In recent years, holopelagic sargassum macroalgae (seaweed) has been increasingly recorded far beyond the Sargasso Sea – named after the prevalent seaweed – in the western subtropical gyre in the North Atlantic (Fidai et al., 2020). Large quantities of sargassum notably appeared across the tropical Atlantic during the summer of 2011 (Franks et al., 2012). Proliferation of sargassum in most years since 2011 has been consequential, as it accumulates and decomposes on beaches, causing damage to ecosystems (van Tussenbroek et al., 2017) and adversely impacting human health and livelihoods (UNEP, 2018). A wide range of uses for sargassum has meanwhile been identified, providing business opportunities (Desrochers et al., 2020). The unpredictable supply of sargassum is however identified as one of five challenges that face stakeholders concerned with potential opportunities for sargassum reuse and valorisation (Oxenford et al., 2021). For resilient adaptation to this environmental challenge, we therefore need to better understand and predict the annual influxes of sargassum to beaches throughout the tropical Atlantic.

Like the Sargasso Sea, pelagic sargassum in the Caribbean comprises two species, Sargassum natans and S. fluitans, including several morphotypes or forms. Early inundations, however, were dominated by a form considered rare in the Sargasso Sea, Sargassum natans VIII (Schell et al., 2015; García-Sánchez et al., 2020). S. fluitans and S. natans have also been sampled along the coast of Ghana since 2009 and not being indigenous to the Gulf of Guinea, were considered to have originated in the Sargasso Sea (Addico and deGraft-Johnson, 2016). These species of sargassum were also identified further to the east, off the coast of Nigeria, in 2012 (Oyesiku and Egunyomi, 2014). The basin-scale extent of these species has led to recognition of the ‘great Atlantic Sargassum belt’ (Wang et al., 2019).

Hydrometeorological processes such as currents and wind not only drive oceanic movement of sargassum but also influence beaching along coastlines. These wide-scale oceanic processes however give way to local tidal, coastal currents, and prevailing wind and wave conditions, that further determine beaching events. In the Caribbean during 2019, for example, beached sargassum was mapped by the Mona Geoinformatics Institute using remote sensing and field validation techniques around the coastline of Jamaica (Anderson et al., 2021). Results revealed those particular shorelines to experience inundation, with much of the east and south coasts seeing some of the highest frequency of sargassum beaching that year (see Figure 1).

Trade winds from the north-east are attenuated by local changes in terrain such as high mountain ranges to the east of Jamaica. Offshore waves undergo refraction and significant energy dissipation on encountering wide, gentle-sloping shelf areas south of the island but far less so to the north of the island where the shelf is predominantly narrow and steep. Combined with a local diurnal wind regime, microtidal conditions and seasonal swells, various coasts are exposed to strong wind and wave conditions at different times of the year (Maxam and Webber, 2010).

While these hydrometeorological conditions will account for some of the beaching observed, some areas were shown on the map to experience significant inundation in both relatively active and calm wave environments. The shoreline of Clarendon (the central coast of south Jamaica – see Figure 1) notably exhibits this paradox. It is expected therefore that local shelf and shoreline changes will provide complexities in coastal geomorphology that will be significant in driving finer-scale processes and intensity of sargassum beaching in specific areas.

The west African coastline, from Senegal to Nigeria is likewise subject to highly variable quantities of beached sargassum, with notable beaching events recorded along the coast of Ghana, especially the western coast (Rhein-Knudsen et al., 2017), in late June of 2020 – see Figures 2A,B. This has been observed along the shore of many fishing villages (Ackah-Baidoo, 2013). Observations indicate that landings are prevalent on gently sloping beaches, especially along the western coast of Ghana. The focus here on large-scale processes should be appreciated in the context of local processes in our oceanographic regions of interest, around Jamaica and off Ghana.

Complementary to monitoring for sargassum at beaches, throughout the region, is remote sensing. Since 2011, innovative satellite remote sensing analysis has provided strong evidence for initial growth and establishment of sargassum in the equatorial Atlantic (Gower et al., 2013; Franks et al., 2016; Wang and Hu, 2016). Based on an ‘Alternative Floating Algae Index’ (AFAI), ‘FA_density’ maps are now provided daily, as the mean of the previous 7 days. The AFAI is an ocean colour index, obtained by utilising the spectral reflectance in red, near infrared and shortwave infrared of floating vegetation per 1-km pixel in images from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument that is currently onboard the Terra and Aqua satellite, translated as a fractional (or %) area covered by floating algae (Hu, 2009). Across the “Central Atlantic Region,” or “Central West Atlantic” (CWA), encompassing 38-63°W, 0-22°N (see Figure 3), the FA_density data can be summed to obtain a total area of sargassum, daily since February 2000, shown here in Figure 3. The inception of summer sargassum blooms in 2011 is clear, with considerable seasonal variability unfolding over subsequent years. Most extensive areas of sargassum developed through 2015, 2018, 2019 and 2020.

Recent proliferation of sargassum across the tropical Atlantic has attracted considerable scientific attention in the marine science community. A series of studies have established that increased presence of sargassum in the Caribbean can be traced back to the equatorial Atlantic, several months earlier (Franks et al., 2016; Brooks et al., 2018; Wang et al., 2019). Subsequent drift calculations based on currents from numerical models and a nominal 1% ‘windage’, provided compelling evidence for the drift of this sargassum from the equatorial Atlantic to the eastern Caribbean (Putman et al., 2018).

Johns et al. (2020) provide an extensive review and update on our understanding of the proliferation of sargassum since 2011. A basin-scale, tropical-extratropical perspective is now established. It is proposed that a historic climate anomaly in 2010 – an exceptionally negative phase of the North Atlantic Oscillation (weak Icelandic low and Azores high pressure centres) – drove sargassum from the western subtropics (the Sargasso Sea) to the northeast, as far as Gibraltar. Swept south in the Canary Current, the hypothesis is that the subsequent appearance of sargassum in the eastern central tropical Atlantic was the trigger for spring growth and the first Caribbean inundation of summer 2011. The sargassum has subsequently been recirculating in the North Equatorial Recirculation Region (NERR) (Franks et al., 2014), forming a belt across the northern tropical Atlantic that follows seasonal migration of the Intertropical Convergence Zone (ITCZ). Seasonal releases from the NERR, in most years since 2011, supply sargassum that subsequently beaches throughout the Lesser Antilles and the Caribbean. Drifter data provide evidence that the NERR also supplies the sargassum that beaches along West Africa (Johnson et al., 2020), consistent with basin-wide extent of the North Equatorial Counter Current in the approximate zone 5–10°N, during the second half of the year. Sargassum may thus be conveyed eastward, joining the Guinea Current to be delivered to west African coastlines from Senegal to Nigeria.

The challenge of predicting sargassum has grown alongside studies of how and why these blooms are developing. On short timescales, of up to several days, satellite imagery and operational 7-day ocean current forecasts are used to predict beaching events (e.g., Maréchal et al., 2017). On much longer timescales, other approaches must be adopted, involving the use satellite observations (Wang and Hu, 2017), drifters (Johnson et al., 2020) and models (Jouanno et al., 2020). Berline et al. (2020) use satellite data to initialise sargassum distributions in an ocean model, subsequently tracked forwards to provide a seasonal hindcast for 2017.

We adopt a similar approach to Berline et al. (2020), extended to forecasting with uncertainty. As part of the interdisciplinary project, Teleconnected SARgassum risks across the Atlantic: building capacity for TRansformational Adaptation in the Caribbean and West Africa (SARTRAC), remote sensing and modelling are being used to develop a wider early warning system for the arrival of sargassum at specific locations on both sides of the tropical Atlantic: Caribbean and West Africa. For fast and practical capacity building and operational implementation, we have developed a forecast system based on data available in the public domain, and hindcast winds and currents obtained with an ocean model of spatial resolution sufficiently high to simulate realistic movement and dispersal of sargassum. We thus address the basin-scale movement of sargassum, subject to complex and chaotic currents, and highly variable winds, at seasonal timescale. The effects on drifting sargassum of growth, mortality and sinking are retrospectively calculated, to refine the predicted quantities of sargassum at specific locations and times. Here, we demonstrate the application of SARTRAC-EFS to years of extensive beach inundation at two different locations, Jamaica and Ghana, exploring the sensitivity of forecasts to variable ocean and climate states, windage, growth and decay.

The paper is organised as follows. In section “MATERIALS AND METHODS,” we outline the use of satellite observations, ocean model hindcasts, particle tracking and basic algorithms for growth mortality and sinking. In section “RESULTS,” with a focus on the Caribbean and Jamaica, we present the basis for initializing sargassum distributions and results of several test predictions, of sargassum inundations in four selected years, 2015, 2018, 2019, and 2020. We further present exploratory predictions for the west African region, and Ghana in particular, although this is currently restricted by more limited satellite observations in the eastern tropical Atlantic. In section “DISCUSSION,” we discuss our proposed framework for seasonal forecasting alongside several quite different methods proposed by others, emphasizing a complementary approach to long-range prediction of annual sargassum conditions across the tropical Atlantic.

Predictions are initialised and also evaluated with satellite observations of floating vegetation through the use of FA_density data. Central to predictions is a high-resolution ocean model hindcast, providing datasets used to efficiently track large numbers of particles that are representative of drifting sargassum mats. Using currents and winds from a hindcast spanning 1988-2010, we obtain 23-member ensembles per forecast, amounting to a basic Ensemble Forecast System (EFS). Hence, we refer to the system as SARTRAC-EFS. We describe in turn the datasets and methods that underpin SARTRAC-EFS, concluding this section with a summary of SARTRAC-EFS and brief comments on practical aspects.

Given the availability of FA_density data coverage, we focus on the Caribbean region and the waters around Jamaica in particular, first illustrating the initialisation of particles in proportion to the observed sargassum across a region encompassing the Jamaican, Eastern Caribbean and Central Atlantic ‘boxes’ (see Figure 4). In a series of 180-day predictions, we consider four recent years when sargassum was prevalent around Jamaica and beached in large quantities around much of the island. With evidence of a role for the Atlantic Meridional Mode of climate variability in two major inundations (2015 and 2018), this influence is further examined in selected ensemble members. Despite the constraints of satellite observation in the eastern tropical Atlantic, a more limited set of predictions are undertaken for this region, indicating wider prospects for seasonal predictions of the arrival of sargassum in coastal waters, across the tropical Atlantic. In two of the figures shown below (Figures 8, 9), we randomly select an example hindcast year, to illustrate time-slice distributions of sargassum areal fraction predicted by SARTRAC-EFS; these selected years are broadly representative of the full 23-year ensemble.

We have used a range of available datasets and methods to develop SARTRAC-EFS, a prototype ensemble forecast system to predict the seasonal arrival of sargassum in selected coastal waters, here around Jamaica and off Ghana. In our hybrid approach, use satellite observations to initialise the location of virtual particles in proportion to the sargassum, across upstream sources. The particles are tracked forwards in time, subject to a plausible range of winds and surface currents from a 23-year hindcast spanning 1988-2010. Forecasts are obtained as time series of the mean and standard deviation in the total area of sargassum, inferred from the aggregation of particles in selected areas. One limitation of this study concerns the unavailability of synoptic wind and current data for more recent years – specifically 2015 and 2018-20 – although we consider that the available hindcast samples a range of ocean and climate variability that is suitably representative, for ensemble forecasting at seasonal timescale.

Tracking particles subject to surface currents under varying degrees of windage (here 0% or 1%) follows the pioneering work of Putman et al. (2018), and we reflect here on specific issues raised by that study in the context of our findings. In broad agreement with Putman et al. (2018), we find that particles initialised in proportion to sargassum in the Central Atlantic region move generally westward to arrive in the Caribbean on a seasonal timescale. Some of the variability between ensemble members of SARTRAC-EFS is likely related to differences in the balance of direct and indirect connectivity between the equatorial Atlantic and the Caribbean that are related to changes in the North Brazil Current system, noting implicit interannual variability in five identified transport routes (Figure 10 in Putman et al., 2018). A variable role of windage across the Caribbean, as suggested by Putman et al. (2018), may explain why windage both improves and degrades forecasts, in different years. Finally, the close correspondence between satellite observations of sargassum coverage in the equatorial Atlantic and subsequent inundations in the eastern Caribbean is consistent with predicted arrivals around Jamaica that are roughly proportional to initialised distributions, in each of our four study years. Other points raised in Putman et al. (2018), regarding the need to more fully account for physical and biological processes, are covered later in our discussion.

This expedient use of model hindcast data and ensemble statistics enables us to account for the uncertainty inherent in winds and currents on timescales from synoptic to seasonal. In this respect, SARTRAC-EFS is complementary to alternative prediction systems, based on statistics of satellite observations (Wang and Hu, 2017), climatological drift observations (Johnson et al., 2020), and a deterministic, physical-biogeochemical model (Jouanno et al., 2020). Each method has advantages and caveats. To the extent that satellite data is an accurate measure of sargassum abundance at basin-scale, past satellite observations can guide the development of a statistical model, but without process-level detail. In situ drift observations, of undrogued drifters, provide the most authentic dynamical description of how sargassum moves through the region, but relies on climatology as a “best guess” of where sargassum will drift in any year, and drifters are known to under-sample some of the flow. Fully coupled models provide a process-rich representation of physical and biogeochemical processes, but the realism may be compromised by a large number of highly uncertain parameters, and multiple interactions can lead to large errors. The pragmatic approach of SARTRAC-EFS blends aspects of these established approaches, and as such we share some strengths and weaknesses. An improved assessment of sargassum developing on seasonal timescales may be best informed by a suite of predictions, accommodating each approach.

Testing SARTRAC-EFS for four recent years, when sargassum proliferated throughout much of the tropical Atlantic, we show that the seasonal forecast in a given year should ideally account for appropriate windage, growth, mortality, sinking and regional climate variability. Improved agreement with observations may be gained in several ways:

Our forecasts are ultimately limited by the availability and quality of satellite derived distribution of sargassum to both initialize and evaluate predictions. In particular, extension of daily FA_density images substantially to the east of 38°W would potentially extend the Jamaica forecast time horizon well beyond 120–180 days. Likewise, westward and southward extensions of the satellite data available for forecasts off west Africa are necessary for Ghana forecasts of up to 90 days. There are now prospects for wider satellite monitoring of sargassum via the Sentinel-2 and Sentinel-3 satellites, with free and open access to Sentinel-3 products via the Copernicus Online Data Access (CODA) Web Service. Météo France, the French national meteorological service, is now using these satellite data to develop sargassum forecasts for the French Caribbean islands. More critically, major improvements in the remote sensing of sargassum in the eastern tropics is necessary to validate seasonal predictions off west Africa. With both Sentinel 2 and Sentinel 3 satellites in orbit, there is now an opportunity to improve temporal coverage and probability of obtaining cloud free images. Furthermore, as indicated in earlier studies (e.g., Gower et al., 2013), the availability of red edge spectral bands in both sensors could be highly useful in detecting the sargassum optical signature and, in turn, accuracy in detection.

On a fundamental level, there is also an urgent need for better understanding of the growth and loss of sargassum in the tropical Atlantic, with existing parameterisations and parameters based on analysis of sargassum from the Sargasso Sea. In the meantime, we highlight scope for improvement of our framework by incorporating more processes in the offline growth algorithm. Different nutrient sources may determine the annual growth rates (Oviatt et al., 2019), depending on pathways followed by sargassum mats in a given year. This growth factor may be incorporated into our seasonal predictions by associating nutrient levels with physical water properties, temperature and salinity.

Regardless of nutrient limitations, sargassum growth is a function of not just temperature – as included here – but also salinity. In laboratory experiments, the growth rate for S. natans increases substantially, from ∼0.1 to ∼0.5 doubling day^–1, as salinity is increased from 24 to 36 psu (Hanisak and Samuel, 1987). Sargassum drifting with the North Brazil Current may encounter salinity variations in this range, associated with the highly variable Amazon plume. Including a functional dependence on salinity, saved along with temperature along each trajectory, therefore offers the prospect for further refinement of predictions. We can furthermore accommodate variable light levels, in relation to local irradiance. Finally, the sinking rate may be varied in relation to local wind conditions, by including Stokes drift in the trajectory calculations.

In addition to in situ processes, seasonal predictions of sargassum movement must account for the interannual variability of winds and currents across the tropical Atlantic. In addition to the Atlantic Meridional Mode and Atlantic Niño that we have identified here, other modes of interannual climate variability may be influential in determining the transport, growth and loss of sargassum in any given year. The North Atlantic Oscillation appears to have exerted a remote influence on the tropical Atlantic in 2010/11 (Johns et al., 2020), but the region is also subject to some influence from the El Niño (Saravanan and Chang, 2000; Xie and Carton, 2004).

Interplay between modes of natural variability and anthropogenic climate change will likely determine the future prevalence of sargassum. Over 2031-50, regional expression of projected global-mean sea surface temperature rises in the range 0.64-0.95°C, (Table CB1.1 in IPCC, 2019), will impact both the growth and mortality of sargassum. Meanwhile, sea surface salinity is subject to ‘pattern amplification’ under an intensifying hydrological cycle, now evident over a long timescale (Gould and Cunningham, 2021). Given the multiple dependencies of sargassum growth and mortality on temperature and salinity, the overall consequence of climate change and variability for sargassum are highly uncertain on a multidecadal timescale.

At a practical level, we finally stress the need to couple the open ocean forecasting systems such as SARTRAC-EFS with coastal nearshore prediction models/frameworks. Only then can the forecasts serve all sector stakeholders, such as those managing sargassum beaching events to protect tourism, coastal livelihoods and health, providing advance warning for fisheries sector and businesses using or planning to use sargassum.

Further experiments and analysis to determine the relative importance of different modes of climate variability, on the prevalence and drift of sargassum, are ongoing. Meanwhile, forecasts with SARTRAC-EFS for summer of 2021 indicate that substantial quantities of sargassum will once again be present north and south of Jamaica, providing useful early warning of beaching events throughout May-August, with peak quantities most likely during June.

Publicly available datasets were analysed in this study. This data can be found here: The satellite datasets analysed for this study can be found at the website of the Optical Oceanography Laboratory at the University of South Florida (https://optics.marine.usf.edu), selecting ‘Satellite Data Products’, and further selecting: ‘E. Caribbean’ and ‘Jamaica’ from ‘Caribbean’; ‘Central Atlantic’ from ‘South America’; ‘Gulf of Guinea’ from ‘West Africa.’ To download all daily AFAI images for the ‘E. Caribbean’, ‘Jamaica’ and ‘Central Atlantic’ regions, we used Matlab scripts that are archived on Zenodo (published with a link to this article, when a DOI is available). Analyzed satellite data, example tracking data, along with past seasonal prediction and prototype forecast data, will be likewise archived on Zenodo, on publication. The hindcast ORCA12 model datasets used in SARTRAC-EFS (surface currents, temperatures, salinities and winds) are archived on a filesystem at the National Oceanography Centre, Southampton, and are available on request. The AMM index data used here can be found at https://www.aos.wisc.edu/_dvimont/MModes/Data.html, while the Atlantic Niño index was computed from NOAA Optimum Interpolation SST (v2) monthly fields, found at https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.html.

RM undertook the calculations and developed the manuscript. KA, P-NJ-Q, HO, AM, and RA provided input on regional contexts in the Caribbean and West Africa and including associated analysis and text. NS provided climate index calculations and accompanying text. JD contributed to text on remote sensing. ET and HO suggested improvements that improved clarity of the manuscript throughout. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.