Apparently Seasonal Variations of the Seawater Sr/Ca Ratio Across the Florida Keys Reef Tract

A 4‐year time‐series of surface seawater Sr/Ca ratios was assembled across a section of the Florida Keys Reef Tract, in order to uncover any variability that might explain previously reported anomalies in regional calibrations of the coral aragonite Sr/Ca paleotemperature proxy. Samples were collected semiannually on a grid of 54 sites, from September of 2016 until January of 2020. The 325‐km2 grid extended from the ocean shore to the forereef wall and from the east end of Long Key to the west end of Marathon. A novel ICP‐AES method was used to measure the Sr/Ca ratio, with ratio calibration and normalization against an in‐house seawater reference, yielding a long‐term precision of better than 0.2%. Significant variations (2%–3%) of the seawater Sr/Ca ratio were found. While it was relatively constant offshore, near the coast the ratio alternated seasonally between higher and lower values, generally resulting in seaward Sr/Ca gradients that were markedly negative in summer but reversed in winter. Inshore seawater Sr/Ca ratios ranged from a summer high of 8.83 mmol/mol to a winter low of 8.54 mmol/mol, the difference corresponding to a potential bias of ∼5.5°C in terms of the coral Sr/Ca paleotemperature proxy. This seasonal variation should diminish the slope of empirical Sr/Ca–SST calibration lines, as has indeed been observed in prior studies with local coral species. Open ocean samples obtained from the Atlantic, Indian, and Pacific enlarge the published Sr/Ca data set for surface seawater and show a much smaller variability of 8.646 ± 0.018 mmol/mol (0.2%).


10.1029/2022GC010728
3 of 19 variability, resulting from strong storms (Khare et al., 2021) or upwelling events, should be more easily detectable, but might still lead to temperature anomalies that are large and hard to quantify (for a possible example, see Reich et al., 2013).
Assessing the likelihood and importance of these effects has been difficult due to the absence of a method allowing routine monitoring of seawater Sr/Ca ratios. We have recently developed an ICP-AES method, based on matrix-matched ratio calibration lines and rigorous sample normalization, that can measure seawater Sr/Ca ratios with a precision of better than 0.15% (Khare et al., 2021). Here we have used this new method to analyze surface seawater samples that were collected in an 11 × 5-site spatial grid on the Atlantic side of the Florida Middle Keys, to the east of the study area of J. M. Smith et al. (2006). The grid was sampled twice per year from September 2016 until January 2020, for a total of four summer and four winter seasons. The data generally show a relatively modest variation of seawater Sr/Ca in the offshore samples, but a strong variation in the nearshore samples, with summer Sr/Ca values being higher than winter values by as much as 0.3 mmol/mol (∼3.5%). This corresponds to a potential temperature bias of 5-6°C and is of precisely the magnitude estimated by J. M. Smith et al. (2006) to explain their anomalous calibration. While a detailed investigation of the underlying mechanism was explicitly beyond the purview of our study, the surface seawater Sr/Ca distribution strongly suggests that these variations are not generated in situ but originate in Florida Bay, directly to the north, and are advected to the coastal ocean by the exchange of water through channels between the islands. Consequently, the distribution alternates between negative and positive seaward gradients in summer and winter, respectively, on what appears to be a seasonal cycle.
We also present Sr/Ca data for bottom samples at sites where instrument packages were deployed to collect seawater samples with much higher temporal resolution, which will be the subject of a separate paper. These results suggest that, unlike in the open ocean, small vertical gradients seem to be related more to tidal mixing between Florida Bay and Atlantic Ocean water than to plankton dynamics. We discuss tidal influences in detail, as they may have altered the observed distributions to some extent yet are not their underlying cause. Finally, we present Sr/Ca data for surface seawater samples around the world that were collected by colleagues over a period of 7 years utilizing our non-contaminating protocol. Their variability is much smaller and consistent with what has been found in earlier surveys. We conclude that the global surface seawater Sr/Ca value is sufficiently stable to allow coral paleothermometry with careful empirical Sr/Ca-SST calibrations. However, in some places like the Florida Keys, significant cyclic or episodic deviations from the global average can cause biases that may greatly complicate or even invalidate such studies. In our opinion it is therefore prudent, by adopting new analytical methods suitable for seawater, to monitor ambient Sr/Ca as an integral component of future research in locations where such deviations may be suspected or anticipated.

Sample Collection
Seawater samples were collected in acid-cleaned 60-mL polyethylene (PE) bottles from the stern platform of a small drifting vessel, as described by Khare et al. (2021). The sample was taken with an all-plastic 30-mL syringe well below the surface film (30-50 cm) and filtered through a 0.2-μm cartridge with Luer-lok fitting. The syringe and cartridge were first flushed with seawater and the bottle was then rinsed several times with filtered sample before filling and capping it with minimal headspace. Nitrile gloves were worn throughout the procedure. Bottles were individually packed in Ziploc bags and refrigerated until return to Chesapeake Biological Laboratory (CBL), where they were unpacked and stored in plastic boxes inside a clean room. A total of 54 samples was collected on a spatial grid comprising 11 offshore lines of five sites each (4 sites on line 10), extending from the east end of Long Key to the west end of Marathon and from the ocean shore to the forereef wall ( Figure 1). The grid was occupied twice annually, in winter and summer, from 2016 to 2020. We will refer to these collections by the last two digits of the year and a W for winter (January), or an S for summer (July-September), for example, 18W for January 2018. Spacing between grid points was roughly 2 km in the N-S direction and 4 km in the W-E direction, for a total surface area of 325 km 2 . Initially, a replicate sample was taken at one station on every offshore line, but this number was later reduced to 1 or 2 for the whole grid, as the reproducibility of our sample protocol was found to be excellent. Some additional samples were collected early on, at off-grid sites of special interest. Further details are given in Section 3.
At each site, water color was rated on an arbitrary scale comprising three shades of green (brown-green, dark green, green) and three shades of blue (turquoise, blue, dark blue), where brown-green typically indicates a turbid or muddy appearance from suspended particulate matter. While the water color is influenced by sea state and cloud cover and therefore somewhat subjective, occasional checks confirmed that ratings were sufficiently consistent among different observers. Water temperature was recorded from the vessel's hull-mounted sensor.
Shown as well in Figure 1 are four sites where osmotic pumps were deployed (Jannasch et al., 2004) for continuous seawater sampling during the entire research period, three in the area of the spatial grid (OS2-4) and one inside Florida Bay (OS1). Data from these so-called OsmoSamplers will be presented in a future publication (and see Khare et al., 2021). Within days of every grid collection, the pumps were serviced by scuba divers, who showing the 54 stations of the sampling grid (black dots), south of Marathon and Long Key, with color shading representing bathymetry. Grid lines are labeled in blue from nearshore to offshore (A-E) and from east to west (1-11). Line 10 has only four stations (A-D). OsmoSamplers were deployed at four sites labeled in red (OS1-4, red dots). Also shown (yellow dots) are some additional sites of special interest, including a very shallow site near station 10A and patch reefs near stations 1D, 2C and 6C. The patch reef at 2C almost coincides with the grid point. Maps created in ODV version 4 (Schlitzer, 2014, http://odv.awi.de).

of 19
also took a discrete seawater sample at each of these four sites. Two syringes were filled just off the bottom, avoiding the entrainment of sediment, and brought up to the vessel. Their contents were then filtered on deck and combined in a single bottle, using the protocol outlined above. An emergency survey was conducted in November 2017, after the passage of Hurricane Irma in September. Although all equipment was lost, except at OS1, bottom samples were taken at the original coordinates. Another set of bottom samples was taken when new equipment was installed at the same sites 2 months later. After grid collections were discontinued, a set of bottom samples was taken when all osmotic pumps were removed and station OS1 was decommissioned (20S). A final set was taken at OS2-4 a year later (21S), when these stations were also decommissioned. No bottom samples were collected in 19S.
In order to constrain the Sr/Ca ratio of saline groundwater, a possible source of the observed seasonal variations on the spatial grid, two well samples were obtained during 17S, following the aforementioned protocol. One sample was taken from the top of a covered open well that supplies seawater to the flow-through culture system of the Keys Marine Laboratory (KML) in Layton, FL. Unfortunately, this sample was collected after a period of heavy rain and found to have very low salinity (S ∼ 1). The second sample was taken from the strainer basket of a system that supplies seawater at the Aquarium Encounters tourist attraction in Marathon, FL. It is brought up from a depth of ∼46 m (150 ft) through a field of 12 wells spaced ∼30 m (100 ft) apart. This seawater is initially anoxic and has a salinity of about 35.
To provide a broader framework for the Florida samples and allow comparison with prior global seawater surveys, colleagues kindly supplied samples from cruises of opportunity at coastal and open ocean locations around the world ( Figure S1 in Supporting Information S1). These were collected with pre-cleaned materials that we provided, utilizing our protocol, and then sent to CBL for analysis. Single samples from Singapore and Taiwan, and a series of samples from the East Java Sea were taken in 2013-2014 from small vessels and by scuba divers. Two samples were collected from bays on opposite sides of Oahu, Hawai'i, in 2017 (Mitchelmore et al., 2019). Four samples were taken from the north, west, east, and south side of Barbados after an undergraduate field geology course in March 2019. The western site was sampled close to shore from a sightseeing vessel, the other three by wading from the beach into knee-deep water. One replicate sample was taken from the 5-m bottle of a CTD/Rosette cast off the R/V Atlantic Explorer at the Bermuda Atlantic Time Series (BATS) station in August 2019. Exact coordinates and dates can be found in Table 3.

ICP-AES Analysis
Analysis of the seawater samples via ICP-AES was described in extensive detail by Khare et al. (2021). Whereas samples were generally diluted 100-fold with nitric acid (∼0.2 M), the KML well water had a much lower salinity and was diluted by a factor of 3. All sample preparations were performed in a class-100 laminar flow fume hood inside a HEPA-filtered clean room. Reagents were of TraceMetal grade (Thermo Fisher) and made with Milli-Q water (18.2 MΩ·cm) from a Millipore Direct-Q 3UV purification system. Diluted samples were analyzed on a PerkinElmer Optima 8300 ICP-AES in radial viewing mode. The plasma source was carefully optimized to yield similar signal strength for the 421.552 nm ionic Sr line and the 422.673 nm atomic Ca line for seawater samples, where Sr is about two orders of magnitude less abundant than Ca. Samples were run in a "block" of 36 (Khare et al., 2021), against an external 5-point ratio calibration line, matrix-matched with Na and Mg to their levels in 100-fold diluted seawater (S = 35). A seawater reference solution was inserted before and after every sample in order to enable a running normalization as described by Schrag (1999). Since a certified reference material (CRM) suitable for this purpose does not currently exist, a large batch of aged and filtered Gulf of Mexico (GoM) seawater, collected in 2005 from a depth of 450 m, was calibrated separately by inductively coupled plasma mass spectrometry (ICP-MS) to create an in-house Sr/Ca standard. In addition, normalized values of this GoM standard were measured before each "group" of six samples, whereby a block contains six groups. To monitor the long-term stability of our method, at least one of the 36 samples was always a coastal seawater CRM (CASS-5 or CASS-6), although their certification does not include Sr and Ca. The entire block of sample and standard solutions was analyzed five times in a run and these block measurements were averaged, after a statistically appropriate removal of outliers, to obtain final normalized Sr/Ca ratios and standard errors (SE). The typical duration of a run was about 6 hr.
The data presented here were generated while the field samples were being collected, in three series of ICP-AES runs by different operators: in December 2017 and February 2018 by AK; from March to October 2019 by HPH; and finally in January and September-October 2021 by JS and KHK. Each used a new batch of GoM, drawn from the primary demijohn. The AK batch was the one calibrated by ICP-MS and assigned a Sr/Ca value of 8.69 ± 0.04 mmol/mol. To ensure analytical continuity throughout this 4-year period and to eliminate inferior data due to random contamination and occasional instrument failure, many samples were rediluted and reanalyzed up to 5 times or more, often by multiple operators. For these samples, average normalized Sr/Ca ratios and SE were derived by pooling every block measurement that passed quality control, from all replicate runs.

Renormalization of the Data to Correct for Reference Drift
The importance and intricacies of normalizing measured seawater Sr/Ca ratios to an external reference were discussed previously. Khare et al. (2021) showed that the normalized Sr/Ca ratios of CASS-5 and CASS-6 were stable at the level of 0.012 mmol/mol (1σ or 0.13% RSD) over a period of 7 months, with that of CASS-6 being lower, on average, by 0.014 mmol/mol. The authors presented their data in terms of relative Sr/Ca ratios since absolute ratios ultimately depend on the value of GoM, which is presently known to an accuracy of only 0.41% and has not yet been independently verified. This approach allowed them to provisionally ignore the fact that the CASS-5 and CASS-6 standards had actually been analyzed by HPH with a different batch of GoM that had initially also been assigned a value of 8.69 mmol/mol without checking it directly against the batch used by AK.
As stated in Section 2.2, the complete set of seawater samples was analyzed in three series of ICP-AES runs, by different operators using separate batches of GoM. Although the constancy of the CASS-5 and CASS-6 Sr/ Ca ratios remained very good within each series, representing periods of less than a year, a critical evaluation of all data revealed that marked shifts in the GoM value had occurred from one batch to the next. Specifically, the value of each new batch was found to be lower than the previous one (Table 1), which we attribute to a gradual improvement in their manner of storage. While the AK batch was kept and used in a laminar flow bench in the main laboratory, the next (HPH) was permanently stored in a 125-mL acid-cleaned PE bottle inside a class-100 clean room and the final one (JS/KHK) in a 500-mL acid-cleaned Teflon PFA bottle. Since the JS/KHK batch was the most cleanly stored and last one drawn, its Sr/Ca ratio is our current best estimate of the true value of GoM. It appears that the Sr/Ca ratio of the AK batch, which had been stored with fewer precautions and used during a year of method development before its calibration using ICP-MS and its first application as a reference standard, exceeds the true GoM value by more than 1%, well outside our analytical precision. A slow upward drift of the value of each batch was also notable, likely due to cumulative contamination as small aliquots were pipetted from the GoM bottles before every ICP-AES run, but this is not significant. The CASS-5 and CASS-6 standards did not seem to be subject to such shifts or drift, probably because they were always kept in the clean room in small Teflon PFA vials that were regularly refilled from the stock bottles. Where listed, n represents the number of block measurements. During each ICP-AES run a block of 36 samples and standards is measured 5 times. The number of measurements may be lower if runs were aborted due to instrument failure or if some measurements were excluded on statistical grounds. b The original AK batch of GoM served as the primary standard whose Sr/Ca ratio was quantified by ICP-MS with the listed uncertainty and used as the reference value for all subsequent analyses. c Standard errors reflect the precision of relative Sr/Ca ratios, but the uncertainty of the absolute Sr/Ca ratios is no better than that of the reference value (0.04 mmol/mol). d The last drawn and most cleanly stored batch of GoM provides our current best estimate of its true Sr/Ca ratio.
As these issues emerged only in hindsight, it was necessary to renormalize all data to a common reference value retroactively. For this, we chose the Sr/Ca ratio of the AK batch, its apparent contamination notwithstanding, because it is the only one determined using a separate method. In a dedicated ICP-AES run, the Sr/Ca ratios of all different GoM batches were therefore measured relative to each other and to the AK batch. This yielded a set of mutually consistent Sr/Ca ratios for each batch, as shown in Table 1. It should be emphasized that while these relative values were determined with high precision, absolute values are still known no more accurately than that of the AK batch itself and this will be the case for all our data until the true value of GoM is independently verified. Once the relative values of all GoM batches were established, the renormalization of the data was readily accomplished with a built-in feature of our custom Excel template (Khare et al., 2021). After renormalization, measured Sr/Ca ratios of GoM standards from the different batches, routinely included as samples in all ICP-AES runs, were fully consistent with the derived set of relative values, validating the procedure. The renormalized values of CASS-5 and CASS-6 (Table 1), now invariant across all runs, are 0.045 mmol/mol lower than those reported by Khare et al. (2021), reflecting the offset between the HPH and AK batches of GoM. Nonetheless, the difference between CASS-5 and CASS-6 is 0.014 mmol/mol, exactly as before, confirming that the renormalization does not affect relative values. Also shown in Table 1 are a few measurements of the older CASS-4, which in turn are 0.014 mmol/mol higher than CASS-5. The declining Sr/Ca ratio of consecutive generations of CASS CRMs may again reflect their age that is, their length of storage, to some degree.

Results
The 54-site spatial grid ( Figure 1) was occupied 8 times over a period of 4 years, 4 times in summer and 4 times in winter. This produced a total of nearly 500 seawater samples, including bottom samples, world ocean samples, and replicates. Samples were collected as much as possible at the same coordinates each season. However, they had to be taken somewhat off-site at 9D in both 17S and 18W, in order to bypass commercial fishing activity. These diversions were recorded and accounted for yet are not large enough to be notable. In 20W, a minor miscommunication with the boat captain resulted in station 5C being skipped and 5B sampled twice. Completion of the grid ideally required two consecutive days of about 6 hr (9 a.m.-3 p.m.), each spanning half of a tidal cycle. Samples were usually collected following offshore lines (1-11; Figure 1), starting either on the eastern or western end of the grid, alternately moving seaward or landward on every next line. In several seasons, rough seas caused the completion of the grid to take more than 2 days, or forced us to skip days or follow longshore lines (A-E; Figure 1) or other more random trajectories. Due to inclement weather, stations 6B-E and 8B-E could not be sampled in 18W. Sampling during rain squalls was avoided as much as possible, but sometimes inevitable. Consequently, individual stations were sampled at arbitrary stages of the tidal cycle and under a wide range of conditions (wind speed and direction, wave height etc.). Although tidal stage and sea state were logged upon departure and return of the vessel, and observations of strong currents, distinct tidal fronts, or sudden changes in the weather were recorded at different times, no attempt was made to correct for these effects.
Apart from routine analytical replicates, whereby a new dilution was made from the same bottle and run as a separate sample, three different types of field replicate were also collected. In some seasons, sites of special interest were occupied close to the grid points (Figure 1), which we will call spatial replicates. Small patch reefs were sampled near stations 1D and 6C in 16S and 17W, and another near station 2C in 17S. During establishment of the grid in 16S, a very shallow site designated 10AA was occupied off the intended line. This site was inadvertently sampled again in 19S, instead of station 10A. Site replicates refer to the collection of two separate samples at the same station, one immediately after the other. Finally, temporal replicates refer to the collection of two separate samples at the same station, one or more days apart in the same season. In addition to testing the reproducibility of the analytical method, these efforts were aimed at assessing the reproducibility of our sample protocol (site replicates) and the effect on the seawater Sr/Ca ratio of the tidal cycle (temporal replicates) or features like small reefs (spatial replicates), as well as more generally its variability in space and time. In this context, we have considered 10AA to be a spatial replicate of station 10A.
A statistical overview of all four replicate types is given in Table S1 of Supporting Information S1 where, for every type, the absolute difference of each replicate pair is compared with the median SE of all sample analyses within each season. Table S1 in Supporting Information S1 lists the number of replicate pairs for which this difference is larger and smaller than twice the median. If a particular type were not a proper replicate, because of analytical and sampling artifacts or tidal effects and spatial heterogeneity, the former number would be consistently larger than the latter. Instead, in almost every case as well as for the whole data set, the opposite is true or the two numbers are nearly equal. For the analytical and site replicates, this testifies to the reliability of our ICP-AES method and sample protocol. For the spatial and temporal replicates, it indicates that variability of the seawater Sr/Ca ratio is negligible in space on the scale of a single station (∼8 km 2 ), and in time on a scale of hours to days. Therefore, mean Sr/Ca ratios for any grid station in the figures incorporate all analytical, site, and temporal replicates. The spatial replicates (Table 2), while indistinguishable from the Sr/Ca ratio at the corresponding stations, are treated as separate sites. They are excluded from the figures since they were not collected in all seasons and their presence perceptibly alters the weighting of the contour plots ( Figure 2 and Figure S2 in Supporting Information S1).
In view of the large number of samples, distributions of seawater Sr/Ca data are presented only in graphical format, except for the spatial replicates and bottom samples (Table 2) and the world ocean samples (Table 3). All numerical data can be freely accessed as specified in the Open Research section (NOAA National Center for Environmental Information, 2022). A typical winter and summer distribution are shown as contour plots in Figure 2, where warm colors (red and yellow) correspond to high Sr/Ca ratios and cool colors (purple and blue) correspond to low Sr/ Ca ratios. Similar plots for the entire time series are shown in Figure S2 of Supporting Information S1. The same data are depicted in Figure 3 as a series of surface profiles along consecutive offshore lines. The mean seaward Sr/ Ca gradients throughout the study are summarized in Figure 4, with red curves corresponding to summer and blue curves to winter. These were calculated by averaging Sr/Ca ratios in each season on every longshore line (A-E), where line A is closest to land. On offshore line 10, which contains only four points, station A was included with line A, whereas stations B-D were included with lines C-E, respectively, based on local bathymetry. In Figure S3 of Supporting Information S1, these gradients are compared with similar gradients comprising the bottom samples, overlain as bar graphs. To highlight the influence of Florida Bay on the Sr/Ca distributions, all surface Sr/Ca data are plotted in Figure 5 as a function of water temperature, divided into 6 bins, and as a function of water color for summer and winter distributions separately. Data for the world ocean samples are shown as a bar graph in Figure 6

Seasonal Patterns of the Spatial Seawater Sr/Ca Distribution
Seawater Sr/Ca ratios on our spatial grid off the Middle Florida Keys (Figure 1) have a distinctive distribution with a seasonally recurring pattern. Figure 2 shows the situation in Jan (18W) and subsequently in Jun/Jul (18S) of 2018, using contour plots that interpolate among data at the grid points. In winter, Sr/Ca ratios are low near the coast and increase offshore, with isolines of constant Sr/Ca (not shown) roughly subparallel to the shoreline. In summer, the distribution is reversed, with high Sr/Ca ratios near the coast and decreasing offshore in a similar manner. It is clear that the Sr/Ca variation is largest near the coast, where Sr/Ca is lower than offshore in winter and higher in summer, and that relatively minor Sr/Ca variations occur in the deeper water of the forereef. It also appears that the lowest Sr/Ca ratios in winter and the highest in summer occur near channels between the islands, forming an open connection between the ocean and Florida Bay, especially the channel between Long Key and Lower Matecumbe Key to the northeast. These two observations are the first indication that the seasonal pattern on the grid may originate in the shallow waters of Florida Bay. Figure S2 in Supporting Information S1 shows all summer and winter distributions from September 2016 to January 2020. It illustrates that the two distributions in Figure 2 are typical of the seasonal pattern, including the two noted characteristics, with the exception of 19W.
There are additional differences in the typical summer and winter patterns from year to year. For example, the winter pattern was more pronounced in 17W and 18W, whereas the summer pattern was more pronounced in 18S and 19S ( Figure S2 in Supporting Information S1). This may indicate some random variability, possibly induced by weather conditions or the tidal cycle, or a more persistent, lower-frequency pattern that was not fully captured within the duration of our time series. The situation in 19W deviates from the seasonal pattern, more closely resembling a muted summer distribution, for reasons that are not immediately obvious. We will return to these issues below. Note. Missing data were not recorded. SE = standard error. See Figure S1 in Supporting Information S1 for maps.
The seasonal reversal of the offshore Sr/Ca gradient is illustrated differently in Figure 3, which shows seawater Sr/Ca data on the grid points of consecutive offshore lines, progressing from east to west, for all distributions. The summer distributions show distinctly descending lines, indicating Sr/Ca ratios decreasing in the offshore direction, while the winter distributions show the opposite, again with 19W being a notable exception. Here as well, the lines are steeper that is, the gradients are stronger in some years than in others, in both seasons. The seasonal pattern emerges perhaps mostly prominently in Figure 4, which captures the entire time series in a single panel. This graph emphasizes the visually linear seaward decline of seawater Sr/Ca in summer (red curves) and the somewhat smaller but equally linear incline in winter (blue curves). An ANOVA and subsequent Tukey's HSD (honestly significant difference) tests of the data (N = 38-45 for each longshore line) confirm that in summer all non-adjacent lines are significantly different from each other (p < 0.01), whereas in winter only line A and line E are significantly different (p < 0.01 vs. p > 0.05 for all other line pairs). Moreover, mean summer and winter Sr/Ca ratios are significantly different on line A (p < 0.01), but not on line E (p > 0.05). Hence, Sr/Ca variability is greatest near the coast (line A) and gradually diminishes to almost constant Sr/Ca ratios, within error, at the deepest stations (line E), in every year. The approximately twofold higher standard deviation of Sr/Ca on line E in summer than in winter, taken over all years, may be explained by the fact that winter samples were always collected in January, whereas summer samples were collected variously between late June (18S) and the middle of September (16S), for logistical reasons. Figure 4 also suggests that the anomalous summer-like Sr/Ca behavior in 19W may have been restricted to the nearshore area (lines A and B).
According to the averages in Figure 4, the difference between the highest summer and the lowest winter Sr/ Ca ratios during the research period is about 0.19 mmol/mol (18S vs. 18W) on the nearshore line A, but only 0.014 mmol/mol (18S vs. 20W) on the offshore line E. The maximum difference on line E is about 0.038 mmol/mol in summer (different months) but only 0.007 mmol/mol in winter (same month). In terms of the full data set of individual grid points, the maximum difference was 0.29 mmol/mol (station 1B in 19S vs. station 3A in 18W). Khare et al. (2021) reported a sudden decrease of the seawater Sr/Ca ratio by 0.16 mmol/mol at station OS1 (Figure 1) over an interval of only 2 days around the landfall of hurricane Irma in September 2017. This was matched by a concurrent decrease in δ 18 O values, both apparently caused by a sudden influx of low-Sr/Ca (in summer) and less evaporated Atlantic water into Florida Bay, associated with the storm surge. As found by Khare et al. (2021) and confirmed here ( Table S1 in Supporting Information S1), the precision of our seawater ICP-AES method is about 0.01 mmol/ mol (1 SE), or <0.2% of the ambient seawater Sr/Ca ratio. The offshore (line E) Sr/Ca variability is evidently not significant (certainly not in winter), but the nearshore variability, the maximum-observed range of Sr/Ca ratios, and the rapid event observed by Khare et al. (2021) all definitely are. We should point out that the focus here is on spatial and temporal Sr/Ca variations that is, differences or relative values, rather than absolute values, since the former are most relevant to coral paleothermometry. We therefore compare differences with the precision of our method and not with its accuracy (0.04 mmol/ mol), which is determined by the present uncertainty of the GoM reference (Table 1). Seawater Sr/Ca variations do not depend on the exact value of the GoM reference, with appropriate corrections for offsets among individual batches (Section 2.3), as emphasized by the constant difference between the certified reference materials CASS-5 and CASS-6 ( Table 1).
The seawater Sr/Ca distributions with large seasonal changes in the nearshore samples and relative stability in the offshore samples (Figures 2-4 and Figure S2 in Supporting Information S1), as well as both the highest summer and lowest winter values occurring around open channels, strongly suggest that the source of the variability is to be found in the waters of Florida Bay. This can be illustrated more directly by showing the relation between our measured seawater Sr/Ca ratios and water temperature, as a function of bottom depth (Figure 5a). The relation emerges because water temperature has the same seasonally alternating distribution (not shown), with shallow nearshore waters being warmer in summer and colder in winter than the deeper offshore waters whose temperature is relatively constant all year. Figure 5a thus reveals a clear separation between warm high-Sr/Ca waters in summer and cold low-Sr/Ca waters in winter in the shallow coastal zone, which is strongly influenced by Florida Bay, but no correlation at stations deeper than about 15 m (note that isobaths are subparallel to the shoreline, Figure 1c). Of course, there is no implication that changes in seawater Sr/Ca are somehow caused by changes in temperature. Instead, both parameters are indicative of the mixing of Florida Bay water with Atlantic seawater in the coastal zone. This fact is reinforced by a similar relation between seawater Sr/Ca and a more subjective and less quantitative property: water color. For each collected sample, the ambient water color was recorded on an arbitrary scale of six shades, ranging from brown-green to dark blue. In Figures 5b and 5c, the relation between seawater Sr/Ca and water color on this scale is shown for summer and winter separately, as a function of bottom depth. It appears that most of the Sr/Ca variability is contained within the murky green water of Florida Bay, whereas little variation is found in the clear blue oceanic waters offshore. Since there is no reason to assume any causal connection between seawater Sr/Ca ratios and color, this again simply reflects physical mixing between these two endmembers.
The seasonal variation of seawater Sr/Ca as an expression of physical mixing is further supported by its behavior as a function of depth. Although the grid samples are all surface seawater (<0.5 m), bottom samples were also collected concurrently at four sites where equipment was deployed on the seafloor (OS1-4; Figure 1).
Whereas in the open ocean the 1%-2% increase in seawater Sr/Ca with depth occurs over the upper 1,000 m (de Villiers, 1999), these samples were necessarily restricted to depths that can be easily accessed by divers (<20 m). Some systematic variation with depth can nonetheless be seen in Figure S3 of Supporting Information S1, where bottom seawater Sr/Ca ratios (vertical bars) are compared with their average variation along the offshore lines (curves) already shown in Figure 4. The placement of the vertical bars corresponds to the position of the four OS stations relative to the grid, whereby station OS1 is to the north of line A, inside Florida Bay (Figure 1c). At first glance, bottom seawater Sr/Ca ratios closely follow those at the surface, decreasing offshore in summer and increasing in winter, suggestive of a well-mixed water column. Inside Florida Bay (station OS1), summer Sr/Ca ratios are generally equal to or higher and winter Sr/Ca ratios equal to or lower than those on line A. In addition, bottom values tend to be higher than at the surface in summer and lower in winter, possibly indicating the presence of denser Florida Bay water at depth, being warmer yet saltier in summer and fresher yet colder in winter. Deviations from these trends did occur and divers occasionally reported abrupt thermoclines associated with surface layers of distinctly different color or visibility. In combination, these observations confirm that variations with depth ( Figure S3 in Supporting Information S1) appear to be more directly related to the presence of Florida Bay water than to biological activity.
Given the major effect of endmember mixing on seawater Sr/Ca ratios it is to be expected that the measured distributions and seasonal variations ( Figure S2 in Supporting Information S1) were influenced to some extent by the tidal cycle. Strong cross-shelf currents were sometimes observed at transitions between high and low tides, especially around the channels, and pronounced tidal fronts were intersected while sampling, visible as a sharp demarcation between clear blue ocean water and murky green coastal water, often accompanied by lines of floating seaweed and organic debris. Some samples were also taken during afternoon thunderstorms or sudden squalls, although this was avoided as much as possible. With sampling of a single station and advancing to the next taking 15-20 min, completing the 54-site grid typically required two 6-hr workdays, which were not always consecutive, each corresponding to half of a tidal cycle. The tidal stage thus inevitably changed as the vessel progressed along the grid. While the tidal stage was always logged upon departure and return, it was not feasible to determine or correct for its evolution with every sample. In general, samples were collected sequentially along offshore lines, alternately moving with or against the tide from east to west (e.g., first sampling A-E on line 1, then E-A on line 2 etc.). However, wind and sea state occasionally forced us to move in the opposite direction, to follow longshore lines, or to adopt a more haphazard sequence. Any systematic aliasing of the tidal cycle was probably minimized, albeit not fully negated, by this inherent randomization. It certainly appears that tidal aliasing is neither a leading cause of, nor able to completely erase the measured seawater Sr/Ca gradients, although it may have suppressed them to some degree. It may also be responsible for some of the year-to-year variation in both summer and winter Sr/Ca distributions ( Figure S2 in Supporting Information S1). On the other hand, these may well be due to synoptic changes in hydrography or atmospheric conditions, related to shifts in the prevailing wind field, the passage of storm fronts, or meandering of the Florida Current (Lee et al., 1992). The most prominent example of this may be the Sr/Ca distribution in 19W, which resembles a summer pattern ( Figure S2 in Supporting Information S1), although we have not been able to tie this anomaly to a specific weather event or sea state. Sampling of the 19W grid was conventional, following alternating offshore and onshore lines, on the first day moving from the center of the grid east (line 6-5) and then from the eastern end back toward the center (line 1-4), continuing the next day from the center to the western end (line 7-11). The unusual inversion of the seawater Sr/Ca gradient appears to be confined to the two most nearshore lines (Figure 4), while the Sr/Ca ratio in Florida Bay (OS1) was clearly lower than on line A ( Figure S3 in Supporting Information S1). To place our samples within the context of the existing global data set of seawater Sr/Ca, we asked colleagues to take samples for us during cruises of opportunity in the subtropical and tropical zones of three oceans. These were collected using the protocol described in Section 2.1 and analyzed in the same manner as our Florida samples. The overall average of the results, listed in Table 3, is 8.646 ± 0.018 mmol/mol (1σ). Remarkably, this is nearly identical to 8.649 ± 0.017 mmol/mol (1σ), the average Sr/Ca ratio across all years and seasons on line E, representing the most offshore and hence the most oceanic waters on our grid. The same comparison is made for the individual world samples in Figure 6, which also shows the total range of seawater Sr/Ca ratios encountered during the research period. Samples from the Indian and Pacific Oceans and the single sample from Bermuda are all close to the line E average. A Student's t-test between the mean seawater Sr/Ca at line E (N = 78, winter and summer) and mean seawater Sr/Ca at each of our world stations ( Figure S1 in Supporting Information S1) confirms that only the samples from Barbados (p < 0.0005, N = 4) and the east coast of Florida (p < 0.0005, N = 5) are statistically different from offshore water in the Florida Keys. The data set is as yet too small to determine whether this indicates a higher level of Sr/Ca variability in the Atlantic Ocean. Conversely, Sr/Ca variability in the global ocean is clearly much lower than what corals experience seasonally on the Florida Keys Reef Tract.
The observed nearshore Sr/Ca variations in the Florida Keys are larger than those found in earlier studies of mostly open ocean seawater. Alibert et al. (2003) reported a maximum range of 0.033 mmol/mol for surface waters on a transect in the Great Barrier Reef, while Shen et al. (1996) observed non-cyclic variability of Sr/Ca at a single reef station in Taiwan with an annual range of 0.034 mmol/mol. A larger range of 0.076 mmol/mol was measured by de Villiers et al. (1994) for seawater from six global reef sites, yet such differences are normally accounted for with location-specific calibrations. These analyses were all obtained with the highly precise technique of thermal ionization MS using double-spike isotope dilution (ID-TIMS). Our observed Sr/Ca variations even exceed the maximum vertical gradients of 0.11-0.15 mmol/mol in the upper 500 m of the southern Atlantic and Pacific Oceans (de Villiers, 1999), yet they resemble gradients of up to 0.19 mmol/mol across the mixed layer (0-100 m) in the northwest Indian Ocean, possibly related to a recent Acantharia bloom, as reported by Steiner et al. (2020) who used an ICP-AES method similar to ours.
Only the comprehensive global survey by  yielded a much larger range of Sr/Ca ratios (7.70-8.80 mmol/mol) throughout the ocean at all depths, in apparent contrast to these observations, with extreme values down to below 4 mmol/mol yet without the distinctive vertical profile shown by de Villiers (1999). The authors state that Sr behaves non-conservatively in the ocean and particularly in coastal areas; that "we cannot absolutely assume that fossil archives using taxa-specific proxies reflect true global seawater chemistry"; and that some disparate Sr/Ca-based reconstructions might be reconciled by "assuming an error of […] 1 to 1.90 mmol/ mol." However, their 2σ precision of 0.14 mmol/mol relative to the value of the IAPSO standard (Lebrato et al., 2021) is of the same order as the total variability in the cited studies and fully half of the maximum difference in our Florida samples. While we agree that the marine geochemistry of Sr is more complicated than the steady-state semi-conservative behavior that is often implicitly assumed, given this level of uncertainty and the lack in their data set of a time dimension as well as spatial resolution on the scale of typical reef systems, we feel that these conclusions are perhaps premature and overly pessimistic.

Implications for Coral Paleothermometry in the Florida Keys and Elsewhere
We determined that corals on the central Florida Keys Reef Tract are subject to a seasonally alternating seaward gradient of surface seawater Sr/Ca ratios with a maximum peak-to-peak amplitude of 2%-3% (0.19-0.29 mmol/mol with a mean of 8.65 mmol/mol). Assuming a median slope of −0.059 ± 0.004 (mmol/mol)/°C (±2SE) for Sr/Ca-SST calibrations of tropical corals (from Table S1 of Ross et al. (2019)), our annual and total Sr/Ca variations, if unaccounted for, represent a potential and plainly unacceptable uncertainty of 3.0-5.3°C in reconstructed SST. In practice, a regular seasonal cycle of seawater Sr/Ca ratios that are high in summer and low in winter will result in a decreased calibration slope for subannual regressions of coral Sr/Ca versus SST.
It is important to recognize that this situation is fundamentally different from the known, more modest spatial variability of seawater Sr/Ca from reef to reef, which is routinely accounted for with location-specific calibrations of coral aragonite Sr/Ca to local SST (Corrège, 2006). That approach explicitly cannot correct for temporal variability and the observed seasonal variation of seawater Sr/Ca is particularly pernicious since it mimicks the SST-induced seasonal fractionation of Sr and Ca in coral aragonite on which the paleotemperature proxy is based. Higher SST in summer leads to relatively less Sr accumulating in the coral skeleton, but this is partly offset by the higher seawater Sr/Ca ratio at that time of the year. In winter, lower SST leads to relatively more Sr accumulating in the coral skeleton when the Sr/Ca ratio of the ambient seawater is lower. Therefore, the seasonal variation of seawater Sr/Ca effectively dampens the dependence of coral aragonite Sr/Ca on SST, making the proxy less sensitive and diminishing the slope of the Sr/Ca-SST calibration.
Coral calibration studies in the Florida Keys illustrate the effects of the apparent seasonal seawater Sr/Ca cycle. J. M. Smith et al. (2006) found a slope of −0.028 (mmol/mol)/°C for the sensitivity of coral aragonite Sr/Ca to the ambient SST signal, when calibrating twentieth-century Orbicella faveolata from Looe Key, to the west of our grid. Applying a calibration slope of -0.047 (mmol/mol)/°C for the same species from outside Biscayne Bay , to the north of their study site, revealed an apparent SST bias of up to 6°C. Tentatively attributing this anomaly to a seasonal variation in seawater Sr/Ca, J. M. Smith et al. (2006) estimated the effect by deconvolving the contribution of measured SST from the coral Sr/Ca signal, using the generic Sr/Ca-SST relation of Kinsman and Holland (1969) and S. V. Smith et al. (1979). They calculated a mean monthly seawater Sr/Ca ratio with a sinusoidal variation from a minimum of ∼8.60 mmol/mol in February to a maximum of ∼8.95 mmol/ mol in August. Although their annual average is somewhat offset from our line E average of 8.65 mmol/mol, the resulting annual cycle and total range of ∼0.35 mmol/mol are in remarkable agreement with our results. Our data currently consist of just two measurements in each of four consecutive years, which is hardly sufficient to establish an annual cycle. While these points seem to have fortuitously captured nearly the entire range of local seawater Sr/Ca variability, sampling at much higher temporal resolution is required to adequately characterize its behavior. Such resolution was achieved with the osmotic pumps deployed at OS1-4 ( Figure 1); however, the resulting large data set (4 years of mostly continuous weekly bottom samples at each site) is part of a differently focused project and will be presented in a separate paper. Nevertheless, our results strongly suggest that the mechanism proposed by J. M. Smith et al. (2006) to explain their anomalous data is real.
Corals from areas in the Florida Keys in less direct communication with Florida Bay do not show a similarly strong attenuation of the seasonal Sr/Ca signal. Much work calibrating the coral Sr/Ca paleothermometry proxy has been conducted in the Upper Keys  and in the Dry Tortugas, the westernmost islands of the chain (DeLong et al., 2011;Maupin et al., 2008). Both places are well known to be less impacted by Florida Bay water (Boyer & Jones, 2002;Lee et al., 2002). The Dry Tortugas are more strongly influenced by the open ocean characteristics of the West Florida Shelf and Loop Current (Lee et al., 1994). In the Upper Keys, the islands themselves form a barrier to water transport and conditions are influenced more by intrusions of the Florida Current onto the shelf (Boyer & Jones, 2002). Slopes of Sr/Ca-SST calibrations for corals from the Upper and Lower Keys are substantially steeper than that found by J. M. Smith et al. (2006) in the Middle Keys, ranging from −0.039 (mmol/mol)/°C (Maupin et al., 2008) to −0.043 (mmol/mol)/°C (DeLong et al., 2011;Kuffner et al., 2017) for Siderastrea siderea and from −0.0392 (mmol/mol)/°C (Flannery & Poore, 2013) to −0.047 (mmol/mol)/°C  for Orbicella faveolata. These slopes are on the lower end of normal for Pacific corals (Corrège, 2006), but comparable to those found in other parts of the Caribbean (Giry et al., 2010;Hetzinger et al., 2006;Xu et al., 2015).
Although our results clearly point to Florida Bay as the source of the seasonal variations, their chemical or biological origin is still speculative and outside the scope of this paper. Florida Bay is a shallow estuary, open to the west and separated from the ocean by the Florida Keys to the south and east, receiving freshwater from the Everglades to the north. A negative water budget due to excess evaporation in summer can yield hypersaline conditions of as much as 70 (Swart & Price, 2002). Millero et al. (2001) calculated aragonite and calcite saturation states from measurements of DIC parameters at various times of the year during 1997-2000. They showed that the Everglades are a source of alkalinity to Florida Bay and that alkalinity and TCO 2 within Florida Bay itself are lowest at times of high salinity, evidently due to the precipitation of calcium carbonate, possibly related to sediment resuspension or biomineralization by calcareous algae. Complementary work by Yates and Halley (2006) and Yates et al. (2007), who measured diurnal calcification cycles on mud banks and within benthic chambers in the intervening channels in central and eastern Florida Bay, yielded a loss of alkalinity during the day (net precipitation) and an excess of alkalinity during the night (net dissolution). Correlations with dissolved O 2 and CO 2 and with light transmission suggested a biological origin for these processes, tracking daily photosynthesis/ respiration cycles.
Regardless of their cause, it is quite possible that the fractionation of Sr and Ca associated with the precipitation, dissolution, and interconversion of aragonite and calcite is a main driver of the observed seawater Sr/Ca variations in Florida Bay. The lack of Sr/Ca variability in offshore waters seems to preclude a major role for seasonal upwelling of deep water. However, alternative explanations such as annual cycles in freshwater supply from the Everglades or groundwater input cannot yet be ruled out. Our two attempts to constrain the Sr/Ca ratio of local saline groundwater were inconclusive. The KML well sample had a distinctive Sr/Ca ratio of 6.857 mmol/ mol, but its low salinity suggests that it was greatly diluted by freshwater from a recent rainstorm that may have dissolved some low-Sr limestone from the surrounding pavement. The deeper Aquarium Encounters well sample had a Sr/Ca ratio of 8.714, somewhat above the mean yet well within the range of seawater on our grid. A more discriminative analysis, such as Sr or Ba stable isotopes (Cao et al., 2016;Katz et al., 1997), may be required to construct a definitive mixing model.
This study provides a new perspective on the coral paleotemperature proxy, however in no way lessens its power or importance as a key tool in climate studies of the recent past, where it continues to be widely used with great success. Our work does serve as a caution that its validity may not be ubiquitous and should not be taken for granted. While the use of favorable coral species and a prudent application of suitable SST data (Corrège, 2006), combined with careful calibrations employing appropriate statistical techniques (Xu et al., 2015), can adequately account for spatial variations in seawater Sr/Ca, it is clear that this may not be sufficient in locations that display regular and long-term temporal variability. The effect of more episodic and short-lived Sr/Ca excursions, for instance as a result of major storms (Khare et al., 2021), may be limited to some degree by the inherently integrative nature of coral biomineralization and would consequently be difficult to detect without continuous chemical monitoring of the ambient seawater. Our seawater ICP-AES method (Khare et al., 2021), in combination with compact, low-maintenance in situ sampling devices (Jannasch et al., 2004), provides an accessible tool to accomplish this and to further ensure and improve the fidelity of coral-based paleoclimate data. Rather than using seawater Sr/Ca variations to correct coral Sr/Ca-SST calibrations, it would seem advantageous to avoid locations where such variations are observed. Given the high cost and logistical challenges of retrieving coral cores from remote and delicate reef systems, we recommend that seawater Sr/Ca ratios should be routinely monitored ahead of such operations, especially in places where variations are made more likely by aspects of the local geography or hydrology.