Dispersal models alert on the risk of non- native species introduction by Ballast water in protected areas from the Western Antarctic Peninsula

Aim: The Western Antarctic Peninsula is challenged by climate change and increasing maritime traffic that together facilitate the introduction of marine non- native species from warmer regions neighbouring the Southern Ocean. Ballast water exchange has been frequently reported as an introduction vector. This study uses a Lagrangian approach to model the passive drift of virtual propagules departing from Ballast water hypothetic exchange zones, at contrasting distances from the coasts. Location: Western Antarctic Peninsula. Methods: Virtual propagules were released over the 2008– 2016 period and at three distances from the nearest coasts: 200 (convention for the management of Ballast Water, 2004), 50 or 11 nautical miles (NM). Results: Results show that exchanging Ballast water at 200 NM considerably reduces the arrival of propagules in proposed marine protected areas of the western side of the Antarctic Peninsula. On the eastern side, propagules can reach north- eastern marine protected areas within a few days due to strong currents for all tested scenarios. Seasonal and yearly variations indicate that exceptional climate events could influence the trajectory of particles in the region. Ballast water should be exchanged at least 200 NM offshore on the western side of the Antarctic Peninsula and avoided on the eastern side to limit particle arrival in proposed marine protected areas. Focusing on Deception Island, our results suggested that the Patagonian crab ( Halicarcinus planatus ) observed in 2010 could have been introduced in case of Ballast water exchange at 50 NM or less from the coast. Main conclusions: This study highlights the importance of respecting Ballast water exchange convention to limit the risk of non- native species introduction. Ballast water exchange should be operated at least at 200 NM from the coasts, which further limits particle arrival in shallow water areas. This is especially important in the context of a more visited and warmer Southern Ocean.

The Southern Ocean, here, is defined as water masses bounded by the Antarctic continent to the south and the Polar Front to the north (Rintoul, 2009). The Polar Front is the most significant of a series of circumpolar marine fronts associated with the eastward-flowing jets of the Antarctic Circumpolar Current (Orsi et al., 1995). Both the Polar Front and the Antarctic Circumpolar Current form physical barriers preventing Antarctic surface water exchanges between the Southern Ocean and northern ocean areas (Aronson et al., 2007;Griffiths, 2010;Sanches et al., 2016), hence blocking the dispersal of most marine organisms (Convey & Peck, 2019;Peck et al., 2014). As a result of the prevalence of such important marine fronts, combined with strong currents and the remoteness from other landmasses, a unique Southern Ocean marine diversity has been shaped (Barnes & Clarke, 2011;Clarke et al., 2005;Crame, 1999;Lawver et al., 1992).
This unique Southern Ocean marine diversity is however currently challenged by the multiple effects of climate change (Ansorge et al., 2014;Henley et al., 2019). Antarctic coastal marine ecosystems are notoriously sensitive as many shallow-water species have limited resilience abilities and limited southward migration capacities towards more suitable areas (Stenni et al., 2017;Cárdenas et al., 2018;Gutt et al., 2018). Direct and indirect impacts of climate change are expected to alter the structure and functions of these marine ecosystems, leading to species distribution shifts, local extinctions, and favourable conditions for colonization by introduced non-native species (e.g. warmer temperatures, Bender et al., 2016;Hughes & Convey, 2010). Anthropogenic impacts induced by fisheries, tourism and research activities have also been shown to facilitate the transport and introduction of non-native organisms through ship hull fouling and Ballast water exchanges (Hughes & Ashton, 2017;Lee & Chown, 2009c;Lewis et al., 2005).

The Commission for the Conservation of Antarctic Marine
Living Resources (CCAMLR) was created in 1982 to manage marine communities in response to an increasing commercial interest in Antarctic fisheries (such as krill and fish resources) (https://www. ccamlr.org/en/organ isati on/home-page, accessed October 2020).
The Scientific Committee and Commission of CCAMLR yearly review and rule on new marine protected areas projects proposed by national experts. To date, two Antarctic marine protected areas have been initiated and include waters off the South Orkney Islands (in 2009) and within the Ross Sea region (in 2016). Antarctic mineral and core resources are not exploited yet (Westermeyer, 2020) but commercial fishing and tourism are on the rise, in particular along the west coast of the Western Antarctic Peninsula (Bender et al., 2016;Lee & Chown, 2009a;McCarthy et al., 2019). During the last five austral summers (2014)(2015)(2016)(2017)(2018)(2019), the yearly number of tourists visiting Antarctica has increased from 36,700 to 55,400 (IAATO & International Association of Antarctica Tour Operators, 2019).
Ship hull fouling and Ballast water release have been reckoned as major vectors of non-native species dispersal and introduction to Antarctic coastal waters (Barnes, 2002;Chan et al., 2015;Hughes et al., 2020;Lavoie et al., 1999;Lewis et al., 2003Lewis et al., , 2005. Ballast water tanks are filled at ships' departure ports in South America to safely navigate across the Drake Passage to Antarctica. Fishing vessels progressively discharge most of their Ballast water as it is replaced by their catch. Cruise ships typically discharge Ballast water to travel faster and regularly take up new water to replace the volume left by fuel consumption (Hughes et al., 2020).
Awaiting for this application, the convention itself (Regulation B-4, adopted in 2004)  In the present study, a Lagrangian model was developed to simulate the drift of virtual particles as they are transported along the water masses. Such an approach has already been developed in other regions of the world (Edwards et al., 2006;Siegel et al., 2003), but not along the Western Antarctic Peninsula. The model calculates particle trajectories (identified here as potential propagules) according to different point locations where Ballast water is exchanged.
This was illustrated by a study on the Patagonian crab, Halicarcinus planatus (Fabricius, 1775), reported in Deception Island (nearby Western Antarctic Peninsula coasts) in 2010. The potential impact of Ballast water exchange on the introduction of non-native species in Antarctic coastal waters was analysed through pluri-annual and multi-seasonal time scales. A map of recommended areas for Ballast water exchange is proposed as a tool to support good practices for Ballast water exchange and for conservation purposes.

| Study area
The study area is enclosed by the strong eastward flowing Antarctic Circumpolar Current (Appendix S1) and includes the Scotia Arc region, the Antarctic Peninsula, and the Weddell Sea, as they concentrate most of the maritime traffic between Antarctica and southern South America and therefore the highest risk of non-native species introduction (Lynch et al., 2010;McCarthy et al., 2019).

| Lagrangian model principle and hydrodynamic settings
In this study, we used a Lagrangian particle model, which combines oceanographic information (e.g. bathymetry, current direction and speed) forced by atmospheric factors (temperatures, winds, atmospheric pressure) (Huthnance, 1991;Robinson & Golnaraghi, 1994) with biological features (e.g. organisms' size, development rate, buoyancy, Van Sebille et al., 2018). The model used in this study is based on the model described in Dulière et al. (2013) and made available as a module of the free and open-source aquatic modelling system COHERENS v2 (Luyten, 2011). This system has already been used to study, among others, oil spill dispersal (Legrand & Dulière, 2014), jellyfish drift  and the movement of harbour porpoises (Haelters et al., 2015). Particles are transported under advective and diffusive processes in three dimensions. The classical fourth-order Runge-Kutta method is used to estimate horizontal transport. The diffusive velocities are obtained from random walk theory with constant horizontal and vertical diffusion coefficients of 10 and 0.0001 m 2 s −1 , respectively. The same diffusion coefficient values are used as in Young et al. (2014) and are equivalent to values observed in the Southern Ocean (empirical values or commonly accepted by modellers; Sheen et al., 2013;Watson et al., 2013). A bouncing condition is used for particles reaching the sea surface or seabed, and particles that leave the model domain through the ocean open boundary are assumed to have left the region. Stranding is not allowed, because it was considered that a larvae should complete the overall development cycle until metamorphosis to survive (Stanwell-Smith et al., 1999). Thus, when a particle reaches a dry cell, its position is set to its previous position at sea. The Lagrangian module is used off-line with a computation time step of 5 min. To ensure the general purpose of this study, the model has been set up with no specific organismal behaviour (i.e. swimming or tidal or diurnal vertical migration). Particles are assumed to move along with water masses (i.e. no buoyancy effect), with an assumed initial position at 10 m depth in the water column. A sensitivity analysis was concentration, thickness and velocity), seawater potential temperature, seawater salinity and ocean mixed layer thickness. These datasets have been generated with NEMO 3.1 and LIM2 EVP models forced with 3-hourly atmospheric forcing from ECMWF (European Centre for Medium-Range Weather Forecasts, https://www.ecmwf. int/). Daily averaged model products are made available after interpolation from the native model grid to a global standard Arakawa C grid of 1/12° horizontal resolution and 50 fixed vertical levels (from 0 to 5000 m). The quality of the Global high-resolution products has been assessed in Lellouche et al. (2019). 3D vertical ocean currents are estimated from the divergence in the horizontal velocity from the PHY_001_024 forcing fields, assuming null surface and bottom vertical velocity.
The model grid was built from a sub-sample of the global grid of the hydrodynamic forcing field from latitude 45°S down to the South Pole. The horizontal resolution of 1/12° (~8 km) was kept and the 50 vertical levels have been adapted to 50 sigma levels for the COHERENS system. The Lagrangian particle model has been previously validated in Dulière et al. (2013), Legrand and Dulière (2014) and a quality analysis of the hydrodynamic forcing is provided in Lellouche et al. (2019).
Raw results and R codes used to process them are publicly provided at https://zenodo.org/recor d/55882 20#.YXEfq R3godU (folder "Ballast water paper"). For each release location, particles were released daily over a 9-year period (from 2008 to 2016) to account for seasonal and inter-annual variabilities. For technical reasons, particles have been released every six grid-cell pixel in latitude (every 1/2°) and every four grid-cell pixel in longitude (every 1/3°), in areas where depth reached at least 200 m (to enable the sensitivity analysis previously described). The number of release locations ranges from 8 (for the South Georgia Islands zone Rz.5 in scenario 11 NM) to 224 (for the Western Antarctic Peninsula zone Rz.1 in scenario 11 NM) release points among the release zones and scenarios. Altogether, it is more than 4.5 million particle trajectories that have been studied in the model. A 6-month duration of the drift was chosen. This matches to the longest duration of larval development periods (corresponding to Southern Ocean species). Larvae should complete this drifting period in the water column before starting metamorphosis and settling down on the seabed (Stanwell-Smith & Clarke, 1998;Stanwell-Smith et al., 1999;White, 1998). and for each Ballast water release scenario, to generate maps of dispersal patterns. Results for averaged years and seasons are first provided to describe the overall dispersal patterns of particles drift.

| Particles trajectory and age: statistical comparisons
Then, inter-annual and seasonal variabilities are described. Due to the different number of released particles among release areas and scenarios, a scaling correction has been applied for statistical analyses (by dividing the predicted densities by the number of particles present on corresponding release lines-zones where particles were launched, i.e. at 200, 50 or 10 NM; which gives a "weighted number of particles").

| Focus on Deception Island
In February 2010, a living and mature female of the brachyuran  Figure 3) is an opportunistic feeder (Boschi et al., 1969) that is commonly found sheltered below intertidal and subtidal rocks (Richer de Forges, 1977;Vinuesa & Ferrari, 2008).
Halicarcinus planatus has a high dispersal potential, with the release of planktonic larvae in the water column that can drift for 45-60 days at temperatures of 11-13°C and 8°C, respectively, before settling on the seabed and triggering metamorphosis (Boschi et al., 1969;Diez & Lovrich, 2010;Richer de Forges, 1977 (Boschi et al., 1969;Diez & Lovrich, 2010). The hypothesis has been tested for all three release scenarios.  Islands areas (Rz.5 and Rz.6) to reach the proposed marine protected areas although they might impact other areas located further east.

| Dispersal patterns according to the different release scenarios
Releasing particles from the Weddell Sea area (Rz.3) also never impacts the proposed marine protected areas.

| Intra-and inter-annual variabilities
Comparison of dispersal patterns among the nine simulated years show inter-annual variations in the extent of dispersal areas: such variation is mainly noticeable in the sub-Antarctic region and in the East Weddell Sea. Inter-annual variation is more obvious in the 11 NM scenario relative to the total extent of the dispersal pattern ( Figure 7; right panel). Interestingly, the dispersal area is broader in

| Invasion risks
Previous results were summarized in a synthesis map (Figure 9) that indicates the release zones and the simulated risk of particle introduction into proposed marine protected areas. We defined a "high risk," when models simulate the arrival of particles every year and every season in all neighbouring marine protected areas. The "no risk" exchange zones correspond to zones where released particles never reach any proposed marine protected areas. A "moderate risk" category was

| A focus on Deception Island
When Ballast water is exchanged from distances exceeding 200 NM from the nearest coasts, the Lagrangian model predicts  (2009,2010,2011,2012) or partly (2008,2013,2014,2015,2016). F I G U R E 9 Ballast water exchange zones and their associated simulated risk (dark green: "no risk"; yellow: "moderate risk"; red: "high risk") for particles to reach proposed marine protected areas. The black solid line represents the Polar Front yearly mean position. The risk was estimated for proposed marine protected areas of the considered region only. Other areas that might also be at risk were not included in this study. Distances to the coasts of the different release scenarios (200,50 or 11 NM) Gaines et al., 2003;Thomas et al., 2014) and to study the spread dynamics of invasive species (Brandt et al., 2008;Brickman, 2014).

| Particle dispersal
In the Southern Ocean, Lagrangian models have already been used to simulate dispersal abilities and the distribution of fish species or top predators, to understand the main key drivers of population connectivity and assess the position of the main foraging areas, in the aim of determining an effective management of natural resources (Della Penna et al., 2017;Young et al., 2012Young et al., , 2014Young et al., , 2015.
In the present work, daily variations of the environment were simulated over a 9-year period and model outputs were analysed to test the significance of dispersal patterns with regard to interseasonal and inter-annual variations. In simulating a large number of particles, Lagrangian models integrate natural variability of hydrodynamic systems . However, the model should rely on assumptions (parameterization of the general environment, of the properties of the simulated particle), which may not be trivial considering the broad spatial scale of the analysis, the overall system complexity and the unknown propagule pressure state (i.e. occurrence and density of non-native species in ship Ballast water).
Furthermore, some propagule traits such as buoyancy, physiology, survival rate and tidal behaviour were hypothesized; the actual traits of invasive species could potentially have a substantial impact on model outputs (Barbut et al., 2019;Miller & Morgan, 2013;North et al., 2008;Stanwell-Smith et al., 1999;Young et al., 2012), as well as the influence of climate change through its impacts on the physiology and dispersal potential of marine larvae (Quinn, 2017). Some sensitivity analysis could also be realized to improve our results, for instance to evaluate the assumption of the larval drift duration.
Model simulations show that in 6 months, particles can drift along the coasts of the Western Antarctic Peninsula up to South Georgia, driven by the power of the Antarctic Circumpolar Current and highlighting the importance of connectivity between Antarctic coasts and the Scotia Sea region (Appendix S1, Rintoul, 2009;Caccavo et al., 2018;Moffat & Meredith, 2018). Other oceanographic features such as the Weddell Sea gyre and major marine fronts off the Scotia Sea  Young et al., 2012Young et al., , 2015. These features also play a crucial role in the connectivity among sub-Antarctic islands (Young et al., 2012).
Comparisons of the three different scenarios indicate that the distance at which Ballast water is exchanged from the nearest coasts has significant impacts on particle trajectories, on the frequency and weighted number of particles reaching the Antarctic coasts ( Figures   4-7). Overall, particles are less likely to reach Antarctic coastal areas when Ballast water is exchanged at least 200 NM away from the nearest coasts (Figures 4 and 5).
Inter-seasonal and inter-annual variability were also shown to have significant effects on modelled dispersal patterns (Appendix S2, Figures 7 and 8). This was expected here, given that Southern to specific climate events and in particular, to regimes of westerly winds in link with El Niño Southern Oscillation events (Carvalho et al., 2005;Ciasto & Thompson, 2008;L'Heureux & Thompson, 2006;Limpasuvan & Hartmann, 1999). Years 2009Years , 2014Years and 2015 were characterized by strong El Niño episodes (warmer temperatures and stronger westerly winds which make the particles drift further east); in contrast, years 2010-2011 were characterized by strong positive Southern Ocean Indexes and with La Niña episodes (weaker westerly winds, dryer and colder atmosphere and a narrower extension of particle drifting area) (Nicolas et al., 2017).

| Results' overview in the general context
The Western Antarctic Peninsula is among the regions on Earth that experience climate warming at the fastest pace, where rising temperatures also directly or indirectly drive other environmental shifts (i.e. glacier melting, phytoplankton community shifts, changes in sea ice duration and extent) (Bers et al., 2013;Convey et al., 2009;Convey & Peck, 2019;Schofield et al., 2017;Schram et al., 2015). This makes the Western Antarctic Peninsula one of the most sensitive regions to potential invasions by introduced species in Antarctica (Hellmann et al., 2008;Hughes et al., 2020;McGeoch et al., 2015;Meredith & King, 2005). Increased temperatures and related environmental shifts indeed may favour the acclimation of non-native species introduced from warmer climates over cold-adapted native taxa (Galera et al., 2018;Hellmann et al., 2008).
For a few decades, maritime traffic has also steadily increased in the Southern Ocean and in the Western Antarctic Peninsula in particular, due to its relative proximity to harbours of southern South America (McCarthy et al., 2019). This increasing traffic has been cited as the main cause for non-native species introduction in coastal waters of the Western Antarctic Peninsula (Avila et al., 2020;Cárdenas et al., 2020;Diez & Lovrich, 2010;Fraser et al., 2018;Lee & Chown, 2007;McCarthy et al., 2019;Tavares & De Melo, 2004;Thatje & Fuentes, 2003 Island indeed occurred through Ballast water of cruise ships sailing southwards from ports of southern South America, which we consider to be a likely scenario, the crab must have been released at a distance equal or less than 50 NM from the Antarctic coasts ( Figure 10, Appendix S3). Another hypothesis could be its introduction by ship hull biofouling (Chan et al., 2015;Hughes & Ashton, 2017;Lee & Chown, 2009c), which has not been tested here. These results could be generalized to other species, with the ensuing consequences of species introductions (Britton et al., 2018;David et al., 2017;Walsh et al., 2016).
Our results also highlight that the variability in climate regimes has a strong effect on dispersal patterns: in certain years, particles may drift further and reach areas that are on average not considered to be potentially impacted by non-native species introduction via Ballast water exchange, as already discussed by Fraser et al. (2018) and Waters et al. (2018) for kelp rafting. Other authors stressed the significance of transient events in longdistance dispersal (Leese et al., 2010;Saucède et al., 2014). Such events may become more frequent in future decades, owing to ongoing climate change, since climate projections for the Southern Hemisphere for the 21st century predict a further southward shift and intensification of storm tracks (Perlwitz, 2011) and therefore hypothesize an increasing threat for potential species introductions (Hughes et al., 2020).

| Future management of marine protected areas
The Antarctic Treaty (ATCM, 2006)   propagule pressure within stored water (Lee & Chown, 2009a). Our results strongly highlight the importance of the upcoming 2024 regulation to further protect Antarctic marine life from non-native species introduction, in complement to conservation measures applied for visitors and ships approaching Antarctic coasts (Hughes et al., 2020;Lennox et al., 2015;McGeoch et al., 2015).

| CON CLUS IONS
This study provides insights on how Ballast water exchange can contribute to the arrival of non-native species in current and proposed marine protected areas of the Western Antarctic Peninsula, being one of the most vulnerable Antarctic regions to biological invasions.
Awaiting for the compulsory settlement of ballast water treatment systems on vessels (in 2024), the existing Ballast water guidelines so far ruled by the Ballast Water Management Convention and Antarctic Treaty (Antarctic Treaty, 1959;ATCM, 2006) are not sufficient to prevent the introduction of species in these marine protected areas, although respecting Ballast water discharges at 200 NM away from the nearest coasts lowers the risk of introduction. This is especially true for Ballast water being exchanged in the areas of the western and eastern Western Antarctic Peninsula. Because of the expected future increase in maritime traffic and the correlated risk of non-native species introduction and invasions potentially increasing due to global warming, we here advocate for delineating Ballast water discharge zones, so that propagules released within Ballast waters would not reach the most fragile Antarctic ecosystems. These discharge zones could be further fine-tuned with more data about maritime traffic and accounting for climatic variability.
We also recommend increasing the ratified distance of Ballast water exchange over 200 NM in the Western Antarctic Peninsula and avoiding discharges in the Eastern Antarctic Peninsula, two recommendations that could be included in future marine protected area proposals. This study shows that Ballast water exchange at 50 NM or closer to the coasts pose a dangerous threat, as these results in drifting propagules reaching Antarctic coasts. This is in particular exemplified by the case study of the introduction of the Patagonian crab H. planatus in Deception Island.
Our results indicate that, if the crab was indeed brought after Ballast water discharge, the Ballast water would have likely been discharged at 50 NM or closer to the Antarctic coast.
Finally, Ballast water regulation is not the only alternative to mitigate non-native species introduction, as they can also be introduced by the biofouling of ships; another serious threat to be considered for next conservation steps.

ACK N OWLED G EM ENTS
This is contribution no. 49 of the vERSO and no. 26 of the RECTO projects (https://recto verso proje cts.be/), both funded by the Belgian Science Policy Office (BELSPO). This work was also supported by a "Fonds pour la formation à la Recherche dans l'Industrie et l'Agriculture" (FRIA) and "Bourse fondation de la mer" grants to C. Guillaumot. We are thankful to Argentina and Chile delegations for QGIS maps of the proposal for a conservation measure establishing a MPA in Domain 1, to Sebastian Cisneros for its contribution to gathering the model forcing and to Emma Young and Léo Barbut for the fruitful discussions on Lagrangian models. Thanks to Ben Wallis for helpful comments during the ideas development.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13464.

DATA AVA I L A B I L I T Y S TAT E M E N T
Codes to run the Lagrangian model are difficult to provide, they rely on a complex network of scripts generally described in https://odnat ure.natur alsci ences.be/coher ens/getti ng-start ed/run-coherens.
You can email Valérie Dulière if you want a compilation of the scripts and some help to run the codes (vduliere@naturalsciences. be). Otherwise, raw results and R codes used to process them are publicly provided at https://zenodo.org/recor d/55882 20#.YXEfq R3godU (folder "Ballast water paper").