Climate and the latitudinal limits of subtropical reef development

Climate plays a central role in coral-reef development, especially in marginal environments. The high-latitude reefs of southeast Florida are currently non-accreting, relict systems with low coral cover. This region also did not support the extensive Late Pleistocene reef development observed in many other locations around the world; however, there is evidence of significant reef building in southeast Florida during the Holocene. Using 146 radiometric ages from reefs extending ~ 120 km along Florida’s southeast coast, we test the hypothesis that the latitudinal extent of Holocene reef development in this region was modulated by climatic variability. We demonstrate that although sea-level changes impacted rates of reef accretion and allowed reefs to backstep inshore as new habitats were flooded, sea level was not the ultimate cause of reef demise. Instead, we conclude that climate was the primary driver of the expansion and contraction of Florida’s reefs during the Holocene. Reefs grew to 26.7° N in southeast Florida during the relatively warm, stable climate at the beginning of the Holocene Thermal Maximum (HTM) ~ 10,000 years ago, but subsequent cooling and increased frequency of winter cold fronts were associated with the equatorward contraction of reef building. By ~ 7800 years ago, actively accreting reefs only extended to 26.1° N. Reefs further contracted to 25.8° N after 5800 years ago, and by 3000 years ago reef development had terminated throughout southern Florida (24.5–26.7° N). Modern warming is unlikely to simply reverse this trend, however, because the climate of the Anthropocene will be fundamentally different from the HTM. By increasing the frequency and intensity of both warm and cold extreme-weather events, contemporary climate change will instead amplify conditions inimical to reef development in marginal reef environments such as southern Florida, making them more likely to continue to deteriorate than to resume accretion in the future.

on the Miami OR and that the deeper portions of the OR, which we were not able to sample in this study, began growing during the Early Holocene, around 10 ka.
As with the OR samples from Palm Beach, we were only able to collect surface samples from the Miami IR. The oldest age of the samples from Miami is 5.3 ka and most cluster around 3.3 ka.
Unlike the samples from Palm Beach, however, the samples from the Miami IR were collected over a narrow depth range of 2.7 m (-8.5 to -5.8 m MSL) 5 . We do not, therefore, have data from the majority of the lifespan of the Miami IR, which is why there is a gap in the record of Acropora palmata reef development in Miami from 7.2 to 3.3 ka (with the exception of the three samples from 5.3-4.5 ka); however, a colony of Montastraea cavernosa found just above the Pleistocene surface in a section of the south Miami IR exposed by a pipeline installation provides an estimate of the depth of the IR initiation surface in this region: approximately -10 m MSL 8 . As we discuss in the main text, this is similar to the depths-to-Pleistocene on the IR of Broward (-9 to -12 m MSL) 3,4 , which suggests that both reefs would have likely initiated by the beginning of the Middle Holocene. Furthermore, surveys of the Miami IR within the Port Miami dredge exposure by WFP using underwater depth gauges suggested that the Holocene reef there extended to at least -15.2 m MSL (the depth of the channel), so the IR-initiation surface in some parts of Miami may be even deeper than in Broward.

Estimates of accretion rates
We estimated the rates of reef accretion during the Early Holocene from A. palmata-dominated reef framework at the Fowey Rocks Outlier Reef in south Miami 9,10 and from the two most reliable sequences sampled on the Broward OR (from Sequences B and C of Lighty et al. 2 ) 6 . All of the samples we collected from the reefs in Palm Beach were surface samples, so it was not possible to derive estimates of reef accretion for that subregion. The base of the first core from Fowey Rocks (BP-FR-1) was dominated by Orbicella spp. rather than A. palmata, and because there were significant age-reversals in this section 10 it was not included in this analysis. Additionally, because five A. palmata ages in this core from the Early Holocene had significant overlap in the 2σ-ranges of adjacent ages (i.e., they were not statistically different from one another) 10,11 , only the oldest and youngest (non-overlapping) A. palmata ages from the Early Holocene were used to estimate reef accretion at this location. Similarly, because the bottom-two ages from the second Fowey Rocks core (BP-FR-2) were not significantly different from one another, we only calculated accretion rates from the two non-overlapping A. palmata ages for the Early Holocene. We also calculated an accretion rate from the interval 9.4-8.0 ka in this core, but we did not include it in any statistical analysis because the upper age was not from A. palmata and it was unclear how much of the interval was deposited during the Early Holocene as opposed to the Middle Holocene.
We were able to estimate accretion rates from the top-and bottom-ages of two sequences (B and C) from the OR of Broward 1,2 . The middle ages in those sequences were excluded because they were not significantly different from adjacent ages (Sequence B) or because of a significant agereversal (Sequence C) 5 . The data from Sequences A and D were not included because the corals in those sequences were likely not in situ 6 .
The reef framework of the SFCRT was dominated by A. palmata during the Early Holocene, when the rates of sea-level rise were most rapid 1,2,5,12 . By the Middle Holocene, however, sea-level rise had slowed, and, as a result, reefs composed of mixed A. palmata and massive-coral framework were common 3 . We considered the rates of reef accretion by A. palmata and massivecoral or mixed facies separately, using published core logs as a guide 3,10 , so that accretion rates of A. palmata reefs could be directly compared between the Early and Middle Holocene. We were able to estimate Middle-Holocene reef accretion from one A. palmata-dominated sequence sampled in this study from the Government Cut dredge exposure (PM-25mE) 5 . Again, because many of the nine ages in this sequence were not significantly different from one another we County. A core collected from the same region by Banks et al. 4 (core BR-IR-B-1) provides one additional estimate of accretion of an A. palmata reef from the IR. Those researchers also collected the only core from the MR (BR-MR-FL-1), and we used that record to estimate accretion by massive corals in this habitat during the Middle-to-Late Holocene. Finally, we calculated accretion rates from the non-Acropora reef framework preserved in the upper layer of each of the cores from Fowey Rocks. We note, however, that accretion on the Fowey Rocks Outlier Reef terminated around the same time as the ORs of the SCRFT 10 . By the Middle Holocene these reefs were in relatively deep water (paleodepths of 7.8 and 10.5 m for BP-FR-1 and BP-FR-2, respectively) 10 , the reef was no longer keeping pace with sea level, and accretion was negligible compared with shallow-water reefs in the region. We did not, therefore, include these data in the statistical analyses of reef accretion on the SFCRT.
We compared our data on accretion of reef framework by A. palmata to framework built by massive/mixed species from the IR of Broward County to test the hypothesis that rates of reef accretion did not differ between these facies. The data were compared with an independent t-test.
The data met the assumptions of normality (Shapiro-Wilk test: W=0.95, p=0.49) and homogeneity of variances (Levene's test: F1,13=1.06, p=0.32) after natural-log transformation. We found that although accretion rates of facies dominated by massive corals were somewhat slower, at 2.2 m 6 ky -1 (range=1.1-4.3 m ky -1 ), than that of A. palmata facies, at 3.5 m ky -1 (range=1.1-7.5 m ky -1 ; Table 1, S2), the difference was not significant (t-test: t13=1. 34, p=0.200). This result supports the conclusion of several previous studies that although annual growth rates of A. palmata are significantly faster than those of massive species, the difference is offset over millennial timescales because of the impacts of processes like reef-framework compaction, bioerosion, and sedimentation 11,13,14 .

Climatic correlates of cold-front frequency in south Florida
A number of studies have demonstrated that the frequency of severe winter cold fronts reaching the southern United States is strongly linked to the relative intensity of meridional versus zonal atmospheric circulation over the continent [15][16][17] (Fig. 5). The association between increased coldfront frequency and intense meridional flow is especially strong in southern Florida, where the occurrence of cold fronts is highly variable from year to year 16,18 . The oscillation between the two modes of atmospheric variability is described by the Pacific North American (PNA) pattern, in which a positive PNA indicates dominance of meridional flow, with increased penetration of winter cold fronts to the south, and a negative PNA indicates more zonal flow and fewer cold fronts 16,19 .
As expected, the PNA index has been a strong predictor of recent cold-front frequency in the southern United States in general [15][16][17] , and Florida in particular 16,18 . Indeed, 80% of citrus freezes in Florida between 1899 and 1989 were associated with atmospheric circulation patterns consistent with a positive PNA 18,20 . Furthermore, in the month preceding the winter cold snap of 2010, the coldest winter on record since 1950, there was an abrupt shift in the PNA to the positive phase, which triggered cold air from the northern United States to flow south 15 . The 2010 cold event was also preceded by a shift in the North American Oscillation (NAO), an index of pressure gradients between the Bermuda High (also called the Azores High) and the Iceland Low in the northern Atlantic, to a negative phase (stronger Iceland Low) 15 . Consequently, there is also a strong association between the NAO index and historic cold-front frequency in south Florida [16][17][18] . Several studies have evaluated the relationship between cold-front frequency and the El Niño-Southern Oscillation (ENSO); however, whereas some analyses have found a link between cold fronts and El Niño 16,19 , the relationship is complex and not always predictable. Other studies found either no relationship between cold fronts and ENSO 16 or that the influence of other climate drivers (i.e., PNA and NAO) were stronger during ENSO-neutral years 17 . Because of the uncertainty related to ENSO as a potential driver of cold-front frequency, we refrain from drawing any conclusions about how changes in ENSO variability during the Holocene may have influenced the millennial-scale changes in the climate of south Florida.
A recently compiled oxygen-isotope record of precipitation from throughout the United States over the Middle to Late Holocene demonstrated that PNA-like oscillations occurred throughout at least the last 8,000 years 19 . Although this record provides a direct proxy for a long-term shift from PNA-to PNA+ by the Late Holocene, shorter-term oscillations in the degree of meridional circulation are less apparent 19 . As a result, we relied on other, high-resolution paleoproxies to infer likely changes in winter cold-front frequency during the Early and Middle Holocene. One such proxy is a record of sea-salt flux to Greenland as recorded in the GISP2 ice core 21 . That study identified five periods of enhanced marine deposition in that record-before 11.3, 8.8-7.8, 6.1-5.0, 3.1-2.4, and 0.6-0 ka-which they interpreted as representing relatively cool periods associated with expansion of the polar vortex and/or enhanced meridional circulation 21 . This record suggests that in addition to millennial-scale shifts in meridional circulation, higher frequency variability may have also played an important role modulating cold-front frequency.
Another potential indicator of the relative degree of meridional flow is the position of the intertropical convergence zone (ITCZ). In the modern environment, the ITCZ tracks meridional heat flux, migrating annually between ~2°N in the boreal winter and 9°N in the boreal summer over the Atlantic, eastern Pacific, and central Pacific 22 . Likewise, over longer, millennial timescales, the mean position of the ITCZ largely followed changes in Northern Hemisphere summer insolation, having occupied its northernmost position during the relatively warm climate of the Holocene Thermal Maximum (Fig. S5; HTM), and having moved to the south as climate Periodic development of intense and persistent upwelling-favorable winds during the late spring and early summer can also cause cold anomalies on the SFCRT 26,27 . These winds bring cold, nutrient-rich water with characteristics similar to deep Gulf Stream water masses onto the outer shelf of the SFCRT 28,29 . The resulting temperatures on the outer shelf can be more than 10°C colder than climatological means 30 , often resulting in harmful algal blooms and mortality of thermophilic species 31 . These periodic cold-water upwelling events may limit modern coral growth and reef development throughout the SFCRT, especially along the OR 32 ; however, because these events are not predictable over millennial timescales, we were not able to evaluate the possibility that upwelling contributed to reef shutdown on the SFCRT during the Holocene.

Holocene temperature reconstructions
In recent years, there have been a number of efforts to combine the available paleotemperature data into a composite temperature reconstruction for the Holocene [33][34][35][36] . The best-known is the reconstruction developed by Marcott et al. 33 , which combined 73 records from around the world to create a composite of global temperature change over the last 11.3 ky (Fig. S5). Since its publication, several follow-up studies have emphasized that this 'global' composite is strongly influenced by records from the North Atlantic and may not truly reflect global temperature 33,36,37 .
The bias is evident when the global temperature stack of Marcott et al. 33 (Fig. S5, black line) is compared with a reconstruction that excludes records from the North Atlantic (Fig. S5, light gray line), which shows no significant change in temperature over the last 10 ky. In contrast, the Holocene temperature trends reflected in the global curve are even more exaggerated when only Northern Hemisphere records are considered (Fig. S5, dark gray line), suggesting that there are broader-scale climate influences on the global trend.
Although it is clear that the trend from warming during the HTM to cooling in the Late Holocene reflected in the Marcott et al. 33 global temperature composite is largely driven by highlatitude, Northern Hemisphere variability, we chose to focus on this reconstruction in our discussion of changes in mean climate during the Holocene for several reasons. Most importantly, environmental variability in the subtropical habitats of south Florida is closely linked to broaderscale, Northern-Atlantic climate via a variety of oceanographic and atmospheric teleconnections 16,19,[38][39][40] . As a result, mean shifts in the climate of south Florida are likely correlated with extra-tropical (30-60°N) to high-latitude Northern Hemisphere variability, more so than are the other, more tropical coral-reef ecosystems of the western Atlantic 11,39 . Recently, Kaufman et al. 35 provided an updated, but lower-resolution, series of temperature reconstructions based on 679 records from around the world (Fig. S5, blue lines). The general trends in the Marcott et al. 33