Simulating transport pathways of pelagic Sargassum from the Equatorial Atlantic into the Caribbean Sea

Since 2011, beach inundation of massive amounts of pelagic Sargassum algae has occurred around the Caribbean nations and islands. Previous studies have applied satellite ocean color to determine the origins of this phenomenon. These techniques, combined with complementary approaches, suggest that, rather than blooms originating in the Caribbean, they arrive from the Equatorial Atlantic. However, oceanographic context for these occurrences remains limited. Here, we present results from synthetic particle tracking experiments that characterize the interannual and seasonal dynamics of ocean currents and winds likely to influence the transport of Sargassum from the Equatorial Atlantic into the Caribbean Sea. Our findings suggest that Sargassum present in the western Equatorial Atlantic (west of longitude 50°W) has a high probability of entering the Caribbean Sea within a year’s time. Transport routes include the Guiana Current, North Brazil Current Rings, and the North Equatorial Current north of the North Brazil Current Retroflection. The amount of Sargassum following each route varies seasonally. This has important implications for the amount of time it takes Sargassum to reach the Caribbean Sea. By weighting particle transport predictions with Sargassum concentrations at release sites in the western Equatorial Atlantic, our simulations explain close to 90% of the annual variation in observed Sargassum abundance entering the Caribbean Sea. Additionally, results from our numerical experiments are in good agreement with observations of variability in the timing of Sargassum movement from the Equatorial Atlantic to the Caribbean, and observations of the spatial extent of Sargassum occurrence throughout the Caribbean. However, this work also highlights some areas of uncertainty that should be examined, in particular the effect of “windage” and other surface transport processes on the movement of Sargassum. Our results provide a useful launching point to predict Sargassum beaching events along the Caribbean islands well in advance of their occurrence and, more generally, to understand the movement ecology of a floating ecosystem that is essential habitat to numerous marine species.


Introduction
Pelagic Sargassum spp. aggregates to form "Lagrangian ecosystems" that drift across the ocean surface. Sargassum provides important habitat for a diversity of invertebrates, fishes, turtles, and birds (Coston-Clements et al., 1991;Wells and Rooker, 2004;Witherington et al., 2012). However, when large accumulations of Sargassum enter coastal waters it may also act as marine debris, disrupting ship traffic, fishing, and tourism, which negatively impacts the economies of coastal communities. Major and unexpected accumulation of Sargassum occurred near and along the islands of the eastern Caribbean starting in 2011 and continued through December of that year (Gower et al., 2013;Hu et al., 2016). A similar event occurred in 2012 (Marechal et al., 2017), followed by even greater Sargassum inundation events throughout the Caribbean in 2014 through 2015 Hu, 2016, 2017). These anomalous events caused widespread negative impacts to the fishing and tourism economies throughout the region. These events have been the subject of considerable scientific investigation using remote-sensing observations from satellite (Gower et al., 2013;Hu et al., 2016;Hu, 2016, 2017) and morphological and genetic T characterization of the Sargassum present (Schell et al., 2015;Amaral-Zettler et al., 2017). Satellite-based observations of large (> 4000 m 2 ) Sargassum mats in the Equatorial Atlantic appear to coincide with the timing of inundation events (Gower et al., 2013;Hu et al., 2016;Wang and Hu 2016;, from which a statistics-based prediction can be made on Sargassum blooms in the eastern Caribbean (Wang and Hu, 2017).
However, oceanographic context for the movement of Sargassum into the Caribbean Sea has received limited attention (Franks et al., 2016). Transport pathways between the Equatorial Atlantic and the Caribbean Sea are well-described, and the North Brazil Current System appears to play an important role (Kelly et al., 2000;Lumpkin and Garzoli, 2005;Johns et al., 2014). There are three main surface dynamical oceanographic features whereby Sargassum travel from the Equatorial Atlantic into the Caribbean Sea: (1) along the continental shelf break of South America via the Guiana Current, (2) across the deep ocean translated via North Brazil Current Rings, and (3) initially moving eastward with the North Brazil Current Retroflection before being swept westward by the North Equatorial Current (Johns et al., 2014) (Fig. 1). With increased understanding of how these advection processes influence the distribution of Sargassum, it may be possible to better assess and predict the occurrence and timing of Sargassum accumulation in the Caribbean Sea, an important first step to assessing the potential for beaching (Marechal et al., 2017) as Sargassum abundance in the Caribbean appears to be directly related to beaching events (Wang and Hu, 2017).
Here, we present results from a series of numerical experiments examining surface transport pathways into the Caribbean Sea from the Equatorial Atlantic. We first assess spatial, seasonal, and annual variability in surface transport from the Equatorial Atlantic into the Caribbean Sea with synthetic particle tracking simulations using the Global Hybrid Coordinate Ocean Model surface currents (HYCOM) (Chassignet et al., 2007) and National Centers for Environmental Prediction Reanalysis (NCEP) winds (Kalnay et al., 1996;Kanamitsu et al., 2002) for the years 2000-2015. We explore the statistical relationships between surface currents and Sargassum coverage in the western Equatorial Atlantic and the occurrence of Sargassum in the eastern Caribbean Sea. We then develop a simple mechanistic model to predict annual Sargassum coverage in the eastern Caribbean Sea by pairing particle transport predictions with remotely-sensed satellite data on Sargassum in the western Equatorial Atlantic. Finally, we assess the relative contributions of three main surface routes to the movement of Sargassum from the Equatorial Atlantic into the Caribbean Sea to provide recommendations for what aspects of ocean circulation could be monitored to serve as a potential tool to provide seasonal warnings to coastal communities that may be negatively affected by future Sargassum accumulation events.

Methods
To simulate Sargassum transport we used output from Global HYCOM for ocean current data and output from the NCEP Reanalysis for wind data. Global HYCOM hindcasts are daily snapshots of ocean velocity at a spatial resolution of 0.08°. HYCOM is an eddy-resolving model that assimilates in situ and satellite observations to depict oceanic conditions that occurred at specific times in the past. The NCEP Reanalysis provides a historical depiction of global atmospheric conditions, including daily averaged surface wind velocity at a spatial resolution of 2.5°. For HYCOM ocean velocities (surface layer) and NCEP winds (10 m above sea level) we extracted daily output for 2000-2015. HYCOM outputs were obtained from the Global Reanalysis (expt. 19.1) for years 2000-2012 and from the Global Analysis (expts. 90.9, 91.0, and 91.1) for years 2013-2015.
Global HYCOM performs reasonably well at portraying the main ocean circulation features, such as fronts, filaments and mesoscale eddies, which are relevant to the transport of marine organisms (Chassignet et al., 2007;Putman and He 2013). Other physical processes occurring at the ocean surface may also play an important role in organismal movement, but are not accounted for in HYCOM (Putman et al., 2016). Such processes include direct forcing of near-surface currents by winds (e.g., the top of the Ekman surface spiral, described more precisely as Stommel shear (Rio et al., 2014;Bonjean and Lagerloef 2002)), direct momentum transfer to floating material from the wind (i.e., windage (Trinanes et al., 2016)) and residual transport due to waves (i.e., Stokes drift (Monismith and Fong 2004)). The influence of these processes on the trajectory of an object is dependent upon its shape, profile, buoyancy, etc. At present, such factors are not well-described for pelagic Sargassum. For instance, Sargassum's buoyancy likely changes with time (age), the amount of biofouling, and ocean conditions (Johnson and Richardson 1977;Woodcock 1993;Zhong et al., 2012). Buoyancy can change the surface expression of Sargassum and thus the cross-sectional area it presents to surface winds; this in turn may change the drag coefficient appropriate to estimating momentum transfer from those winds.
As a first approximation to explore the sensitivity of transport predictions to such effects, we performed simulations using the Global HYCOM surface velocity and a simple "windage" factor of 1.0% of the 10 m wind speeds from the NCEP reanalysis that was applied in the same direction as the wind. This value falls within the range of possible windage factors that might be expected for Sargassum, or for a variety of floating objects, perhaps towards the lower end of what is likely (Breivik et al., 2011;Nero et al., 2013;Trinanes et al., 2016). Regardless, we reiterate the value was chosen as a simple test of the sensitivity of Global HYCOM predictions to aspects of particle transport not contained in that model. The NCEP data were interpolated to the 0.08°HYCOM grid to perform simulations that included windage. We used ICHTHYOP v.2 particle tracking software (Lett et al., 2008) to simulate the movement of Sargassum in the numerical experiments described below. ICHTHYOP used a Runge-Kutta 4th-order time-stepping method to compute trajectories at 30-minute intervals. Particle trajectories were simulated for up to one year, following the initial analyses of Franks et al. (2016), which suggested strong connectivity between the Equatorial Atlantic and Caribbean Sea on this time-scale.  (Lumpkin and Garzoli 2005). The blue arrow depicts the North Brazil/Guiana Current, the purple arrows show North Brazil Current Rings (which translate northwestwards toward the Caribbean on the order of several times per year), the thick red arrow represents the North Brazil Current Retroflection, and the thin red arrows indicate the potential westward return via the North and South Equatorial Currents (Kelly et al., 2000). The dashed lines indicate the Equatorial Atlantic (0-10°N) where extensive and anomalous Sargassum was detected by satellite from April through September 2011 (Gower et al., 2013). Ocean currents in this region are hypothesized to play an important role in the beaching events that occurred in the Caribbean that same year (Franks et al., 2016). The dotted lines represent the area of ocean where long-term, monthly measurements of Sargassum coverage have been obtained (2000-2015) (Wang and Hu 2016). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Backtracking
To assess which areas in the Caribbean could be impacted by Sargassum originating in the Equatorial Atlantic, we released synthetic particles across the Caribbean Sea during the years 2003-2015 and backtracked them for up to one year. For the seasons in which Sargassum was reported as first entering the Caribbean during 2011 (Wang and Hu 2017;Franks et al., 2016), 2500 synthetic particles were released at randomly assigned locations at 5-day intervals across the Caribbean Sea, resulting in 30,000 synthetic particles backtracked each spring (April-June or AMJ), summer (July-September or JAS), and autumn (October-December or OND). Particles were then aggregated into 0.5°× 0.5°release zones and the percentage of particles in each zone that were backtracked to the Equatorial Atlantic was computed. We defined backtracked particles that crossed east of 58°W and south of 10°N as having an Equatorial Atlantic origin (Fig. 1).

Forward tracking
To examine from which areas of the Equatorial Atlantic transport of Sargassum into the Caribbean is likely, we released synthetic particles between 0°-10°N and bounded by the coasts of South America and Africa ( Fig. 1). This region has been identified as the most likely source of Sargassum in the 2011 event (Gower et al., 2013;Franks et al., 2016). Particles were released at randomly assigned locations at 5-day intervals, 500 particles per release, and forward tracked for one year. This resulted in 9000 synthetic particles tracked during each winter (JFM), spring (AMJ), summer (JAS), and autumn (OND) for the years 2000 through 2014. For analysis, particles were aggregated into 1°× 1°release zones and the percentage of particles in each that entered the Caribbean Sea (i.e., the same area from which synthetic particles were backtracked in the preceding section) within one year was determined.

Statistical exploration of annual variation in the role of ocean transport on Sargassum abundance
We used statistical models to explore the relationships between the results of synthetic forward-tracking experiments (described above) and annual variation in Sargassum abundance in the Equatorial Atlantic and the Caribbean Sea. Satellite-based Sargassum observations were provided in Wang and Hu (2016) for the region between 0-22°N and 63-38°W (Figs. 1 and 2). These data are reported as the percentage of Sargassum cover across a 0.5°× 0.5°area for each month from 2000 to 2015. We obtained an annual value of Sargassum percent coverage (SPC) in the eastern Caribbean Sea (10-22°N, 58-63°W) and an annual value of SPC in the region where forward-tracking of synthetic particles and SPC data overlap (Equator-10°N, 60-38°W; Fig. 1). To assess variability in transport, we computed the annual percentage of synthetic particles released from this region of the western Equatorial Atlantic that entered the Caribbean Sea (see above).
We used Spearman correlation (non-parametric) tests (Spearman, 1904) to determine whether the relative annual variation in transport predictions into the Caribbean were related to relative annual variation in SPC values in the eastern Caribbean Sea (e.g., whether years of high transport probability were also years of high SPC values). For comparison, we performed the same tests to examine the relationship between western Equatorial Atlantic SPC values and eastern Caribbean Sea SPC values. Using multiple regression and variance partitioning procedures (Legendre and Legendre, 1998), we calculated the relative contributions ocean transport and western Equatorial Atlantic SPC values for predicting eastern Caribbean SPC values. For these analyses, SPC data were log-transformed.

A Sargassum transport index
We developed a simple mechanistic model of Sargassum transport by combining synthetic particle transport predictions from the western Equatorial Atlantic with SPC values across this region. We aggregated the synthetic particles released into 1°× 1°bins for each winter (JFM), spring (AMJ), summer (JAS), and autumn (OND). For each seasonal bin, we multiplied the probability of transport into the Caribbean Sea within one year (derived from synthetic particle tracking experiments) by the seasonally aggregated percent coverage of Sargassum for the same period. This allowed us to scale our transport predictions relative to the amount of Sargassum present at a given location. Thus, if a 1°× 1°bin was half covered with Sargassum (SPC = 0.5) and the probability of transport from that bin to the Caribbean was 10%, the index value associated with that bin would be 0.05. From these data we produced a seasonal index that indicates the amount of Sargassum likely to move into the Caribbean Sea from the western Equatorial Atlantic, where forward-tracking and SPC data overlap (Equator-10°N, 60-38°W), which is also the region where the highest concentrations of Sargassum tended to initially occur (Figs. 1 and 2). It is important to note that this transport index makes no attempt to account for growth, sinking, grazing, physical aggregation or dispersion of mats, or other processes that may change the amount of Sargassum that enters the Caribbean Sea. Our focus is entirely on assessing the role of surface currents and winds on the movement of Sargassum between regions.
We assessed whether this Sargassum transport index from the western Equatorial Atlantic could predict SPC values in the eastern Caribbean for a given year. We used Spearman correlation (non-parametric) tests (Spearman, 1904) to determine whether the Sargassum transport predictions into the Caribbean were related to relative annual variation in SPC values in the eastern Caribbean Sea, i.e., whether years with a high Sargassum transport index were also years of high SPC values in the Caribbean or vice versa. Additionally, these data were logtransformed to perform linear regressions to assess what percent of variance in eastern Caribbean SPC values could be explained by the Sargassum transport index. Because this mechanistic model explicitly were rare prior to 2011, after which such occurrences have become increasingly common. Satellite data east of 38°W is still being analyzed and thus was not available for these models. Data derived from (Wang and Hu 2016).
links oceanic transport and Sargassum, variance partitioning analyses would not be meaningful to perform in this instance.

Oceanographic pathways
A key benefit of the mechanistic model we developed is the ability to assess the oceanographic pathways that Sargassum might take to reach the Caribbean Sea from the Equatorial Atlantic. There are three main routes available ( Fig. 1): (1) along the continental shelf break of South America via the Guiana Current (Condie, 1991), (2) across the deep ocean translated via North Brazil Current Rings (Goni and Johns, 2001), and (3) initially moving eastward with the North Brazil Current Retroflection before being swept westward by the North Equatorial Current (Johns et al., 2014). To infer the transport pathways of particles we first determined which particles entered the Caribbean Sea. Of these particles, we then determined the maximum eastward longitude the particle traveled and computed a "straightness index" for each particle (the straight-line distance between the start location and where it first entered the Caribbean Sea divided by the sum of length of 5-day track segments along the particle's path). The following criteria were chosen after visually inspecting a subset of trajectories to identify values that would be conservative with respect to assigning false positives to each of the three categories. Particles traveling east of longitude 35°W were placed into the "retroflection" category. Owing to the characteristic "looping" translation motion of objects traveling with rings (Richardson, 2005), particles with relatively low straightness index values are likely to be associated with North Brazil Current Rings whereas particles with high straightness index values are more likely to travel more directly to the Caribbean Sea via the Guiana Current. We assumed particles with a straightness index of less than 0.6 were associated with North Brazil Current Rings and assumed particles with a straightness index of greater than 0.8 traveled with the Guiana Current. Those particles that did not meet these criteria were excluded from the "oceanographic pathways" analysis (mean = 34% of trajectories). We then computed the percentage of particles released within the western Equatorial Atlantic (Equator-10°N, 60-38°W) that followed each of these three routes into the Caribbean Sea for each season, weighted by the seasonal SPC values at the start locations of the particles.

Backtracking
Results obtained here indicate that Sargassum present in the eastern Caribbean has a high probability of originating proximately in the Equatorial Atlantic (Fig. 3). Including a 1% windage factor in the computation of trajectories tended to increase the probability of backtracked particles originating from the Equatorial Atlantic, particularly in the western Caribbean Sea. Simulations including windage also resulted in greater seasonal differences than those only using the original HYCOM velocity fields.
In recent years, there are stark differences in the probability of transport from the Equatorial Atlantic into and across the Caribbean Sea (Fig. 4). Sargassum beaching has been most pronounced July through September (Wang and Hu, 2017). Interannual differences in inputs from the Equatorial Atlantic predicted by Global HYCOM simulations (no-windage) appear consistent with observed regional patterns of Sargassum beaching. In 2011 Sargassum beaching was reported across the eastern Caribbean, and in 2012 beaching was more restricted to the southeastern Caribbean. For both of these years, HYCOM simulations suggest that transport from the Equatorial Atlantic to the western Caribbean would be unlikely ( Fig. 4A and C), whereas simulations including windage predicted higher than usual inputs to the western Caribbean ( Fig. 4B and D). In 2013 no major Sargassum beaching was reported, but in 2014 and 2015, widespread Sargassum beaching was reported throughout the entire Caribbean Wang and Hu, 2016). During the summer of 2014, much greater transport from the Equatorial Atlantic was predicted by HYCOM, especially into the western Caribbean.

Forward-tracking
Transport into the Caribbean from the western Equatorial Atlantic is highly likely (Fig. 5). Including a windage factor of 1% increased the probability of transport into the Caribbean, extending eastward the area where transport was most likely (Fig. 5). Seasonal variation in transport predictions was more apparent when not including windage: the area of the western Equatorial Atlantic with the highest probability (> 80%) of being a potential source for Sargassum in the Caribbean broadened to nearly the 30°W meridian in the winter and shrunk westward to the 55°W meridian in the summer.
Anomalies for the transport predictions into the Caribbean Sea for the spring months (AMJ) of 2011 through 2014 (the season with the greatest average Sargassum coverage in the western Equatorial Atlantic) suggest that ocean circulation processes were fairly similar in those years (Fig. 6) to the long-term mean for spring months in other years (Fig. 5). Albeit in 2011, the year when Sargassum beaching was first reported in the Caribbean, increased transport to the Caribbean from east of longitude 30°W was predicted ( Fig. 6A and B).

Statistical relationships between transport predictions and Sargassum observations
As seen previously (Wang and Hu, 2017), SPC values in the western Equatorial Atlantic were positively correlated with those in the eastern Caribbean Sea (Spearman r = 0.564, p = 0.028, n = 15). In contrast, however, a strong negative relationship between transport predictions without windage and Caribbean SPC values was detected (Spearman r = −0.796, p = 0.00038, n = 15). Transport predictions assuming 1% windage did not correlate strongly with SPC values in the eastern Caribbean Sea (Spearman r = −0.418, p = 0.121, n = 15), though the relationship was also negative. Multiple regression and variance partitioning procedures indicated that 90.8% (89.1%) of the annual variation in eastern Caribbean SPC values could be accounted for by western Equatorial SPC values and no-windage (with windage) transport predictions from that region. Variance partitioning of the statistical models uniquely ascribed 48.9% (67.2%) of the variation in Caribbean SPC values to western Equatorial Atlantic SPC values and 1.8% (0.2%) to transport predictions; the remaining 40.1% (21.7%) of variation in Caribbean SPC values represents interactions between Sargassum occurrence and the no-windage (with windage) transport predictions Our initial expectation was that increased transport from the equatorial Atlantic into the Caribbean Sea would correlate with increased Sargassum in the eastern Caribbean Sea. Detecting the opposite relationship, however, may suggest a possible mechanism responsible for the increased concentration of Sargassum in the Equatorial Atlantic. One hypothesis is that when transport to the Caribbean Sea is reduced within the western Equatorial Atlantic, Sargassum and the nutrients that facilitate its growth might be allowed to concentrate in this region, eventually resulting in a major "bloom." Given that oceanic transport from the western Equatorial Atlantic into the Caribbean Sea is, essentially, inevitable (e.g., synthetic particles (Fig. 5), drifters (Lumpkin and Garzoli, 2005), water masses (Kelly et al., 2000), nutrients and larval fish (Johns et al., 2014), and young sea turtles (Mansfield et al., 2017)), the initial reduction in transport would eventually result in an increase in Sargassum reaching the Caribbean Sea. Consistent with this possibility, we found a strong negative correlation between the annual percentage of synthetic particles transported into the Caribbean Sea and the SPC of the western Equatorial Atlantic in the following year (Spearman r = −0.789, p = 0.0008, n = 14). No such correlation was obtained when using transport predictions that included windage (Spearman r = −0.345, p = 0.227, n = 14).

A Sargassum transport index
The region of highest probability of transport into the Caribbean (Fig. 5) was also a region where Sargassum remote sensing observations were available (Figs. 1 and 2) (Wang and Hu 2016). Weighting transport predictions of particles released in the western tropical Atlantic by SPC values at their starting locations showed there was minimal input into the Caribbean between 2000 and 2010, followed by a dramatic increase in 2011, 2012, and 2014. A significant correlation with Sargassum observations in the eastern Caribbean was found when a windage factor of 1% was assumed (Spearman r = 0.557, p = 0.031, n = 15). However, this relationship was weaker without accounting for windage (Spearman r = 0.421, p = 0.117, n = 15). These mechanistic models show similar results to statistical models (Wang and Hu, 2017) that indicated SPC values in the western Equatorial Atlantic are positively correlated with those in the eastern Caribbean Sea (Spearman r = 0.564, p = 0.028, n = 15).
Regression analyses indicated that the Sargassum transport index can account for much of the variation in annual SPC values in the eastern Caribbean (no-windage scenario: R 2 = 0.798, p < 0.00001, n = 15; 1% windage scenario: R 2 = 0.875, p < 0.000001, n = 15) (Fig. 7). The variance explained by these mechanistic models was similar to the statistical model that considered only western Equatorial Atlantic SPC values (R 2 = 0.889, p < 0.000001, n = 15). The general agreement between the Sargassum transport index, statistical models (see Section 3.3), and observations ( Fig. 7) suggests that annual variation of Sargassum entering the eastern Caribbean Sea is closely related to surface transport processes from the tropical western Atlantic.

Oceanographic pathways
The movement of Sargassum from the Equatorial Atlantic into the Caribbean Sea is predicted to be strongly influenced by the dynamics of the North Brazil Current System (Figs. 1, 5 and 8). Sargassum is in greatest abundance in the western Equatorial Atlantic in spring and summer (Fig. 2). Transport of Sargassum in the spring is most likely associated with the Guiana Current, whereas in the summer, North Brazil Current Rings are the primary pathway. The relative contributions of both pathways, over the course of a year, are similar (Fig. 8). Movement of Sargassum into the Caribbean that was initially retroflected eastward is possible within one year but is relatively rare. Adding a windage factor of 1% increases the probability of particles following each of these three routes to the Caribbean, but the relative contributions are similar (Fig. 8). Sargassum traveling with the Guiana Current in the spring would, on average, arrive to the Caribbean in approximately 2 months. This predicted timing coincides well with the increase in Sargassum observed in the Caribbean in the summer months. In contrast, Sargassum associated with NBC Rings averages closer to a 6month transit time, resulting in an influx of Sargassum that would occur in the fall. Transit times for the retroflection route average approximately 9 months, but this may be an underestimate given that we only tracked particles for one year. Depending on the seasonal timing, particles may go back and forth across the Atlantic between Brazil and Africa several times before either entering the Caribbean or passing north of the Antilles via the North Equatorial Current.

Discussion
Our simulations suggest that Sargassum in the western Equatorial Atlantic would consistently enter the Caribbean Sea. This finding is in agreement with remote sensing data that suggest the recent major Sargassum events likely result from inflow from the Equatorial Atlantic (Wang and Hu 2017). Additionally, our work builds upon and complements the study by Franks et al. (2016). They used Global HYCOM to backtrack 5 particles from each Sargassum beaching site that was reported in 2011 and found all trajectories passed through the Equatorial Atlantic (south of 10°N). Likewise, their analysis of drifter data supports the conclusion that ocean circulation dynamics of the Equatorial Atlantic play a central role in the transport of Sargassum into the Caribbean Sea (Franks et al., 2016). Our work indicates that those conclusions are robust to annual and seasonal variability in ocean circulation. Likewise, the general patterns of transport from the Equatorial Atlantic into the Caribbean are observed irrespective of whether a windage factor of 1% is included in simulations.
Beyond finding agreement in basin-scale patterns of connectivity, including a windage factor of just 1% to surface velocity fields results in considerable changes to transport predictions. When the influence of windage is simulated the mean probability of movement into the western Caribbean is increased (Fig. 3), as is the area in the Equatorial Atlantic over which transport into the Caribbean is most likely (Fig. 5). This effect can be attributed to the strong and consistently westward component of surface winds in the tropical Atlantic (between latitudes 20°S-20°N and longitudes 90°W-10°W) and the northward component of surface winds in the southern half of the equatorial Atlantic (between the equator and 5°N) (Atlas et al., 1996).
Comparisons of our simulations' predictions of transport from the western Equatorial Atlantic to SPC values in the eastern Caribbean suggest that including windage results in better agreement (Fig. 7).
However, it is premature to conclude that this windage factor better accounts for Sargassum movement than the surface layer of HYCOM alone. Examination of transport prediction anomalies across the Caribbean (Fig. 4) suggest that the windage scenario modeled may overestimate Sargassum transport into the western Caribbean. In 2011, Sargassum beaching was only reported east of 71°W (Franks et al.,  2016), yet including 1% windage greatly increased the probability of transport west of this longitude ( Fig. 4A and B). In 2012, Sargassum beaching was restricted to areas in the Caribbean even further to the east (Marechal et al., 2017). The scenario without windage is consistent with this observation, predicting much less transport from the Equatorial Atlantic into the Caribbean than usual (Fig. 4C). In contrast, the scenario including windage predicts greater transport into the western Caribbean than average (Fig. 4D). The 2014 Sargassum beach inundation events occurred throughout the Caribbeanas far west as Mexico Wang and Hu 2017). For that year, the no-windage scenario predicts much greater inputs into the western Caribbean than is typical (Fig. 4G), whereas the windage scenario predicts lower probabilities of transport into the western Caribbean in 2014 than in 2011 (Fig. 4H).
At present, we only speculate as to why including windage brings forward-tracking simulations of Sargassum transport into better agreement with SPC values in the eastern Caribbean (Fig. 7), but the nowindage scenario of backtracking appears to better account for variation in beaching events in the western Caribbean (Fig. 4). Perhaps Sargassum becomes less buoyant as it ages and moves east to west across the Atlantic and into the Caribbean and is less influenced by wind. The discrepancy between SPC values and transport predictions may be subject to other biological processes (e.g., growth) and biologicalphysical interactions that we do not model here. Nonetheless, the sensitivity of our predictions to a windage factor of just 1% shows that further research to understand the movement of Sargassum at and near the ocean surface is needed. We anticipate that numerical and in situ experiments to improve estimates of windage and other forcing terms acting on Sargassum mats will greatly increase the ability to simulate its movement and predict its distribution (Breivik et al., 2011).
Indeed, resolving the effect of processes at the air-sea boundary, which are not accounted for in most ocean circulation models, is an important area of investigation in marine ecology (Monismith and Fong, 2004;Werner et al., 2007). Such efforts may yield important insight into a number of ecological and evolutionary processes, including rare occurrences of transoceanic colonization of marine plants (Smith et al., 2018), meteorological and oceanographic drivers of population connectivity (Monzón-Argüello et al., 2012), the role of swimming behavior on the movements of small marine animals (Putman et al., 2016;Mansfield et al., 2017), and the distribution of anthropogenic debris in the ocean (van Sebille et al., 2015). Our results represent an initial test of the sensitivity of transport estimates to one such process, windagedirect, linear transfer of momentum from the wind to Sargassum. We anticipate that studies that test the sensitivity of transport estimates to other processes, including the Stommel shear (the upper part of the "Ekman spiral", which is latitude dependent), residual transport due to waves (Stokes drift), and possible variations in the appropriate drag coefficient will help guide and prioritize empirical research on the life cycle, buoyancy characteristics, and aggregation patterns of Sargassum.
Regardless of differences in transport predictions between windage scenarios, however, our simulations imply that Sargassum transport into the Caribbean Sea is closely tied to the North Brazil Current System (Fig. 8). Further examination of the circulation processes associated with this region (Fig. 1) may yield increased predictive power regarding the timing of Sargassum entry into the Caribbean Sea. Variability in beach inundation events are likely primarily the result of high variability in Sargassum occurrence (Figs. 2 and 7), but variability in ocean circulation appears to also play a role. Indeed, statistical analyses indicate that Sargassum abundance in the Caribbean Sea may be linked to surface transport processes in complex ways. Our results suggest that reduced transport to the Caribbean Sea correlates with increased Sargassum coverage in the Equatorial Atlantic, resulting in an eventual increase in Sargassum coverage within the Caribbean Sea. For instance, in 2011 there was 180% greater Sargassum coverage in the western Equatorial Atlantic, south of latitude 10°N, than the 15-year average. This translated into 255% greater Sargassum coverage observed in the eastern Caribbean west of 60°W relative to previously measured years. In 2012, there was even more Sargassum in the western Equatorial Atlantic, 255% greater coverage than average. Yet, this only translated into a 100% increase in Sargassum coverage in the eastern Caribbean relative to the average. A likely cause for this difference is the different states of the North Brazil Current (Fig. 9). In 2011, annual eastward flow of the NBC retroflection was 10% lower than average (12.6 Sverdrups, Sv = 10 6 m 3 /s), whereas in 2012 eastward flow increased by 24% (17.4 Sv). Thus, oceanic conditions in 2011 likely favored northwestward transport of Sargassum into the Caribbean, but in 2012 much of the Sargassum was likely carried eastward toward the central Equatorial Atlantic. This may also explain the hiatus in Sargassum entry into the Caribbean in 2013, its limited presence in the western Equatorial Atlantic, and its eventual return in 2014, as Sargassum was advected back into this region via the North and South Equatorial Currents.
Such an influence of ocean circulation processes on the entry of Sargassum into the Caribbean suggests that pairing remote sensing data with particle tracking simulations within ocean circulation models could become an effective means to predict future inundations. Indeed, our analyses indicate that much of the annual variation in Sargassum abundance (as observed from satellites) in the eastern Caribbean can be accounted for by scaling particle transport predictions to Sargassum coverage in the western Equatorial Atlantic. That better than 79% of the annual variation in Sargassum abundance in the eastern Caribbean Sea can be accounted for by Sargassum coverage and particle tracking simulations in the western Equatorial Atlantic, suggests that further refining this paired approach may allow for predictions and forecasts of Sargassum beaching events. Moreover, assessing the differences in the transport time for specific oceanic pathways from the Equatorial Atlantic into the Caribbean (Fig. 8b) suggests that timing of Sargassum beaching might be predictable well in advance of its occurrence. These results show the importance of monitoring North Brazil Current System dynamics in relation to Sargassum distribution (e.g., Figs. 2 and 8), which could allow the development of an early-warning system that estimates the probability of Sargassum reaching specific areas of the Caribbean Sea within a certain time (e.g., Fig. 3).
Of course, Sargassum entering the eastern Caribbean might not be exclusively of western Equatorial Atlantic origin. The floating algae may grow as it moves towards the Caribbean and inputs from other oceanic regions may add to it. Expanding the analyses pairing remotesensing and particle-tracking eastward into the Equatorial Atlantic, as well as into the Gulf of Mexico and Sargasso Sea (other areas of known Sargassum accumulation (Gower et al., 2013)) will provide a more comprehensive understanding of the dynamics of Sargassum throughout the Atlantic. Likewise, field campaigns are needed to obtain data on the biophysical properties of Sargassum (e.g., depth, buoyancy, growth rates, sinking rates) to parameterize numerical experiments and validate remote-sensing observations. Such work will provide a launching point to understand the ecosystem-level impacts of pelagic Sargassum as its movements may play an important role in the population dynamics of numerous species that depend upon it as habitat (Coston-Clements et al., 1991;Wells and Rooker 2004;Mansfield and Putman 2013).  (Fig. 1). Darker bars show results using the surface layer of HYCOM, lighter bars show results that include 1% windage from the NCEP Reanalysis.
Error bars indicate the range (maximum and minimum) of average seasonal values. The large error bars reflect the strong seasonal change in circulation patterns in this region, where the Guiana Current is the predominant pathway during January-June and NBC Rings are the predominant pathway during July-December (Supplemental Fig. 2). (B) For 2011-2014 (years when substantial Sargassum was observed (Fig. 7)), we computed the mean number of days to reach the Caribbean for particles released east of 50°W (the region where Sargassum is greatest (Fig. 2)) by each transport route and season. Error bars represent the 95% confidence interval of the mean. Other conventions and color scheme as in (A). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Fig. 9. Percent difference from annual mean values of monthly averaged Sargassum percent coverage across the eastern Caribbean Sea (10-22°N, 58-63°W) (gray), monthly averaged Sargassum percent coverage across the western Equatorial Atlantic (Equator-10°N, 60-38°W) (black), and retroflected flow (in Sv) from the North Brazil Current (white), based on sea height anomaly fields from satellite altimetry and climatological data (www.aoml.noaa.gov/phod/ altimetery/cvar/nbc/transport.php). For retroflected flow, positive values indicate an increase in eastward transport.