Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 The ocean's biological carbon pump is a key component of the global carbon cycle. Only a small fraction of the carbon fixed by primary production is exported to the deep ocean, yet this flux sets to first order the efficiency with which carbon is sequestered out of further contact with the atmosphere on long time scales. Here we examine global patterns in particle export efficiency (PEeff), the proportion of primary production that is exported from the surface ocean, and transfer efficiency (Teff), the fraction of exported organic matter that reaches the deep ocean. Previous studies have found a positive correlation between Teff and deep ocean calcite fluxes recovered from sediment traps, implying that ballasting by calcium carbonate may play an important role in regulating Teff. An alternative explanation is that this correlation is not causative, as regions where the dominant biomineral phase is calcite tend to be subtropical systems, which are hypothesized to produce sinking aggregates highly resistant to degradation. We attempt to distinguish between these alternative hypotheses on the control of Teff by examining the relationship between Teff and biomineral phases exported from the upper ocean, rather than those collected in deep traps. Global scale estimates derived from satellite data show, in keeping with earlier studies, that PEeff is high at high latitudes and low at low latitudes, but that Teff is low at high latitudes and high at low latitudes. However, in contrast to the relationship observed for deep biomineral fluxes in previous studies, we find that Teff is strongly negatively correlated with opal export flux from the upper ocean, but uncorrelated with calcium carbonate export flux. We hypothesize that the underlying factor governing the spatial patterns observed in Teff is ecosystem function, specifically the degree of recycling occurring in the upper ocean, rather than the availability of calcium carbonate for ballasting.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The biological carbon pump transfers large amounts of organic carbon from the upper ocean to the deep interior, playing an important role in regulating atmospheric CO2 levels. In the sunlit euphotic zone, primary producers convert CO2 to O2 and particulate organic carbon (POC). Of the POC generated through primary production (PP), the majority is remineralized by metabolic processes in the epipelagic zone (∼0–200 m). A small fraction of this POC survives respiration and is exported below the thermocline. Further remineralization occurs at mesopelagic depths (∼200–1000 m), so that the flux of POC to the deep ocean is only a tiny fraction of the POC resulting from PP. As the deep ocean is a major sink for atmospheric CO2 and sequesters carbon on timescales of hundreds to thousands of years, quantifying the efficiency of the biological carbon pump is key to our understanding of the global carbon cycle [Falkowski et al., 1998; Sabine et al., 2004].
 The earliest attempt to quantify the export ratio, i.e., the ratio of organic matter exported to that created by primary production, was the seminal work of Dugdale and Goering . They defined the f-ratio as new production/total production (i.e., new plus regenerated production). New production is the primary production supported by new sources of nitrogen, supplied through, e.g., winter mixing or atmospheric deposition. Regenerated production is supported by forms of nitrogen previously recycled through the planktonic food web, such as ammonium or urea. The definition of thef-ratio assumes that over sufficiently long time and space scales, new production equals export production (the fraction of PP that is exported), and requires knowledge of the magnitude of all sources of new and regenerated nitrate. This has become increasingly complicated by the recent recognition of the roles of upper ocean nitrification [Yool et al., 2007] and nitrogen fixation [Gruber, 2004], which are additional sources of regenerated and new nitrogen, respectively. An alternative method to estimate the export ratio is to directly measure the particles sinking below the euphotic zone, either with shallow drifting sediment traps [e.g., Lampitt et al., 2008] or using isotope tracer methods (234Th, 210Po or 210Pb [e.g., Murray et al., 2005]). A compilation of export measurements made using a combination of shallow trap and 234Th methods led Dunne et al. [2005, 2007] to define a particle export ratio (POC export/PP), which was parameterized in terms of sea surface temperature (SST) and chlorophyll. Similarly, Henson et al. compiled estimates of thorium-derived POC flux to define a particle export ratio based on SST. Additional approaches to estimating export ratios rely on inverse modeling of observed nutrient and oxygen fields [Schlitzer, 2000, 2004] or oxygen or carbon isotope mass balance [Emerson et al., 2001]. Estimates of the global mean export ratio range from ∼10% [Henson et al., 2011] to ∼40% [Eppley and Peterson, 1979].
 The fraction of exported organic carbon that survives remineralization during sinking to reach the deep ocean is quantified by the transfer efficiency (Teff) and is defined as deep POC flux/exported POC [Francois et al., 2002]. Data on the POC flux reaching the deep ocean comes from moored sediment traps that collect sinking particles on timescales of months to years. The development of algorithms to estimate deep POC flux from surface conditions arose from the need to estimate flux over larger spatial scales than those represented by a sediment trap. Early attempts necessarily relied on a limited number of traps and were generally defined in terms of PP and entailed power law decreases of POC flux with depth [e.g., Suess, 1980; Pace et al., 1987]. Perhaps the most influential and enduring equation to estimate deep POC flux is that of Martin et al. , which uses exported flux combined with an exponent ‘b’ that describes the rate at which POC flux decreases with depth. More recently, Lutz et al.  developed a POC flux parameterization that incorporates the seasonal variability of PP. By combining deep POC fluxes and estimates of export, several recent papers have assembled sufficient estimates of the Teff to examine regional differences [Francois et al., 2002; Klaas and Archer, 2002]. These suggest that Teff is generally high (i.e., much of the organic material exported from the euphotic zone reaches the deep ocean) at low latitudes and vice versa.
 Observations of a strong correlation between POC fluxes and the flux of minerals (biogenic silica, calcium carbonate and lithogenic silica) in deep sediment traps led to the development of the ‘ballast hypothesis’ [Francois et al., 2002; Klaas and Archer, 2002; Armstrong et al., 2002]. High Teff was found to be associated with an accompanying deep flux of calcium carbonate [Francois et al., 2002; Klaas and Archer, 2002]. Conversely, the deep flux of biogenic opal was found to be essentially uncorrelated with Teff. Francois et al.  and Klaas and Archer  concluded that carbonate minerals act as effective ballast for organic carbon, either by increasing the density of sinking particles or by providing some protection against degradation. A ‘packaging factor’ was also hypothesized [Francois et al., 2002], with carbonate-dominated systems containing organisms that produce hydrodynamic fecal pellets capable of efficiently delivering organic carbon to deep waters. This is in contrast to opal-dominated systems that contain organisms which produce loose aggregates that disintegrate readily. Here we attempt to distinguish between the two hypotheses on the dominant factor controlling theTeff: ecosystem structure or presence of calcareous material.
 Here we use estimates of biomineral export from the surface ocean to address the factors affecting Teff, in contrast to the analyses that led to the ballast hypothesis which were based on fluxes captured in deep (>2 km) sediment traps. However, our rationale for using surface biomineral fluxes is that the processes occurring in the mesopelagic (base of the euphotic zone to ∼1000 m depth) are believed to be key in setting the Teff [e.g., Buesseler et al., 2007], as below ∼1000 m the reduction in POC flux with depth is relatively small. These processes which link the shallow export of organic carbon and the arrival of a much-reduced fraction in deep waters are poorly understood. The premise of the ballast hypothesis, that mineral fluxes provide protection against degradation or increase the density of sinking particles, implies that the association of the minerals with POC must take place in the upper ocean. Thus, comparison of transfer efficiency with the upper ocean export flux of minerals (rather than the deep flux) provides an additional test of the ballast hypothesis.
 In this study, we compare multiple permutations of satellite data-derived export production (EP) and deep POC flux algorithms to large databases of in situ measurements to determine which combination most closely reproduces the in situ data. Global scale estimates of export efficiency (POC export/PP) and transfer efficiency (organic carbon to deep ocean/POC export) are then developed from satellite data and compared to measures of ecosystem structure, as well as the global distribution of calcium carbonate and opal export flux. Our aim is to inform the debate on the processes controlling transfer efficiency by examining the ballast hypothesis in the upper mesopelagic region of the ocean.
2.1. Estimating Particle Export Efficiency
 The particle export efficiency, PEeff, is defined as POC export at 100 m/PP. To estimate POC export we use a large database (n = 419) of thorium-derived POC export measurements (mgC m−2 day−1) collated from the literature (see Table S1 in Text S1 in the auxiliary material). Thorium-234 is the radiogenic daughter product of the naturally occurring soluble isotope uranium-238, which is proportionally conserved in seawater. Unlike238U, 234Th is insoluble in seawater and readily adheres to particulate matter. As particles sink, a radioactive disequilibrium between 238U and 234Th arises which, when combined with data on the ratio of POC concentration to 234Th activity, allows an estimate of POC export to be made (see Buesseler  for details). The thorium tracer approach provides direct estimates of particle export, avoiding issues associated with the 15N and f-ratio methods [seeYool et al., 2007]. The thorium approach does, however, have inherent uncertainties arising from the estimation of POC:234Th ratios which can vary geographically, with depth or by sampling method [Buesseler et al., 2006]. Additional uncertainty arises from the use of steady state assumptions in some estimates of thorium-derived export. The locations of the measurements compiled from the literature for this study are plotted inFigure 1.
 Estimates of PEeff made using the thorium method to estimate downward POC flux usually use in situ PP measured during 12 or 24 h incubations. However, the residence time of 234Th is considered to be ∼2–20 days [Coale and Bruland, 1985], and therefore the export estimated using this method represents an integrated value over a period of days to a couple of weeks. This introduces a mismatch in time scales between POC flux and PP estimates. In a recent paper, Henson et al.  accounted for the integration time scale of the 234Th-derived export by using satellite-derived PP integrated over the 16 days prior to the in situ measurement. This is also the approach we follow here. To evaluate the satellite data-derivedPEeff, we compare estimates to a data-based value ofPEeff. There are 226 occasions in the database when concurrent in situ PP was measured alongside 234Th-derived export, allowing data-basedPEeff to be calculated.
2.2. Particulate Organic Carbon Fluxes at 2000 m
 Sediment traps provide the only source of in situ data on deep POC flux. Moored sediment traps have been deployed in all of the major ocean basins, often repeatedly over many years. Here we use a compilation of annual integrated POC flux data from numerous deep sediment traps deployed between 1976 and 2004, reported by Honjo et al. . Only traps which recorded a full annual cycle of export were included (n = 134). All POC flux data were normalized by Honjo et al.  to a depth of 2000 m using a Martin curve with b = −0.86. In Figure 1, the locations of the measurements of POC at 2000 m (2000FCorg, gC m−2 year−1) are plotted. Transfer efficiency is defined, following Francois et al. , as Teff = 2000FCorg /POC export at 100 m. We wish to estimate Teff using in situ data in order to evaluate our estimates of Teff based on satellite data. However, 2000FCorg and POC export are rarely measured concurrently in the same location, and in addition, particles collected by a sediment trap at 2000 m depth can originate from a large region around the site, determined by prevailing currents (the ‘statistical funnel’ [e.g., Siegel and Deuser, 1997]). Therefore, to calculate Teff from in situ data, 2000FCorgmeasurements located within a 3° × 3° box centered on the location of a thorium-derived POC export measurement were selected. This resulted in 41 data-derived estimates ofTeff, which are used to validate satellite data-derivedTeff estimates. Note that the 2000FCorg fluxes reported by Honjo et al. are annual totals, and so were divided by 365 to convert into daily rates prior to calculating data-basedTeff estimates.
 In addition to estimates of Teff, we present estimates of b (in the sense of Martin et al. ), which also represents the extent to which particles are remineralized in the water column. The Martin et al. equation (equation S4) is very commonly used in both data and model applications to describe the depth-dependent reduction in POC flux. Using data from six stations in the Northeast Pacific,Martin et al.  estimated that b ≈ −0.858. Due principally to a lack of data, b is generally assumed to be spatially constant, and the value of −0.858 has been applied globally in several biogeochemical models to describe the rate of particle remineralization in the water column [e.g., Doney et al., 2004]. In this study, in situ data-based estimates ofb are calculated using 2000FCorgmeasurements located within a 3° × 3° box centered on the location of a thorium-derived POC export measurement, resulting in 41 data-derived estimates ofb.
2.3. Satellite Data Derived Estimates of Export and 2000FCorg
 Several different algorithms have been proposed for estimating PP, EP and 2000FCorgfrom satellite data. These all require as inputs some combination of chlorophyll, SST and incident light. Climatological monthly mean Level-3 SeaWiFS chlorophyll-a concentration, photosynthetically available radiation and AVHRR SST data were downloaded fromhttp://oceancolor.gsfc.nasa.gov and http://pathfinder.nodc.noaa.gov/, respectively. All data were spatially averaged to a 1° grid.
 To select the combination of algorithms that produced estimates of POC export and 2000FCorg closest to observed values, we tested every permutation of the following algorithms. PP was estimated using the Vertically Generalized Productivity Model (VGPM) [Behrenfeld and Falkowski, 1997], Carr model [Carr, 2002] and Marra model [Marra et al., 2003]. Particle export ratios were calculated using the empirical algorithms of Henson et al.  and Laws et al. , both based on SST [Dunne et al., 2005], based on SST and chlorophyll concentration, and the food web model of Laws et al. , which requires SST and PP as inputs. Note that the algorithms of Laws et al.  are based on the f-ratio and so include contributions to export from both particulate and dissolved phases, whereas theDunne et al.  and Henson et al.  parameterizations consider only the particulate component of export. The organic carbon flux at 2000 m, 2000FCorg, was estimated by applying the Martin curve to export estimated by one of the above algorithms with a constant b = −0.858 [Martin et al., 1987] and the Suess , Pace et al.  and Lutz et al.  parameterizations, all of which use PP as an input. In addition, Lutz et al.  includes terms relating to the seasonality of PP. This results in 12 possible permutations for POC export and 21 for 2000FCorg. Details of all the algorithms and permutations can be found in the auxiliary material.
 Satellite data-derived estimates of POC export and2000FCorgwere validated against databases of in situ measurements of thorium-derived export [Henson et al., 2011] and 2000FCorg [Honjo et al., 2008]. For POC export, the measured thorium-derived export was compared to the mean of satellite-derived export in a 1° × 1° box centered on the measurement location, and integrated over the 16 days prior to the in situ measurement. For2000FCorg, the measured annually integrated POC flux at 2000 m was compared to the satellite-derived annually integrated value averaged spatially over a 3° × 3° box centered on the measurement location.
 The combination of algorithms which produced the best fit to the data (judged by highest Pearson correlation coefficient and lowest error variance) were Carr + Henson for POC export and Carr + Lutz for 2000FCorg. The selection of Carr + Henson as the best estimator for export is unsurprising, as the algorithm was derived from a subset of the database we use here for in situ export (i.e., Table S1). Figure 2 plots the observed POC export and 2000FCorgagainst the satellite-derived estimates (correlation coefficient, r = 0.45 for POC export and r = 0.55 for2000FCorg, p < 0.01 for both). Statistics for all the other algorithm permutations tested can be found in Tables S2 and S3. Note that all permutations of the various satellite algorithms tested here resulted in similar latitudinal distributions of PEeff and Teff, although in general the rejected algorithms overestimated both EP and 2000FCorg. The error on the estimates of globally integrated EP and 2000FCorgare reported as the median relative error of measured versus predicted values. In the remainder of the manuscript satellite-derivedPEeff and Teffrefer to ratios derived solely from satellite data-based estimates of PP, EP and2000FCorg using the algorithms described above.
2.4. Estimation of Calcium Carbonate and Opal Export Fluxes
 The magnitude of calcium carbonate and opal export flux is estimated following Sarmiento et al. [2002, 2004a]. The export of opal and organic nitrogen is equal, in steady state, to the uptake of silicate and nitrate, which in turn is supplied by transport. In cases where there are no large horizontal gradients, the export ratio of opal to organic nitrogen is represented by the vertical gradients in the nutrients
where the subscript s refers to the annual mean tracer concentration in the depth range 0–100 m and d to the annual mean tracer concentration in the range 100–200 m. The ndenotes salinity-normalized tracer,nN = 35.(N/S), where S is salinity. Global climatological fields of N, Si and S were taken from the World Ocean Atlas 2005 [Antonov et al., 2006; Garcia et al., 2006]. The total flux of opal is obtained by multiplying the right hand side of equation (1) by: Fnitrate = EP.RCorg:POC. RN:C where RCorg:POC is the ratio of total organic carbon export to POC export. Here we use the global mean value of 1.25 suggested by Hansell and Carlson . RN:C is the stoichiometric ratio of nitrate to carbon for which we use the Redfield ratio of 116:16. Similar to opal export, the export of CaCO3 and organic carbon is estimated from the supply of potential alkalinity and nitrate
where Palk is potential alkalinity taken from the GLODAP program [Key et al., 2004] (http://cdiac3.ornl.gov). The total flux of CaCO3 is obtained by multiplying the right hand side of equation (2) by: FCorg = EP.RCorg:POC.
3.1. Comparison of Satellite-Derived and Data-Based Efficiency Estimates
 The satellite-derived estimates of EP,2000FCorg, PEeff, Teff and Martin's b are compared to estimates based solely on in situ measurements in Figure 3. In order to make them directly comparable, the satellite-derived estimates are calculated on the same time scales as the in situ data, i.e., thorium-derived particle export represents the integrated export over ∼2 weeks and the2000FCorg is reported as the annual total flux. The results are presented as a function of latitude to more readily identify large scale spatial patterns.
 The broad geographical patterns in POC export (Figure 3a) are similar in the data and satellite-derived estimates, with higher export at high latitudes and minima in the low latitudes. The satellite-derived estimates lie around the mean of the in situ data, but show lower spread than the data at the same latitude. The latitudinal distribution of2000FCorgis also very similar in the trap data and the satellite-derived estimates (Figure 3b). Minima in 2000FCorg occur in the Southern Ocean and high latitude Northern Hemisphere. At low latitudes, there is a high degree of variability, with 2000FCorgvarying by up to eightfold in the Arabian Sea (∼0–20°N) in both satellite-derived and in situ estimates. Peak values of2000FCorgoccur in the Arabian Sea and the Benguela and California upwelling regions. The satellite-derived2000FCorgcaptures the large variance evident in the in situ data. The particle export efficiency has a distinct latitudinal pattern in both the data and the satellite-derived estimates (Figure 3c), with higher PEeff at high latitudes decreasing toward the equator. The exception is a few relatively high values of PEeff near the equator that stem from the Galapagos Current region [Murray et al., 1989]. Note that the use of instantaneous PP to estimate the data-basedPEeff sometimes results in ratios >1, i.e., EP exceeds PP, because of the integration time scale of 234Th. The Teff shows the opposite latitudinal pattern as the PEeff, with low Teff at high latitudes and vice versa (Figure 3d). The latitudinal distribution of Martin's b parameter shows a similar distribution as Teff with lowest values in the Southern Ocean and a peak in low latitudes (Figure 3e).
 The limited number of locations where particle export and 2000FCorghave been spatially coincidentally measured in situ makes comparison with the satellite-derivedTeff and bestimates difficult, but generally speaking the patterns are similar in both. Disparities between the data-based and satellite-derived estimates occur in the Southern Ocean and northern hemisphere high latitudes, where the satellite-based estimates underestimateTeff. This may arise because of the difference in time and space scales over which the in situ and satellite data are collected. The in situ data are point measurements made in a spatially complex environment, whereas the satellite data is averaged spatially over a larger area (3° × 3°). Additionally, the data-basedTeff is the ratio of export integrated over the residence time of the 234Th tracer to the annually integrated 2000FCorg converted to a daily rate simply by dividing by 365, i.e., assuming no seasonality in flux. This will give a fair estimate of Teff if POC export is invariant throughout the year, but not in regions of strong seasonal variability. Finally, the export and 2000FCorg measurements were not usually collected in the same year and none of the 2000FCorgdata overlaps in time with the satellite era (here a climatology of 1998–2007 is used) introducing uncertainty in both the data-based and satellite-derived estimates in regions of strong interannual variability.
 Overall, the agreement between latitudinal patterns in data-based and satellite-derived export efficiency and transfer efficiency is good. This demonstrates a robust broad geographical pattern of highPEeff but low Teff at high latitudes, and low PEeff but high Teff at low latitudes.
3.2. Global Estimates of Export Efficiency and Transfer Efficiency
 The agreement between the satellite-derived and data-based estimates ofPEeff and Teffsuggest that we can use satellite-derived data to examine global-scale spatial patterns in both export and transfer efficiencies. The satellite-derived global maps of export production, POC flux to 2000 m, export efficiency, transfer efficiency and Martin'sb parameter are plotted in Figure 4. In contrast to Figure 3, the maps in Figure 4 are calculated using total annual PP, EP or 2000FCorg as relevant. The global maps confirm the broad spatial patterns observed in the limited, spatially patchy, in situ data (Figure 3).
 The global map of particle export (Figure 4a) shows highest rates (∼20–35 gC m−2 yr−1) in the high latitude Northern Hemisphere, in shelf areas, the Arabian Sea and in upwelling regions. Particle export in the Southern Ocean is mostly moderate (∼10–20 gC m−2 yr−1) with increased export downstream of continental landmasses and islands. At low latitudes, particle export is generally very low (∼1–10 gC m−2 yr−1), with slightly higher levels of export in equatorial upwelling regions. The globally integrated total particulate carbon export is 4 Gt C yr−1 (as with Henson et al. ). Contributions to export flux also come from the export of dissolved organic carbon (DOC), which, as a global average, contributes ∼20% to export in the open ocean [Hansell and Carlson, 1998]. Adding in this contribution brings our estimate of globally integrated export to 5 (±2.2) GtC yr−1.
 The flux of POC at 2000 m depth is presented in Figure 4b and displays some of the same geographical patterns as particle export. 2000FCorg is ∼2.5 gC m−2 yr−1 in Northern Hemisphere high latitudes, lower than the 2000FCorg estimated for equatorial upwelling areas and the Arabian Sea (∼3–4 gC m−2 yr−1). The oligotrophic gyres and Southern Ocean have very low 2000FCorg of ∼0.5 gC m−2 yr−1. The globally integrated total POC flux to 2000 m depth is 0.66 (±0.3) Gt C yr−1.
 In Figure 4c, the global map of PEeffshows highest values in the Southern Ocean and sub-Arctic regions (∼0.3), indicating that ∼30% of PP is exported out of the euphotic zone. The low latitudes have lowPEeff (∼0.01–0.1), including the equatorial and upwelling regions which have higher particle export than neighboring oligotrophic gyres (Figure 4a). The high latitude Northern Hemisphere has moderate PEeff of ∼0.15, with larger values in some coastal regions. The global mean particle export efficiency is 0.1, i.e., 10% of PP is exported below the euphotic zone.
 The global distribution of Teff (Figure 4d) has very low values in the Southern Ocean (∼0.05), i.e., only 5% of organic carbon exported from the euphotic zone survive to 2000 m depth. Moderate values (∼0.1–0.2) are found in midlatitudes and North Hemisphere high latitudes. Highest Teff (∼0.25–0.4) occurs in equatorial upwelling regions, the Indian Ocean and western equatorial Pacific. The global mean transfer efficiency is 0.19, i.e., 19% of exported organic matter survives dissolution to reach 2000 m depth.
 The global map of Martin's b parameter (Figure 4e) shows similar patterns to Teff, where more negative values correspond with greater remineralization of material in the water column. Lowest values of b(∼−1 to −1.2) occur in the Southern Ocean, with moderate values (∼−0.6 to −0.8) in the midlatitudes and North Hemisphere high latitudes and high values (∼−0.3 to −0.5) in the sub-tropics. Theb parameter as estimated here is clearly spatially heterogeneous. Values of b of ∼−0.858 (the original and widely used value of Martin et al. ) occur only at midlatitudes. The range of b globally is from ∼−1.18 to −0.24, with a mean of −0.639.
3.3. Patterns of Ecosystem Structure
 Both the global satellite-derived maps and limited in situ data suggest opposing geographical distributions of export efficiency and transfer efficiency. These are captured inFigures 5 and 6, which show scatterplots of PP against export and export against POC flux at 2000 m, respectively. The relationship between PEeff and latitude is clear in Figure 5a. The Southern Ocean has highest values, with moderate PEeff in the high latitude Northern Hemisphere and low values at low latitudes. Although more scattered than the corresponding PEeff plot, Figure 6a, which shows POC export against 2000FCorg, indicates that low transfer efficiency is found at high latitudes, both in the Northern and Southern Hemispheres, with higher Teff at low latitudes.
 Some of the latitudinal differences in physico-chemical conditions between high and low latitudes are described by the ‘biological pump efficiency’ (in the sense ofSarmiento et al. [2004b]). This describes the amount of upper ocean nitrate consumed during the year, as a proportion of the total nitrate available: ([NO3−]deep − [NO3−]surface)/[NO3−]deep where surface and deep nitrate are the annual means in the 0–100 m and 100–200 m range, respectively (nitrate from the World Ocean Atlas 2005 [Garcia et al., 2006]). Figures 5b and 6billustrate the interaction between this term and the export and transfer efficiencies. Large values of the ‘biological pump efficiency’ indicate that most available nitrate is consumed – implying a combination of low initial nutrient levels and near-continuous stratification, conditions typical of low latitudes where regenerated nitrate fuels much of the PP. InFigure 5b, this coincides with low export efficiency (∼2–10%). Low values of ‘biological pump efficiency’ indicate that only a small fraction of available nutrients are consumed – suggesting high initial nutrient levels, seasonal mixing and restratification resupplying nitrate, and/or other growth limiting factors (e.g., iron limitation). These conditions are typical of high latitudes where new nitrate fuels most PP and coincide in Figure 5b with high export efficiency. A similarly strong correspondence is not observed between ‘biological pump efficiency’ and transfer efficiency (Figure 6b), although the general pattern is that low ‘biological pump efficiency’ corresponds to low transfer efficiency, and vice versa.
 These differences in nutrient supply and physical conditions are reflected in the community structure of the primary producers. Typically, diatoms dominate in low light, unstable, high nutrient environments (i.e., at high latitudes) and comprise only a small component (if present at all) of the phytoplankton community in sub-tropical regions. An estimate of the contribution of diatoms to total chlorophyll was obtained from the NASA Ocean Biogeochemistry Model (NOBM). The NOBM is a biogeochemical model that assimilates satellite-derived chlorophyll data, and explicitly models the abundances of 4 phytoplankton groups as contributions to total chlorophyll: diatoms, coccolithophores, chlorophytes and cyanobacteria (full details of the model parameterization are given byGregg and Casey  and Gregg et al. ). Figures 5c and 6c show how this varies with export efficiency and transfer efficiency. Where diatoms dominate the phytoplankton community, export efficiency is generally high, and vice versa. Again the correspondence is not as strong for transfer efficiency, with high diatom abundance corresponding to a range of Teff values.
 The global maps of calcium carbonate and opal export flux are presented in Figure 7. The export flux of opal has distinct regions of high opal export (>0.5 mol Si m−2 yr−1) in the North Pacific and throughout the Southern Ocean. In contrast, CaCO3 export occurs throughout the world ocean, with the exception of the Southern Ocean. The global mean PIC:POC ratio is 0.05, consistent with estimates by Sarmiento et al.  and Balch et al. . An upper bound estimate of the contribution of diatoms to total export, using a RSi:N of 1:1 (indicative of a ‘healthy’ diatom [Brzezinski et al., 1998]), is 4 Gt C yr−1, i.e., 80% of total organic carbon export. However, the silica:nitrogen ratio of diatoms is unlikely to be 1 everywhere [e.g., Takeda, 1998]. A lower bound estimate, using a RSi:N of 4:1 (as found in the iron limited Southern Ocean [Brzezinski et al., 2003]), is 1 Gt C yr−1, 20% of total organic carbon export, i.e., diatoms may contribute between 20 and 80% of export and thus potentially dominate the flux of organic matter. In keeping with these estimates, use of a spatially explicit RSi:N field, derived from a global biogeochemical model, suggests that diatoms contribute 43% of total organic carbon export [Jin et al., 2006].
4.1. Spatial Gradients in Export Efficiency and Transfer Efficiency
 The spatial gradients in PEeff and Teffimply that in high latitudes, relatively more carbon generated by primary production is exported below the mixed layer than in low latitudes, but that this material is not effectively sequestered below 2000 m. In high latitude regions, export efficiency is relatively high, indicating that ∼15–25% of the particulate organic matter produced by PP is exported below the euphotic layer. This suggests that relatively little (compared to low latitudes) remineralization occurs in the surface waters and that most of the organic carbon is associated with dense, fast-sinking particles. However, the transfer efficiency at high latitudes is low, indicating that little of the material that is exported (∼1–10%) reaches 2000 m depth, i.e., the majority is remineralized in the mesopelagic zone. Note that2000FCorg is still higher in, for example, the North Atlantic compared to the oligotrophic gyres, but it is only a small percentage of the total exported organic carbon, resulting in a low Teff. Conversely, at low latitudes only a small proportion (∼1–5%) of particulate organic matter produced by PP is exported and the majority is remineralized within the upper ocean (<100 m). However, ∼20–35% of the material that is finally exported survives to a depth of 2000 m. This suggests that more (relative to high latitudes) of the particulate matter exported at low latitudes is refractory and undergoes little further remineralization at mesopelagic depths.
 Previous studies of regional differences in transfer efficiency based on sediment trap records suggest similar patterns as observed here. For example, Antia et al.  concluded that POC fluxes to >3000 m are fairly constant in the North Atlantic between 33 and 55°N along the 20°W meridian (cf. Figure 4b). However, they also found increasing surface export flux at higher latitudes, implying decreasing Teff with increasing latitude. Consistent with our estimates, a compilation of JGOFS data by Berelson  suggested that b is larger (i.e., Teff is greater) in the equatorial Pacific and Arabian Sea than in the Southern Ocean or North Atlantic. Francois et al.  also found similar broad geographical differences in Teffbetween studies carried out in the Southern Ocean (0.02 at the AESOPS site), the productive Northwest Pacific (0.04 at KNOT), the oligotrophic Northwest Pacific (0.07 at CP-6) and the Arabian Sea (0.12 at AS-3). These spatial differences are consistent with those displayed in our globally resolved maps ofTeff (Figure 4d). The degree of spatial variability exhibited in our estimate of b (Figure 4e) belies the assumption often made in global ocean biogeochemical models of a constant remineralization length scale (usually, b= −0.858) to estimate deep POC flux (e.g., in all the OCMIP-2 model runs [Doney et al., 2004]). Our results confirm that a constant b value or remineralization length scale is not appropriate to represent deep carbon fluxes in global biogeochemical models.
 Our result of low Teff associated with high PEeff at high latitudes, and vice versa, is consistent with the pattern proposed by Francois et al. . Their conclusion was based on the Laws et al.  export algorithm, which relates f-ratio to SST. Recently however, the estimation of export production using thef-ratio has been called into question due to the recent recognition of the importance of upper ocean nitrification [Yool et al., 2007]. The Henson et al. estimate of export used here bypasses this issue because of the use of direct particle export measurements. As the estimate of global carbon export derived from thorium-based export estimates is lower than that based on other methodologies, it is therefore reassuring that we come to the same conclusion asFrancois et al.  regarding the spatial patterns of transfer efficiency.
4.2. Comparison of Particle and Transfer Efficiencies With Ecosystem Structure
 The comparison of PEeff with indices of ecosystem structure suggest some correlation with relative diatom abundance and ‘biological pump efficiency’ in the sense of Sarmiento et al. [2004b]. Regions where the available nitrate is not fully consumed (i.e., low ‘biological pump efficiency’) are associated with high export efficiency but low transfer efficiency and high relative diatom abundance (Figures 5b and 5c). Not only do these regions support high rates of new production, relative to regenerated production (thus promoting high PEeff), but the characteristics of diatoms themselves also contribute to high export efficiency. Some diatoms are relatively large and dense and so may sink rapidly enough to avoid significant remineralization in the surface ocean [Martin et al., 2011]. However, Teff is low in these regions suggesting that the characteristics of the exported particles make them prone to relatively rapid remineralization.
 Conversely, in low latitudes, new production is low, principally due to a lack of mixing to supply nutrients, and instead PP is supported by regenerated nitrogen resulting in low PEeff. Diatoms tend not to thrive in these conditions and comprise only a small fraction of the total phytoplankton population. The phytoplankton that dominate in these high light, low nutrient regions are generally small, and thus sink very slowly. Much of the organic carbon they contain is therefore remineralized in surface waters and not exported. These regions do have high Teff, however, so the particles that are exported must have characteristics that make them relatively resistant to remineralization.
 Regional studies have previously demonstrated the importance of the proportion of diatoms to determining the particle flux from the surface ocean [Boyd and Newton, 1995, 1999; Buesseler, 1998]. Correlations between diatom abundance and thorium-derived POC flux have been noted in the Arabian Sea [Ducklow et al., 2001] and Baffin Bay [Amiel et al., 2002]. In the North Pacific, a correspondence between stations that experienced greater diatom abundances and reduced particle remineralization below the mixed layer, i.e., higher export efficiency, was noted [Buesseler et al., 2009]. Two studies conducted in the Southern Ocean reached similar conclusions, one observing increased export efficiency after the stimulation of a diatom bloom by iron fertilization [Buesseler et al., 2005], and another finding large spatial gradients in export efficiency along a transect from the ice-edge to north of the Polar Front, related to differing relative abundance of diatoms [Buesseler et al., 2003].
 The importance of ecosystem structure in determining Teff was also emphasized by Francois et al. who proposed that a ‘packaging’ effect may be at work. They hypothesized that hydrodynamic, fast-sinking fecal pellets are more likely to be produced in carbonate-dominated regions, versus loose aggregates in opal-dominated regions. The loosely aggregated, slowly sinking detritus produced in opal-dominated regions would thus be more readily remineralized in the mesopelagic, reducingTeff. In the next sections we examine further how the ecosystem structure may govern the transfer efficiency.
4.3. Comparison of Particle and Transfer Efficiencies With Biomineral Fluxes
 The ballast hypothesis is derived from correlations between mineral fluxes and POC captured in deep sediment traps [Armstrong et al., 2002; Klaas and Archer, 2002]. However, if the hypothesized mechanism for this correlation is correct, that is that CaCO3 protects POC against degradation or increases the density of sinking particles, then the association of CaCO3 with the POC flux must occur prior to, or rapidly after, export, i.e., in the upper ocean. Thus, if a similar correlation between CaCO3 and Teff was found in upper ocean export fluxes, this would support the key tenet of the ballast hypothesis.
 A multiple linear regression of the mineral components of the deep flux against the transfer efficiency was used by Francois et al.  to develop the ballast hypothesis. We performed a series of similar analyses using the upper ocean export of CaCO3 and opal. We first compare POC flux reaching 2000 m with upper ocean biomineral export; we then correlate the transfer efficiency with the POC:biomineral export ratios; finally with correlate Teff with biomineral export. Francois et al.  and Klaas and Archer  also included a term for lithogenic particles in their analyses, although both studies concluded that lithogenic material played a minor role in setting Teff or 2000FCorg. However, Dunne et al.  concluded that lithogenic material is the second most important mineral phase (after CaCO3) in determining 2000FCorg. As we have no data on upper ocean export of lithogenic particles, we excluded the regions most likely to be influenced by lithogenic input from our analysis, i.e., coastal and inland sea regions. The predictor variables were randomly sub-sampled from the global maps to create a data set of ∼300 points and, prior to regression, were normalized so that the magnitude of the regression coefficients can be directly compared. The random sub-sampling and regression were repeated 1000 times with replacement to yield error estimates on the coefficients.
 The multiple linear regression of 2000FCorg against upper ocean biomineral export yields (r = 0.72, p < 0.01)
This suggests that the magnitude of POC flux reaching 2000 m depth is higher in regions of high CaCO3 export. However, performing a similar analysis of Teff against the POC:biomineral ratio yields r = 0.70, p < 0.01)
The regression coefficient for opal export is larger than for CaCO3 export and repeating the analysis with only the opal export term results in almost no decrease in explanatory power (r = 0.68, p < 0.01). Similarly, for the regression of Teff against biomineral export (r = 0.68, p < 0.01)
Again, the regression coefficient for opal export is larger than for CaCO3 export and repeating the analysis with only the opal export term results in almost no decrease in explanatory power (r = 0.67, p < 0.01). An improved regression is obtained by fitting an exponential function (r = 0.78, p < 0.01)
where the opal export is in mol Si m−2 yr−1. Including the CaCO3 term has no effect on the explanatory power of this regression. These analyses suggest that the upper ocean export of CaCO3 has very little influence on Teff, but opal export is strongly negatively correlated with transfer efficiency. Although the magnitude of flux reaching the deep ocean is greater in regions of CaCO3 export, association of POC with CaCO3 in the upper ocean does not increase transfer efficiency, i.e., our results imply that ballasting of POC by CaCO3 in the upper ocean does not decrease the remineralization rate of particles sinking through the mesopelagic zone.
 This conclusion complements the ballast hypothesis as formulated by Francois et al. , which found a positive correlation between deep CaCO3 flux and Teff, by highlighting a different perspective on the controls on Teff. In the next section, we contrast the results obtained here for the upper ocean biomineral flux with previously published observations of the deep flux and consider what this may tell us about controls on transfer efficiency.
4.4. Large-Scale Controls on Transfer Efficiency
 Our results suggest that particle export efficiency is high, but transfer efficiency is low, at high latitudes. In these regions, conditions are suitable for dominance of the phytoplankton community by diatoms (Figure 5c). In low latitude regions, particle export efficiency is low, but transfer efficiency is high. Low nutrient levels and stratified conditions ensure that these regions are unsuitable for supporting large diatom populations, confirmed by the map of opal export flux (Figure 7b). We find a negative correlation between upper ocean opal export flux and Teff, but no correlation between Teff and upper ocean CaCO3 export flux. This is because CaCO3 export is roughly homogenous (with the exception of the Southern Ocean), whereas opal export only occurs in regions where the phytoplankton community is dominated by diatoms. Therefore, in low latitudes, high Teff is not quantitatively associated with the presence of CaCO3 export flux, but rather the absence of opal export flux.
 One could argue that this supports the ballast hypothesis by consideration of the different remineralization length scales of opal and CaCO3. Dissolution of biogenic opal can begin as shallow as 100 m [e.g., Tréguer et al., 1995]. Indeed, observations in the Gulf of Alaska suggest that up to 50% dissolution of diatoms occurred within the top 200 m of the water column [Boyd et al., 2004]. Abiotic dissolution of carbonate does not occur above the lysocline, typically at ∼2000 m depth. This implies that any protection against degradation provided by association of POC with opal is removed at fairly shallow depths, increasing the exposure of POC to remineralization processes occurring in the mesopelagic, and resulting in relatively low transfer efficiency. The association of POC with CaCO3 provides protection against degradation below the depth where many of the processes that act to remineralize POC occur (i.e., the mesopelagic). This results in relatively more of the exported POC reaching the deep ocean, and hence high Teff, consistent with the positive relationship between transfer efficiency and the proportion of CaCO3 collected in deep traps found by Francois et al. . The difference in remineralization length scale of opal and CaCO3 also implies that POC associated with opal dominates the supply of carbon to the mesopelagic, while POC associated with CaCO3 dominates the transfer of carbon to the abyssal ocean.
 However, although one could argue that the presence of opal in high latitude systems results in low Teff, the absence of opal in low latitude systems is not the cause of the high Teff observed in these regions. Instead, we suggest that the key characteristic that determines the spatial differences in Teff is the ecosystem structure and specifically the extent of recycling of organic carbon in the upper ocean. High Teff in low latitudes is associated with low PEeff, suggesting extensive recycling of organic matter in the euphotic zone. The material that is finally exported after having been processed multiple times in the euphotic zone is likely to be highly refractory and to undergo relatively little further remineralization in the mesopelagic, resulting in high transfer efficiency. Conversely, at high latitudes PEeff is high implying that much of the organic carbon generated through primary production is exported. However, the low Teff values suggest that this material is relatively labile and fresh and is readily remineralized in the mesopelagic zone. This conclusion is consistent with the hypothesis of Francois et al. that fast-sinking pellets dominate POC flux at low latitudes, while loose aggregates are predominant at high latitudes. We argue on the basis of the results presented here that it is the ecosystem structure, rather than a ballasting effect, which is the dominant control on transfer efficiency.
4.5. Implications for the Ballast Hypothesis
 Our conclusions suggest that upper ocean ballasting is not the ultimate factor determining transfer efficiency. Evidence from other work based on in situ studies is inconclusive. Support for an upper ocean ballast effect came from a study in the Atlantic that measured POC and biominerals at multiple depths in the euphotic zone [Sanders et al., 2010]. Positive correlations between POC export and both CaCO3 and opal export were found, suggesting that both minerals have a ballasting effect. However, this conclusion was later revised after additional measurements of the slow sinking fraction of POC revealed that it was unballasted by either CaCO3 or opal [Riley et al., 2012]. An earlier study of the upper ocean mineral fluxes and POC export found no evidence for a ballast effect in particles exported from the surface [Lam and Bishop, 2007]. Based on comprehensive measurements in the mesopelagic at two study sites in the Southern Ocean, Lam and Bishop  concluded that there was no direct effect of mineral composition on transfer efficiency. Similarly, observations at K2 in the North Pacific and Station ALOHA suggested that biominerals were not present in higher proportions in the fast sinking fraction of the export flux, although insufficient data to completely rule out an upper ocean ballasting effect were collected [Trull et al., 2008]. Mixed evidence for an upper ocean ballast effect comes from data collected during an Atlantic Meridional Transect cruise that suggests export efficiency is correlated with opal export, but no enhancement of POC flux with high calcite export occurs [Thomalla et al., 2008]. The authors concluded that the relationship between export and opal flux was strongest in high export regions, consistent with our conclusion that regions of high export efficiency coincide with high opal export flux. Finally, a study based on a global compilation of data suggested that a correlation between mesopelagic POC flux and CaCO3 does exist, but the authors conclude that ecosystem structure is the dominant factor in setting the Teff [Lam et al., 2011]. The authors reach a conclusion consistent with ours, that regions of low Teff are those dominated by diatoms, whereas regions of high Teff are dominated by small cells.
 The interpretation that the correlation between deep CaCO3 flux and POC [Francois et al., 2002; Klaas and Archer, 2002] confirms the role of ballasting in setting Teff has been questioned. An alternative explanation for the correlation between POC and minerals may be a result of POC itself providing a glue to aggregate minerals which otherwise would not sink [Passow, 2004; Passow and De La Rocha, 2006; De La Rocha et al., 2008]. In this scenario, deep waters contain minerals which are insufficiently dense to sink and remain suspended until they are scavenged by sinking aggregates [Brzezinski and Nelson, 1995; Passow et al., 2001]. This is consistent with Thomalla et al. , who concluded that the potential for effective ballasting of particle flux increases with depth. Rather than originating from coccolithophore production in the upper ocean, Lam and Bishop  instead suggested that the correlation between CaCO3 and POC observed in deep traps arises from the foraminiferal contribution to CaCO3. The authors suggest that biological processes which affect the fragmentation and remineralization of large particles, i.e., particle packaging and (dis)aggregation mediated by zooplankton grazing, are the most important factors affecting transfer efficiency in the mesopelagic. Finally, the observed correlation between deep CaCO3 flux and transfer efficiency may not be causative. As exported CaCO3 particles occur almost everywhere (Figure 7a), a larger fraction of the exported organic matter may become associated with CaCO3than with opal, which only occurs in specific regions. Another, non-causative explanation is that in low latitudes, the exported material is relatively refractory, having been processed many times in the upper ocean, while the material exported in high latitudes is relatively fresh and labile. This could result in highTeff at low latitudes, without the need for invoking a ballasting effect. Our analysis supports this last interpretation.
 A corollary of the ballast hypothesis is that, under future climate change scenarios with increased atmospheric CO2, the predicted shoaling of the lysocline [Tyrrell, 2008] will result in decreasing POC flux to depth, providing a positive feedback to increasing CO2 concentration. If however, the ecosystem structure is more important in setting Teff, then the projected global warming-driven changes in stratification, nutrient supply, temperature etc. will strongly impact efficiency of POC transfer to depth. Increased stratification (and hence reduced nutrient supply) will promote expansion of the subtropical gyres, i.e., recycling-dominated systems [Bopp et al., 2005; Henson et al., 2010; Sarmiento et al., 2004c], which would result in reduced total carbon export, but larger regions of high transfer efficiency. Currently, at high latitudes, Teffis low and therefore substantial remineralization occurs at relatively shallow depths, but if these regions shift toward a more subtropical-like, recycling-driven system in the future,Teff may increase, delivering carbon into deeper water masses. This could result in longer storage times, i.e., carbon would be sequestered out of contact with the atmosphere for longer. Indeed, modeling studies have shown [e.g., Kwon et al., 2009] that a modest increase in remineralization depth results in a substantial reduction of atmospheric CO2, due to the redistribution of remineralized carbon from intermediate to bottom waters.
 Compilations of hundreds of in situ measurements of thorium-derived particle export and POC flux to 2000 m [Honjo et al., 2008] enabled us to undertake an inter-comparison of several satellite data-derived algorithms for export production and deep POC flux. The combination of algorithms best able to reproduce the in situ data were Carr (for PP) + Henson for export flux [Carr, 2002; Henson et al., 2011] and Carr + Lutz for POC flux [Lutz et al., 2007]. These were applied to global satellite data fields to produce maps of export efficiency, transfer efficiency and Martin's b parameter. The distinct spatial patterns revealed by the global maps allowed us to examine two contrasting hypotheses about the processes controlling transfer efficiency: ballasting by biominerals or ecosystem structure.
 In order for a ballast effect to provide protection against degradation, the association of organic material with biominerals must happen in the upper ocean. Our analysis finds a strongly negative correlation between transfer efficiency and upper ocean opal export, consistent with Francois et al.  who concluded that opal may provide little protection from degradation (in comparison to CaCO3). However, CaCO3 export flux is not positively correlated with Teff, suggesting that association of POC with CaCO3 in the upper ocean does not reduce degradation via a ballasting effect at mesopelagic depths.
 At high latitudes, where diatoms often dominate the phytoplankton community structure, export efficiency is high. However, transfer efficiency is low, implying that the organic matter exported by these systems is relatively labile and is prone to remineralization in the upper mesopelagic zone. In low latitude regions, where an effective microbial loop ensures that much organic matter is recycled before it is exported, export efficiency is low. However, transfer efficiency is high, implying that the small fraction of primary production that is finally exported is refractory and undergoes comparatively less degradation at mesopelagic depths. On the basis of these observations, we surmise that ecosystem structure, rather than the availability of CaCO3 for ballast material, is the key factor controlling the efficiency of the biological carbon pump.
 SeaWiFS data were provided by GSFC/NASA in accordance with the SeaWiFS Research Data Use Terms and Conditions Agreement. This work was supported by NERC grant NE/G013055/1 to SAH.