Abstract
 Top of page
 Abstract
 1. Introduction
 2. Methods
 3. Results
 4. Discussion
 5. Summary
 Acknowledgments
 References
 Supporting Information
[1] We compiled and standardized sediment trap data below 1000 m depth from 52 locations around the globe to infer the implications of the Armstrong et al. [2002] “ballast” model to the ratio of organic carbon to calcium carbonate in the deep sea (the rain ratio). We distinguished three forms of mineral ballast: calcium carbonate, opal, and lithogenic material. We concur with Armstrong et al. [2002] that organic carbon sinking fluxes correlate tightly with mineral fluxes. Based on the correlations seen in the trap data, we conclude that most of the organic carbon rain in the deep sea is carried by calcium carbonate, because it is denser than opal and more abundant than terrigenous material. This analysis explains the constancy of the organic carbon to calcium carbonate rain ratio in the deep sea today, and argues against large changes in the mean value of this ratio in the past. However, sediment trap data show variability in the ratio in areas of high relative calcium carbonate export (mass CaCO_{3}/mass ratio > 0.4), unexplainable by the model, leaving open the possibility of regional variations in the rain ratio in the past.
1. Introduction
 Top of page
 Abstract
 1. Introduction
 2. Methods
 3. Results
 4. Discussion
 5. Summary
 Acknowledgments
 References
 Supporting Information
[2] Model studies have proposed an increase in ocean pH as a mechanism to explain the decrease in atmospheric pCO_{2} during glacial time [Archer et al., 2000]. The increase in pH of the ocean could be driven by an increase in the weathering minus shallow water CaCO_{3} deposition [Opdyke and Walker, 1992], but the lysocline signature of an increase in deep sea CaCO_{3} burial during the last glacial maximum is not seen [Archer and MaierReimer, 1994; Catubig et al., 1998]. The pH of the ocean may also be sensitive to the ratio of particulate organic carbon (POC) to CaCO_{3} or particulate inorganic carbon (PIC) raining to the deep sea (the POC/PIC rain ratio), which may have a smaller impact on the distribution of CaCO_{3} on the seafloor [Archer et al., 2000].
[3] Export of POC into the deep ocean is thought to be driven primarily by growth and sedimentation of diatoms following nutrient injection events due to winter convection or upwelling [Peinert et al., 1989; Smetacek et al., 1990; Ittekkot et al., 1991; Honjo, 1996; Takahashi et al., 2000]. CaCO_{3} production, mediated primarily by coccolithophores, is thought to increase when diatom growth is limited by availability of silica or iron [Margalef, 1978; Tyrell and Taylor, 1996]. However, if organic carbon export to the deep sea were associated primarily with diatom production at the sea surface, we would expect regional variability in the rain ratio that is not observed in trap data, and could not be tolerated by models of CaCO_{3} preservation on the seafloor [Archer, 1996]. Armstrong et al. [2002] and François et al. [2002] propose that organic carbon export is determined by the presence of ballast minerals (opal, calcium carbonate and lithogenic material). The model of Armstrong et al. [2002] partitions POC in sinking matter into two fractions: a ballastassociated fraction with remineralization scale similar to the dissolution scales of the minerals, and a “free” POC fraction which decays in the top kilometer or so. Either the ballast increases the sinking speed of the organic matter, as implied by the term ballast, or it could provide some physical protection from degradation through adsorption or other interaction between organic matter and minerals. Armstrong et al. [2002] found significant geographical differences in the POC fraction associated with the mineral fraction and hypothesized that such differences might be related to differences in composition of the ballast material. The purpose of this study is to synthesize a global longterm sediment trap data set to investigate the relationship between mineral fractions (opal, calcium carbonate and lithogenic material) and the transfer of POC to the deep ocean, based on the model of Armstrong et al. [2002].
3. Results
 Top of page
 Abstract
 1. Introduction
 2. Methods
 3. Results
 4. Discussion
 5. Summary
 Acknowledgments
 References
 Supporting Information
[9] Scatterplots of POC versus mineral constituents of annual fluxes for trap experiments below 1000 m depth are given in Figure 3 and indicate a tendency toward increasing POC with mineral flux (CaCO_{3}, opal, and lithogenic material). Multiple correlation analysis of POC fluxes as a function of mineral fraction (in mass units) for different trap data sets is highly significant with a correlation coefficient varying from 0.956–0.985 (Table 2). The carrying coefficients are highest for CaCO_{3} and lowest for opal mass fluxes. Using different data sets for the analysis does not change significantly carrying coefficients for the different mineral fractions, however, the carrying coefficients tend to decrease when using deep trap data, suggesting that a small fraction of the POC transported to 1000 m is remineralized before reaching the sediments. Overall carrying coefficients determined using trap experiments below 2000 m seem to show the least uncertainty. If the data set is reduced to sediment trap experiments carried out in open ocean areas far from influence of “exogenous” sources such as resuspended sediments from continental shelves or icerafted debris (“selected” sediment trap experiments, Figure 1), the coefficients for CaCO_{3} and opal do not change much. The coefficient for the lithogenic fraction shows the highest variability between data sets and the highest uncertainties (Table 2). These results could indicate that the fraction of carbon associated with the lithogenic fraction is sensitive to the sources and possibly mineralogy of lithogenic particles. The analysis results for the selected trap data set shows the least variability in the carrying coefficients for lithogenic material and, therefore, seem to give the best results for open ocean conditions. In view of these results following analysis will, therefore, be limited to the selected trap data set.
Table 2. Carrying and Correlation Coefficients for Multiple Correlation Analysis of POC Fluxes Versus Mineral Fluxes^{a}  CaCO3  Opal  Lithogenic  r  N 


All traps < 1000 m  0.094 ± 0.010  0.025 ± 0.011  0.035 ± 0.006  0.962  107 
TrapsSP < 1000 m  0.090 ± 0.010  0.023 ± 0.011  0.052 ± 0.022  0.956  105 
Traps < 2000 m  0.075 ± 0.011  0.029 ± 0.009  0.052 ± 0.018  0.973  78 
Traps < 3000 m  0.077 ± 0.012  0.030 ± 0.010  0.039 ± 0.035  0.970  50 
Selected < 1000 m  0.083 ± 0.011  0.023 ± 0.010  0.068 ± 0.028  0.976  46 
Selected < 2000 m  0.074 ± 0.010  0.025 ± 0.008  0.071 ± 0.030  0.985  34 
Selected < 3000 m  0.070 ± 0.025  0.026 ± 0.009  0.065 ± 0.048  0.985  20 
[10] Carrying coefficients obtained using multiple linear correlation analysis are all significant. However, we find that a singleballast model (following Armstrong et al. [2002]) is almost as good at predicting the overall variability of POC fluxes as the threeballast model presented above (Table 3). These results seem to contradict the hypothesis that regional difference in POC flux and POC/PIC rain ratios could be determined by changes in ballast composition. However, comparison of measured versus predicted values (Figure 4) indicates that the threeballast model is better for predicting POC/PIC rain ratios. Consequently, the single ballast model overestimates POC fluxes in regions where opal dominates ballast fluxes such as the Bering Sea, the Subarctic Pacific and the Southern Ocean and underestimates POC fluxes in areas where CaCO_{3} and lithogenic material dominate mineral fluxes. None of the models are very good at reproducing geographical variability of PIC/POC rain ratios in areas of intermediate to high CaCO_{3} versus opal and lithogenic material export fluxes (mass CaCO_{3}/total mass > 0.4, Figure 5), although the threeballast model gives a better estimate of average values for both POC fluxes and POC/PIC rain ratios (Figure 5, Table 4).
Table 3. Variance Explained (in %) by the Models for POC Fluxes and POC/PIC Rain Ratios^{a}  Three Fractions  One Fraction 

POC  POC/PIC  POC  POC/PIC 


<1000 m  84  80  67  78 
<2000 m  89  87  71  86 
<3000 m  87  89  75  95 
Table 4. Average POC/PIC Rain Ratios (±1 Standard Deviation) and Variance Explained (in %) by the Models for POC Fluxes and POC/PIC Rain Ratios^{a}  <1000 m  <2000 m  <3000 m 


ThreeFractions Model^{b} 
Percent of POC variance explained  83  88  84 
Percent of POC/PIC variance explained  Ns  Ns  Ns 
Average POC/PIC  0.852 ± 0.101  0.790 ± 0.114  0.765 ± 0.104 

OneFraction Model^{c} 
Percent of POC variance explained  79  87  80 
Percent of POC/PIC variance explained  Ns  Ns  Ns 
Average POC/PIC  0.714 ± 0.110  0.653 ± 0.102  0.574 ± 0.098 

Measured^{d} 
Average POC/PIC  0.814 ± 0.241  0.768 ± 0.197  0.719 ± 0.215 
[11] To investigate the causes for the additional regional variability of POC fluxes and POC/PIC rain ratios in areas of high relative CaCO_{3} fluxes, we tested the possibility that the “free” POC fraction might still be a significant component of the fluxes. These tests assumed the presence of free POC, representing most of the export production near the surface, proportional to surface estimates of primary productivity (PP) [Antia et al., 2001]. Primary productivity was derived from estimates by Antoine et al. [1996] and Behrenfeld and Falkowski [1997]. The decay scale for free POC was adopted from Armstrong et al. [2002] (efolding scale of 560 m):
alternatively,
As for the carrying coefficients, the proportionality factor (β) determining free POC from primary productivity data can be solved through multiple regression analysis.
[12] Multiple correlation analysis of the selected trap data set indicates that the primary productivity term was not significant for trap experiments below 2000 m depth. When analyzing the selected data set with trap experiments below 1000 m depth, the free POC coefficient is significant only for the model assuming an inverse relationship between free POC and depth (β.PP/z). The improved model explains at most 22% of POC/PIC rain ratio variability for traps with high mass CaCO_{3}/total mass ratios (still not very good). This corresponds to an total increase from 80 to 84% of the total variance in POC/PIC rain ratios explained for trap experiments below 1000 m depth, not better than results obtained using trap experiment below 2000 m and 3000 m depth (Table 3). In short, including a free POC fraction in the analysis reduces differences between results using data collected at different depth intervals, rather than improving significantly the predictions for POC/PIC rain ratios at high mass CaCO_{3}/total mass ratios. The same tests were done using only trap data with mass CaCO_{3}/total mass >0.4, based on the assumption that in systems with high mass CaCO_{3}/total mass export flux the carrying coefficients for the different mineral fractions might differ. The results are similar to the previous analysis: variability in POC/PIC rain ratios cannot be explained by the three fractions ballast model independent of the depth range of the trap experiments used. In addition, the derived equations assuming a residual free POC fraction improved POC/PIC rain ratio predictability by at most 24%, not better than models including trap data with low mass CaCO_{3}/total mass ratios (<0.4).
[13] We also tested the possibility that the ballasted POC decays with depth, rather than sinking conservatively. In order to quantify the depth effect we compared the ratio between predicted values and measured values to depth:
[14] Results indicate a significant relationship with depth for trap data independent of the depth interval considered for the analysis. Comparison of predicted versus measured values indicates that some of the variability in the POC/PIC rain ratios from trap experiments with high mass CaCO_{3}/total mass ratios can be explained by ballasted POC degradation at depth (Table 5) depending on trap data used. Including a depth dependent remineralization of ballasted organic carbon does not, however, improve predictions of POC fluxes and POC/PIC rain ratios when trap experiments with low mass CaCo_{3}/total mass ratios are included in the analysis (Tables 3, 4 and 5).
Table 5. Variance Explained (in %) by the Models for POC Fluxes and POC/PIC Rain Ratios and Average POC/PIC Rain Ratio (± 1 Standard Deviation)^{a}  <1000 m  <2000 m  <3000 m 


All Trap Data 
Percent of POC variance explained  74  78  86 
Percent of POC/PIC variance explained  36  72  87 
CaCO_{3} > 0.4 
Percent of POC variance explained  59  78  88 
Percent of POC/PIC variance explained  12  16  33 
Average predicted POC/PIC^{b}  1.019 ± 0.414  0.852 ± 0.203  0.738 ± 0.114 
Average measured POC/PIC^{c}  0.814 ± 0.241  0.768 ± 0.197  0.719 ± 0.215 
[15] In regions of high mass CaCO_{3}/total mass export flux, most of the variability in the POC/PIC rain ratio in the deep sea is not predictable by any version of the ballast model that we have been able to devise and is not related to differences in primary productivity (direct comparison of POC/PIC rain ratios in areas of high mass CaCO_{3}/total mass ratios with levels of primary production is significant but corresponds only to at most 6% of the variability).
[16] The importance of CaCO_{3} in transporting carbon to the deep ocean contradicts the hypothesis based on plankton dynamics and seasonal sediment trap flux showing high opal versus CaCO_{3} ratios during periods of high POC flux [Ittekkot et al., 1991; Honjo, 1996; Takahashi et al., 2000]. We, therefore, collected individual cup measurements from the traps used in the analysis of annual data (Figure 1) to investigate variability of POC fluxes as a function of composition in the mineral fraction on short timescales (4–61 days). Correlation of POC fluxes with the different mineral fractions shows similar results as for annual data (Figure 6). Again, the highest correlation coefficient is obtained for CaCO_{3}. POC versus opal fluxes show variations dependent on the mass ratio of opal/total minerals. Applying the models obtained from the analysis of annual data to the cup data gives good results (for all trap data save SP* r = 0.899 and r = 0.902, and for the reduced data set r = 0.909 and r = 0.982; for POC and POC/PIC respectively). These results confirm that the carrying coefficient as determined for the annual flux data are not an artifact of averaging but reflect mechanisms affecting POC fluxes both on an annual scale as well as on the scale relevant to biological and chemical processes.