Fig. 6a) and is due to the decrease in S associated with the core of LSW (see Fig. 3 in Piron et al., 2016). In the layer 1000–1400 m, the increase in B (Fig. 6a) is mainly explained by the increase in B_S (Fig. 6c), which follows the increase in S in the transition from LSW to Iceland Scotland Overflow Water (ISOW). This transition layer will be referred to hereinafter as the deep halocline. The preconditioning of the water column is usually analyzed in terms of heat (e.g., Piron et al., 2015, 2017). The decomposition of B in B_θ and B_S reveals that θ governs B in the layer 0–800 m. S tends to reduce the stabilizing effect of θ in the layer 800–1000 m and reinforces it in the layer 1000–1400 m by adding up to 1×10⁹ J m² to B.

Figure 7Decomposition of the profiles of buoyancy to be removed (B, continuous lines) in its thermal (B_θ, dotted lines) and salinity (B_S, dashed lines) components in (a) the SECF region; (b) the Labrador Sea. The B_S components for W2016 and W2018 were added to show the evolution of the depth of the deep halocline.

Download

In order to further understand why the SECF region was favorably preconditioned during the winters of 2016–2018, we compare the B_θ and B_S of W2017, which was the most favorably preconditioned winter, with the mean 2008–2014 (Fig. 7a). From the surface to 1600 m, B_θ and B_S were smaller for W2017 than for the mean 2008–2014. There are two additional remarkable features. First, in the layer 500–1000 m, the large reduction of B_θ compared to the 2008–2014 mean mostly explains the decrease in B in this layer. Second, the more negative value of B_S in the layer 1100–1300 m, compared to the 2008–2014 mean, eroded the B_θ slope, making the B profile more vertical for W2017 than for the mean. The more negative contribution of B_S in the layer 1100–1300 m comes from the fact that the deep halocline was deeper for W2017 (1300 m; see orange dashed line in Fig. 7a) than for the mean 2008–2014 (1000 m; see blue dashed line in Fig. 7a). Finally, we note that the profiles of B(z_i), B_θ(z_i) and B_S(z_i) for W2016 and W2018 are more similar to the profiles of W2017 than to those of W2015 or to the mean 2008–2014 (see Fig. 6), which indicates that the water column was also favorably preconditioned for deep convection in W2016 and W2018 for the same reasons as in W2017.

Figure 8Evolution of vertical profiles of monthly anomalies of (a) σ₀, (b) θ and (c) S at 58^∘ N, 40^∘ W. The anomalies were estimated from the ISAS database (Gaillard et al., 2016) and were referenced to the monthly mean estimated for 2002–2016.

Download

The origin of the changes in B is now discussed from the time evolutions of the monthly anomalies of θ, S and σ₀ at 58^∘ N–40^∘ W, which is at the center of the SECF box (Fig. 8). The time evolutions there are similar to those at any other location inside the SECF box. These anomalies were computed using ISAS Gaillard et al., 2016) and were referenced to the monthly mean of 2002–2016. A positive anomaly of σ₀ appeared in 2014 between the surface and 600 m (Fig. 8a) and reached 1200 m in 2015 and beyond. This positive anomaly of σ₀ correlates with a negative anomaly of θ. The latter, however, reached ∼1400 m of depth in 2016, which is deeper than the positive anomaly of σ₀. The negative anomaly of S between 1000 and 1500 m that appeared in 2015 and strongly reinforced in 2016 caused the negative anomaly in σ₀ between 1200 and 1500 m (the density anomaly caused by the negative anomaly in θ between 1200 and 1400 m does not balance the density anomaly caused by the negative anomaly of S).

The θ and S anomalies in the water column during 2016–2018 explain the anomalies of B, B_θ and B_S and can be summarized as follows. On the one hand, the properties of the surface waters (down to 500 m) were colder than previous years and, despite the fact that they were also fresher, they were denser. The density increase in the surface water reduced the density difference with the deeper-lying waters. The intermediate layer (500–1000 m) was also favorably preconditioned due to the observed cooling. Additionally, in the layer 1100–1300 m, the large negative anomaly of B_S with respect to its mean is explained by the decrease in S in this layer, which caused a decrease in σ₀ and consequently reduced the σ₀ difference with the shallower-lying water. The decrease in S also resulted in a deepening of the deep halocline.

Figure 9Horizontal distribution of the anomalies of S (a, b, c), θ (d, e, f) and σ₀ (g, h, i) in the layer 1200–1400 m in February 2015 (a, d, g), February 2016 (b, e, h) and February 2017 (c, f, i). The monthly anomalies were estimated from the ISAS database and were referenced to the period 2002–2016.

4.4 Atmospheric forcing versus preconditioning of the water column

We now use the estimates of the accumulated atmospheric forcing ($B_{\mathrm{surf}}^{*}+{\mathrm{BF}}_{\mathrm{ek}}$) from 1 September to 31 March the year after (see Fig. S4) to predict the maximum convection depth for a given winter based on September profiles of B. The predicted convection depth is determined as the depth at which B(zi) (Fig. 6a) equals the accumulated atmospheric forcing. The associated error was estimated by propagating the error in the atmospheric forcing (0.05×10⁹ J m⁻²). The accumulated atmospheric forcing amounted to $-\mathrm{3.21}\times {\mathrm{10}}^{\mathrm{9}}\pm \mathrm{0.05}$  $-\mathrm{2.21}\pm \mathrm{0.04}\times {\mathrm{10}}^{\mathrm{9}}$, $-\mathrm{2.01}\pm \mathrm{0.05}\times {\mathrm{10}}^{\mathrm{9}}$ and $-\mathrm{2.47}\pm \mathrm{0.05}\times {\mathrm{10}}^{\mathrm{9}}$ J m⁻² for W2015, W2016, W2017 and W2018, respectively. We found predicted convection depths of 1085±20, 1285±20, 1415±20 and 345±20 m for W2015, W2016, W2017 and W2018, respectively. We consider the aggregate maximum depth of convection to be the observed estimate of the MLD (Table 1). The predicted MLD agrees with the observed MLD within ±200 m. The differences could be due to errors in the atmospheric forcing (Josey et al., 2018), lateral advection and/or spatial variation in the convection intensity within the box not captured by the Argo sampling.

The satisfactory predictability of the convection depth with our 1-D model indicates that deep convection occurred locally. In spite of the fact that the atmospheric forcing was close to mean (1993–2016) conditions during W2016, W2017 and W2018, convection depths > 1300 m were reached in the SECF region. This was only possible thanks to the favorable preconditioning.