The observed seasonal preferences of Loop Current eddy shedding, more in summer and winter and less in fall and spring, are shown for the first time to be due to a curious combination of forcing by the seasonal winds in the Caribbean Sea and the Gulf of Mexico. The conditions are favorable for the Loop to shed eddies in summer and winter when strong trade winds in the Caribbean produce large Yucatan transport and Loop's intrusion, and concurrently when weak easterlies in the Gulf offer little impediment to eddy shedding. The conditions are less favorable in fall and spring as the trade winds and Yucatan transport weaken, and the strengthening of the Gulf's easterlies impedes shedding.
 Early studies of the Loop Current in the Gulf of Mexico in the 1960's∼1980's suggest that it may vary seasonally. The northward penetration of the Loop Current was bimodal: maximum penetrations occur, on average, in winter (Dec∼Jan) and summer (Jun∼Jul) [Leipper, 1970; Behringer et al., 1977; Molinari et al., 1978; Sturges and Evans, 1983]. Molinari et al.  concluded that the seasonal intrusion of the Loop Current varied with the geostrophic transport through the Yucatan Channel. Sturges and Evans suggested that the Loop Current varied in response to wind. These pioneering authors also recognized that there were substantial deviations from the seasonal cycle, and intrusions and eddy-sheddings can occur in virtually any month of the year. That the Loop Current can intrude into the Gulf and eddies can separate from it without the need for a seasonal forcing such as the inflow transport was first demonstrated numerically byHurlburt and Thompson , and since then confirmed by numerous studies using more elaborate models.
 The idea of a seasonal Loop Current is nevertheless very attractive; the system is more predictable, and understanding the underlying mechanisms can lead to improved predictions of the strong currents and heat content associated with the Loop, which have practical significance. In this work, the old problem of a seasonal Loop Current is revisited taking advantage of the order-of-magnitude increase in data coverage from satellite, advent in models and forcing data, and improved theoretical understanding of Loop Current dynamics.
2. Observed Loop Current Shedding Events
 The dates of Loop Current eddy separation from 1974 to 1992 are from Vukovich , Sturges  and Sturges and Leben using a combination of satellite-SST images as well asin situ and ship measurements to identify eddy separations. From 1993 to 2010, satellite altimetry data from AVISO [http://www.aviso.oceanobs.com/] is used. For shedding period shorter than 2 months (one in 1993, the other one in 2002), the two consecutive events are taken as the same event, and the first shedding is recorded. There are 47 eddy shedding events from 1974 to 2010. Figure 1a sorts the number of shedding events by months (a seasonal histogram or SeH) and indicates that eddy shedding has a bimodal (biannual) seasonal signal: maximum in summer (Jul∼Sep) and winter (Mar), and minimum in late fall (Nov∼Dec) and late spring (May∼Jun). The maximum difference in eddy count (Mde) is 7 between the period of most and least eddies. Approximately 40% of the eddies are shed during summer, but only one eddy is shed in the late fall (Nov∼Dec). However, most of the seasonal signal is for the record after 1993 (bars in Figure 1a); summer eddy sheddings then account for 45% of the total, and no eddies were shed in Nov∼Dec. This difference suggests a shift in the Loop Current's behaviors between the two periods - a point we will comment on later. The seasonal preference of eddy-shedding suggests that the system is, at least in part, forced. Such a possibility was anticipated byChang and Oey [2010, hereinafter CO2010; see also Oey et al., 2003, hereinafter OLS2003] whose process experiments show the effects of wind on Loop Current eddy-shedding.
 Another way of displaying the eddy-shedding data is to plot the eddy-shedding histogram (ESH;Figure 1b). The ESH has peaks (e.g., 6, 9 months etc), but most importantly it shows wide-ranging shedding periodsPfrom 4∼19 months: the eddy-shedding process appears to be chaotic. However, the “broad-spectrum” ESH can be a consequence of the seasonal shedding preferences of eddy-shedding. The argument is straightforward as summarized inFigure 1c. For example, suppose the forcing is such that the Loop sheds eddies in August and September, the ESH then shows values at 1, and 11–13 months. By including only 4 observed, preferred shedding months: March, July, September and October (3,7,9,10 in Figure 1c), a broad-spectrum ESH with periods from 1–20 months can exist. The solution is not unique, but this is not central to our argument. The point here is that an orderly, seasonally forced Loop Current that sheds eddies only in certain months is consistent with the existence of a broad spectrum of shedding periods; in other words, a chaotic system is not necessary for the existence of the broad spectrum. In addition to possible contribution from some natural shedding periods which depend on internal physics [e.g.,Hurlburt and Thompson, 1980; OLS2003], peaks in the ESH may then be thought of as the result of some interannual variations of the forcing that perturb the shedding month from one year to the next, or even no shedding at all until the following year. That the Loop Current and eddy-shedding system may be non-chaotic was first suggested byLugo-Fernandez .
 The contrary is not necessarily true. In other words, a chaotic Loop Current with a broad-spectrum ESH which may contain some prominent peaks (Figure 1b) does not in general lead to seasonal preferences of eddy-shedding (Figure 1a). With steady forcing a modeled Loop Current can display a natural period (e.g., CO2010); on the other hand, experiments can be designed to produce a chaotic system with a broad-spectrum ESH (OLS2003). Assuming such a system exists in the observed world, that the corresponding ESH has a broad spectrum with prominent peaks around some natural periods, what then can be deduced about its SeH? Given P, the month Msh when shedding occurs is:
where FP = 12/gcd(12, P) is the number of peaks in the SeH for that P, gcd = greatest common divisor, P = 1, 2, 3, …, 19, 20 months, and Msh0 = the month of the first shedding; Figure 1d shows the case for Msh0 = 1. It is readily shown that, despite the presence of biannual and/or annual peaks in the shedding periods (that may therefore favor a seasonal SeH), the existence in the observed ESH (Figure 1b) of a P = PFull = 5, 7, or 11, etc for which gcd(12, PFull) = 1, can yield a non-seasonal SeH (details in theauxiliary material).
 The simple calculations above demonstrate the importance of order in the shedding events. It appears that nature has selected an order that, in the case of the Loop Current, is largely non-random. In other words, the shedding process is largely controlled by some form of external forcing, such as the winds. Model experiments support this assertion.
3. Processes That Control the Seasonal Shedding of the Loop Current Eddies
 The importance of wind forcing on eddy-shedding has previously been noted (OLS2003; CO2010). We now demonstrate that the existence of a bimodal SeH (Figure 1a) is caused by a curious complementary effect (on the Loop Current) of the zonal component of the seasonal winds in the Caribbean Sea and the Gulf of Mexico.
3.1. Seasonal Winds
 Winds in the Caribbean Sea vary depending on the movement and intensity of the North Atlantic Subtropical High and, in winter, on the North American High also (Figures S2–S3). In the Gulf of Mexico, winds are additionally modified by the North American monsoon in summer, the high pressure over the northeastern US in fall, and the low pressure over the western US in spring. The combined effect is that the seasonal winds are 180° out of phase in the two regions: the Caribbean easterly is strong in winter and summer and weak in spring and fall while the Gulf's easterly wind is stronger in fall and spring and weak in summer and winter (Figure 2a).
3.2. Numerical Experiments
 This out-of-phase relation between the seasonal winds in the Caribbean Sea and the Gulf of Mexico is central to the understanding of why the Loop Current tends to shed more eddies in some months than others. Within the Gulf, easterly wind forces an eastward return flow across the middle of the basin which counters the westward-growing Loop Current by Yucatan inflow and Rossby-wave dynamics and delays eddy-shedding [Chang and Oey, 2010]. We may expect then that the easterly peaks in the Gulf of Mexico in Oct∼Nov and, to a lesser degree, in Apr∼May, would delay eddy-shedding, which would be consistent with the observed SeH (Figure 1a) that less eddies are shed in those months. However, explanations based on the Gulf's forcing alone are incomplete; the dynamics of the Caribbean Sea are necessary.
 The NW Atlantic Ocean model (5°–50°N and 98°W–55°W; see Figure S4 in the auxiliary material which also contains model descriptions) that we have previously tested (e.g., OLS2003; CO2010) for studying Loop Current dynamics is set up to run various experiments to isolate processes. The “Basic” experiment is forced by the CCMP wind stresses (0.25° × 0.25°, 6-hourly satellite+NCEP blended dataset ) from 1988–2009. The “NoWind” experiment has no wind. In the “Atl” experiment, the wind is applied to the east of 82°W only, and the experiment “GOM+NWCar,” has wind applied to the west of 82°W only. Finally, the “GOM+NWCarNoCurl” also has winds applied west of 82°W but they are zonal only and are spatially constant averaged over the Gulf of Mexico and the NW Caribbean Sea (Figure 2a). This last experiment has the essentials of the out-of-phase relation between the seasonal winds in the Caribbean Sea and the Gulf of Mexico. Each experiment was conducted for 22 years (1988–2009). To ensure robustness of our results, the Exp.Basic, Atl and GOM+NWCarNoCurl were repeated for additional 22 years with different initial fields and with a reduced Smagorinsky's constant (0.05 instead of 0.1) for the horizontal viscosity.
 The Exp.NoWind yields P ≈ 7∼10 months around a peak ≈ 8 months (e.g., OLS2003; CO2010). Its SeH is basically full (no seasonal preference with small standard deviation (sd) = 0.5 and an Mde of only 1; not shown) as may be anticipated from the discussions (Figures 1c and 1d) of the previous section. Exp.Atl also gives a full SeH, also with small sd = 0.4 and Mde =1 (Figure 3a, grey). Remote winds in the eastern Caribbean Sea and the North Atlantic Ocean are therefore unlikely to force a seasonal shedding. The Exp.Basic (Figure 3a, solid) has sd = 1.8 and Mde= 6; it shows eddy-shedding preferences in winter (Feb∼Mar) and summer (Jul∼Aug), with less shedding in late spring (May, 4 less) and early fall (Oct∼Nov, 6 less), in general agreements with observations. This suggests that the seasonal eddy-shedding is wind-forced. This deduction is confirmed by the SeH from Exp. GOM+NWCar (Figure 3b; sd = 1, Mde=4), which shows similar winter (Mar) and summer (Aug) shedding preferences. Experiments GOM+NWCar and Exp.Atl show that it is the regional wind in the Cayman Sea (i.e., NW Caribbean Sea) and the Gulf of Mexico that influences the seasonal eddy-shedding of the Loop Current. Finally, when the wind stress curl is removed, Exp. GOM+NWCarNoCurl (Figure 3c; sd = 1.3, Mde = 5) shows that the zonal component of the wind alone can explain the seasonal preferences with more sheddings in winter (Mar) and summer (Jul∼Sep). While there are some differences in the preferred months of shedding amongst the three experiments, we do not consider them to be significant.
3.3. Why Can Wind Force a Seasonal Preference in the Shedding of Loop Current Eddies?
 Yucatan transport (TrYuc) also varies biannually: stronger in summer and winter and weaker in spring and fall [Molinari et al., 1978; Rousset and Beal, 2010]. Simulated TrYuc and Caribbean wind stress (τo, and wind stress curl ∇ × τo) are significantly correlated with wind leading by 0∼3 months. Correlation maps show that winds in the Cayman Sea are effective in driving transport fluctuations (Figures 2d and 2e): westward wind stress (τox < 0) and negative ∇ × τo drive stronger TrYuc. The TrYuc is positively correlated with τox in the eastern Gulf: TrYuc decreases as westward wind in the Gulf becomes stronger (CO2010).
 The seasonal preferences of eddy-shedding can now be explained. It is well-known that the Loop Current tends to shed eddies more readily when it extends northward into the Gulf, and that once the Loop is in the extended state and ready to shed, the process is relatively fast (a few weeks [e.g., OLS2003]). The fundamental variable for the Loop's intrusion is TrYuc. In summer and winter, TrYuc increases as the negative wind stress and wind stress curl in the Caribbean Sea increase (see wind plots in Figure S3 in the auxiliary material); the easterly peaks (Jul and Jan) in the Caribbean correspond well to the peaks in TrYuc especially for summer (Figures 2a and 2b). The larger TrYuc leads to stronger inflow velocity vo and cyclonic vorticity ζo on the western (∼50 km) portion of the Yucatan Channel, and a more extended Loop Current [Oey, 2004; OLS2003]. The ζo/f (f = Coriolis parameter) is an excellent predictor of the Loop Current's northern boundary with high R2 = 0.83 for their linear regression (Figure 2c). While this linear relation agrees with the Reid's formula [Reid, 1972; OLS2003], we treat it to be merely an empirical one. The Loop Current therefore tends to be extended in summer and winter. As TrYuc decreases (Sep and Mar) when the Caribbean (westward) windstress weakens (Jul∼Sep, and Jan∼Mar), the Loop retracts as ζo also weakens. The mass influx (Qi) feeding the Loop also decreases, providing a favorable condition for the westward Rossby wave speed of the extended Loop (Ci ∼ −βRd2, where Rd = Rossby radius based on the depth of the matured Loop) to overcome Qi, hence also a favorable condition for eddies to separate [Nof, 2005]. The weakening of the wind (and transport) are abrupt especially in summer (Figures 2a and 2b). Moreover, because the Gulf of Mexico's easterlies are weak during those periods (Figure 2a), the eastwardmomentum flux that impedes eddy-shedding (CO2010) is also weak. This combination of strong Caribbean easterly, abrupt weakening, and weak easterly in the Gulf of Mexico favors a larger proportion of eddies being shed from Jul∼Aug and Feb∼Mar (Figure 3). In fall and spring, TrYuc and the Caribbean easterly remain weak but at the same time westward wind in the Gulf of Mexico intensifies (Oct and May; Figure 2a). The Loop Current's expansion and eddy-shedding are now impeded by the eastward momentum flux that intensifies along the mid-latitudes within the Gulf. These factors lead to a reduced number of eddies being shed in fall and spring (Figure 3). These processes are summarized schematically in Figure 4. In the auxiliary material, the dynamics are further examined using a simple reduced-gravity model (Exp.RG). The Exp.RG confirms that easterly wind in the NW Caribbean Sea drives a seasonal shedding. The Gulf's easterly wind accentuates the seasonality by delaying eddy-shedding in fall and spring: it increases the summer-fall (or winter-spring) difference in the number of eddies shed. We also compared the RG experiments with the 3D Exp.Basic (and Exp.GOMCarNocurl) using the ensemble averaging idea of the Loop Current Cycle described byChang and Oey . In the 3D experiments, we found that on average eddy-shedding follows shortly (∼1 month) after the maximum Yucatan transport, but that in Exp.RGCarib there is an additional time-lag of 1∼2 months. The RG response is similar to the EOF modes 1 + 2 of the 3D experiments while interestingly the EOF mode 3 accelerates the shedding in the 3D experiments and closely resembles the Campeche Bank instability mode [Oey, 2008]. Therefore, dynamical instability takes part in the eddy-shedding process, but it does not control the seasonal timing.
4. Summary and Conclusions
 The Loop Current is observed to shed more eddies in summer and winter. Numerical experiments also yield seasonal preferences with more sheddings in winter and summer, and less in fall and spring in agreement with observations. The seasonal preferences are forced by the seasonal winds in the Caribbean Sea and the Gulf of Mexico. The Loop sheds more eddies in summer and winter in response to intensified Yucatan transports driven by the stronger trade winds in the Caribbean, and concurrently when weak easterlies in the Gulf offer little impediment to eddy shedding. The conditions are reversed in fall and spring when the Caribbean's (Gulf's) easterlies weaken (strengthen). Since wind plays a central role, our results suggest the existence of an interannual variation of the eddy-shedding process. Indeed,Figure 1a indicates that the biannual seasonal preferences are much less distinct for the first half of the data period from 1974–1992. The second half (1993–2010) has more shorter (biannual) periods, and why that is so may be due to a basic change in the wind. This and other consequences will be examined in a future study.
 We gratefully acknowledge the supports by the Bureau of Offshore Energy Management contract M08PC20007 and the Portland State U. contract 200MOO206.
 The Editor thanks the anonymous reviewers for their assistance in evaluating this paper.