Spatial patterns and temporal variability of seagrass connectivity in the Mediterranean Sea

The endemic seagrass Posidonia oceanica is a key component of the coastal seascapes of the Mediterranean Sea, where it provides crucial ecosystem services and promotes the assembly of diverse ecological communities. Although protection policies exist, P. oceanica meadows have been steadily declining in the recent past because of human activities and climate change. Here, we quantitatively analyse basin‐wide patterns of seagrass connectivity over a 30‐year‐long period and identify connectivity hotspots that may serve as priority targets for conservation actions.


| INTRODUC TI ON
Posidonia oceanica (L.) Delile is a seagrass species endemic to the Mediterranean Sea; it inhabits the coasts of the entire Mediterranean basin in a 1-45 m depth range, except for large estuaries and regions where extreme thermal and salinity conditions are not favourable for its persistence (Gobert et al., 2006;Telesca et al., 2015). P. oceanica plays a pivotal ecological role as a habitat-forming species. Its vast underwater meadows shape the submarine seascape in coastal areas (Montefalcone, Albertelli, Bianchi, Mariani, & Morri, 2006).
They also create favourable conditions for the assembly of diverse and complex communities that include also many commercially important fish species (Pergent et al., 2016). For these reasons, P. oceanica can be considered an ecosystem engineer, that is a species providing crucial ecosystem services, such as water oxygenation, carbon sequestration, nutrient cycling, water purification and protection from coastline erosion, and offering shelter or nursery to other species (Campagne, Salles, Boissery, & Deter, 2015;Vassallo et al., 2013). Despite remaining the most widespread seagrass in the Mediterranean Sea, P. oceanica populations have been sharply declining in recent decades due to multiple stressors, including the localized effects of climate change and human activities, with an estimated 13%-50% decrease in areal extent over the past 60 years (de los Santos et al., 2019;Marbà, Díaz-Almela, & Duarte, 2014;Telesca et al., 2015). In this respect, P. oceanica shares the same fate as the majority of seagrass species in coastal waters across the globe (Orth et al., 2006;Waycott et al., 2009). Owing to its importance in the context of the Mediterranean Sea coastal ecosystems, associated with the current trends of species distribution decline, P. oceanica has been identified as a key target for conservation by European institutions since the 1990s (Boudouresque et al., 2012).
Posidonia oceanica can reproduce sexually, producing seed-carrying, positively buoyant fruits that may be carried by marine currents and thus represent the main dispersal agents for this species (McMahon et al., 2014). Data on the dispersal distances of P. oceanica fruits are scarce. As an example, Arnaud-Haond et al. (2007) showed that the dispersal of P. oceanica fruits can be restricted to the scale of a few metres in some meadows, in spite of the apparent potential for larger-scale seed dispersal. In the same study, the authors showed the existence of genetic structure within individual seagrass meadows and genetic differentiation among populations on scales ranging from tens of kilometres up to the great divergence between populations inhabiting the eastern and western basins of the Mediterranean Sea. However, evidence also exists suggesting that dispersal distances for P. oceanica may be significantly greater and more variable than previously considered. For instance, Serra et al. (2010) reported dispersal distances up to 50 km. Realized connectivity in P. oceanica is likely mainly limited by the episodic nature of flowering, sexual reproduction and the overall low production rate of fruits in most locations (e.g. Balestri & Cinelli, 2003;Balestri, Vallerini, & Lardicci, 2017;Procaccini, Orsini, Ruggiero, & Scardi, 2001). This implies that local management alone may not be enough for P. oceanica and that spatial planning should not dismiss connectivity out of hand. In fact, sexual reproduction and fruit dispersal, even at a low rate, can play a critical role in the colonization of new sites, recovery after disturbance and establishment of new genotypes in existing seagrass populations. Because of the widespread distribution of this foundation species along Mediterranean coasts and the reported large-scale trajectories of loss, conservation strategies should be planned at a whole-basin scale, with priority being given to sites that play a key role in structuring population connectivity, thus supporting effective conservation and restoration strategies. This coordinated effort would also be coherent with habitat-based policies that constitute the cornerstone of Europe's nature conservation policy (e.g. "Habitats Directive," 92/43/EEC).
Designing prioritization strategies at the scale of the whole Mediterranean Sea is necessary to channel resources where interventions are most urgently needed and/or likely to be effective.
However, this requires a comprehensive framework able to capture the complexity of both patterns and processes, and duly accounting for the challenges imposed by such a large spatial extent.
This problem is further exacerbated by the trans-boundary nature of the Mediterranean Sea, the complex social, cultural and political conditions of the countries surrounding it and its high sensitivity to global climate change (Lejeusne, Chevaldonné, Pergent-Martini, Boudouresque, & Perez, 2010;Micheli et al., 2013). Indeed, a regional approach for the protection and enhancement of the status of the marine environment in the Mediterranean Sea requires a close cooperation among states and international organizations, which is one of the founding principles of the Barcelona Convention ("Convention for the Protection of the Marine Environment and the Coastal Region of the Mediterranean"). Also, the Marine Strategy Framework Directive (MFSD, 2008/56/EC) has established detailed criteria and methodological standards according to which each member state has to take the necessary measures to achieve or maintain "Good Environmental Status" of seagrass conservation actions in the Mediterranean large marine ecosystem, a challenging environment characterized by complex socio-economic boundary conditions and high sensitivity to the localized effects of global climate change.

K E Y W O R D S
biophysical modelling, conservation hotspots, dispersal, Lagrangian simulations, marine connectivity in the marine environment. In general, identifying areas for protection in large marine ecosystems requires techniques involving multiple scales of analysis. The spatial structuring imposed by the interplay between local environmental conditions and basin-wide circulation patterns, in particular, calls for the design of marine protected areas ensuring and promoting seascape connectivity (Planes, Jones, & Thorrold, 2009). In fact, if a protected area is not sufficiently connected to others, it may not effectively receive/ send propagules (such as larvae or seeds), thus possibly thwarting natural recovery (McCook et al., 2009). In other words, spatial planning of marine protection should be conceived as the design of a coherent network of protected areas ecologically connected at various spatial scales, in order to fulfil ecological aims more effectively than single individual sites could do (Boero, 2015;WCPA/ IUCN, 2007). Assessing the functional connectivity of species that are target of protection efforts, such as P. oceanica, is thus of paramount importance to large-scale conservation planning (Jahnke et al., 2017;Kendrick et al., 2017).
In this work, connectivity patterns of P. oceanica are evaluated at the scale of the whole Mediterranean Sea over a 30-year-long time span. We propose a definition of species-specific functional connectivity (suitability-weighted connectivity-for brevity, s-connectivity) accounting for both local suitability conditions and dispersal patterns driven by marine currents. This definition aims to account not only for the amount of propagules potentially exchanged between marine sites but also for the environmental conditions that may influence local suitability for the species under study. To that end, a Lagrangian approach is used to build a biophysical model for the dispersal of P. oceanica. Following the methodological framework proposed by Melià et al. (2016), s-connectivity is then evaluated on top of the results of Lagrangian simulations to single out the strongest and most time-persistent ecological connections for P. oceanica across the Mediterranean Sea, specifically in terms of the possible functional roles that a local population can play in the context of a larger metapopulation, namely retainer, sink and source. The multi-decadal temporal span of the simulation exercise also allows the study of temporal variability in P. oceanica connectivity and the identification of trends, as well as the investigation of the possible relationships between connectivity and meteorological fluctuations.
The ultimate goal of the analysis is to improve spatial prioritization F I G U R E 1 The biophysics of Posidonia oceanica connectivity in the Mediterranean Sea. (a) Species-specific suitability map (Giannoulaki et al., 2013). Colour-coded scores represent estimated probabilities of P. oceanica presence. (b) Example of circulation field. Colours represent the speed of surface currents for 1 January 2014, obtained through bilinear interpolation of data from a physical reanalysis of Mediterranean circulation (Lecci et al., 2017;Simoncelli et al., 2014). (c) Marine sectors for the analysis of P. oceanica connectivity (centroids, coloured dots). Colours represent the number of simulated Lagrangian trajectories starting from each sector in each year of the simulation time span . This variability reflects the small-scale spatial heterogeneity in the distribution of suitable sites. Black dots are suitable sites that fall outside the spatial domain of the physical reanalysis and that are thus not used in the numerical simulations

| Biophysical simulations of dispersal
Basin-wide potential connectivity for P. oceanica in the Mediterranean Sea is estimated through Lagrangian simulations (Van Sebille et al., 2018), with dispersing agents representing P. oceanica fruits. Lagrangian particles are released at marine sites that are suitable for P. oceanica meadows, are transported by currents and may eventually settle at some suitable sites.
Release sites are determined based on a species-specific suitability map produced by the MediSeH project (Giannoulaki et al., 2013), in which binary observations of P. oceanica presence-absence (Telesca et al., 2015) and a set of 36 mapped predictor variables (encompassing bathymetry and geographical features, physico-chemical characteristics, nutrient and pollutant concentrations, as well as human impact indicators) were used to train a random forest algorithm (Breiman, 2001)  In each year, timing of release is set to match the fruit-release season of P. oceanica (typically, January throughout April, for a total of n d = 120 days; see Melià et al., 2016;Jahnke et al., 2017). For each day in this season, a fixed number of particles (n p = 15) are released from each pixel of the suitability map that has strictly positive suitability and that lies, at least in part, inside the domain of the physical reanalysis. While the release of 15 particles per site and day might seem quite a low figure, the size of the spatial domain of our Lagrangian exercise (the whole Mediterranean basin) is such that matters of computational feasibility become relevant. A total of n r ≈ 5.69 × 10 5 release sites are in fact identified following the selection criteria outlined above. All in all, an excess of n t = n y n d n p n r ≈ 30 billion Lagrangian particles is tracked over the whole numerical assessment. The initial position of particles within each release site is randomly assigned to uniformly span the area of the pixel and a depth interval of 0-1 m (P. oceanica fruits are freefloating and positively buoyant; Serra et al., 2010).
The longitudinal and latitudinal components of the position of each particle are updated by assuming passive transport driven by marine circulation fields, while particle depth is not updated.
Numerical integration is performed with a Runge-Kutta fourthorder scheme with adaptive step size (Dormand & Prince, 1980). At each time step, three-dimensional trilinear interpolation of the longitudinal and latitudinal components of the velocity field is performed. Note that the spatial grain of the circulation model (1/16°) is much coarser than the suitability map used to identify release sites (1/240°). As such, the effects of releasing a large number of particles from each pixel of the latter, higher-resolution grid would likely be dampened by the necessity of interpolating current velocities from the former, lower-resolution grid. The position of each particle is tracked for a period of time corresponding to the duration of the dispersing stage of P. oceanica, after which fruits dehisce and release their seeds.
Duration estimates for the floating phase of the fruits of this seagrass species vary from one/two weeks (e.g. Aliani, Gasparini, Micheli, Molcard, & Peirano, 2006;Buia & Mazzella, 1991) up to 4 weeks (Serra et al., 2010). Here, we use a value of 28 days, which is towards the maximum reported length of the dispersal window and allows an assessment of potential connectivity (an upper bound for realized connectivity). This value has consistently been used in all previous modelling studies addressing P. oceanica dispersal dynamics (Jahnke et al., 2017;Melià et al., 2016;Serra et al., 2010).
In our modelling framework, a dispersal event is considered successful only if a particle reaches a suitable site at the end of its dispersing phase. Although this approach differs from what has been proposed in the literature to describe connectivity in other seagrass species (see Appendix S1 for details), it has consistently been used in previous works addressing P. oceanica fruit dispersal via Lagrangian simulations (see again Jahnke et al., 2017;Melià et al., 2016;Serra et al., 2010).

| Posidonia oceanica-specific connectivity
The strength of P. oceanica connectivity is assumed to be proportional to the number of successful dispersal events n AB (y) that link (directionally) any two suitable sites (say, site A to site B) in year y.
This quantity is clearly influenced by species-specific traits such as timing of fruit release and duration of the dispersing phase, but is mostly concerned with the hydrodynamics of passive propagule dispersal by marine currents. To account for small-scale heterogeneities in the quality and spatial distribution of suitable sites across the Mediterranean Sea, we define an ecologically motivated measure of connectivity in which successful dispersal events are weighted according to the suitability scores of both release (source) and settling (sink) sites. Suitability-weighted connectivity (s-connectivity, from here on out) between two sites A and B in year y is thus defined as C AB s (y) = s A n AB (y) s B . In this way, not only species-specific dispersal patterns but also local suitability conditions are effectively taken into consideration and integrated in a comprehensive measure of functional connectivity. For this reason, s-connectivity can represent an informative tool to evaluate the ecological value of marine sites, at least from the perspective of the potential connectivity of the species being studied. year (say y) is defined as.
These scores can be suitably organized in a time-varying s-connectivity matrix C s (y) = [C ij s (y)], which in turn can be characterized as a directed weighted graph, with nodes and edges being, respectively, marine sectors and time-varying s-connectivity scores.

| Connectivity metrics
In the theory of complex networks, two simple yet powerful metrics of connectivity are the indegree and the outdegree of network nodes, defined for weighted graphs as the sum of the incoming and outgoing links' weights, respectively (Newman, 2010). Thus, in the context of P. oceanica dispersal, indegree and outdegree measure the tendency of the sites within a marine sector to function as potential sinks or sources for P. oceanica fruits, that is to be successful at receiving/sending propagules from/to suitable sites lying in other sectors. In ecological applications, another important metric based on connectivity scores is self-retention (e.g. Melià et al., 2016). For the problem at hand, self-retention quantifies how many P. oceanica fruits both are released and settle within a given marine sector (say i), with the release and settling sites being possibly different, but both lying in sector i.
Technically, the diagonal elements C ii s (y) of the s-connectivity matrix thus represent the local self-retention (SR) of each marine sector (SR i (y) = C ii s (y)), while indegree (ID) and outdegree (OD) can be easily evaluated as the column or the row sums of the s-connectivity matrix, that is ID i (y) = ∑ j≠i C s ji (y) and OD i (y) = ∑ j≠i C s ij (y), with the condition j ≠ i being imposed to avoid multiple counting of self-retention. Because of the time-varying nature of the s-connectivity matrix (as determined by the temporal variability of circulation fields), self-retention, indegree and outdegree are all time-varying quantities too.

| Identification of s-connectivity hotspots
The metrics described above can be used to identify the hotspots of s-connectivity for P. oceanica across the Mediterranean Sea. To do so, we follow the methodological framework proposed by Melià et al. (2016) to assign each marine sector a synthetic s-connectivity score recapitulating its capacity to simultaneously act as retainer, sink and source. For a given metric, two percentile scores can thus be assigned to each sector: the first defined as the percentage of sectors that have intensity equal to or lower than the one being considered and the second as the percentage of sectors that have equal or higher variability. Therefore, the sector endowed with highest acrossyear mean will receive an intensity score of 100 and the one with the lowest a score of zero; conversely, the sector with lowest coefficient of variation will receive a variability score of 100 and the one with the highest a score of zero. This procedure leads to the definition of six percentile scores for each sector (say i): pSR i Ave and pSR i CV for self-retention, pID i Ave and pID i CV for indegree and pOD i Ave and pOD i CV for outdegree.
Afterwards, for each sector i, we introduce summary percentile scores for self-retention, indegree and outdegree. They are defined conservatively as the minimum between the two relevant percentile scores pertaining to the intensity and variability of each metric, that is pSR i = min (pSR i Ave , pSR i CV ), pID i = min (pID i Ave , pID i CV ) and pOD i = min (pOD i Ave , pOD i CV ). A synthetic percentile s-connectivity score pCS i is then assigned to each of the n m marine sectors by taking the minimum (again, conservatively) among the three summary scores for self-retention, indegree and outdegree, that is pCS i = min (pSR i , pID i , pOD i ). This final percentile value can thus be interpreted as an overall s-connectivity score, and sectors where pCS is highest can be considered hotspots of s-connectivity for P. oceanica in the Mediterranean Sea and possible priority candidates for species protection. We stress again that the aggregation procedure used to evaluate pCS reflects a conservative strategy by which hotspots are marine sectors identified based on their ability to outperform others in the s-connectivity metric/ indicator in which they are weakest.
Finally, the top-k s-connectivity hotspots (with k being the number of target hotspots) are identified as the k sectors whose synthetic percentile s-connectivity score exceeds the (n m − k)th order statistic of the score distribution.
The final results of the hotspot identification procedure do evidently depend upon the spatial scale of analysis, which in the problem at hand is defined by the size of marine sectors. Sensitivity analysis can be used to check whether the procedure is robust to variations in spatial scales, namely to changing the way local (self-retention) versus in/outbound (in/outdegree) connections are defined (details in Appendix S1). Other approaches to investigate connectivity within a network of seagrass populations, thus also possibly identifying connectivity hotspots, have been explored in the literature, namely based on tools proposed in the context of complex network theory (Grech et al., 2018). A discussion of such approaches is also available in Appendix S1.

| Evaluation of s-connectivity temporal variability
To evaluate temporal variability in s-connectivity and to assess whether there may exist temporal trends, linear regression is per-

| Connectivity metrics and s-connectivity hotspots for Posidonia oceanica
Across-year s-connectivity indicators (mean values and coefficients of variation of self-retention, indegree and outdegree for each marine sector; see Figures S1 and S2 in Appendix S2) represent the basis to F I G U R E 2 Examples of time-varying Posidonia oceanica dispersal kernels. In each panel, colours code the relative frequency of successful dispersal events linking the selected marine sector (corresponding to the labelling of the black circles in the top inset) with other suitable sectors lying at a given distance during a specific dispersal season. The nine sample sectors (a-i) have been selected so as to span over different spatio-temporal scales of dispersal, that is encompassing sectors characterized by relatively short/long dispersal distance (acrossyear average of mean dispersal distance approximately half/double the across-sector mean value) and low/high temporal variability (acrossyear coefficient of variation of mean dispersal distance approximately half/double the across-sector mean value). The values of the acrossyear average (Ave) and coefficient of variation (CV) of mean dispersal distance for the nine sample sectors are reported on top of the panels

| Temporal trends in s-connectivity at the local scale
Temporal trends in local s-connectivity metrics are shown in

| D ISCUSS I ON
In this work, we have performed a basin-wide, multi-decadal connectivity assessment for P. oceanica, an iconic primary producer species F I G U R E 4 Hotspots of s-connectivity for Posidonia oceanica in the Mediterranean Sea. (a) Synthetic percentile s-connectivity score, evaluated for each suitable marine sector as the minimum among its percentile scores for intensity and variability of self-retention, indegree and outdegree ( Figure S4 in Appendix S2). (b) Top-k s-connectivity hotspots, with k = 100 or k = 500. The top-500 sectors do obviously include the top-100 as well Reef, Australia, to evaluate the potential of seagrass dispersal in the area. Based on the results of Lagrangian simulations, different metrics of network connectivity (node-degree distribution, self-retention and outdegree) and node centrality (betweenness and PageRank) were used to identify seagrass meadows acting as retainers, sources or stepping stones for dispersal, and that could serve as priority candidates for conservation (Grech et al., 2018).
We believe that our attempt to integrate habitat suitability (also a proxy for the actual distribution of the species being studied) directly into the evaluation of connectivity metrics might be especially promising when the size of the study area (the whole Mediterranean Sea basin, in our case) makes it impractical (or simply not possible) to compare potential connectivity patterns, obtained by biophysical modelling of current-driven dispersal, against realized connectivity patterns, estimated, for example, through genetic analyses. Our network-based approach could also serve as an effective starting point for the detection of communities within the time-varying graphs describing the dispersal patterns of P. oceanica (Newman, 2010), which in turn could assist in the definition of separated management units, that is clusters of local seagrass meadows that should be managed separately to ensure their long-term persistence (Grech et al., 2018;Jahnke et al., 2018).
The results of our assessment suggest that spatio-temporal variability is an important component of P. oceanica s-connectivity ( Figures 2 and 3, Movie M1). Clearly, such variability makes the identification of connections that are both sufficiently strong and timepersistent to be ecologically relevant a completely non-trivial task.
This difficulty has been overcome by applying a recently proposed methodological framework that allows to determine connectivity hotspots based on their potential to simultaneously function as effective retainers, sinks and/or sources for the dispersing agents of the target species (Melià et al., 2016). This approach accounts for the different functional roles of dispersal, is based on easily interpreted connectivity metrics, relies on a simple and conservative aggregation scheme and allows to effectively take into consideration both spatial and temporal variability in dispersal, thus representing a balanced framework to quantitatively discuss spatial conservation strategies at a basin scale. According to this identification procedure, hotspots of s-connectivity for P. oceanica (Figure 4)  is interesting to note that the highest ranked sectors (e.g. top-100 hotspots) are consistently surrounded by sectors also endowed with high s-connectivity (e.g. top-500 hotspots), which is suggestive of the fact that s-connectivity analysis is robust enough to be relevant for policymaking.
By contrast, it is crucial to remark that the results of our hotspot identification procedure do not necessarily provide a complete picture of P. oceanica connectivity in the Mediterranean Sea.
As However, the study of this type of intergenerational dynamics will require the development of an integrative modelling approach in which the basin-wide metapopulation dynamics of P. oceanica can effectively be explored by coupling the dispersal means provided by marine currents with local-scale demographic processes, such as shoot survival, vegetative growth and sexual reproduction.
All these considerations highlight the importance of cross-validating measures of potential connectivity, albeit corrected for local suitability conditions like in the present study, with measures of realized connectivity, as obtained through analysis of effective gene flow. In the case of P. oceanica, in fact, the former is often found to be possibly quite overestimated with respect to the latter (Jahnke et al., 2017;Serra et al., 2010), which in turn is thought to be relatively low overall (e.g. Arnaud-Haond et al., 2014;Procaccini et al., 2001). Comparing the findings presented in published studies of genetic connectivity for P. oceanica in the Mediterranean Sea with our assessment of basin-wide potential connectivity may not be a trivial task. Some of those studies were in fact conducted over relatively small spatial domains, spanning from single meadow (e.g. Migliaccio, Martino, Silvestre, & Procaccini, 2005) to regional scales (e.g. Jahnke et al., 2017;Procaccini et al., 2001).
Mediterranean-wide analyses of P. oceanica genetic connectivity exist (Arnaud-Haond et al., 2007;Rozenfeld et al., 2008;Serra et al., 2010), but in all those cases the number of sampled meadows was understandably limited to a few dozen at most. In general, whenever genetic connectivity was evaluated at a whole-basin scale, the Strait of Sicily was identified as a contact zone between the genetically partitioned seagrass populations inhabiting the western and eastern basins of the Mediterranean Sea. This region (the coast of Tunisia, in particular) is also highlighted as one of the richest in s-connectivity hotspots by our modelling approach (Figure 4), as well as one endowed with relatively long potential dispersal distances (see again Figure 3).
The multi-decadal temporal span of this study has allowed to ascertain the existence of recent trends in P. oceanica s-connectivity across the Mediterranean Sea ( Figure 5). Statistically significant temporal trends in self-retention, indegree and outdegree seem to be quite infrequent among all suitable sectors (they have been detected in less than 6% of marine sectors), but relatively more frequent in sconnectivity hotspots (with frequencies ranging up to 20%; Table 1).
Here, contrasting directions of change are actually found: for instance, decreasing self-retention and increasing in/outdegree are all more frequently observed in the top-100 s-connectivity hotspots than in non-hotspot sectors, possibly a sign that somewhat small changes in circulation patterns around key strategic sites may have important consequences for P. oceanica dispersal dynamics at large spatial scales. Also, while basin-averaged connectivity values do not show any statistically significant trends, in/outdegree in, for example, the top-500 s-connectivity hotspots do ( Figure S9 in Appendix S2). All these findings suggest that the role played by s-connectiv- Like all modelling studies, ours is not devoid of limitations. One such source of possible inaccuracies in our analysis is perhaps the use of a static suitability map to both initialize Lagrangian simulations and evaluate s-connectivity scores. As a matter of fact, P. oceanica meadows have declined rapidly in several areas of the Mediterranean basin, possibly also as a result of decreased habitat suitability in response to the localized effects of climate change, water quality degradation, coastal modification and other sources of human pressure (Chefaoui, Duarte, & Serrão, 2018;de los Santos et al., 2019;Marbà et al., 2014;Telesca et al., 2015). In this respect, airborne and satellite imagery could provide a dependable, deployable and cost-effective tool to produce updated distribution maps for P. oceanica, as testified by the growing number of related applications (Borfecchia et al., 2013;Fornes et al., 2006;Matta et al., 2014;Pasqualini et al., 2005;Traganos et al., 2018).
Although most of these studies refer to relatively small areas within the Mediterranean Sea, the most recent one (Traganos et al., 2018) proposes a workflow for regional-scale mapping of seagrasses powered by remote sensing, machine learning and cloud- and cultures-all of which makes the problem of setting priorities for regional conservation planning a highly non-trivial task . This caveat notwithstanding, we believe that the present study may represent a step forward in the application of a quantitative, scalable and replicable methodological framework for the prioritization of conservation actions, with the overarching goal of saving more with less.

ACK N OWLED G EM ENTS
The authors acknowledge support from the European Union's Horizon 2020 research and innovation programme through the ECOPOTENTIAL project, grant agreement No. 641762. The authors also wish to thank the anonymous referees for their valuable and constructive comments.

DATA AVA I L A B I L I T Y S TAT E M E N T
The suitability map for P. oceanica is available at http://www.emodn et-seabe dhabi tats.eu. The physical reanalysis of circulation field is available online at http://marine.coper nicus.eu/. MOI1 and MOI2 data are available online at https ://cruda ta.uea.ac.uk/cru/data/moi/,