The salinity and temperature of the Florida Current are key parameters affecting the transport of heat into the North Atlantic, yet little is known about their variability on centennial timescales. Here we report replicated, high-resolution foraminiferal records of Florida Current surface hydrography for the last millennium from two coring sites, Dry Tortugas and the Great Bahama Bank. The oxygen isotopic composition of Florida Current surface water (δ18Ow) near Dry Tortugas increased 0.4‰ during the course of the Little Ice Age (LIA) (∼1200–1850 A.D.), equivalent to a salinity increase of 0.8–1.5. On the Great Bahama Bank, where surface waters are influenced by the North Atlantic subtropical gyre, δ18Ow increased by 0.3‰ during the last 200 years. Although a portion (∼0.1‰) of this shift may be an artifact of anthropogenically driven changes in surface water ΣCO2, the remaining δ18Ow signal implies a 0.4–1 increase in salinity after 200 years B.P. The simplest explanation of the δ18Ow data is southward migration of the Atlantic Hadley circulation during the LIA. Scaling of the δ18Ow records to salinity using the modern low-latitude δ18Ow-S slope produces an unrealistic reversal in the salinity gradient between the two sites. Only if δ18Ow is scaled to salinity using a high-latitude δ18Ow-S slope can the records be reconciled. Variable atmospheric 14C paralleled Dry Tortugas δ18Ow, suggesting that solar irradiance paced centennial-scale migration of the Inter-Tropical Convergence Zone and changes in Florida Current salinity during the last millennium.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The Yucatan Peninsula [Hodell et al., 2005] and the Cariaco Basin [Haug et al., 2001] experienced unusually dry Little Ice Age conditions. The records can be interpreted as a southward shift of the Atlantic Inter-Tropical Convergence Zone (ITCZ), either as a result of a greater meridional temperature gradient in the North Atlantic or persistent El Niño–like conditions in the equatorial Pacific [Haug et al., 2001]. Shifts in mean ITCZ position affect evaporation and precipitation in the tropical Atlantic and may act to stabilize perturbations to the Atlantic overturning circulation by influencing Gulf Stream salinity [Vellinga et al., 2002].
 Here we present evidence from high-resolution sediment cores that Florida Current salinity increased during the Little Ice Age. The simplest explanation for this pattern is a southward shift of the Atlantic ITCZ and increased evaporation minus precipitation (E – P) in the tropical Atlantic. These data support the suggestion that southward ITCZ migration caused dry LIA conditions in Central and South America [Haug et al., 2001; Hodell et al., 2005] and are consistent with simulations of increased aridity in the tropical North Atlantic in response to Northern Hemisphere cooling [Vellinga et al., 2002; Lohmann, 2003; Vellinga and Wu, 2004]. On the eastern edge of the Florida Current, δ18Ow was anomalously low during the Little Ice Age, perhaps because of southward movement of the Hadley circulation.
2. Oceanographic Setting
 We used four sediment cores for this study, two from each side of the Florida Current, the portion of the Gulf Stream confined to the Straits of Florida. Cores KNR166-2-62MC-A and W167-79GGC were retrieved near Dry Tortugas, on the northwestern edge of the Florida Current (Figure 1). Cores KNR166-2-118MC-A and KNR166-2-125MC-D were retrieved from the Great Bahama Bank, on the eastern edge of Florida Current. Although separated in distance by only 400 km, the surface water at each location represents distinct oceanographic regimes. On the western side of the Florida Current, surface water is fresh because of to the advection of low-salinity waters from the tropical Atlantic [Schmitz and Richardson, 1991]. Today, tropical Atlantic and Caribbean surface salinity is controlled largely by the excess rainfall relative to evaporation under the ascending branch of the Hadley circulation, the ITCZ. Rainfall associated with the annual migration of the ITCZ between 5°S and 15°N [Waliser and Gautier, 1993] is the primary reason surface water salinity is low in this region (Figure 1). This water is advected through the Caribbean and into the Straits of Florida via prevailing wind-driven and meridional overturning transports. Annual average salinity near Dry Tortugas is 36.1 with a seasonal and interannual range of approximately ±0.1 and ±0.2, respectively (J. A. Carton and B. S. Giese, SODA: A reanalysis of ocean climate, unpublished manuscript, 2006, available at http://www.atmos.umd.edu/%7Ecarton/pdfs/carton-giese06.pdf, hereinafter referred to as Carton and Giese, 2006).
 The Great Bahama Bank core sites reflect the influence of both the Florida Current and the Santaren Current, which is relatively salty because of the influence of the subtropical gyre. Higher salinities in the subtropical gyre reflect positive E – P conditions that occur under the dry, descending branch of the Hadley circulation and the convergence of surface water because of Ekman transports. Annual average salinity near the 118MC and 125MC core sites is 36.2 (Carton and Giese, 2006). The seasonal and interannual range in salinity is approximately ±0.05 and ±0.15, respectively (Carton and Giese, 2006), somewhat smaller than the Dry Tortugas region.
3. Methods and Core Top Results
 The hydrographic reconstructions presented here are based on paired measurements of oxygen isotopes and Mg/Ca in fossil shells of planktonic foraminifera. The δ18O of calcite (δ18Oc) precipitated by foraminifera is a function of temperature (T) and δ18O of the seawater (δ18Ow). By using paired measurements of oxygen isotopic composition [δ18Oc ≈ f(T, δ18Ow)] and magnesium concentration [Mg/Ca ≈ f(T)], it is possible to estimate δ18Ow, which is related to salinity through evaporation and precipitation at the ocean surface [Mashiotta et al., 1999]. We measured δ18Oc and Mg/Ca in the surface dwelling foraminifer Globigerinoides ruber in cores from each side of the Florida Current to reconstruct salinity variability over the last millennium.
3.1. Age Control
 Down-core age control is based on multiple planktonic foraminiferal radiocarbon dates in each core (Table 1). Each date is based on an average of 8 mg G. ruber specimens from the >250 μm size fraction. All three multicore core tops in this study have >1 fraction modern radiocarbon, indicating that deposition occurred primarily after 1950 A. D. The one gravity core (79GGC) has a 14C core top age of 200 years [Lund and Curry, 2004], also indicating the presence of bomb radiocarbon, but to a lesser extent than the multicores. Raw radiocarbon ages were converted to calendar ages using CALIB 4.3 [Stuiver et al., 1998], assuming a surface ocean reservoir age of 400 years. Average sedimentation rates for the cores range from 21 to 42 cm/kyr (Table 1). We sampled 118MC and 125MC every 1 cm, equivalent to time intervals of 24 and 48 years, respectively. Samples were taken every 0.5 cm in 62MC and 79GGC, or every 16 and 20 years, respectively.
Table 1. Radiocarbon Ages for Cores KNR166-2-118MC-A, KNR166-2-125MC-D, and KNR166-2-62MC-A
Sample Depth Range, cm
Raw 14C Age, years
Analytical Error, years
Calendar Age, years B.P.
Average Sedimentation Rate, cm/kyr
Average Sampling Interval, years
Core tops contain more radiocarbon than expected in modern samples, indicating contamination with bomb 14C and post-1950 A.D. deposition ages. We calibrated the down-core raw 14C values to calendar ages using CALIB 4.3 [Stuiver et al., 1998] and a reservoir age of 400 years. Each date is based on G. ruber from the >250 μm size fraction. Average sedimentation rate is 25 cm/kyr and average sampling interval is 20 years for core W167-79GGC. See Lund and Curry  for the W167-79GGC age model.
 To address the potential bias of running δ18O and Mg/Ca on separate groups of foraminifera, we decided to use one common population of approximately 100 individual G. ruber specimens (212–250 μm size fraction) from each depth interval in 62MC, 118MC, and 125MC. Each sample was sonicated in methanol, siphoned, dried at 70°C, crushed, weighed, and then split for isotopic and Mg/Ca analyses. The crushed sample was evenly distributed on a tray, and split using a fine brush to ensure the subsamples contained approximately equal portions of fine and coarse material. The weight equivalent of 40 G. ruber tests was removed and then split into four separate samples for stable isotopic determinations. We set the remainder aside for Mg/Ca analyses.
3.3. Stable Isotopes
 Stable isotope analyses for 79GGC were run on a Finnigan MAT 252 coupled to a Kiel II carbonate device. Analyses for 62MC, 118MC, and 125MC were run on a Finnigan MAT 253 coupled to a Kiel III carbonate device. Calibration to VPDB scale was made using NBS-19 (δ18O = −2.20‰ and δ13C = 1.95‰). Reproducibility (1σ) of NBS-19 (n = 461) for the MAT 253 system is ±0.08‰ for δ18O and ±0.04‰ for δ13C. The 1σ values for NBS-19 on the MAT 252 were ±0.07‰ for δ18O and ±0.03‰ for δ13C (n > 2200 [Ostermann and Curry, 2000]).
 The isotope splits for 62MC, 118MC, and 125MC can be used to determine the degree of sample heterogeneity caused by the splitting procedure. If each isotope split represents the δ18O of the sample population, then the variance of the residual δ18O at each depth should be similar to the variance of the external standards. Residual values were determined by subtracting the mean δ18O value at each depth from the individual measurements. The standard deviation of residual values for the cores is 0.09‰, close to the analytical error for NBS-19. The similarity of these statistics indicates that our splitting procedure created subsamples that are representative of the G. ruber population in each sample. A similar comparison is not possible for 79GGC since the variance of δ18O analyses based on multiple measurements of 10 individuals is dominated by sampling rather than analytical uncertainty [Lund and Curry, 2004].
3.4. Mg/Ca Analyses
 The Mg/Ca analyses were based on either 50–60 individual foraminifera (79GGC) or their weight equivalent from a sample of 100 foraminifera (62MC, 118MC, and 125MC). To minimize bias due to matrix effects [Rosenthal et al., 1999], we weighed each Mg/Ca split individually to ensure it fell between 150 and 200 μg. Each sample was cleaned using a sequential three-step process, including clay removal, metal oxide reduction, and organic matter oxidation [Boyle and Keigwin, 1985; Rosenthal et al., 1997]. All samples were run on a Finnigan Element 2 single collector ICP-MS.
 Samples were corrected for drift using a 2% HNO3 blank and a 50 ppm calcium standard. Analytical error was determined using three external standards with Mg/Ca ratios of 1.6, 3.2, and 4.8 mmol/mol. The analytical precision (1σ) for each standard was ±0.03 (n = 12), ±0.04 (n = 11), and ±0.05 mmol/mol (n = 11), respectively. Mg/Ca ratios for the unknowns were typically in the 4 to 5 mmol/mol range (Figures 2 and 3). The average precision (1σ) for Mg/Ca splits, which reflects both analytical and sample processing uncertainty, is ±0.12 mmol/mol (n = 82 pairs). Poorer precision for the unknowns primarily reflects sample heterogeneity due to the cleaning procedure. A Mg/Ca analytical precision of ±0.05 mmol/mol translates to an analytical temperature uncertainty of approximately ±0.1°C (1σ), whereas the combined analytical and sample processing precision (±0.12 mmol/mol) corresponds to a SST uncertainty of ±0.25°C (1σ).
 We converted Mg/Ca values to sea surface temperature using the general calibration equation of Anand et al. , which is based on sediment trap data from 10 planktonic foraminiferal species [Mg/Ca(mmol/mol) = 0.38*exp(0.09*SST(°C))]. We chose this equation because it is not subject to dissolution biases of core top calibration data sets and is therefore more appropriate for the shallow (<700 m) core sites discussed in this paper. We used the generalized sediment trap equation because the errors associated with its preexponential and exponential constants are smaller than for the G. ruber–specific calibration [Anand et al., 2003].
 Calculated Dry Tortugas core top SST values are approximately 27.5 (62MC) and 28.5°C (79GGC) (Figures 4b and 4e). The annual average sea surface temperature at these sites is 26.9°C (Carton and Giese, 2006). On the Great Bahama Bank, the core top Mg/Ca-derived SST for 118MC and 125MC is 28.0°–28.5°C (Figures 5b and 5e), compared to an annual average sea surface temperature of 27.2°C (Carton and Giese, 2006). Taking into account the ±1°C (1σ) calibration uncertainty [Anand et al., 2003], it appears that foraminiferal Mg/Ca produces reasonable absolute temperature estimates for these locations. All four core top values are warmer than observed SSTs, however, a discrepancy that is only amplified if G. ruber calcifies closer to 25 m water depth [Fairbanks et al., 1982; Deuser, 1987], where temperatures are ∼0.5°C cooler than at the surface (Carton and Giese, 2006).
 The calibration based on multiple species may create a warm bias in the SST estimates. If we instead use the calibration for G. ruber, [Mg/Ca(mmol/mol) = 0.34*exp(0.102*SST(°C))] [Anand et al., 2003], the time series become 2°C cooler than in the general calibration case, and the core tops fall into better agreement with modern observations. The ∼1°C offset between 62MC and 79GGC SSTs is likely due to the strong temperature gradient along the northern edge of the Florida Current which exposes 79GGC to warmer waters and yields higher Mg/Ca–derived SSTs. Spatial SST variability at this scale (∼10 km) is not captured by the 0.5° × 0.5° gridding of the Carton and Giese (2006) database.
3.5. The δ18Ow Estimates
 We calculated past changes in δ18Ow using the isotopic fractionation versus temperature relationship from Bemis et al.  as presented by Lynch-Stieglitz et al. . Similar results were obtained using the Kim and O'Neil  paleotemperature equation. To evaluate whether the δ18Ow values are reasonable, we compare the core top δ18Ow to estimates of surface water δ18Ow based on modern δ18Ow-salinity observations. Since the Dry Tortugas sites are heavily influenced by tropical Atlantic surface waters, we used a δ18Ow/salinity regression based on data from the open ocean tropical Atlantic [δ18Ow(‰) = 0.26*S(psu) − 8.44] (G. A. Schmidt et al., Global seawater oxygen 18 database, 1999, available at http://www.giss.nasa.gov/data/o18data/, hereinafter referred to as Schmidt et al., 1999). Assuming annual average salinity values of 36.1 near Dry Tortugas (Carton and Giese, 2006), the predicted core top δ18Ow is 0.95‰, similar to the 62MC and 79GGC core top δ18Ow values of 0.8 ± 0.1‰ and 1.1 ± 0.2‰, respectively (Figures 4c and 4f). For the Great Bahama Bank sites, the annual average salinity (36.2 (Carton and Giese, 2006)) and subtropical gyre δ18Ow/salinity regression [δ18Ow(‰) = 0.17*S(psu) − 5.13] (Schmidt et al., 1999) yield a modern δ18Ow of 1.02‰. This value is slightly higher than the 0.7 ± 0.1‰ estimate for the 118MC and 125MC core tops (Figures 5c and 5f).
 Although uncertainty in each core top and surface water δ18Ow value may account for some of the discrepancy, three of the four core top values underestimate surface water δ18Ow (62MC, 118MC, and 125MC). One potential source of systematic error is the use of crushed foraminiferal tests for stable isotope analyses. The δ18O values based on crushed foraminifera from 62MC are consistently 0.2‰ lower than δ18O values of whole foraminifera from the same samples (Figure 6). We observe no significant offset in δ13C between the whole and crushed samples, so it is unlikely due to fractionation of CO2 in the mass spectrometer. Keigwin and Jones  found a similar pattern: the δ18O of ground foraminifera was 0.2‰ lower than the δ18O of whole foraminifera, with no obvious difference in δ13C values. This offset is likely due to either variable partitioning of oxygen isotopes among the products of the calcite-phosphoric acid reaction or diffusion of reaction products away from the particle surface associated with the production of CO2 bubbles [Walters et al., 1972; Keigwin and Jones, 1990]. Given that no fractionation occurs when crushed samples are acidified under elevated pressure to prevent foaming, it appears that the fractionation is most likely tied to the rate at which gas evolves during acidification [Walters et al., 1972]. Swart et al.  suggest that production of gas bubbles may encourage isotopic exchange between CO2 and phosphoric acid, thus affecting the average isotopic value of the evolved carbon dioxide.
 The apparent grain-size isotopic effect is analogous to temperature-driven changes in the fractionation factor. As the temperature and the reaction rate increase, the fractionation factor (18O/16Ocalcite/18O/16OCO2) decreases, leading to lower δ18O values [e.g., Swart et al., 1991]. For reasons that remain unclear, faster reaction rates associated with crushed samples apparently lead to a greater proportion of 16O trapped as CO2 and lower δ18O. Therefore the low δ18Oc and δ18Ow in 62MC relative to 79GGC are most likely a function of crushing the 62MC samples prior to analysis. If the ∼0.2‰ offset between the cores were real, it would imply a salinity difference similar to the contrast observed between the subtropical gyre and tropical North Atlantic (∼1 psu) (Figure 1). Given that 62MC and 79GGC are located only 10 km apart, it is highly unlikely that their surface water δ18Ow values are offset to this degree. Therefore we conclude that the offset between core top and modern surface water δ18Ow can largely be attributed to the use of crushed samples for stable isotope analyses. Since the isotopic fractionation versus temperature relationship from Bemis et al.  is based on whole foraminifera, 0.2‰ should be added to the δ18Oc data to compensate for the effect of crushing. Applying this correction to 62MC produces a core top δ18Ow value of 1.0 ± 0.1‰, indistinguishable from the modern seawater estimate (0.95‰). Adding 0.2‰ to the 118MC and 125MC data yields core top δ18Ow values of 0.9 ± 0.1‰, also very similar to the seawater estimate for these sites (1.02‰).
4. Down-Core Results and Discussion
4.1. Surface Water δ18Ow
 Planktonic δ18Oc and SST estimates from 62MC and 79GGC yield consistent changes in Florida Current δ18Ow during the last 1000 years. In 62MC, G. ruber δ18Oc increased by approximately 0.2‰ over the past 900 years, with the largest change occurring from 400 years B.P. to present (Figure 4a). The δ18Oc increase is significant at the 95% confidence level and implies that Florida Current surface waters became progressively cooler, saltier, or both. Mg/Ca analyses of G. ruber indicate that SST was stable at 24.5°C from 900 to 500 years B.P., and then increased monotonically to ∼25.5°C by 200 years B.P. (Figure 4b). This SST maximum was followed by an abrupt 1°C decrease from 200 to 100 years B.P., and then a steady increase over the past 100 years. Reconciliation of the δ18Oc and SST records requires δ18Ow variability on the order of 0.4‰ (Figure 4c). Foraminiferal δ18Oc in nearby core 79GGC was nearly constant from 1400 to 500 years B.P. but then increased by 0.2–0.3‰ at 500 years B.P. (Figure 4d). Similar to 62MC, Mg/Ca–derived SSTs for 79GGC generally increase from 800 to 200 years B.P., although 79GGC is 0.9°C warmer on average over the past 800 years. This difference is similar to the core top offset which reflects the strong SST gradient near Dry Tortugas (section 3).
 The 79GGC time series extends the Dry Tortugas δ18Ow record to 1400 years B.P. and shows a distinct 0.1–0.2‰ increase at 900 years B.P., in addition to the larger 0.4‰ shift at 400 years B.P. (Figure 4f). Although the absolute values for the Dry Tortugas δ18Ow records are different, the magnitude and timing of their δ18Ow variability is nearly identical. The parallel changes in the two records support their fidelity and indicate that each provides a robust signal of δ18Ow variability at this location. The 0.2‰ offset between 62MC and 79GGC is caused by the difference in δ18Oc from using whole versus crushed foraminifera for the stable isotopic analyses in these two cores (section 3).
 On the Great Bahama Bank, significant changes in δ18Oc and SST occurred during the past millennium, but with different timing than the Dry Tortugas sites. The largest change in 118MC δ18Oc occurred from 200 years B.P. to present (Figure 5a). A similar shift occurred about 100 years earlier in the 125MC record (Figure 5d). Since 118MC has better sampling resolution and chronological control than 125MC, the δ18Oc increase most likely occurred after 200 years B.P. Mg/Ca–derived SSTs in 118MC increased gradually by 0.5°C from 500 years B.P. to present (Figure 5b). Core 125MC extends the Great Bahama Bank SST record to 1400 years B.P. and shows a ∼0.5°C cooling at 800 years B.P., near the beginning of the LIA, and then a ∼0.5°C warming at ∼300 years B.P. (Figure 5e). The combined δ18Oc and SST records require that δ18Ow along the Great Bahama Bank increased by ∼0.3‰ during the past 200 years (Figures 5c and 5f).
4.2. Sea Surface Temperatures
 The Great Bahama Bank results are inconsistent with foraminiferal SST estimates from the Sargasso Sea and the magnitude of LIA cooling inferred from Caribbean corals. Rather than increasing after 400 years B.P., SSTs in the Sargasso Sea apparently decreased [Keigwin, 1996]. This discrepancy may be due to the use of different SST proxies: δ18Oc for the Sargasso Sea and Mg/Ca for the Great Bahama Bank. The data presented here suggest that high Sargasso Sea δ18Oc after 200 years B.P. may have been a result of high δ18Ow rather than low SSTs. Estimates from Caribbean corals indicate SSTs were 1°–3°C cooler than today during the coldest LIA intervals [Winter et al., 2000], significantly larger than the cooling along the Great Bahama Bank. The disagreement between the coral and Bahamas SST estimates may be due to the brief (∼10 years) time windows covered by the corals and signal smoothing in the foraminiferal records caused by bioturbation.
 Interestingly, both Great Bahama Bank and Dry Tortugas SSTs warmed by ∼0.5°C during the course of the LIA, from about 800 years B.P. to 200 years B.P. (Figure 7). As a result, the ∼1.0°C cross-current SST gradient remained nearly constant during this time. Prior to 800 years B.P. and after 200 years B.P., however, the cross-current gradient was 1.5°–2.0°C, significantly larger than during the LIA interval. We believe this most likely reflects variable formation of Tortugas eddies, which today can shift SSTs at the 62MC and 79GGC core sites by >3°C for months at a time [Fratantoni et al., 1998] (see also APL Ocean Remote Sensing, Gulf of Mexico region, available at http://fermi.jhuapl.edu/avhrr/gm/index.html). Therefore the LIA may have been characterized by reduced Tortugas eddy formation relative to the preceding and subsequent time intervals. This inference is opposite that based only on δ18Oc from the Dry Tortugas sites [Lund and Curry, 2004]. The apparent SST differences in 62MC and 79GGC over the past 600 years (Figure 7) are not significant at the 95% confidence level (Figures 4b and 4e).
4.3. SST and δ18Ow Error Analysis
 We estimated the 95% confidence limits for SST in two ways (1) based on variance in the raw Mg/Ca data only, and (2) based on both the sample variance and calibration uncertainty. In the latter case we used a Monte Carlo approach that accounts for both measurement error and uncertainty in the preexponential and exponential constants for the general Mg/Ca–SST calibration [Anand et al., 2003]. The resulting 95% confidence limits of approximately ±2°C (not shown) are dominated by the calibration uncertainty. Given the down core SST variability can be replicated to within 0.5°C (Figure 7), we believe incorporating the absolute calibration uncertainty yields artificially conservative errors for estimating relative changes in SST. The calibration error, which largely reflects uncertainty in the calcification temperatures used for calibration [Anand et al., 2003], appears to be most useful for comparing core top SST estimates with modern observations.
 The error envelopes for the δ18Ow time series in Figures 4 and 5 are based on Monte Carlo estimates, where at each depth in each core, δ18Oc and SST values were randomly sampled from a distribution constrained by the 95% confidence limits for δ18Oc and SST. For example, the error for 62MC δ18Ow (Figure 4c) is based on random data from within the error envelopes in Figures 4a and 4b. This approach, which excludes calibration error, yields 95% confidence limits for δ18Ow of approximately ±0.1‰. If Mg/Ca–SST calibration error is included, then the uncertainty for δ18Ow expands to approximately ±0.4‰ (95% confidence limit), similar to that calculated by Schmidt et al. [2004a] for data from the Caribbean.
 Whether the δ18Ow variability in 62MC is significant depends on the method used to calculate confidence limits. If we incorporate the calibration error, then the δ18Ow signal does not exceed the 95% confidence level. At the one-sigma level (± 0.2‰), however, the δ18Ow increase from 700 to 200 years B.P. is significant (Figure 4c). If we choose to exclude the Mg/Ca–SST calibration error, then smaller-scale features, such as high δ18Ow at ∼200, 400, and 550 years B.P., begin to appear. Similar arguments can be made for 79GGC, 118MC, and 125MC. Overall, the ±0.4‰ uncertainty for δ18Ow appears to be too conservative for these core sites. For the remainder of this paper, we assume the down core errors for δ18Ow (and SST) are better represented by the confidence limits based on sampling uncertainty (Figures 4 and 5). Although this is not entirely satisfactory, the issue can only be resolved through refinement of the Mg/Ca–SST calibration and verification of the records presented here using additional high-resolution δ18Ow time series. We are encouraged, however, by the strong similarity of the replicate δ18Ow records at both Dry Tortugas and the Great Bahama Bank (section 5.2).
4.4. Carbonate Ion and 13C Suess Effects
 Great Bahama Bank δ18Oc (and hence calculated δ18Ow) may be influenced by decreasing surface water pH associated with the input of anthropogenic CO2 over the past 200 years. As pH decreases, the greater proportion of bicarbonate to carbonate ion leads to isotopically heavier dissolved inorganic carbon, since bicarbonate is the more enriched of the two species [Zeebe, 1999]. In laboratory culturing experiments, G. ruber δ18Oc varies systematically with [CO32−] by a slope of −0.002 ± 0.001‰/(μmol/kg) [Spero et al., 1999]. During the last 200 years, surface North Atlantic subtropical gyre dissolved inorganic carbon (DIC) has increased ∼60 μmol/kg because of anthropogenic CO2 emissions [Sabine et al., 2004]. Assuming constant near-surface alkalinity between the LIA and present (2400 μEq/kg; GEOSECS station 31), we estimate that [CO32−] decreased by approximately 40 μmol/kg over this time interval (see http://geosci.uchicago.edu/~archer/cgimodels/pco2.html), equivalent to a carbonate ion isotope effect on G. ruber δ18Oc of +0.08 ± 0.04‰. Therefore approximately 0.1‰ of the δ18Ow increase from the LIA to today may be attributed to decreasing surface water pH. If this is true, then the remainder (0.2‰) reflects increasing salinity of the surface North Atlantic subtropical gyre since the LIA.
 Since carbonate ion concentration affects both the oxygen and carbon stable isotopes of foraminiferal calcite, it should be possible to verify the effect of pH on δ18Oc using δ13Cc. Using a [CO32−] decrease of 40 μmol/kg and a G. ruber δ13Cc/[CO32−] relationship of −0.009 ± 0.001‰/(μmol/kg) [Spero et al., 1999], the expected shift in δ13Cc from LIA to present would be +0.35‰, similar to that observed in 118MC and 125MC (Figure 8 top). However, if G. ruber δ13Cc reliably records δ13C of dissolved inorganic carbon (δ13CDIC), it should reflect not only pH variability but also the influence of 13C-depleted CO2 from fossil fuel and biomass burning (the 13C “Suess effect”). Surface Atlantic δ13CDIC decreased by 0.8‰ over the past 200 years [Bohm et al., 1996], consistent with model estimates of mixed layer δ13CDIC driven by decreasing δ13C of atmospheric CO2 [Francey et al., 1999]. Taking into account both the carbonate ion and Suess effects, the change in G. ruber δ13Cc from 200 years B.P. should be approximately −0.4‰, as opposed the observed signal of +0.3–0.4‰.
 The 0.8‰ offset between predicted and observed δ13Cc is either due to real changes in δ13Cc or is an artifact of G. ruber vital effects. Increasing surface layer ∑CO2 since the LIA would likely enhance biological carbon isotope fractionation and increase δ13Cc [Jasper and Hayes, 1990; Hofmann et al., 1999]. An increase in export production of 13C-depleted organic matter could also enrich mixed layer 13C of DIC. The situation is further complicated by the tendency of G. ruber δ13Cc to be lower than ambient seawater δ13CDIC. This offset is most likely a function of foraminiferal respiration, metabolic and growth rates, and symbiont photosynthesis, all of which are ecologically sensitive to oceanographic variability [Curry and Matthews, 1981; Spero and Lea, 1993; Russell and Spero, 2000]. Regardless of the exact cause, it is clear G. ruber δ13Cc on its own cannot be used to validate the effect of pH on δ18Oc.
 The problematic nature of using G. ruber δ13Cc as a proxy for δ13CDIC is further highlighted by the clear existence of the 13C Suess effect in benthic foraminiferal δ13C records from the Great Bahama Bank (Figure 8 bottom). At water depths ranging from 250 to 700 m, four separate benthic time series show anomalous 13C depletion during the past 200 years. The decrease in δ13Cc near the core tops is unprecedented during the last millennium and the largest 13C depletion occurs in the shallowest cores (<500 m), consistent with a surface ocean source of 13C-depleted CO2. This is the expected response to input of anthropogenic CO2 into midlatitude to high-latitude North Atlantic surface waters and subsequent thermocline ventilation and transport along isopycnals into the subtropical gyre.
5. Sources of Salinity Variability
 Florida Current δ18Ow variability during the last millennium was most likely due to large-scale changes in oceanic precipitation and evaporation. While variable flow of the Mississippi River and altered routing of oceanic currents may have played a minor role, the observed δ18Ow signal is too large to be explained by these mechanisms. Instead, southward migration of the Atlantic Hadley circulation during the Little Ice Age, in combination with a mean shift in thermocline δ18Ow, are the most likely mechanisms responsible for changes in δ18Ow near Dry Tortugas and the Great Bahama Bank.
5.1. Local/Regional δ18Ow Effects
 Input of isotopically light Mississippi River water may have influenced the Dry Tortugas δ18Ow records, but only to a small extent. Today, the Mississippi River has an annual average flow rate of 0.015 Sv [Walker et al., 1994] and a δ18Ow signature of −7‰ SMOW [Coplen and Kendall, 2000]. Transport in the upper 100 m of the Florida Current is approximately 9 Sv [Schmitz and Richardson, 1991] with a δ18Ow signature of ∼0.9‰ (section 3).
 If the isotopic signature of the oceanic and fluvial end-members is held constant, the δ18Ow increase in 62MC from 700 to 200 years B.P. (0.4‰) would require Mississippi discharge to change by a factor of 33 (e.g., 0.015 Sv at 200 years B.P. and 0.5 Sv at 700 years B.P.). During the 2-month peak of the 1993 flood event, one of the most severe floods of the last century, the Mississippi discharge rate was 0.03 Sv [Walker et al., 1994], a doubling of the average rate, making even the 1993 flood too small to influence Florida Current δ18Ow in a significant way. Given the unrealistic river flows required to create the observed changes in δ18Ow, and that these flow rates would have to persist for centuries, it is highly unlikely that variability in Mississippi River discharge significantly influenced Florida Current δ18Ow during the last millennium.
 On the Great Bahama Bank, variable current routing may have affected surface water δ18Ow because the 118MC and 125MC core sites are located near the confluence of the Santaren and Florida Currents. Today, the Santaren Current supplies on average approximately 2 Sv of warm, salty water to the main flow of the Florida Current [Atkinson et al., 1995; Leaman et al., 1995]. Lower Santaren Current flow during the LIA would likely result in lower salinity at the core sites. Despite the 0.2 psu difference in salinity on either side of the Florida Current, their δ18Ow values are nearly equal (section 3). Therefore reduced flow of the Santaren Current relative to the Florida Current during the LIA would not necessarily lead to lower surface water δ18Ow along the Great Bahama Bank. If anything, increased Florida Current salinity during the LIA should lead to saltier conditions at the 118MC and 125MC core sites. What is remarkable is that LIA salinities on the GBB are anomalously low relative to core top values, despite the proximity of the cores to the Florida Current (Figure 1).
5.2. Hadley Cell Migration
 The salinity changes on the Great Bahama Bank were very different than those at Dry Tortugas during the past 1200 years. From 400 to 200 years B.P., when δ18Ow was higher than today near Dry Tortugas, it was lower than today on the Great Bahama Bank (Figure 9). Beginning at 200 years B.P., δ18Ow increased by ∼0.2‰ on the Great Bahama Bank, while it decreased by 0.2–0.3‰ at Dry Tortugas. The largest cross-current difference in δ18Ow occurred during the Little Ice Age, while the smallest offsets occurred from approximately 0 to 100 years B.P. and 900 to 1200 years B.P. North-south movement of the Atlantic Hadley circulation is the most likely explanation of the observed δ18Ow variability.
 Given that Florida Current surface salinity is heavily influenced by E – P in the tropical Atlantic, reduced precipitation in this region during the Little Ice Age is the most logical cause of the observed changes. This interpretation is consistent with anomalously dry LIA conditions in the Yucatan Peninsula [Hodell et al., 2005] and northern South America near the Cariaco Basin [Haug et al., 2001]. Comparison of Dry Tortugas δ18Ow to the aridity record from the Cariaco Basin suggests a strong climatic link between the two (Figure 10). Cariaco Basin percent Ti, which is proportional to fluvial input, shows a long-term decrease beginning ∼600 years B.P., with distinct minima clustered between 400 and 150 years B.P. [Haug et al., 2001]. The long-term drying of the Cariaco Basin region ended abruptly at 200 years B.P. Overall, the changes observed in 62MC and 79GGC δ18Ow parallel those in Cariaco Basin percent Ti. One mechanism that could cause this common change is southward ITCZ migration, which would dry the watershed for the Cariaco Basin [Haug et al., 2001], and remotely force increased salinity in the Florida Current.
 On an interannual basis, southward migration of the Atlantic ITCZ can be caused by either an increased Atlantic cross-equatorial SST gradient or El Niño conditions in the equatorial Pacific [Chiang et al., 2002]. Even slight (∼0.2°C) cooling of the North Atlantic can force southward migration of the ITCZ, which in turn reduces precipitation in the tropical North Atlantic and creates a positive salinity anomaly in this region [Vellinga et al., 2002; Vellinga and Wu, 2004]. The Florida Current δ18Ow maxima and Cariaco Basin percent Ti minima from 400 to 150 years B.P. coincide with an interval of unusually cool conditions in several Northern Hemisphere records [Lamb, 1995; deMenocal et al., 2000; Dahl-Jensen et al., 1998; Huang et al., 2000]. This correspondence suggests that LIA cooling in the North Atlantic enhanced the cross-equatorial SST gradient and triggered a southward shift in mean ITCZ position.
 Beginning at ∼900 years B.P., δ18Ow increased on the Florida Margin and decreased on the Great Bahama Bank, potentially reflecting southward migration of the Hadley cell and increased E – P in the tropical Atlantic and decreased E – P in the subtropical gyre. Several paleoclimate records indicate cooling in the North Atlantic region at about this time [Dahl-Jensen et al., 1998; Esper et al., 2002; deMenocal et al., 2000]. The sharp decrease in δ18Ow after 200 years B.P. near Dry Tortugas and increase on Great Bahama Bank is consistent with warming since the LIA and an abrupt movement of the Hadley circulation to the north, causing relatively higher salinity in the subtropical gyre and lower salinity in the tropical Atlantic.
5.3. Surface Ocean Versus Thermocline δ18Ow-S
 The modern relationship between δ18Ow and salinity in the tropical and subtropical Atlantic surface waters is described by a δ18Ow/S slope of approximately 0.2‰ per unit salinity change (Figure 11a). In the subtropical gyre, δ18Ow is high relative to the tropics since evaporation discriminates against 18O. The resulting 16O-enriched vapor rains out in the tropics and high-latitude Atlantic. At high latitudes, δ18O decreases because of the preferential condensation of 18O at cooler temperatures, making the δ18Ow/S slope steeper. Since these waters (up to ∼45°N) ventilate the thermocline, the thermocline δ18Ow/S relationship (0.5‰ per unit salinity change) is similar to that for the surface high-latitude North Atlantic.
 Today, the Dry Tortugas and Bahamas sites differ by ∼0.1‰ in δ18Ow and ∼0.2 in salinity (Figure 11a). During the LIA, δ18Ow increased by 0.1‰ at Dry Tortugas and decreased by ∼0.2‰ at the Great Bahama Bank. If this was a function of simply moving along the low-latitude δ18Ow-S slope, this would imply LIA salinities of 36.5 at Dry Tortugas and 35.7 at the Bahamas (Figure 11b). This is unrealistic since it requires the surface North Atlantic subtropical gyre to be fresher than the tropical Atlantic (i.e., a reversal of the Hadley cell distribution of E – P). A more reasonable scenario involves southward migration of the Hadley circulation and its associated E – P patterns (Figure 11c). In this case, the tropical Atlantic becomes saltier along the 0.2‰ slope, but the subtropical gyre becomes fresher through greater influence of 18O-depleted surface water from higher latitudes (equivalent to the thermocline δ18Ow-S line). The situation depicted in Figure 11c allows δ18Ow at Great Bahama Bank to be lower than Dry Tortugas and minimizes the LIA salinity difference between the two sites. Although less extreme than the first case, this scenario still requires a salinity gradient the reverse of today.
 Surface water δ18Ow is not only a function of evaporation and precipitation, but also the δ18Ow of water with which it mixes. For the surface North Atlantic, this mixing component is the thermocline. If thermocline δ18Ow and salinity varied during the LIA, it would likely be along its modern slope of 0.5‰. Since the thermocline dominates the surface layer from a volumetric standpoint, the entire low-latitude surface Atlantic δ18Ow/S relationship would also move along a thermocline slope to retain mass balance [Craig and Gordon, 1965]. In this way, a 0.1‰ increase in mean thermocline δ18Ow could account for the 0.1‰ increase at Dry Tortugas during the LIA (Figure 11d). The thermocline scenario would also require a positive shift in Great Bahama Bank δ18Ow, however, opposite that observed.
 The scenario most consistent with the sign and magnitude of δ18Ow variability in our records involves a combination of the thermocline and Hadley circulation models (Figure 11e). Higher mean thermocline δ18Ow reduces the implied LIA salinity change at Dry Tortugas and southward migration of the Hadley cell and provides a mechanism for introducing 18O-depleted surface waters into the subtropical gyre. This approach avoids the pitfall of a reversed salinity gradient by moving each site in opposite directions along lines with 0.5‰ slopes. In fact, the only way to accommodate the opposing changes in δ18Ow on either side of the Florida Current is to invoke the influence of high-latitude and thermocline waters. Using today's low-latitude δ18Ow-S relationship to estimate past changes in salinity yields unreasonable results (Figure 11b). If we use instead the high-latitude slope, then LIA salinities at Dry Tortugas and Great Bahama Bank were nearly equal. This does not mean that salinities in the subtropical gyre and tropical Atlantic were also equal. Today the salinity offset between these regions is on the order of ∼1, whereas the offset between the coring sites is only ∼0.3. During the LIA therefore the salinity gradient between the tropics and subtropical gyre was likely lower than today, but probably not zero.
5.4. El-Niño Southern Oscillation
 Of course, the North Atlantic does not act independently of the remainder of the climate system. Anomalous atmospheric convection in the eastern equatorial Pacific during El Niño events forces subsidence and decreased precipitation over the tropical North Atlantic [Chiang et al., 2002]. El Niño events also cause the main locus of rainfall to migrate into the central equatorial Pacific, away from its more typical location in the western equatorial Pacific (WEP) (see http://www.cdc.noaa.gov/ENSO). Coral δ18O records show a decreased equatorial Pacific zonal SST gradient during the LIA, possibly caused by more frequent El Niño events during this time [Hendy et al., 2002; Cobb et al., 2003]. If the interannual model can be applied to longer timescales, El Niño–like conditions would lead to decreased precipitation in the northern tropical Atlantic, as is observed, but also in the WEP. Available evidence shows that this did not occur. Low δ18Ow in the WEP during the LIA [Stott et al., 2004] is inconsistent with the simple interannual model (Figure 10). Either ENSO played little role in Atlantic ITCZ movement or the MD81 δ18Ow record reflects climatic processes other than ENSO.
5.5. Role of Solar Variability
 Reduced sunspot counts [Eddy, 1976], higher atmospheric Δ14C (Δ14Catm) [Stuiver et al., 1991] and greater 10Be flux [Bard et al., 2000] all imply solar irradiance was lower than today during the LIA. High Δ14Catm during the Maunder, Spörer, and Wolf sunspot minima matches intervals of high Florida Current δ18Ow and low western equatorial Pacific δ18Ow (Figure 10). The δ18Ow records appear to lag Δ14Catm by 50–100 years, assuming surface reservoir ages were similar to modern values over the past 1200 years. A similar lag is observed when comparing δ18Ow with modeled estimates of radiocarbon production rate [Bond et al., 2001; Marchal, 2005]. If this lag is correct, then it appears that solar variability paced not only migration of the North Atlantic subpolar front [Bond et al., 2001] but also centennial-scale hydrologic variability in both the tropical Atlantic and Pacific.
 Numerical simulations of the evaporation-precipitation (E – P) response to reduced solar irradiance lack the positive E – P anomaly in the tropical North Atlantic and negative E – P anomaly in the tropical South Atlantic characteristic of southward ITCZ migration [Rind and Overpeck, 1993]. An imposed 1°–2°C cooling across the entire North Atlantic is capable of creating the tropical Atlantic E – P dipole pattern, however [Rind and Overpeck, 1993]. Dynamics of the ocean and atmosphere coupling not yet included in solar irradiance models may create very different temperature and E – P anomalies. Reduced northward surface heat transport associated with weaker meridional overturning or wind-driven subtropical gyre circulation may provide the cooling necessary to force ITCZ movement. Available paleoclimate evidence for reduced North Atlantic Deep Water production during the LIA is equivocal [Keigwin and Boyle, 2000].
 The surface hydrography of the Florida Current was highly variable during the last millennium. Near Dry Tortugas, salinity increased by 0.8–1.5 during the course of the Little Ice Age. Similar although larger changes in tropical Atlantic salinity and Northern Hemisphere temperature are observed on glacial-interglacial and millennial timescales [Schmidt et al., 2004a, 2004b]. Thus this pattern of climate variability is not necessarily a manifestation of large continental ice sheets as it can occur under climatic boundary conditions very similar to today. Variable solar irradiance apparently paced shifts in Florida Current salinity during the last millennium, implying the transport of salt into the North Atlantic is sensitive to modest changes in solar forcing on centennial timescales. The lack of evidence for variable Atlantic overturning during the LIA [Keigwin and Boyle, 2000] suggests that southward ITCZ migration and increased Gulf Stream salinity may stabilize changes in North Atlantic Deep Water production, as is predicted by model studies [Vellinga et al., 2002; Vellinga and Wu, 2004].
 On the Great Bahama Bank, where surface waters reflect the influence of the North Atlantic subtropical gyre, SST and δ18Ow records display different variability than their Dry Tortugas counterparts. Sea surface temperatures decreased by 0.5°C at the beginning of the LIA, consistent with the timing of cooling in several Northern Hemisphere temperature records. SSTs increased by 0.5°C about 200 years prior to the LIA termination, however, suggesting a more heterogeneous temperature anomaly pattern at the end of the LIA. If δ18Ow on the Great Bahama Bank represents primarily salinity rather than anthropogenically driven pH effects, then surface salinity increased ∼0.4 during the last 200 years, making the modern values the saltiest of the last millennium.
 We would like to thank Howie Spero, David Lea, Olivier Marchal, Lloyd Keigwin, Delia Oppo, and Jerry McManus for their valuable comments and suggestions. We also thank Dorinda Ostermann, Marti Jeglinski, Paige Cerulli, Simon Thorrold, and Scot Birdwhistle for technical support. We are grateful to the WHOI core lab for sample collection and archiving, the captain and crew of the R/V Knorr, and the Sea Education Association for access to their vessel Westward. This work was supported by NSF grant OCE-0096469.