Shipwreck rates reveal Caribbean tropical cycloneresponse to past radiative forcingValerie Troueta,1, Grant L. Harleyb, and Marta Domínguez-Delmásc,d

aLaboratory of Tree-Ring Research, University of Arizona, Tucson, AZ 85721; bDepartment of Geography and Geology, University of Southern Mississippi,Hattiesburg, MS 39402; cDepartment of Botany, University of Santiago de Compostela, 27002 Lugo, Spain; and dDepartment of History I, University ofHuelva, 21071 Huelva, Spain

Edited by Kerry A. Emanuel, Massachusetts Institute of Technology, Cambridge, MA, and approved February 2, 2016 (received for review October 2, 2015)

Assessing the impact of future climate change on North Atlantictropical cyclone (TC) activity is of crucial societal importance, but thelimited quantity and quality of observational records interferes withthe skill of future TC projections. In particular, North Atlantic TCresponse to radiative forcing is poorly understood and creates thedominant source of uncertainty for twenty-first-century projections.Here, we study TC variability in the Caribbean during the MaunderMinimum (MM; 1645–1715 CE), a period defined by the most severereduction in solar irradiance in documented history (1610–present).For this purpose, we combine a documentary time series of Spanishshipwrecks in the Caribbean (1495–1825 CE) with a tree-growthsuppression chronology from the Florida Keys (1707–2009 CE). Wefind a 75% reduction in decadal-scale Caribbean TC activity duringthe MM, which suggests modulation of the influence of reducedsolar irradiance by the cumulative effect of cool North Atlantic seasurface temperatures, El Niño–like conditions, and a negative phaseof the North Atlantic Oscillation. Our results emphasize the needto enhance our understanding of the response of these oceanicand atmospheric circulation patterns to radiative forcing and cli-mate change to improve the skill of future TC projections.

Caribbean | tropical cyclone | Maunder Minimum | dendrochronology |documentary data

Landfalling tropical cyclones (TCs) bring devastation to natu-ral and human landscapes with floods, winds, and storm

surges. In recent decades, TC-related human mortality andeconomic losses have risen in step with increasing populations inhigh-risk coastal communities (1). TC damage is expected tofurther increase in the near future with rising exposure andprojected anthropogenic climate change (2). This is particularlythe case for the North Atlantic Basin, which is one of the mostTC-active basins globally. The development of successful adap-tation and mitigation strategies relies on skillful projections ofNorth Atlantic TC activity, as well as an improved understandingof the drivers of its variability.Modeling studies of twenty-first-century global TC activity

generally converge in their projections of increased TC intensityand decreased frequency, but the magnitude range of projectedNorth Atlantic TC variability is wide (3). Uncertainties in twenty-first-century North Atlantic TC projections are largely driven bythe chaotic nature of the climate system and by our limited un-derstanding of TC response to radiative forcing, including an-thropogenic greenhouse gases and aerosols, as well as naturalvariability in volcanic and solar activity (4). Response uncertaintyis the dominant source of uncertainty toward the end of thetwenty-first century (4), with different model runs resulting in TCresponses of opposing sign to projected radiative forcing (3). Ourunderstanding of TC response to radiative forcing—and thus theskill of future TC projections—is restricted by limitations in thetime-series length and quality of observational records (5) thathinder trend detection and attribution (3).To attribute significant TC changes to specific climate forc-

ings, recent TC activity needs to be placed in a longer-termcontext (6). The Maunder Minimum (MM; 1645–1715 CE), the

most severe change in solar irradiance in documented history (7,8), is of particular interest in this context, but TC proxy recordsthat cover this period are scarce, often present a conservativeestimate of the total number of storm events (9), and largely haveinsufficient time resolution to distinguish the MM (6, 10–12).Documentary data sets are the main source of paleotempestologyinformation of appropriate temporal resolution, but most docu-ment-based TC studies have primarily focused on long-term TCclimatology (e.g., seasonality, recurrence intervals) rather thaninterannual or decadal-scale variability (13, 14). Here, we combinetwo annual resolution proxy records—a documentary time seriesof Spanish shipwrecks in the Caribbean (TCship) and a tree-growthsuppression chronology from the Florida Keys (TCsupp)—to ex-tend the observational Caribbean TC (CTC) record back over thelast 500 y and thus to cover the MM.Over the past centuries, TCs have caused destruction of hu-

man settlements and wreaked havoc at sea. In the Caribbean,TCs were the primary documented cause of shipwrecks in thesixteenth through eighteenth centuries (15) and they left theirmark on regional history. For instance, Spanish hegemony overCuba was secured in 1640 after a hurricane decimated a Dutchfleet poised to attack Havana (15), leading to an additionalcentury of Spanish monopoly over trade between the Caribbeanand Europe. We make use of the well-documented maritime TClegacy in the Caribbean region to reconstruct CTC variability.Our reconstruction (TCship) is based on a comprehensive docu-mentary compilation (16) of 657 ships of Spanish originthat wrecked in the Caribbean Basin (Fig. 1A and Tables S1and S2) over the period 1495–1825 CE due to storms orunspecified factors.

Significance

Twenty-first-century North Atlantic tropical cyclone (TC) pro-jections are crucial for the development of adaptation andmitigation strategies, but they are subject to large uncer-tainties, particularly with respect to TC response to radiativeforcing. We used a combination of tree-ring data and historicalshipwreck data to show that TC activity in the Caribbean wasdistinctly suppressed during the Maunder Minimum (1645–1715 CE), a period when solar irradiance was severely reduced.This solar fingerprint on decadal-scale Caribbean TC variabilityimplies modulation by a combination of basin-wide climaticphenomena. Our findings highlight the need to enhance ourunderstanding of the response of atmospheric circulation pat-terns to radiative forcing and climate change to improve theskill of future TC projections.

Author contributions: V.T., G.L.H., and M.D.-D. designed research, performed research,analyzed data, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. Email: trouet@ltrr.arizona.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519566113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1519566113 PNAS | March 22, 2016 | vol. 113 | no. 12 | 3169–3174

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

TCship does not overlap in time with the North AtlanticHurricane Database (HURDAT; 1851–2010 CE) (17) and there-fore we used TCsupp (1707–2009 CE) to assess its validity as aCTC proxy. TCsupp is based on 38 south Florida slash pine (Pinuselliottii var. densa) trees from Big Pine Key that show commonpatterns of suppressed growth (Fig. 1B). The primary causes oftree-growth suppressions in the Florida Keys are high-energywinds and storm-surge-induced saltwater intrusion and TCsupp

thus reflects interannual variability in landfalling CTCs. Owingto the geography and regional positioning of the Florida Keysrelative to the Caribbean Sea and Atlantic Ocean (Fig. 1A) andto the decadal-scale relationship between landfalling and basin-wide TC dynamics (18, 19), TCsupp can also be interpreted to re-flect Caribbean basin-wide TC dynamics.

TCsupp captures 40 of the 44 storms—occurring during 31 outof a total of 34 storm years—that tracked within 160 km of BigPine Key over the instrumental period (Figs. 1A, 2A, and S1;Tables S3 and S4). Widespread suppressions in tree growth (highTCsupp values) corresponded to storm years (χ2 = 255, P < 0.001)and typically occurred the same year (t0; P < 0.001) or 1 y after(t+1; P < 0.01) a TC event (Figs. 1C and 2A). We interpret thedouble peak in the superposed epoch analysis (SEA) results torepresent the immediate or delayed effect of wind and stormsurge damage on tree growth. The timing of a TC event duringthe hurricane season (August to October) relative to the growingseason of trees (February to November) (20) likely dictateswhether suppression in growth occurs at t0 or t+1, but a signifi-cant pattern was not found.

Lag year

0 1 2 3-3 -2 -1

PD

SI (

depa

rtur

e ac

tual

from

si

mul

ated

)

0.2

0

0.4

0.6

-0.2

A

B C

Fig. 1. Geographical location, storm-induced tree growth suppression, and its climatic signal at Big Pine Key, Florida. (A) Geographical location of the BPKtree-ring site (red dot), HURDAT-derived (1851–2010 CE) (17) category 1–5 TCs (white lines; n = 44) that tracked within 160 km (red buffer ring) of the site, andcountries/states for which shipwrecks were recorded (black shading) (16); (B) Slash pine (P. elliottii var. densa) section (dated over 1707–1829 CE) with tree-growth suppressions (multiple consecutive narrower-than-average annual rings) resulting from TC events indicated by date of recorded TC event; (C) SEA(1895–2009 CE) of TCsupp event series (defined as years when >75% of samples showed growth suppression; n = 12) with July–November Palmer DroughtSeverity Index (PDSI) for Florida climate division 7.

3170 | www.pnas.org/cgi/doi/10.1073/pnas.1519566113 Trouet et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

A similar pattern occurs when comparing TCsupp to TCshipevents: anomalously high TCsupp values occurred in the year of(t0; P < 0.001) or the year after (t+1; P < 0.05) severe TCshipevents (χ2 = 470, P < 0.0001; Fig. 2B), suggesting a high syn-chronicity between years of damage at sea (shipwrecks) and onland (pines). Comparable patterns were found when selectingshorter but better replicated TCsupp time series (Fig. S2) andwhen using a TCship time series based only on shipwrecks thatwere documented to be caused by storms (Fig. S3). Furthermore,this “double peak” SEA pattern also occurs when using a com-bined storm event list (1707–2010 CE) including TCship, docu-mented historical storms (14), and HURDAT data (Fig. 2C).TCsupp and TCship combined provide an annual resolution

record of CTC variability that covers the Little Ice Age (ca.1500–1850 CE), a period characterized by unusually cool globaltemperatures driven largely by a combination of strong volcanicactivity and fluctuations in solar irradiance (21). Over this pe-riod, TCship shows distinct decadal-scale variability (Fig. 3A) thatcorresponds to decadal-scale variations in an independent ship-wreck compilation (Fig. S4). By calculating decadal-scale per-centages of shipwrecks per number of active North AtlanticSpanish ships (TCship%), we verified that decadal-scale fluctua-tions in TCship were not a function of fluctuations in Spanishtransatlantic shipping activity (Figs. 3B and S5). Both TCship andTCship% reveal a distinct reduction of CTC activity during theMM (Fig. 3 A–D), which is also expressed as a reduction in thenumber of years with shipwrecks per decade (Fig. 3C). Thisfinding is supported by historical evidence from ship logbooks(14) suggesting increased CTC activity during the 1550s–1640sfollowed by a marked reduction until the 1720s (Fig. 3B) andby a lake sediment geochemical record from Jamaica (22) (Fig.3E). Reduced seventeenth-century CTC activity has also been

recorded in an array of Caribbean and Gulf of Mexico lower-resolution TC proxies (10–12, 23–25) and is concurrent withincreased ocean stratification—possibly linked to reduced TC-induced mixing—near Great Bahama Bank (26) and increasedaridity in various proxy records from the circum-Caribbean re-gion (27, 28).Modern CTC variability is influenced by four key large-scale

climate state variables (3, 29–31): (i) absolute tropical Atlanticsea surface temperatures (SSTs) (32); (ii) tropical Atlantic SSTsrelative to tropical mean SSTs (relative SSTs) (33); (iii) the ElNiño–Southern Oscillation (ENSO); and (iv) the North AtlanticOscillation (NAO). A combination of proxy records suggests thatthe states of these climate factors converged during the Little IceAge, and the MM in particular, to create a drastically unfavor-able environment for CTC development. The distinct reductionin CTC activity during the MM can be attributed to cool absoluteSSTs in the Caribbean (34, 35) and in the main developmentregion (36): warmer ocean waters provide more thermal andkinetic energy and thus increase the potential for a greaternumber of severe storms (31) (Fig. 3E). Moreover, west AtlanticSST anomalies during the MM were on average −0.21 °C coolerthan SST anomalies in the other (Indian, west Pacific, and eastPacific) tropical ocean basins (based on data from ref. 37; Fig.3E), resulting in cool relative SSTs. Anomalously cool absoluteand relative SSTs resulting from generally low sunspot activityduring the MM (8) could therefore have impacted the thermo-dynamic environment, vertical wind shear, and troposphericstability, thus creating conditions adverse to CTC development.The MM was likely also a period of broad-scale climate shifts.

Model simulations suggest a southerly position of the intertropicalconvergence zone, an El Niño–like mean state, and negative NAOconditions (7) during the late seventeenth century, that have been

A B

C

Fig. 2. Instrumental and reconstructed CTC activity. SEA of the TCsupp time series with instrumental [HURDAT, ref. 17; 1851–2010 CE] (A); and reconstructed(TCship; 1495–1825 CE) (B) CTC events. SEA was performed for TCsupp contemporaneous to (year 0) and lagging (1 y before, 5 y following) the TC event years.HURDAT TC event years were selected based on category 1–5 storms that tracked within 160 km radius from Big Pine Key (A). TCship event years were definedas years with five or more shipwrecks (16) (B). The combined event list (C) consisted of TCship, HURDAT, and documented historical storms (14) (1826–1849 CE;asterisks above TCsupp).

Trouet et al. PNAS | March 22, 2016 | vol. 113 | no. 12 | 3171

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

confirmed by proxy evidence (35, 36, 38, 39). Modern CTCclimatology—with generally suppressed TC activity during El Niñoyears and negative NAO phases—alludes to a likely contributionof the state of these major circulation patterns to TC suppressionin the Caribbean during the MM. The majority of sedimentary TCproxies on the northeastern US coast, however, did not record aTC suppression during the MM (9, 36), suggesting an importantinfluence of regional oceanic conditions and shifting genesis lo-cations as well as a potential modulation—and northeastwardredirection—of the prevailing storm tracks related to the above-described changes in atmospheric circulation patterns (30).Our high-resolution CTC proxy records suggest that a com-

bination of cold Caribbean SSTs, El Niño–like conditions, and a

negative NAO during the MM might have provided a pathwayfor an unfavorable effect of reduced solar activity on decadal-scale CTC activity. However, on interannual and even daily timescales, the solar–TC relationship is complex, spatially explicit,and modulated by oceanic heat content (40). In regions with highoceanic heat content (e.g., the twentieth-century Caribbean), TCactivity is typically reduced during years of strong solar activity,when tropopause temperatures are warmed and the verticaltemperature differential responsible for tropospheric convectionis weakened (41). In contrast, in regions where ocean heat contentis limited in summer (e.g., the twentieth-century east Atlantic),strong solar activity can provide a marginal SST increase necessaryfor cyclogenesis, resulting in more TCs. This modulation of the

A

B

C

D

E

Fig. 3. Reduced CTC activity during the Maunder Minimum. (A) Decadal-scale TCship and instrumental (HURDAT) time series sums and TCsupp averages;(B) TCship% and decadal documentary Caribbean storm counts (14); (C) number of years per decade with more than one or four shipwrecks in TCship; (D) spectral solarirradiance (8); (E) tropical Atlantic SST, relative SST proxies (calculated as the residual between tropical Atlantic SST anomalies and averaged Indian, west, and eastPacific Oceans SST anomalies), and storm proxies from Cariaco Basin foraminifera (34), tropical Atlantic corals (35), a Jamaica lake sediment geochemical record(22), mean SST anomalies over the Main Development Region (36), and tropical west Atlantic, west and east Pacific, and Indian Oceans corals (37). Gray-shaded areaindicates the MM (1645–1715 CE). Smoothing techniques in E include loess (34) and 30-y spline (36, 37). The extended hurricane activity index in E is calculated overthe region 15–25°N and 70–80°W (22).

3172 | www.pnas.org/cgi/doi/10.1073/pnas.1519566113 Trouet et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

solar–TC relationship by ocean heat content can also be applied todecadal temporal scales rather than spatial scales: tropical westernAtlantic SST reconstructions suggest that MM Caribbean SSTswere up to 1 °C cooler than present (34, 37; Fig. 3E) and as suchrepresent a potential MM shift in the ocean heat budget and thesolar–TC relationship.Our findings suggest an abrupt CTC suppression in response

to a severe reduction in solar irradiance and thus provide abenchmark for modeling studies of North Atlantic TC responseto radiative forcing. The unequivocal strength and sign of theCTC response to reduced solar activity imply modulation bysynchronized oceanic and atmospheric circulation patterns. Tofurther improve our understanding and to constrain estimatesof future North Atlantic TC activity, additional high-resolutionproxy records and targeted climate modeling experiments areneeded that can support a process-based investigation of theresponses of North Atlantic TC activity, SSTs, ENSO, and NAOto radiative forcing and future climate change.

DataTCship. We compiled a total of 656 Spanish shipwreck events(1495–1825 CE) that occurred in the Caribbean realm and wereincluded in Marx (16). This data set is not complete, as someshipwrecks were never documented and some documents likelylost due to deterioration with age, war, or fire (15), but it is acomprehensive collection of shipwrecks in the Americas thatincludes ca. 4,000 entries cataloged by year and location. Ships ofBritish, French, Dutch, and other nationalities also crossed theAtlantic and wrecked in the Caribbean from the seventeenthcentury onward, but to avoid trends in TCship due to increasingnumbers of ships, we compiled only shipwrecks of Spanish origin.We compiled all Spanish shipwreck events that were recorded tohave occurred (i) due to storm activity (66%) or for unspecifiedcauses (34%), (ii) during the hurricane season (July–November)or with unknown seasonality, and (c) in Florida (21%), on theAtlantic Coast of Mexico (25%), in Hispaniola (22%), Cuba(13%), the Lesser Antilles (5%), and Puerto Rico, Texas, SouthCarolina, Louisiana, Mississippi, Jamaica, the Cayman Islands,Bermuda, and the Bahamas (less than 5% each; Fig. 1A).Quantities were recorded as “many,” “several,” or in similarterms on six occasions in the record and we counted these as fiveshipwrecks, which is the median number of ships reported tohave wrecked during storms when more than one ship wrecked(range: 2–29 shipwrecks; n = 39). The TCship time series thusconsists of the number of Spanish ships that wrecked per year inthe Caribbean region due to storm activity or for unspecifiedcauses (Fig. 2).We calculated decadal sums based on TCship (Fig. 3A) and

verified that decadal-scale fluctuations were not a function offluctuations in the number of active transatlantic Spanish shipsby calculating decadal-scale percentages of shipwrecks per numberof active ships in the Caribbean (TCship%; Fig. 3B). For this pur-pose, we compiled a time series of the total number of shipsdeparting from Seville to the Caribbean and Gulf of Mexico duringthe period 1511–1784 CE based on the General Archive of theIndies (AGI) (pares.mcu.es). Within the AGI, we tallied each shipdeparture among the 643 registers in the Registros de Ida. Becauseof discrepancies in the AGI related to low ship log entries duringearly and late decades, we calculated annual ship occurrences forthe period 1590–1780 CE only (n = 3,881; Fig. S5). The data setwas quality checked for potential duplicate records by consideringthe ship name, type, and captain name, and duplicate records wereexcluded from the final time series. The record of annual shipcounts was binned into decadal sums (Fig. S5) and used to calcu-late TCship% (Fig. 3B).We further compared the decadal-scale TCship time series to a

shipwreck time series (1515–1815 CE; number of shipwrecks =232) for Florida based on the data compiled in ref. 42. This

decadal-scale Florida time series correlated strongly (r = 0.77,n = 31, P < 0.001) with the Florida component of the TCship timeseries (Fig. S4).

TCsupp. We aimed our sampling strategy at using radial growthanomalies from Pinus elliottii var. densa (hereafter slash pine)trees to detect TCs in the Florida Keys/Caribbean region. Forthis purpose, we collected partial (live trees) or full (dead trees)cross-sections from 38 mature old-growth pines that showed nosigns of visible injury to the tree (i.e., fire scar, beetle gallery,lightning scar) and were located at the highest elevations (≥ 2 ma.s.l.) within the island interior of Big Pine Key (BPK; Fig. 1A).Trees located at higher BPK elevations are able to survivemultiple storm surge events because salt water has a short resi-dence time at the surface (43). Moreover, BPK slash pine treesexhibit a different growth form compared with individuals onmainland Florida and windthrow-induced mortality associatedwith TCs is uncommon. We used standard dendrochronologicaltechniques (44) for sample preparation, measurement, and cross-dating of the tree-ring data.Slash pine ring-width data can contain a weak late summer

(September) moisture availability signal (45), but this was not thecase at the BPK site (Fig. S6) and we focused our analysis ontree-growth suppressions (Fig. 1B) in the slash pine data set. Weexpected a TC signal due to loss of tree biomass (leaves,branches) to be manifested within the tree-ring record as a pe-riod of tree-growth suppression (46) and therefore definedsuppression events as periods in which raw ring width was ≥25%below the 10-y running mean of all growth rings (n = 1,228 for 38tree samples over the period 1707–2009 CE). We than calculatedan annual time series TCsupp as the percentage of samples in a givenyear t that contained a suppression. Widespread tree growth sup-pressions (defined as suppression in ≥75% of samples) occurredduring wet [positive Palmer Drought Severity Index (PDSI)] sum-mers, most likely caused by TC-induced rainfall (Fig. 1C).

MethodsThe relationship between TCsupp and instrumental and reconstructed TCevents was evaluated in a contingency analysis and in an SEA. We comparedthe frequency distribution of growth suppressions in TCsupp to storms in theHURDAT (17) and shipwrecks in the TCship time series in a contingencyanalysis and calculated statistical significance levels using a Pearson’s χ2 (χ2)test. The HURDAT time series (1851–2010 CE) consisted of the number ofcategory 1–5 storms that tracked within 160 km of BPK per year (Fig. 2A).Large storms within this radius are documented to cause major storm surgeson BPK (47). For instance, Hurricane Wilma passed at 145 km from BPK andcaused growth suppression in 75% of the TCsupp samples in 2005 (Table S3).

We compared lagged relationships in an SEA by superposing a windowof contemporaneous and lagged annual growth suppressions over eachHURDAT (1851–2010 CE) or TCship-based (1707–1825 CE) TC event year (48).Event years were defined as years in which category 1–5 TCs tracked within160 km of BPK (n = 34) for HURDAT and as years with five or more shipwrecksevents (n = 9; Fig. 2B) for TCship. Results are consistent when using smaller (100km) and larger (300 km) tracking radii for HURDAT TCs (Fig. S1) or shorter butbetter replicated TCsupp time series (Fig. S2). In addition to this, we used acombined event list of TCship, documented TCs/storms from ref. 14 (n = 2; 1826–1849 CE), and HURDAT (Fig. 2C). Monte Carlo simulations were used (n =1,000) to develop bootstrapped confidence intervals to determine whether asignificantly higher than average number of trees showed growth suppressionin a 7-y window (t–1 through t+5) centered around the TC event year (t0).

ACKNOWLEDGMENTS. We thank Justin L. Hart and Mary Glueck forassistance with data collection and development. We are grateful toP. J. Baker, L. Bakkensen, H. Diaz, and J. Esper for providing suggestions. G.L.H.was supported by a grant from the Office of Research from the Universityof Southern Mississippi, by the National Science Foundation under Grants1002479 and 0538420, and by the United States Fish and Wildlife Service.M.D.-D. was supported by a University of Arizona Agnese N. Haury VisitingFellowship and by a Marie Sklodowska Curie Innovative Training NetworksFellowship (ForSEAdiscovery project, PITN-2013-GA-607545). The idea forthis collaborative project originated during the Second AmericanDendrochronology Conference.

Trouet et al. PNAS | March 22, 2016 | vol. 113 | no. 12 | 3173

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

1. Pielke RA, et al. (2008) Normalized hurricane damage in the United States: 1900-2005.Nat Hazards Rev 9:29–42.

2. Mendelsohn R, Emanuel K, Chonabayashi S, Bakkensen L (2012) The impact of climatechange on global tropical cyclone damage. Nat Clim Chang 2(3):205–209.

3. Knutson TR, et al. (2010) Tropical cyclones and climate change. Nat Geosci 3(3):157–163.

4. Villarini G, Vecchi GA (2012) Twenty-first-century projections of North Atlantic trop-ical storms from CMIP5 models. Nat Clim Chang 2(8):604–607.

5. Landsea CW, Harper BA, Hoarau K, Knaff JA (2006) Climate change. Can we detecttrends in extreme tropical cyclones? Science 313(5786):452–454.

6. Mann ME, Woodruff JD, Donnelly JP, Zhang Z (2009) Atlantic hurricanes and climateover the past 1,500 years. Nature 460(7257):880–883.

7. Shindell DT, Schmidt GA, Mann ME, Rind D, Waple A (2001) Solar forcing of regionalclimate change during the Maunder Minimum. Science 294(5549):2149–2152.

8. Lean J (2000) Evolution of the sun’s spectral irradiance since the Maunder Minimum.Geophys Res Lett 27(16):2425–2428.

9. Donnelly JP, et al. (2015) Climate forcing of unprecedented intense‐hurricane activityin the last 2000 years. Earths Futur 3(2):49–65.

10. Denommee KC, Bentley SJ, Droxler AW (2014) Climatic controls on hurricane patterns:A 1200-y near-annual record from Lighthouse Reef, Belize. Sci Rep 4:3876.

11. Lane P, Donnelly JP, Woodruff JD, Hawkes AD (2011) A decadally-resolved paleo-hurricane record archived in the late Holocene sediments of a Florida sinkhole. MarGeol 287(1-4):14–30.

12. Frappier AB, et al. (2014) Two millennia of tropical cyclone-induced mud layers in anorthern Yucatan stalagmite: Multiple overlapping climatic hazards during the MayaTerminal Classic “megadroughts”. Geophys Res Lett 41(14):5148–5157.

13. Chenoweth M, Divine D (2008) A document-based 318-year record of tropical cy-clones in the Lesser Antilles, 1690-2007. Geochem Geophys Geosyst 9(8):Q08013.

14. Garcia-Herrera R, Gimeno L, Ribera P, Hernandez E (2005) New records of Atlantichurricanes from Spanish documentary sources. J Geophys Res 110(D3):D03109.

15. Rappaport EN, Fernandez-Partagas JJ (1997) History of the deadliest Atlantic tropicalcyclones since the discovery of the New World. Hurricanes: Climate and SocioeconomicImpacts, eds Diaz HF, Pulwarty RS (Springer-Verlag, Berlin), pp 93–108.

16. Marx RF (1983) Shipwrecks in the Americas (Crown Publishers, New York), p 482.17. Landsea CW, Franklin JL (2013) Atlantic hurricane database uncertainty and pre-

sentation of a new database format. Mon Weather Rev 141(10):3576–3592.18. Kozar ME, Mann ME, Emanuel KA, Evans JL (2013) Long-term variations of North

Atlantic tropical cyclone activity downscaled from a coupled model simulation of thelast millennium. J Geophys Res Atmos 118(24):13383–13392.

19. Reed AJ, Mann ME, Emanuel KA, Titley DW (2015) An analysis of long-term rela-tionships among count statistics and metrics of synthetic tropical cyclones downscaledfrom CMIP5 models. J Geophys Res Atmos 120(15):7506–7519.

20. Harley GL, Grissino-Mayer HD, Franklin JA, Anderson C, Kose N (2012) Cambial activityof Pinus elliottii var. densa reveals influence of seasonal insolation on growth dy-namics in the Florida Keys. Trees Struct Funct 26(5):1449–1459.

21. Crowley TJ (2000) Causes of climate change over the past 1000 years. Science289(5477):270–277.

22. Burn MJ, Palmer SE (2015) Atlantic hurricane activity during the last millennium. SciRep 5:12838.

23. Brandon CM, Woodruff JD, Lane DP, Donnelly JP (2013) Tropical cyclone wind speedconstraints from resultant storm surge deposition: A 2500 year reconstruction ofhurricane activity from St. Marks, FL. Geochem Geophys 14(8):2993–3008.

24. Ercolani C, Muller J, Collins J, Savarese M, Squiccimara L (2015) Intense southwestFlorida hurricane landfalls over the past 1000 years. Quat Sci Rev 126:17–25.

25. Donnelly JP, Woodruff JD (2007) Intense hurricane activity over the past 5,000 yearscontrolled by El Niño and the West African monsoon. Nature 447(7143):465–468.

26. Lund DC, Lynch-Stieglitz J, Curry WB (2006) Gulf Stream density structure andtransport during the past millennium. Nature 444(7119):601–604.

27. Burn MJ, Palmer SE (2013) Solar forcing of Caribbean drought events during the lastmillennium. J Quat Sci 29(8):827–836.

28. Lane CS, Horn SP, Orvis KH, Thomason JM (2011) Oxygen isotope evidence of Little IceAge aridity on the Caribbean slope of the Cordillera Central, Dominican Republic.Quat Res 75(3):461–470.

29. Sabbatelli TA, Mann ME (2007) The influence of climate state variables on AtlanticTropical Cyclone occurrence rates. J Geophys Res Atmos 112(D17), 10.1029/2007jd008385.

30. Kozar ME, Mann ME, Camargo SJ, Kossin JP, Evans JL (2012) Stratified statisticalmodels of North Atlantic basin-wide and regional tropical cyclone counts. J GeophysRes Atmos 117:D18103.

31. Goldenberg SB, Landsea CW, Mestas-Nunez AM, Gray WM (2001) The recent increasein Atlantic hurricane activity: Causes and implications. Science 293(5529):474–479.

32. Webster PJ, Holland GJ, Curry JA, Chang HR (2005) Changes in tropical cyclonenumber, duration, and intensity in a warming environment. Science 309(5742):1844–1846.

33. Vecchi GA, Swanson KL, Soden BJ (2008) Climate change. Whither hurricane activity?Science 322(5902):687–689.

34. Black DE, et al. (2007) An 8-century tropical Atlantic SST record from the CariacoBasin: baseline variability, twentieth-century warming, and Atlantic hurricane fre-quency. Paleoceanography 22(4):PA4204.

35. Kilbourne KH, Alexander MA, Nye JA (2014) A low latitude paleoclimate perspectiveon Atlantic multidecadal variability. J Mar Syst 133:4–13.

36. Mann ME, et al. (2009) Global signatures and dynamical origins of the Little Ice Ageand Medieval Climate Anomaly. Science 326(5957):1256–1260.

37. Tierney JE, et al. (2015) Tropical sea surface temperatures for the past four centuriesreconstructed from coral archives. Paleoceanography 30(3):226–252.

38. Trouet V, et al. (2009) Persistent positive North Atlantic oscillation mode dominatedthe Medieval Climate Anomaly. Science 324(5923):78–80.

39. Haug GH, Hughen KA, Sigman DM, Peterson LC, Röhl U (2001) Southward migrationof the intertropical convergence zone through the Holocene. Science 293(5533):1304–1308.

40. Hodges RE, Elsner JB (2012) The spatial pattern of the sun-hurricane connection acrossthe North Atlantic. ISRN Meteorol 2012:1–9.

41. Hodges RE, Elsner JB (2011) Evidence linking solar variability with US hurricanes. Int JClimatol 31(13):1897–1907.

42. Singer SD (1992) Shipwrecks of Florida (Pineapple Press, Sarasota, FL).43. Ross MS, Obrien JJ, Sternberg LDL (1994) Sea-level rise and the reduction in pine

forests in the Florida Keys. Ecol Appl 4(1):144–156.44. Stokes MA, Smiley TL (1968) An Introduction to Tree-Ring Dating (Univ of Chicago

Press, Chicago).45. Harley GL, Grissino-Mayer HD, Horn SP (2011) The dendrochronology of Pinus elliottii

in the lower Florida Keys: Chronology development and climate response. Tree RingRes 67(1):39–50.

46. Hart JL, Grissino-Mayer HD (2008) Vegetation patterns and dendroecology of a mixedhardwood forest on the Cumberland Plateau: Implications for stand development.For Ecol Manage 255(5-6):1960–1975.

47. Saha AK, et al. (2011) Sea level rise and South Florida coastal forests. Clim Change107(1-2):81–108.

48. Haurwitz MW, Brier GW (1981) A critique of the Superposed Epoch Analysis method -Its application to solar weather relations. Mon Weather Rev 109(10):2074–2079.

3174 | www.pnas.org/cgi/doi/10.1073/pnas.1519566113 Trouet et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020