Data compilation of ﬂuxes of sedimenting material from sediment traps in the Atlantic Ocean

. We provide a data set assemblage of directly observed and derived ﬂuxes of sedimenting material (total mass, POC, PON, bSiO 2 , CaCO 3 , PIC and lithogenic / terrigenous ﬂuxes) obtained using sediment traps. This data assemblage contains over 5900 data points distributed across the Atlantic, from the Arctic Ocean to the Southern Ocean. Data from the Mediterranean Sea are also included. Data were compiled from a variety of sources: data repositories (e.g. BCO-DMO, PANGAEA ® ), time-series sites (e.g. BATS, CARIACO), published scientiﬁc papers and data provided by the originating principal investigators (PIs). All sources are speciﬁed within the combined data set. Data from the World Ocean Atlas 2009 were extracted to coincide with ﬂux data to provide additional environmental information where available. Speciﬁcally, contemporaneous data were extracted for temperature, salinity, oxygen (concentration, AOU and percentage saturation), nitrate, phosphate and silicate. Data show a broad range of ﬂux estimates, with marked di ﬀ erences between ocean domains. Data erences total mass ﬂux, which is relevant towards understanding the factors that control the of the biological This data set has been submitted to the data repository PANGAEA ® (http: // www.pangaea.de), who have made it available under doi: 10.1594 / PANGAEA.807946.


Introduction
The export of particulate organic carbon (POC) from the sunlit upper layers to the ocean interior (deep waters and deep ocean sediments), known as the biological carbon pump (BCP), is an important component of the global carbon cycle. However, the BCP is not well constrained, with estimates of carbon transport ranging from 3.4-4.7 (Eppley and Peterson, 1979) to ∼ 20 Gt C yr −1 depending on the method used (e.g. Laws et al., 2000). In the North Atlantic alone, estimates of the BCP range fourfold from 0.55 to 1.94 Gt C yr −1 , representing 10-20 % of global export fluxes (Sanders et al., 2014). The uncertainty in these estimates reflects the sparseness of observations and to a smaller extent the variety of methods employed.
Currently, several independently compiled data sets of export flux exist but they reside in separate data repositories as individual data sets. Given the importance of the BCP for carbon storage and the need to further constrain its magnitude, we have attempted to bring together into a single compilation, all data currently available on POC fluxes (and related from the originating PIs. All sources are specified within the data set.

Data
Observational studies of export often make use of sediment traps. These are deployed at depths of interest in order to collect particles sinking through the water column. Upon recovery, particles are then analysed primarily to determine their carbon content, but additionally the nitrogen content is also easily obtained. The quantities of ballasting material such as calcium carbonate and opal are also increasingly important variables. The flux rate of material captured by a trap can be calculated from knowledge of the duration a trap is deployed for and the aperture of the trap itself. By obtaining export fluxes at different depths, the efficiency of the carbon pump can be evaluated simply as the fraction of the flux leaving the surface that makes it to a given depth.

Data sources
The data assemblage presented here includes variables commonly measured in studies concerning export production. These include total mass (Tot_Mass) flux, POC flux, particulate organic nitrogen (PON) flux, biogenic silica (bSiO 2 ) flux, calcium carbonate (CaCO 3 ) flux, particulate inorganic carbon (PIC) flux, and terrigenous or lithogenic (Terr/Litho) material flux. Most data were obtained from data repositories and individual time-series websites. A total of 2679 data points (45 % of total) were derived from 32 smaller data sets obtained from the Data Publisher for Earth and Environmental Science (PANGAEA ® ) at http://www.pangaea.de; 111 data points (1.9 %) were obtained from the Biological and Chemical Oceanography Data Management Office (BCO-DMO) at http://bcodmo.org (Lee et al., 2009a, b); 1755 data points (29.5 %) were obtained from the Carbon Retention In A Colored Ocean Project Ocean Time Series (CARIACO) at http://www.imars.usf.edu/CAR (Montes et al., 2012); and 784 data points (13.2 %) from the Bermuda Atlantic Time Series (BATS) at http://bats.bios.edu. Data (428 data points, 7.2 %) were also obtained from published journal articles (e.g. Bory et al., 2001;Hwang et al., 2009). Finally, a few data sets (190 data points, 3.2 %) were directly obtained from Principal Investigators (Bauerfeind et al., 2009;Lampitt et al., 2010). These combined data sets provide 5947 data points of variables related to export production. In this compilation, a "data point" is an independent measurement or calculation of a given variable. The full data set is organised in columns as indicated in Table 1. The names and acronyms of variables as presented in this table are the most commonly used and least ambiguous terms. Though not all individual data sets contain all the variables listed in Table 1. The total number of data points per variable are presented in Table 2. POC flux contains the largest number of observations (5206 data points) and PIC flux the lowest (1048 data points).

Methods commonly employed
Across data sets there is considerable inconsistency in the names of variables and reported units. For this compilation we have attempted, to the best of our knowledge, to identify the variable that was actually being dealt with. For instance "biogenic particulate silicon" vs. "biogenic particulate silica", with the first referring to the element silicon (Si) and the second referring to the mineral silica (SiO 2 ). Upon careful examination of such cases, when enough information was available to identify the correct variable, that variable was standardised so that it is internally consistent within this compilation (e.g. all Si related fluxes are consistently reported as SiO 2 ). In cases where available information was insufficient to clarify ambiguity, data sets were excluded from the compilation.

Sample collection procedure
Before deployment, the collecting cups of sediment traps are filled with ambient seawater. NaCl is typically added to increase the salinity to 40 (Antia et al., 1999; Bory and Newton, 2000;Fischer et al., 2002;Fahl and Nöthig, 2007;Neuer et al., 1997). Sufficient formalin to yield 2-3 % formaldehyde (wt/vol) or mercuric chloride (0.14 % final solution) is commonly added to poison the sample to preserve the content (Antia et al., 1999;Fischer et al., 2002;Fahl and Nöthig, 2007;Helmke et al., 2005;Bauerfeind et al., 2009). Following recovery of sediment traps, swimmers (i.e. zooplankton that feed on sedimenting material) are identified and removed from collecting cups (Antia et al., 1999;Bory et al., 2001;Lampitt et al., 2010). Sometimes the samples are sieved (1 mm mesh) to remove large swimmers (e.g. . Also, samples are sometimes centrifuged following the removal of swimmers and the supernatant is then analysed in order to take into account of any possible dissolution of the material collected (e.g. Waniek et al., 2005). Samples from trap cups are typically split to generate subsamples for the different types of analysis and filtered through preweighed filters which are rinsed with ammonium formate to remove salt and excess formalin (e.g. Bory et al., 2001). The reader is referred to the source references for details of a particular deployment.

Total mass flux
Total mass is obtained by weighing the dried matter collected on a filter. As such, it is sometimes referred to as "dry mass". Total mass flux (Tot_Mass flux , mg m −2 d −1 ) is calculated as where M w is the mass dry weight (mg), F w is the filter weight (mg), T is the deployment time (days), and A is the aperture trap area (m 2 ) (e.g. Bahr et al., 1997).
As a term, POC is frequently used interchangeably with "organic carbon (C org )" (e.g. Fischer et al., 1996;Lampitt et al., 2001) or total organic carbon (Wefer and Fischer, 1991;Jonkers et al., 2010). POC, is sometimes estimated as C org = C total − C CaCO 3 (e.g. Romero et al., 2002), where C total is the total carbon content of a sample and C CaCO 3 is the carbon content in calcium carbonate. Similarly, PON is sometimes referred to as total nitrogen (e.g. Lampitt and Antia, 1997;Jonkers et al., 2010). We suggest POC and PON are the most appropriate terms by reasons of method and most common usage in literature.

Calcium carbonate flux and particulate inorganic Carbon (PIC) flux
CaCO 3 and PIC fluxes are typically derived based on molar mass ratios. CaCO 3 has been estimated by multiplying PIC by 8.34  or by 8.33 Fischer et al., 2002); i.e. CaCO 3 C ≈ 8.33 (though this is not usually explicitly stated). It has also been calculated as (C total −C org ) × 8.33 (e.g. Wefer and Fischer, 1991;Helmke et al., 2005) following CHN analysis; that is, PIC × CaCO 3 C . It has been also determined through mass loss following acidification and then weighing (e.g. Fahl and Nöthig, 2007). Hwang et al. (2009) refer to "biogenic CaCO 3 ", which they estimated by multiplying the "biogenic Ca" by 2.5; i.e. the molar mass ratio CaCO 3 Ca . In turn, they obtained "biogenic Ca" as the difference between total Ca and lithogenic Ca. The latter being 0.5 × Al, based on the ratio of Ca to Al of the average continental crust composition (e.g. Wedepohl, 1995;Rudnick and Gao, 2003). Total inorganic carbon is determined by coulometric titration (Hwang et al., 2009). CaCO 3 flux is sometimes corrected if organisms containing calcium carbonate, such as pteropods, are present in the sample (e.g. Bauerfeind et al., 2009).
PIC has been calculated as 12 % carbonate by weight (Antia et al., 1999;Bauerfeind et al., 2009), i.e. the C content in CaCO 3 . PIC content has also been calculated from total Ca concentrations in samples as CaCO 3 (Bory et al., 2001). It is also estimated as C total − C org ; i.e. the difference between the C measured in filtered samples without removal of carbonate, and the C measured in samples treated with HCl. The term "inorganic carbon" as equivalent of PIC is sometimes used in the literature too (e.g. Lampitt and Antia, 1997;Lampitt et al., 2001).

Biogenic silica flux
Biogenic silica requires more care relative to other variables. This derives from the fact that multiple names and terms are used rather ambiguously. For this compilation we attempted, to the best of our knowledge, to identify the variable that was being dealt with in each instance. We based our evaluation on the methods used and in some cases by contacting originating PIs.
Biogenic silica is typically measured as dissolved silicon with colorimetric methods following extraction from particulate material. Several methods exists (e.g. Eggimann et al., 1980;DeMaster, 1981;Mortlock and Froelich, 1989;Müller and Schneider, 1993), but the methods most commonly used are based on an the alkaline digestion method of Mortlock and Froelich (1989) (e.g. Antia et al., 1999;Bory et al., 2001;Salter et al., 2010) or the sequential leaching method of De-Master (1981) as modified by Müller and Schneider (1993) (e.g. Fischer, 1991, 1993;Romero et al., 2002;Helmke et al., 2005). The former is based on the extraction of opaline silicon into a 2 M solution of Na 2 CO 3 at 85 • C for 5 h, after which the digested sample is measured with  Thomsen and von Bodungen (2001a, b, c) standard photometric methods using an autoanalyser (Mortlock and Froelich, 1989). The latter is an automated method designed to extract Si from a broader range of compounds.
The extraction is carried out with a 1 M solution of NaOH also at 85 • C, but the digestion solution is cycled from and to a digestion vessel; a proportion runs through an autoanalyser and another fraction is circulated back to the digestion vessel until extraction is completed (Müller and Schneider, 1993). There is an issue however, associated with the use of either method. Since the end product of the extraction is Si, this then needs to be "scaled" back to silica (SiO 2 ). The method by Mortlock and Froelich (1989) uses a factor of 2.4 (the molar mass ratio of SiO 2 ·0.4H 2 O Si ), which accounts for the average water content of diatomaceous silica. The method by Müller and Schneider (1993) instead uses the molar mass ratio of SiO 2 Si ≈ 2.139. Another method used is that of Koning et al. (2002) (e.g. Jonkers et al., 2010, which is based on the method by Müller and Schneider (1993). Given the molar mass ratios adopted, and given that some data sets report biogenic Si rather than biogenic SiO 2 , which we converted using the molar mass ratio of 2.1, we include a column in the data assemblage where the ratio is indicated. We did this based on whether a study or source used either of the methods above, except when a "conversion factor" was explicitly stated independently of the digestion method used. When information provided was "unclear" about the ratio used, this is pointed out.
In some cases, dissolved silica is first measured in the water used for the collecting cups. Dissolved silica is then measured again following the trap's recovery in order to correct for any opal dissolution (e.g. Jonkers et al., 2010). However, we note that not all the biogenic silica data in this compilation includes such a correction or its application to the data was not clear. We have flagged these data as follows: 0 when corrections were made, 1 when corrections were not made, and 2 when information was not available to ascertain either.
Since the ballasting effect of the mineral SiO 2 (opal) is most germane to this database, and since the interest resides in identifying the opal related to particles of biological origin (hence the term "biogenic"), the data we report here is referred to as "biogenic silica" or "biogenic opal". When data was found to be reported as Si, these were converted to SiO 2 using the molar mass ratio SiO 2 Si . We suggest the use of lower-case "b" to refer to "biogenic" in combination with the chemical formula SiO 2 (i.e. bSiO 2 ), since upper-case B is the chemical symbol for the element Boron. We also suggest that whether samples of bSiO 2 are corrected for dissolution or not in the trap-collecting cups should clearly be stated in future studies.

Lithogenic and/or terrigenous material flux
Lithogenic fluxes are typically estimated as the difference between the total mass flux and what is termed either "biogenic flux", "biogenic matter" or "organic matter flux"; i.e. CaCO 3 + POC + BSiO 2 fluxes (e.g. Antia et al., 1999;Bauerfeind et al., 2009). For this purpose, in deriving "organic matter", POC is sometimes multiplied by 2 Bauerfeind et al., 2009) or 2.5 (e.g. Hwang et al., 2009), as this is considered to give a more representative flux of organic matter but, in the literature, this adjustment would benefit from a fuller explanation. Thus, biogenic flux (Bio flux ) is Bio flux = 2×POC flux +CaCO 3flux +Opal flux . Lithogenic flux (Litho) is given by Litho = Tot_Mass flux − Bio Flux . Some researchers estimate the lithogenic material from Al concentrations, under the assumption it contains 8.4 % Al (Bory et al., 2001) or by multiplying the Al concentration by 12.15 (e.g. Hwang et al., 2009). The latter is based on the assumption of a crustal Al composition of 8.2 % (1/12.15 = 0.082).

Data standardisation
This data assemblage contains fluxes from short-duration deployments (hour-days) to longer-duration deployments lasting from months to a year, or over a year. Hence, in order to standardise the data set, all values from long-term deployments, typically reported in grams per square metre per year (g m −2 yr −1 ) (e.g. Fischer, 1991, 1993;Fischer, 2005;Peinert et al., 2001), were converted to daily values, i.e. milligrams per square metre per day (mg m −2 d −1 ), which is the unit most commonly reported. Long-term deployments, however, can be easily identified; a column is provided which specifies the duration of the deployment. A few daily values were reported in grams per square metre per day (g m −2 d −1 ), and these were also converted to milligrams per square metre per day (mg m −2 d −1 ) for consistency. In a few instances, POC, PIC, and bSiO 2 were reported in moles per square metre per year (mol m −2 yr −1 ) (e.g. Antia et al., 1999;Dymond and Lyle, 2003a, b;Honjo and Manganini, 2003d, e, f;Fahl and Nöthig, 2007) or millimole per square metre per day (mmol m −2 d −1 ) (Martin, 2003a, b, c). Again, for consistency, these were converted to milligrams per square metre per day (mg m −2 d −1 ) using the appropriate molecular masses and or molecular mass ratios as required. A column of notes is included, and where unit conversions were done, these are pointed out.

Quality control
Given that the data compiled here derives from research already published, we assume that the originating authors have already undertaken steps necessary to assure data quality. We point out however, that attention should be paid to the fact that the use of slightly different "conversion factors" for a given variable inherently adds error to the data, with up to 20 % in the case of lithogenic flux when derived as the difference between total mass flux and "organic matter" flux, and where "organic matter" is calculated using conversion factors of 2 or 2.5 (Sect. 2.2.7). In the case of bSiO 2 , the error generated is ∼12 % , resulting from the use of 2.1, 2.139 or 2.4 when estimated from Si (Sect. 2.2.6). We did not attempt to "harmonise" the data by using a unique factor for a given variable, since this would involve modifying the data from that found in the original sources. However, in the light of the different deployment durations and traps used, different analytical methods employed, different calculation approaches and different units reported, here we have tried, as best as possible, to put the data together in a manner allowing users to trace original data sources for further scrutiny and so that users can decide how to handle the data further for the specific questions they may choose to tackle.

Ancillary data
Where possible, data from the World Ocean Atlas 2009 (WOA09) were extracted to coincide with flux data to provide additional environmental information (http:// www.nodc.noaa.gov/OC5/WOA09/pr_woa09.html). Specifically, data were extracted for temperature, salinity, oxygen (concentration, AOU and percentage saturation), nitrate, phosphate and silicate. The extraction involves linear interpolation of WOA09 data to the latitude, longitude and depth of the flux data. Each environmental variable is a weighted average over the period of deployment. Note that as WOA09 is a climatology it cannot provide data for specific years. For example, if a mooring collected flux data from 1 November 1990 until 15 January 1991, the WOA data at the relevant point is averaged over the 76 days comprising the annual climatologies for November (for 30 days), December (for 31 days) and January (15 days). For temperature, salinity and oxygen variables, monthly climatologies are used above 1500 m and annual ones below. For nutrients, monthly climatologies are only available and used above 500 m. The distribution of WOA09 climatologies does not extend close to the coasts. Hence, given the proximity of the CARIACO timeseries station to the mainland, ancillary data is not available for this site from WOA09. Figure 1 shows the distribution of the sediment-trap deployments compiled in this data set. Data coverage spans from 1982 to 2011, with the largest amount of observations between 1990 and 2010 (Fig. 2). Figure 3 shows a map with the number of data points available on a 5 • × 5 • grid. The most abundant contributions to this data set derive from es-  (Miquel et al., 2011); the Porcupine Abyssal plain (PAP), 49 • N, 16 • 30 • W, 366 data points, 6.2 % of total ; the North Atlantic Bloom Experiment (NABE), from 34 • N, 21 • W to 48 • N, 21 • W, 170 data points, 2.9 % of total (Honjo and Manganini, 1993;Martin, 2003a, b, c); and the European Station for Time series in the OCean (ESTOC), 29 • N, 15.5 • W, 124 data points, 2.1 % of total (Neuer et al., 1997). Missing in this compilation are data from the Progamme Océan Multidisciplinaire Méso Echelle (POMME, 30-60 • N, 0-30 • W), which are not yet publicly available. We are also aware of a few more recent data sets from the ESTOC, but these have copyright restrictions (e.g. Neuer et al., 2007).

Data distribution
Observation depths span from 15 down to 5031 m (Table 2). Tot_Mass, CaCO 3 and Terr/Litho exhibit the broadest range of export fluxes (0.0-5585 and 0.0-4529 mg m 2 d −1 , respectively), while PON and PIC exhibit the narrowest range (0.0-57.9 and 0.04-81.4 mg m 2 d −1 ). The largest number of observations, with higher vertical resolution, have been made within the first 1000 m of the water column, particularly in the upper 500 m (Fig. 4). At depths greater than 1000 m a preference for sampling at 3000 m is apparent in the data. With the exception of the PIC flux and lithogenic/terrigenous flux, which contain the lowest number of data points (Table 2), all other variables of particle export have been sampled at a similar vertical resolution (Fig. 4). The overall pattern of particle export fluxes show the expected decrease from upper layers to depth as sinking particles decay and dissolve (Fig. 5). This is particularly clear in the Atlantic Ocean and Mediterranean Sea data (red and orange symbols in range of values within the upper 1250 m which reduce substantially at greater depths. The proportion each export variable makes to total mass flux is shown in Fig. 6. The largest contributions of up to 80 % are provided by the Terr/Litho, CaCO 3 and bSiO 2 fractions but there is a broad range in the contribution each fraction makes to the total mass flux at all sampled depths and within the four ocean domains. In the case of CaCO 3 and Terr/Litho the range in the contribution from these fractions to total mass flux appears to narrow with depth which may reflect the attenuation of other variables with depth rather than any systematic change to Terr/Litho and CaCO 3 contributions. The largest contribution (up to 90 %) made by bSiO 2 to total mass flux derive from trap deployments in the Southern Ocean with apparent peaks at depths of 500 and 4500 m; elsewhere the bSiO 2 contribution is smaller but nevertheless a major component of the downward particle flux. The contribution made by POC to total mass flux is broadly similar within all four ocean domains and decreases from ∼ 80 % in the upper 500 m to < 20 % at 5000 m revealing a marked attenuation with depth. Both PON and PIC typically contribute < 20 % to total mass flux, but, whilst the contribution from PIC remains fairly constant with depth, there is vertical attenuation of PON with depth such that at depths > 500 m the PON contribution is < 10 %.

Conclusions
We have assembled a data set of over 5900 data points of particle flux across the wider Atlantic Ocean and adjacent seas, which will be invaluable in determining seasonal and geographical variability in the biological carbon pump. Our initial examination of this data set already indicates important  differences in the flux estimates between ocean domains and in the contribution particular flux variables make to the total mass flux, which may in turn indicate important differences in the strength of the BCP due to local environmentaland ecosystem-level forcing. Exploring the reasons for such differences remains a major scientific and societal problem particularly given projected changes to the future ocean and this data set will help in this endeavour. This data set has been submitted to the data repository PANGAEA ® (http: //www.pangaea.de), where it has been made available under doi:10.1594/PANGAEA.807946.

List of compiled data sets
Here we list all individual data sets. PANGAEA ® digital object identifiers are also given.