Dramatic loss of seagrass habitat under projected

Accepted Article

DR. ROSA CHEFAOUI (Orcid ID : 0000-0001-5031-4858) PROF. ESTER A. SERRAO (Orcid ID : 0000-0003-1316-658X)

Article type

: Primary Research Articles

TITLE Dramatic loss of seagrass habitat under projected climate change in the Mediterranean Sea

Loss of seagrass habitat under climate change

AUTHORS Rosa M. Chefaoui1*, Carlos M. Duarte2, Ester A. Serrão1*

1.

CCMAR - Centro de Ciências do Mar, CIMAR Laboratório Associado, Universidade do

Algarve, Campus de Gambelas, 8005-139 Faro, Portugal 2.

King Abdullah University of Science and Technology (KAUST), Red Sea Research Center

(RSRC), Thuwal, 23955-6900, Saudi Arabia This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/gcb.14401 This article is protected by copyright. All rights reserved.

Accepted Article

*

Correspondence authors:

Rosa M. Chefaoui Tel: +351 289 800 051 e-mail: [email protected] Ester A. Serrão Tel: +351 289 800 051 e-mail: [email protected]

KEYWORDS climate change; Cymodocea nodosa; ecological niche modelling; genetic diversity; Mediterranean Sea; Posidonia oceanica; range shift; seagrass decline

PAPER TYPE Primary Research Article ABSTRACT Although climate warming is affecting most marine ecosystems, the Mediterranean is showing earlier impacts. Foundation seagrasses are already experiencing a well-documented regression in the Mediterranean which could be aggravated by climate change. Here, we forecast distributions of two seagrasses and contrast predicted loss with discrete regions

This article is protected by copyright. All rights reserved.

Accepted Article

identified on the basis of extant genetic diversity. Under the worst-case scenario, Posidonia oceanica might lose 75 % of suitable habitat by 2050, and is at risk of functional extinction by 2100, whereas Cymodocea nodosa would lose only 46.5 % in that scenario as losses are compensated with gained and stable areas in the Atlantic. Besides, we predict that erosion of present genetic diversity and vicariant processes can happen, as all Mediterranean genetic regions could decrease considerably in extension in future warming scenarios. The functional extinction of Posidonia oceanica would have important ecological impacts and may also lead to the release of the massive carbon stocks these ecosystems stored over millennia.

INTRODUCTION The effect of climate change on latitudinal or elevational shifts, colonization, and extinction of species is widely documented (eg., Burrows et al., 2011; Chen et al., 2011; Poloczanska et al., 2013; Bates et al., 2014; Pecl et al., 2017) . Although climate warming affects all biomes on Earth, the velocity of climate change and the shift in seasonal timing of temperatures are higher in the ocean than land at some latitudes (Burrows et al., 2011; Poloczanska et al., 2013), which may prevent species with limited dispersal from rescuing lineages located at threatened refugia (Assis et al., 2017). This, and the good match between marine species latitudinal ranges and their thermal limits (Sunday et al., 2012) allow us to predict future distributional range shifts of marine species and to assess the impact of warming on the loss of genetic diversity with some confidence (Neiva et al., 2015; Assis et al., 2016, 2018; Buonomo et al., 2018) . Genetic diversity plays a key role in resisting and adapting to environmental changes (Jump et al., 2009) but it is poorly understood. Climate change can


Accepted Article

affect biodiversity at all levels, and although there is a strong focus on the loss of species richness, genetic diversity below species level is the basis of adaptive and evolutionary potential and should therefore be assessed. The Mediterranean Sea, due to its semi-enclosed nature, is warming faster than the global average (Coll et al., 2010; Jordà et al., 2012; Templado, 2014), with the increase of sea temperature since the 70’s enhancing seawater stratification, mass mortality events, species shifts and the spread of invasive species (Coma et al., 2009; Lejeusne et al., 2010; Marbà et al., 2015) . Synergistic interactions between climate change and other anthropic pressures (e.g. pollution, habitat degradation) might be converting the Mediterranean into one of the most impacted seas in the world (Templado, 2014). Moreover, the presence of the European landmass constrains the movement of species and blocks the poleward biogeographic shift of endemic species (Poloczanska et al., 2013) as well as their potential dispersal routes (Burrows et al., 2011). As a result, Mediterranean species, especially endemic ones, have a higher risk of declining and even of functional extinction with climate change. Among the rich biota extant in the Mediterranean Sea, seagrass ecosystems play an important role harbouring species, improving water quality, dissipating wave energy and protecting coastlines (Duarte, 2002, 2011; Orth et al., 2006). This is particularly the case for the endemic Mediterranean seagrass species Posidonia oceanica (L.) Delile, which forms some of the most productive and valuable ecosystems in the Mediterranean. Cymodocea nodosa (Ucria) Ascherson is also widespread in the Mediterranean, but occurs on some adjacent northeast Atlantic coasts (Chefaoui et al., 2016) . Both subtidal seagrasses are protected by the Appendix I of the Bern convention. In addition to various national legislations, P. oceanica is also protected by the Barcelona convention (Annex II) and the


Accepted Article

Annex 1 of the European Union’s Habitats Directive, while C. nodosa is listed by the OSPAR convention. Despite their ecological value as important foundation species, their meadows are experiencing a well-documented regression in the Mediterranean due to anthropogenic pressure and invasive species (e.g., Marbà et al., 1996; Boudouresque et al., 2009; Marbà et al., 2014; Oprandi et al., 2014; Holon et al., 2015; Telesca et al., 2015; Burgos et al., 2017), which will probably be aggravated by the effect of climate change. Documented seagrass mortality, mesocosms experiments on thermal limits, and projections of the increase in SST under climate change scenarios have pointed to the possible disappearance of P. oceanica meadows with future Mediterranean warming (Marbà & Duarte, 2010; Jordà et al., 2012; Olsen et al., 2012; Marbà et al., 2015) . The genetic structure of both seagrasses in the Mediterranean Sea shows a similar cleavage of differentiated genetic groups between the western and eastern-Mediterranean basins, in both P. oceanica (Arnaud-Haond et al., 2007; Rozenfeld et al., 2008; Serra et al., 2010) and C. nodosa (Alberto et al., 2008; Masucci et al., 2012) . Previous studies found that regions retaining the highest genetic diversity were congruent with the location of putative cold climate refugia during the Last Glacial Maximum for these species (Chefaoui & Serrão, 2017; Chefaoui et al., 2017a) . According to these models, colder palaeoclimatic conditions probably caused southwards shifts in the Mediterranean meadows of these seagrasses followed by postglacial northwards migrations (Chefaoui & Serrão, 2017; Chefaoui et al., 2017a) . Thus, it is expected that further range shifts could take place in the near future as a consequence of ongoing climate warming. However, the climatic response of the two species to warming may differ. Experimental tests demonstrated a higher vulnerability to thermal stress of P. oceanica (a temperate seagrass) when compared with C. nodosa (of tropical origin) regarding their growth and demography (Olsen et al., 2012). In This article is protected by copyright. All rights reserved.

Accepted Article

addition, the poleward shift of P. oceanica is precluded by the European landmass (Burrows et al., 2011) as it is endemic in the Mediterranean, whereas C. nodosa occurs in the Atlantic and may therefore more easily expand polewards along the Eastern Atlantic coasts. Here we examine how predicted range dynamics caused by climate change could affect the distribution and genetic diversity of P. oceanica and C. nodosa in the Mediterranean. We use Ecological Niche Modelling (ENM) to project the future distributions of the seagrasses under a range of climatic scenarios proposed by the Intergovernmental Panel on Climate Change (IPCC). ENMs relate statistically the environmental space of the species data with that of the study area to produce a prediction in that geographical space. Given the conformity of marine species’ latitudinal ranges to their thermal tolerances (Sunday et al., 2012), the success of ENM in modelling past and current Mediterranean seagrass distribution (Chefaoui et al., 2016, 2017a; Chefaoui & Serrão, 2017), and the availability of predictive scenarios of future sea surface temperature, ENM-based forecasts provide a promising approach to assess the future of Mediterranean seagrasses. The use of ENM coupled with genetic diversity data has already provided prospects on the future persistence of rich genetic populations of other coastal species (Neiva et al., 2015; Assis et al., 2016, 2017, 2018; Buonomo et al., 2018) . Our approach will allow us to estimate the future ranges of Mediterranean seagrass species under different future climatic scenarios and infer how the expected range dynamics would erode the current genetic diversity of these seagrass populations.


Accepted Article

MATERIALS AND METHODS Species and environmental data We compiled occurrence data for Cymodocea nodosa from an extensive literature review, Algaebase (www.algaebase.org/), and the Global Biodiversity Information Facility (GBIF, www.gbif.org/). The dataset compiled for C. nodosa is available as Supporting Information S1. Posidonia oceanica's occurrences were obtained from available published data (Chefaoui et al., 2017a)

. One occurrence for each grid cell was used in the modelling process: 709

presence cells for P. oceanica (obtained from 1139 records), and 209 cells for C. nodosa (from 299 records compiled) (Fig. 1). We georeferenced species and environmental data to the same grid resolution (9.2 km ~ 0.083° × 0.083°). We used the 30 arc-seconds resolution GEBCO’s gridded bathymetric data set (http://www.gebco.net/) to delimit a 40 m depth threshold for seagrass occurrence along the Mediterranean, Black Sea and adjacent northeast Atlantic coasts. We used a Pearson correlation test to check an initial list of variables composed by derived measures of sea surface temperature (SST) and salinity. These variables are both available for present and future climate scenarios, and were found to be relevant to model the distribution of the species in the past (Chefaoui et al., 2016, 2017a; Chefaoui & Serrão, 2017) . The minimum, maximum, mean, and range values of sea surface temperature (SST), and the mean salinity for current conditions were obtained from the Bio-ORACLE dataset (Tyberghein et al., 2012). Predictors for future conditions were ensembled from five Ocean General Circulation Models (Table 1) pertaining to the Coupled Model Intercomparison Project Phase 5 (CMIP5; http://cmip-pcmdi.llnl.gov/cmip5/). We used two Representative Concentration Pathways (RCPs), the RCP 2.6 and the RCP 8.5, providing the lowest and


Accepted Article

highest greenhouse gas concentration scenarios, respectively (Moss et al., 2008; Meinshausen et al., 2011) . The monthly averaged data by 2050 (from 2030 to 2050) and 2100 (from 2080 to 2100) were computed for each scenario.

Ecological niche modelling To predict the current and future climate scenarios for both seagrass species, we implemented an ensemble niche modelling approach using the “biomod2” package (Thuiller et al., 2016) . We computed six presence-absence algorithms: generalized additive model (GAM), flexible discriminant analysis (FDA), generalized boosting model (GBM), multiple adaptive regression splines (MARS), generalized linear model (GLM), and randomForest (RF) models. The same number of pseudo-absences as presence cells were extracted at random for each species, due to the relatively small study area (17785 cells), but three sets of pseudo-absences were created to increase randomness. A total of 120 models were computed for each species (10 iterations x 2 pseudo-absence sets x 6 modelling techniques). Data were split into a calibration (70 %) and a validation set (30 %) in each of the 10 iterations performed for each set of pseudo-absences and model. We evaluated model performance using the area under the receiver operating characteristic (ROC) curve (AUC), and ROC-derived sensitivity (presences correctly predicted) and specificity (absences correctly predicted) measures (Fielding & Bell, 1997) , and the true skill statistic (TSS; Allouche et al., 2006), using the threshold at which the sum of the sensitivity and specificity is highest (Thuiller et al., 2016). We computed the average of binary predictions (“committee averaging”) ensemble for current conditions by using just the models achieving TSS ≥ 0.7. The ensembles were projected to the four scenarios for the future (RCP 2.6 and RCP 8.5 by 2050 and 2100). We computed a clamping mask to check the difference in the values of the variables between the training range and the novel scenarios and ascertain the uncertainty involved in the extrapolation of models to future


Accepted Article

conditions. Besides, the relevance of each variable in the ensemble was calculated by means of the correlation between the full model and a model rearranged without one of the variables (Liaw & Wiener, 2002; Thuiller et al., 2016). We used three iterations for this process which gave an importance value from 0 to 1 (highest importance). We converted the predicted probabilities of occurrence into binary outputs (presence-absence) using the threshold which optimized TSS and ROC scores after testing a set of thresholds for each ensemble (Thuiller et al., 2016). We estimated the gained, lost and stable regions for each species by comparing the binary predictions between the present and each future scenario.

Genetic diversity data Data on genetic structure and diversity were gathered from Alberto et al., (2008) Cymodocea nodosa populations, and from Arnaud-Haond et al., (2007) (2010)

for

and Serra et al.

for Posidonia oceanica. Alberto et al., (2008) assigned 47 populations to four

genetic clusters: high-latitude Atlantic (AH), low-latitude Atlantic (AL), Western Mediterranean basin (WM), and Eastern Mediterranean basin (EM). For P. oceanica, ArnaudHaond et al., (2007)

identified three clusters in the Mediterranean Sea using 34 populations:

Western (W), Central (C) and Eastern (E). Serra et al., (2010) subdivided the Central cluster in two regions (C.I. and C.II.) by means of a finer genetic structure analysis using 27 populations of P. oceanica (Fig. 1). From these studies we estimated the mean of the different measures of genetic diversity for each genetic cluster: allelic richness (Â, number of alleles), expected heterozygosity (He, gene diversity) and private alleles, following procedures in Chefaoui et al., (2017a)

and Chefaoui & Serrão (2017)

(see Figs. S2.1 and S2.2 in

Supporting Information S2). Finally, we estimated the number of cells of suitable habitat in each genetic region through the various scenarios of climate change.


Accepted Article

RESULTS We retained three uncorrelated variables (r ≤ |0.70|, p < 0.001) for our models: the minimum and maximum SST, and salinity. The evaluation of the ensembles’ performance gave the following mean results: AUC = 0.917, TSS = 0.712, sensitivity = 94.783, and specificity = 76.434 for C. nodosa; and AUC = 0.933, TSS = 0.756, sensitivity = 96.377, and specificity = 79.187 for P. oceanica. Scores obtained for the entire set of models computed are available in Table 2. The clamping masks only detected differences between the values of the training range and the future scenarios regarding one variable and one scenario (RCP 8.5 by 2100). This uncertainty can be considered low since it was limited to the Red Sea, where the species are not present (Fig. S2.3 in Supporting Information S2). The most relevant variable on average for the ensemble of the present distribution of P. oceanica was minimum SST (mean of scores = 0.438), followed by salinity (0.230), and maximum SST (0.184). For C. nodosa, maximum SST was the most relevant variable (mean scores = 0.574), followed by minimum SST (0.521), and salinity (0.289). The ensembles for the present time approximate the actual distribution of the species (Fig. S2.4 in Supporting Information S2). The future projections showed a progressive decline of the suitable habitat with high probability of occurrence for both seagrass species across time (Figs. 2, 3, and Fig. S2.5 in Supporting Information S2). P. oceanica showed a steeper loss of suitable habitat than C. nodosa, which also had a major compensation with gained and stable areas (Fig. 4). For P. oceanica, no coast with high probability was found outside the Mediterranean Sea in any of the future scenarios. By 2100, the ensembles showed a possible disappearance, i.e. a possible extinction, of P. oceanica and a relevant decline of C. nodosa meadows under the most pessimistic greenhouse gas emissions


Accepted Article

scenario (RCP 8.5) (Figs. 2, 3 and 4). According to our models, P. oceanica might lose 70-75 % of suitable habitat by 2050 when compared with its extent under present conditions. By 2100, P. oceanica could lose 82 % of its suitable habitat under the lower emissions (RCP 2.6) scenario, and the totality of its meadows under the worst-case (RCP 8.5) scenario (Figs. 2, 3 and 4). The predicted persistence of meadows was low for P. oceanica (27.4 % under RCP 2.6 by 2050 and 17.2 % by 2100), and the gained habitat only reached 2.7 % by 2050 and 0.8 % by 2100. For Cymodocea nodosa we also found a relevant loss of suitable habitat in comparison with the model for the present; the loss varied between 20.8 % by 2050 and 46.5 % under the worst-case scenario by 2100. However, C. nodosa showed a greater gained and conserved area than P. oceanica. The models for C. nodosa estimated an increase in suitable habitat from 34.9 % to 38.8 % by 2050, and around 32 % by 2100. The stable habitat would vary between 40.4 % by 2050 and 21.5 - 36.2 % by 2100, depending on the scenario (Fig. 4). The predicted disappearance of habitat under climate change would affect considerably all genetic regions identified for P. oceanica, even in the best-case scenario by 2050 (Figs. 2, 3 and S3). The C.I. and C.II. regions were specially affected, showing a 92 – 83.4 % and 45.1 – 63.4 % decrease in suitable area by 2050, respectively, or a nearly total disappearance by 2100 (Figs. 2, 3 and S3). The model did not find any relevant habitat gain in any of the regions of P. oceanica. According to all future scenarios, Cymodocea nodosa would decrease its extension in the two Mediterranean genetic regions, eastern (EM) and western (WM) (Figs. 2 and 3). The seagrass could reach a 68.2 % decrease in EM, and 60 % in WM by 2100 (RCP 8.5). In the Atlantic regions, C. nodosa could experience some increase in relation to its present suitable


Accepted Article

habitat, except under the scenario RCP 8.5 by 2100, under which the Iberian Atlantic region (AH) could disappear and the African Atlantic range (AL) would decrease its extension. By 2050, C. nodosa would increase by 36.52 - 56.9 % in the AL region; and by 25 - 59 % in the AH region, depending on the scenario. Under the RCP 2.6 by 2100, C. nodosa would increase a 55 % in the AL region, and 62 % in the AH region. In contrast, a 27.7 % decrease in the AL region, and a 98.8 % decrease in the AH region were predicted under the more severe RCP 8.5 scenario by 2100.

DISCUSSION Mediterranean warming will lead to significant population declines and genetic loss of Posidonia oceanica and Cymodocea nodosa. This loss is so severe that it may involve the complete loss of habitat and genetic diversity of P. oceanica by 2100, involving its functional and possibly species extinction, under the most severe scenario of climate change, and a 70% decrease in habitat under the best-case scenario by 2050. The Mediterranean meadows of C. nodosa will also suffer an important range contraction (ranging from 20.8 % to 46.5 %), only partially compensated by increase in future new habitat in the north-western African coast, the Black Sea and the Adriatic Sea. Overall genetic diversity in the Mediterranean populations of both seagrasses could be eroded under all scenarios by 2100, together with the north-eastern Atlantic populations of C. nodosa if the most pessimistic scenario is realized. Given the characteristic of these two species as foundational species, our results are alarming for the future of the Mediterranean marine ecosystems. According to the low probability of occurrence found for Posidonia oceanica in the Atlantic Ocean, this endemic This article is protected by copyright. All rights reserved.

Accepted Article

will continue confined to the Mediterranean Sea in the future. The unsuitable environmental conditions in the Atlantic, along with the oceanographic barrier that the Strait of Gibraltar entails for dispersal between the Atlantic and the Mediterranean Sea for many species (see Patarnello et al., 2007; Burrows et al., 2011) , particularly those with buoyant propagules such as Posidonia oceanica, imply that the expansion of this seagrass species outside the Mediterranean will be unlikely. In addition, the realized connectivity among P. oceanica meadows is lower than its potential estimations (Jahnke et al., 2017) . Thus, its endemic condition in the Mediterranean, experiencing faster warming rate than the rest of oceans (Coll et al., 2010; Jordà et al., 2012; Templado, 2014) , renders P. oceanica extremely vulnerable to ocean warming, with a substantial extinction risk. The low extension of gained or stable habitat in comparison to the predicted range retraction will probably lead to the disappearance of the current genetically rich populations of P. oceanica. The central Mediterranean regions (C.I. and C.II.) harbor nowadays the highest allelic richness (ArnaudHaond et al., 2007; Serra et al., 2010) probably due to their persistence since the Last Glacial Maximum (Chefaoui et al., 2017a). These central Mediterranean regions encompass now a relatively small habitat and could be especially affected by warming regardless of the future scenario considered. During the Last Glacial Maximum, glacial refugia were probably located in the southern Mediterranean (Chefaoui et al., 2017a) . Thus, the genetic diversity preserved throughout time in those southern meadows could be lost under projected warming. Even under the case of persistence of some northern populations, as projected by 2050 and the best-case scenario by 2100, warming will result in depleted gene pools which could make the species more vulnerable to invasions or diseases. In the case of Cymodocea nodosa, the loss of habitat across its entire range will not be counterbalanced by the eventual gain in the north-western African coast, the Adriatic and This article is protected by copyright. All rights reserved.

Accepted Article

the Black Sea. The eastern-Mediterranean region (EM), retaining the highest genetic diversity nowadays, was inferred to have been a likely glacial refugium together with its Atlantic populations at its southernmost limit (AL) in Africa-Canaries (Chefaoui & Serrão, 2017). The genetically richest populations found in the western (WM) and eastern (EM) Mediterranean regions would suffer a 60% and 68.2% decrease in habitat by 2100, respectively, under the most pessimistic scenario. This decline would be concentrated in the southern Mediterranean populations and would lead to the loss of unique gene pools in this area (Alberto et al., 2008) . Western Mediterranean populations could colonize future new habitats in the Black and the Adriatic Seas. Although this putative spread would be essential for the preservation of a portion of gene pools, founder events during range expansion (as happened during colonization of the Canary islands, Alberto et al., 2006) could also erode present genetic diversity. The Atlantic meadows of C. nodosa will be favoured under all future scenarios, except for the most pessimistic one by 2100. This could represent an increase of around 56 % and 60 % in habitat in the Atlantic genetic regions along the southernmost Africa-Canaries range (AL) and Iberia-Morocco (AH), respectively. The new suitable Atlantic habitats could be colonized mainly from populations pertaining to the AL region, and to a lesser extent by those of the AH region. Despite that increase in the Atlantic, the southern range edge (in Senegal) would be at risk of being lost, together with the entire AH region under RCP 8.5. No substantial northwards expansion in the Atlantic leading edge would occur in the Iberian Peninsula, where prevalent upwelling leads to seasonally cold waters around much of the Portuguese and Spanish coast. Interestingly, the most pessimistic scenario by 2100 would render a future disjoint distribution of C. nodosa meadows. These stable or gained areas would be located in the AL region (Canary Islands and corresponding mainland African This article is protected by copyright. All rights reserved.

Accepted Article

coasts) and in the northern Mediterranean and Black Sea. These populations are currently very differentiated and disconnected in the present (Alberto et al., 2008; Masucci et al., 2012) , and in the absence of stepping stone habitats would become even much less connected than under present time, which could eventually lead to a future vicariant process supporting, if persisting over a long time, eventual possible speciation. The Mediterranean seagrass decline predicted by our ensemble is consistent with previous studies predicting a decline of Posidonia oceanica under climate change (Marbà & Duarte, 2010; Jordà et al., 2012; Olsen et al., 2012; Marbà et al., 2015) . Although both seagrass species will be affected by warming, we have found that P. oceanica is more vulnerable to future climate change, with a risk of extinction for this species under severe climate change scenarios. This is consistent with entirely independent predictions that P. oceanica could face functional extinction (density decline by more than 90%) in the Western Mediterranean by 2050 (Jordà et al., 2012) . Differences found between the two seagrass species regarding their SST tolerance could be attributed to their different evolutionary origins. They differ in photosynthetic responses to heat stress, consistent with their evolutionary history: the gene expression of C. nodosa (evolved in the genus Cymodocea, which is typically tropical) was less affected than that of the temperate P. oceanica under heat stress (Marín-Guirao et al., 2016) . Similarly, P. oceanica seedlings were more vulnerable to warming than those of C. nodosa during experiments in mesocosms (Olsen et al., 2012) . Experimental heat stress exposure has been recently found to reduce leaf growth rates, but to trigger P. oceanica’s flowering (Ruiz et al., 2017). A relationship between flowering induction and thermal stress has also been found using field observations of these flowering rare events (Diaz-Almela et al., 2007). However, it remains unanswered if thermally induced sexual reproduction could enhance plasticity and adaptive capacity of P. This article is protected by copyright. All rights reserved.

Accepted Article

oceanica by the generation of genetic diversity, and increase its persisting chances under ocean warming. The projections presented here are subject to common uncertainties related to niche modelling. Uncertainty concerning the occurrence data available was already detected in the North African coast for Cymodocea nodosa (Chefaoui et al., 2016; Chefaoui & Serrão, 2017), and in a long portion of coast located mostly in the eastern Mediterranean for P. oceanica (Telesca et al., 2015; Chefaoui et al., 2017a) . We found low uncertainty regarding the difference in range between the variables used to model the future distribution and those used to calibrate the models, since only the Red Sea - a region not occupied by the species seemed affected. The variability among the General Circulation Models used to generate projections of potential distribution also contributes to uncertainty (Buisson et al., 2010; Chefaoui & Serrão 2017), which we reduced by using an ensemble of models. Other uncertainties such as the resolution of the available marine variables, the threshold used to produce the binary transformations, and the election of the modeling algorithm are common factors affecting the predictions of marine species (see Chefaoui et al., 2017a; Chefaoui & Serrão 2017) . Mediterranean warming, together with other stressors, could produce an irreversible regression of Posidonia oceanica (Jordà et al., 2012) . In fact, the meadows of P. oceanica are already suffering a wide-spread decline as a consequence of the effects of a range of anthropogenic pressures including coastal development, dredging, pollution, fish farming, moorings, and invasive species (e.g., Marbà et al., 1996; Boudouresque et al., 2009; Marbà et al., 2014; Oprandi et al., 2014; Holon et al., 2015; Telesca et al., 2015; Burgos et al., 2017) . Indeed, the joint effect of anthropogenic stresses and Mediterranean warming could


Accepted Article

lead to an earlier decline than predicted by our models, as predicted by Jordà et al., (2012). Despite the difficulties in distinguishing the effects produced by seawater warming from those produced by other anthropogenic stressors at some areas, there is already evidence of P. oceanica loss in relatively pristine areas of the Western Mediterranean Sea. The impact of warming on seagrass shoot mortality and net rates of shoot population loss in the Balearic Islands (Marbà & Duarte, 2010), included evidence of warming-associated decline of P. oceanica in a Marine Protected Area (Cabrera National Park, Spain), where other anthropogenic stressors are minimal. Indeed, Jordà et al., (2012) forecasted the decline of P. oceanica in the Balearic Islands at rates similar to those predicted here, and found that removing anthropogenic stressors had a small effect on the projected losses, that is insufficient to avoid functional extinction of the Posidonia meadows with warming. The recovery rate of the meadows of P. oceanica, one of the slowest-growing angiosperms in the planet, is so slow that it would require centuries to millennia (e.g., Duarte 1995). In contrast, C. nodosa populations can recover over decades (Duarte, 1995), and a relative replacement of the declining P. oceanica by C. nodosa has already been reported at some regions of the Mediterranean Sea (Burgos et al., 2017). The finding that Mediterranean warming may lead to the extinction of Posidonia oceanica, a major foundation species in the Mediterranean Sea, is of major concern. In addition to its ecological role as foundation species, Posidonia oceanica has developed the largest organic carbon stocks in their sediments among all seagrasses (Fourqurean et al., 2012), much of which may be emitted as CO2 upon loss of the seagrass cover (Lovelock et al., 2017) . It has been suggested that habitat fragmentation erodes genetic variation and increases interpopulation genetic divergence (Young et al., 1996). The loss of genetic diversity has also shown to be related to a lower tolerance to environmental changes (Jump This article is protected by copyright. All rights reserved.

Accepted Article

et al., 2009). Prior to extinction, the erosion of genetic diversity predicted here could diminish the potential for adaptation to climate change , which could, along with the pressures derived from other anthropogenic stressors, accelerate the predicted rate of decline. In field experiments, genetic diversity has been found to enhance the resistance of Zostera marina to disturbance by grazing (Hughes & Stachowicz 2004) and to increase its rate of recovery after perturbation (Reusch et al., 2005). However, further empirical research needs to be done to establish the real adaptive capacity of seagrasses to ocean warming. Shall the goals of the Paris Agreement be met, particularly the most ambitious goal to contain warming at 1.5 oC (~ RCP 2.6; Meinshausen et al., 2011) , extinction could be avoided, but the safety margin in doing so leaves no margin for the added loss resulting from anthropogenic pressure and genetic decline. In the Ligurian Sea, several studies combining modelling and historical data have shown that protection measures undertaken decelerated the decline of P. oceanica (Burgos et al., 2017), and that certain recovery of the extents of its meadows is possible (Oprandi et al., 2014). However, total recovery of P. oceanica meadows after correcting the impact could happen at a very low rate and be prolonged in time to nearly 100 years (González-Correa et al., 2005). Hence, avoiding the extinction of Mediterranean Posidonia oceanica meadows require that removal of local stressors be achieved while reduced emissions, and possibly CO2 capture technology, is deployed to avoid global warming exceeding 1.5 oC.


Accepted Article

ACKNOWLEDGEMENTS We thank anonymous reviewers for their useful comments and suggestions. This study received funds from a Pew Marine Fellowship (USA) to EAS and Fundação para a Ciência e a Tecnologia (FCT, Portugal) through postdoctoral fellowship “SFRH/BPD/85040/2012” to RMC. The authors have no conflict of interest to declare.

REFERENCES Alberto F, Arnaud-Haond S, Duarte CM, Serrão EA (2006) Genetic diversity of a clonal angiosperm near its range limit: the case of Cymodocea nodosa at the Canary Islands. Marine Ecology Progress Series, 309, 117–129. Alberto F, Massa S, Manent P, Diaz-Almela E, Arnaud-Haond S, Duarte CM, Serrão EA (2008) Genetic differentiation and secondary contact zone in the seagrass Cymodocea nodosa across the Mediterranean-Atlantic transition region. Journal of Biogeography, 35, 1279– 1294. Allouche O, Tsoar A, Kadmon R (2006) Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). Journal of Applied Ecology, 43, 1223– 1232. Arnaud-Haond S, Migliaccio M, Diaz-Almela E et al. (2007) Vicariance patterns in the Mediterranean Sea: East-west cleavage and low dispersal in the endemic seagrass Posidonia oceanica. Journal of Biogeography, 34, 963–976. Assis J, Coelho NC, Lamy T, Valero M, Alberto F, Serrão EA (2016) Deep reefs are climatic refugia for genetic diversity of marine forests. Journal of Biogeography, 43, 833–844. Assis J, Berecibar E, Claro B, Alberto F, Reed D, Raimondi P, Serrão EA (2017) Major shifts at the range edge of marine forests: the combined effects of climate changes and limited dispersal. Scientific Reports, 7, 44348. Assis J, Araújo MB, Serrão EA (2018) Projected climate changes threaten ancient refugia of kelp forests in the North Atlantic. Global Change Biology, 24, e55–e66. Bates AE, Pecl GT, Frusher S et al. (2014) Defining and observing stages of climate-mediated range shifts in marine systems. Global Environmental Change, 26, 27–38. Boudouresque CF, Bernard G, Pergent G, Shili A, Verlaque M (2009) Regression of


Accepted Article

Mediterranean seagrasses caused by natural processes and anthropogenic disturbances and stress: a critical review. Botanica Marina, 52, 395–418. Buisson L, Thuiller W, Casajus N, Lek S, Grenouillet G (2010) Uncertainty in ensemble forecasting of species distribution. Global Change Biology, 16, 1145–1157. Buonomo R, Chefaoui RM, Bermejo Lacida R, Engelen AH, Serrão EA, Airoldi L (2018) Predicted extinction of unique genetic diversity in marine forests of Cystoseira spp. Marine Environmental Research, 138, 119–128. Burgos E, Montefalcone M, Ferrari M, Paoli C, Vassallo P, Morri C, Bianchi CN (2017) Ecosystem functions and economic wealth: Trajectories of change in seagrass meadows. Journal of Cleaner Production, 168, 1108–1119. Burrows MT, Schoeman DS, Buckley LB et al. (2011) The pace of shifting climate in marine and terrestrial ecosystems. Science, 334, 652–655. Chefaoui RM, Serrão EA (2017) Accounting for uncertainty in predictions of a marine species: integrating population genetics to verify past distributions. Ecological Modelling, 359, 229–239. Chefaoui RM, Assis J, Duarte CM, Serrão EA (2016) Large-scale prediction of seagrass distribution integrating landscape metrics and environmental factors: the case of Cymodocea nodosa (Mediterranean–Atlantic). Estuaries and Coasts, 39, 123–137. Chefaoui RM, Duarte CM, Serrão EA (2017a) Palaeoclimatic conditions in the Mediterranean explain genetic diversity of Posidonia oceanica seagrass meadows. Scientific Reports, 7, 2732. Chefaoui RM, Casado-Amezúa P, Templado J (2017b) Environmental drivers of distribution and reef development of the Mediterranean coral Cladocora caespitosa. Coral Reefs, 36, 1195–1209. Chen I-C, Hill JK, Ohlemuller R, Roy DB, Thomas CD (2011) Rapid range shifts of species associated with high levels of climate warming. Science, 333, 1024–1026. Coll M, Piroddi C, Steenbeek J et al. (2010) The biodiversity of the Mediterranean Sea: estimates, patterns, and threats. PLoS ONE, 5, e11842. Collins W, Bellouin N, Doutriaux-Boucher M et al. (2008) Evaluation of HadGEM2 model. Meteorological Office Hadley Centre, Technical Note, 74, 47 pp. Coma R, Ribes M, Serrano E, Jiménez E, Salat J, Pascual J (2009) Global warming-enhanced stratification and mass mortality events in the Mediterranean. Proceedings of the National Academy of Sciences of the United States of America, 106, 6176–6181. Diaz-Almela E, Marbà N, Duarte CM (2007) Consequences of Mediterranean warming events


Accepted Article

in seagrass (Posidonia oceanica) flowering records. Global change biology, 13, 224–235. Duarte CM (1995) Submerged aquatic vegetation in relation to different nutrient regimes. Ophelia, 41, 87–112. Duarte CM (2002) The future of seagrass meadows. Environmental Conservation, 29, 192– 206. Duarte CM (2011) Seagrass meadows. at Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Ecological Conservation, 24, 38–49. Fourqurean JW, Duarte CM, Kennedy H et al. (2012) Seagrass ecosystems as a globally significant carbon stock. Nature Geoscience, 5, 505–509. González-Correa JM, Bayle JT, Sánchez-Lizaso JL, Valle C, Sánchez-Jerez P, Ruiz JM (2005) Recovery of deep Posidonia oceanica meadows degraded by trawling. Journal of Experimental Marine Biology and Ecology, 320, 65–76. Holon F, Mouquet N, Boissery P, Bouchoucha M, Delaruelle G, Tribot A-S, Deter J (2015) Fine-scale cartography of human impacts along French Mediterranean coasts: a relevant map for the management of marine ecosystems. PLoS One, 10, e0135473. Hughes AR, Stachowicz JJ (2004) Genetic diversity enhances the resistance of a seagrass ecosystem to disturbance. Proceedings of the National Academy of Sciences of the United States of America, 101, 8998–9002. Jahnke M, Casagrandi R, Melià P, Schiavina M, Schultz ST, Zane L, Procaccini G (2017) Potential and realized connectivity of the seagrass Posidonia oceanica and their implication for conservation. Diversity and Distributions, 23, 1423–1434. Jordà G, Marbà N, Duarte CM (2012) Mediterranean seagrass vulnerable to regional climate warming. Nature Climate Change, 2, 821–824. Jump AS, Marchant R, Peñuelas J (2009) Environmental change and the option value of genetic diversity. Trends in Plant Science, 14, 51–58. Lejeusne C, Chevaldonné P, Pergent-Martini C, Boudouresque CF, Pérez T (2010) Climate change effects on a miniature ocean: the highly diverse, highly impacted Mediterranean Sea. Trends in Ecology & Evolution, 25, 250–260. Liaw A, Wiener M (2002) Classification and Regression by randomForest. R news, 2, 18–22. Lovelock CE, Atwood T, Baldock J et al. (2017) Assessing the risk of carbon dioxide emissions from blue carbon ecosystems. Frontiers in Ecology and the Environment, 15, 257–265. Marbà N, Duarte CM (2010) Mediterranean warming triggers seagrass (Posidonia oceanica) This article is protected by copyright. All rights reserved.

Accepted Article

shoot mortality. Global Change Biology, 16, 2366–2375. Marbà N, Duarte C, Cebrián J, Gallegos M, Olesen B, Sand-Jensen K (1996) Growth and population dynamics of Posidonia oceanica on the Spanish Mediterranean coast: elucidating seagrass decline. Marine Ecology Progress Series, 137, 203–213. Marbà N, Díaz-Almela E, Duarte CM (2014) Mediterranean seagrass (Posidonia oceanica) loss between 1842 and 2009. Biological Conservation, 176, 183–190. Marbà N, Jordà G, Agustí S, Girard C, Duarte CM (2015) Footprints of climate change on Mediterranean Sea biota. Frontiers in Marine Science, 2, 56. Marín-Guirao L, Ruiz JM, Dattolo E, Garcia-Munoz R, Procaccini G (2016) Physiological and molecular evidence of differential short-term heat tolerance in Mediterranean seagrasses. Scientific Reports, 6, 28615. Masucci AP, Arnaud-Haond S, Eguíluz VM, Hernández-García E, Serrão EA (2012) Genetic flow directionality and geographical segregation in a Cymodocea nodosa genetic diversity network. EPJ Data Science, 1, 11. Meinshausen M, Smith SJ, Calvin K et al. (2011) The RCP greenhouse gas concentrations and their extensions from 1765 to 2300. Climatic Change, 109, 213–241. Moss R, Babiker W, Brinkman S et al. (2008) Towards new scenarios for the analysis of emissions: climate change, impacts and response strategies. Intergovernmental Panel on Climate Change Secretariat (IPCC), Geneva, Switzerland. Neiva J, Assis J, Coelho NC, Fernandes F, Pearson GA, Serrão EA (2015) Genes left behind: Climate change threatens cryptic genetic diversity in the canopy-forming seaweed Bifurcaria bifurcata. PLoS ONE, 10, e0131530. Olsen YS, Sánchez-Camacho M, Marbà N, Duarte CM (2012) Mediterranean seagrass growth and demography responses to experimental warming. Estuaries and Coasts, 35, 1205– 1213. Oprandi A, Montefalcone M, Vacchi M et al. (2014) Combining modelling and historial data to define the status of Posidonia oceanica meadows. In: 5th Mediterranean Symposium on Marine Vegetation (eds Langar H, Bouafif C, Ouerghi A), pp. 119–124. UNEP/MAPeRAC/SPA, Tunis. Orth R, Carruthers T, Dennison WC et al. (2006) A global crisis for seagrass ecosystems. Bioscience, 56, 987–996. Patarnello T, Volckaert FAMJ, Castilho R (2007) Pillars of Hercules: Is the AtlanticMediterranean transition a phylogeographical break? Molecular Ecology, 16, 4426– 4444.


Accepted Article

Pecl GT, Araújo MB, Bell JD et al. (2017) Biodiversity redistribution under climate change: Impacts on ecosystems and human well-being. Science, 355, eaai9214. Poloczanska ES, Brown CJ, Sydeman WJ et al. (2013) Global imprint of climate change on marine life. Nature Climate Change, 3, 919–925. Raddatz TJ, Reick CH, Knorr W et al. (2007) Will the tropical land biosphere dominate the climate–carbon cycle feedback during the twenty-first century? Climate Dynamics, 29, 565–574. Reusch TBH, Ehlers A, Hämmerli A, Worm B (2005) Ecosystem recovery after climatic extremes enhanced by genotypic diversity. Proceedings of the National Academy of Sciences of the United States of America, 102, 2826 LP-2831. Rozenfeld AF, Arnaud-Haond S, Hernandez-Garcia E, Eguiluz VM, Serrao EA, Duarte CM (2008) Network analysis identifies weak and strong links in a metapopulation system. Proceedings of the National Academy of Sciences, 105, 18824–18829. Ruiz JM, Marín-Guirao L, García-Muñoz R et al. (2017) Experimental evidence of warminginduced flowering in the Mediterranean seagrass Posidonia oceanica. Marine Pollution Bulletin. (in press) doi.org/10.1016/j.marpolbul.2017.10.037 Serra IA, Innocenti AM, Di Maida G et al. (2010) Genetic structure in the Mediterranean seagrass Posidonia oceanica: Disentangling past vicariance events from contemporary patterns of gene flow. Molecular Ecology, 19, 557–568. Sunday JM, Bates AE, Dulvy NK (2012) Thermal tolerance and the global redistribution of animals. Nature Climate Change, 2, 686–690. Telesca L, Belluscio A, Criscoli A et al. (2015) Seagrass meadows (Posidonia oceanica) distribution and trajectories of change. Scientific Reports, 5, 12505. Templado J (2014) Future trends of mediterranean biodiversity. In: The Mediterranean Sea: Its History and Present Challenges, pp. 479–498. Springer Netherlands, Dordrecht. Thuiller W, Georges D, Engler R, Breiner F (2016) biomod2: Ensemble platform for species distribution modeling. Tyberghein L, Verbruggen H, Pauly K, Troupin C, Mineur F, De Clerck O (2012) Bio-ORACLE: a global environmental dataset for marine species distribution modelling. Global Ecology and Biogeography, 21, 272–281. Voldoire A, Sanchez-Gomez E, Salas y Mélia D et al. (2013) The CNRM-CM5.1 global climate model: description and basic evaluation. Climate Dynamics, 40, 2091–2121. Young A, Boyle T, Brown T (1996) The population genetic consequences of habitat fragmentation for plants. Trends in Ecology & Evolution, 11, 413–418.


Accepted Article

TABLES Table 1 Description of the five General Circulation Models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5) used to predict the future distribution of Cymodocea nodosa and Posidonia oceanica. We used data from scenarios RCP 2.6 and RCP 8.5 to produce the future projections by 2050 and 2100.

GCM

Reference

Original resolution (degrees)

HadGEM2-ES

Collins et al., 2008

MPI-ESM-MR

Raddatz et al., 2007

1, 1

IPSL-CM5A-LR

http://icmc.ipsl.fr/

1, 1

NASA/GISS

http://data.giss.nasa.gov/modelE/ar5

CNRM-CM5

Voldoire et al., 2013

1, 0.837

2.5, 2

1, 1

Table 2 Mean validation scores of the 20 models obtained with each technique for each species. The best models are highlighted in bold. (s.d.= standard deviation). AUC: the area under the receiver operating characteristic curve; TSS: the true skill statistic; GLM: generalized linear model; GBM: generalized boosting model; GAM: generalized additive model; FDA: flexible discriminant analysis; MARS: multiple adaptive regression splines; RF: randomForest.


Accepted Article

Model

AUC Mean (± s.d.)

TSS Mean (± s.d.)

Sensitivity (± s.d.)

Specificity (± s.d.)

71.508 (± 5.692)

Cymodocea nodosa

GLM

0.824 (± 0.023)

0.571 (± 0.051)

85.797 (± 5.126)

GBM

0.886 (± 0.019)

0.649 (± 0.049)

84.420 (± 4.846)

80.635 (± 5.307)

GAM

0.834 (± 0.032)

0.582 (± 0.066)

81.812 (± 7.038)

76.508 (± 4.720)

FDA

0.856 (± 0.018)

0.611 (± 0.042)

82.101 (± 4.128)

79.206 (± 3.991)

0.618 (± 0.053)

80.652 (± 5.147)

81.429 (± 4.980)

0.892 (± 0.016)

0.663 (± 0.032)

86.232 (± 4.211)

80.159 (± 2.874)

GLM

0.894 (± 0.011)

0.703 (± 0.019)

98.344 (± 0.950)

72.188 (± 2.250)

GBM

0.916 (± 0.013)

0.721 (± 0.020)

95.142 (± 2.482)

77.214 (± 3.145)

0.703 (± 0.020)

97.397 (± 3.496)

72.195 (± 4.298) 72.681 (± 2.875)

MARS

RF

0.860 (± 0.026)

Posidonia oceanica

GAM

0.892 (± 0.013)

FDA

0.895 (± 0.013)

0.695 (± 0.021)

96.256 (± 2.941)

MARS

0.901 (± 0.014)

0.706 (± 0.025)

94.414 (± 2.561)

75.755 (± 3.860)

RF

0.919 (± 0.012)

0.735 (± 0.023)

95.008 (± 3.340)

78.154 (± 4.170)


Accepted Article

FIGURE CAPTIONS Fig. 1 Occurrence data used to develop niche models of (a) Posidonia oceanica, and (b) Cymodocea nodosa (in blue). The maps also show the genetic populations (orange squares) and genetic regions identified by Arnaud-Haond et al., (2007)

and Serra et al., (2010) :

Western, Central (C. I and C. II) and Eastern; and by Alberto et al., (2008)

: high-latitude

Atlantic (AH), low-latitude Atlantic (AL), Western Mediterranean (WM), and Eastern Mediterranean (EM). Fig. 2 Predicted habitat change under contrasting scenarios (RCP 2.6 and RCP 8.5) by 2100 for Posidonia oceanica and Cymodocea nodosa. The maps show the predicted future loss, stable, and gained habitat. The genetic regions identified by Arnaud-Haond et al., (2007) and Serra et al., (2010)

for P. oceanica, and by Alberto et al., (2008)

for C. nodosa are

also depicted. C. I. and C. II.: Central I and Central II subregions of the Central Mediterranean Sea; AH: high-latitude Atlantic; AL: low-latitude Atlantic; WM: Western Mediterranean; EM: Eastern Mediterranean. Fig. 3 Suitable habitat for Posidonia oceanica and Cymodocea nodosa under present and future conditions in each one of the genetic regions. The extension of the habitat is expressed as the number of cells ( ~ 9.2 km) found for each region after a binary classification. Fig. 4 Habitat change between the ensembles predicted under present and future scenarios (RCP 2.6 and 8.5) by 2050 and 2100 for Posidonia oceanica and Cymodocea nodosa. The change is expressed as the number of cells ( ~ 9.2 km) with a probability of presence above the threshold used for the binary clasification of each ensemble. Colors used correspond to those used in Fig. 2.


Accepted Article This article is protected by copyright. All rights reserved.

