Organic Carbon Export and Loss Rates in the Red Sea

The export and fate of organic carbon in the mesopelagic zone are still poorly understood and quantified due to lack of observations. We exploited data from a biogeochemical‐Argo float that was deployed in the Red Sea to study how a warm and hypoxic environment can affect the fate of the organic carbon in the ocean's interior. We observed that only 10% of the particulate organic carbon (POC) exported survived at depth due to remineralization processes in the upper mesopelagic zone. We also found that POC exported was rapidly degraded in a first stage and slowly in a second one, which may be dependent on the palatability of the organic matter. We observed that apparent oxygen utilization (AOU)‐based loss rates (a proxy of the remineralization of total organic matter) were significantly higher than the POC‐based loss rates, likely because changes in AOU are mainly attributed to changes in dissolved organic carbon. Finally, we showed that POC‐ and AOU‐based loss rates could be expressed as a function of temperature and oxygen concentration. These findings advance our understanding of the biological carbon pump and mesopelagic ecosystem.


Introduction
The ocean carbon cycle is regulated by physicochemical and biogeochemical processes collectively called carbon pumps (Boyd et al., 2019). Among these, the biological carbon pump (BCP) exports organic matter in particulate and dissolved form to the deep ocean (Boyd et al., 2019). The export of particulate organic carbon (POC) is attributed to particles sinking to depth (Boyd et al., 2019;Buesseler et al., 2007Buesseler et al., , 2020. The BCP is driven by phytoplankton cells in the euphotic layer that uptake CO 2 to produce organic carbon, which is then transported downward to the mesopelagic zone (100 to 1,000 m) through a wide range of processes (i.e., biological gravitational pump, mixed layer pump, eddy subduction pump, and vertical migration pump) (Boyd et al., 2019;Buesseler et al., 2020). The BCP is mostly responsible for the annual export of surface POC that occurs during episodic events, such as seasonal phytoplankton blooms (Honjo et al., 2014;Lutz et al., 2007). The efficiency of the BCP is usually quantified based on two main metrics: (1) the attenuation of the export flux, which is typically estimated from the exponent of a "Martin's curve fit" to the flux profile (i.e., a power law fit, Martin et al., 1987), and (2) the transfer efficiency corresponding to the fraction of POC exported in the mesopelagic zone with respect to that at the base of the euphotic zone (Buesseler et al., 2020;Llort et al., 2018).
In general, most of the organic carbon exported into the mesopelagic zone is lost due to heterotrophic remineralization processes (Cavan et al., 2017;Giering et al., 2014). Many studies revealed that remineralization of sinking organic particles is affected by local environmental conditions by either affecting particle size (fragmentation of particles) or the rate of remineralization (Lopez-Urrutia et al., 2006;Marsay et al., 2015;Martinez-Garcia et al., 2014). Under lower oxygen concentration, a larger fraction of organic carbon is thought to be unyielding to remineralization (Robinson, 2019;Weber et al., 2016). Temperature is also known to influence remineralization rates, as the heterotrophic processes responsible of the organic carbon remineralization are temperature dependent (Marsay et al., 2015;Robinson, 2019).
However, most of these studies were not performed in situ which means that changes in oxygen or temperature could have been masking other drivers. Therefore, the roles of oxygen concentration and temperature on heterotrophic remineralization remain unclear (Laufkotter et al., 2016). A warm and/or oxygen-depleted ocean may affect the behavior of heterotrophs and thus the attenuation of organic carbon flux. Since the exported organic carbon is remineralized by heterotrophs back into inorganic carbon, oxygen and temperature changes may also impact the ocean's capacity to control the concentration of atmospheric CO 2 (Briggs et al., 2020;Kwon et al., 2009). Under a warm environment, most remineralization may occur in the upper layer of the mesopelagic zone (100-300 m), modifying how carbon and nutrients are fixed and recycled into the ocean's interior (Kwon et al., 2009). Thus, a better understanding of the processes controlling biological carbon fluxes is fundamental to predict the ocean's capacity in regulating climate change (Boyd et al., 2019;Briggs et al., 2020).
The attenuation of the flux of particles should be balanced by community processes responsible for the loss of organic carbon in the mesopelagic layer (Boyd et al., 2019;Giering et al., 2014). Several previous studies have shown that the carbon budget in the mesopelagic zone remains unbalanced (Baltar et al., 2009;Burd et al., 2010), with the notable exception of Giering et al. (2014) who were able to balance the attenuation of POC flux with the community metabolism (heterotrophic respiration). Recent studies identified additional mechanisms, such as fragmentation of particles by zooplankton, which may contribute to balancing the carbon budget in the mesopelagic zone (Briggs et al., 2020). Overall, we only have a limited knowledge of the role of the heterotrophic community in the turnover of organic matter and its responses to changes in environmental conditions (i.e., oxygen and temperature sensitivity), which compromises our predictions of carbon storage in the global ocean (Burd et al., 2010).
In the past decade, several studies demonstrated the potential of using autonomous platforms equipped with optical sensors to estimate export flux by providing particle measurements over sustained periods of time in diverse areas of the global ocean (e.g., Briggs et al., 2020;Dall'Olmo & Mork, 2014;Estapa et al., 2017Estapa et al., , 2019. For instance, using biogeochemical-Argo (BGC-Argo) floats, estimates of carbon sequestration were improved as these provide information to track particle accumulation rates in the water column (Briggs et al., 2020).
While the Red Sea is known as a warm environment and characterized by highly stratified waters in the upper layers, its northern part behaves as a temperate ecosystem with deepening of the mixed layer depth (MLD) during winter and highly stratified waters during summer (Gittings et al., 2018(Gittings et al., , 2019. The Red Sea is the northernmost tropical sea of the global ocean and displays a strong latitudinal gradient of decreasing temperature from south (26-32°C) to north (21-28°C) (Chaidez et al., 2017;Kheireddine et al., 2017Kheireddine et al., , 2018. The Red Sea has been recently identified as a fast-warming oceanic region of the global ocean, especially in its northern part (Chaidez et al., 2017). The Red Sea is also experiencing warming-related decreases in oxygen levels and hypoxic conditions in some areas of the basin (Breitburg et al., 2018;Stramma et al., 2010).
Such conditions may impact the entire Red Sea ecosystem (i.e., phytoplankton community structure, primary production, nutrient availability, temperature, and oxygen concentration), which in turn would affect the organic carbon export flux by influencing the remineralization rates of sinking particles and dissolved organic carbon (Cavan et al., 2017;Guidi et al., 2015;Marsay et al., 2015;Puigcorbe et al., 2017). Marsay et al. (2015) demonstrated a strong correlation between the Martin's exponent and the median upper 500 m water temperature for various regions of the global ocean, suggesting a strong temperature dependence of the rate of attenuation of the export flux of POC.
In this study, we examined the effect of temperature and oxygen on the remineralization of total organic carbon (TOC) and POC using measurements collected by a BGC-Argo float in the northern Red Sea. Our results suggest that remineralization processes mainly occurred within the upper layer of the mesopelagic zone and that both temperature and oxygen may control the heterotrophic respiration of the organic carbon entering the mesopelagic zone.

Deployment, Sampling Strategy, and Measurements
In September 2015, a biogeochemical profiling float was deployed in the northern Red Sea at 34.73°N and 27.66°E where the bathymetry is variable and shallow (<1,000-m depth) (Figure 1a). This float, identified Figure 1. Trajectory of the BGC-Argo float over the period of deployment (a). The location of the deployment is indicated by the red star. The thick white line indicates the portion of the trajectory during the period of this study (1 January to 31 December 2016). The colored lines represent the average modeled trajectories of particles that were tracked within the productive layer (z p ) (b) and isopycnal layers 28. 30-28.40 (c), 28.40-28.52 (d), 28.52-28.55 (e) Table 1 and its trajectory is displayed in Figure 1a. The maximum parking depth was set to 1,000 m. The float was programmed to surface at noon every 2, 5, or 10 days depending on the season. Data acquisition was nominally at 10-m resolution between 250 and 1,000 m of depth, 1-m resolution between 10 and 250 m, and increased to 0.20-m resolution between 10 m and the surface . Float data were downloaded from the global Argo database (ftp://ftp.ifremer.fr/ifremer/argo/).
Vertical profiles of chlorophyll-a (chla) fluorescence and the volume scattering function (measured at 124°a nd at 700 nm) were obtained using a Wet Labs ECO sensor. The photosynthetically available radiation (PAR, see acronym definitions in Table 2) was measured with a Satlantic OCR radiometer. Raw counts for each parameter were transformed into geophysical quantities by applying the manufacturer-provided scaling factor and dark count. Then, as recommended by the Argo data management procedures (Wong et al., 2015), each profile was quality controlled by applying methods that have been specifically developed for each variable Thierry, Bittig, Gilbert et al., 2018;Thierry, Bittig, & The Argo-Bgc Team, 2018, and references therein). Dissolved oxygen concentrations (O 2 ) were determined using an Aanderaa optode sensor (model 4330).
Following Schmechtig et al. (2015), vertical profiles of chla concentration (mg m −3 ) were adjusted for non-zero deep values and corrected for non-photochemical quenching according to Xing et al. (2018). The chla values were further divided by a factor of 2 to correct for the overestimation observed by standard  Wet Labs fluorometers as described in Roesler et al. (2017). The particulate backscattering coefficient at 700 nm (b bp (700) in m −1 ) was calculated as described in Schmechtig et al. (2016). First, the particulate volume scattering function was calculated by removing the contribution of pure seawater that depends on both temperature and salinity (Zhang et al., 2009). Then, the particulate volume scattering function was converted into b bp as recommended in Boss and Pegau (2001) and Sullivan et al. (2013). Spikes were removed from the chla and b bp measurements as described in Briggs et al. (2011). The PAR vertical profiles acquired during unstable sky and sea conditions as well as during sporadic atmospheric clouds were discarded following Organelli et al. (2016). Then, the remaining PAR profiles with non-zero dark measurements at depth and wave focusing in the upper part of the profiles were removed (Zaneveld et al., 2001). O 2 measurements were processed by applying a locally derived factor (~1.06) to each profile based on the comparison between the float's percent oxygen saturation (O 2 sat) values and those from the monthly World Ocean Atlas climatology (WOA09) ( Figure S1), as described by Takeshita et al. (2013). Finally, quality-controlled vertical profiles of chla, b bp , PAR, and O 2 were binned in 1-m intervals and used for the analyses in this study.
Apparent oxygen utilization (AOU) was computed as the difference between O 2 and the oxygen concentration at saturation (O 2 sat) following the method of Garcia and Gordon (1992). Then, positive changes in AOU were converted to changes in TOC using the molar ratio ΔC:ΔO 2 = 0.688 (Anderson & Sarmiento, 1994).
The MLD was calculated as the depth at which density exceeded the density at 10 m by 0.03 kg m −3 (de Boyer Montegut et al., 2004). The euphotic depth was calculated as the depth at which PAR was 1% of its surface value (Morel & Berthon, 1989) which corresponded to the depth below which less than 5% of chlorophyll-containing particles were detected. Finally, the bottom of the productive layer (z p ), defined as the region where particles can be produced by photosynthesis, was computed as the maximum between the euphotic depth and the MLD.

From b bp and AOU to Organic Carbon Stocks and Fluxes
Particle backscattering is sensitive to the presence of particles in the size range 0.2-20 μm in general (Stramski et al., 2004) and more specifically between 1 and 10 μm within the surface of oligotrophic waters . In this study, b bp was used to estimate the concentration of POC using an empirical relationship ([POC] = 40,233 b bp − 11) established for Red Sea waters based on in situ measurements collected during different cruises performed between 2014 and 2016. The slope of this relationship was within the range of those reported in previous studies (Cetinic et al., 2012;Rasse et al., 2017;Stramski et al., 2008). Details about the POC versus b bp (700) relationship are provided in the supporting information.
To compute stocks and net rates of POC accumulation within homogeneous water masses, density measurements were used to define different layers within the mesopelagic zone. To verify the homogeneity of these layers along the trajectory of the float, the spiciness metric was calculated according to Flament (2002). This variable allowed us to identify changes in water masses with distinct thermohaline properties but having similar density and thus to ensure spatial and temporal homogeneity within the chosen isopycnal layers. The float remained in nearly homogeneous water masses only during the year 2016 (see supporting information), and thus, only data from this year are further considered in this study. Because the float moved significantly over the period of the analysis (Figure 1a), Lagrangian particle tracking, based on previously validated model results, was used to further test the extent to which the float sampled homogeneous water masses during the analysis period. Essential information regarding the modeling study is described in the supporting information. Briefly, the general (1 km) circulation in the Red Sea was simulated based on a three-dimensional high-resolution MIT general circulation model (MITgcm) (Marshall et al., 1997) configured for the Red Sea (Yao et al., 2014a(Yao et al., , 2014b. The trajectories of numerous passive particles from a point-source region corresponding to the float location were then simulated and tracked using the connectivity modeling system (CMS) (Paris et al., 2013).
POC values were integrated over the productive layer and over different isopycnal depths (28.30 to 28.57 kg m −3 ) below the productive zone (Dall'Olmo & Mork, 2014): (1) from the surface (z 0 ) to isopycnal depth 28.57 kg m −3 (z 28.57 ) (z 0 − z 28.57 , iPOC 28:57 0 ) corresponding to the whole water column (excluding bottom effects), (2) within the productive layer (z 0 − z p , iPOC zp 0 ), (3) within the entire mesopelagic layer defined as the isopycnal layer 28.30-28.57 kg m −3 (z 28.30 − z 28.57 , iPOC 28:57 28:3 ), and (4) over gradually deeper isopycnal layers of the mesopelagic zone (iPOC 28:57 28:30 þ zi , with z i = 0.1, 0.22, 0.25, and 0.27 kg m −3 isopycnals). The different isopycnal layers were chosen based on their physical properties and the vertical resolution of the float in order to have enough data points to integrate POC values over gradually deeper layers of the mesopelagic zone. We paid particular attention to exclude sediment resuspended from the bottom from the analysis to avoid interpreting these resuspensions as POC fluxes.
The net POC flux (E) within the isopycnal layers 28.30 + z i was then estimated from the rate at which POC stocks in the different layers of the mesopelagic (iPOC 28:57 28:30 þ zi ) varied as a function of time, and E zp represents the net POC flux just below the productive layer (Dall'Olmo & Mork, 2014). As the float did not sample at high vertical and temporal resolution, it was not possible to extract information about sinking of large particles in the mesopelagic layer (Briggs et al., 2020). Thus, the potential export of large particles/aggregates was not considered in this study. Our estimates of POC fluxes are mainly attributed to small slowly sinking particles, given that de-spiked b bp measurements are not expected to be sensitive to large particles (Briggs et al., 2011).
The vertical variation of E was then used to estimate the so-called Martin's exponent, b, using the power law Martin et al., 1987). While this model remains debated (Guidi et al., 2015;Marsay et al., 2015), estimates of the b exponent allow us to compare the results from our study with other previously studied regions of the global ocean. The fraction of surface POC flux exported to the different layers of the mesopelagic zone, T zi (transfer efficiency), was computed as the ratio E zi =E zp (Buesseler & Boyd, 2009).
The seasonal net export rate in the mesopelagic layers (E zi ) was also estimated by fitting a linear leastsquared model from the time when iPOC 28:57 28:3 was minimal (late January) to the time when iPOC 28:57 28:3 reached its maximum value (late February). The fast maximum (instantaneous) net carbon export, E ziF , within the mesopelagic zone was determined from the slope of the straight line fitted by a linear least-squared model over the time period of fast POC export (late February, about 6 days).
Removal of POC occurred in two stages. The first stage was observed during a short period of time qualified as "fast removal" and the second was observed over a longer period of time qualified as "slow removal." Thus, carbon fast and slow net loss rates, L ziF and L ziS , respectively, were estimated as the time rate of change of POC stocks (linear fit), when POC stocks within a given layer of the water column were observed to decline for each stage. Fast losses occurred in early March (for about 6 days), while slow ones occurred from March to September (about 205 days). Fast and slow carbon-specific loss rates, L ziF * and L ziS *, respectively, were computed by dividing L ziF or L ziS by the average POC stocks during the period of removal. Finally, the loss rates in TOC (derived from AOU), L AOU zi , were estimated as the time rate of change of AOU (linear least-square fit) from early March (minimum of the average of AOU) to early June (maximum of the average of AOU) within a given layer of the water column.

Physical Characteristics Along the Float Trajectory
All particles released in the model at the location of the float on 1 January 2016 (34.27°N and 27.20°E) within the productive layer and the four gradually deeper isopycnal layers of the mesopelagic zone followed the western boundary currents along the African coast up to a latitude of 24°N. Their average trajectories within each layer are presented in Figures 1b-1f and agree reasonably well with the float's trajectory. They generally exhibited low dispersion in their pathway. The mean transport velocities and the related drift times were related to the initial release depth. Specifically, particles in the productive layer (145 ± 37 m) needed 53 ± 26 days to reach the position of the float at the end of the period considered in the analysis (Figure 1b), whereas this time increased to 113 ± 18 days at 180 ± 43 m depth ( Figure 1c) and to 334 ± 7 days for those released deeper (300-485 m) (Figure 1f). Even if the surface particles traveled faster than the float, this is not a major problem for our study because particles were injected into the mesopelagic from the surface layer only during a localized event (late February 2016). Furthermore, the rest of the dynamics we analyzed occurred in the mesopelagic zone, where the float followed water masses in a quasi-Lagrangian fashion.
Temperature-salinity diagrams produced from the Conductivity-Temperature-Depth (CTD) measurements collected along the float trajectory, as well as a compilation of historical stations (see supporting information) collected in the Red Sea, are presented in Figure S4. Different water masses were identified with distinct physical and chemical characteristics: the Red Sea surface water (RSSW), the Red Sea outflow water (RSOW), and the Red Sea deep water (RSDW; Sofianos & Johns, 2007). The RSSW mass corresponds to the surface waters and is characterized by relatively high variability in salinity. On the other hand, the RSOW and the RSDW are deeper water masses characterized by constant temperature and salinity.
Each of the four mesopelagic layers studied was characterized by temperatures and salinities that corresponded to the RSOW and RSDW masses ( Figures S4 and S5). During 2016, the four isopycnal layers sampled in mesopelagic zone exhibited fairly stable temperatures (22.15°C ± 0.10°C; 21.85°C ± 0.09°C; 21.68°C ± 0.05°C; and 21.63°C ± 0.04°C, respectively) and salinities (40.47 ± 0.05; 40.52 ± 0.03; 40.53 ± 0.01; and 40.53 ± 0.01, respectively) along the float trajectory ( Figure S5). In addition, the spiciness metric was used to reveal the underlying isopycnal mixing that could alter the variability of biogeochemical properties on the same isopycnal layers ( Figure S6). We found that each of the four mesopelagic layers was also characterized by low spiciness variability (Table 3), whereas high spiciness variability was observed in response to unstable vertical salinity and temperature gradient immediately near the surface layer ( Figure S6).

Temporal Evolution of chla and b bp Vertical Distribution
The seasonal dynamics of chla and b bp are displayed in Figure 2. In the upper productive layer, chla and b bp exhibited a similar seasonal pattern with highest values during winter and lower values during the rest of the year. These concomitant high chla and b bp values within the surface layer were associated with an increase  of phytoplankton biomass following the deepening of the MLD (Figure 2). In the summer stratified period, chla showed a pronounced deep chlorophyll maximum DCM (Figure 2a).

Export Fluxes and Transfer Efficiency
After the phytoplankton biomass increased in the productive layer, the POC stock in the mesopelagic zone (iPOC 28:57 28:3 ) increased to its maximum value (~333 mmol POC m −2 , Figure 3). About 7 days later, this POC stock decreased by about 166 mmol POC m −2 . Then, over the spring-summer seasons, a slow decline of the stock was observed that amounted to around 83 mmol POC m −2 .
The seasonal net carbon export flux below the productive layer, E zp , was 233 ± 46 mmol POC m −2 y −1 and was mainly associated with a fast export event (25.8 ± 5.2 mmol POC m −2 d −1 ) that occurred just after the summer stratification was established (Figure 3). Vertically resolved seasonal and fast values of carbon export below the productive layer, E zi and E ziF , respectively, decreased exponentially as a function of depth (Figures 4a  and 4b). Most of the POC was degraded in the upper part of the mesopelagic layer (about 72% between 72and 157-m depth), while less than 10% of the POC reached the bottom of the mesopelagic zone (Figure 4c).
The seasonal estimates of POC flux were fitted to the equation of Martin et al. (1987) to determine the exponent of the attenuation flux, b, which had a mean (±standard deviation) value of 1.53 ± 0.16.

Oxygen-Based Respiration Rates
The AOU inventory exhibited a ventilation-respiration cycle ( Figure 5). When the mixed layer reached its maximum depth in January-February, the AOU inventory was ventilated. During this mixing period, ephemeral restratifications were observed that caused an anomalous rise in the average AOU values, but these were excluded from the rest of the analysis. During the period of stratification, the AOU inventory increased by 7% to 55% throughout the winter spring until the end of spring (early June). Then, over the summer, the inventory in AOU remained relatively stable and increased again from October. The symbols *, **, ***, and **** refer to the same isopycnal layers presented in Figure 4. Cyan solid and dashed lines: linear regressions used to estimate seasonal and daily fast carbon export rate, E zi and E ziF , respectively. Gray dashed and dotted lines: linear regressions used to estimate daily fast and slow carbon loss rates, L ziF and L ziS , respectively.

Organic Carbon Losses
In early March, the POC stock just below the productive layer decreased approximately from 333 to 166 mmol POC m −2 in about 7 days (Figure 3), corresponding to a fast POC loss rate L zpF of 24 ± 6 mmol POC m −2 d −1 . POC stocks decreased further by around a factor of 2 over the spring-summer (~205 days), corresponding to a , respectively (see also Figure 2).

10.1029/2020GB006650
Global Biogeochemical Cycles slow loss rate L zpS of 0.41 ± 0.16 mmol POC m −2 d −1 (Figure 3). As observed for E zi and E ziF , L ziF and L ziS also decreased exponentially as a function of depth (Figures 6a and 6b and Table 4). Average AOU just below the productive layer increased from~44 to~66 μmol kg −1 until the end of the spring season (~92 days), corresponding to a loss of TOC at a rate, L AOU zp , of 1 ± 0.25 mol C m −2 y −1 (10.8 ± 2.7 mmol C m −2 d −1 ).
L AOU zi also decreased exponentially as a function of depth (Table 4).

Assumption of Spatial Homogeneity
The model-based Lagrangian particle tracking suggested that deep mesopelagic waters moved approximately at the same speed as the float, while surface waters moved faster than the float (Figure 1). This Figure 6. Fast and slow loss rates L ziF (a) and L ziS (b) over gradually deeper isopycnal layers of the mesopelagic zone. Error bars represent one standard deviation. The symbols *, **, ***, and **** refer to the isopycnal layers 28.30-28.40, 28.40-28.52, 28.52-28.55, and 28.55-28.57 kg m −3 , respectively (see also Figure 4). suggests that the float followed in a quasi-Lagrangian manner the same mesopelagic water masses during the analysis period. This conclusion is further supported by the spatiotemporally coherent physical characteristics (temperature, salinity, and spiciness) that we observed within each layer of the mesopelagic zone. Overall, these results support our use of the float data as a time series.

Small Particle Dynamics
Large post-bloom sinking organic aggregates were likely fragmented by biological and/or physical mechanisms (Burd & Jackson, 2009) and generated the small particles we observed in this study. These small slowly sinking particles are expected to be mostly remineralized by heterotrophic bacteria in the mesopelagic zone (Goutx et al., 2007;Herndl & Reinthaler, 2013), because only low concentrations of zooplankton can be found in the Red Sea mesopelagic waters (Dypvik & Kaartvedt, 2013). In addition, bacterial activities, including remineralization rates, are enhanced by warm environments as the Red Sea (Lopez-Urrutia et al., 2006;Sarmento et al., 2010).

POC Fluxes and Transfer Efficiency
We observed that about only 10% of the material exported reached depths greater than 220 m (Figure 4c). This suggests that most of the particulate matter was relatively labile and was remineralized in the upper mesopelagic zone, which is consistent with the relatively high value of Martin's b exponent we estimated. Marsay et al. (2015) found a b value of 1.59 in the North Atlantic subtropical gyre and hypothesized that a strong attenuation of the flux may occur in the upper mesopelagic in warm waters due to strong recycling of organic material by heterotrophic respiration. In our study, the high value of 1.53 ± 0.16 suggests that strong remineralization processes occurred in the upper layer of the mesopelagic zone, possibly as a response to the high temperature of the Red Sea.

Organic Carbon Losses
To the best of our knowledge, this study presents the first seasonal estimates of fast and slow net POC loss rates from small slowly sinking particles in the mesopelagic zone. Losses of POC through the mesopelagic zone are mainly attributed to remineralization by bacteria (Herndl & Reinthaler, 2013;Robinson et al., 2010). However, the temporal and spatial variability of bacterial remineralization rates in the mesopelagic zone still needs to be quantified accurately and leads to uncertainties in the mesopelagic carbon budget (Giering et al., 2014). The contribution of small slowly sinking particles could be a potential supplementary carbon source capable of balancing carbon budgets in the mesopelagic zone (Boyd et al., 2019;Giering et al., 2014). Thus, our estimates of net POC loss rates from small particles could be used to further constrain the mesopelagic carbon budget as well as to parameterize biogeochemical models.
Our oxygen-based remineralization rates are significantly higher (25-fold) than our POC-based loss rates. These differences likely arise because changes in AOU integrate processes and carbon stocks that we cannot detect with b bp . Also, they could be due to differences observed in the timing between POC and AOU losses which may be due to variability and sensitivity of the carbon pool (type and stoichiometry). Among these additional carbon stocks detected by AOU, but not b bp , the most important is the pool of dissolved organic carbon that is known to support most (50-90%) of the remineralization (Abell et al., 2000;Hansell & Carlson, 2001). In agreement with these previous studies, we found that slow-sinking POC could account for less than 10% of the overall organic matter remineralization recorded by losses in oxygen. Recently, Torfstein et al. (2020) observed high POC flux values (1.7 to 4.6 mol C m −2 y −1 ) within the mesopelagic layer in the Gulf of Aqaba in the Red Sea which suggest that a large fraction of POC flux attributed to large fast-sinking particles (>10 μm) might not be detected here. The remaining part of these losses should thus be assigned to the consumption of DOC and larger particles (>10 μm).

Drivers of Organic Carbon Loss
Our estimated POC-and AOU-based loss rates decreased with depth, which could be due to multiple factors that also decrease as a function of depth. These depth-dependent factors include (1) the concentration of POC and DOC that are the substrates for remineralization; (2) the lability of carbon (e.g., their composition, Roder et al., 2013); and (3) temperature and oxygen concentration (Robinson, 2019, and references therein) that affect microbial rates. To attempt to explain the variability of our loss rates with depth (~65% of variation), we computed POC-specific loss rates (rates per unit POC stock). Unfortunately, we do not have in situ measurements of TOC, so we could not normalize the AOU-based loss rates. The POC-specific loss rates still varied over a significant range (~40% of variation), suggesting that the POC variability with depth could not explain all the observed variations in the organic carbon loss rates. In other words, the variability in loss rates with depths could not be explained entirely by a decrease in the available substrates.
We then investigated the variability of POC-specific and AOU-based loss rates as function of temperature (Brewer & Peltzer, 2017;Figures 7 and 8). POC-specific loss rates and AOU-based loss rates were significantly correlated with temperature, despite the relatively narrow range of temperatures observed (   Using the Arrhenius law (Arrhenius, 1889), we estimated the activation energy (E a ) from the slope of a plot of the natural logarithm of the change in the AOU or POC averaged at each time step within a given isopycnal layer of the mesopelagic zone versus the inverse absolute temperature in Kelvin (1/T) (Figure 9, Brewer & Peltzer, 2017). The activation energy is expected to be inversely related to the lability of the substrate of a chemical reaction (in this case, the remineralization of organic matter): the more a compound is recalcitrant, the higher is E a . Furthermore, E a can vary as a function of the organisms degrading the organic matter (Davidson et al., 2000;Davidson & Janssens, 2006). No significant differences between E a reported for POC and TOC loss rates were found (Figure 9). However, we found higher E a values (>1,760 kJ mol −1 ) compared to those reported for other regions of the global ocean (60.8-758.1 kJ mol −1 ) as well as the Red Sea (392.5 kJ mol −1 ), even though our carbon loss rates are within the same range than those reported for the Red Sea by Brewer and Peltzer (2017) (Figure S7).
Our estimates of E a are, however, similar to those reported for the eastern part of the Mediterranean Sea (1,663 kJ mol −1 , Brewer & Peltzer, 2017). These authors suggested that their Mediterranean site had experienced a full column overturn and thus suspected that their data set was not in a steady-state condition due to a deep mixing event, which could explain the high E a value obtained. Our analyses of the stability of the water column ( Figure S6 and Table 3) indicate that this was not the case for our study. Thus, our relatively high E a values are consistent with slow remineralization of relatively recalcitrant organic matter and/or might be an artifact of the very narrow range in temperature observed in our study. Alternative explanations for the differences in E a observed between our study and that of Brewer and Peltzer (2017) could be that our loss rates were (1) estimated using different techniques and (2) estimated within the northern Red Sea compared to the whole basin in Brewer and Peltzer (2017). Indeed, the data used for the Red Sea in Brewer and Peltzer (2017) are based on AOU measurements, which were collected during spring and fall seasons along the central axis from south to north. Further investigations are needed to explain these differences in activation energies.
Although it is tempting to interpret the strong correlations between loss rates and temperature presented in Figures 7 and 8 as causal links, the real driving factors of the variability in L zi * and L AOU zi could covary with and thus be confounded by oxygen. Indeed, POC-specific loss rates were also strongly correlated with oxygen, which varied over a wider range than temperature ([L ziF * = 0.008O 2 − 0.06], R 2 = 0.91, Figure 7b; [L ziS * = 0.004O 2 + 0.05], R 2 = 0.98, Figure 7d; N = 4, p < 0.01). We also found a robust correlation between AOU-based loss rates and oxygen ([L AOU zi = 0.094O 2 − 1.02], R 2 = 0.92, Figure 8b; N = 4, p < 0.01). While oxygen was strongly correlated with temperature ( Figure S8), given the greater dynamic range of oxygen with respect to temperature, we have more confidence in relating oxygen than temperature with organic carbon losses rates as a function of depth. This is consistent with previous studies that observed remineralization processes taking place mainly in the upper layer of the mesopelagic zone and slower remineralization rates under lower oxygen concentration (Laufkotter et al., 2017;Le Moigne et al., 2017;Robinson, 2019). Based on these observations, we surmise that temperature and oxygen might be key parameters controlling the respiration of the organic carbon pool entering the mesopelagic zone by heterotrophic organisms. Hence, the sensitivity of heterotrophic organisms to variations in ocean temperature and oxygen concentration may have implications for controlling concentration of atmospheric CO 2 and thus regulating the climate.
Although oxygen was strongly related to carbon loss rates, the stoichiometry of organic matter could also explain the variability observed in loss rates. The C:N ratio of organic matter increases with depth in the Red Sea (Calleja et al., 2019;Roder et al., 2013), indicating that organic matter becomes more recalcitrant at depth (Paulmier et al., 2009;Thomas, 2002). As consequence, we would expect lower remineralization rates deeper in the water column. Unfortunately, we do not have coincident particulate or dissolved organic nitrogen measurements to test this hypothesis.

Why Are There Fast and Slow Net POC Loss Rates From Small Slowly Sinking Particles?
Small slowly sinking particles (<10 μm) can be degraded rapidly by mesozooplankton and bacteria, or degraded slowly (weeks to months) by bacteria (Goutx et al., 2007;Sempere et al., 2000). Our results showed that more than 85% of the POC exported to the mesopelagic was degraded within a few days, while the remaining portion of POC (less than 15%) was degraded over the course of weeks to months. We also showed that the fast loss of POC occurred at temperatures similar to those measured during the slow decline. Thus, we can infer that the observed differences in fast and slow loss rates are not due to temperature, but most likely to variations in the organisms consuming the carbon (e.g., zooplankton and bacteria) within the mesopelagic zone (Calleja et al., 2018) and/or to changes in the lability of the POC. In other words, we may have observed slow loss rates of POC over the spring-summer seasons, because after the recorded fast losses, the remaining small particles might have become less nutritious and thus less palatable for bacteria. Moreover, the bacteria and zooplankton communities consuming particles during the periods of fast and slow loss rates might have been different.

Mesopelagic Carbon Budget
Closing the mesopelagic carbon budget has been challenging due to the complexity of the mesopelagic food web and the numerous processes involved in the export and remineralization of organic carbon. Numerous studies have found a strong imbalance in the mesopelagic zone by up to an order of magnitude between heterotrophic carbon demand and the flux of sinking carbon (Anderson & Ryabchenko, 2009;Burd et al., 2010;Steinberg et al., 2008). We found that the POC exported via small particles was balanced by its consumption (sum of the fast and slow small particle POC losses) over the seasonal cycle ( Figure S9). Our results regarding export and losses of POC are consistent with the findings of Giering et al. (2014) for the Northeast Atlantic Ocean, where they suggest that the main role of zooplankton is to fragment aggregates into smaller particles that are easily degradable by prokaryotes. Recently, Briggs et al. (2020) mentioned that fragmentation of particles into smaller slowly sinking particles by zooplankton could play an important role in supplying organic carbon to balance the carbon demand by bacteria in the mesopelagic zone. Additional studies are needed to look at the role of zooplankton in mesopelagic waters and to determine particle size and composition in the Red Sea.

Red Sea Waters Versus Other Oceanic Regions
Our mean transfer efficiency at 100-m depth below z p (T 100 ) and remineralization exponent b were significantly different from values reported in the literature for different oceanic regions (Figures 10a and 10b). T 100 was lower than other values reported in some of the oceanic regions of the global ocean ( Figure 10b). Consistently, b was higher than those reported in different environments, even in other tropical and oligotrophic ecosystems (Figure 10a). The POC flux just below the productive layer (E zp ) was lower than those reported in other areas of the global ocean except for the eastern part of the Mediterranean Sea (Figure 10c). Recent studies showed that most of the biomass of zooplankton in the Red Sea is mainly found in the upper layer of the water column (0-100 m) due to the presence of low oxygenated waters below 100 m (Karati et al., 2019). The scenario of low abundance of mesozooplankton in Red Sea waters may explain these differences. We can assume that most of the particles within the productive layer are fragmented into smaller size by zooplankton and are then degraded by bacteria in the mesopelagic zone. Regarding the TOC remineralization rates, our values agree well with those reported for some oceanic regions of the global ocean ( Figure 10d). However, no clear information regarding the temporal and spatial variability of the ΔC:ΔO 2 Redfield ratios in the Red Sea is known. Uncertainties should be considered when using conversion factors to estimate TOC based on the Redfield ratio. It is known that temporal and regional differences in ΔC:ΔO 2 Redfield ratio may appear depending on which biogeochemical processes predominate (Burd et al., 2010;Reinthaler et al., 2006). The common value 0.688 has been used to convert AOU into TOC here, which might not be appropriate of the Red sea and could result in higher or lower estimates of TOC consumption. Thus, while our estimates are within the range of those reported in other areas of the global ocean, they should be considered with caution. Future work should investigate quantitatively how these uncertainties affect our understanding of the carbon fluxes in the mesopelagic layer.
While it is known that POC remineralization varies with time and space in the oceans (Buesseler & Boyd, 2009), little is known about this variability and most global biogeochemical models use a constant b = 0.86 and a fixed remineralization rate (in depth and time) (0.025 d −1 ) to estimate ocean carbon sequestration from sinking POC (DeVries & Weber, 2017;Dutkiewicz et al., 2005;Segschneider & Bendtsen, 2013). A constant b value and a fixed remineralization rate are clearly not representative of natural conditions and thus likely lead to uncertainties in predictions of global carbon sequestration. Our observed TOC remineralization rates (loss rates) converge with previous estimates using different approaches reported in the literature (Figure 10d). However, many of those depend on conversion parameters which may vary in time and space as well inducing high variability. Future work should investigate quantitatively how these uncertainties affect our understanding of the ocean carbon cycle.

Conclusion
In this study, we found that POC exported in the mesopelagic zone was rapidly degraded in a first stage and slowly in a second stage likely by heterotrophic organisms. We hypothesized that these differences in POC loss rates may depend on the palatability of the organic matter (from labile and nutritive at the start vs. recalcitrant and less palatable later). Regarding the TOC degradation (AOU-based), we found higher loss rates compared to POC loss rates, which are likely due to the consumption of dissolved organic carbon tracked by AOU and perhaps on the differences observed in the timing between POC and TOC loss.
We then observed that organic loss rates could be potentially expressed as a function of temperature and/or oxygen and thus could be used to predict their changes in response to climate change. We also showed a significant correlation between b values and the median upper 500 m water temperature and AOU with greater b values in warm waters where we observed high respiration processes by heterotrophs. Such regional variability in b showed that a uniform b value in the open ocean might not be necessarily appropriate to estimate or predict the feedback of ocean carbon remineralization on atmospheric carbon in the future as suggested by Marsay et al. (2015) and Weber et al. (2016). As we observed a rapid remineralization of the organic matter within the upper layer of the mesopelagic zone, further studies of the processes controlling the size and nature of the particles between 0 and 500-m depth are necessary to study the size and the nature of the particles in such environment. Currently, BGC-Argo floats do not provide this information. The next configuration of the BGC-Argo floats will enable a better understanding of the nature of the particles and the role of the heterotrophic organisms (zooplankton and bacteria) with a particular focus on the upper mesopelagic layer (Boyd et al., 2019). We also found that the sum of fast and slow POC losses led to a balanced budget for the small slowly sinking particles in the mesopelagic zone. The contribution of these particles could be a supplementary carbon source capable of balancing carbon budgets in the mesopelagic zone. Heterotrophic communities play an important role in regulating the efficiency of the BCP through controlling POC export by grazing, fragmenting, and degrading sinking particles in the global ocean. An outstanding scientific challenge is to understand the role of the heterotrophic communities in the mesopelagic zone. This will be possible with the development of new sensors on autonomous platforms (Boyd et al., 2019).

Data Availability Statement
The data presented here are freely available by the International Argo Program and the national programs that contribute to it (http://argo.jcommops.org).