The first bimonthly time series of paired δ18O and Sr/Ca from the slow-growing coral Siderastrea siderea, from the Dry Tortugas, Florida, has been generated that documents that robust proxy climate records of the tropical Atlantic and IntraAmerican Seas can be produced from this massive coral. The time series contain a 20-year-long calibration window (1973–1992) for both δ18O and Sr/Ca and a 73-year-long verification window (1900–1972) for Sr/Ca. These time series permit the quantification of the relationship between coral δ18O-SST and Sr/Ca-SST and the assessment of the stability of the proxy relationships over time. Both coral geochemical records are highly correlated with the augmented instrumental SST record through the calibration period, and Sr/Ca remains highly correlated through the verification period both at the bimonthly (r = −0.97) and annual average resolution (r = −0.72). Coral δ18O and Sr/Ca are highly reproducible within the same core, and Sr/Ca exhibits no extension-related vital effects. This study sets the stage for generating multicentury scale climate records from the tropical Atlantic Ocean using the skeletal geochemistry of this massive, but slow growing coral.
 The establishment of a robust, subannually resolved, multicentury coral-based archive of climate variability in the tropical Atlantic has lagged behind successful efforts on this front using corals from the Pacific [Gagan et al., 1998; Quinn et al., 1998; Gagan et al., 2004; Kilbourne et al., 2004a, 2004b; Correge, 2006]. Efforts using Atlantic corals to reconstruct climate have also been hindered by (1) questions concerning the overall validity of using coral Sr/Ca as a tracer of sea surface temperature (SST) and (2) physical problems associated with sampling of the coral skeletons in such a way that is not time-transgressive.
 Initially, the massive Atlantic coral species Montastraea spp. was investigated for their ability to act as a proxy species for generation of Sr/Ca- and δ18O-based records of tropical Atlantic and Caribbean climate, as they are continuously growing, long-lived, widely distributed throughout the region, and exhibit clear annual density banding for external chronological control [Hudson et al., 1976; Leder et al., 1996; Swart et al., 1996; Smith et al., 2006]. The massive Atlantic coral Diploria strigosa has recently been investigated for its use as a climate archive and early results look promising [Hetzinger et al., 2006].
 Two previous studies have investigated the origins of stable isotopic variations within the skeleton of Siderastrea siderea [Guzman and Tudhope, 1998; Gischler and Oschmann, 2005]. Guzman and Tudhope  examined isotopic variations in the skeletons of multiple colonies of S. siderea from Panama over a 14 month period and found significant intercolony differences in mean δ18O over the study period, indicating that care must be taken if using multiple colonies to generate a continuous δ18O record or when comparing modern and fossil coral δ18O values. Gischler and Oschmann  presented a continuous δ18O record from S. siderea from near the southern Belize coast spanning 1870–1999 with a maximum sampling resolution of eight samples per year. These authors concluded that the δ18O record exhibits significant correlations with the SST as extracted from the GISST 2.3b gridded SST product, but may also be significantly influenced by rainfall and riverine discharge.
 Additionally, the primary sampling strategy employed thus far to avoid any signal contamination from heterogeneous or time-transgressive skeletal elements in Montastraea has been to exclusively target the thecal wall of the corallite [Leder et al., 1996; Swart et al., 2002; Smith et al., 2006]. However, workers have admitted that such inclusions may be impossible to avoid using millimeter-scale sampling [Swart et al., 2002; Smith et al., 2006].
 Recent work done by Cohen and Thorrold  on a slow growing (<5 mm a−1) colony of Montastraea franksii from Bermuda utilized laser ablation inductively coupled plasma mass spectrometry to discretely sample nighttime-deposited centers of calcification (COCs) at a temporal resolution of ∼5–10 days per sample. The resulting data, numerically smoothed to 35 days per sample, along with coeval data from colonies of Diploria labyrinthiformis, were related to monthly Hydrostation S SST data (0–20 m in depth) to yield a reproducible equation relating 35-day smoothed Sr/Ca to monthly SST with a slope ∼3 times as large as those previously published for these same species [Cohen and Thorrold, 2007]. This increase in slope was attributed to the avoidance of measuring Sr/Ca ratios in daytime calcification or thickening deposits of ambiguous age and temperature sensitivity. However, such a strategy fails to address the critical need of many paleoclimate workers to reconstruct sea surface conditions through the paired analysis of δ18O and Sr/Ca from powdered, homogenized sample material.
 The challenge of extracting multicentury climate records using millimeter-scale sampling techniques from massive Atlantic corals is approached here by investigating the thus far underutilized species of coral, S. siderea. The coral S. siderea is a massive, slow-growing (<1 cm a−1) species commonly found in shallow Caribbean, tropical West Atlantic, and Gulf of Mexico waters. Colonies are often >1 m across and exhibit cerioid corallites 3–4.5 mm in diameter with thickened septocostae forming poorly defined, dense corallite walls [Veron, 2000].
 No previous studies have investigated the origin of Sr/Ca variations within the skeleton of S. siderea, hence unambiguous, subannually resolved coral records of SST variations are lacking in the tropical North Atlantic, Caribbean, and Gulf of Mexico regions. Multicentury records of SST variability are needed to extend instrumental records of climate indices and allow further interpretation of important modes of climate variability, such as Western Hemisphere Warm Pool variability, ENSO teleconnections, and the so-called Atlantic Multidecadal Oscillation [Delworth and Mann, 2000; Mann, 2001; Wang and Enfield, 2001, 2003; Gray et al., 2004; Kerr, 2005; Knight et al., 2006; Wang, 2006; Wang et al., 2006].
 Here we develop a new high-resolution climate archive for the tropical West Atlantic and IntraAmerican Seas by demonstrating that reproducible, bimonthly resolved Sr/Ca and δ18O measurements in S. siderea from the Dry Tortugas are capable of robust reconstruction of SST variability within a 20-year calibration window, and extending the Sr/Ca time series to include a 73-year verification interval of SST reconstruction. The Dry Tortugas present an ideal location to begin such work. The open ocean surroundings provide isolation from potential terrestrial and groundwater interferences on coral growth and geochemistry. Additionally, there is strong coupling of local SST to regional, Gulf of Mexico SST and Northern Hemisphere surface temperatures, as well as to North American continental moisture transport [Wang et al., 2006]. These factors make coral-based SST reconstructions from this location of significant paleoclimatological value.
2. Background and Methods
 The Dry Tortugas (∼24°39′N, 82°52′W) are a series of modern and relict coral reef and sand shoal features adjacent to the southwest Florida continental margin and the Straits of Florida [Mallinson et al., 2003]. Hydrographic features are dominated by the Florida Current and the local formation and persistence of large (100–200 km in diameter) cyclonic eddies that control surface circulation in the Dry Tortugas region [Fratantoni et al., 1998]. The quasi-stationary, closed-circulation cold core eddies may remain stationary in the region for up to 100 days before being interrupted by cold core frontal perturbations in the Loop Current, which migrate southward along the Loop Current edge adjacent to the West Florida Shelf [Fratantoni et al., 1998].
 The SEAKEYS/C-MAN station (DRYF1) time series from the Dry Tortugas National Park (http://ndbc.noaa.gov/station_page.php?station = DRYF1) contains short, and somewhat discontinuous, records of hourly SST and sea surface salinity (SSS) variations. For this study, the SST data were averaged to produce a monthly time series for the period 1996–2003. Mean annual SST for this time interval is 26.07°C (±0.38°C, 1σ) with an average annual cycle of 9.04°C (±1.10°C, 1σ). The warmest and coolest temperatures in the year occur during August and February, respectively. SSS data available from the SEAKEYS program for the Dry Tortugas from 1993 to 2000 reveal a mean SSS of 35.6 (±1.42, 1σ), with more saline values occurring during boreal spring and summer months, and fresher values occurring during boreal fall and winter (Figure 1a). SST data used for the calibration and verification intervals of the Sr/Ca record were extracted from the HADISST1 gridded data set at the 1° × 1° grid box of 24–25°N and 82–83°W [Rayner et al., 2003]. These data were corrected using an ordinary least squares (OLS) regression relationship between the HadISST1 gridded data product and the DRYF1 time series of SST [Smith et al., 2006]. The corrected data are hereafter referred to as an augmented instrumental data set (Figure 1b). Regional SST variability in the GOM agrees well with the local, augmented SST data set over the interval from 1900 to 1992 (r = 0.77), indicating that the Dry Tortugas are an excellent target location for proxy reconstructions of SST variability.
2.2. Coral Sampling
 A colony of S. siderea, >1 m in height, located in the Dry Tortugas, immediately south of Long Key (∼24°37′N, 82°52′W), was cored along the axis of maximum growth in the summer of 1993. This core, hereafter referred to as 93DRYSS-1, was cut lengthwise into 0.5 cm-thick slabs and x-radiographed (Figure 2). The coral slabs were mounted to a computer-aided triaxial sampling platform and samples for paired elemental and isotopic analyses were milled out of the coral thecal wall using a 1.4 mm dental drill bit along continuous paths. Samples were extracted from corallite walls most parallel to the lengthwise axis of the slabs to minimize the possibility of time-transgressive sampling. One sample was taken every 0.75 mm of linear skeletal extension, corresponding to approximately six to eight samples per year as estimated by the existence of annual cycles in the resulting geochemical data. The width and depth of the sample transects were 2 mm and ∼1.5 mm, respectively.
 An additional sampling path, 4.575 cm in length, was drilled at an arbitrary location down-core parallel to the initial path in order to assess geochemical reproducibility within the core. A second additional sampling path near the core top was drilled at an interval of 0.25 mm of extension per sample in order to assess the effect of multiple sampling resolutions on the geochemical signal.
2.3. Elemental and Isotopic Analyses
 Analyses of Sr/Ca were performed by dissolving ∼100–300 μg of drilled coral powder in a volume of 2% HNO3 appropriate to dilute the Ca concentration of the sample to ∼20 ppm. Measurements of sample Sr/Ca ratios were made using a Perkin-Elmer 4300 Dual View Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES) housed at the Paleoceanography, Paleoclimatology and Biogeochemistry (PPB) Laboratory of the University of South Florida College of Marine Science. The Sr/Ca of a gravimetrically prepared standard solution (IGS) was measured between each dissolved coral sample in order to correct sample Sr/Ca for instrumental drift and noise [Schrag, 1999]. The average corrected precision of the IGS standard was 0.015 mmol/mol, based on batches of seven consecutive measurements (1σ, n = 278 corrected values) performed at the beginning and end of each 50 samples measured per instrument run containing 93DRYSS-1 samples. A second standard consisting of homogenized powder from a Porites lutea coral dissolved in 2% HNO3 was analyzed for Sr/Ca every sixth sample. The average precision of this second standard within instrument runs containing 93DRYSS-1 samples was also 0.015 mmol/mol (1σ, n = 277).
 Stable isotopic analyses of oxygen and carbon were performed by dissolution of powdered samples in phosphoric acid at 70°C in a Kiel III automated carbonate preparation device connected to a ThermoFinnigan Delta PlusXL gas-source dual-inlet stable isotope ratio monitoring mass spectrometer, also housed at the PPB Laboratory. All resulting isotope values are reported in delta notation relative to the VPDB isotopic standard. Analyses of the isotopic standard NBS-19 were measured within sample runs to monitor instrumental precision at a ratio of three standard to 20 sample measurements. The resulting average precision of this standard was 0.06 ‰ for δ18O and 0.03 ‰ for δ13C (1σ, n = 20 measurements of NBS-19).
2.4. Data Analysis
 The x-radiographs of the coral slabs show none of the clear annual density band couplets that are often observed in other species of coral [Alibert and McCulloch, 1997; Cohen et al., 2004; Smith et al., 2006]. Hence, geochemical variations with depth were converted to variations in time by maximizing agreement between the Sr/Ca (inverted) and augmented SST data sets, both of which exhibit clear annual cycles, using the AnalySeries program [Paillard et al., 1996]. Age-modeled data were then evenly sampled at bimonthly resolution, resulting in a bimonthly resolved coral δ18O and δ13C time series spanning 1973 through 1992 (Figure 3) and a Sr/Ca time series spanning 1900–1992. The origin of δ13C variations in coral skeletons is ambiguous [Swart et al., 1996, 2005] and will not be discussed further. The remaining two geochemical time series were directly compared to the augmented instrumental SST data, which were averaged from monthly to bimonthly to match the coral data. Extension rates were calculated by counting the number of data points that fell within an annual cycle prior to linear interpolation at a maximum resolution of 0.75 mm (sampling resolution).
 The first 20 years (1973–1992) of the coral Sr/Ca and δ18O time series were regressed against the bimonthly augmented SST data set in order to yield equations relating Sr/Ca and δ18O to SST. The resulting Sr/Ca equation was used to predict bimonthly and mean annual SST from the geochemical time series, allowing assessment of the skill of the Sr/Ca thermometer in this coral over a 73-year verification window. The time-averaging sampling method employed here (continuous routing of one sample representing ∼2 months of time at constant extension rate) likely results in a minimum estimate of the Sr/Ca-SST slope. This hypothesis is tested in the next section by the results of the sampling resolution experiment.
3.1. Sampling Resolution and Reproducibility
 The effect of increasing sampling resolution from approximately six samples per year (0.75 mm per sample) to approximately 18 samples per year (0.25 mm per sample) on the amplitude of the geochemical signal recovered over three annual cycles is negligible (Figure 4). The mean amplitude of Sr/Ca values over the 3 years sampled is 0.303 mmol/mol at 0.75 mm per sample and 0.332 mmol/mol at 0.25 mm per sample, values that agree within 2σ of instrumental precision for Sr/Ca measurements.
 The geochemistry of contemporaneous drill paths reveals a high level of reproducibility between respective variables (Figure 5). The mean Sr/Ca values for the two parallel paths are 9.008 ± 0.003 (σμ) and 8.992 ± 0.003 (σμ) mmol/mol, and the mean δ18O values are −3.24 ± 0.01 (σμ) and −3.25 ± 0.01 (σμ) ‰. The average absolute difference between the geochemical values in the parallel paths, 0.027 mmol/mol for Sr/Ca and 0.11 ‰ for δ18O, are within 2σ of the analytical precision for either variable (Sr/Ca: ±0.030 mmol/mol, δ18O: ±0.12 ‰).
3.2. Skeletal Sr/Ca Geochemistry and Extension Rate
 The average annual amplitude and overall mean Sr/Ca values for the entire 93-year time series are 0.2664 ± 0.046 (1σ) mmol/mol and 8.999 ± 0.103 (1σ) mmol/mol, respectively. The average annual extension rate, estimated from skeletal geochemistry, is 4.94 ± 0.73 (1σ) mm a−1 over the entire time series. The lowest and highest observed annual extension rates were 3 mm a−1 and 6 mm a−1, respectively.
3.3. Calibration of Geochemistry With SST
 The generalized regression equations [York et al., 2004] were calculated for the relationships between bimonthly Sr/Ca-SST and δ18O-SST (Figure 6) for the calibration period from 1973 to 1992 and are given below:
The slope for the Sr/Ca equation is within the span of published slopes for Montastraea spp. from the Florida Keys, which range from −0.023 to −0.047 mmol/mol °C−1 [Swart et al., 2002; Smith et al., 2006], and another slow growing coral, Diploria labyrinthiformis, from Bermuda, which range from −0.0359 to −0.0436 mmol/mol °C−1 for multiple colonies [Goodkin et al., 2007]. The slope reported here is within error of the slopes found for a single Diploria strigosa colony from Guadalupe, which range from −0.041 to −0.042 mmol/mol °C−1 [Hetzinger et al., 2006].
 Slopes for published equations relating δ18O to SST for Montastraea spp. from the Florida Keys range from −0.085 to −0.22 ‰ °C−1 [Leder et al., 1996; Smith et al., 2006] and range from −0.184 to −0.196 ‰ °C−1 for D. strigosa from Guadalupe [Hetzinger et al., 2006], and the slope of −0.138 ‰ °C−1 found here is comparatively and significantly lower. Results of Leder et al.  suggest that sampling rates of less than 50 samples per year may be responsible for the relatively lower than predicted δ18O-SST slope, as is found here. However, the δ18O of coral aragonite (δ18Oaragonite) is a function of both SST and the δ18O of seawater (SSS), hence simultaneous changes in SST and SSS may provide destructive and/or constructive influences on the coral δ18O signal depending on the sign of these environmental changes. In the Dry Tortugas, changes in SST and SSS are positively correlated (Figure 1a). The net effect of this is to dampen the coral δ18O signal because increases in temperature yield decreases in coral δ18O, whereas increases in salinity yield increases in coral δ18O. This relationship at the Dry Tortugas would serve to change the character of the δ18O signal such that accurately recovering the “true” δ18O -SST relationship is highly unlikely even if the coral skeleton could be perfectly sampled.
 The average absolute difference between bimonthly interpolated augmented SST and Sr/Ca-estimated SST data during the calibration interval is 0.65 ± 0.55 (1σ) °C, whereas the average absolute difference between bimonthly interpolated augmented SST and δ18O-estimated SST data is 0.97 ± 0.75 (1σ) °C.
 The agreement of the two proxy time series between each other is a further confirmation of their robustness (Figure 6), especially given that the δ18O record was not tuned to the SST record. The δ18O thermometer in coral-derived carbonates is a widely accepted and utilized, thermodynamically constrained proxy for the temperature of aragonite growth in regions where seawater δ18O changes are small relative to temperature [Leder et al., 1996; Correge, 2006;]. The degree to which paired Sr/Ca and δ18O measurements agree significantly in both bimonthly time series and anomaly time series, where the average annual cycle has been removed (r = 0.91 and r = 0.44, respectively), lends further credence to the idea that coral δ18O and Sr/Ca ratios are being driven by the same environmental forcing, changes in seawater temperature at time of aragonite precipitation.
3.4. Verification of Proxy Skill
 There is no independent, in situ SST data set available with the appropriate length to verify the skill of the proxy-SST relationships. Instead, the 1900–1972 portion of the augmented SST data set is used here to observe the ability of the calibrations and geochemical data to hindcast SST at both the bimonthly (Figures 7a and 7b) and annual average resolution (Figure 7c). The average absolute difference between bimonthly interpolated augmented SST and Sr/Ca estimated SST data during the verification interval is 0.56 ± 0.49 (1σ) °C. The average absolute difference between annually averaged augmented SST and annually averaged Sr/Ca-estimated SST during the verification interval is 0.27 ± 0.25 (1σ)°C. The entire 93-year annual average records exhibit an average absolute difference of 0.27 ± 0.25 (1σ)°C for Sr/Ca-SST relative to the annual average augmented SST.
4.1. Removal of SST Annual Cycle
 The SST signal in the Dry Tortugas region and the Sr/Ca-SST signal from the coral time series both contain pronounced annual cycles, and each are forced to have a 0° phase difference from one another as a result of the age modeling process (see section 2). The net effect is a lack of independence between the two bimonthly resolved time series and therefore inflated covariance from serial correlation. Here the 93DRYSS-1 bimonthly Sr/Ca-SST is compared to the augmented bimonthly instrumental SST time series over the twentieth century in anomaly space, allowing the confirmation of a significant correlation between the two time series using independent data (Figure 7d). Anomalies were calculated by removal of the mean bimonthly annual cycle from 1973 to 1992 for each time series from the entire twentieth century bimonthly record. The resulting correlation between the two time series at the bimonthly resolution is r = 0.56, which is significant at the 99% confidence level and indicative of common variability despite the lack of serial correlation. Disagreement between the two anomaly time series may be augmented by uncertainties in assigning time to coral core depth at subannual resolution.
 The relatively high correlation (r = 0.72) between annual average Sr/Ca-SST and the annual average augmented SST time series further support the conclusion that coral Sr/Ca variations are driven by SST variations. The annual average time series are also independent of the serial correlation as the seasonal cycle is removed by averaging (as opposed to subtraction), but contains fewer degrees of freedom than the bimonthly anomaly time series comparison. The high degree of correlation between the two time series suggests that this proxy species can be utilized as an accurate indicator of climate variability on timescales ranging from interannual to centennial.
4.2. Vital Effects
 Given the agreement and correlation between Sr/Ca and augmented SST time series at both the bimonthly and annual average resolution, the reproducibility between parallel paths, and the stability of the signal through the twentieth century, there is no reason to suspect that so-called vital or growth rate effects [Cohen et al., 2002; Cohen and McConnaughey, 2003; Goodkin et al., 2007] are significant drivers of Sr/Ca variability in this record. On the basis of the results of Cohen and McConnaughey , which show decreased extension rates corresponding to relative enrichment in Sr/Ca values, one would expect to see a significant, persistent inverse relationship between annual average Sr/Ca and annual extension rate. Such a relationship is not observed in this study.
 The overall correlation between annual extension rate and annual average Sr/Ca is r = −0.14, which is insignificant at the 99% confidence level. However, given the limited resolution of our extension rate data (0.75 mm), we augmented our analysis by performing running correlations at different timescales in order to observe whether this overall insignificant inverse relationship remains throughout the last century of coral growth, and whether or not there are periods of strengthened covariance between the two variables. Figure 8 clearly shows that there is no systematic correlation between Sr/Ca and extension rate and that the expected inverse relationship does not persist.
4.3. Capture of Regional Climate Variability
 The Gulf of Mexico (GOM), as a component of the Atlantic Warm Pool (AWP), is an important source of heat and moisture to the continental United States, and variability on interannual and multidecadal timescales has broad-reaching climate effects [Mestas-Nunez et al., 2005; Wang et al., 2006; Mestas-Nunez and Enfield, 2007]. Therefore, the ability of a proxy record of climate within the GOM to reflect variability within the region as a whole is an important consideration when assessing the relevance of such records to the understanding of climate variability and change.
 In order to assess the correlation between the GOM regional SST and the local record Sr/Ca-SST record from the Dry Tortugas, SST data were extracted from the HadISST1 data set for the GOM region, 24–28°N and 84–96°W for the 20th century [Rayner et al., 2003]. The regional time series agrees well with the local, augmented SST data set over the interval from 1900 to 1992 (r = 0.77), and similarly well with the local coral record on the annual average level (r = 0.67) (Figure 9), suggesting that coral records of SST variability from the Dry Tortugas have regional significance and can therefore be utilized in reconstructions of IAS or AWP thermal behavior on a variety of timescales, perhaps in concert with representative proxy records from other AWP components, such as the Caribbean and tropical West Atlantic.
 Here we present time series of geochemical variations from the coral S. siderea from which the following was concluded:
 1. Both Sr/Ca and δ18O variations are highly reproducible within the skeleton.
 2. The Sr/Ca and δ18O variations agree well with augmented SST data from 1973 to 1992 (r = −0.96 and r = −0.92, respectively), allowing calibration of geochemistry to SST. However, the nature of seasonal changes in SST and SSS at the Dry Tortugas acts to produce destructive influence on the coral δ18O signal, which impacts the observed coral the δ18O -SST slope.
 3. Sr/Ca variations agree well with SST during the 20th century in bimonthly (r = −0.96), bimonthly anomaly (r = 0.56), and annual average (r = 0.72) SST space, indicating a relationship between SST and Sr/Ca that is stable over the 20th century.
 4. There is no evidence that extension rate related vital effects are present in this coral record.
 5. The coral record agrees well with regional scale GOM climate variability (r = 0.67).
 6. Our results suggest that the massive Atlantic coral S. siderea can be used to develop continuous, subannually resolved multicentury proxy based records of SST variability in the Caribbean and tropical West Atlantic.
 The authors would like to acknowledge the enabling, contributions, and efforts of Ethan Goddard, Gene Shinn, Hali Kilbourne, Julie Richey, Kristine DeLong, and Mark Eakin. This project was partially supported with funding from the NOAA Coral Reef Watch Program.