4.1. Particle Export
 Given that the export flux of biogenic matter out of the euphotic zone is one of the primary drivers of ocean biogeochemical cycling, it is important to determine its magnitude. A synthesis of these fluxes is presented in terms of ocean regions in Table 2 and compared with previous estimates in Table 3. Our central estimate of POC export of 9.6 ± 3.6 Pg C a−1 is slightly lower than the satellite-based Laws et al.  estimate of 11.1 Pg C a−1. However, the Laws et al.  estimate includes export through transport of dissolved organic carbon (DOC) in addition to particle sinking, which ours does not. Hansell and Carlson  estimate that about 20% of the net community production occurs in the form of DOC, which would increase our value for new production to 12.3 ± 4.0. As seen in Table 3, our central estimate of POC flux is extremely close to the estimates of a number of circulation models [Aumont et al., 2003; Heinze et al., 2003; Gnanadesikan et al., 2004] and to the value of the inverse model of Schlitzer . The recently published model of Moore et al.  is somewhat lower than these estimates.
 Gnanadesikan et al.  noted that diagnostic models which restore surface nutrients toward observations produce rates of biological cycling that depend strongly on the vertical diffusion coefficient and suggested that satellite-based observations might be used to put constraints on the rate of vertical exchange. Gnanadesikan et al.  found that the particle export in a suite of such models varied between a low value of 6.9 GtC a−1 (low vertical diffusion and weak vertical exchange in the Southern Ocean) to a high value of 13.1 GtC a−1 (high vertical diffusion). While this study is unfortunately not sufficiently precise to constrain mixing within forward models, it does suggest an increasing broad consensus between approaches of global POC export being 10 ± 3 Pg C a−1.
 This degree of consensus is not yet available for mineral CaCO3 and has been recently described in detail by Berelson et al. . Our estimate of the export of CaCO3 of 0.52 ± 0.15 Pg C a−1 is significantly less than the Milliman and Troy  estimate of 0.70 Pg C a−1, and much less than the Lee  estimates of 0.9–1.1 Pg C a−1. Berelson et al.  give a range of estimates of 0.5–1.6 Pg C a−1 based on a satellite-based estimate of plankton calcification rates, 1.4–4.7 Pg C a−1 based on surface sediment traps, and 1.6 Pg C a−1 based on the combination of an estimate of upper water column dissolution of 1.0 Pg C a−1 and an estimate of the export to sediment traps at 2000 m of 0.6 ± 0.4 Pg C a−1. Model results range from 0.38 to 1.64 Pg C a−1 [Heinze et al., 1999, 2003; Gnanadesikan et al., 2004; Moore et al., 2004; Jin et al., 2006] (Table 3). This discrepancy is particularly important in the Southern Ocean, where a number of models using the OCMIP-like protocols suggested a rain ratio near 0.08 [e.g., Jin et al., 2006], while the work of Feely et al.  and Sarmiento et al.  suggest a much lower value. If one removes CaCO3 fluxes south of 50°S within the models of Gnanadesikan et al. , the resulting global export flux ranges between 0.46 and 0.66, essentially overlapping our estimate. As discussed above for the North Atlantic, we suspect that the large alkalinity supply to surface waters may lead to significant underestimation of the CaCO3 to POC ratio by the Sarmiento et al.  method in that region. Alternatively, Jin et al.  have also argued that this method underestimates the net CaCO3 to organic carbon ratio by up to 50% because of the vertical gradient method neglecting horizontal transport processes, specifically DOC transport, and because their mean ratios were computed by area-weighted averaging rather than flux-weighted averaging, of particular importance in the Southern Ocean.
 Much of the analysis hinges on assumptions regarding the dissolution depth scale of CaCO3. Chung et al.  and Feely et al.  have suggested that much of the excess alkalinity in the upper waters could be due to the dissolution of particles within the upper water column. Friis et al.  suggested that the excess alkalinity in the upper water column could be derived from dissolution of CaCO3 from the sediment. The role of river-derived alkalinity on these estimates is also uncertain, but potentially important [e.g., Lee et al., 2006].
 Quantification of global SiO2 export has also been somewhat controversial. As shown in Table 3, literature values of SiO2 export flux range between 70 and 185 Tmol Si a−1 Our estimate of SiO2 export of 100.1 ± 35.5 Tmol Si a−1 is similar to the lower databased values [Treguer et al., 1995; Nelson et al., 1995; Gnanadesikan, 1999] which range between 70 and 140 Tmol Si a−1. It is also consistent with the model-based values proposed by Gnanadesikan  for models which include the Gent and McWilliams  eddy parameterization and have low vertical diffusivity.
 One major source of disagreement in the models is the SiO2 export in the Southern Ocean. Models differ greatly in how much convection they predict in the Southern Ocean, with simulations which do not have the Gent and McWilliams  parameterization giving a lot of convection and high inferred Southern Ocean SiO2 export (69–171 Tmol Si a−1 [Gnanadesikan, 1999]) and simulations which include it giving low values (23–38 Tmol Si a−1 [Gnanadesikan, 1999]). Coarse resolution models in which convection is the dominant mechanism by which radiocarbon is delivered to the deep Southern Ocean have SiO2 export values that are much higher than the 33 Tmol Si a−1 we obtain (Table 3). Our estimates agree with those of Nelson et al.  but conflict with higher in situ estimates by Pondaven et al.  and Nelson et al.  of 50–80 Tmol a−1. Because all in situ measurements (whether sediment traps, bottom SiO2 flux, or SiO2 incorporation) require broad geographical extrapolation over a highly heterogeneous environment, it is difficult to determine the relative robustness of these estimates.
 Another area where models often disagree both with each other and with observations is the equatorial Pacific. Modeling studies [e.g., Gnanadesikan and Toggweiler, 1999; Gnanadesikan et al., 2002, 2004] have shown the character of this upwelling to be very sensitive to the details of vertical diffusion, and that high vertical diffusion coefficients generally lead to excessive production. Our estimate of 4.5 Tmol a−1 east of the dateline between 10°S and 10°N (implying an average rate of only about 0.2 mol m−2 a−1) is in line with the model estimate of Jiang et al.  but are threefold lower than the modeled estimates of Gnanadesikan  and six- to eight-fold lower than the modeled estimates of Heinze et al. . This discrepancy suggests that either the satellite-based estimates of POC flux are too low or that both of the models used by Gnanadesikan  and Heinze et al.  have excessive tropical upwelling. We suggest this issue as a focus for future modeling experiments.
 We provide a more detailed analysis of variability in Figure 8. Variability in monthly particle export (standard deviation of monthly values over the 1998–2004 period divided by the mean over the 1998–2004 period multiplied by 100) is shown in Figure 8a. This treatment of variability illustrates the high month-to-month variability in polar regions as expected because of the solar cycle but also some high areas of variability in the tropical Atlantic and tropical Indian oceans and the subpolar frontal regions but relatively low variability in the equatorial Pacific. Much of the overall variability is diminished when data are averaged over the annual cycle (standard deviation of annually averaged values over the 1998–2004 period divided by the mean over the 1998–2004 period multiplied by 100), as shown in Figure 8b. The great exception to this is along the Antarctic coastline where two- to three-fold variations in annual particle export are seen from year to year. However, it is not clear how much of the variability is real as some of it may be due to aliasing because of poor satellite coverage due to clouds. Further analysis of annual versus interannual timescales of variability (standard deviation of annually averaged values over the 1998–2004 period divided by the standard deviation of monthly values over the 1998–2004 period multiplied by 100; Figure 8c) demonstrates that globally the seasonal variability in particle export dwarfs the interannual variability, with interannual variability accounting for only about 10–30% of the total variance in most regions. The one exception to this general trend is the central tropical Pacific, where interannual variability accounts for 50–70% of the monthly averaged particle export variability.
Figure 8. Global ocean particulate organic carbon export (PE) variability in terms of percent relative standard deviation. (a) Variation in monthly particle export relative to the mean (σPE/PEave × 100). (b) Variation in annual particle export relative to the mean (σann.ave.PE/PEAve × 100). (c) Extra-annual variation relative to the total monthly particle export variability (σann.ave.PE/σPE × 100).
Download figure to PowerPoint
 This analysis provides a fascinating look at the nature of particle export variability. It suggests that most variability in particle export relative to the mean state occurs in polar, coastal and gyre boundary regions (Figure 8a), but that only in the ice-influenced Southern Ocean does year to year variability in total export vary more than 30% (Figure 8b). Still, in all subpolar and polar regions, it suggests that intra-annual variability swamps interannual variability overall, with the only areas with strong interannual variability being the equatorial Pacific and the subtropical gyres of the Atlantic and Indian oceans.
4.2. Penetration of Export Flux to Deep-Sea Sediments
 In representing observed vertical gradients in POC fluxes, the Martin et al.  function provides a critical improvement over exponential decay in that it guarantees that a certain fraction of surface production penetrates through the water column to the seafloor. The Klaas and Archer  treatment quantifies a mechanistic explanation for this functionality, arguing that POC does not penetrate through the water column unless it is protected by mineral, similar to but importantly different than arguments made by Armstrong et al.  and Francois et al. . Whether this protection mechanism is through the increase in sinking velocity via ballasting or through a decrease in the remineralization rate constant via biological unavailability is still an open question. Because the protected portion can be regionally larger or smaller than that assumed in the Martin curve depending on SiO2, CaCO3 or lithogenic export, it provides a means through which the biogenic cycles of the elements are coupled.
 Our analysis of the satellite-derived export estimates and comparison with benthic oxygen utilization and sediment POC burial estimates demonstrates a surprisingly strong level of internal consistency between these approaches given the uncertainties involved. Because this internal consistency was only achievable through the application of the full mineral protection model of particle flux penetration including protection by CaCO3, SiO2 and lithogenic components, our analysis provides a new line of support for the theory of mineral protection of POC as a means of transporting it to the deep sea in accounting for regional variability in flux penetration length scales. Furthermore, the comparison between lithogenic fluxes through atmospheric deposition and sediment burial demonstrates the importance of both atmospherically and riverine derived lithogenic material in POC cycling. Klaas and Archer  estimated the protection of organic material flux to the seafloor below 1000 m to be dominated by CaCO3 almost exclusively, with CaCO3 accounting for 80–83% of mineral protection, followed by SiO2 (11–15%) and lithogenic material (clay; 5–6%). Our analysis suggests that lithogenic material plays a much larger role than previously considered, with CaCO3 still accounting for a majority (56%; sensitivity range 42–76%) of mineral protection but with lithogenic material (38%; sensitivity range 16–51%) following close behind and SiO2 (6%) falling a distant third. This result suggests that models which exclude the role of lithogenic material will significantly underestimate the penetration of POC to the deep seafloor by approximately 38% globally, and by a much larger fraction in areas with low productivity. However, a great deal of uncertainty remains in both lithogenic fluxes in particular and the regional variability in mass accumulation in sediments as a whole. We consider this topic to be in great need of further, focused observational and synthesis work.