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 Amazon River plume is a highly seasonal feature that can reach more than 3000 km across the tropical Atlantic Ocean, and cover ∼2 million km2. Ship observations show that its seasonal presence significantly reduces sea surface salinity and inorganic carbon. In the western tropical North Atlantic during April–May 2003, plume-influenced stations exhibited surface DIC concentrations lowered by as much as 563 μmol C kg−1 (∼28%) and pCO2 as low as 201 μatm. We combine our data with other data sets to understand the annual uptake and seasonal variability of the plume-related CO2 sink. Using flux estimates from all seasons with monthly plume areas determined by satellite, we calculate the annual carbon uptake by the outer plume alone (28 < S < 35) to be 15 ± 6 TgC yr−1. Diazotroph-supported net community production enhanced the air-sea CO2 disequilibrium by 100x and reversed the typical CO2 outgassing from the tropical North Atlantic. The carbon sink in the Amazon plume depends on climate-sensitive conditions that control river hydrology, CO2 solubility, and gas exchange.
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.
 Previous estimates of annual plume carbon uptake [Körtzinger, 2003; Ternon et al., 2000] are based on DIC-S relationships from one or two offshore plume transects, spatially extrapolated using ship-based, monthly sea surface salinity (SSS) climatologies. Until now, limited spatial and temporal coverage has hampered assessment of Amazon plume DIC variability. In this study, we improve current coverage by applying a satellite-based model of plume S to four published data sets and one new data set. We investigate temporal changes in river supply and offshore processes to explain observed seasonality. This work extends our mechanistic understanding of the tropical Atlantic CO2 sink, which will assist future predictions of sink behavior under changing climate conditions.
 Total alkalinity can be altered nonconservatively by biological activity [Morel, 1983; Stumm and Morgan, 1996]. Photosynthesis and respiration alone do not alter TA, but other biogeochemical changes can. Nutrient uptake or remineralization causes small TA changes, and some plankton lower TA by secreting carbonate tests. Assuming Redfield stoichiometry and a soft-to-hard tissue production ratio of 4:1, biological activity changes TA by 28% of the net DIC change [Broecker and Peng, 1982]. If the 4:1 ratio increases (up to 40:1 in some regions [Pätsch et al., 2002; Hood et al., 2006]), then the biologically associated TA change would be a much lower percentage of the DIC change. Higher ratios are expected in regions dominated by diatom blooms [Rixen et al., 2006].
 We report a new data set for WTNA surface mixed layer DIC, TA, and underway pCO2 measured during spring 2003 and use this data with historical data from the WTNA [Ternon et al., 2000; Körtzinger, 2003; Cooley and Yager, 2006] and Amazon main stem [Devol et al., 1995] to examine seasonality in the Amazon plume-associated CO2 sink. We quantify potential seasonality in riverine supply by modeling TA at the river mouth with two independent methods. We also quantify the biological impact on plume DIC over four seasons and then apply climatological control to calculate seasonal gas exchange. Combining seasonal CO2 flux estimates with satellite-based plume areas provides an estimate of the total annual plume CO2 uptake that includes observed variability.
2.1. Sample Collection and Analysis
 The spring 2003 (Sp03) samples were collected during daily stations aboard the R/V Seward Johnson I on the eighth MANTRA/PIRANA (“MP8”) cruise (6°N–13°N, 48°W–58°W; Figure 1 and Table 1). Water samples for DIC and TA analysis were collected between 1 and 1000 m and stored according to standard protocols [Department of Energy, 1997; Cooley and Yager, 2006]. In addition, a single sample of Amazon surface water was collected from a small boat near Macapá, Brazil (0°N, 51°W; Figure 1) on 25 March 2002 using a pole sampler. The sample was not fixed but was sealed tightly and shipped to Georgia, where it was refrigerated in the dark until analysis in 2005.
 During the cruise, surface ocean pCO2 was measured continuously using an automated underway analyzer plumbed into the ship's uncontaminated seawater line. The system's primary components included a 10 L showerhead-type seawater equilibrator [Bates et al., 1998; Sweeney et al., 2000; Takahashi et al., 1997] and an infrared CO2 detector (LI-6262, Li-Cor, Lincoln, Nebraska). Seawater flow of 8 L min−1 produced a system e-folding time of 35 minutes (S. Cooley, unpublished data, 2006). Measurements were automatically calibrated at least hourly against five CO2-N2 gas mixtures containing 0−510.85 ppm CO2 (certified by the National Oceanographic and Atmospheric Administration Climate Monitoring and Diagnostics Laboratory, NOAA CMDL, Boulder, Colorado, USA). Instrument accuracy was ±1 μatm. pCO2 was calculated at in situ temperature [Takahashi et al., 1993]. A Lanczos filter [Duchon, 1979] (100 weights, cutoff frequency = 9.3 × 10−5 Hz) was used to remove occasional brief high-pCO2 excursions related to short interruptions in the ship's seawater supply during high seas.
 DIC measurements were performed on a SOMMA [Johnson et al., 1993; Cooley and Yager, 2006] connected to a CO2 coulometer (UIC, model 5011) and equipped with a Sea-Bird conductivity cell (SBE-4). Salinity (S, practical salinity scale, unitless) was calculated on the basis of cell calibration to IAPSO standards. Precision based on analysis of replicate standards or replicate samples [Department of Energy, 1997] was ±0.51 and ±1.58 μmol C kg SW−1, respectively.
 The Macapá sample was centrifuged at 8500 rpm for 20 minutes to remove the larger suspended sediment particles before chemical analysis. Salinity was determined using the SOMMA conductivity cell, and TA was measured as described above.
 Following previous analyses [Cooley and Yager, 2006; Lentz and Limeburner, 1995; Geyer et al., 1996; Lentz, 1995b], plume stations were identified by low surface salinities (<35) and shallow haloclines overlying oceanic-S water. Surface mixed layer (SML) depths used to select the plume-influenced data were calculated from CTD profiles using a density gradient criterion (Δρ/Δz > 0.01 [Brainerd and Gregg, 1995]). The critical value throughout the analysis for statistical significance was α = 0.05, and mean values are presented with their standard error unless otherwise noted.
 Estimated TA at the river mouth from 1991–2003 was modeled by predicting main stem TA at an upstream river gauging station (Manacapuru) based on Andean discharge, and propagating this TA signal downstream to the river mouth. Following the method of Devol et al. (1995), a multiple regression equation was developed to predict Manacapuru TA (1700 km inland, Figure 1) using measured Manacapuru TA from 1982–1992 [Devol et al., 1995], daily Manacapuru discharge (QMan, Table 1), and percent Andean discharge (A%). Andean discharge to the Amazon has been defined as main stem discharge at Sao Paulo de Olivença (QSPdO [Devol et al., 1995; Quay et al., 1992; Richey et al., 1990] (Table 1 and Figure 1)), making A% = (QSPdO/QMan)*100%.
 For the periods in 1991–2003 where Hidroweb records of Manacapuru discharge are not continuous (1991–1996, 2002–2003), we estimated QMan as a fraction of Óbidos discharge (QObi, Table 1, continuous for 1969–2005). We calculated the daily ratio of Manacapuru discharge to Óbidos discharge (M = QMan/QObi) when discharge records overlapped (June 1972 to December 1984, January 1996 to December 2001). Daily ratios from all 17 years were averaged into mean daily ratios (MD), and QMan data missing from the Hidroweb record were filled with QMan, predicted = MD*QObi (75% of Hidroweb QMan data was randomly selected to calculate MD, and the remaining 25% was compared to QMan, predicted to check prediction accuracy). Daily Manacapuru TA (TAMan) was calculated for 1991–2003 with A%, the continuous QMan, and the multiple regression equation.
 To obtain river mouth TA, the Andean signal was diluted into the total river discharge at Macapá (location on Figure 1), assuming no lowland contributions to alkalinity. For each month of 1991–2003, mean TAMan and QMan were calculated from daily values, and monthly TA at Macapá was predicted by the equation: TAMacapá = TAMan*QMan/QTotal, where QTotal is modeled monthly total Amazon discharge using precipitation data (see next paragraph). Monthly TAMacapá was further combined into an interannual climatology to show average seasonal variability. Error associated with mean discharge and multiple regression equation-modeled TA was propagated throughout the dilution model by a Monte Carlo method [see Yager et al., 1995].
2.3.3. Modeled Main Stem Discharge
 Previous Amazon hydrology studies [e.g., Zeng, 1999; Labat et al., 2004] suggest that interannual and annual precipitation signals dominate river discharge data. To capture maximum interannual variation, total Amazon discharge was modeled by combining a seasonal discharge climatology [Dai and Trenberth, 2002] (based on streamflow coupled to a runoff database) with integrated precipitation over the watershed. Two climatologies based on monthly precipitation databases were used (GPCP and CMAP, Table 1). Precipitation was lagged and linearly correlated to the Dai and Trenberth  climatology. Adjusted discharge was then calculated using the precipitation climatology and the linear relationship. The resulting two linear models coupling discharge and precipitation climatologies were applied to interannually varying precipitation data to estimate total Amazon discharge at the river mouth (Qtot). For comparison, the Óbidos gauge was projected to the river mouth using a linear regression against the Amazon River mouth climatology. Data from the Óbidos gauging station obtained from SAGE Global River Discharge Database www.sage.wisc.edu/riverdata) were used to validate the model. As above, linear models were also constructed on the basis of the relation between interannually varying precipitation and Óbidos data projected to the river mouth.
2.3.4. Mixing Model: River End-Members and Biological DIC Drawdown
 A previously described conservative mixing model [Cooley and Yager, 2006] was used to determine the proportions of river water and seawater in each sample based on S. We used this information, non-plume-influenced seawater end-member TA (TAs), and observed sample TA (TAobs) to calculate the river end-member TA (TAr). The mixing-modeled TAr was then compared to discharge-modeled TAMacapá.
 The mixing model was also used to calculate biological effects on DIC for all plume-influenced samples for which DIC and TA were available [Cooley and Yager, 2006]. W01 and fall 2002 (F02) data were not analyzed this way because W01 sampling did not encounter the river plume [Cooley and Yager, 2006], and the F02 published data set did not include DIC and TA. The model calculates expected DIC (DICexp) of each sample, which is the DIC value if conservative river-ocean mixing were the only influence. The difference between DICexp and measured DIC (DICobs) is attributed to biological activity and is ΔDICbio. Positive ΔDICbio values indicate net production, and negative values indicate net respiration. To avoid skewing the seasonal analysis, we only used data for outer plume water with salinities between 28 and 35, where all seasons had good coverage. Seawater end-members used in the mixing model were average values of pooled nonplume (S > 35) SML samples from all five cruises (Ss = 36.04 ± 0.22, TAs = 2367 ± 18 μmol kg−1, DICs = 2014 ± 19 μmol kg−1; mean ± s.d., n = 254). Error associated with average end-member values was propagated through the mixing model by a Monte Carlo method [see Yager et al., 1995].
2.3.5. The pCO2 and CO2 Flux Estimates
 In situ pCO2 (hereafter pCO2obs) was calculated for W01 and Su01 [Cooley and Yager, 2006] from discrete surface DIC and TA values using CO2SYS [Lewis and Wallace, 1998]. Measured pCO2 from other seasons and years (this study and Ternon et al. ) was used in the rest of the analysis. In addition, the pCO2 expected if conservative river-ocean mixing were the only influence on the system (pCO2mix) was calculated for plume-influenced data sets using CO2SYS and measured temperature, S, nutrients, TAobs, and DICexp output from the mixing model, with CO2SYS settings as described above.
 Air-sea fluxes into the Amazon plume were calculated for pCO2obs and pCO2mix to assess seasonal biological enhancement of the sink. Assuming pCO2obs represents the combined effects of mixing plus biology, the difference between fluxes is attributed to biology. Both fluxes were calculated for each plume-influenced data set using solubility [Weiss, 1974], piston velocity (short-term wind formulation of Wanninkhof and McGillis ), and the air-sea pCO2 gradient (ΔpCO2 = pCO2 atm − pCO2 sea). Atmospheric pCO2 was assumed to be 370 μatm, near the average value (368.4 ± 2.3 μatm, n = 10) of atmospheric values at Ragged Point, Barbados during the cruises. Winds [Atlas et al., 1996] (Table 1) were averaged over space (4°N–13°N, 40°W–60°W) and time to yield mean daily wind speeds for the study region. Climatological daily winds for the WTNA were then determined using all 1995–2005 data. Standard deviations from averaging wind speeds were propagated through air-sea flux calculations and plume uptake estimates (below) by Monte Carlo as described above.
 We examined air-sea CO2 fluxes and parameters influencing them (ΔDICbio, ΔpCO2obs, ΔpCO2mix, solubility, and climatological winds) for temporal or S-related trends. Plume-influenced data was grouped by cruise (Sp95, F96, Su01, Sp03) or by S (28–30, 30–32, 32–35). Equality of group variances was tested with Bartlett's test for homogeneity of variances [Sokal and Rohlf, 2000]. When group variances were equal (see Table 2), group means were compared using an ANOVA. When group variances were unequal, group data distributions were compared with the more appropriate nonparametric Kruskal-Wallis test for differences in location of ranked data [Sokal and Rohlf, 2000].
Table 2. Testing for Differences in Mean (ANOVA) or Data Distributions (Kruskal-Wallis; KW) Between Cruises or Salinity Groups (28–30, 30–32, 32–35) for Parameters Related to Air-Sea CO2 Transfera
ANOVA P Value
KW P Value
ANOVA P Value
KW P Value
Top part of table includes all four cruises. Bottom part of table tests for differences between spring 2003 (Sp03), summer 2001 (Su01), and fall 1995 (F95) only. Significant effects at the >95% confidence level are indicated in boldface.
Sp03, Su01, F95
 Outer plume surface area was obtained from SeaWiFS satellite derived maps of diffuse attenuation coefficient at 490 nm (K490) using a S-K490 relationship [Del Vecchio and Subramaniam, 2004] (also A. Subramaniam et al., Amazon River enhances diazotrophy and carbon sequestration in the tropical North Atlantic Ocean, submitted to Nature, 2007, hereinafter referred to as Subramaniam et al., submitted manuscript, 2007). Briefly, monthly climatological binned 9-km level 3 data of K490 for the WTNA were extracted on the basis of the relationship between in situ measurements of K490 and S [Del Vecchio and Subramaniam, 2004]: K490 = 0.021(±0.002)*S + 0.768(±0.057), (mean ± s.d., r2 = 0.93). This formulation is only valid for 28 < S < 35 in the Amazon plume-influenced WTNA. High CDOM levels associated with Orinoco waters did not affect the 2001–2003 study area [Del Vecchio and Subramaniam, 2004].
 Monthly carbon uptake in the outer plume for conditions of mixing + biology (observed) and mixing-only (expected) were then calculated by multiplying the monthly plume surface area (Table 3) by the mean daily flux (based on climatological winds) and the number of days in the month. Because field observations in the plume were not available for all 12 months, we tried to match observed cruise S distributions with the climatologies. Sp03 fluxes represented May–June, Su01 fluxes were used for July–August, and F95 fluxes were applied to September–December. We used Sp96 data to represent the winter period (January–April) because the unusually low salinities observed east of the Amazon mouth that year (see Figure 1) generally correspond to conditions intermittently observed in February and March [Geyer et al., 1996], when northerly winds impede along-shelf plume movement [see also Nikiema et al., 2007].
Table 3. Plume Surface Area and Mean Monthly Plume CO2 Uptake (±s.d.)
December–March uptakes were calculated from spring 1996 fluxes; April–May from spring 2003; June–August from summer 2001; September–November from fall 1995.
−9.1 ± 22.5
−4.5 ± 22.8
−8.3 ± 20.6
−4.2 ± 20.8
−11.1 ± 27.7
−5.6 ± 28.0
12.6 ± 49.1
176.8 ± 150.6
19.2 ± 74.6
268.7 ± 228.9
17.6 ± 65.0
207.8 ± 208.4
18.3 ± 67.6
216.0 ± 216.6
17.3 ± 63.9
204.3 ± 204.9
−17.7 ± 50.0
99.6 ± 103.9
−10.2 ± 28.9
57.6 ± 60.1
−5.7 ± 16.1
32.0 ± 33.4
−7.7 ± 19.1
−3.9 ± 19.4
Total, Gmol yr−1
15.2 ± 162.8
1244.5 ± 474.3
 Regions of low S (Figure 1), DIC, TA, and pCO2 (Figure 2) extended north and northwest or northeast of the Amazon mouth for all cruises. In Sp03, DIC, TA, and pCO2 were similar to those in other plume-influenced seasons [Ternon et al., 2000; Cooley and Yager, 2006]. Minimum observed Sp03 DIC and TA values at S = 24.3 were 1457 and 1706 μmol kg−1, respectively, 555–662 μmol kg−1 below mean seawater values (Figure 3). Surface pCO2 in the southwesternmost study area fell as low as 201 μatm where S was lowest, and also exhibited lows in the moderate salinities of the mid- to northern study zone (Figures 1 and 2). Underway Sp03 pCO2 was 201–402 μatm (mean = 337 ± 0.4, n = 10169), and underway SSS was 21.9–36.4 (mean = 33.5 ± 0.03, n = 10169). The Sp03 underway sea surface temperature (SST) ranged only from 26.3° to 28.6°C (mean = 27.7 ± 0.004°C; auxiliary Figure S1). Low-carbon, low-S regions moved between seasons, but linear relationships between DIC or TA and S from 28 < S < 35 were highly conserved (r2 = 0.827 and 0.997, respectively [Ternon et al., 2000; Cooley and Yager, 2006]).
3.2. River Discharge Model
 The best regressions between discharge and precipitation climatologies (r2 = 0.94 and 0.95 for GPCP and CMAP, respectively; auxiliary Figure S2) occurred with a 3 month lag between variables, consistent with the findings of Zeng  for the Amazon. The climatology constructed from Óbidos gauge data (1979–1995) was correlated at r = 0.99 with theDai and Trenberth  climatology. The linear models coupling precipitation and discharge climatologies (CMAP and GPCP based) closely matched the projected discharge based on the Óbidos river gauge alone. The resultant discharge time series were correlated at r = 0.87 and 0.86 with the Óbidos data projected to the river mouth between 1979 and 1995. Linear models were also constructed on the basis of the relation between interannually varying precipitation and Óbidos data projected to the river mouth. The differences between precipitation-based and Óbidos-based discharge estimates were modest and indistinguishable between data sets and methods (rms = 0.6 or 11%). Relatively high correlation coefficients and low errors suggest that the method reasonably captured gross variability in the seasonal cycle associated with interannual climate patterns. Greater model error occurred at high precipitation and maximum discharge, where there is more scatter in the relationship (auxiliary Figure S2).
3.3. TA Prediction Model
 Andean discharge peaks in May and declines through the summer to its minimum in September (Figure 4a). The multiple regression equation relating TAMan (Figure 4b) to QMan and A% was: TAMan = 224.6(±69.23) + 5.981(±0.8435)*A% − 1.161 × 10−3(±4.983 × 10−3)*QMan(±s.d., n = 74, r2 = 0.42). The regression was significant, whereas relationships between TAMan and other chemical or hydrographic variables were not [Devol et al., 1995]. For the 25% of the data set withheld, predicted values (QMan,pred) were not significantly different from Hidroweb observations: QMan,pred = 0.967 ± 0.033 (QMan) + 2612 ± 3305 (r2 = 0.94, n = 53). As with Andean discharge, QTotal also peaked in late spring (Figure 4c), but it showed a minimum in November. Discharge predictions made using the two models, CMAP and GPCP, were not significantly different from each other.
 Dilution-modeled interannual average monthly TAMacapá was 295 ± 61 μmol kg−1 (mean ± amplitude; Figure 4d), about a third lower than Óbidos values (360−600 μmol kg−1 [Devol and Hedges, 2001]). The bimodal pattern was a product of the May river discharge peak (Figures 4a and 4c) and the sinusoidal Andean TA supply (maxima in March and December, Figure 4b [Devol et al., 1995]). Mixing-modeled average TAr values from oceanic cruises (Figure 4d) nicely overlapped with interannual average values of dilution-modeled TAMacapá when backed up by 1 month, allowing transit from the mouth. Modeled values also agree with historical data [Gibbs, 1972; Stallard and Edmond, 1983] during the spring and summer (Figure 4d), but are below measured values during the fall and winter when discharge is lowest. Our single measured 25 March 2002 Macapá TA hand sample (440.71 ± 0.37 μmol kg−1; ±s.d., n = 2 at S = 0.0 ± 0.0, n = 3) was higher than the predicted April 2002 TA (353.5 μmol kg−1; Figure 4d). After applying the TA = 0.82*DIC relationship to the Macapá DIC reported from November 1991 (363 μmol L-1 [Druffel et al., 2005]), a TA of 298 μmol kg−1 was close to modeled TAMacapá (272 μmol kg−1; Figure 4d), but significantly lower than observations from the 1960s [Gibbs, 1972]. Applying the 0.82 ratio to dilution-modeled TA predictions suggests yearly DICMacapá was 360 ± 55 μmol kg−1 (mean ± amplitude), which agrees well with the few available observations [Stallard and Edmond, 1983; Druffel et al., 2005].
 Interannual variability in Andean and total (Macapá) discharge (Figures 4a and 4c) related to precipitation and/or runoff caused small interannual variability in TAMacapá (Figure 4d). In 1991 and 1995, Andean discharge was significantly below the lower 95% confidence interval of the interannual average, and total river discharge was variable. This lower Andean discharge delivered less carbonate alkalinity to the main stem, and TAMacapá was at or below the 95% confidence interval. In 2001, Andean discharges were above the interannual average, but total main stem discharges were average, raising TAMacapá slightly above average levels.
3.4. Biological DIC Drawdown
 Observed variations in WTNA plume DIC not attributable to conservative mixing (ΔDICbio [Cooley and Yager, 2006]) averaged 26.1 ± 3.0 μmol C kg−1 for all seasons, with individual deficits up to 109 μmol C kg−1 (Figure 5a). A Model I linear regression [Sokal and Rohlf, 2000] of individual data plotted against day of year indicates no significant increase in ΔDICbio with time (r = 0.20, n = 76, p = 0.09). We did see a significant difference between cruises (Table 2; KW) due to the systematically lower Sp96 values. Mean seasonal values for ΔDICbio during Sp03, Su01, and F95 were not significantly different (p = 0.16) although data distributions varied (KW p = 0.05). There was no difference in ΔDICbio related to plume S groups (28–30, 30–32, 32–35). Since there was little seasonality in biological carbon drawdown within the 28 < S < 35 outer plume, any observed seasonal inorganic carbon changes must therefore be related to the size, shape, and location of the plume or plume phytoplankton community composition.
 As in Su01 [Cooley and Yager, 2006; Subramaniam et al., submitted manuscript, 2007], ΔDICbio was greatest in Sp03 at stations where Richelia intracellularis/Hemiaulus hauckii diatom-diazotroph assemblages (DDAs) were found. ΔDICbio at the surface correlated significantly with integrated Richelia abundance (r = 0.62, n = 57, p = 0.01). DDAs were observed in moderate S plume waters regardless of season (Subramaniam et al., submitted manuscript, 2007), so large ΔDICbio values in moderate-S, F95 plume waters suggest that DDAs may also have contributed to the biological carbon drawdown observed then. Other diazotrophs, such as Trichodesmium, did not show as large an impact on ΔDICbio in Sp03 or Su01 [Cooley and Yager, 2006].
3.5. Plume Uptake of CO2
 Significant differences (p < 0.01) were observed among all cruises for solubility, observed and mixing-only ΔpCO2, climatological winds, and the observed and mixing-only CO2 fluxes (Table 2). As with ΔDICbio, however, differences in ΔpCO2obs, the gradient (ΔΔpCO2bio, the difference between ΔpCO2obs and ΔpCO2mix), and the flux attributable to biology (ΔFluxbio, the difference between Fluxobs and Fluxmix) were due to the very different Sp96 data set (see Figure 5). Differences between Sp03, Su01, and F95 were not significant for ΔpCO2obs, ΔΔpCO2bio, or ΔFluxbio (Table 2). Significant differences were found only among S groups for ΔpCO2mix and CO2 fluxmix (Table 2), suggesting that the biological impact was not sensitive to S at the level we grouped (28–30, 30–32, 32–35). Because we did not detect significant S-associated differences in most variables, mean values for each season are subsequently presented for the entire 28 < S < 35 zone.
 CO2 solubility (Figure 5b and Table 2) was higher in Sp03 than in Su01 or F95 owing to cooler SST. Solubility during Sp03 was also higher than Sp96, which experienced anomalously warm SST related to Pacific-Atlantic atmospheric teleconnections [Illig et al., 2006]. The overall temperature increase and solubility decrease from spring to fall caused seasons with similar ΔDICbio to have lower ΔpCO2obs later in the year (Figure 5c and Table 3).
 Except during Sp96, values for cruise-averaged ΔpCO2obs were positive (pCO2atm > pCO2sea; Figure 5c) and favored plume uptake from the atmosphere. The air-sea pCO2 gradient expected from mixing only (ΔpCO2mix) was near zero or slightly positive (Figure 5c), favoring very little CO2 uptake in the outer plume. Climatological winds were highly seasonal (Figure 5d) and dropped by about 20% between spring and summer. Spring 1996 wind variability was likely related to the anomalous atmospheric conditions then [Illig et al., 2006].
 Air-sea CO2 fluxes calculated for observed conditions (Fluxobs, to which both mixing and biology contribute) were significantly higher than fluxes expected from mixing only (Fluxmix) in Sp03, Su01 and F95 by 4–7 mmol m−2 d−1 (Figure 5d). Although differences between observed and predicted CO2 fluxes varied somewhat across seasons (with a maximum in Sp03), the main difference was due to spring 1996 (Table 2). Available data does not indicate the exact cause of low spring 1996 ΔDICbio, which caused near zero ΔΔpCO2bio and ΔFluxbio. Neither ΔΔpCO2bio nor ΔFluxbio differed significantly between Sp03, Su01, and F95 (Table 2 and Figures 5c and 5e). Observed fluxes were generally positive, indicating uptake within the 28 < S < 35 zone (auxiliary material Figure S3). No S effect (Table 2) emerged for Fluxobs or ΔFluxbio. Only Fluxmix varied significantly with S. Without significant seasonal changes in biology, observed flux seasonality could be attributed entirely to solubility and wind speed decreases from spring to fall.
 When summed over the entire plume area across all seasons, biological activity enhanced the Amazon outer plume CO2 sink by nearly 100x (Table 3). Under observed (mixing plus biology) conditions, the outer plume took up 1.245 ± 0.474 Tmol C yr−1 (15 ± 6 Tg yr−1; Table 3). If mixing were the only influence, the outer plume would take up 0.015 ± 0.163 Tmol C yr−1 (0.2 ± 2 Tg yr−1; mean ± s.d.), for a difference of ∼15 Tg C yr−1.
4.1. Conservative Versus Nonconservative Alkalinity Changes
 Because we know little about the carbonate system at the Amazon mouth, key assumptions of our analysis are that Andean runoff is the primary source of alkalinity to the main stem, that TA is dominated by carbonate alkalinity, and that TA travels conservatively through the lower main stem and offshore plume. Our models based on these ideas predicted TA at the river mouth by dilution (TAMacapá) and offshore mixing (TAr) with good agreement over the annual cycle (Figure 4d). Our models also agree well with some available data, including spring and summer 1963–1964 [Gibbs, 1972], summer 1976–1977 [Stallard and Edmond, 1983] values near the mouth, and the November 1991 Macapá hand sample [Druffel et al., 2005] after applying the TA/DIC = 0.82 ratio [Devol et al., 1995]. Agreement between seasonal Macapá TA patterns from both models (Figure 4d) suggests that plume TA seasonality can be explained largely by variations in basin drainage rather than by nonconservative behavior. Since most of the flux into the plume occurs between April and August, it is important to get the end-member right during those months.
 Some measurements indicate that river-mouth TA may be above our conservative model predictions, especially during winter. Our April 2002 Macapá hand sample TA was significantly higher than the dilution-model prediction (Figure 4d). This sample of opportunity may differ because it was handled differently from other samples, but storage conditions should have prevented carbonate alkalinity changes. Remineralization of noncarbonate ions such as ammonium and phosphate could have contributed somewhat to the higher measured TA value, but they could not raise TA the 100 μmol kg−1. Historical bicarbonate data at the river mouth [Gibbs, 1972] (see Figure 4D) also suggests that fall and winter TA could be up to 250 μmol kg−1 higher than our predictions. The lack of a recent river mouth time series or winter plume data offshore hinders our ability to test this idea. Significant wintertime alkalinity contributions from lowland rivers could raise TA at the mouth above the prediction, but winter TA data for these tributaries are unavailable. Offshore carbonate-precipitating plankton could have caused us to underestimate river TA by as much as 28% of the maximum ΔDICbio (109 μmol kg−1) or ∼30 μmol kg−1, which is significant, but not enough to make the difference. Likely it is a combination of both factors. An increase in TAr or DICr would similarly raise our estimates of NCP and biologically mediated gas exchange offshore.
 In any case, our analysis affirms that Macapá TA is likely well below the 600 μmol kg−1 estimate used in previous studies (TAr = 600 μmol kg−1 and DICr = 600–744 μmol kg−1 [Ternon et al., 2000; Körtzinger, 2003]) that set expected carbonate system mixing lines far above observed data. These studies explained the difference between mixing and observed regressions by invoking large production-related DIC losses from the early plume [Ternon et al., 2000], possibly supplemented by carbonate equilibria shifts [Körtzinger, 2003]. DIC decreases owing to NCP are very likely [DeMaster and Pope, 1996; DeMaster and Aller, 2001], and apparent DIC deficits of 200 μmol kg−1 (almost twice the maximum of our outer plume ΔDICbio) can indeed be created on the continental shelf with observed inner-plume transit speeds and primary production rates [Ternon et al., 2000]. Less likely are systematic plume TA decreases attributed to biological carbonate production off the shelf [Ternon et al., 2000], because near-surface coccolithophorid abundance relative to other tropical Atlantic phytoplankton is quite low [Aller and Aller, 1986; Kinkel et al., 2000] and they are not favored in the nearshore, high-nutrient zone, where diatoms thrive [Iglesias-Rodriguez et al., 2002]. Neither coccolithophorids nor foraminifera were observed in the offshore plume during our cruises (E. J. Carpenter, personal communication, 2005). Along the entire plume-ocean mixing gradient, phosphate depletion implies that at least 96 μmol kg−1 DIC is biologically removed during transit [Körtzinger, 2003]. Our average NCP (26 ± 3 μmol kg−1) for the outer 20% of the 0–36 S gradient is slightly higher than 20% of this phosphate-associated DIC removal. Thus the enhanced carbon drawdown due to biology calculated with our end-members is in line with biologists' observations and geochemists' predictions.
 Estimates of biological drawdown and flux calculations were also sensitive to our assumptions about the TA:DIC ratio at the river mouth [Richey et al., 1991; Ternon et al., 2000; Cooley and Yager, 2006]. Carbonate equilibria may shift when river and ocean water meet [Körtzinger, 2003], changing the TA:DIC ratio in the early plume, but the ratio is not likely to exceed 1 before S rises above zero. Given that the Druffel et al.  DIC sample fits within our predicted TA range after applying the 0.82 ratio, the assumption seems reasonable until further data can be collected in the river mouth and early plume. Applying a ratio of 1, however, would bring it closer to the bicarbonate observations of Gibbs . The closer the TA:DIC ratio of river water is to 1, the lower its pCO2. If we had used a ratio of 1, our estimated mixing-only fluxes would be greater, decreasing by 30% the mean difference between mixing-only and mixing + biology fluxes from the current 4.8 ± 0.5 mmol C m−2 d−1 (n = 76) to 3.3 ± 0.5 mmol C m−2 d−1 (n = 76), and similarly decreasing our estimated annual biological enhancement of the Amazon plume carbon sink.
4.2. Seasonality in the Carbon Sink
 Inorganic carbon behavior along the river-sea continuum is influenced by both biological and physical processes, which vary enough in other estuarine and coastal systems to affect regional carbon source/sink status over an annual cycle [Cai et al., 2003; Frankignoulle and Borges, 2001; Thomas et al., 2004; Tsunogai et al., 1999]. Seasonal variations in WTNA biological processes would affect the carbon sink by variably enhancing the air-sea pCO2 gradient initially created by mixing, while changes in physical processes will influence gas transfer mechanics by changing piston velocity, solubility, or the amount of time surface water spends in contact with the atmosphere.
 We observed little ΔDICbio seasonality in the outer plume. The ΔDICbio reflects the instantaneous balance between NCP and air-sea replacement [Cooley and Yager, 2006]. Therefore ΔDICbio changes over time may result from a combination of change in both biology and gas exchange. Spring 2003 total primary production rates from outer plume stations (57 ± 7 mmol C m−2 d−1, n = 29; D. Capone, personal communication, 2006) tended to be somewhat (∼15%) higher than but were not significantly different (T-test; p = 0.26) from Su01. Gas exchange rates slowed from spring to fall owing to solubility and wind speed changes, which would suggest that NCP slowed somewhat since ΔDICbio did not change significantly. High primary production rates year-round in the Amazon plume are likely supported by nitrogen fixation [DeMaster and Aller, 2001], and diazotroph species and abundance did not vary seasonally in the offshore plume (Subramaniam et al., submitted manuscript, 2007). Unfortunately, we have no primary production or community composition data from fall 1995 or spring 1996 to compare.
 Our annual calculation of plume uptake is less sensitive to our choice of which cruise's ΔDICbio applies to each month than it is to monthly plume areas and climatological conditions. Thus the most significant “biologically driven” seasonality we detected did not result from seasonal changes per square meter, but rather was caused by scaling biologically influenced fluxes up to the seasonally changing surface area of the plume. Since ΔDICbio varied little with S or season, conditions favorable to diazotroph-supported NCP must be linked to more complex processes, such as nutrient supply (Subramaniam et al., submitted manuscript, 2007), and more research is needed to understand these dynamics and their sensitivity to perturbation.
 Seasonal changes in TAr and DICr within the dilution-modeled annual range would cause small changes in outer plume distribution, and therefore impact our calculations of the biological impact. The Gibbs  data set raises the possibility that a large uncharacterized wintertime carbonate system shift may occur, greatly affecting this system. Our mixing model indicates that outer plume samples contain a small proportion of river water (average = 9% ± 0.5, n = 76). Increasing or decreasing mean TAr and DICr by 2 standard deviations (±212 and ±259 μmol kg−1, respectively; n = 76), for example, would increase or decrease plume TAexp by an average of 0.7% (6 μmol kg−1), and DICexp by 3.4% (59 μmol kg−1). Increased DICexp would lead to a similarly greater ΔDICbio. The nearshore plume (S < 28), where we did not sample, contains a greater (>10%) proportion of river water. DIC changes due to river supply would therefore affect a larger proportion of a plume sample.
 Mixing affects plume DIC primarily by altering air-sea pCO2 gradients. The vertical flux of CO2 into the plume from underlying ocean water was assumed to be zero, because the strong pycnocline between them discourages vertical mixing and attenuates most of the normal upward carbon flux [Cooley and Yager, 2006; Pailler et al., 1999; Sprintall and Tomczak, 1992]. In contrast, lateral mixing between river and ocean water dominates plume DIC distributions [Cooley and Yager, 2006] and causes a seaward decrease in both air-sea pCO2 gradients and fluxes. In general, the plume moves quickly (0.54 m s−1) from the river mouth to the retroflection zone (6°N) and slows thereafter (0.12 m s−1 [Hellweger and Gordon, 2002, and references therein]), with additional speed changes due to seasonal winds and currents [Lentz, 1995a]. Since WTNA air-sea CO2 transfer (0–10 mmol C m−2 d−1, Figure 5e) can lag biological production (2–30 mmol C m−2 d−1 [Maranon et al., 2003]), the faster an individual water parcel travels from the Amazon mouth to the retroflection zone, lesser time is available for replacing DIC deficits. Faster plume speeds may enable a biologically induced carbon deficit to outlast the plume's physical structure. Low-S, low-pCO2 zones extending to 25°W with no evidence of ongoing biological activity [Lefevre et al., 1998; Bakker et al., 1999] point to the far reach of plume effects.
4.3. Implications of Results on Carbon Sequestration
 Primary production based on nitrogen fixation is a key pathway for atmospheric carbon sequestration in the ocean [Eppley and Peterson, 1979]. Diazotrophy-associated production permits the offshore Amazon plume to take up significant quantities of atmospheric carbon. Under observed conditions that include positive NCP, the enhanced sink took up 15 ± 6 TgC yr−1, which agrees well with estimates of diazotroph production in the plume (20.4 TgC yr−1, calculated by rates of nitrogen fixation and a Redfield C:N ratio (Subramaniam et al., submitted manuscript, 2007)) and with sediment trap data suggesting a downward flux of 1.5–15 TgC yr−1 (Subramaniam et al., submitted manuscript, 2007)). Since our outer plume spanned a maximum surface area of 1.3 × 106 km2, this corresponds to an average flux of 2.63 mmol C m−2 d−1, which is nearly double the Körtzinger  estimate of 1.33 mmol C m−2 d−1. This result be partially due to our choices of end-members, but as a single transect, the Körtzinger  estimate appeared less influenced by-productive patches. We believe sampling diazotroph-rich waters in Sp03 and Su01 increased our mean flux estimates, given that DDA production (averaging 93 mmol C m−2 d−1 [Carpenter et al., 1999]) is several times greater than open ocean primary production in the WTNA. Improved precision requires better knowledge of river end-members and better quantification of biological effects by sampling both irregular, patchy offshore blooms and inshore plume waters with S < 28 where production and respiration rates may be much higher.
4.4. Interannual Variability
 Our analysis begins to uncover interannual differences in WTNA inorganic carbon cycling. Spring 1996 data was collected during an anomalously warm period [Illig et al., 2006], while F95 data was collected during a slightly negative phase (El Niño) of the Southern Oscillation Index (SOI; http://www.cpc.noaa.gov/products/precip/CWlink/MJO/enso.shtml). The other samples were collected during near-neutral periods. The western tropical North Atlantic SST remained near normal for most of 1995 and early 1996 [Illig et al., 2006], likely driving normal moisture transport into the Amazon basin in the first months of 1996 (described by Marengo and Nobre ) and leading to near-average Amazon discharge and predicted TA in the spring 1996 data sets (Figure 4). Anomalously warm spring 1996 conditions decreased solubility and increased wind speed variability (Figures 5b and 5d). Spring 1996 also appears to be a period during which NCP and the biological sink enhancement were minimal. Higher temperatures then may have triggered greater plume respiration or lower nutrient levels may have discouraged primary production.
 Future climate-driven changes influencing biological activity, meteorological conditions, or plume size will most affect the Amazon plume carbon sink. Unless the carbonate fraction of river TA changes significantly, variation in riverine TA or DIC will not likely change the sink offshore. Changes in river discharge or nutrient chemistry, however, may indirectly influence the offshore carbon sink through the biological community (Subramaniam et al., submitted manuscript, 2007). Drought in the Amazon basin, as in 2005, may be linked to rising western tropical North Atlantic SST [e.g., Nobre and Shukla, 1996]. Therefore global warming could decrease the Amazon plume carbon sink by shrinking plume surface area and decreasing the air-sea CO2 gradient by reducing solubility. Enhanced blooms or export due to increased nitrogen fixation, biogenic silica generation, or grazing would increase ΔDICbio. Such changes would increase plume CO2 uptake only if they occurred early in the year, when prevailing meteorological conditions enable sink uptake. That warmer spring 1996 conditions could have led to lower ΔDICbio values suggests that interannual changes can decrease biological sink enhancement according to local meteorological conditions. Alternatively, cooler or windier fall conditions, such as those in La Niña years [Wang, 2005], could increase total annual uptake by lowering plume pCO2 and raise the air-sea transfer rate, promoting uptake for a longer period. Higher wind speeds may also encourage faster plume lateral mixing, reducing air-sea contact time and the air-sea CO2 gradient, thereby countering some of the uptake due to a higher piston velocity.
 Although the WTNA is traditionally considered a CO2 source to the atmosphere, biological activity in the Amazon plume permits uptake of an excess 15 ± 6 Tg C yr−1 from the atmosphere due to a combination of biological and physical controls. We examined both riverine and oceanic data to quantify biological and physical controls on plume inorganic carbon. By a seasonally changing combination of Andean carbonate dissolution and lowland dilution, the Amazon River delivers at least 295 ± 61 (mean ± amplitude) μmol kg−1 TA to the river mouth. Modeled alkalinities at the mouth agree very well with historical measurements made during the critical spring and summer months. Assuming DIC was supplied to the plume in proportion to TA, we quantify biological impacts on DIC offshore. Biological production, primarily associated with diatom-diazotroph assemblages (DDA), enables the offshore plume to take up CO2 throughout the year at a rate (2.63 mmol C m−2 d−1) nearly twice that previously reported in studies not likely encountering DDA-favoring conditions. Errors potentially associated with our assumptions about alkalinity would generally increase our estimates of biological production. Even so, our estimates of plume sink magnitude are within range of those determined with other methods [Körtzinger, 2003; Ternon et al., 2000]. The apparent biological enhancement of the plume carbon sink is steady throughout the annual cycle, but air-sea CO2 fluxes are heavily governed by physical factors. As a result, fall conditions are not as conducive to carbon sequestration as those in spring and summer. Changes in meteorological conditions (such as lower temperatures and higher wind speeds) during the fall would enhance plume uptake and have a greater impact on the sink than if they occurred in the spring or summer.
 This research was supported by the Office of Science (BER), U.S. Department of Energy, grant DE-FG02-02ER63472, NOAA Office of Global Programs grant NA16GP2689, and by faculty research support from the University of Georgia to P. L. Y. Financial support was provided to S. R. C. by the University of Georgia Presidential Graduate Fellowship and the NASA Earth System Science Fellowship. A. F. Michaels, D. G. Capone, and E. J. Carpenter graciously invited us into the MANTRA/PIRANA Biocomplexity Project, and with M. Neumann, shared data and provided excellent sampling access. S. Pluvinage assisted with sampling. B. Bergquist collected the Macapá hand sample. C. M. Tilburg assisted with sample analysis. C. E. Tilburg provided technical assistance on the Lanczos data filter. F. le Hingrat at IFREMER provided the ETAMBOT/SABORD data sets. A. Körtzinger kindly shared the Meteor 55 pCO2 data.