JOURNAL OF GEOPHYSICAL RESEARCH,VOL 104,NO. Cl, PAGES 1431-1453,JANUARY 15, 1999

The connectivity of eddy variability in the Caribbean Sea, the Gulf of Mexico, and the SylviaJ. Murphy and Harley E. Hurlburt Naval Research Laboratory, Stennis Space Center, Mississippi

James J. O'Brien Centerfor OceanAtmospheric Pr~diction Studies, Florida StateUniversity, Tallahassee

Abstract. A set of numericalsimulations is usedto investigatethe connectivityof mesoscalevariability in the Atlantic Ocean,the Caribbean,and the Gulf of Mexico. The primitive equationmodels used for thesesimulations have a free surfaceand realistic coastlinegeometry including a detailedrepresentation of the LesserAntilles island arc. Two simulationshave 1/4° resolution and includea 5.5-layerreduced gravity and a 6-layer model with realisticbottom topography.Both are wind forced and include the global thermohalinecirculation. The third simulationis from a 1(2°linear wind-drivenmodel. In the two nonlinearnumerical simulations, potential vorticity from decayingrings shedby the North retroflection can be advectedthrough the LesserAntilles. This potentialvorticity actsas a finite amplitudeperturbation for mixedbarotropic and internal modebaroclinic instabilities, which amplify mesoscalefeatures in the Caribbean.The eddiesassociated with the CaribbeanCurrent are primarily anticyclonicand transit a narrowcorridor acrossthe Caribbeanbasin along an axis at 14°to 15°Nwith an average speedof 0.15m/s. It takesthem an averageof 10 monthsto transit from the Lesser Antilles to the YucatanChannel. Along the way,many of the eddiesintensify greatly.The amountof intensificationdepends substantially on the strengthof the CaribbeanCurrent and is greatestduring a multiyear period when the current is anomalouslystrong owing to interannualvariation in the wind forcing. SomeCaribbean eddies squeeze through the YucatanChannel into the Gulf of Mexico,where they can influencethe timing of eddy-sheddingevents. There is a significantcorrelation of 0.45between the Loop Current eddyshedding and the eddiesnear the LesserAntilles with a time lag of 11 months.However, Canobean eddies show no statisticallysignificant net influenceon the meaneddy-shedding period nor on the sizeand strengthof shededdies in the Gulf of Mexico.Additionally, no significantcorrelation is found betweeneddy shedding in the Gulf of Mexicoand transpott variationsin the Florida Straits,although transport fluctuationsin the Florida Straits at 27°Nand the YucatanChannel and showeda correlationof about 0.7 with a lag of 15 days.The linear solution exhibiteda multiyear anomalyin the strengthof the Canobeancirculation that wasconcentrated in the central and easternCaribbean due to a multiyearanomaly in the wind field over the basin. In the nonlinearsimulation this anomalyextended into the westernCanobean and acrossthe entire Gulf of Mexico.This westwardextension resulted from the nonlinearity and instabilityof the CaribbeanCurrent, the westwardpropagation of the eddies,and the passageof Canobeaneddies through the YucatanChannel into the Gulf of Mexico.

1. Introduction of inflow through the various Caribbean passages.The Carib- bean Current exits the basin to the northwest through the The Caribbean is a semienclosed sea bounded on the east Yucatan Channel and flows into the Gulf of Mexico to form and the north by a chain of closely spaced islands that act as a the Loop Current (Figure 3). Both the Caribbean Current and sieve for inflow from the Atlantic Ocean. The islands from the Loop Current are essential components of the Guadaloupe south to Grenada (Figure 1) comprise the Lesser system,which is partly wind driven but which is also augmented Antilles. while the larger islands to the north (Cuba, Hispani- by a contnoution from the global ola, and Puerto Rico) are referred to as the Greater Antilles. [Schmitz and Richardson, 1991; Schmitz, 1995]. Westward flow between the islands of the Lesser Antilles en- Many aspectsof Canobean circulation, including mesoscale ters the Caribbean and forms th~ westward flowing Caribbean variability, are summarized by Kinderet al. [1985). Maul [1993) Current (Figure 2), the principal current within the region. A also presents an overview including the environmental and great deal of uncertainty exists, however, about the distnoution socioeconomic implications of global climate change in the Copyright1999 by the AmericanG.:ophysical Union. Caribbean. In this paper we investigate the connectivity of eddy vari- Paper number 1998JC9(XX)10. 0 148-{)227/99/1998Jcxxx) 10$09.(X) ability in the Caribbean Sea, the Gulf of Mexico, and the 1431 1432 MURPHY ET AL: CONNECrlVITY IN INTRA-AMERICA SEAS , MURPHY ET AI..: CONNECfIVITY IN INTRA-AMERICA SEAS 1433

1966;Kinder, 1983;Lemming, 1971; Molinari et ai., 1981;Nys- tuen andAndrade, 1993},but these studies have been limited in spatial and temporal resolution and coverage.The utilization of a global numerical model aUowsfor research at time and space scalescurrently unfeasible observationaIly. Three versions of the Naval Research Laboratory (NRL) Layered Ocean Model (NLOM) [WaI1craft,1991} are used in this study. All are global in domain and contain detailed Ca- ribbean geometry. The simulations include a reduced gravity, thermodynamic version (simulation 1, Table 1) with 1/40reso- lution and a finite depth, hydrodynamic version (simulation 2, Table 1) also with 1/40 resolution. Finally, a linear, reduced gravity version (simulation 3, Table 1) with 1/20resolution is utilized. After spin-up to statistical equilibriUin, all of the sim- ulations were forced by daily averaged winds from 1981 to 1994, so that interannual variability is also present. The for- mulation for the NLOM is presented in section 2 with detailed equations for the thermodynamic version. The modifications necessaryto create a hydrodynamic version are also presented 64W 60W 56W as well as a more detailed discussionof the design of the three simulations specifically used in this study. These simulations contain different physics,which are used to include or exclude dynamical processesthat may be relevant to the simulation of mean currents and variability in the Ca- ribbean. For instance, simulation 2 contains realistic bottom topography. Simulation 1 is reduced gravity and therefore ex- cludes the barotropic mode and the bottom topography. Since eddies are observed in both simulations, these differences pro- vide a means to investigate (1) the influence bottom topogra- phy and baroclinic instability involving the barotropic mode on the simulated eddies in the Caribbean and (2) topographic effects on eddy formation and pathways of mean currents. Simulation 1 is also thermodynamic and therefore contains more accurate stratification that can influence eddy generation and propagation speeds.This simulation also has a more ac- curate representation of the return flow of the global thermo- haline circulation, which affects the mean transports through the Lesser Antilles. The linear simulation excludesflow insta- bility dynamics, providing a linear, deterministic ocean model 50 66 82 98 114 130 146 responseto the atmospheric forcing. It is used to demonstrate inm the impact of nonlinearity on a multiyear anomaly seen in the interannual simulations and to isolate the causeof this anomaly. 0.28..vs This investigation can be viewed as an attempt to use the Plate 1. Snapshotsof upper layer thickness (meters) and ve- model simulations to extend our understanding of a region locity vectors (meters per second) from simulation 1 for (a) with limited observations after comparing the simulations with September 2, 1992, where a ring from the North Brazil ret- the observational data currently available. However, additional roflection is impacting the Lesser Antilles, and (b) September observations are required to adequately assessthe model re- 8, 1992, where potential vorticity from the ring is advected through the Lesser Antilles. sults. Hence some of the results must be viewed as model predictions subject to future verification. The connectivity of eddy variability in the Caribbean, Gulf of Mexico, and Atlantic Ocean is discussedin section 3. Topics include (1) the influence of the Atlantic Ocean on the forma- Atlantic Ocean. This includes (1) the influence of eddies that tion of Caribbean eddies; (2) a description of the eddies, their impinge on the eastern side of the Lesser Antilles, principally rings shed from the retroflection [Johns mean pathways, and velocities; (3) a comparison between Ca- et al., 1990;FraJan/oni et al., 1995]; (2) the role of mesoscale ribbean eddies observed by Geosat altimetry and model ed- flow instabilities in the Caribbean Current; and (3) the influ- dies; -(4) the relationship between Caribbean eddies and the ence of the Caribbean eddies on eddy shedding by the Loop eddy shedding from the Loop Current in the Gulf of Mexico; Current in the Gulf of Mexico, including eddies from the (5) discussionof an intradecadal cycle simulated in Caribbean easternmost Caribbean. variability and its potential origins; and, finally (6) a discussion Caribbean eddies have been discussedin the literature [Fu of variability connectivity within the Gulf Stream system.Sec- and Holt, 1983; Hebum et al., 1982; Ingham and Mahnken, tion 4 contains the summary and conclusions. , 1434 MURPHY ET AL.: CONNEcrrvITY IN INtRA-AMERICA SEAS

0.75m/s 50 66 82 98 114 130 146 ~ inm Plate 2. Snapshotsof upper layer thickness (meters) and velocity vectors (meters per second) from simu- lation 1 on (a) August 18, 1991,when a North Brazil Current (NBC) ring impacts the Lesser Antilles, and (b) September 4,1991, when fully formed eddy A is observed.The eddy labeled A is tracked through subsequent plots.

2. Ocean Model and Numerical Experiments '" " ~ The NRL Layered Ocean Model [WaLLcraft,1991] is used for ~~~~ { G~V{/,/ - HJ + h/VGk/] the numerical simulations in this study. It is based on the "'1 semi-implicit, free surface model of HurlbLtrt and Thompson [1980]. The model is formulated using an Arakawa C grid [Mesingel.and Arakawa, 1976], an explicit numerical scheme for the reduced gravity simulations, and a semi-implicit scheme for the finite depth simulations. The equations for the n layer + max (0, -(J)A-I)VA-I - [max (0, (J)t-J finite depth,thermodynamic model are, for layersk = 1,"', n, +max (0. -(&I.)]Yt + max (0, (&I.)YA'" avt +(v'Vt+Vt.V}Vt+ fc XfVk at + max (0., -CMt&lA-J(Vk-1- v.) MURPHY f:r AL.: OONNECI1VITY IN INTR..\-AMERICA SEAS 1435

.., . 40.

30° ~LORIDA f/aULFSTREAM

~~ ~! '\~ "- ANTILL~S \LOOP -:::- ~~ "-. "" CURRENT '" '"'" "'" . "~0". ~~~CUI" ~ '.., 20. PUB.TO RICO. :~~:):)~ QI." --- N I \. ~ VI.CIII'. -- \.. HII'AHIOLA IILAIIDI f' YUCATAN \.. CARIBBEAN MA.T1HIQUB\ ""'- - -- 8T. LUCIA: . -/<' ~ --:;:o~"OQU;;- . CURRENT Of'

.,- U-~~VENEZUELA ~" ~,~i' .~""".. 10' , "" ~---i .-" ~ , ,/' ~ GUYANA ~ ~URRENT O' AMAZON.IV~R ~ r--v,,-- 5"8 90° 80° 700W 60. 50° .0. 100. Figure 2. SchematicCaribbetn crculation drawnfrom pilot charts[from Duncanet al., 1982].

f Coriolis parameter; g acceleration due to gravity; Gkl=g, 1 ~ k; Gkl=g - g(Pk - P/)/PO'I < k; hk kth layer thickness; apt DIU (0, (lJJ h; kth layer thickness at which entrainment starts; M + Vt. V Pt = (Pt+1- Pk) ht h; kth layer thickness at which detrainment starts; Ho= 100 m and the reference layer thickness that fOnDSpart of a density relaxation e-folding timescalecoefficient that equals1/0- p when Ho = h,,; Hk kth layer thickn~ at rest; H,,=D(~, 6) - ~i--:HI; i, j, { unit vectors positive eastward, northward, and where upward, respectively; 1 a 1 a KH coefficient of horizontal density diffusivity; v =ia cos9a";j;+J;a8; Q sutface heat flUX; t time; a radius of the Earth (6371 kn1); AH coefficient of horizontal eddy ~i~; 11.. kth layer velocity; Cb coefficient of bottom friction; Vk hk11k; Cf coefficient of interfacial friction; 6 latitude; C M coefficient of additional interfacial friodion 41 longitude; associatedwith entrainment; Pk kth layer den~ty; jJ( 41, 9) kth layer density climatology; D( 1/>. 9) ocean depth at rest; 1436 MURPHY ET AL: CONNEcnVITY IN INTRA-AMERICA SEAS

Figure 3. Topographyof the 2ZOCisothenital surface, August 4-18. 1%6 (Alaminoscruise 66-A-II) [from Uipper, 1970).

Po constant reference density; (IIk=O. k = O. n; 0'p reference coefficient of density climatology (Ilk = (II; - (Ilk - °k("k. k - 1 ".n - 1; relaxation; (II; =&lk [max (0. h; - ht;)/h;j2; T., wind stress; &I; =&It;[mu (0. hk - h;)/h;)2; 't.=T.,. k = 0; I"k = (II; f f «(II; - (11;:)/Jf Ok; T.=C.Polv. - Vk+ll(p. - P.+I)' k = I"'n - 1; &lk kth intetface referencediapyaJal mixing velocity-; Tk=CbPol.~lp~.k = II;

Table 1 World Ocean Simulations

Simulation

2 3

.4., mz/s 300 300 500 Stratificationu, kg/m3 varies in time and space 25.24,26.47,26.99,27.23,27.39,27.77 25.45.27.55 Layers 5.5 6 1.5 Type nonlinear, reduced gravity Thermoflalioe cilQllation, a Sv nonlinear,realistM: bottom linear.reduccdgravity 14 8 '" not present H. 80, 1ro. 100, 250, 350, ~ 155, 185, 210. 225, 225, variable 250.'"' -- A is a)efficient of horizontal eddy vjII;X)Sity;H k is kth ~ thaness at rest. . "Nonh Atlantic outflow pons used have a net venial profile of I, 1,21,2.5, and 1.4 Sv (for k ~ I, ..., 5) MURPHY ET AL: CONNEcrIVrrY IN INTRA-AMERICA SEAS 1437

Table 2. Additional Model Parameters Simulation

Definition )

Ch bottom drag coefficient 0 2x 10-3 0 Ck kth interfacial stress coefficient 0 0 0 g, m/s~ acceleration due to gravity 9.8 9.8 9.8 h;, m thickness of layer k at which entrainment starts 50 (k = I), 40 (k = 2 f) 50 (k = 1), 40 (k = 2 .51 N/A 48, deg latitudinal grid resolution 0.25 0.25 0.50 4<1>,deg longitudinal grid resolution 0.3515625 0.3515625 0.703125 idA:'cm/s kth interface vertical reference mixing velocity 0.001 for all layers 0.001for all layers N/A

N/A is not applicable.

.(}k( 4J, 8) kth interface weighting factor for global A modified version of the 1/2° ETOPO5 bottom topography diapycnal mixing designed to conserve volume [National Oceanic and Atmospheric Administration (NOAA), within a layer in compensation for explicit 1986] was used in the finite depth simulation that included diapycnal mixing due to hk -< h: (e.g., Cd: - co;) realistic bottom topography. The topography data set was first and net transport through the lateral boundaries interpolated to the model grid and then smoothed with a of layer k. Qine-point real smoother. This smoothing is to reduce energy generation on small scales that are poorly resolved by the A hydrodynamic version of this model is obtained by setting p = constant within a layer. This includes neglecting the den- model. The model includes realistic coastline geometry, based on sity equation and retaining only the hydrodynamic part of the EfOPO5, which has been extensively modified in the Carib- pressure gradient in the momentum equation: bean region by Youtsey[1993]. In most locations this geometry is determined by the 200-m depth contour of the topography, which is the minimum depth in the model and which represents the nominal shelf break. The model domain is global. exclud- ing the Arctic, and extends from 72OSto 71°N. The horizontal In reduced gravity simulations with n active layers. the equa- resolution is 0.25° in latitude and 45/1ZSOin longitude for each tions are identical to the (n + 1) layer finite depth equations. variable for the two main simulations and is 0.500by 45/64°for except that the linear 1.5-layer reduced gravity simulation. H n = constant In the finite depth model simulation the amplitude of the Gk/=g(Pn+l - Pk)/PO 1:S k; topography above the maximum depth of 6500 m was multi- Gkl=g(Pn+l - pJ/po 1 > k; plied by 0.78 to confine it to the lowest layer. This removes the 1'k=1'w. k = 0; numerical difficulty that arises when moving interfaces inter-. 1'k=CkPolvk- Vk+ll(Vk - Vk+l)' k s t...n; sect sloping topography [Hurlburt and Thompson, 1980]. Flow C1Jk=O k = 0; through straits with a shallow sill is constrained to small values C1Jk=C1J;- C1Jk - °kWk. k = 1 "'11. below the sill depth.

0.50 MIS .

9OW 8crN 7OW &OW , Figure 4. Mean currentsover 1981-1991for layer 1 of the nonlinearreduced gravity simulation. 1438 MURPHY ET AL: CONNEcnvrrv IN INTRA-AMERICA SEAS

Volume is conserved within each layer by requiring no net ward Deep Western in the abyssallayer and transfer of volume acrosseach layer interface. The net volume by a northward return flow in the upper ocean which contrib- flux acrossinterface k, due to (1) outcropping and/or (2) any utes13 of the 32 Sv (1 Sv = IW m3 S-I) through the Straits of net transport of volume into or out of the layer above via ports Florida at 2~ [Schmitz and Richardson, 1991]. The relation in the lateral model boundaries,is balancedby the domain inte- of the Caribbean Sea to the region of the path of this return gral of cross-interfacialmixing acrosslayer interface k (f f °k"'k)' flow makes the global thermohaline circulation an essential The \veight function for this cross-interfacial moong Ok is element in this study. nonuniform in the NRL global model and has both positive In the model, deepwater formation is parameterized (1) by and negativevalues basedon oxygensaturation, as discussedin using ports along the northern boundary with prescribed out- detail by Shriver and Hurlbwt [1997]. This allows broad, slow flow in the upper ocean and prescribed inflow in the abyssal upwelling over much of the world ocean interior in addition to layer and (2) by downwelling in the Labrador Sea via the regions of downwelling, e.g., the Labrador Sea. The model oxygen-basedcross-interfacial mixing scheme described ear- boundary conditions are kinematic and no slip at solid lier. Both simulations 1 and 2 have an imposed outflow of 8 Sv boundaries. in the top five layers at two ports, one located between Green- The thermal forcing in the thermodynamic version of the land and Scotland, the second in the Davis Strait (Table 1). In model is relaxation to annual mean density values derived from the finite depth simulation (simulation 2), 8 Sv is injected into Levitzls [1982]. Temporal variations in the layer interface the bottom layer at ports locate

146 0.75 mjs 50 66 82 98 114 130 - inm Plate 3. Eddy A on (a) October 6,1991, when it has transited to 66.5OW,and (b) November 26,1991, when it has transited to 71OWand intensified. respectively (9.05 Sv total) [Wilson and Johns, 1997]. The cor- The transport of the has been quite accu- responding simulated total mean transports from simulations 1 rately measured at 2rN [Larsen, 1992] with a mean value of and 2 are 9.9 and 7.7 Sv, respectively. 32.3 Sv. The model transport in this region for simulations 1 Transport through Windward Passagehas been estimated and 2 are 22.7 and 20.7 Sv, respectively. using an inverse model to have a mean inflow value of 7 Sv The preceding comparisons indicate that in the southern [Roemmich, 1981]. A more recent estimate is 4 Sv [Schmitz, Caribbean the simulations agree quite well with the latest 1996]. Both simulations have contradictory flows with mean observational data. Since this is where the Caribbean eddies outflow transports for simulations 1 and 2 of 0.8 and 3 Sv, form and intensify, this is an important conclusion. The dis- crepanciesin the transports of the Florida Current at 2TN, the respectively. The mean transport of 29 Sv through Yucatan Channel has Yucatan Channel, and Windward Passageare caused by a been estimated based upon its presumed contribution to the known error in the topographic configuration of the eastern mean transport of the Florida Current [Schmitz and Richard- coastlineof Florida and the geometryof the Bahamas[Hurl- son, 1991].The mean simulated value of the Yucatan Channel burt and Townsend, 1994] and do not affect the results pre- transport for simulations 1 and 2 is 21.4and 21.5Sv, respectively. sented in this paper regarding the transit of Caribbean eddies 14-40 MURPHY ET AL.: CONNEcrIvrrY IN IN1RA-AMERICA SEAS

0.75m/s 50 66 82 98 114 130 146 inm Plate 4. Eddy A on (a) January 4, 1992, when it is advected northwestward by the Can'bbean Current to 75OW,and (b) February 16, 1992,when it is advected southwestward to 79"W.

into the Gulf of Mexico. This phenomenon has been simulated energy; and stronger Caribbean eddies. As will be discussed in other versions of the NRL model that possessaccurate later, flow instabilities are required to demonstrate the North transports and topography in the western Caribbean, Bahamas, Brazil Current ring and Caribbean eddy connectivity that is the and Straits of Florida. Several of these versions are configured primary focus of this paper. Additionally, by using a reduced with significantly higher resolution than the ones used in this gravity simulation, we are able to show that bottom topography study, but unfortunately, none of these simulations was forced and baroclinic instability between the barotropic and the in- by daily interannual wind forcing for more than a decade. ternal modes are not essential in our results and, in fact, play All three of the model simulations described above are im- no role. Simulation 1 also has a more accurate representation portant to this study, although many of the results presented in of the return flow of the global thermohaline circulation, which the paper depend primarily upon simulation 1. Simulation 1 is also resultsin higher eddy kinetic energyas well as more accurate more realistic because it is thermodynamic and has more ac- mass transport through the passagesof the Lesser Antilles. curate stratification than simulation 2. Simulation 1 also has a Simulation 3 has a very specific purpose in this paper. It is thinner upper layer; higher current velocities in this layer, used (1) to isolate the cause of an anomaly in the Caribbean which results in greater flow instabilities; higher eddy kinetic Current (discussedlater) becauseany anomalies in this simu- MURPHY ET AL.: CONNECllVnY IN INTRA-AMERICA SEAS 1441

95W 90W 8SW sow 75W 70W 65W &Ow

0.75mls 50 66 82 98 114 130 146 - Inm Plate S. Eddy A on (a) March 3,1992, when it moves northward and starts to merge with a large eddy west of Jamaica, and (b) March 28, 1992, when the merged eddy transits to 81OW. ration are a linear, deterministic response to the atmospheric Lesser Antilles from the Atlantic Ocean. The source of this forcing, and (2) to help demonstrate the impact of nonlinearity potential vorticity can be seen in Plate 1a, where a large anti- and eddies on this anomaly. cyclonic ring shed from the retroflection of the North Brazil Current (NBC) is observed in the model on the eastern side of the Lesser Antilles. Plate 1b demonstrates flow from the NBC 3. Results ring into the Caribbean and the formation of an eddy on the 3.1. NBC Rings and Caribbean Eddy Formation western side of the Lesser Antilles. An analysis of potential The mean Caribbean circulation represented in Figure 2 is vorticity for 1986on either side of St. Vincent passageresulted simulated by the numerical model including the Caribbean in a cross-correlation value of 0.54 at lag 90 daysfor simulation Current, cyclonic circulation in the Colombian Basin, and the 1. This indicates that potential vorticity is not a constant Loop Current in the Gulf of Mexico (Figure 4). through the passage,and at least some of the vorticity from the Mesoscale features in the Caribbean are observed in both ea$tern side is advected to the west. nonlinear model simulations. The formation of these features Since a pioneering investigation into NBC rings by Johns et is due in part to potential vorticity being advected through the aL.[1990], they have been observed in Geosat data [Didden and 1442 MURPHY ETAL: CONNEcnVITY IN INTRA-AMERICA SEAS

0.75m/s 50 66 82 98 114 130 146 inm Plate 6. Eddy A on (a) June 12, 1992,when it is approaching the Yucatan Channel. The previous eddy has transited through the Yucatan Channel and into the Loop Current Eddy. (b) Eddy A on June 27,1992, when material from the leading edge of it passesthrough the Yucatan Channel into the Loop Current.

Schott, 1993], by drifters [Richardson et at., 1994], by current perturbation on which flow instabilities can grow. A beta meter arrays, and in numerical models [Frotantoni et at., 1995]. Ros..,bynumber analysis of both nonlinear simulations was In all of these studies, approximately three NBC rings per year used as one test of the hypothesis that the eddies form as a transit up the coast toward the Lesser Antilles. Frotantoni et at. result of barotropic instability. The beta Rossby number is [1995] reported that both the observed and modeled eddies defined as v/ /3r:!, where v is the maximum swirl velocity of an decayedwithin 120 days of reaching the ridge between Tobago eddy averaged around the eddy, /3is the variation of the Co- and Barbados (essentially the Lesser Antilles). It appears from riolis parameter with latitude, and r is the mean radius of the the model simulation (Plate 1) that as the NBC rings decay, eddy from the center to the maximum swirl velocity. The di- some of their vorticity can be advected into the Caribbean. ameter (2r) and maximum swirl velocities were determined for six representative eddies. The averagediameter for simulation 3.2. Eddy Amplification Within the Caribbean 1 was 330 kIn, and for simulation 2 it was 365 kIn. The beta The initial eddy formation is weak, but the potential vorticity Rossby numbers for these eddies were order 1 for both simu- advected through St. Vincent passage is a finite amplitude lations with an average of 0.47 for simulation 1 and 0.77 for MURPHY ET AL: CONNECrIVrrY IN INTRA-AMERICA SEAS 1443 simulation 2. A beta Rossby number of the order of 10 is of part of the eddy through the Yu~tan Channel. Also, it is indicative of baroclinic instability involving the barotropic ea.~ily followed on a Hopfmuller diagram of eddy tracks mode, while a beta Rossby number of order 1 is indicative of (section 3.5). an equivalent barotropic instability of the first baroclinic mode Many Caribbean eddies only exhtoit some of the features [see McWilliams and Flier/, 1979; Hurlburt and Thompson, seen during the life span of eddy A Compared with other 1982, 1984]. Caribbean eddies, however, eddy A is not one of the strongest, The low beta Rossby number in both simulations 1 and 2, nor is it the most dramatic example of NBC ring intrusion into the similarity of the eddies in the two simulations, and the the Caribbean nor of an eddy passing thr~ugh the Yucatan similarity of the eddy trajectories serve as justification for Channel and influencing eddy shedding in the Gulf of Mexico. studying the Caribbean eddies with a reduced gravity simula- Many of these points are discussedin greater detail throughout tion. Thesesimilarities imply that there is little or no baroclinic the remainder of section 3. instability between the barotropic and first baroclinic modes On average,an eddy takes approximately 10 months to tran- nor influence of the bottom topography. However, a vertical sit the basin. Throughout Plates 2-7 other anticyclonic eddies analysisof layer thicknessanomaly revealed a mixture of baro- can be seenin various stagesof tran$it. This is indicative of the tropic and internal mode baroclinic instability between layers 1 complex and continuous flow of mesoscale variability that and 2. This was indicated by some eddies being barotropic with crossesthe region. depth while others were baroclinic while they strengthened. Additionally, several eddies were observed to change from 3.4. Model Data Comparison barotropic to baroclinic and back again in the course of a life There is a paucity of obsetVatiQnaldata in the Caribbean cycle. Finally, one double-lobe.d eddy was observed with one with which to conduct model-data comparisons. In one case, lobe baroclinicwith depth and the other barotropic. Note that this eddies observed in the reduced gravity simulation can be re- analysiswas performed during periods of eddy intensification. lated to eddiesobserved by Geosat. It should be noted that the 3.3. Lire Cycle or Caribbean Eddies model eddiesare not aused by a deterministic responseto the atmospheric forcing, and therefore a comparison between 0b- The resulting anticyclonic eddies are qbseIVed to transit served and modeled eddies canbe done only in a statistical from the Lesser Antilles along an axis of 14°-16~ until they sense, not as individual eddies. Nystuen and Andrade [1993] are deflected northward by the Nicaraguan coast (plates 2-5). reported two eddy tracks (Figure 5) over a 1-month time span: Here the eddies often merge and intensify until they pass June-July 1987 and April-May 1988. The Geosat eddy A through the Yucatan channel (Plate 6b). (1987) tracks approximately 10 farther south and 20 farther Plates2-7 are a sequenceof model snapshotsthat represent west than the model eddy A (Figure 6a). a case study for a typical anticyclonic eddy track through the For the second Geosat sequence (Figure 6b), satellite eddy Can"bbeanfrom simulation 1. Snapshots are available at the C tracks 10farther north and T farther east than the model frequency of the model output (3.05 days), and the position of eddy C (Figure 6b). Intensity variations are approximately the the eddy has been conclusivelyverified in every snapshot. For same as the previous sequence. the sake of brevity, however, only 12 out of the possible 100 snapshotsare presented. Plate 2a shows an NBC ring impacting the Lesser Antilles 3.5. Caribbean Eddy Trajectories with flow from the ring entering the Caribbean. The eddy that The model-simulated anticyclonic Caribbean eddies transit is beginning to form on the western side of the islands has been westward in a fairly narrow corridor. This corridor was deter- labeled eddy A In Plate 2b a fully formed eddy A is obseIVed. mined by tracking upper layer thickness maxima averaged In Plates 3a and 3b, eddy A is obseIVed to continue its west- within a sequenceof small rectangular boxes along the corri- ward transit and it begins to intensify. This intensification is dor. The size of each box was chosen to best capture the due to flow instabilities of the Caribbean Current. In Plates 4a average pathway of the maxima so that the corridor encom- and 4b the path of eddy A is perturbed first northward and passedmost of the ensemble of eddy tracks. There are a total then southward by the Can"bbeanCurrent. Eddy A begins to of 32 rectangles, each containing a time series of upper layer merge with the large eddy near Jamaica (Plate 5a). In Plate 5b thicknessaveraged over the box (12 years long). The long-term the merged eddy A has transited farther westward. Note also mean (1981-1992) is subtracted Figure 7 depicts these boxes, that in Plate Sb, there is a large eddy west of eddy A. In Plate shaded together to illustrate the corridor. 6a the eddy west of eddy A has squeezedthrough the Yucatan A phase plot of the upper layer thickness anomalies along Channel In this figure, eddy A transits to the Yucatan Chan- the corridor for simulation 1 (Plate 8) clearly traces the Ca- nel In Plate 6b the leading edge of eddy A is squeezing ribbean eddies from the Lesser Antilles (0 kID) through the through the Yucatan Channel, but as we can see from Plate 7a, Yucatan Channel (3000 kin) and into the Loop Current eddy- the entire eddy does not make the transit before the Li>op shedding regime. The propagation speed of the eddies along Current Eddy is shed In Plate 7b the Loop Current Eddy is the track is fairly uniform for the first 2000 kID of the track. At making its westward transit across the Gulf of Mexico. This 2000 kID (approximately SOOW)the track turns northward and eddy is labeled eddy A since part but not all of eddy A is the eddies intensify and accelerate. The average propagation present. This particular eddy track was chosen. because it speed to the Yucatan Channel (3(XX)kID) is 0.15 m/s. The shows a wide variety of typical features associated with the average transit time from the Lesser Antilles to the Yucatan simulated Caribbean eddies, including (1) formation from an Channel is 10 months, with values as short as 7 months and as NBC ring fragment that passesinto the Caribbean through a long as 17 months. The black dots sta$ng in 1991 in Plate 8 narrow passagein the LesserAntilles. (2) intensification due to correspond to the positions of eddy A observed in Plates 2-7, instability of the Caribbean Current as it propagates westward while the black square cofJ"espondsto the Loop Current Eddy through the Caribbean, (3) eddy merging, and (4) passage observed in Plate Th. Plate 8 also demonstrates that only part 1- MURPHY ET AL: CONNEcrlVITY IN INTRA-AMERICA SEAS

.~-

.

. . -

~ N b= '0 "- r= '- ~ ..; ) >- ;..- "3 r-.: - u ~ 1=1 =' v - ~ u C r= 0

\'.. . ;;\'

~ f~~.\ - t . .) coe<' L ~~ 1~ ( °1 ;~ ~'"

§:.; £: ~ ~~ ~.5 ~=~ 0 t~ 8c - ~8 . "'00 . ~- .~ 00 -0- ~- 8 ~ of- 3: 1=1- . ~) ~ '" ~ -0 \ ) X~ ~'O () u,. ~a ~O; .d ~ 01)0 .-M ~ U "'r .d:= .' ~ 1. uyo. b ,./ ) ... .!~ sc..:: ",- 9; ~=

:'!J: =- . 0- :!Jf\ u~ ~,.. C.t: '-' -, o"tJ <~, I Inr= f, ~( f~ =f- .00 )X7' ~~ ~ -°~ G ~ .. I~ \ c--

j\\. / /°--~ _I\j

~ MURPHY ET AL: CONNECTIVITY IN INTRA-AMERICA SEAS 1«5

1

" I ., ,-, ~';', 0 .,.', !,.,"- ~-- " ,~.','...,.; r. r.:. "'- ,'"\ ,;. u(~ 15N ,"" '" '.~ ~.Oi-- " - v' o ~'-~' " \'~ '---0/1{ ~ Q '" "", :1,

, -;.. -~ \ . . .- -' ") ," .\}(\.. :~ 1. .

7OW &OW lOW 7OW 6OW Figure 6. SSH anomalies, corresponding in time with Figure 5, from simulatjon 1. Contour interval is 10 cm, and dashed Jines are negative contours. of eddy A transits the Yucatan Channel. Part of the eddy west so there ~ a false break in the resulting track depicted in Plate 8. of eddy A also transits through the Yucatan channel (Plate 5b). The correspondingplot for sin1ulation2 (Plate 9) is similar, ex- That eddy transits the entire Can"bbeanbut does not stay on cept for the frequency of eddy sheddingby the Loop Current the averageeddy track which is depicted graphically in Figure 7; 3.6. Loop Current Eddy Shedding The observed frequency for separation of the Loop Cunent eddies is of the order of a year. The data sets of ring separa- tions are gappy, the eddy-shedding period varies, and an in- sufficient number of ring separations have been observed.This hinders statistical analysis, and different analysesshow differ- ent spectral peaks. For example, Vukovich [1995] compiled two data sets, one 22-year data set (1972-1993) and one 17-year data set (1977-1993), from various sources.Sturges [1992] also reported a .technique to develop a continuous data set Sturges found a primary spectral peak at 8.5 months and secondary peaks at 6, 13.4, and 25 months. Multiple spectral peaks were also reported by Vukovich [1988],Maul and Vukovich [1993] (6, 10.9, and 17 months), and Vukovich [1995] (9, 14 months). A singlefrequency of 8-9 monthsWdS reported by Maul et al. [1985]. It is unclear whether this variability in frequency reflects the natural variability of ring separation or problems associated with the data sets, such as their length. For example, Vukovich [1995] reported a primary spectral peak at 9 months and a secondary peak at 14 months. The mean period of the Loop Current eddy shedding (11 months) would have a frequency equal to the secondary spectral peak, with only one more data occurrence at 11 months, and it would have a frequency equal to the primary spectralpeak with only two more data occurrences. In the 12.5years of reduced gravity model results, there were 18 Loop Current eddies. This number was determined by av- eraging the reduced gravity upper layer anomalies at 89.SOW from 21.0 to 30.00N (Plate 10). This resulted in a mean period of 8.3 months between eddies. In the finite depth simulation there were 24 eddies in 12 years (Plate 9) or an averageof 6.0 months between eddies. The eddy shedding in simulation 2 may have a weak contribution from baroclinic instability. Hurl- Figure 7. Path representing the average eddy track through bul1 and Thompson [1982] show that such a contribution can the Caribbean. increase the westward propagation speed of the Loop Current 1446 MURPHY ET AL: CONNEcnVlTY IN INTRA-AMERICA SEAS

3.7. Caribbean Eddies and Loop Curl"ent Eddy Shedding

Connections between Loop Current eddy shedding and Ca- ribbean eddies in simulation 1 are evident graphically in Plates 6-8. In addition, a cross correlation between the upper layer thickness time series of increment 27 in the corridor (repre- senting the Loop Current Eddy) and the first increment (rep- resenting the initial input of eddies into the Caribbean) shows a significant value of 0.45 at a lag of 11.4months for simulation 1. Statistical significance was determined using the technique descn"bedin the appendix. A comparison, however, between the five Loop Current ed- dies shed betWeenMarch 1984and November 1987 (the period of high Caribbean eddy activity from Figure 8) and the five eddies shed between December 1987 and March 1991 in sim- ulation 1 (the period of low Caribbean eddy activity from Figure 8) showed no influence of Caribbean eddy activity on the mean eddy-shedding period (8.8 months in both cases). Additionally, there was no significant difference in the average Loop Current eddy propagation speed as determined from Plate 10, nor in eddy diameter measured from the maximum swirl velocity averaged around the eddy. Finally, there was no significant difference in the averagemaximum swirl velocities. This might not be evident from Plate 10, but note that the background sea surface height was higher during the period of high Canbbean eddy activity. Note that significance was deter- Figure 8. Sum of total upper layer thickness anomaly mined by a single classification analysisof variance (ANOV A) (metersX 1Q2)along the path depictedin Figure 7 for 1981- [Sokal and Rohlf. 1981]. Thus we see Caribbean eddies that (1) 1992for(a) simulation 1 and (b) simulation 2. Total track cross the entire Canobean from east to west, (2) squeeze distanceis approximately3800 km. through the Yucatan Channel, (3) influence the strength of the closed anticyclonic circulation inside the Loop Current, and (4) influence the timing of individual eddy-shedding events into an unstable configuration. In their paper the eddy peri- (including a statistically significant lagged correlation with ed- odicity was about 60 days in simulations where baroclinic in- dies in the easternmost Canobean). We do not see, however, a net impact on the mean eddy-sheddingperiod nor on the size stability was dominant. The observed eddy-shedding peri~ [Vukovich, 1995] are much more variable than those exhibited or strength of eddies that are shed, a topic which bears further in the model, although 8.3 months matches the primary spec- investigation with higher-resolution models. tral peaks reported by Maul et al. [1985], Sturges[1992], and Vukovich [1995]. 3.8. Multiyear Anomalies in the Caribbean Historically, ring separation was thought to result from sea- A closer examination of Plates 8 and 9 reveals substantial sonal variations in transport through the Yucatan Channel, interannual variation in the Caribbean eddy formation and which affected the penetration distance of the Loop Current propagation. The sum of the upper layer thickness anomalies into the Gulf of Mexico [Cochrane, 1965]. Using a set of nu- along the corridor (Figure 8) showsthe extent of this variabil- merical simulations, Hurlburt and Thompson [1980, 1982]dem- ity. In addition, large interannual variations in the strength of onstrated that an eddy-shedding period similar to that ob- the Caribbean Current are found from the Lesser Antilles to served was possible with constant inflow transport. They also the Yucatan Channel (Figure 9). The mean pattern of a mul- demonstrated that a 15-layer reduced gravity model was the tiyear anomaly where the Caribbean Current strengthened simplest model that could reproduce the basic eddy-shedding (simulation 1) is depicted in Plate 11a. The variations in the cycle, eliminating baroclinic instability as a determining factor strength of the Caribbean Current and the flow through the in the periodicity. Instead, they found horizontal shear insta- Lesser Antilles affect the strength of the mesoscalevariability bility in the first internal mode. It should be noted that while forming in the Caribbean (Plate 8). The anomaly depicted in horizontal shear instability was used to explain the final eddy Plate 11a is mainly confined to the Caribbean and Gulf of pinch-off in that model, the authors showed that the westward Mexico. The part in the central Caribbean, south of Hispaniola bending of the Loop Current and its tendency to bend back on and Jamaica, is consistent with a strong anomaly in the re- itself could be understood without invoking an instability gional wind stress curl (Figure 10), with a negative anomaly mechanism.Thus the eddy-shedding frequency was not deter- north of 15°N and a positive anomaly south of 15°N.To further mined by a flow instability but, rather, by the time required for investigate the ocean model's response to this anomaly, the the loop to penetrate into the Gulf of Mexico and bend west- nonlinear results from simulation 1 are compared to corre- ward into an unstable configuration. This timescale is propor- sponding results from a linear simulation (simulation 3), which tional to the diameter of shed eddies divided by the nondis- has an entirely deterministic responseto the wind forcing and persive internat Rossby wave speed. The dynamics are which tends to adjust toward Sverdrup flow closed by western discussed in more detail by Hurlburt and Thompson [1982, boundary currents [Sverdrup, 1947;Munk, 1950]. Compared to section 5]. the linear solution anomaly (Plate lIb), the nonlinear anomaly MURPHY m AL.: CONNEcrIVrrY IN INTRA-AMERICA SEAS 1447

0.75 m/s 50 66 82 98 14 130 146 . inm Plate 7. The Loop Current Eddy on (a) July 9,1992, when it begins to separate, and (b) August 27,1992, when the Loop Current has shed an eddy.

is stronger in the western Caribbean and extends across the impact does not affect the mean eddy-shedding period, Loop entire Gulf of Mexico (Plate IIa). The westward extension of Current eddy diameter, or maximum swirl velocity. In addition, the anomaly in the nonlinear simulations is a consequenceof in the south central Caribbean it shows the ability of these the strengthened flow instabilities in the Caribbean Current, nonlinear dynamics to increase the strength of the cyclonic the associatedstrong eddy generation, the westward propaga- gyre in the Colombian Basin south of the Caribbean Current. tion of these eddies, and their ability to squeeze through the Yucatan Channel (Plate 6). Most of these aspects are well 3.9. Connections to Florida Current Variability illustrated in Plate 8. The comparison of linear and nonlinear Finally, several cross correlations were calculated between anomalies in Plate 11 illustrates the large impact that nonlin- the transport of the Florida Current at 27°N (Subtropical At- earity, flow instabilities, and eddies can have on forced anom- lantic Oimate Studies (STACS» and aspectsof the circulation alies with much larger space and timescales than mesoscale in the Caribbean and Gulf of Mexico. These results are report- variability. It also shows the large impact that variability in the edly due to the continued interest in understanding the trans- Caribbean can have on anomalies in the Gulf of Mexico via port variability at 2rN. A cross correlation between transport nonlinear dynamical mechanisms.As noted in section 3.7, this through the Yucatan Channel and transport at 2rN yielded a 25 <0 N 10 co Q'I 0» 0) ..- .- 95 80 It) ~ «J 0) 65 0) 0) ... ~ 50 35

~ ~ 20 m 0) .- .- 5 -10 I"') 0'1 00 CX) m 0'1 -25 .- - -40 -55 N a) ~ 0) -70 -0) , -85

~ ~ -100 CX> ~ m 0'1 -115 ~ .,.,. -130

0 1000 2000 JOOO 0 1000 2000 3000 inm

Plate 9. Sameas Plate 8, but for simulation2. MURPHY BTAL;CQNNBCnVn'Y IN INTRA-AMERICA SEAS 1449

4.5

5.0

4.0 u

1982 1988 1~ 1982 1986 1990 Figure 9. Upper layer transport time series in sverdrups (106m3 S-I) for various combinations of Canobean passagesfrom simulation 1. Values were sampled daily and filtered using a I-year running mean to emphasize interannual variability. (a) Windward plus Mona plus Anegada (northern Caribbean passages),(b) all Lesser Antilles, (c) Guadeloupe plus Dominique plus Martinique (northern passagesof the Lesser Antilles), (d) Grenada plus St. Vincent plus St. Lucia (southern passagesof the Lesser Antilles), (e) Yucatan Channel, and (f) Florida Current at 2~N.

value of 0.64 at a lag of 15 days for simulation 1 and 0.77 at a lag of 15 days for simulation 2. No significant correlation was found between the Loop Current eddy shedding and transport at 2JON. 4. Summary and Conclusions The connectivity of mesoscale variability in the Atlantic Ocean, Caribbean, and Gulf of Mexico has been studied using three simulations by the Naval Research Laboratory's global ocean model. Two simulations have a horizontal resolution of 1/40for each variable and include a S.S-layerreduced gravity and 6-layer finite depth with realistic bottom topography. The third simulation is a linear 1.S-layerreduced gravity, which has horizontal grid resolution of 1/2°. All were forced by daily ECMWF hybrid winds over 1981-1994.Mesoscale features are observed in both of the nonlinear simulations. These features dominate the geostrophically balanced surface layer variability of the Caribbean circulation (Plates 2-7). The formation of the Caribbean eddies is due in part to potential vorticity that originates in rings from the retroflection of the North Brazil Current (NBC) and that is advected through the Lesser Antilles. This potential vorticity acts as a finite amplitude perturbation for mixed barotropic and inter- nal mode baroclinic instabilities (Plate 2a), and mesoscalefea- 1450 MURPHY ET AL: CONNEcnVn'Y1N INTRA-AMERICA SEAS

-6 -4 -2 0 2 4 6 8 10

-12 0 12 24 incm Plate 11. The 1985-1987 mean minus 1981-1991 mean SSH anomaly in centimeters for (a) simulation 1 with a contour interval of 1.0 cm and (b) simulation 3 with a contour interval of 3.0 cm.

tures form as a result. The Caribbean eddies are primarily traced across the Caribbean and through the Yucatan Chan- anticyclonic and transit a fairly narrow corridor westward nel. In addition, a statistically significant (11.4-month lagged) across the basin (Figure 7) with an average speed of 0.15 mls. correlation of 0.45 was found between eddy shedding by the This results in a 100monthtravel time from the Lesser Antilles Loop Current in the Gulf of Mexico and eddies in the east- to the Yucatan Channel. ernmost Caribbean. However, no net influence of Caribbean With only limited drifter tracks, satellite altimetry, and eddies was found on the mean eddy-shedding period in the sparse hydrographic data, there is a scarcity of observational Gulf of Mexico nor on the size and strength of shed eddies. data in the Caribbean region with which to conduct model-data Considerable interannual variation exists in eddy formation comparisons. However, eddies similar to those in the model and strength. The initial intensity of the Caribbean eddies is (Figure 5) have been observed by Geosat altimetry [Nystuen weak compared with their NBC counterparts, but as the Ca- and Andrade, 1993]. In a model phase plot of upper layer ribbean eddies transit the basin, they intensify into larger-scale thickness (Plate 8), eddies from the Lesser Antilles can be features (with average diameter of 330 km in simulation 1). MURPHY ET AL: CONNECrIVrrY IN INTRA-AMERICA SEAS 1451

tl~t' I I ~ I I :p,'~.{

-- .~. ----\ " !~ ~.-~ \, ' L:--- "

I i; ".;;. ,'"' " ~-j:~ I'f(~~~~ ~",-:---,o I \; , - ," " ~~j~~;~~~~ J. I " J

, --;:.- ~ ~;.a~~\~"'\--. '.., ... .. ' .. ,..'. c:::...t::::::..- ..':' .-' ;~- \

f ~ ,. \ ' .

. .- " ,"" - ':::::::'-~O:'-- 8 r ,-- ', : ~ -'. 8 1 ,. i ~:::.:::~' ~;.: -- \ .-

,.,~:( -' -- -'. 12 --' 1 ~8 . ..." " .. -:--" " " \-.:-:::1~ r <~~~~~- ". ". P;"-=::.. 1

~ :=:=:!J ._~..~ ~~-.,. .. .. Q. ~~ .., -..a - - . . ~ E': '-'E --"' ' it.o ~ ~ ~ .'- -'.~-",... ~~~f~' - ,.Y/~~" ~~ ~" .~,.-'""' ?- "'". ~_. ..~--- - ~,-., ~ , ~~-- ,.,w~. ...g.~.~..-~~:~_:: ~~~ --""~i';~:::~-;;///"""'..v...;,.1~"'~~~_.,.- .~~ '~--~ '. ~~-::-~- -- - - .~ J~-~~ -.- . . , I '-r'//,--,,~ ..- I Figure 1~14"/'/Y~'~"10. The 1985-1986I I mean minus 1981-1991 mean European Centre for Medium-Range Weather

Forecasts (a) wind stress curl with contour intervals at 6.0 Pa/m x 10-8 and (b) wind stress in newtons per square meter.

The amount of intensification is strongly affected by interan- latter exhibited significantly less eddy kinetic energy in the nual varialions in the strength of the Caribbean Current. which Can"bbean(E. J. Metzger, personal communication, 1995). lies along the axis of eddy propagation (Plate lla). Similar A CarIbbean model that is forced by a constant inflow interannual variability exists in the flow through the Lesser through the Lesser Antilles [Hebum et al., 1982) can produce Antilles (Figure 9), which also plays a role in Caribbean eddy some eddies but does not captwe the periodicity, connectivity, strength. or cycles of can"bbean eddies. Even prescnDing a variable The anomaly depicted in Plate lla is consistent with a strong inflow similar to the transports obseIVCdin the model (Figure anomaly in the regional wind stresscurl (Figure 10). The linear 9) would mask the dynamical contributions of the Atlantic, simulation, which is forced by the same winds, also depicts an unless they came from an Atlantic model. Any successfulmod- anomaly in the central Canbbean (Plate lIb). A striking dif- eling of CanDbcan variability must therefore consider the in- ference exists between the areal extent of the anomaly in the fluence of the Atlantic. linear and nonlinear simulations. The nonlinear anomaly (sim- The results of this study depend on the ability of the model ulation I) is stronger in the western Caribbean and extends to represent the necessarymesoscale variability and the geom- across the entire Gulf of Mexico. The anomaly in the linear etry of the islands surrounding the Caribbean. especiaIIy the simulation (simulation 3) is confined to the Caribbean. This Lesser Antilles island arc. With higher model resolution one difference is a consequence of the flow instabilities in the would expect greater mesoscalevariability and more accurate Caribbean Current, the associatededdy generation, the west- representation of the island arc. However, great care was ex- ward propagation of the eddies, and their ability to squeeze ercised in representing the island arc present in the model and through the Yucatan Channel (Plate 6). detailed maps were used to make many improvements over the Both nonlinear simulations show similar complex eddy standard ETOPO5 topography used by most ocean modelers. fields. These simulations were forced by daily winds for 14 These results show that the model represents the geometry years. The variability of these winds greatly enhanced the Ca- well enough and has sufficient variability to demonstrate the ribbean mesoscalevariability over a similar global simulation main features of interest, including (1) production of strong forced by Hellerman and R~n.'tein climatological winds. The rings from the retroflection of the North Brazil CurreQt, (2)

~ 1452 MURPHY ET AI-: roNNEcnvnY IN INI'RA-AMERlCA SFAS connectivity between NBC rings and eddy generation in the perirnent Station, Vicksburg. MDsissippi. The C!X) was used under a eastern Canbbean that is strongly affected by the Lesser An- computer time grant from the Defense Department High Performance tilles, (3) strong mesoscaJevariability with the Canbbean. (4) Computing Initiari\oe. James J. O'Brien receiYed support from the Physical Oceanography Program of ONR under the Secretary of the eddies that transit the entire Caribbean and amplify in the Navy Chair Grant. Sylvia Murphy split her effort between Florida State processowing to flow instabilities. and (5) eddy shedding in the Un~ty and NRL during the conduct of this research and was Gulf of Mexico. Accurate representation of the island arc is supported by an NRL feUowship while at Florida State. We thank critical. In the present study the NBC rings decay on the Tamara L Townsend for her helpful comments and suggestions,Alan J. WaIJcraft for his work on developing of the NRL g1oba1ocean eastern side of the LesserAntilles becausethe island arc serves model, and Ashley McManus for her assistancewith the figures. This as a barrier to propagation into the Can"bbean.If the Lesser work is NRL contnbution JAfl322-96-0030. Antilles were eliminated or represented with unrealistically large gaps. the NBC rings could continue a northwestward pathway into the Caribbean unimpeded. References As described earlier, we find a less direct influence of NBC Cocbrane, J. D., The Yucatan Current and equatorial currents of t~ rings in the Caribbean and Gulf of Mexico because some western Atlantic, Ref. (65-177), pp. 20-27, Dep. of Oceanogr., Tex. potential vorticity from the rings is advected into the Carib- A&M Univ., College Station, 1965. Didden, N.. and F. Schott. Eddies of the Nonh Brazil CUn'CDtret- bean through passagesof the Lesser Antilles, providing finite roflection region observed by Geosat altimetry, J. GeopIIys.&s., 98, amplitude perturbations for flow instabilities within the Can"b- 2O,Ul-20,131, 1993. bean. In the model this impact is sufficient to yield a significant Duncan, c. p.. S. G. SchIadow, and W. G. Williams, Surface currents correlation (0.45) between eddy shedding in the Gulf of Mex- near the Greater and Lesser Antilles, Int. Hydrogr. Rev., LIX(2), ico and eddies near the Lesser Antilles. However, no signifi- 67-78, 1982- European Centre for Medium-Range Weather Forecasts (ECMWF), cant correlation is found between eddy shedding in the Gulf of The description of the ECMWF/WCRP level III-A global atmo- Mexico and transport variations in the Straits of Florida, al- spheric data an:biYe, report, 72 PP.. Reading. England, 1994. though the correlation between transport ftuctuations in the Fratantoni, D. M.. W. E. Johns, and T. L Townsend, Rings of the Yucatan Channel and the Straits of Florida at 2~N is about 0.7 North Brazil Current: Their structure and behavior inferred from with a lag of 15 days. observations and a numerical simulation, J. GeopIIyof.&s., 100, 10,633-10,654,1995. Fu, L L, and B. Holt. Some examples of detection of oceanic me- ~ eddies by the SEASAT synthetic-aperture radar,I. ~ Appendix: Significance Test Re.f.,88, 1844-1852, 1983. Hammeniey, J. M.. and D. c. Handsromb, Mmte Corio Methods,178 Given two time seriesT(j) and S(j) (wherej = 1, N and pp., OIapman and Hall, New York, 1964. N is the total numberof observations)and their time-lagged Hebum, G. W., T. H. Kinder, J. H. Allender, and H. E. Hurlburt, A cross-correlation function CCF( - L, + L) (L is maximum numerM:aJmodel of eddy generation in tbe southeastern Can"bbean Sea, in HydTodynIJmicoSof Semi-Enclosed Sear, edited by J. C. J. lag), a significancetest on CCF can be conductedusing the Nihoul, pp. 299-328, Elsevier Sci.. New York, 1982. following procedure:first, computethe autocorrelationfunc- HeUerDlan, S., and M. Rosenstein, NorDlai monthly wind stressover tion a of T; second,calCulate surrogate T time seriesSTS by the world ocean with error estimates, J. Phy.f.Ocearlogr., 13, 1~ 1104, 1983. STS(j) = aSTS(j) + kV") Hurlburt, H- E., and J. D. Thompson, A numerM:aJstudy of ~ (AI) Current intrusions and eddy sbedding, J. PhyoJ.Oceanogr., 10. 1611- where k is a random number between 0 and 1 andj = 1, 1000. 1651, 1~. Hurlburt, H. E., and J. D. Thompson, The dynamics of the Loop The surrogate time series preserves the autocorrelation func- Current and shed eddies in a numerical model of t~ Qulf of Mex- tion of the original function, and the addition of k is equivalent ico, in Hyfirodynamla of Semi-Enclosed Sear, edited by J. c. J. to the addition of random noise. Nihoul, pp. 243-297, Elsevier Sci., New York, 1982. Third, calculate the cross-correlation function between each Hurlburt, H. E., and J. D. Thompson, Preliminary results from a numerical study of the New England Seamount Chain influence 011 of the surrogate time series and the other original seriesS and tbe Qulf Stream, in Predictability of Fluid Motion.f, edited by Q. rank the values in ascendingorder by lag. The 975th and 25th Holloway and B. J. West. pp. 489-504, Am. Inst. of Phys., College rankings represent two-tailed 95% confidence limits about the Park, Md., 1984. original CCF. Hurlburt, H. E., and T. L Townsend, NRL effons in the Nonh At- The addition of random numbers is frequently used in lantic, in Dilla Assimilation and Model E~'QIUQtiQIIExpm:lllml.J- North Atlantic Basin: Prelimina,,' Experiment Pian, edited by R. C. Monte Carlo techniques [Hammersley and Handscomb, 1964) Willems, pp. 30-39, Cent. for Ocean Atmos. Modeling. Univ. of to create surrogate time series. The ranked correlations be- South. Miss., Stennis Space Center, ]994. tween the surrogate and S represent the highest values one Hurlburt, H. E., A. J. Wallcraft, W. J. Schmitz Jr., P. J. Hoglln, and would statistically expect for a correlation with noise. There- E. J. Metzger, Dynamics of the Ku~hio/ system using eddy-resolving models of the Nonh Pacific Ocean,J. GIo'op#Iys. fore, if the original CCF is higher than a correlation for noise, Res., 10/,941-976, 1996. it is considered significant. Ingham, M. C., and C. V. W. Mahnken, Turbulence and productivity near St. Vincent Island. BWI, A preliminary repon, CaribbeanJ. Sci., 6(3-4),83-87, 1966. Ac~ts. This work was performed as pan of the 6.1 Jacobs, G. A., H. E. Hurlburt, J. C. Kindle. E. J. Metzger. J. L project Dynamics of U>Wlatitude Western Boundary Currents. It is Mitchell, W. J. Teague, and A. J. Wullcraft. Decade-scale trans- also a contribution to the 6.2 Gk>baJOcean Prediction Systemproject, Pacific propagation and warming effects of an EI Nino anomaly, modeling task, as pari of the Naval Ocean Modeling and Prediction Nature, 370, ~-363, 1994. (NOMP) program. The Office of Naval Research (ONR) sponsored Johns, W. E.. T. N. Lee, F. A. Schott, R. J. Zantopp, and R. H. Evans, t~ two projet.1Sunder program elements ro115N and 62435N, re- The Nonh Brazil Current retroRection: Seasonalstructure and eddy spectively. The ocean model :;imuliltions werc run by E. Joseph variubility, J. GC'ophys.Rt's., 9.~.22,103-22.120. I~. Metzger and performed on the CRAY Y-MP 8/8128 at the Naval Kinder, T. H., Shallow currents in the Caribbean Sea and the Gulf of Oceanographic Office, Stennis Space Center, Mississippi, and the Mexico as observed with satellite-tracked drifters, BulL Mar. Sci.,33, CRAY C90/161024at the Army Corps of EngineersWaterways Ex- 239-246, 1983. MURPHY ET AL: CONNEcrIVrrY IN INfRA-AMERICA SEAS 1453

Kinder, T. H~ G. W. Heburn, and A. W. Green, Some aspectsof the Roemmich. D.. Circulation of the Caribbean Sea: A well resolved Caribbean Circulation, Mar. Geol., 68, 25-51, 1985. invenie probl.:m, J. Geoph,vs.Re.f., 86,7993-8005. 1981. Larsen,J. c..Transport and heatflux of the Florida Currentat 2~ Schmitz, W. J.. Jr., On the interbasin-scale thermohaline circulation, derived from cross-streamvoltages and profiling data: Theory and Rev. ~"S~ 33(2), 151-173, 1995. observations,Phuos. -Trans. R. Soc. London, SeT.A, 338A, 169-236, Schmitz, W. 1.. JrryOn the world ocean circulation, vol. I, Some global 1992. features/North Atlantic Circulation, Tech.Rep. WHOI-96-03,Woods uipper, D. F., A sequenceof current patterns in the Gulf of Mexico, Hole Ocean02r. Inst.. Woods Hole. Mass., 1996. J. GeopilYS.Res., 75, 937-948, 1970. Schmitz, W. 1.. J~ryand P. L Richardson, On the sourcesof the Aorida umming, T. D., Eddies west of the Southern Lesser Antilles, in Current, Deep Sea Res., Pan A, 38, supp. I, S379-S409, 1991. Symposiumon Investigationsand Resourcesof the Caribbean Seaand Shriver, J. F.. and H. E. Hurlburt, The contribution of the global Adjacent Regions,pp. 113-120, UNESCO, Paris, 1971. thermohaline circulation to the Pacific to throughflow Levitus, S~ Oimatological atlas of the world ocean, NOAA Prof Pap. via Indonesia.J. Geoph.vs.ReSry 102, 5491-5511,1997. 13, 173 pp., U.S. Govt. Print. Off., Washington, D. C, 1982. Sokal, R. R.. and F. J. Rohlf, Biometery,859 ppryW. H. Freeman, New Maul, G. A (Ed.), Climate Changein the Intra-Americas Sea, 389 pp., York, 1981. Chapman and Hall, New York, 1993. Stalcup, M. C.. and W. G. Metcalf, Current measurements in the Maul, G. A, and F. M. Vukovich, The relationship betwcen variations passagesof the Lesser Antilles, J. Geopilys. Res.. 77, 1032-1049, in the Gulf of Mexico Loop Current and the Straits of Florida 1972. volume transport, J. Phys. Oceanogr.,23, 785-796, 1993. Stalcup, M. C.. W. G. Metcalf, and M. E. Zemanovic, Current mea- Maul, G. A., D. A Mayer, and S. R. Baig, Comparisons between a surements in the Lesser Antilles, Tech. Rep. WHOI-7/-5I, Woods continuous 3-year current.meter observation at the sill of the Yu- Hole Oceanogr. Inst, Woods Hole, Mass., 1971. catan Strait, satellite measurementsof Gulf Loop Current area, and Sturges, W., The spectrum of Loop Current variability from gappy regional sea level, J. Geophys.Res., 90, 9089-9096, 1985. data, J: Ph.I's.Oceanogr., 22, 1245-1256, 1992- Mazeika, P. A, T. H. Kinder, and D. A Burns, Measurements of Sverdrup, H. Ury Wind-driven currents in a baroclinic ocean; With subtidal flow in the Lesser Antilles passages,J. GeophYS.Res., 88, application to the equatorial currents of the eastern Pacific, Proc. 4483-4488, 1983. Nall Acad. Sci. U. S. A., 33, 318-326, 1947. McWilliams, J. C., and G. R Flierl, On the evolution of isolated Toggweiler, J. R., and B. Samuels,New radiocarbon constraintson the nonlinear vortices,J. Phys. Oceanogr.,9,1155-1182, 1979. upwelling of abyssal water to the oceans surface, in The Global Mesinger, F., and A Arakawa, Numerical methods used in atmo- Carl:IonCycle, NATO ASl Ser.,Ser. I, voL 15, edited by M. Heimann, spheric models, GARP PubL Ser. 17, 64 pp., World Meteorol. pp. 303-331, Springer-Verlag, New York, 1993. Organ., Geneva, 1976. Vukovich, F. Mry Loop Current boundary variations, J: Geopilys.Res., Metcalf, W. G., M. C. Stalcup, and M. E. Zernanovic, Hydrographic 93, 15,585-15,591,1988. station data from Atlantis II cruise 56 to the southeastern ap- Vukovich, F. M., An updated evaluation of the Loop Current's eddy- proaches to the Caribbean Sea, February-April 1970, Tech. Rep. shedding frequency, J: Geophys.Res., ]00, 8655-8659, 1995. WHOI-71-13, Woods Hole Oceanogr. Inst., Woods Hole, Mass., Wallcrait, A. J., The Navy Layered Ocean Model Users Guide; 1971. NOARL Rep.35, 21 pp., Nav. Res. Lab., Stennis SpaceCent, Miss., Metzger, E. J., H. E. Hurlburt. J. C. Kindle, z. Sides, and J. M. 1991. Pringle, Hindcasting of wind-driven anomalies using a reduced- Wilson, W. Dry and W. E. Johns, Velooty structure and transport in gravity global ocean model, Mar. TechnoLSoc. J., 26(2), 23-32, 1992. the Windward Island Passages,Deep SeDRes., Pan I, 44, 487-520, Molinari, R. L, M. Spillane, L Brooks, D. Atwood, and C. Duclcett, 1997. Surface currents in the Canobean Sea as deduced from Lagrangian Youtsey, W. J.. Report detailing modifications to the 1/8 degree global observations,J. GeOphYs.Res., 86, 6537-6542, 1981. bathymetry, Sverdrup TechnoL Inc., Stennis Space Cent, Missry Munk, W. H., On the wind-driven ocean circulation, J. Meteorol., 7, 1993. 79-93, 1950. National Oceanic and Atmospheric Administration, ETOPO5 digital H. E. Hurlburt and S. J. Murphy, Naval Research Laboratory, Code relief of the surface of the earth, Data AnnOUIlCe.86-MGG-O7, Natl. 7320, Stennis Space Center, MS 39529. ([email protected]) Geophys. Data Cent, Washington, D. C., 1986. J. J. O'Brien, Center for Atmospheric Prediction Studies,2035 East Nystuen, J. A, and C A Andrade, Tracking mesoscaleocean features Paul Dirac Drive, R. M. Johnson Building, Suite 200. Tallahassee,FL in the Caribbean Sea using Geosat altimetry, J. Geophys.Res., 98, 32310. 8389-8394, 1993. Richardson, P. L, G. E. Hufford, R. Limeburner, and W. S. Brown, North Brazil Current retroflection eddies,J. Geophys.Res., 99, 5081- (Received Mardt 6, 1998; revised August 13, 1998; 5093, 1994. accepted August 31, 1998.)