Modulation of the vertical particle transfer efﬁciency in the oxygen minimum zone off Peru

. The fate of the organic matter (OM) produced by marine life controls the major biogeochemical cycles of the Earth’s system. The OM produced through photosynthe-sis is either preserved, exported towards sediments or degraded through remineralisation in the water column. The productive eastern boundary upwelling systems (EBUSs) associated with oxygen minimum zones (OMZs) would be expected to foster OM preservation due to low O 2 conditions. But their intense and diverse microbial activity should enhance OM degradation. To investigate this contradiction, sediment traps were deployed near the oxycline and in the OMZ core on an instrumented moored line off Peru. Data provided high-temporal-resolution O 2 series charac-terising two seasonal steady states at the upper trap: suboxic ( [ O 2 ] < 25 µmol kg − 1 ) and hypoxic–oxic (15 < [ O 2 ] < 160 µmol kg − 1 ) in austral summer and winter–spring, respectively. The OMZ vertical transfer efﬁciency of particulate organic carbon (POC) between traps ( T eff ) can be classi-Published by Copernicus Publications on behalf of the European Geosciences Union.

fied into three main ranges (high, intermediate, low). These different T eff ranges suggest that both predominant preservation (high T eff > 50 %) and remineralisation (intermediate T eff 20 < 50 % or low T eff < 6 %) configurations can occur. An efficient OMZ vertical transfer (T eff > 50 %) has been reported in summer and winter associated with extreme limitation in O 2 concentrations or OM quantity for OM degradation. However, higher levels of O 2 or OM, or less refractory OM, at the oxycline, even in a co-limitation context, can decrease the OMZ transfer efficiency to below 50 %. This is especially true in summer during intraseasonal wind-driven oxygenation events. In late winter and early spring, high oxygenation conditions together with high fluxes of sinking particles trigger a shutdown of the OMZ transfer (T eff < 6 %). Transfer efficiency of chemical elements composing the majority of the flux (nitrogen, phosphorus, silica, calcium carbonate) follows the same trend as for carbon, with the lowest transfer level being in late winter and early spring. Regarding particulate isotopes, vertical transfer of δ 15 N suggests a complex pattern of 15 N impoverishment or enrichment according to T eff modulation. This sensitivity of OM to O 2 fluctuations and particle concentration calls for further investigation into OM and O 2 -driven remineralisation processes. This should include consideration of the intermittent behaviour of OMZ towards OM demonstrated in past studies and climate projections.

Introduction
Eastern boundary upwelling systems (EBUSs) are generally known to be highly productive (Chavez and Messié, 2009), associated with significant primary production (479 to 1213 gC m −2 yr −1 ) and elevated concentrations of chlorophyll a (1.5 to 4.3 mg m −3 ). The high production is caused by prevailing equatorward alongshore coastal winds triggering the dynamic upwelling of cold nutrient-rich waters from the subsurface to the well-lit surface layer. The associated intense biological surface activity produces a large amount of organic matter (OM). Part of the OM will sink and be degraded by catabolic processes. Therefore, subsurface OM degradation contributes to the consumption of oxygen (O 2 ). In conjunction with poor ventilation of the water mass, O 2 consumption leads to the formation of oxygen minimum zones (OMZs), characterised in the global ocean by a suboxic layer between 100 and 1000 m in depth (Paulmier and Ruiz-Pino, 2009). OM degradation associated with O 2 consumption via respiration or remineralisation may influence biological productivity (fixed nitrogen loss). OM degradation may also influence climate on both short and long timescales (Buesseler et al., 2007;Law et al., 2012;Moffitt et al., 2015;Chi Fru et al., 2016) via modulation of the airsea exchange of climatically important gases (e.g. CO 2 , N 2 O and CH 4 ). Moreover, these impacts on climate and ecosys-tems may be significant when remineralisation stimulated by high surface productivity takes place in waters that feed the upwelling close to the ocean-atmosphere interface (Helmke et al., 2005;Paulmier et al., 2008). Although poorly documented, the OM fate in OMZs stands out as a major issue, due to O 2 deficiency and its effect on remineralisation processes. Progress will depend on two different hypothesised mechanisms. On the one hand, weak oxygenation appears to decrease OM degradation because anaerobic remineralisation is considered to be of an order of magnitude less efficient than aerobic remineralisation (Sun et al., 2002). This low remineralisation efficiency suggests a tendency toward OM preservation and enhanced sediment export. On the other hand, the intense and diverse microbial activity (Devol, 1978;Lipschultz et al., 1990;Azam et al., 1994;Ramaiah et al., 1996;Lam et al., 2009;Stewart et al., 2012;Roullier et al., 2014) may induce efficient remineralisation and/or respiration. This may particularly be the case in the more oxygenated, warmer upper OMZ layer associated with the oxycline, leading to substantial OM recycling. Remineralisation, involving a relatively variable stoichiometry in the OMZ , depends on several factors. In addition to its quantity, OM recycling relies on quality (e.g. lability) and its sinking time through the OMZ layer. The depth of euphotic zone with OM production compared to the depth of oxycline that defines O 2 availability is of particular importance, together with particle size and shape (Paulmier et al., 2006;Stemmann et al., 2004). The conditions that control particle export and remineralisation also affect oxygen distribution and biogeochemical cycles. A better understanding of the processes that constrain particle export should help to improve estimations of OMZ development and maintenance (Cabré et al., 2015;Oschlies et al., 2017). It is also important to explore the detailed O 2 feedback effect on particles.
The EBUS off Peru is one of the most productive systems, accounting for 10 % of the world's fisheries (Pennington et al., 2006;Chavez et al., 2008), with the shallowest oxycline and one of the most intense OMZs (Fig. 1a-b;. Thus, it provides perfect conditions for investigating the relative importance of the aforementioned mechanisms. In order to examine the particle fluxes and their variability, this study focusses on the analysis of a time series compiled from moored sediment traps deployed in the Peruvian OMZ (Fig. 1c). This dataset is part of the AMOP ("Activities of research dedicated to the Minimum of Oxygen in the eastern Pacific"; see Sect. 2) project.

Methods
A fixed mooring line was deployed in January 2013 by R/V Meteor ∼ 50 km off Lima at 12 • 02 S; 77 • 40 W (Fig. 1). It was recovered in February 2014 by R/V L'Atalante within the framework of the AMOP project ("Activities of research dedicated to the Minimum of Oxygen in the eastern Pacific"; http://www.legos.obs-mip. fr/recherches/projets-en-cours/amop, last access: 16 August 2018). Sediment traps (PPS3 from Technicap) were deployed along the line in the oxycline-upper OMZ core (34 m) and in the lower OMZ core (149 m) in order to study particle flux through the water column (Figs. 1c, 2 and S1 in the Supplement; Tables 1 and S1 in the Supplement). The line was also equipped with five sensors measuring pressure, temperature, salinity and oxygen (SMP 37-SBE63), one sensor for fluorescence (ECO FLSB) and four additional temperature sensors (SBE56; Fig. 1c). The oxygen sensors have a resolution (smallest change detection) of 0.2 µmol kg −1 and an initial accuracy and detection limit of 3 µmol kg −1 (Fig. 3, Table 2). The resolutions and initial accuracies for the pressure, temperature and salinity sensors (0.2-0.7 and 0.1-0.35 dbar; 0.002 and 0.0001 • C; 0.0003 and 0.00001; respectively) induce an estimated resolution and accuracy for density ( Fig. 4a-b) of 0.01 kg m −3 for both according to the standard TEOS-10 equation. Each sediment trap was equipped with an inclinometer, allowing any incline to be recorded, which is fundamental for data interpretation. Also, to avoid OM decay (e.g. grazing) before analysis, the OM was collected in a poisoned solution of sea water with 5 % of formaldehyde. The traps sampled particles simultaneously over a period of 7 days, during the 3 months of austral summer (AMOP summer period: 6 January to 31 March 2013). The mooring was serviced in June 2013 and then re-deployed on 26 June 2013; the collection of material in traps re-sumed on 28 June. The sampling interval was extended to 11 days to fit the planned recovery date and to cover a wider period including two seasons (austral winter-spring during AMOP winter-spring ). The traps were full on 6 November 2013 but the mooring could not be recovered until February 2014 by R/V L'Atalante. Note that the SMP 37-SBE63 sensors started recording on 5 January at 34 m, on 7 January at 76 m, on 8 January at both 147 and 160 m (Fig. 3), and on 27 June 2013 at 50 m only (due to a technical breakdown).
Before analysing particle samples, we removed the swimmers, which could have actively entered the trap and thus would not represent the strict vertical sinking mass flux. After freeze-drying, the mass flux (dry weight; Fig. S1, Table S1) was determined with an accuracy of ±3 %. Total carbon (Ctot), particulate organic and isotopic carbon (POC, δ 13 C), and nitrogen (PON, δ 15 N) were analysed via an isotope ratio mass spectrometer (IRMS) IsoPrime100 paired with an elementary analyser (EA) Elementar vario PYRO cube. The carbon and nitrogen content (Figs. 2 and S2; Tables 3a, S2a-b and S3) was measured with an accuracy of ±0.2 %, and the isotopic δ 13 C and δ 15 N measurements (Tables 3c and S4) with an accuracy of ±0.006 ‰, and 0.007 ‰, respectively. Phosphorus and silica (Tables 3a, S2c-d and S3) were measured by colorimetry, using a spectrophotometer SPECORD 250 plus. Particulate organic phosphorus (POP) was analysed using the standard method (Strickland and Parsons, 1972) with an accuracy of ±3 %. The biogenic silica (BSi) was extracted with an alkaline dissolution at 95 • C using a kinetic method (DeMaster, 1981) with an accuracy Table 1. POC flux, transfer efficiency T eff and b for each main T eff range. T eff is determined from the %Flux 149 m / Flux 34 m ratio and is given as a percentage (Eq. 1). b is the coefficient from the Martin's curve theory (Suess, 1980;Martin et al., 1987). Italic and non-italic values correspond to the fluxes at 34 and 149 m, respectively. In the last lines of the table, POC fluxes, T eff and b are averaged for low, intermediate and high T eff , respectively, with the relative standard deviation among samples (±SD %). Analysis accuracy on the POC fluxes is ±0.2 %, inducing an absolute uncertainty on its vertical transfer efficiency estimated from a logarithmic expansion of ±0.2 % (see Sect. 2 Table S5) was determined from inductively coupled plasma-optical emission spectroscopy (ICP-OES) analysis with an accuracy of ±3 %. Systematic replicates for all sediment trap parameters have been analysed to estimate reproducibility, which mainly represents the heterogeneity of the sample. The reproducibility estimated from the standard deviation (SD) of the replicates for the total mass flux determination (0.12 %) is generally lower than the accuracy, except for δ 13 C. Daily satellite ASCAT wind measurements (Fig. 4cd), produced by remote-sensing systems (http://www.remss. com, last access: 16 August 2018; sponsored by the NASA Ocean Vector Winds Science Team) were used with an accuracy of 2 m s −1 . The satellite wind data are consistent with the available in situ wind measurements taken from R/V Meteor during the initial mooring deployment. Wind direction corresponds to alongshore winds favourable to upwelling Ekman transport. The mixed-layer depth (MLD; Fig. 4c-d) was estimated from a difference of temperature of 0.5 • C following De Boyer Montegut et al. (2004), in phase with the 0.2 and 0.8 • C criteria.
In situ pH sw (Fig. S3) and calcite saturation state ( calcite ) were calculated with the CO2SYS program (Lewis and Wallace, 1998), using discrete dissolved inorganic car- bon measured using a LI-COR 7000 and potentiometric pH sw measurements (25 • C). The dissociation constants proposed by Lueker et al. (2000) were used for the calculation with an estimated precision of ±0.04 units for in situ pH sw and ±0.2 units for calcite . Certified reference material (CRM) from Dr. Andrew Dickson's laboratory at Scripps Institution of Oceanography (University of California, San Diego) was used for assessing the precision and accuracy of measurements. The reference material gave a rela-tive difference averaging 2.2 ± 1.1 µmol kg −1 , with a peak of 4 µmol kg −1 (0.2 % error) with respect to the certified value. The analysis accuracy for the sediment trap samples is indicated in Fig. 2. The analysis accuracy has been estimated by propagation of the accuracy of each parameter from a logarithmic expansion for the molar ratios (C : P, N : P, N : P, Si : C, Si : N and Si : P) and all the calculated vertical transfer efficiencies as T eff for POC fluxes (see Eq. 1). SD among Table 3. Organic elemental fluxes and the corresponding molar ratios as well as inorganic and isotopic fluxes, and their transfer efficiencies for each main T eff range. (a) Organic elementary fluxes in mg m −2 d −1 and their transfer efficiency as a percentage (T eff , T effPON , T effPOP , T effBSi ; see Table 1 caption for calculation) in terms of particulate organic carbon (POC), nitrogen (PON), phosphorus (POP) and biogenic silica (BSi) for each three main T eff ranges (low in red, intermediate in yellow, high in blue) with the temporal standard deviation among samples (±SD %). Italic and non-italic values correspond to the fluxes at 34 and 149 m, respectively. Analysis accuracies on the elementary fluxes are ±0.2 % for both POC and PON, ±3 % for POP, and ±5 % for BSi, inducing an absolute uncertainty of ±0.2 % (T eff ), ±0.2 % (T effPON ), ±3 % (T effPOP ) and ±5 % (T effBSi ) on the transfer efficiency (see Sect. 2). (b) Values of elementary ratios (C : N, C : P, N : P, Si : C, Si : N, Si : P) and transfer efficiency of these ratios as a percentage (T effC : N , T effC : P , T effN : P , T effSi : N , T effSi : N , T effSi : P ; see Table 1 caption for calculation) for each three main T eff ranges with the temporal standard deviation between samples (±SD %). Italic and non-italic values correspond to the fluxes at 34 and 149 m, respectively. Analysis accuracies on the elementary ratios are ±0.4 %, 3.2 %, and 3.2 % and ±5.2 %, 5.2 %, and 8 % for C : N, C : P, and N : P and Si : C, Si / N, and Si : P, respectively, inducing an absolute uncertainty of ±0.9 % (T effC : N ), ±6.3 % (T effC : P ), and ±6 % (T effN : P ) and ±12 % (T effSi : C ), ±13.4 % (T effSi : N ), and ±19 % (T effSi : P ) on the transfer efficiency (see Sect. 2). Classical (reference) molar ratios have been reported on the second lines from Redfield et al. (1963) and Brzezinski (1985).
(c) Fluxes of inorganic calcium carbonate (CaCO 3 ) in mgCa m −2 d −1 and of carbon isotopic ratio (δ 13 C) and nitrogen isotopic ratio (δ 15 N) in ‰ and their transfer efficiency in % (T effCaCO 3 , T eff 13 C , T eff 15 N ; see Table 1 caption for calculation) for each three main T eff ranges (low in red, intermediate in yellow, high in blue) with the temporal standard deviation among samples (±SD %). Italic and non-italic values correspond to the fluxes at 34 and 149 m, respectively. Analysis accuracies on the elementary fluxes are ±3 % for CaCO 3 , ±0.006 ‰ for δ 13 C and ±0.007 ‰ for δ 15 N, inducing an absolute uncertainty of ±3 % (T effCaCO 3 ), ±0.06 % (T eff 13 C ) and ±0.26 % (T eff 15 N ) on the transfer efficiency (see Sect. 2).     samples representing the variability over the total dataset (AMOP summer + AMOP winter-spring ) or a given subset of data (e.g. corresponding to high, intermediate or low T eff ranges) has also been indicated in Tables 1, 2, 3, S1, S2, S3, S4 and S5). The different present relative SD values are higher than the total uncertainties (TU = accuracy + reproducibility) for all considered parameters.
Data are available at different time resolutions: 15 min (O 2 , density from temperature and salinity), 30 min (fluorescence), 1 day (satellite ASCAT wind), 7 or 11 days for AMOP summer and AMOP winter-spring datasets, respectively. The sediment trap fluxes include the percentage of Polychaetes relative to all other collected swimmers, denoted as %Poly. All fluxes (for the total mass of particles, POC, PON, POP, BSi, CaCO 3 , δ 13 C and δ 15 N, as well as %Poly), corresponding to a collection period of 7 and 11 days for the AMOP summer and AMOP winter-spring periods, respectively, have been normalised and expressed per day. Hereafter, we will use 7 days as the nominal weekly period. Different averages have been performed to compare with different temporal resolution of other data: from 15 min resolution for O 2 , density, and MLD and from 30 min resolution for the fluorescence to 1-day resolution (Fig. 4), and from 15 min to 7-day (for AMOP summer ) or 11-day (for AMOP winter-spring ) resolution for O 2 (Table 2). We verified that different ways to time-average did not modify the main findings of this study. Note however that daily-average MLD (Fig. 4c-d) presents a magnitude ∼ 9 times smaller than with the 15 minfrequency MLD. This is mainly due to biases induced by the vertical resolution according to the mooring sensor depths. For oxygen, the 1, 7 and 11 day averages have been denoted [O 2 [O 2 ] to be at the same temporal resolution (Table 2). This ratio corresponds to an availability index in terms of POC flux to be degraded according to the oxygen concentration availability. When  (Tables 1-3 and S1-S5).
3 Results and discussion 3.1 Particle transfer efficiency through the OMZ

Temporal modulation of particles and POC fluxes and their transfer efficiencies
The transfer of particles through the Peruvian OMZ is studied using data collected at the two fixed sediment traps, one located in the oxycline-upper core (34 m) and the second in the lower core (149 m). Seasonally, at 34 m, the mass fluxes during summer are about 60 % lower than during winter-spring (AMOP winter-spring : 986 mg m −2 d −1 on average; see Fig. S1, Table S1). At 149 m, the mass fluxes during summer are about 80 % higher than during winter-spring (AMOP winter-spring : 95 mg m −2 d −1 on average; see Fig. S1, Table S1). Intra-seasonally, during summer (AMOP summer ), the variability in fluxes at 34 and 149 m is 3 times and 40 % lower than during winter-spring (AMOP winter-spring ; see Table S1) with a SD = 144 % (39 < 4647 mg m −2 d −1 ) and 138 % (12 < 488 mg m −2 d −1 , respectively. The POC flux (Table 1) is globally proportional to the total particle flux (R 2 = 0.92) (Fig. S2a). Therefore analysis of the total particle flux and the POC flux will lead to similar results. To investigate the influence of the oxygendeficient layer between both traps for each season, we use the POC transfer efficiency (T eff ) introduced by Buesseler et al. (2007), defined as The transfer efficiency allows the determination of the ability of the system to preserve OM quantity and to foster the export of carbon from the productive layer. The higher the transfer efficiency, the higher the proportion of particles reaching the deeper trap. Therefore, T eff is an index of the relative amount of carbon that reaches the deeper trap. The mean transfer efficiency appears to be relatively similar for both datasets (T eff ∼ 45 %). However, T eff values present a strong temporal variability, with T eff being more than 3 times more variable for AMOP winter-spring than for AMOP summer (Fig. 2; Table 1).
T eff can be higher than 100 %, referring to particle accumulation between both traps. T eff > 100 % is potentially attributable to advection of particles or to primary or secondary production between traps. T eff is never higher than 100 % during AMOP summer but it is 3 times higher during AMOP winter-spring (AMOP winter-spring -S1, AMOP winter-spring -S3, AMOP winter-spring -S6; T eff = 135 %, 106 % and 149 %, respectively). Due to the potential bias affecting these values without considering the OMZ influence, values of T eff > 100 % were discarded. Excluding these very high T eff results, transfer efficiency varies between 1 % and 71 %. Three ranges of variation can be defined: high, intermediate and low. The high range with relatively efficient transfer (T eff > 50 %) corresponds to a predominance of OM preservation. This preservation is observed for a third of the total samples, namely the first three samples of AMOP summer (AMOP summer -S1, AMOP summer -S2 and AMOP summer -S3), the sample between 17 and 24 March (AMOP summer -S11), and two samples in winter (AMOP winter-spring -S2 and AMOP winter-spring -S5). Conversely, the other samples correspond to a predominance of potential OM degradation or remineralisation. In late winter-early spring, the proportion of particles reaching the deeper trap is very low (low range of T eff < 6 %). This period represents half of the AMOP winter-spring period (from 2 September to 6 November) and seems to correspond to the main period of OM degradation. Between these extreme high and low values, T eff presents an intermediate range between 20 % and 50 %. Intermediate T eff occurs mainly in summer from 27 January to 31 March (AMOP summer -S4, AMOP summer -S5, AMOP summer -S6, AMOP summer -S7, AMOP summer -S8, AMOP summer -S9, AMOP summer -S10 and AMOP summer -S12). Within the 20 < 50 % T eff range, the six samples in summer (AMOP summer -S4, AMOP summer -S6, AMOP summer -S8, AMOP summer -S9, AMOP summer -S10 and AMOP summer -S12), and one sample in winter (AMOP winter-spring -S4) present a low intermediate T eff range of 20 < 38 %.
Note that the vertical profile of POC flux is assumed to follow a power law (Suess, 1980;Martin et al., 1987).
The b coefficient represents the attenuation of the curve and therefore an index for OM respiration during sinking.  Devol and Hartnett, 2001), in the Peruvian OMZ (0.66±0.24, Martin et al., 1987) and in the Arabian Sea OMZ (0.22, Roullier et al., 2014;0.59 ± 0.24: Keil et al., 2016). The temporal modulation of b is comparable to that associated with the spatial switch from the coast to the open ocean off Peru (Packard et al., 2015) as well as from high-to lowlatitude regions (Marsay et al., 2015). Globally, and in line with the transfer efficiency, the strongest attenuation was observed in spring (b = 2.43 ± 15 %) and the weakest in winter (b = 0.47 ± 57 %). One can notice that the attenuation is also 4 times stronger and 2.5 times more variable in spring than in summer, where b is on average 0.62 ± 39 %. Considering the ranges previously defined, b = 2.43(±0.37) for low T eff (< 6 %), b = 0.74(±0.15) for intermediate T eff (20 < 50 %) and b = 0.36(±0.06) for high T eff (> 50 %).

Significance of transfer efficiency
While the transfer efficiency (T eff ) between the upper (34 m) and lower (149 m) sediment traps allows a mathematical distinction among ranges of POC export efficiency, it is crucial to investigate the physical significance of this T eff . Particles sampled at 149 m are not necessarily associated with the same vertical flux as those previously sampled at 34 m over the 7-or 11-day period. This is due to the following three main processes: (a) horizontal transport, (b) vertical sinking speed defining the finite time between both traps and (c) particle production between both traps involving different trophic levels.
T eff may be affected by horizontal advection of particles as well as by sediment trap line inclination, in response to the coastal current system (e.g. shear due to the northward flow at the surface and southward flow in the subsurface layers). These typical methodological biases in sediment trap studies are known to potentially affect the collection of particles and their efficiency. Here, the mean alongshore (poleward) current reaches ∼ 12 cm s −1 (slower than 15 cm s −1 ) over the duration of the experiment in the vicinity of the sediment traps. This is based on in situ data (AMOP cruise), climatology (Chaigneau et al., 2013) and estimates from climatological regional model simulations (Montes et al., 2010;Dewitte et al., 2012). Therefore, the collection of particles is considered to be marginally affected by currents in this transfer layer (Baker et al., 1988). This is confirmed by a small inclination of the mooring line (< 5 • ). However, the only three samples presenting values of T eff > 100 % (AMOP winter-spring -S1, AMOP winter-spring -S3 and AMOP winter-spring -S6) are characterised by a relatively high inclination anomaly (related to the mean inclination) of the mooring line. This high inclination anomaly can be assigned to a significant modification of the horizontal currents' mean state. Zonal advection of particles from a more productive region in the lower trap could explain anomalous high transfer.

M. Bretagnon et al.: Modulation of the vertical particle transfer efficiency
Vertically, we assume that upwelling or downwelling events (velocity below 0.5 m d −1 ) do not significantly impact particle sinking speed (ranging from 1 to 2700 m d −1 ; Siegel et al., 1990;Waniek et al., 2000). In addition, the quantity of matter collected by the sediment trap in each cup at 149 m may be different from that collected in the corresponding cup at 34 m. This depends on the vertical velocity of specific particles. Particle velocity also determines exposure time to degradation activity in the water column. Therefore, the probability of particle degradation may increase for slower (generally smaller) particles. They spend a longer time in the subsurface active remineralisation layer between 34 and 149 m, which could be the case for samples with T eff < 50 %. Conversely, the large amount of matter collected at 149 m for the high T eff range (> 50 %) might be explained by the presence of potentially less degraded particles, resulting from a faster sinking velocity (McDonnell et al., 2015). In theory, the sinking velocities of biogenic particles depend on the three-dimensional properties of the flow field as well as on various intrinsic factors (such as their sizes, shapes, densities and porosities; Stemmann and Boss, 2012). These intrinsic factors can be modified along their fall by complex biophysical processes (e.g. aggregation, ballasting and trimming by remineralisation). Note that processes like aggregation (e.g. flocculation) or disaggregation may affect the vertical transfer, as they modulate the sinking rate. Indeed, while disaggregation transforms fast-sinking large particles into small suspended particles, aggregation of small particles will induce their sinking. However, the samples from the present study are mainly composed of faecal pellets, and in a relatively equal proportion for both traps. Therefore, biophysical processes do not appear to be the main factor that modulates transfer efficiency.
Finally, subsurface particle production between 34 and 149 m can affect T eff . For instance, T eff can be affected by the presence of a deep or secondary chlorophyll maximum (SCM), which can sometimes be more intense than the primary maximum in OMZ areas (Garcia-Robledo et al., 2017). The fluorometer data at 31 m suggest an intermittent increase in fluorescence around this depth (Fig. 4e-f). However, the high fluorescence values could be attributable to the detection of a SCM or to deepening of the surface mixed layer, mixing the surface chlorophyll with the subsurface layer. To complement fluorescence data, δ 13 C values (Table S4) provide some information about particle production and typical processes linked to surface productivity. Here, the transfer efficiency of δ 13 C is roughly constant (between 88 % and 105 %, Table S4). This suggests no significant primary production below the top trap and therefore no significant contribution of an SCM. In addition, the water column between 34 and 149 m is mostly below the euphotic layer in which primary producers are mainly active. Thus, particle production is considered to be related to higher trophic levels only, in particular zooplankton (e.g. detritus due to excretion and mortality). Zooplankton can also induce active vertical trans-port of particles not directly considered here. However, zooplankton production should be pointed out for AMOP summer -S1. AMOP summer -S1 is characterised by an oxygen event (> 10 µmol kg −1 , Fig. 4a) but with an unexpectedly high T eff (71 %) suggesting preferential OM preservation. This sample is different from the others previously mentioned, the number of swimmers collected being 3 times higher than the average of the AMOP summer dataset, especially in the lower trap. This AMOP summer -S1 specificity is raising the influence of the active transfer between both traps and potentially secondary production.
In this study, we assume that (a) similar particles are collected in the upper and lower traps for each sampling period; (b) all particles collected in the deep trap are, for each sampling period, subject to a comparable time in the OMZ layer. Note that all the AMOP summer samples in summer and the first six AMOP winter-spring samples (from AMOP winter-spring -S1 to AMOP winter-spring -S6) in winter appear to be subject to a comparable trophic level effect. Indeed, samples are composed of around 60 %-70 % faecal pellets, therefore spending a similar amount of time in the OMZ. However, for the last six AMOP winter-spring samples (from AMOP winter-spring -S7 to from AMOP winter-spring -S12) in late winter-early spring, the proportion of faecal pellets and marine snow inverts. For the AMOP winter-spring dataset, low T eff (from AMOP winter-spring -S7 to AMOP winter-spring -S12) can be explained by the higher proportion of small particles (as marine snow), which is potentially easily degradable.
Transfer efficiency (T eff ) is controlled by the degradation of particles occurring below the productive layer. OM degradation can be due to macro-organisms feeding (Lampitt et al., 1990) or to microbial activity (e.g. Devol, 1978;Lam et al., 2009;Stewart et al., 2012;Roullier et al., 2014). This degradation implies a dependence on oxygen availability. Low oxygen availability could constrain zooplankton in a specific layer, therefore limiting feeding. Low oxygen availability could also reduce the microbial activity. Aerobe remineralisation is considered to be 10 times more efficient than anaerobe remineralisation (Sun et al., 2002;Taylor et al., 2009). However, in addition to the main requirement of catabolic energy fuelled by O 2 availability, OM bioavailability should feed the substrate anabolic requirement of the heterotrophic microbial community controlling remineralisation activity. This argument is in line with previous studies showing microbial nitrogen cycling regulated by OM export (Kalvelage et al., 2013). Therefore, for intermediate (20 % < T eff < 50 %) and high T eff (> 50 %) values, OM degradation is considered limited, whereas for low T eff (< 6 %) it is not. The role of oxygenation and OM availability in OM degradation was explored. This was to provide a better estimation of whether the quantity of carbon remains available for surface production and air-sea exchange, or whether it is preserved and exported toward the sediment.

Key parameters modulating particle transfer efficiency
The respective roles of oxygen and OM in modulating transfer efficiency will be evaluated.

The role of oxygen
Transfer efficiency (T eff ) shows a variation among seasons, as well as at an intraseasonal level. The role of oxygen is investigated by considering temporal changes in oxygenation and whether they could be a potential factor associated with changes in remineralisation activity. Could this explain the T eff modulation? Vertical and temporal [O 2 ] changes mainly occur near the oxycline and upper OMZ core (upper trap) rather than in the lower OMZ (lower trap). In the lower OMZ, O 2 concentration remains stable, reaching the lowest detection limit (Fig. 3). Close to the upper trap, oxygen concentration can then be a key factor triggering limitation of remineralisation.
Seasonally, mean oxygen concentration appears to be ∼ 10 times lower for AMOP summer (∼ 5 µmol kg −1 ) than for AMOP winter-spring (∼ 60 µmol kg −1 ; Fig. 4a-b). The dailyaveraged oxygen concentration at 34 m highlights the existence of two steady states regarding oxygenation: (i) the suboxic conditions occurring in summer, in which [O 2 ] stays below 25 µmol kg −1 , and/or with a shallower oxycline; (ii) the hypoxic-oxic conditions occurring in winter and early spring, in which [O 2 ] is always above 15 µmol kg −1 , and/or with a deeper oxycline (Fig. 4a-b). Suboxia corresponds to limiting conditions for both aerobe micro-and macro-biological (e.g. bacteria and zooplankton) OM degradation, thereby impacting the vertical transfer efficiency (T eff ). This is confirmed by the fact that the abundance of swimmers during AMOP summer is half the amount as during AMOP winter-spring . This is also confirmed by the relative abundance of polychaetes, known to better tolerate suboxic conditions than copepods. The number of reported polychaetes is 22 times higher at the oxycline during AMOP summer than during AMOP winter-spring (Table 2). Oxygen concentration may also indirectly impact T eff . More oxygenated conditions (e.g. during AMOP winter-spring ) allow micro-organisms such as copepods to colonise depths between both traps, and therefore potentially produce particles within this layer through sloppy feeding, faecal pellets and carcasses sinking. The latter mechanism may explain the T eff higher than 100 %.
In addition to concentration considerations, [O 2 ] for AMOP summer is 10 times less variable (SD = 2.6 µmol kg −1 for [O 2 ] 7 days_15 min ) than for AMOP winter-spring (SD = 28 µmol kg −1 for [O 2 ] 11 days_15 min ; Table 2). This difference regarding variability highlights less intense and 2 times less frequent oxygenation events during AMOP summer than during AMOP winter-spring (Fig. 4a-b). The more elevated O 2 conditions observed during AMOP winter-spring are favourable to OM degradation through both micro-and macroorganisms. This is therefore consistent with a lower T eff (< 6 %), at which no limitation of the degradation mechanisms is considered.
At higher frequency, within the weekly period, the oxygenated conditions present oxygenation episodic events with (i) a higher daily occurrence (≥ two per week; up to six events for AMOP summer -S8) and (ii) often relatively intense ([O 2 ] 1 day_15 min reaching 24.2 µmol kg −1 for AMOP summer -S9; Fig. 4a). In contrast, the less oxygenated conditions present only less frequently occurring oxygenation events (≤ two per week), which are generally less intense.
The oxygenation events, reported for both AMOP summer and AMOP winter-spring , are linked with density minima (< 26.1 kg m −3 ) and are relatively consistent with a deepening of the MLD. This suggests vertical diapycnal mixing with surface water (Fig. 4a-d). Induced vertical mixing appears to be driven by an increase in wind intensity, frequency (more than one wind pulse per week) and duration (∼ 10 days). Globally, the averaged density for AMOP winter-spring is lighter than AMOP summer by ∼ 0.07 kg m −3 . Short wind-driven mixing events are followed by a longer re-stratification period associated with an [O 2 ] decrease (of ∼ 5 µmol kg −1 for AMOP summer and > 20 µmol kg −1 up to 100 µmol kg −1 for AMOP winter-spring ) and density increase (of > 0.1 kg m −3 , up to 0.4 kg m −3 for AMOP 2), then stabilisation. The sequences of mixingstratification and oxygenation-deoxygenation could have been induced by sequences of stirring (or downwellingupwelling). These sequences are typically observed during upward transportation of deeper, denser and lower [O 2 ] water, in response to a modulation in alongshore winds favourable to Ekman transport. A propagation of coastal trapped waves, with in-phase vertical fluctuations in the density and oxygen isopleths, can also take place (Sobarzo et al., 2007;Dewitte et al., 2011;Illig et al., 2014). These winddriven oxygenation events during the lowest seasonal steadystate oxygenation as in AMOP summer potentially modulate the intensity of remineralisation at an intra-monthly frequency. In fact, during summer, transfer efficiency varies up to a factor of 2 (24 < 57 %) associated with oxygenation events. These oxygenation events allowing less O 2 limitation are consistent with a relatively lower intermediate T eff between 20 % and 40 % (e.g. for AMOP summer -S4, AMOP summer -S6, AMOP summer -S8, AMOP summer -S9 and AMOP summer -S10).
The transfer efficiency (T eff ) decreases from high (> 50 %) to low intermediate (20 < 38 %) when [O 2 ] at the oxycline, or in the upper OMZ, increases during oxygenation events. This T eff decrease occurs at (i) a seasonal scale from the limit of detection of [O 2 ] 1 day_15 min higher than ∼ 5 up to ∼ 25 µmol kg −1 in summer, and (ii) an intraseasonal scale from less (∼ 5 µmol kg −1 in summer) to more (∼ 60 µmol kg −1 in winter-spring) oxygenated mean states. However, for the similar winter-spring hypoxic-oxic conditions at the oxycline, the modulation of T eff (between 1 % and 68 %) suggests that a factor other than oxygen is restricting the mechanism of OM degradation and remineralisation.

The role of OM
In addition to oxygen, transport mechanisms, sinking time and trophic transfer having an effect, other processes that depend on the nature of particles may explain the contrast in transfer efficiency (T eff ). Collected particles are marine and organic. Collected particles can mainly be considered to be OM, based on the similar modulation of T eff for POC and the transfer efficiency for the total particles (Figs. S1, S2 and S4). Indeed C : N ratios at 34 m (between 5.7 and 10.1, Tables 3b and S3a) are always below 20, which is characteristic of a marine origin (Mayers, 1993), although approximately 13 % higher than the canonical Redfield values (Redfield et al., 1963). Carbon isotopic signatures (δ 13 C) are between −22.7 ‰ and −17.4 ‰ and δ 15 N between 3.5 ‰ and 13.1 ‰ (Tables 3c and S4). These δ 13 C values are consistent with marine organic compounds and inconsistent with terrigenous influence (Degens et al., 1968;Ohkouchi et al., 2015;Bardhan et al., 2015).
Variability in the exported OM acts as anabolic biogeochemical forcing, supplying the OMZ with particles to be degraded and remineralised. OM variability thereby poten- tially mitigates the transfer efficiency (T eff ). The variability in the particle flux collected in the upper trap is thus considered in order to understand the role of OM quantity and quality on transfer efficiency (T eff ). Quantitatively, the POC flux at 34 m presents a seasonal variability. POC flux values are 40 % higher during AMOP winter-spring than AMOP summer (on average 131 and 93 mgC m −2 d −1 , respectively). POC flux also presents a stronger intra-seasonal variability during AMOP winter-spring , being more than 1.5 times more variable than during AMOP summer . This is confirmed by the fluorescence measurement at 31 m, higher for AMOP winter-spring than for AMOP summer (Fig. 4e-f). In fact, a deepening of the MLD as a response to the wind strengthening can increase the fluorescence values at 31 m by stronger vertical mixing of the chlorophyll produced at the surface. Thus, mixing of the surface productivity with the subsurface layers could contribute to an increase in fluorescence in the subsurface.
During the AMOP winter-spring winter period, low light availability and high mixing (Fig. 4d) induce low surface productivity according to lower fluorescence values at 31 m. The lower fluorescence contributes to the low POC flux recorded by the upper trap (Fig. 4f). Conversely, in spring, the water column stratifies (MLD decrease; Fig. 4d). The surface productivity increases in line with higher fluorescence values (globally higher than 1 µg L −1 ). This productivity increase leads to a higher concentration of particles and a POC flux about 10 times stronger than in winter (239.58 < 24.79 mgC m −2 d −1 ; Figs. 2b and 4f). The T eff decrease from winter to early spring, characterised by high and intermediate (> 20 %) and low (< 6 %) values, respectively, follows a power tendency line (Fig. 5). An increase in remineralisation activity as a consequence of a primary productivity increases could be suspected, as previously reported for the anoxic basin of Cariaco (Thunell et al., 2000).
During AMOP summer in suboxic conditions at the oxycline and O 2 limitation of OM degradation, the events of slight oxygenation have supported the modulation of T eff from high (> 40 %; AMOP summer -S2, AMOP summer -S3,  Fig. 2) with error bars corresponding to the standard deviation. (b) Sector diagrams of particle composition (mol%) in terms of particulate organic carbon (POC), particulate organic nitrogen (PON), particulate organic phosphorus (POP) and biogenic silica (BSi). The values indicated as a percentage correspond to the abundance of one element relative to the sum of the four other elements analysed here for the OM. For more details, see Tables S1 and S2. Note that due to its specific OM quality at 34 m, the AMOP2-S4 sample has been extracted from the intermediate T eff range and represented separately. AMOP summer -S5, AMOP summer -S7, AMOP summer -S11) to low (< 40 %; AMOP summer -S6, AMOP summer -S8, AMOP summer -S9, AMOP summer -S10) intermediate values, except for AMOP summer -S1 (considered apart; see Sect. 3.1.2). Now, the variability in OM quantity together with O 2 availability is analysed to identify conditions potentially leading to remineralisation-like and preservation-like configurations (T eff below or above 50 %). The highest POC fluxes (> 85 mgC m −2 d −1 ) at 34 m occur from AMOP summer -S1 to AMOP summer -S6 and AMOP summer -S8 to AMOP summer -S9 (Figs. 2a and 4e). For AMOP summer -S7 and from AMOP summer -S10 to AMOP summer -S12, POC fluxes are ∼ 30 % lower than the seasonal average. The low OM quantity could therefore explain a weaker remineralisation and thus a slightly higher T eff (up to 57 %). However, limitation of both O 2 and OM, simultaneously (co-limitation) or not, should be considered. When the ratio POC flux / [O 2 ] is far from the mean range, POC flux / [O 2 ] values define a severe limitation in OM only (< 7 for AMOP summer -S11) or O 2 only (> 27 for AMOP summer -S2 and AMOP summer -S3). In these cases, T eff becomes high. Conversely, the ratio POC flux / [O 2 ] remains closer to the mean range (7 < 27) for AMOP summer -S4, AMOP summer -S5, AMOP summer -S6, AMOP summer -S7, AMOP summer -S8, AMOP summer -S9, AMOP summer -S10 and AMOP summer -S12, except for AMOP summer -S1 (considered apart; see Sect. 3.1.2). In these cases, T eff remains interme-diate (20 < 50 %), associated with a potentially balanced co-limitation in O 2 and OM on OM degradation.
Qualitatively, the evolution of elemental fluxes at 34 m should be considered in investigating whether the composition of a more or less labile OM can affect transfer efficiency (T eff ) in addition to the OM availability. For both datasets, POC and PON fluxes show a strong linear correlation with the total particle mass fluxes at the upper trap as well as for the lower trap (R 2 = 0.98 for both traps; Fig. S2a, b). To study the influence of particle quality on transfer efficiency, the composition was averaged for the three main ranges of T eff . Also, as the matter collected in the trap is mainly organic, only the four main components (POC, PON, POP and BSi) were considered here. Whatever the range of T eff , the particle flux is dominated by POC. Then, BSi, PON and POP contribute in different proportions to the particle flux ( Fig. 6b; Tables 3a and S2). For low T eff , POC dominates with only 49 %. For intermediate (including AMOP winter-spring -S4) and high T eff ranges, POC remains relatively constant reaching 65 %-66 % of the total POC + BSi + PON + POP. Conversely, BSi reaches 43 % for low T eff . For the relatively constant intermediate and high T eff ranges, BSi only reaches 25 %. Whatever the T eff range, PON and POP have a relatively stable contribution of 7 %-8 % and 1 %-2 %, respectively. Between intermediate and high T eff ranges, the relative constancy in the composition of the par-ticles does not allow the investigation of the influence of the quality on transfer efficiency.
Nevertheless, for the remineralisation event observed in AMOP winter-spring -S4, while OM quantity was expected to limit remineralisation, the influence of quality should be pointed out as another factor acting on its low intermediate T eff (32 %; Fig. 6, Table 1). Indeed, this sample is specifically characterised by a relatively low BSi content (∼ 19 %, 50 % lower than the winter average) and the highest PON and POP proportions (35 % and 5 times higher than the winter average, respectively; Fig. 6b; Table S2). For this sample, the relative proportion of BSi decreases and that of PON increases compared to the other intermediate, low and high T eff samples, leading to less refractory and more labile matter, preferentially degraded. The difference in composition for this sample could also be seen in terms of calcium carbonate. The AMOP winter-spring -S4 CaCO 3 upper flux exhibits a maximum at this date compared to other intermediate and high T eff (Table 3c) and about 5 times higher (20 % of the total mass flux) than the average for the entire dataset (Table S5). A difference in the phytoplankton community could be at the origin of this distinction. Indeed, at the beginning of the AMOP winter-spring dataset and up to this sample, the MLD is relatively stable. No strong MLD deepening is observed as a consequence of wind intensification. As the surface productivity is mainly due to diatoms, this long period with no strong mixing events can induce silica depletion at the surface, limiting diatom growth. This hypothesis is supported by the analysis of the phytoplankton functional types around the mooring location using MODIS data and the algorithm developed by Hirata et al. (2011). For AMOP winter-spring -S4, prymnesiophytes become the most influent phytoplanktonic group. The dominance of prymnesiophytes in this sample could have induced a change in the composition of the sinking particles. In particular, their BSi proportion decreases, leading to a more labile matter.
The composition of the matter at the upper trap can also be observed as a function of the particulate molar ratios to identify the relative elemental excess or deficit. Whatever the considered range of T eff , BSi appears to be in excess. Si : C, Si : N and Si : P are on average 3.9, 4.4 and 3.0 times higher than the classical ones, respectively ( Table S3b). The strongest BSi excess can be assessed for low T eff (ratio ∼ 5 times higher than the classical ones; Table 3b). For the other elemental ratios, low T eff appears to be different from the other T eff ranges. Indeed, low T eff presents a relative POP deficit (C : P and N : P ∼ 20 % higher than classical ones) with a C : N ratio equal to the classical one (6.67). Conversely, the other T eff ranges present a relative POP excess (C : P and N : P about half the amount of classical ones), with a PON deficit relative to POC (C : N) between 12 and 24 %.
Molar ratios at 34 m for AMOP winter-spring -S4 confirm the analysis of elemental composition (Fig. 6b). In particular, BSi deficit (e.g. Si : C and Si : N about twice as low and Si : P ∼ 4 times lower than classical ones) and P excess (C : P and Figure 7. Mean transfer efficiency for the main components of the particle fluxes (related to POC, PON, POP, BSi, CaCO 3 , and particulate δ 13 C and δ 15 N), as a function of the three main T eff ranges defined from POC fluxes. The transfer efficiencies for PON (T effPON ), POP (T effPOP ), BSi (T effBSi ), CaCO 3 (T effCaCO 3 ), δ 13 C (T eff 13 C ) and 15 N (T eff 15 N ) are derived from Eq. (1), as for T eff . Error bars represent the associated standard deviation of the elemental transfer for the considered T eff ranges (more details in Table 3). N : P ∼ 8 times higher than classical ones; Table S3) are reported. However, Si : P, C : N, C : P and N : P present strong intra-seasonal variability. Relative SD reaches 60 % for the intermediate T eff range. In particular, reported Si : P, C : N, C : P and N : P values are below or above standard reference levels, whatever the T eff range.
Therefore, the OM quantity produced above the oxycline appears to have a stronger influence on transfer efficiency than the OM quality of the sinking particles. However, more or less labile materials can also contribute to better preservation and export of particles towards sediment or their remineralisation in the upper layers of the ocean. Together with oxygenation conditions, OM quantity is a major factor in triggering strong remineralisation. It may be strengthened or mitigated by OM quality, which is considered here to be a secondary factor. The significance of cofactors is consistent with the fact that oxic or anoxic conditions have a different effect on OM degradation (faster or slower, respectively) depending on the OM components considered (e.g. Taylor et al., 2009).

Vertical flux modulation for elemental and isotopic components
The analysis of particle transfer efficiency through the OMZ has been focussed on the carbon element, defining three main T eff ranges. Here we study the transfer efficiencies of other particle components (T effPON for PON, T effPOP for POP, T effBSi for BSi, T effCaCO 3 for CaCO 3 , T eff 13 C and T eff 15 N for isotopic signatures) as well as their modulation and distribution. All transfer efficiencies for the elemental composition present a low range (< 15 %), clearly dissociated from the intermediate (15 < 55 %) and high (55 < 80 %) ranges ( Table S2). T effPOP alone shows a more gradual transition be-the T effCaCO 3 decrease could be associated with a stimulation of significant water column dissolution. Since the intermediate T eff range corresponds to a large predominance of ∼ 90 AMOP summer samples (∼ 90 %, Table 1), the explanation is focussed on AMOP summer only and based on the following consideration.
The CaCO 3 transfer efficiency could be modulated partly by pH conditions, and partly as a consequence of ballast. Indeed, OMZs are characterised by low-pH conditions (Paulmier and Ruiz-Pino, 2009;Paulmier et al., 2011;León et al., 2011) and may induce calcite dissolution (e.g. Orr et al., 2005). As low pH was recorded in a cross-shore section during the AMOP cruise (austral summer 2014, Fig. S3), CaCO 3 dissolution could potentially be considered as a factor acting on CaCO 3 transfer in AMOP summer samples. Moreover, because of the refractory nature of CaCO 3 , it could accumulate along the water column. This potential CaCO 3 accumulation could explain the high transfer efficiency. Ballasting could therefore explain the transfer efficiency of over 100 % for some samples. Note however that the error bar on T effCaCO 3 is high, and thus does not allow precise differentiation among these invoked processes.
In line with palaeoceanographic studies, it should be interesting to focus on the evolution of particulate δ 13 C and δ 15 N as a function of the T eff ranges. T eff 13 C and T eff 15 N remain around 100 % whatever the T eff ranges (Fig. 7, Table 3c). While δ 13 C appears to slightly decrease during particle sinking (average T eff 13 C of 95 %; Table S4), no significant T eff 13 C distinction among low, intermediate and high T eff ranges could be made. For all T eff ranges and all considered seasons, carbon isotope is slightly heavier in the lower trap (Table S4). The enrichment in heavy isotopes with depth potentially indicates an increasing influence of inorganic carbon with depth. The T eff 15 N variation appears to be stronger than T eff 13 C , varying between 83 % and 267 % (average of 125 % ± 37 %). T eff 15 N remains high (∼ 100 % and higher) with variations of up to 40 %. The highest T eff 15 N (156 %) occurs for the intermediate T eff range. But T eff 15 N is lower for low and high T eff ranges (107 % and 113 %, respectively). The T eff 15 N differences between ranges could be linked to the vertical alteration of particulate δ 15 N distribution associated with the OMZ oxycline and core structure (Libes and Deuser, 1988). The three T eff range configurations could correspond to different (i) chemical and isotopic PON compositions associated with different metabolic pathways of OM synthesis, or (ii) oxycline depths and/or oxygenations impacting the microbial activities and the OM degradation processes. The alteration to particulate δ 15 N is potentially due to the microbial activities and chemical and isotopic PON compositions associated with the different metabolic pathways of OM synthesis. In addition, the depth and oxygenation of the oxycline should play an important role in explaining the differences among low, intermediate and high T eff ranges. However, one should be cautious with these results, as the variability in δ 15 N for the entire dataset does not allow strict differentiation among the ranges.
Therefore, in the OMZ, short-term [O 2 ] fluctuations and particle loading appear to be significant. They should be considered for further studies on OM fate, carbon sequestration, nutrient regeneration, and the production or consumption of greenhouse and toxic dissolved gas. These effects call for the development of variable molar ratio models for OM production mechanisms, and OMZ remineralisation processes, and for the revised interpretation of palaeo-proxies.

Summary and conclusion
The seasonal and intraseasonal analysis of particles collected using sediment traps in the oxycline and the core of the Peruvian EBUS confirms that the OMZ can behave as either a recycling or a preservation system for organic matter (OM). Transfer efficiency (T eff ) presents variations that can be classified into three main characteristic ranges: high with 50 < T eff < 75 % associated with preservation capacity, intermediate with 20 < T eff < 50 % or low with T eff < 6 %, associated with remineralisation capacity. These T eff variations represent a more or less efficient carbon export through the OMZ. Seasonally, two different steady states are defined for oxygen conditions. The suboxic regime ([O 2 ] < 25 µmol kg −1 ) occurs mainly in austral summer. The hypoxic-oxic regime occurs in austral winter and early spring (15 < [O 2 ] < 160 µmol kg −1 ). Suboxia is expected to foster OM preservation and therefore enhance transfer efficiency. But low-O 2 conditions occurring in austral summer can also induce slight remineralisation, as a consequence of episodic wind-driven oxygenation events. In addition to oxygenation conditions, sinking particles from the oxycline play a role in transfer efficiency. Indeed, the high POC flux (> 80 mgC m −2 d −1 ) end of winter-early spring can provide enough substrates to sustain the anabolic requirement of the microbial activity and shut down the vertical transfer (T eff < 6 %). In contrast, the extreme deficits in oxygen ([O 2 ] < 5 µmol kg −1 at the oxycline) or OM (< 40 mgC m −2 d −1 ) are considered a limitation for OM degradation activity (e.g. microbial remineralisation and zooplankton feeding on particles). These configurations correspond to the most efficient POC transfer (T eff > 50 %) for both summer and winter seasons. Between high and low T eff , higher levels of O 2 or OM, even in a co-limitation context, can lead to slightly decreased OMZ transfer efficiency (20 < 50 %), especially in summer (20 < 40 %). For all sampling seasons, particle composition could be considered stable, mainly composed of POC and BSi. The stable particle composition thus does not allow a full investigation of the question of the impact of OM quality. In the time and spatial location covered by this study, OM quality does not seem to be the main factor leading to T eff modulation. Only punctually could the occurrence of nitrogen-rich organic compounds in relatively well-oxygenated water strengthen remineralisation activity, with low intermediate T eff (32 %). This study reconciles two opposite views concerning the effects of OMZ behaviour on OM cycling. It supports the existence of both dynamic and static balanced biogeochemical states defined as states with and without significant remineralisation and O 2 consumption, respectively. The key microbial feedback on particles including their elemental composition as well as detailing the role of OM quality should be further investigated. This is expected to lead to a better understanding of the vertical OM transfer efficiency of the OMZ and its modulation. Climate projections and paleoceanography studies should therefore consider the intermittence of OMZ preservation or recycling capacity, which is crucial for global biogeochemical budgets.
Data availability. The data can be obtained by contacting the author (aurelien.paulmier@legos.obs-mip.fr).