The suspended small-particle layer in the oxygen-poor Black Sea: a proxy for delineating the effective N2-yielding section

The shallower oxygen-poor water masses of the ocean confine a majority of the microbial communities that can produce up to 90 % of oceanic N2. This effective N2yielding section encloses a suspended small-particle layer, inferred from particle backscattering (bbp) measurements. It is thus hypothesized that this layer (hereafter, the bbp-layer) is linked to microbial communities involved in N2 yielding such as nitrate-reducing SAR11 as well as sulfur-oxidizing, anammox, and denitrifying bacteria – a hypothesis yet to be evaluated. Here, data collected by three BGC-Argo floats deployed in the Black Sea are used to investigate the origin of this bbp-layer. To this end, we evaluate how the key drivers of N2-yielding bacteria dynamics impact the vertical distribution of bbp and the thickness of the bbp-layer. In conjunction with published data on N2 excess, our results suggest that the bbp-layer is at least partially composed of the bacteria driving N2 yielding for three main reasons: (1) strong correlations are recorded between bbp and nitrate; (2) the top location of the bbp-layer is driven by the ventilation of oxygen-rich subsurface waters, while its thickness is modulated by the amount of nitrate available to produce N2; and (3) the maxima of both bbp and N2 excess coincide at the same isopycnals where bacteria involved in N2 yielding coexist. We thus advance that bbp and O2 can be exploited as a combined proxy to delineate the N2-yielding section of the Black Sea. This proxy can potentially contribute to refining delineation of the effective N2-yielding section of oxygendeficient zones via data from the growing BGC-Argo float network.


Introduction
Oxygen-poor water masses (O 2 < 3 µM) host the microbial communities that produce between 20 % and 40 % of oceanic N 2 mainly via heterotrophic denitrification and anaerobic oxidation of ammonium (Gruber and Sarmiento, 1997;DeVries et al., 2013;Ward, 2013). The shallower oxygen-poor water masses (∼ 50-200 m) are the most effective N 2 -producing section because this is where the microbial communities that condition the process mainly develop and generate up to 90 % of the N 2 (Ward et al., 2009;Dalsgaard et al., 2012;Babbin et al., 2014). These microbial communities include nitrate-reducing SAR11 and anammox, denitrifying, and sulfur-oxidizing bacteria (e.g., Canfield et al., 2010;Ulloa et al., 2012;Ward, 2013;Tsementzi et al., 2016;Callbeck et al., 2018). It is thus important to unravel the biogeochemical parameters that trigger the accumulation of such bacteria in the ocean's oxygen-poor water masses. This information is crucial for understanding and quantifying how bacterial biomass and related N 2 -yielding bacteria can respond to the ongoing expansion of oceanic regions with low oxygen (Keeling and Garcia, 2002;Stramma et al., 2008;Helm et al., 2011;Schmidtko et al., 2017). Ultimately, greater accuracy in this domain can contribute to improving mechanistic predictions on how such expansion will affect the oceans' role in driving the Earth's climate by sequestering atmospheric carbon dioxide (e.g., Oschlies et al., 2018).
In oxygen-poor water masses, the biogeochemical factors that can affect the abundance of denitrifying and anammox bacteria are the levels of O 2 , organic matter (OM), nitrate (NO − 3 ), ammonium (NH + 4 ), and hydrogen sulfide (H 2 S) (Murray et al., 1995;Ward et al., 2008; gers the confinement of such bacteria, we need to investigate how the above biogeochemical factors drive their vertical distribution, with high temporal and vertical resolution. To this end, we should develop multidisciplinary approaches that allow us to permanently monitor the full range of biogeochemical variables of interest in oxygen-poor water masses. Optical proxies of tiny particles can be applied as an alternative approach to assess the vertical distribution of N 2 -yielding microbial communities in oxygen-poor water masses (Naqvi et al., 1993). For instance, nitrate-reducing SAR11 and anammox, denitrifying, and sulfur-oxidizing bacteria are found as free-living bacteria (0.2-2 µm) and can be associated with small suspended (> 2-30 µm) and large sinking (> 30 µm) particles (Fuchsman et al., 2011(Fuchsman et al., , 2012a(Fuchsman et al., , 2017Ganesh et al., 2014Ganesh et al., , 2015. Therefore, particle backscattering (b bp ), a proxy for particles in the ∼ 0.2-20 µm size range (Stramski et al., 1999(Stramski et al., , 2004Organelli et al., 2018), can serve to detect the presence of these free-living bacteria and those associated with small suspended particles.
Time series of b bp acquired by biogeochemical Argo (BGC-Argo) floats highlight the presence of a permanent layer of suspended small particles in shallower oxygenpoor water masses (b bp -layer) (Whitmire et al., 2009;Wojtasiewicz et al., 2020). It has been hypothesized that this b bplayer is linked to N 2 -yielding microbial communities such as nitrate-reducing SAR11 and denitrifying, anammox, and sulfur-oxidizing bacteria. However, this hypothesis has not yet been clearly demonstrated. To address this, the first step is to evaluate (1) potential correlations between the biogeochemical factors that control the presence of the b bp -layer and such arrays of bacteria (O 2 , NO − 3 , OM, H 2 S; Murray et al., 1995;Ward et al., 2008;Fuchsman et al., 2011;Ulloa et al., 2012;Dalsgaard et al., 2014;Bristow et al., 2016) and (2) the possible relationship between the b bp -layer and N 2 produced by microbial communities. This first step is thus essential for identifying the origin of the b bp -layer and, ultimately, determining whether BGC-Argo observations of b bp can be implemented to delineate the oxygen-poor water masses where such bacteria are confined. The Black Sea appears a suitable area for probing into the origin of the b bp -layer in low-oxygen waters in this way. It is indeed a semi-enclosed basin with permanently low O 2 levels where N 2 production and related nitrate-reducing SAR11 and denitrifying and anammox bacteria are mainly confined within a well-defined oxygen-poor zone (Kuypers et al., 2003;Konovalov et al., 2005;Kirkpatrick et al., 2012). In addition, a permanent b bp -layer is a typical characteristic of this region, which is linked to such microbial communities and inorganic particles (Stanev et al., 2017(Stanev et al., , 2018; see details in Sect. 2).
The goal of our study is therefore to investigate the origin of the b bp -layer in the oxygen-poor waters of the Black Sea using data collected by BGC-Argo floats. More specifically, we aim to evaluate, within the oxygen-poor zone, how (1) two of the main factors (O 2 and NO − 3 ) that drive the dy-namics of denitrifying and anammox bacteria impact the location and thickness of the b bp -layer, (2) NO − 3 controls the vertical distribution of b bp within this layer, (3) temperature drives the formation of the b bp -layer and consumption rates of NO − 3 , and (4) particle content inferred from b bp and N 2 produced by microbial communities can be at least qualitatively correlated. Ultimately, our findings allow us to infer that b bp can potentially be used to detect the presence of the microbial communities that drive N 2 production in oxygenpoor water masses -including nitrate-reducing SAR11 and sulfur-oxidizing, denitrifying, and anammox bacteria.
Raw data of fluorescence and total backscattering were converted into chlorophyll concentration (chl) and particle backscattering (b bp ) following standard protocols, respectively (Schmechtig et al., 2014(Schmechtig et al., , 2015. Spike signals in vertical profiles of chl and b bp and due to particle aggregates were removed by using a median filter with a window size of three data points (Briggs et al., 2011). NO − 3 , HS − , and O 2 data were processed following BGC-Argo protocols (Bittig and Körtzinger, 2015;Johnson et al., 2018;Thierry et al., 2018). Sampling regions covered by the three floats encompassed most of the Black Sea area ( Fig. 1 and Appendix A). However, we only used data collected during periods without a clear injection of small particles derived from the productive layer and Bosporus plume (e.g., advection of water masses, Stanev et al., 2017). This restriction allowed us to focus on the in situ 1D processes driving local formation of the b bplayer, with minimal interference from any possible external sources of small particles.
We only describe the time series of data collected by float 6901866 because this was the only float carrying a NO − 3 /HS − sensor. Data acquired by floats 6900807 and 7900591 are described in Appendix A and nevertheless used as complementary data to those of float 6901866 to corroborate (1) qualitative correlations between O 2 levels and the location of the b bp -layer and (2) consistency in the location of the b bp maximum within the b bp -layer.

Defining the oxygen-poor zone, mixed-layer depth, and productive layer
We used O 2 and NO − 3 to, respectively, define the top and bottom isopycnals of the oxygen-poor zone where denitrifying and anammox bacteria are expected to be found. To set the top isopycnal, we applied an O 2 threshold of ∼ 3 µM because denitrifying and anammox bacteria seem to tolerate O 2 concentrations beneath this threshold (Jensen et al., 2008;Dalsgaard et al., 2014;Babbin et al., 2014). The bottom isopycnal was defined as the deepest isopycnal at which NO − 3 was detected by the SUNA sensor (0.23 ± 0.32 µM). NO − 3 was used to set this isopycnal because heterotrophic denitrification and subsequent reactions cannot occur without NO − 3 (Lam et al., 2009;Bristow et al., 2017). HS − was not used to delimit the bottom of this zone because the maximum concentration of HS − that denitrifying and anammox bacteria tolerate is not well established (Murray et al., 1995;Kirkpatrick et al., 2012; see also Sect. 4.1).
Mixed-layer depth (MLD) was computed as the depth at which density differed from 0.03 kg m −3 with respect to the density recorded at 1 m depth (de Boyer Montégut et al., 2004). We used chl to define the productive layer where living phytoplankton were present and producing particulate organic carbon. The base of this layer was set as the depth at which chl decreased below 0.25 mg m −3 . This depth was used only as a reference to highlight the periods when surface-derived small particles were clearly injected into the oxygen-poor zone.  (Fuchsman et al., 2008(Fuchsman et al., , 2019 were exploited to complement discussion of our results. N 2 produced by anaerobic microbial communities (N 2 excess, µM) was estimated from N 2 : Ar ratios and argon concentrations at atmospheric saturation (Hamme and Emerson, 2004). N 2 excess data were used to (1) describe the oxygen-poor zone where N 2 is expected to be predominantly produced and (2) highlight qualitative correlations between N 2 excess, the location of the b bp -layer, and vertical distribution of small particles within the b bp -layer.
4 Results and discussion

Description of the oxygen-poor zone
The top and bottom of the oxygen-poor zone are located around isopycnals (mean ± standard deviation) 15.79± 0.23 and 16.30 ± 0.09 kg m −3 , respectively. The two isopycnals therefore delimit the oxygen-poor water masses where nitrate-reducing SAR11 and denitrifying, anammox, and sulfur-oxidizing bacteria are expected to be found (zone hereafter called the OP DA , Fig. 2; Kuypers et al., 2003;Lam et al., 2007;Yakushev et al., 2007;Fuschman et al., 2011;Kirkpatrick et al., 2012). The top location of the OP DA shows large spatial-temporal variability ranging between 80 and 180 m (or σ θ between 15.5 and 15.9 kg m −3 , Fig. 2). Similarly, the OP DA thickness varies between 30 and 80 m, which corresponds to a σ θ separation of ∼ 0.50 kg m −3 . The bottom of the OP DA is slightly sulfidic (HS − = 11.4 ± 3.53 µM, n = 86) and deeper than suggested (e.g., σ θ = 16.20 kg m −3 and H 2 S ≤ 10 nM; Murray et al., 1995). However, our results coincide with the slightly sulfidic conditions of the deepest isopycnal at which anammox bacteria can still be recorded (σ θ = 16.30 kg m −3 and H 2 S ≥ 10 µM; Kirkpatrick et al., 2012).

NO −
3 , O 2 , and MnO 2 as key drivers of the thickness and location of the suspended small-particle layer The permanent b bp -layer is always confined within the two isopycnals that delimit the OP DA (Fig. 2). It follows that the thickness and top location of this layer demonstrate the same spatial and temporal variability as the one described for the OP DA (Fig. 2 and Appendix A). This correlation indicates that variations in the thickness and top location of the b bplayer are partially driven, respectively, by (1) the amount of NO − 3 available to produce N 2 inside the OP DA via the set of bacteria communities involved and (2) downward ventilation of oxygen-rich subsurface waters ( Fig. 2 and Appendix A).
NO − 3 and O 2 are two of the key factors that modulate the presence of (1) denitrifying and anammox bacteria working in conjunction with nitrate-reducing SAR11 (Fuschman et al., 2011;Ulloa et al., 2012;Tsementezi et al., 2016;Bristow et al., 2017), and probably with chemoautotrophic ammoniaoxidizing bacteria (in this case, only with anammox, e.g., γ AOB; Ward and Kilpatrick, 1991;Lam et al., 2007), and (2) sulfur-oxidizing bacteria (e.g., SUP05 and potentially Epsilonproteobacteria Sulfurimonas; Canfield et al., 2010;Glaubitz et al., 2010;Fuschman et al., 2011Fuschman et al., , 2012bUlloa et al., 2012;Kirkpatrick et al., 2018). Therefore, the results described above highlight that at least a fraction of the b bp -layer should be due to this array of bacteria. This notion is supported by three main observations. Firstly, the top location of the b bp -layer is driven by the intrusion of subsurface water masses (S ≤ 20.36 ± 0.18 psu) with O 2 concentrations above the levels tolerated by denitrifying and anammox bacteria (O 2 ≥ 3 µM, Jensen et al., 2008;Babbin et al., 2014;Fig. 2). As a result, in regions where O 2 is ventilated to deeper water masses, the top location of the b bp -layer is also deeper. The contrary is observed when O 2 ventilation is shallower ( Fig. 2 and Appendix A). Secondly, nitrate-reducing SAR11 and denitrifying, anammox, and sulfur-oxidizing bacteria reside between isopycnals 15.60 and 16.30 kg m −3 (Fuchsman et al., 2011(Fuchsman et al., , 2012aKirkpatrick et al., 2012), while the b bp -layer is formed between isopycnals ∼ 15.79 and 16.30 kg m −3 . We can thus infer coexistence of such bacteria between the coincident isopycnals where the b bp -layer is generated. Thirdly, NO − 3 declines from around isopycnal 15.79 kg m −3 to isopycnal 16.30 kg m −3 due to the expected N 2 production via the microbial communities involved (Figs. 2-3 and Kirkpatrick et al., 2012).
The ventilation of subsurface O 2 is also key in driving the depth at which MnO 2 is formed (O 2 ≤ 3-5 µM; Clement et al., 2009) and can thus contribute to setting the characteristics of the b bp -layer via its subsequent accumulation and dissolution (Konovalov et al., 2003;Clement et al., 2009;Dellwig et al., 2010). Thus, in regions where subsurface O 2 (e.g., O 2 ≥ 3-5 µM and S ≤ 20.36±0.18 psu) is ventilated to deeper water masses, both the formation of MnO 2 and top location of the b bp -layer can be expected to be deeper and vice versa (Fig. 2). Finally, the dissolution of MnO 2 should also influence the thickness of the b bp -layer because it occurs just beneath the maxima of the optical particles inside this layer (Konovalov et al., 2006; see the explanation in Sect. 4.3).
Overall, the qualitative evidence presented above points out that particles of MnO 2 as well as nitrate-reducing SAR11 and denitrifying, anammox, and sulfur-oxidizing bacteria appear to define the characteristics of the b bp -layer (Johnson, 2006;Konovalov et al., 2003;Fuchsman et al., 2011Fuchsman et al., , 2012bStanev et al., 2018). This observation leads us to argue, in the next section, that the b bp -layer is partially composed of the main group of microbial communities involved in N 2 yielding as well as of MnO 2 . We propose that the removal rate of NO − 3 is a key driver of the vertical distribution of small particles and N 2 excess within the OP DA . This is because the vertical profiles of small particles and of N 2 excess are qualitatively similar, and both profiles are clearly related to the rate at which NO − 3 is removed from the OP DA (Figs. 3-4). For instance, maxima of N 2 excess and b bp coincide around isopycnal 16.11 ± 0.11 kg m −3 (Fig. 3; Konovalov et al., 2005;Fuchsman et al., 2008Fuchsman et al., , 2019. At this isopycnal, the mean concentration of NO − 3 is 1.19 ± 0.53 µM. We thus propose that this NO − 3 threshold value splits the OP DA into two sub-zones with distinctive biogeochemical conditions (e.g., nitrogenous and manganous zones; Canfield and Thamdrup, 2009). Ultimately, these two different sets of conditions drive the rates at which NO − 3 and small particles are removed and formed within the OP DA , respectively ( Fig. 3 and explanation below).
The first sub-zone is thus located between the top of the OP DA (σ θ = 15.79 kg m −3 ) and around isopycnal 16.11 kg m −3 . Here, removal rates of NO − 3 (−0.16 ± 0.10 µM m −1 , Fig. 4) are likely to be boosted by (1) high content of organic matter (dissolved organic carbon = 122 ± 9 µM, Margolin et al., 2016) and NO − 3 (≥ 1.19 ± 0.53 µM) and (2) O 2 levels staying between a range that maintains the yielding of N 2 (0.24 ± 0.04 µM ≥ O 2 ≤ 2.8 ± 0.14 µM, n = 100, the means of the minima and maxima of O 2 , respectively, in the first sub-zone) and promotes the formation of MnO 2 (e.g., maximum of Mn(II) oxidation is at O 2 levels ∼ 0.2 µM; Clement et al., 2009). Consequently, the formation of biogenic and inorganic small particles (and related N 2 excess) increases from the top of the OP DA to around isopycnal 16.11 kg m −3 (Fig. 3). This hypothesis is (1) in part confirmed by significant and negative power-law correlations between the suspended small-particle content and NO − 3 in this sub-zone (Fig. 3) and is (2) in agreement with the progressive accumulation of MnO 2 from around isopycnal 15.8 kg m −3 to isopycnal 16.10 kg m −3 (e.g., Konovalov et al., 2006).
The second sub-zone is located between isopycnal 16.11 kg m −3 and the bottom of the OP DA (σ θ = 16.30 kg m −3 , Fig. 3). Here, NO − 3 is low (≤ 1.19 ± 0.53 µM) and O 2 is relatively constant (0.23 ± 0.02 µM, n = 2284; mean of O 2 calculated in the second sub-zone for all profiles) or lower than the minimum of O 2 recorded by this sensor (0.22 ± 0.02 µM, n = 89). These constant (or lower) levels of O 2 roughly correspond to those at which anammox and heterotrophic denitrification are inhibited by ∼ 50 % (0.21 and 0.81 µM, respectively; Dalsgaard et al., 2014). In addition, low levels of NO − 3 necessarily promote the microbial use of Mn(IV) as an electron acceptor, ultimately dissolving the particles of MnO 2 into Mn(II) (e.g., manganous zone; Konovalov et al., 2006;Yakushev et al., 2007;Canfield and Thamdrup, 2009). As a result, this sub-zone exhibits a decline in removal rates of NO − 3 (−0.04 ± 0.01 µM m −1 , Fig. 4) along with inhibited formation of biogenic small particles and dissolution of MnO 2 . Ultimately, both the content of small particles and related N 2 excess decrease from around isopycnal 16.11 kg m −3 to the bottom of the OP DA (Fig. 3). These results are in agreement with (1) significant and positive exponential correlations computed between the small-particle content inferred from b bp and NO − 3 within this sub-zone (Fig. 3) and (2) the overlap of nitrogenous and manganous zones in this sub-zone because the content of MnO 2 particles and dissolved Mn(II) concurrently declines and increases just beneath isopycnal 16.11 kg m −3 , respectively (e.g., Murray et al., 1995;Konovalov et al., 2003Konovalov et al., , 2005Konovalov et al., , 2006Yakushev et al., 2007;Canfield and Thamdrup, 2009).
Strong-positive linear correlations are also recorded between b bp and T in the first sub-zone of the OP DA (Fig. 4). This is likely to indicate that the formation of small particles is sensitive to very tiny increments in T (0.003 ± 0.001 • C m −1 , n = 133). We thus infer a tendency for the decline rates of NO − 3 and related production of N 2 to increase with T . This hypothesis is at least partially supported by the significant correlation between NO − 3 decline rates and T increase rates in this sub-zone (Fig. 4). Within the second sub-zone, T continues increasing while b bp decreases, likely due to inhibition of the formation of small particles for the reasons described above (Fig. 4). These observations suggest that the production of small particles is likely to have first-and second-order covariations with NO − 3 and T , respectively -a likelihood backed up by a lack of correlation between NO − 3 decline rates and T increase rates in this subzone (Fig. 4). Finally, more information is needed to investigate the physical and/or biogeochemical processes driving the correlation between the increase rates of T and declines rates of NO − 3 in the first sub-zone. This is however beyond the scope of our study.
To summarize, BGC-Argo float data combined with a proxy of N 2 production suggest that in regions without the Bosporus plume influence, the b bp -layer systematically tracks and delineates the effective N 2 -yielding section independently of (1) the biogeochemical mechanisms driving N 2 yielding and (2) the contribution that MnO 2 and other microorganisms can be expected to make to the formation of the b bp -layer (e.g., Lam et al., 2007;Fuchsman et al., 2011Fuchsman et al., , 2012aKirkpatrick et al., 2018). It is thus finally in-ferred that this b bp -layer is at least partially composed of the predominant anaerobic microbial communities involved in the production of N 2 , such as nitrate-reducing SAR11 and anammox, denitrifying, and sulfur-oxidizing bacteria. These results also suggest that N 2 production rates can be highly variable in the Black Sea because the characteristics of the b bp -layer show large spatial-temporal variations driven by changes in NO − 3 and O 2 (Figs. 2 and 4). Finally, we propose that b bp and O 2 can be exploited as a combined proxy for defining the N 2 -producing section of the oxygen-poor Black Sea. We consider that this combined proxy can delineate the top and base of this section by applying an O 2 threshold of 3.0 µM and the bottom isopycnal of the b bp -layer, respectively. This section should thus be linked to free-living bacteria (0.2-2 µm) and those associated with small suspended

New perspectives for studying N 2 yielding in oxygen-deficient zones
The conclusions and inferences of this study, especially those related to the origin and drivers of the b bp -layer, primarily apply to the Black Sea. However, these findings may also have a wider application. In particular, the shallower water masses of oxygen-deficient zones (ODZs) are similarly characterized by the formation of a layer of suspended small particles that can be optically detected by b bp and the attenuation coefficients of particles (Spinrad et al., 1989;Naqvi et al., 1993;Whitmire et al., 2009). This layer is mainly linked to N 2 -yielding microbial communities because (1) its location coincides with the maxima of N 2 excess, microbial metabolic activity, and nitrite (NO − 2 , the intermediate product of denitrification and anammox that is mainly accumulated in the N 2 -yielding section: Spinrad et al., 1989;Naqvi et al., 1991Naqvi et al., , 1993Devon et al., 2006;Chang et al., 2010Chang et al., , 2012Ulloa et al., 2012;Wojtasiewicz et al., 2020), and (2) MnO 2 is not accumulated as in the Black Sea (Martin and Knauer, 1984;Johnson et al., 1996;Lewis and Luther, 2000). Therefore, our findings suggest that highly resolved vertical profiles of b bp and O 2 can potentially be used as a combined proxy to define the effective N 2 -production section of ODZs. Such a definition can be key to better-constrained global estimates of N 2 loss rates because it can allow us to (1) accurately predict the oxygen-poor water volume where around 90 % of N 2 is produced in the ODZ core (Babbin et al., 2014) and (2) evaluate how the location and thickness of the N 2 -yielding section vary due to changes in the biogeochemical factors that modulate anammox and heterotrophy denitrification.
Global estimates of N 2 production differ by 2-3-fold between studies (e.g., 50-150 Tg N yr −1 , Codispoti et al., 2001;Bianchi et al., 2012Bianchi et al., , 2018DeVries et al., 2012;Wang et al., 2019). These discrepancies are caused in part by inaccurate estimations of the oxygen-poor volume of the N 2production section. Other sources of uncertainties arise from the methods applied to estimate the amount of particulate organic matter (POM) that fuels N 2 production. For instance, POM fluxes and their subsequent attenuation rates are not well resolved because they are computed, respectively, from satellite-based primary-production algorithms and generic power-law functions (Bianchi et al., 2012(Bianchi et al., , 2018DeVries et al., 2012). POM-flux estimates based on these algorithms visibly exclude (1) POM supplied by zooplankton migration (Kiko et al., 2017;Tutasi and Escribano, 2020), (2) substantial events of POM export decoupled from primary production (Karl et al., 2012), and (3) the role of small particles derived from the physical and biological fragmentation of larger ones (Karl et al., 1988;Briggs et al., 2020), which are more efficiently remineralized by bacteria in ODZs (Cavan et al., 2017). In addition, these estimates do not take into con-sideration the inhibition effect that O 2 intrusions may have on N 2 -yielding rates (Whitmire et al., 2009;Ulloa et al., 2012;Dalsgaard et al., 2014;Peters et al., 2016;Margolskee et al., 2019).
Overall, mechanistic predictions of N 2 production misrepresent the strong dynamics of the biogeochemical and physical processes that regulate them. Consequently, it is still debated whether the oceanic nitrogen cycle is in balance or not (Codispoti, 2007;Gruber and Galloway, 2008;DeVries et al., 2012;Jayakumar et al., 2017;Bianchi et al., 2018;Wang et al., 2019). The subsiding uncertainty points to a compelling need for alternative methods that allow accurate refinement of oceanic estimations of N 2 production.
Our study supports the proposition that robotic observations of b bp and O 2 can be used to better delineate the N 2yielding section at the appropriate spatial (e.g., vertical and regional) and temporal (e.g., event, seasonal, interannual) resolutions. In addition, POM fluxes and N 2 can be simultaneously quantified using the same float technology (BGC-Argo, Bishop and Wood, 2009;Dall'Olmo and Mork, 2014;Reed et al., 2018;Boyd et al., 2019;Estapa et al., 2019;Rasse and Dall'Olmo, 2019). These robotic measurements can contribute to refining global estimates of N 2 production by better constraining both the oxygen-poor section where N 2 is produced and POM fluxes that fuel its loss. Ultimately, O 2 intrusions into the N 2 -yielding section can potentially be quantified by BGC-Argo floats to assess their regulatory effect on N 2 production.

Conclusions
Our results along with those from previous studies suggest that the b bp -layer of the oxygen-poor Black Sea is at least partially composed of nitrate-reducing SAR11 and anammox, denitrifying, and sulfur-oxidizing bacteria. The location and thickness of this layer show strong spatial-temporal variability, mainly driven by the ventilation of oxygen-rich subsurface waters and nitrate available to generate N 2 , respectively. Such variations in the characteristics of the b bplayer highlight that N 2 -production rates can be highly variable in the Black Sea. We therefore propose that highresolution measurements of O 2 and b bp can potentially be exploited as a combined proxy to delineate the effective N 2yielding section of ODZs. This proposition is in part supported by evidence that the b bp -layer and a majority of N 2yielding microbial communities are both confined in the shallower oxygen-poor water masses of ODZs. We however recommend investigation into the key biogeochemical drivers of the b bp -layer for each ODZ. This information will be critical for validating the applicability of the b bp -layer in assessing spatial-temporal changes in N 2 production.
Finally, it is evident that BGC-Argo float observations can acquire essential proxies of N 2 production and associated drivers at appropriate spatial and temporal resolutions.
The development of observation-modeling synergies therefore has the potential to deliver an unprecedented view of N 2 -yielding drivers if robotic observations become an integrated part of model validation. Ultimately, this approach could prove essential for reducing present uncertainties in the oceanic N 2 budget.   ), respectively. SU, A, W, and SP stand for summer, autumn, winter, and spring, respectively. The colored horizontal line at the bottom indicates the sampling site for a given date (Fig. A1). The horizontal white lines in (d) are the profiles used to (1) delimit the OP DA and (2) find the isopycnals at which b bp is maximum in the OP DA . chl is set to zero in the OP DA due to fluorescence contamination (Stanev et al., 2017).
Data availability. These data were collected and made freely available by the International Argo Program and the national programs that contribute to it (Argo, 2020). The Argo Program is part of the Global Ocean Observing System. Data on N 2 : Ar ratios are freely available at https://doi.org/10.1029/2018GB006032 (Fuchsman et al., 2019).
Author contributions. RR conceptualized the study, wrote the original draft, and generated all the figures. HC contributed to tuning the study's conceptualization and figure design. AP processed all BGC-Argo float data. RR and HC reviewed and edited the final manuscript.
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. This study was conducted under the framework of the Marie Skłodowska-Curie Individual Fellowship awarded to Rafael Rasse (NOCEANIC project). This study is a contribution to the REMOCEAN project (H. Claustre), and the final writing was funded by the REFINE project (H. Claustre). We finally thank Clara A. Fuchsman and the anonymous reviewer for their accurate and constructive feedback, which allowed us to significantly improve the original version of the manuscript.
Financial support. This research has been supported by the European Union's Horizon 2020 research and innovation program (NO-CEANIC project, grant no. 839062), and the European Research Council, Seventh Framework Programme (REMOCEAN project: grant no. 246777;and REFINE project: grant no. 834177).
Review statement. This paper was edited by Aninda Mazumdar and reviewed by Clara A. Fuchsman and one anonymous referee.