Do Oceanic Hypoxic Regions Act as Barriers for Sinking Particles? A Case Study in the Eastern Tropical North Atlantic

In oxygen minimum zones (OMZs), the attenuation rates of particulate organic carbon (POC) fluxes of large particles are known to be reduced, thus increasing the efficiency with which the biological carbon pump (BCP) transfers carbon to the abyss. The BCP efficiency is expected to further increase if OMZs expand. However, little is known about how the POC fluxes of small particles—a significant component of the BCP—are attenuated inside OMZs. In this study, data collected by two BGC‐Argo floats deployed in the hypoxic OMZ of the eastern tropical North Atlantic were used to estimate net instantaneous fluxes of POC via small particle during 3 years. This information was analyzed together with meteorological data and published POC fluxes of large particles and allowed us to conclude that (1) major pulses of surface‐derived small particles toward the OMZ interior coincided with seasonal changes in wind stress and precipitation; (2) a permanent layer of small particles, presumably linked to microbial communities, was found in the upper section of the OMZ which might play a key role attenuating POC fluxes; and (3) fluxes of large particles were attenuated less efficiently inside this poorly oxygenated region than above it, while attenuation of small‐particle fluxes were equivalent or significantly higher inside the OMZ. These results highlight that more information about the processes controlling the fluxes of small and large particles in hypoxic OMZs is needed to better understand the impact of hypoxic OMZs on the BCP efficiency.


Introduction
The biological carbon pump (BCP) is a set of mechanisms that removes atmospheric carbon dioxide by exporting particulate organic carbon (POC) from the upper ocean toward its interior (Volk & Hoffert, 1985). As it sinks to the abyss, most POC is remineralized by bacteria back to CO 2 and reexchanged with the atmosphere on time scales that depend on the depth at which remineralization takes place (Kwon et al., 2009). Thus, understanding what controls POC fluxes and their remineralization is crucial to mechanistically predict how the BCP controls atmospheric CO 2 concentration and in turn mitigates the Earth's climate (Falkowski et al., 1998;Parekh et al., 2006;Sarmiento & Gruber, 2006).
In the North Atlantic, fragmentation-derived small-particle fluxes contribute between 40% and 85% of the total POC fluxes in the first 650 m depth, and these fluxes can show a strong seasonality linked to changes in the mixed layer depth (MLD) (Dall'Olmo & Mork, 2014;Durkin et al., 2015;Giering et al., 2016). These results demonstrate that the export of these tiny particles is a significant component of the BCP that varies over time. Little is known however about POC fluxes via small particles and their attenuation rates in OMZs. POC fluxes measured by conventional sediment traps are mostly due to large sinking particles because small particles are not efficiently captured by these tools (Arístegui et al., 2009;Buesseler et al., 2007;Gardner et al., 1985;Trull et al., 2008). Although methods exist to collect small particles (e.g., Marine Snow Catcher, Durkin et al., 2015;Giering et al., 2016;Riley et al., 2012), these are still labor intensive and time consuming and provide data with limited temporal, vertical, and spatial resolution. Therefore, complementary techniques are needed to assess what drives the temporal variations and attenuation rates of POC fluxes by small particles in OMZs.
Net instantaneous fluxes of POC by small particles can be estimated from quasi-continuous measurements of particle backscattering (b bp , a proxy of POC; Stramski et al., 1999Stramski et al., , 2004 collected by autonomous profiling floats (i.e., "BGC-Argo," Dall'Olmo & Mork, 2014). Even though a typical range of 0.2-20 μm is assumed for these small particles (Dall'Olmo & Mork, 2014), recent findings have demonstrated that the small particles detected by b bp should range approximately between~1 and 10 μm at least in the sunlit oligotrophic ocean (Organelli et al., 2018). Thus, the dynamics of these small particles can be investigated using data from BGC-Argo floats to fill knowledge gaps about the BCP (e.g., Bishop & Wood, 2009;Dall'Olmo et al., 2016;Estapa et al., 2017Estapa et al., , 2019Lacour et al., 2017;Mignot et al., 2018). However, so far BGC-Argo observations have not been used to study particle dynamics inside OMZs.
The aim of this study was to investigate particle export fluxes and their attenuation rates in the hypoxic OMZ of the ETNA. We used data collected from two BGC-Argo floats deployed in the ETNA OMZ to estimate vertically resolved net instantaneous fluxes of POC by small particles during 3 years. This information, together with complementary meteorological data retrieved from remote sensing and published POC fluxes by large particles, allowed us to demonstrate that (1) major pulses of small particles derived from the fragmentation of large aggregates were related to seasonal changes in wind stress and precipitation; (2) a permanent layer rich in small particles, presumably linked to bacteria, was present between the upper and intermediate layers of the OMZ and might have a key role attenuating the fluxes of small-particle POC; (3) in the poorly oxygenated waters studied, the attenuation rates of small-particle POC fluxes were equivalent or significantly higher relative to those in the well-oxygenated regions above the OMZ, whereas large particles fluxes were attenuated less efficiently inside the OMZ than above it.

Site Description
In the ETNA, the seasonal meridional migration of the intertropical convergence zone (ITCZ, Figure 1) drives wind stress and precipitations (Nobre & Shukla, 1996). This forcing contributes to controlling primary production and the export of particles from the surface to the abyss (Pastor et al., 2013;Schlosser et al., 2014;Signorini et al., 1999).
±0.5°and longitude ±0.5°) for a backward period of 5 days. This backward period started from the day of the profile and ended 5 days before it. These wind products were generated by the European Centre for Medium-Range Weather Forecasts by using scatterometers Advanced scatterometer (ASAT) and Oceansat-2 scatterometer (OSCAT) retrievals with a horizontal resolution of 0.25 × 0.25 degrees.
Similarly, cumulative daily area-averaged precipitation rates (mm d −1 ) were also extracted at the profile locations and for the 5 days before each profile. These data were retrieved from the TRMM multisatellite precipitation analysis with a temporal and spatial resolution of 3 hours (3B42RT) and 0.25 × 0.25 degrees, respectively (https://giovanni.gsfc.nasa.gov/giovanni/).

Bio-optical and Physicochemical Parameters
To assess how the ITCZ migration drives particle dynamics in this region, we used data collected with high temporal and vertical resolution by two Biogeochemical-Argo (BGC-Argo) floats. These floats are uniquely identified by World Meteorological Organization (WMO) numbers 6901174 and 6901175 and profiled every 5 days in the first 1000 m depth of the ETNA OMZ since November 2014 until January 2018 (n = 235-245 upward profiles for each BGC-Argo float, Figure 1). The two floats remained within a relative small area that ranged between 310×390 and 250×690 km 2 , respectively, and were separated by a meridional distance of approximately 600 km ( Figure 1). This separation allowed us to assess how the seasonal location of the ITCZ (e.g., Figure 1) impacted surface stratification and related production and export of particles in two regions of the ETNA OMZ.
BGC-Argo floats were equipped with a CTD, Aanderaa oxygen optode, and WETLabs sensors to measure upward profiles of temperature (T), conductivity, depth, dissolved oxygen (O 2 ), chlorophyll fluorescence (chl), and total optical backscattering (particles + pure seawater) at 700 nm. Raw data of total backscattering and fluorescence were converted into particle backscattering (b bp ) and chl following standard protocols (Schmechtig et al., 2014(Schmechtig et al., , 2015. In addition, b bp and chl spikes were removed by applying a median filter with a window size of three data points (e.g., Briggs et al., 2011;Dall'Olmo & Mork, 2014).

Minimizing Uncertainties in b bp Estimates
Biases in raw b bp due to instrumental drift were first removed (see supporting information). Furthermore, an additional potential problem was identified in the b bp data from one of the floats. For the first 14 months of operation, the b bp meters of both floats were programmed to collect measurements with a vertical resolution of 10 dbars below 600 meters depths (close to isopycnal 27.2 kg m −3 ). Subsequently, the b bp meter resolution was increased to 1 dbar (supporting information). For float 6901175, this change in the vertical resolution of b bp measurements did not affect the mean b bp values determined in the water masses below 600 meters depth (supporting information). In contrast, for float 6901174, during the last 4 months (August 2015 and January 2016) of the sampling period with low vertical resolution, the mean b bp were~20% lower in the same deepwater masses than after the change in resolution (supporting information). The reason for this change is unclear but could be similar to the problem described for these sensors by . These authors hypothesized that this decrease in b bp likely could have resulted from (1) a longer time response of the sensor to achieve optimal performance and/or (2) variations in the light source intensity due to changes in the water temperature (Sullivan et al., 2013;Wojtasiewicz, Trull, et al., 2018). Thus, the b bp data collected during this period were not used to calculate the stocks of POC. The latter data were only exploited to describe the small-particle dynamics above 600 m depth (0.2-20 μm, e.g., sections 2.2 and 2.3 and Figures 4-6) because above this depth, there were no changes in the vertical resolution of b bp measurements.

Delimiting MLD, Productive Region, and OMZ
Potential density (σ ɵ ) was derived from T and S (salinity) and the MLD was computed using a σ ɵ threshold difference of 0.03 kg m −3 from a depth of 1 meter (de Boyer Montégut et al., 2004). To derive a proxy for stratification, we calculated the difference between σ ɵ at 100 m and 10 m depth (σ ɵ, e.g., Lozier et al., 2011). This proxy allowed us to synthesize in a single metrics the effects that wind stress and precipitation had on surface stratification to investigate the related production of surface particles (e.g., Figures 2 and 3). The bottom of the productive region (z p ) was estimated as the depth at which chl decreases below 0.05 mg m −3 (Figures 2  and 3). We used chl to define the productive layer where living phytoplankton are present because our technique to compute export requires that no production is taking place below a given layer of the water column. We found that z p was always deeper than the MLD with average values (±standard deviation, WMO number) of 139 m (±12, 6901174) and 130 m (±13, 6901175); z p approximately followed the 26.6 kg m −3 isopycnal (Figures 2 and 3).
Dissolved oxygen (O 2 ) concentrations were quality controlled following published procedures (supporting information, Takeshita et al., 2013). No well-defined method to delimit hypoxic OMZs exists because of the wide range of O 2 concentrations encompassed by them (e.g., <60-120 μM, Stramma et al., 2008). Here, to define the top, center, and bottom of the OMZ, isopycnals were used because they allowed us to consistently delimit, in both sampled regions, the water mass characterized by hypoxic conditions (oxygen concentrations <60-90 μM). To this end, the isopycnal corresponding to the local minimum of dissolved oxygen concentration was used as the reference isopycnal for the center of the OMZ core. For the whole time series, we found that the minimum O 2 concentration (although different in magnitude) occurred approximately along the same central isopycnal (σ ɵ = 27.05 kg m −3 ) in the two regions sampled by the two floats.
After having identified the isopycnal 27.05 kg m −3 as the one corresponding to the minimum oxygen concentration along the vertical profile, we selected two additional isopycnals above and below the central isopycnal to define the water mass that contained the OMZ core. These two additional isopycnal were manually selected and differed by ±0.15 kg m −3 from the central isopycnal. Although subjective, this choice allowed us to consistently define the water mass that contained the hypoxic region (O 2 < 60-90 μM). Thus, the top (OMZ top ), center (OMZ center ), and bottom (OMZ bottom ) of the OMZ were defined as the isopycnals 26.9, 27.05, and 27.2 kg m −3 , respectively (Figures 2, 3, and section 2.1). Although subjective, our choice allows us to define the OMZ core consistently with the subdivision of the water column (by isopycnals) that we adopted to compute POC fluxes (section 1.2.5).

POC Stocks and Export
No published relationships between b bp and POC were available for the mesopelagic ETNA OMZ; thus b bp was converted into POC concentrations (POC, mg m −3 ) by using a published bio-optical POC-b bp relationship for the surface Atlantic (log 10 (POC) = 0.870*log 10 (b bp ) + 4.28; Rasse et al., 2017). Predicted POC concentrations (8-26 mg m −3 ) were similar to those previously measured in the mesopelagic region (≥150-900 m) of the ETNA OMZ (18-60 mg m −3 , Martiny et al., 2014).
Water column layers were then delimited using isopycnals separated by a potential density (σ ɵ ) threshold of 0.05 kg m −3 (e.g., between isopycnals 26.6 and 26.65 kg m −3 ). The thickness of each layer was calculated as the mean distance separating the two isopycnals during the time series.
Stocks of POC (mg m −2 ) in progressively deeper layers of the water column were estimated by multiplying the mean thickness of the layers (m) by their corresponding average concentration of POC (mg m −3 ) for each profile. Specifically, the stocks of POC were computed for the layers delimited from selected isopycnals (σ ɵi ) to the deepest isopycnal sampled by the float (27.5 kg m −3 , σ max ). We indicate these POC stocks with the notation iPOC 27:5 σθi , where σ ɵi = 26.6 kg m −3 + n, n = 0, 0.05, 0.1, … , 0.8 kg m −3 ; and 27.5 kg m −3 is the deepest isopycnal sampled (e.g., Figures 4 and 5). In addition, to track the temporal variations in surface production of small particles, surface POC stocks were calculated between 1 m and z p .
Net instantaneous fluxes of POC just below a given isopycnal (σ ɵi ) were computed as the time rate of change of iPOC 27:5 σ θi (E σ θi ¼ ∂iPOC 27:5 σ θi =∂t) after applying a pseudo-Gaussian smoothing filter to iPOC 27:5 σ θi (e.g., orange and red lines in Figures 7 and 8). Here the main assumptions are a negligible export of particles below σ max and spatial homogeneity of the particle field (Dall'Olmo & Mork, 2014). Thus, positive E σ θi values correspond to the net rate at which small particles accumulated below a given isopycnal. Uncertainties and details of the method applied to predict POC fluxes are discussed in Dall'Olmo and Mork (2014).

Global Biogeochemical Cycles
Transfer efficiencies (T eff , %) of POC fluxes below z p and OMZ top were calculated as the ratio between Eσ ɵi : E ref multiplied by 100 (Buesseler & Boyd, 2009). E ref indicates the POC fluxes just beneath z p or OMZ top , while Eσ ɵi indicates fluxes at progressively deeper isopycnals below z p or OMZ top .
To compare T eff values in poorly and well-oxygenated water layers that differed in thickness, a new metric denominated "depth-specific attenuation rate" was developed (ΔT, % m −1 ). This new metric allows one to compare the attenuation of POC fluxes between water layers, independently of the thickness of these layers.
More specifically, ΔT was calculated as: Here T eff-R and T eff-σ are the transfer efficiencies at the reference and progressively deeper isopycnals, respectively (e.g., T eff-σ = Eσ ɵi :E ref ); d is the distance between the reference and deeper isopycnals.
For water layers above the OMZ (ΔT above ), the isopycnal following the depth of the productive region was used as reference and T eff-σ was calculated between isopycnals >26.6 and ≤ 26.9 kg m −3 (e.g., Figures 9,  10). Similarly, inside the OMZ layers (ΔT inside ), the isopycnal at the top of the OMZ was used as reference and T eff-σ was computed between isopycnals >26.9 and ≤ 27.2 kg m −3 (e.g., Figures 9, 10).
ΔT values inside and above the OMZs were compared and discussed for each float separately, because O 2 concentrations differed by 30% between the two sampled regions (e.g., section 1.2.4, and Figures 11, 12). Our main purpose was to assess how the attenuation rates of POC fluxes varied in poorly and well-oxygenated regions relative to the in situ O 2 levels that characterized each sampled region.
Finally, ΔT values were also calculated by using published absolute POC fluxes measured by drifting sediment traps and mostly due to large particles in the region sampled by float 6901175 between March and April 2014 (Engel et al., 2017, Figures 9, 10). Thus, ΔT served as normalized metric to independently quantify and compare the remineralized proportion of small and large POC particles in the waters above and inside

10.1029/2019GB006305
Global Biogeochemical Cycles the OMZ. However, when comparing ΔT values derived from net and absolute fluxes, it should be kept in mind that these fluxes will differ between them. The reasons of these differences are discussed in section 3.3.

Particle Dynamics and Surface POC Stocks Inferred from b bp
In the ETNA OMZ, maxima and minima of surface stratification were respectively triggered by (1) high precipitation and weak wind stress (June-November) and (2) strong wind stress and no significant precipitation (December-April). These changes in stratification coincided with temporal changes in the dynamics of small particles (0.2-20 μm; Figures 4 and 5). For instance, surface POC stocks reached maximum values when the wind-rain driven surface stratification began between June and September (e.g., Figures 6-8). Weaker increases in surface POC stocks were also observed between the late winter and spring in some of the years, likely due to vertical mixing of nutrients from underlying waters (January-May, period of lower stratification, e.g., Figures 6-8). During this period, surface POC stocks tended to increase only between February and May of 2015 and 2016 in the region sampled by float 6901175 (Figure 8).
A permanent layer relatively rich in small particles was observed in both regions between the upper and intermediate layers of the OMZ (i.e., between isopycnals 26.9-27.1 kg m −3 ), whereas fewer particles were found below the OMZ bottom (σ ɵ = 27.2 kg m −3 , Figures 4 and 5). In this permanent layer, mean POC concentrations tended to increase mostly when surface POC stocks accumulated (e.g., Figure 6).

Surface-Mesopelagic Stocks and Net Instantaneous Fluxes of POC
Surface and mesopelagic POC stocks tended to increase between June-September and January-May (Figures 7, 8). During these periods, large particles could have been exported beneath z p . Ultimately, the fragmentation of these large particles likely generated pulses of small particles beneath z p and throughout the mesopelagic (Figures 7, 8). These pulses of small particles beneath z p are hereafter referred to as summer-autumn and winter-spring pulses, respectively.

Global Biogeochemical Cycles
When these pulses of small particles occurred, net instantaneous POC fluxes beneath z p (E zp ) ranged between 0.02-12 and 0.1-31 mg C m −2 d −1 for floats 6901175 and 6901174, respectively (Figures 7 and 8).
Flux attenuation rates inside the OMZ were significantly higher than those above the OMZ for 56% of the profiles assessed for float 6901174 (p < 0.05, paired t test, number of profiles assessed: 33, Figure 11) and 70% of the profiles assessed for float 6901175 (p < 0.05, paired t test, number of profiles assessed: 27, Figure 12).
There were the periods during which small-particle stocks decreased and net instantaneous POC fluxes beneath z p were negatives (i.e., particles were lost below a given isopycnal). These negatives fluxes ranged between −22.6 and −0.18 mg C m −2 d −1 for float 69011744 and between −27.3 and −0.03 mg C m −2 d −1 for float 69011745 (Figures 7 and 8). In addition, there were also specific periods when POC stocks beneath z p did not change significantly with time (e.g., September-November 2016 for float 6901175, Figure 8).

Key Drivers of Net Instantaneous Fluxes of Small Particles
Several forcing mechanisms may contribute to producing and transporting biogenic particles from the sunlit surface toward the ocean interior (e.g., strong summer-export, mixed-layer, eddy-subduction, and local-/ large-scale subduction pumps, Boyd   Here, we hypothesize that two forcing mechanisms might have partially contributed to generate pulses of small particles from z p to the OMZ: (1) strong summer export events that coincided with higher stratification periods triggered by elevated precipitations and low wind stress (summer-autumn pulses, Figures 7, 8) and (2) local subduction pump linked to wind-driven vertical mixing of nutrients from underlying waters (winter-spring pulses, e.g., Ekman pumping, Boyd et al., 2019;McClain & Firestone, 1993;Pastor et al., 2013;and Figures 7, 8).
Strong summer export events that coincided with stronger stratification appeared to be the most important mechanism related to the net accumulation of small particles inside the productive layer and their export beneath z p (summer-autumn pulses, Figures 7, 8). This export of POC beneath z p ranged between 0.  center (OMZ center ), and bottom (OMZ bottom ) of the OMZ, respectively. White circles and colored squares in (c) are the profiles (n = 27) used to calculate the depthspecific transfer efficiencies (T, % m −1 ) described in Figures 9-11. Red, yellow, and gray squares are the profiles randomly selected for three temporal events of net instantaneous fluxes of small particles beneath z p (these temporal events were (1) the summer-autumn pulse of 2016, (2) the winter-spring pulse of 2017, and (3) the summer-autumn pulse of 2017). The latter profiles were used to prepare Figure 11 and those in the supporting information. Black double arrows at the bottom of (c) indicate the periods when there were no positive net pulses of small particles beneath z p .
Summer-autumn pulses contributed on average around 65% (±16%) to the net annual fluxes of small particles beneath z p (230 ± 54 mg C m −2 year −1 ). These pulses of small particles were likely triggered by a complex synergy of meteorological, physical, and biogeochemical mechanisms. For instance, in this region, Saharan dust and related iron are efficiently scavenged and deposited on the surface ocean by wet deposition (Kieber et al., 2003(Kieber et al., , 2005Schlosser et al., 2014;Shi et al., 2012). As a result, the surface ocean becomes enriched in dissolved iron (DFe) and Trichodesmium abundance tends to increase by 1-2 orders of magnitude (Schlosser et al., 2014;Snow et al., 2015). Ultimately, Trichodesmium blooms collapse and might generate strong episodic exports of large sinking particles possibly reinforced by ballasting effect (Iversen & Robert, 2015;Pabortsava et al., 2017;van der Jagt et al., 2018).
The above hypothesis was partially supported by the progressive accumulation of surface small particles after precipitation events (and inferred related DFe deposition) and subsequent stratification began during the summer in most of the years (e.g., between June and September, Figures 7, 8). Accumulation of small particles likely followed physical and biological fragmentation of large sinking particles generated after phytoplankton blooms collapsed (e.g., Trichodesmium). Ultimately, fragmentation might have increased smallparticle POC in the mesopelagic region during summer and autumn (Figures 7, 8).
Net annual POC fluxes beneath z p (E zp ) due to winter-spring pulses were on average~2 times lower than those generated by summer-autumn pulses. In addition, the former E zp values showed a high temporal variability (Figures 7, 8). We hypothesize that this temporal variability in the E zp values was at least in part due to changes in the amount of nutrients injected from underlying waters to z p (e.g., due to variations in wind-driven vertical mixing of nutrients, McClain & Firestone, 1993; Pastor et al., 2013).

Global Biogeochemical Cycles
Negative net instantaneous fluxes of POC were observed after the end of the winter-spring and summerautumn pulses beneath z p (E zp , Figures 7, 8). The observed temporal variability in these negative fluxes was likely driven by the differences between the rates at which surface small particle was produced and remineralized. For example, the most negative rates of E zp were found between September-December of 2015 and April-May of 2016 for float 69011745 (Figure 8). This probably indicated that the remineralization rates of small particles beneath z p were more significant than their surface production rates during this period (e.g., between September 2015 and February of 2016, Figure 8). For the same float, the decrease of POC stocks was less steep between September and December of 2016 (E zp ranged between −3.0 and −0.7 mg C m −2 d −1 , Figure 8). We hypothesize that these differences between the removal and production rates of small particles were low because episodic events of production of surface POC occurred during this period (e.g., October and November of 2016, Figure 8).
In summary, two main mechanisms could have contributed to surface production of small particles and related carbon transfer from z p toward the OMZ interior: (1) a strong summer export that coincided with higher stratification periods triggered by low wind stress and elevated precipitations (plus inferred related DFe deposition, Trichodesmium blooms, and their collapse) and (2) the local subduction pump linked to wind-driven vertical mixing of nutrients from underlying waters. However, as we will discuss in next sections, these mechanisms might not be as efficient in exporting small-particle POC (0.2-20 μm) to the ocean interior as previously suggested for large particles in the same region (e.g., Engel et al., 2017).

Dynamics of the Permanent Layer of Small Particles Inside the OMZ
Anoxic OMZs (O 2 < 5 μM) are characterized by a permanent layer of suspended particles likely related to anaerobic microbial communities (e.g., Günter et al., 2008;Karl & Knauer, 1991;Ulloa et al., 2012;Whitmire et al., 2009;. Here, to the best of our knowledge, we report for the first time that this feature might not be unique to anoxic OMZs because a permanent layer relatively rich in small particles was found between the upper and intermediate layers of the hypoxic OMZ of the ETNA (Figures 4, 5).
Within the OMZ, the concentrations and stocks of POC reached their maxima after the winter-spring and summer-autumn pulses occurred (Figures 6-8). However, the permanent OMZ layer of small particles persisted for the whole time series in both regions (Figures 4, 5). This persistence, even outside of periods with clear particle fluxes from the surface, suggests small particles might be autonomously produced in the upper layers of the OMZ (Figures 2 and 3). This production is likely linked to the activity of free-living aerobicmicrobial communities (e.g., chemosynthesis by Archaea, Engel et al., 2017;Herndl et al., 2005;Löscher et al., 2012;Raven, 2009;Reinthaler et al., 2006;Wright et al., 2012).
In support of this hypothesis, there is evidence of an enhancement in the activity of aerobic-anaerobic heterotrophic bacteria associated with large sinking particles in the upper layers of the OMZ. Engel et al. (2017) reported an increment in (1) the ratio between particle hydrolyzable amino acids (PHAA, a chemical marker of aerobic-anaerobic heterotrophic bacterial growth efficiency) and POC in large sinking particles in the center of the OMZ sampled by float 6901175 between March and April 2014 (Engel et al., 2017, e.g.,~400 m, Figure 3) and (2) the contribution of γ-amino butyric acid (a proxy of aerobic-anaerobic heterotrophic bacterial decomposition activity; Lee & Cronin, 1982) to the PHAA in sinking large particles in the top of the

Global Biogeochemical Cycles
OMZ in the same region and sampled period (Engel et al., 2017, e.g.,~300 m, Figure 3). These two pieces of evidence are consistent with large particles intercepting a large number of free-living aerobic-heterotrophic bacteria in the same part of the water column where we observed the permanent layer of small particles (e.g., Archea, Löscher et al., 2012). However, we cannot rule out that at low oxygen levels (<90 μM), anaerobicheterotrophic bacteria colonization of the large particles interior could have been induced by the expected expansion of their anoxic core. Ultimately, this could also contribute to increasing the concentration of both chemical markers (PHAA and γ-amino butyric acid, Bianchi et al., 2018;Klawonn et al., 2015).
During the periods without pulses of small particles, we registered negative values of net instantaneous fluxes of POC in just the center of the OMZ (Figures 7, 8). These results suggest that (1) within the permanent layer of particles presumably linked to aerobic-heterotrophic bacteria, net consumption of exported small particles was significant during these periods and (2) this layer was not a net source of small particles to deeper layers (>500 m) of the ETNA. On the contrary, as discussed in the next section, this layer might have acted as a barrier for small particles derived from the fragmentation of large sinking particles.

Depth-Specific Attenuation Above and Inside the OMZ
The hypothesis that the potential expansion of anoxic and hypoxic OMZs will increase the efficiency with which POC is transported to the ocean's interior appears to be strengthening (e.g., Cavan et al., 2017;Engel et al., 2017;Le Moigne et al., 2017). However, inside the ETNA OMZ, the attenuation rates of net POC fluxes by small particles were significantly higher than those above the OMZ for 62% of the profiles (p < 0.05, paired t test, Figures 11, 12). For the remaining 38% of the profiles, the attenuation rates of the

10.1029/2019GB006305
Global Biogeochemical Cycles fluxes were not significantly different above and inside the hypoxic OMZ (p > 0.05, paired t test, Figures 11,  12). Thus, our results suggest that the above hypothesis might need to be revised for small-particle fluxes in hypoxic OMZs like the ETNA OMZ.
In addition, inside the ETNA OMZ, flux attenuation tended to decrease with depth (Figures 9, 10). This might indicate that inside this layer (1) small particles derived from the fragmentation of large sinking particles became more recalcitrant with depth and/or (2) flux attenuation was low below the permanent layer of small particles, likely because most of the labile POC is consumed in this layer.
When we compared our estimates to flux attenuations computed from published data , we found differences that are worth discussing. Specifically, the mean ΔT values estimated above the OMZ were quantitatively similar to those estimated from absolute fluxes of POC measured by drifting sediment traps in the same region (p > 0.05, t test), although these measurements were collected during a different period of the year (March-April 2014; Engel et al., 2017, and Figures 9, 10, 13, and supporting information). In contrast, inside the OMZ, our mean ΔT was significantly higher (p < 0.05, t test) than that derived for absolute fluxes of POC in the study of Engel et al. (2017) (Figures 9, 10, Figure 13). To understand these differences, we need to remember that our ΔT values were derived from net instantaneous fluxes of small particles instead of absolute fluxes of large sinking particles.
The net instantaneous POC fluxes beneath z p reported in our study (0.02-31 mg C m −2 d −1 ) were lower than the published absolute fluxes at similar depths (49-76 mg C m −2 d −1 ). We propose three main reasons to explain these differences: (1) our net fluxes of POC are only measurable when the production of small particles is larger than their loss and thus are a lower bound for the absolute fluxes (POC fluxes at the steady state cannot be quantified).
(2) Our flux estimates only include the contribution of particles smaller than at most 20 μm (Organelli et al., 2018). Finally, (3) to estimate the rate of change of POC stocks, data were smoothed, which decreased even more the net instantaneous fluxes of POC via small particles. These differences must be kept in mind when comparing ΔT values derived from net and absolute fluxes.

Global Biogeochemical Cycles
Despite these differences, our results highlighted that attenuation rates of small particles in poorly oxygenated regions of the OMZ can be equivalent (p > 0.05) or significantly higher (p < 0.05) than those in well-oxygenated regions (Figures 11, 12). On the other hand, our analysis of the data from Engel et al. (2017) demonstrated that absolute POC fluxes of large particles were less attenuated inside the OMZ (p < 0.05, t test, Figure 13).
We thus expect that inside the OMZ, the contribution of small particles to the total POC fluxes (small + large particles) would be reduced relative to the shallower well-oxygenated regions (e.g., ≤280 m depth). We hypothesize that this is because the permanent layer of small particles, partially linked to heterotrophic bacteria, consumes surface-derived small-particle POC more efficiently than large particles. In addition, it cannot be ruled out that large sinking particles fragmented less in this region.
With the current data, we cannot test the above hypothesis, but if it was proven true, it would have implications for the efficiency of the BCP. This is because the contribution of small particles to the overall POC fluxes is significant in oxygenated regions of the mesopelagic Atlantic at subtropical (41 ± 20%, 11-64 μm, Durkin et al., 2015) and high latitudes (86 ± 23 %, <100 μm, Giering et al., 2016). However, as indicated above, this contribution would be reduced within the hypoxic OMZs. We therefore recommend that more information about POC fluxes by small and large particles and the processes that drive their distribution throughout the water column of the hypoxic OMZs is needed (e.g., remineralization, aggregation/fragmentation rates of large particles). Ultimately, this additional information will allow us to understand how the potential expansion of the hypoxic OMZs will affect the efficiency of the oceans in sequestering atmospheric CO 2 .
Finally, a new metric, the depth-specific attenuation (ΔT) rate, was presented based on an extension of the method proposed by Buesseler and Boyd (2009). This metric can provide estimates of POC-flux attenuation rates that are independent of the thickness of the layer over which the fluxes are determined. Thus, it can allow one to compare attenuation rates in different layers of the water column. We exploited this new metric to compare flux attenuation rates above and within the ETNA OMZ.

Conclusions
We conclude that (1) the main factor that promoted net instantaneous POC fluxes via small particles toward the OMZ was the strong summer pulse of small particles that coincided with stronger stratification. This stronger stratification was triggered by temporal changes in precipitation (plus inferred related DFe deposition, Trichodesmium blooms, and their collapse) and wind stress; (2) similarly to anoxic OMZs, a permanent layer rich in small particles of presumably free-living anaerobic-heterotrophic bacteria was found in the upper part of the hypoxic ETNA OMZ; (3) this permanent layer did not appearto have a measurable net input of POC toward deeper water masses; and (4) inside the hypoxic region of the ETNA, the attenuation rates of net instantaneous small-particle fluxes were equivalent (p > 0.05) or significantly higher (p < 0.05) than those in well-oxygenated regions. These results differ from those

10.1029/2019GB006305
Global Biogeochemical Cycles obtained from absolute POC fluxes measured by drifting sediment traps and mostly due to large particles. These differences highlight the need to better understand the processes controlling the stocks and fluxes of small and large particles in hypoxic OMZs. Progress in this direction will help to better understand the role that hypoxic OMZs have in attenuating POC fluxes. Finally, our results confirm that BGC-Argo floats are powerful tools to fill knowledge gaps related to particle dynamics and their role in the oceanic BCP.

Data Availability Statement
Data from Biogeochemical-Argo floats are freely available at ftp.ifremer.fr/ifremer/argo. These data were collected and made freely available by the International Argo Program and the national programs that contribute to it (http://www.argo.ucsd.edu, The Argo Program is part of the Global Ocean Observing System. Data of wind stress and daily area-averaged precipitation rates applied in this study are also freely available at the websites of the Copernicus Marine Environment Monitoring Service (http://marine.copernicus.eu/services-portfolio/access-to-products/) and Giovanni-NASA (https://giovanni.gsfc.nasa.gov/giovanni/), respectively.