Nutrient transport pathways in the Lower St. Lawrence Estuary: seasonal perspectives from winter observations

The St. Lawrence Estuary connects the Great Lakes with the Atlantic Ocean. The accepted view, based on summer conditions, is that the Estuary’s surface layer receives its nutrient supply from vertical mixing processes. This mixing is caused by the estuarine circulation and tidal-upwelling at the Head of the Laurentian Channel (HLC). During winter when ice forms, historical process-based studies have been limited in scope. Winter monitoring has been typically confined to vertical profiles of salinity and temperature and near-surface water samples collected from a helicopter for nutrient analysis. In 2018, however, 5 the Canadian Coast Guard approved a science team to sample in tandem with its icebreaking and ship escorting operations. This opportunistic sampling provided the first winter turbulence observations, which covered the largest spatial extent ever measured during any season within the St. Lawrence Estuary and Gulf. The nitrate enrichment from tidal mixing resulted in an upward nitrate flux of about 30 nmol m−2s−1, comparable to summer values obtained at the same tidal phase. Further downstream, deep nutrient-rich water from the Gulf was mixed into the subsurface nutrient-poor layer at a rate more than an 10 order of magnitude smaller than at the HLC. These fluxes were compared to the nutrient load of the upstream St. Lawrence River. Contrary to previous assumptions, fluvial nitrate inputs are the most significant source of nitrate in the Estuary. Nitrate loads from vertical mixing processes would only exceed those from fluvial sources at the end of summer when fluvial inputs reach their annual minimum. 1 https://doi.org/10.5194/os-2021-59 Preprint. Discussion started: 7 July 2021 c © Author(s) 2021. CC BY 4.0 License.

1 Introduction Table 1. Overview of measurement campaigns. All campaigns included CTD profiling with pumped Seabird instruments (SBE9 and/or SBE19plus V2). Bottles provided in-situ nutrients concentrations, in particular nitrate. We present phosphate and silicate only during winter.
The accuracy of the SBE19's temperature and conductivity, were 0.005 • C and 0.0005 S/m. For the SBE9 and SBE-3F/SBE4-C on-board the VMP, the accuracy of temperature and conductivity were 0.001 • C and 0.0003 S/m. The accuracy of the pressure sensors were 0.015% of the full-scale for the SBE9, 0.1% for the SBE19 and VMP. depth profiler (CTD), nutrient concentrations derived from in-situ water samples. We also collected turbulence microstructure profiles in the LSLE and the Gulf since the USLE was too shallow for turbulence sampling ( Figure 1).

95
At each station, a SBE9 CTD profiler (Seabird Electronics) was mounted to a rosette that was operated through the ship's moon-pool. Another pumped CTD (SBE19plus, Seabird Electronics) was regularly deployed from the vessel's side shortly after the SBE9. The SBE19-CTD's purpose was mainly to measure Photosynthetically Active Radiation (PAR) in the upper 16-m of the water column. We generally collected in-situ water samples at 10 m (i.e., 2 m beneath the ship's haul), 25 m, 50 m, 100, 150 m, 250 m and bottom, water depth permitting. These sampling depths were adjusted at shallower stations within 100 the USLE and LSLE. At the station closest to Quebec City (U37) in 18 m depth, a single water sample was collected at 14 m depth with the rosette. Salinity was calculated from the temperature and conductivity sensors, which we present in practical salinity units (psu) herein.
All water samples obtained from the CTD-Rosette were filtered through a 0.7 µm GF/F filter using acid-washed syringes and Swinnex. The samples were analyzed immediately on board the vessel to derive nutrient concentrations. Concentrations 105 of NO3-+NO2-, NO2-, PO4-and Si(OH)4 were determined using a colorimetric method adapted from Hansen and Koroleff (2007) with a Bran and Luebbe Autoanalyzer III. We calculated NO3-concentrations by differencing NO2-from the NO3and +NO2-readings. Working standards were prepared at each station and were checked against reference standard material (KANSO CRM, lot.CH, high-Atlantic) only for the marine stations with the highest salinities. The detection limits were 0.03, 0.02, 0.03 and 0.05 µmol L −1 for NO3-+NO2-, NO2-, PO4-and Si(OH)4, respectively. The precision of the triplicates over 110 the observed range of concentrations was the same as or better than these detection limits. The samples yielded nitrate, nitrite, phosphate and silicate concentrations at different depths for each station visited.
At nine of the stations located between the Saguenay Fjord and Cabot Strait (Figure 1), we collected temperature, conductivity (salinity), and turbulence profiles with a VMP-500 manufactured by Rockland Scientific Ltd. We operated the Vertical Microstructure Profiler (VMP) from the ship's front deck. The VMP was fitted with two airfoil shear probes, two fast-response 115 thermistors (FP07, GE Thermometics), one micro-conductivity sensor and pressure and it sampled at 512 Hz. Only data from the shear probes (5% accuracy) and the pressure sensor are presented herein. The VMP was also equipped with a high-accuracy temperature (SBE-3F) and conductivity (SBE-4C) sensors from Seabird Electronics, which sampled at 64 Hz. These measurements enabled calculating the vertical salinity and density gradients using the pressure sensor on board the VMP.
Because of the Coast Guard's operations, the VMP and CTD-Rosette sampling occurred at different phases of the tides 120 ( Figure 2). With the VMP, we were unable to cover a complete semi-diurnal tidal cycle at any station. We obtained the best temporal coverage of the tidal cycle at station G163. Fourteen VMP profiles were collected during flood tide ( Figure 2). We attempted to cover another tidal cycle at station G294 on 19 February, but an ice-breaking request at the Magdalen Islands halted sampling after collecting seven profiles during ebb tide. This station was revisited on 21 February 2018 during the ebbing tides. For all other stations, the VMP collected two to three consecutive profiles ( Figure 2). 125

Historical monitoring surveys
Fisheries and Oceans Canada run the Atlantic Zone Monitoring Program (AZMP) multiple times each year (Therriault et al., 1998;Blais et al., 2019;Galbraith et al., 2019). Their monitoring consists of surveys in March, June, August, and November, which cover the Gulf and the LSLE (Figure 1a). Here, we present nitrate and salinities measured in summer and fall preceding our boreal winter campaign (Table 1). For completeness, we also present nitrate and salinity observations from the fall moni-130 toring survey of the St. Lawrence Ecosystem Health Research and Observation Network. These nitrate samples were analysed using the same procedures and equipment as those collected during the Odyssee 2018 winter field campaign.
The AZMP's ship-based fall and summer surveys provided standard CTD profiles, along with nutrient concentrations at similar depths to our winter campaign, in addition to samples closer to the surface at 2.5 and 5 m depth. The summer nitrate measurements were collected from two different AZMP cruises (Table 1). The first cruise, during June, collected profiles in the 135 downstream reaches of the USLE and near the HLC. The second cruise was in August and focused on the LSLE and the Gulf.
Similarly to the Odyssee winter cruise, the nitrate concentrations were derived using colorimetric methods. The AZMP uses CSK standards from the Sagami Research Center, Japan and participates in inter-calibration exercises through the International Council for the Exploration of the Sea (Mitchell et al., 2002). From the CSK standards, the AZMP's nitrate concentrations have an accuracy (rms) of 3.1, 1.7 and 1.8 % at concentrations of 5, 10 and 30 µmol L −1 , respectively. The reported values represent 140 the triplicates' average. Their coefficient of variations were on average less than 1.2%. The majority (95%) of the 136 sets of triplicates had coefficient of variations below 3%. The summer measurements are provided to contrast with our winter nutrient observations. We use the fall measurements to estimate the nitrate inventory generated between the fall monitoring survey and our winter observations within the LSLE.
We also present observations from the winter heli-survey conducted in mid-March 2018, a few weeks after our winter 145 program in February 2018 (Table 1)  Our paper focuses on establishing the main transport pathways for nutrients in the LSLE. In this region, nitrate concentrations are generally lower than silicate and, therefore, more likely to limit primary production (Tremblay et al., 1997;Jutras et al., 2020). Our subsequent analysis thus focuses on tracking nitrate inputs into the LSLE's surface layer. We treat the upper-75 m of the LSLE as a box. The box receives fluvial (horizontal) inputs of nitrate from the USLE and vertical inputs entering its base through mixing processes ( Figure 3a). These mixing processes include tidal-upwelling at the HLC and shear-induced mixing water column since mixing within the box redistribute the fluvial nitrate entering from the USLE. This shear-induced mixing at the base of the box is associated mainly with the estuarine circulation and occurs at the interface between the different layers of water. The vertical mixing caused by the estuarine circulation is more persistent than tidal-dependent upwelling mixing at the HLC. The vertical nitrate fluxes will be quantified using the techniques described in §3.2, and applied to the respective surface 160 areas of the HLC and the LSLE. These vertical nitrate loads will then be compared to the horizontal fluvial inputs described in § 4.3.1. Our analysis obviously ignores consumption of nutrients within the system, or any export. Our goal, however, is to compare the fluvial nitrate loads with those entering the box from vertical mixing processes.

Fluvial nitrate loads
We estimated the advected (horizontal) fluvial contributions of nitrate into the LSLE using nitrate concentrations and flow rates 165 at Quebec City. The nitrate loads were thus calculated at the upstream extent of the USLE, where water is fresh. We assume that the fluvial loads in the LSLE are representative of those at Quebec City, or alternatively, that our box includes the USLE.
The fluvial nitrate loads, expressed in mol s −1 , in the St. Lawrence River is calculated from: using the flow rate Q (m 3 s −1 ) and the nitrate concentration N (mmol m −3 ) at Quebec City. These flow rates were calculated 170 on a daily basis using the inverse modelling techniques recently developed by Bourgault and Matte (2020a, b). The approximate error on these daily and monthly flow rates are 0.25 ×10 4 m 3 s −1 and 0.16 ×10 4 m 3 s −1 , respectively (Bourgault and Matte, 2020a). This method of estimating flows replaces the older and less accurate method of Bourgault and Koutitonsky (1999), which is currently issued on a monthly basis by the Government of Canada.
The historical dissolved nitrate concentrations at Quebec City were digitized from published sources (

Turbulent vertical nitrate fluxes
We combined the VMP's turbulence profiles with the nutrients measurements to obtain vertical fluxes along the St. Lawrence during winter via: where z is the height above the free surface. The mixing rates K were derived from the VMP's measurements ( §3.2.1), while 185 the vertical background concentration gradients ∂N /∂z were derived for nitrate from the VMP profiles via a nitrate-salinity relationship developed by analysing the rosette's water samples ( §3.2.2).

Diapycnal mixing rates K
The most commonly-used model for estimating K was proposed by Osborn (1980) for shear-induced mixing: and requires estimating the rate of dissipation of turbulent kinetic energy and the background buoyancy frequency N = (−g/ρ)(∂ρ/∂z). A constant mixing efficiency Γ = 0.2 is often assumed, despite mounting evidence that it varies (e.g., Monismith et al., 2018). Several parametric models have been proposed (e.g., Ivey et al., 2018), and debated (e.g., Gregg et al., 2018), to relate Γ with external parameters.
During the winter field program, the temperature gradients were generally gravitationally unstable, i.e., cold water overlaying 195 warmer water. These unstable temperature gradients were stabilized by the salinity gradients. In these situations, doublediffusive convection (DDC) is possible. However, our observations lacked the presence of distinctive large (∼ several meters high) steps that are typically suggestive of DDC. Even when DDC dominates in weakly sheared flows, equation 3 can be used to estimate K by increasing Γ ∼ 1 (see Hieronymus and Carpenter, 2016;Polyakov et al., 2019). In these situations, buoyancy is the main source of mixing. Our turbulence levels were much higher than those reported by Polyakov et al. (2019) and the 200 strong tides are more conducive to shear-induced mixing than quiescent DDC mixing. We thus assume the custom value of Γ = 0.2 for shear-induced mixing. Our chosen Γ is consistent with field observations at low /(νN 2 ) (e.g., Holleman et al., 2016;Monismith et al., 2018). Here, ν represents the kinematic viscosity of seawater, while the ratio /(νN 2 ) is proportional to ratio of largest and smallest turbulent overturns in a stratified fluid. During our field campaign, /(νN 2 ) were 95% of the time less than 500.

205
To obtain the mixing rate K, we estimated using the methods described by Bluteau et al. (2016). Each profile was split into 4096 samples (8 s) that overlapped by 50% before computing the velocity gradient spectra. The VMP's profiling speed was derived from its pressure sensor to convert the spectra between the frequency and wavenumber domain. We applied the multivariate technique of Goodman et al. (2006) to remove motion-induced contamination from these spectra, which were then integrated over the viscous subranges to obtain . The VMP typically profiled at around 0.65 m s −1 , thus providing turbulence 210 estimates at a resolution of about 2.5 m given the 50% overlap when splitting the cast. Turbulence estimates near the end and the beginning of a cast were discarded because of the VMP's deceleration and acceleration, respectively. We also discarded estimates within 25 m of the surface because of the turbulence induced by the ship. To derive the density gradients ∂ρ/∂z, we relied on the high-accuracy temperature and conductivity sensors (SBE-3F and SBE-4C) aboard the VMP. We first low-pass filtered these signals with a Butterworth filter using a cutoff period of 8 s. We then centred-differenced these smoothed profiles 215 before averaging them over the same segments used for getting -yielding the mean vertical gradients necessary for using equations 2 and 3.

Proxy nitrate concentrations for estimating ∂N /∂z
Proxy nitrate concentrations are required to derive vertical nitrate gradients from the VMP's measurements (equation 2). This proxy made use of the winter dependency between nitrate and salinity. When limited amounts of nutrients are being consumed 220 (or generated), mixing and advection processes govern the spatial distribution of nutrients. Their concentrations vary linearly with salinity, as reflected by our winter observations in Figure 4f. A concave nitrate-salinity curve, such as observed during summer and fall 2017, indicates nitrate was being consumed along the USLE (gray points in Figure 4d,e). These trends in the nitrate-salinity diagrams are supported by incubation experiments that quantify the nitrate consumption (Villeneuve, 2020). than during winter. Nitrate consumption downstream of Quebec City was less than 0.50 nmol m −3 s −1 and about half as much in the LSLE (Figure 16 of Villeneuve, 2020). Hence, a nitrate-salinity relationship is thus justified to estimate nitrate from the VMP profiler's salinity measurements during winter.
Nitrate concentrations in winter depended on salinity but also on the location in the St. Lawrence (Figure 4). The nitrate variations are associated with the water masses in the region. The temperature-salinity diagram suggests two, possibly three, 230 significant water masses across the region (Figure 4a). The first water mass is nutrient-rich waters from the USLE that mix in the LSLE before eventually mixing with nutrient-poor surface water downstream in the Gulf (Figure 4d). This mixing resulted in nitrate N concentrations decreasing with salinity for S < 31.2 psu. For higher salinities, S > 31.9 psu, which includes samples deeper than 50-m in the LSLE and almost all samples in the Gulf, nitrates increased proportionally with salinity ( Figure   4f). This vertical nutrient distribution resembles the expected waters exposed to the open ocean, which we associate with the than water with comparable salinity in the USLE (Figure 4). This third water mass created a more rapid decrease in nitrate between the 31.2 to 32 psu salinity range before increasing again due to mixing with Gulf waters within the LSLE (Figure 4f).
From the observed water masses in winter, we created three separate nitrate-salinity relationships to obtain proxy nutrient concentrations the VMP's salinity measurements: (4)

245
The first relationship, applicable for S < 31.2 psu, reflects the nutrient-rich water in the USLE mingling with saltier water downstream. The third relationship included all samples with S > 31.9 psu. It was applied to turbulence profiles within the Gulf except for the deep data near Cabot Strait (station G540, 250m). This data corresponds to a fourth water mass, which we exclude from our vertical nitrate fluxes analysis. The second relationship links the other two using a quadratic fit to the samples outside of the Gulf within the range 31.2 < S ≤ 31.9. This relation reflects contributions from the Saguenay Fjord, which was 250 applied solely to turbulence profiles collected in the LSLE. Outside of the 31.2 < S ≤ 31.9 range, we applied the first or third relationship over their applicable salinity ranges. Typically, the first relationship was applied to surface waters in the LSLE that originated from the nutrient-rich USLE upstream. In contrast, the third relationship was used for deeper waters entering from the Gulf. A mean relative error of 5.5% was obtained for the predicted N after applying equation 4 to all the 64 samples in Figure 4f. The poorest agreement is with the low surface nitrate concentrations measured in the Gulf, particularly the furthest 255 downstream near Cabot Strait (station G540 in Figure 4f). The proxy nitrate concentrations calculated from (4) were then used to estimate the vertical nitrate gradients necessary for obtaining vertical fluxes from equation 2.

Vertical mixing contributions of nitrate
To convert the vertical nitrate fluxes into mass loadings, we rely on the same techniques as Cyr et al. (2015) to determine the tidal-upwelling zone. This mixing process creates a surface signature of cooler water during summer (see Figure 13 of Cyr

Winter conditions
The salinity transect highlight that we visited the HLC during an upwelling event (Figure 3b). The salinity was relatively homogeneous across the depth when we sampled at L0 during the flooding tide ( Figure 2). The water was nonetheless 2 psu 270 more saline than near-surface waters (i.e., 15 m) measured at stations both upstream and downstream from the HLC ( Figure   3b). The higher salinities at the HLC, station L0, cannot be attributed to inputs from the Saguenay Fjord. This waterway provides a freshwater source, confirmed by the AZMP's annual helicopter survey a few weeks later (Figure 4b). We attribute the relatively high salinities near the HLC to tidal upwelling and mixing that characterize this region throughout the year (Ingram, 1983;Galbraith, 2006;Cyr et al., 2015). At the HLC, nitrate concentrations in the upper 50-m were also lower than 275 water both upstream and downstream at comparable depths ( Figure 3a). This presence of low nitrate and high salinity water, especially in the upper 50-m, further supports that water was being tidally-upwelled.
The transects illustrate more clearly, than the nitrate-salinity diagrams, the presence of a sub-surface nitrate minimum in the LSLE (Figure 5b). During winter, nitrate concentrations were relatively high near the surface but decreased to reach a subsurface minimum at the 50-m deep sample. These subsurface samples were near or slightly more saline than 31.2 psu, 280 the corresponding break between nutrient-rich fluvial waters and the relatively nutrient-poor water downstream seen in the nitrate-salinity diagram (Figure 4a). In the LSLE, a nitrate-poor subsurface layer is overlaid by fresher and nitrate-rich surface waters.

Seasonal nitrate variations
We observed this subsurface nitrate-poor layer during other seasons (Figure 5b-d). Climatological averages show that a sub-285 surface nitrate minimum is typical of the LSLE's lower reaches during the ice-free months (see Figure 3 of Cyr et al., 2015).
Above this layer, near-surface nitrate concentrations are higher, especially during winter (e.g., ∼ 350 km in Figure 5b). These high nitrate concentrations were associated with low salinities (e.g., station L34 and L176 in Figure 4), reflecting the input of nutrient-rich fluvial water from the USLE. During winter, these fluvial waters likely extended downstream to 400-km from Quebec City (station L176; Figure 5b), but extended less far in the fall (300 km, Figure 5c). The subsurface nitrate minimum 290 that the input of fluvial waters creates was more evident in the winter and fall than in summer given the higher consumption (200-300 km, Figure 5d). However, the nitrate-salinity diagrams show that the fluvial waters may be perceptible just as far downstream into the LSLE during summer than in fall (Figure 4d-e). Upstream in the USLE, nitrate concentrations were highest in winter, followed by the fall and summer (Figure 5b-d). Overall, the USLE supplied nitrate-rich fluvial waters into the surface layer of the LSLE.

295
The nutrient inventory created in the upper 75-m of the LSLE during the fall and winter measurements is illustrated in Figure   5a. The inventory is obtained by depth-integrating the difference between the fall and winter measurements (Figure 5b and c).
This analysis assumes no nitrate consumption between these two seasons, i.e., no sink term. For completeness, we also added the inventory created between winter and summer, which is about 25% larger than the nitrate accumulated between winter and rates during these two seasons compared to summer (Figure 16 of Villeneuve, 2020). The LSLE's inventory includes mainly contributions from the nitrate-rich and fresher water originating from the USLE (i.e., S < 31.2 psu in Figure 4). The high values at 420 km reflect the nitrate-rich and fresher water from USLE during winter since the near-surface nitrate is relatively depleted during fall at this location (Figure 5a-b). The spatially-averaged inventory between these two seasons was 280 mmol m −2 , which translates to an equivalent vertical flux of 32 nmol m −2 s −1 given the 100 days between the fall and winter

Nitrate transport pathways
We now present the annual cycle of fluvial nitrate loads, so they can be compared with the vertical loads entering the LSLE from vertical mixing processes. This value is slightly higher than the 17-year average for February and also higher than the historical average of 300 ± 20 mol s −1 for the winter months of November through to February (Figure 6b). The fluvial nitrate loads peaked in April when the historical averages reach 500 mol s −1 before decreasing to their annual minimum in August and September. During these 325 late summer months, nitrate loads are below 150 mol s −1 (Figure 6b). These low nitrate loads coincide with the period when biological consumption is greatest upstream of Quebec City (Hudon et al., 2017) and flow rates are at their lowest. The fluvial inputs during the field experiment of Cyr et al. (2015) were about 90 mol s −1 , much lower than historical averages (Figure 6b).
Their fluvial nitrate loads were more than three times less than those during our winter campaign. The vertical nitrate fluxes were determined from the direct turbulence measurements along the Lower St. Lawrence Estuary and Gulf (Figure 7). The turbulence observations were most energetic in the upwelling-driven polynya near the HLC. At station L0, at the sill near the HLC, dissipation exceeded 10 −7 W kg −1 near the bottom and the surface (Figure 7a), while the mixing rates K exceeded 10 −4 m 2 s −1 throughout most of the water column (Figure 7b). The other stations were relatively quiescent with of the order 10 −9 W kg −1 . Elevated diapycnal mixing rates were also found in the surface mixed layer (depth <50 335 m), especially in the Gulf, presumably due to winter convective mixing processes reaching the deeper pycnocline (Figure 7d).
Elsewhere in the water column, outside of the energetic HLC, mixing rates were more typical of the ocean interior (O(10 −5 ) m 2 s −1 ; Waterhouse et al., 2014). Near the HLC, our winter "hot-spot", diapycnal mixing estimates were within the ranges observed during the summer months (e.g., Figure 11 of Cyr et al., 2015). They would have likely been higher if collected at high tide given the late summer observations of Cyr et al. (2015). They measured K ≈ 10 −2 m 2 s −1 at high tide between    We revisit the notion of tidal-upwelling and mixing processes, especially at the HLC, acting as the nutrient pump for the LSLE. The low consumption during winter provides a better representation of the physical mechanisms transporting nitrate into the LSLE than in summer. Summer observations invariably track both physical and biogeochemical processes. Studies

375
have reached variable conclusions about the importance of tidal-upwelling and mixing at the HLC in supplying nutrients to the LSLE, and ultimately the Gulf (see section 1). All of these studies concluded that vertical mixing processes dominate the supply of nutrients in the LSLE (e.g., Steven, 1974;Sinclair et al., 1976;Savenkoff et al., 2001;Cyr et al., 2015;Jutras et al., 2020;Greisman and Ingram, 1977 Cyr et al. (2015), yet their summer fluxes were some of the highest reported in the world (see Table 1 of Cyr et al., 2015).
These biogeochemical studies contrast with our winter observations since the fluvial nitrate loads dominated the supply of  Future field campaigns should focus on documenting the physical processes responsible for upwelling nitrate into the euphotic zone at the HLC. Specifically, estimating the magnitude of the vertical nitrate fluxes during an entire semi-diurnal tidal cycle for both neap and spring tides. Peak tides could pump nitrate into the surface layer at a much higher rate, as shown by Cyr et al. (2015). Differences in magnitude could exist between neap and spring tides (e.g., Sharples et al., 2007;Green et al., 2019). The goal of these additional measurements would be to remedy the short portion of the tidal cycle resolved during

Conclusions
The inaugural Odyssée campaign aboard the CCGS Amundsen icebreaker provided the first winter turbulence measurements in the St. Lawrence Estuary and Gulf. We collected these opportunistic measurements in tandem with the vessel's icebreaking 410 and ship escorting operations. These mixing measurements covered the most considerable spatial extent of the Estuary and Gulf during any season. Our analysis shows that tidal-upwelling appears to be a less effective mechanism for supplying nutrients into the euphotic zone during winter than in summer. In fact, for most of the year, we expect higher nitrate loads from fluvial sources than from tidal-upwelling. In particular, during freshet as the snowpack melts. Code and data availability. The code and computed flow rates are also publicly available at the codeocean repository (Bourgault and Matte, 2020b).
Author contributions. Conceptualisation of winter field campaign by CB, DB and PG. CB collected and processed the winter turbulence data observations, in addition to the formal analysis and data visualisation for the manuscript. VV collected and curated the nutrient observations scientific program of Quebec-Ocean. CEB thanks NSERC for funding a fellowship to carry out the work and acknowledges that the research contributes to the project "Quantifying and parameterising ocean mixing" funded by the Australian Research Council (DP180101736). We thank Pascal Guillot from Quebec-Ocean who quality-controlled the CTD aboard the rosette during the Odyssee cruise. We thank the Captain and crew of the CCGS Amundsen, and the staff from UQAR, Université Laval, who aided in the collection of data.