Biogeographical network analysis of plant species distributionin the Mediterranean region

Maxime Lenormand,1, ∗ Guillaume Papuga,2, 3, ∗ Olivier Argagnon,2 Maxence

Soubeyrand,1 Guilhem De Barros,2 Samuel Alleaume,1 and Sandra Luque1

1Irstea, UMR TETIS, 500 rue JF Breton, 34093 Montpellier, France2Conservatoire botanique national mediterraneen de Porquerolles, Parc scientifique Agropolis,

2214 boulevard de la Lironde, 34980 Montferrier sur Lez, France3UMR 5175 CEFE, CNRS, 1919 route de Mende, 34293 Montpellier cedex 5, France

The delimitation of bioregions helps to understand historical and ecological drivers of speciesdistribution. In this work, we performed a network analysis of the spatial distribution patternsof plants in south of France (Languedoc-Roussillon and Provence-Alpes-Cote d’Azur) to analyzethe biogeographical structure of the French Mediterranean flora at different scales. We used anetwork approach to identify and characterize biogeographical regions, based on a large databasecontaining 2.5 million of geolocalized plant records corresponding to more than 3,500 plant species.This methodology is performed following five steps, from the biogeographical bipartite networkconstruction, to the identification of biogeographical regions under the form of spatial networkcommunities, the analysis of their interactions and the identification of clusters of plant species basedon the species contribution to the biogeographical regions. First, we identified two sub-networks thatdistinguish Mediterranean and temperate biota. Then, we separated eight statistically significantbioregions that present a complex spatial structure. Some of them are spatially well delimited, andmatch with particular geological entities. On the other hand fuzzy transitions arise between adjacentbioregions that share a common geological setting, but are spread along a climatic gradient. Theproposed network approach illustrates the biogeographical structure of the flora in southern France,and provides precise insights into the relationships between bioregions. This approach sheds light onecological drivers shaping the distribution of Mediterranean biota: the interplay between a climaticgradient and geological substrate shapes biodiversity patterns. Finally this work exemplifies whyfragmented distributions are common in the Mediterranean region, isolating groups of species thatshare a similar eco-evolutionary history.

INTRODUCTION

The delimitation of biogeographical regions orbioregions based on the analysis of their biota has beena founding theme in biogeography, from the pioneerwork of Wallace (1876), Murray (1866) or Wahlen-berg (1812) to the most recent advances of Cheruvelilet al. (2017); Ficetola et al. (2017). Describing spatialpatterns of biodiversity has appeared fundamental tounderstand the historical diversification of biota, andgain a better understanding of ecological factors thatimprint spatial patterns of biodiversity (Graham &Hijmans, 2006; Ricklefs, 2004). Additionally, it hasbecome a key element in the identification of spatialconservation strategies (Funk et al., 2002; Mikolajczaket al., 2015; Rushton et al., 2004). To divide a giventerritory into meaningful and coherent bioregions, theoverall aim is to minimize the heterogeneity in taxo-nomic composition within regions, while maximizingdifferences between them (Kreft & Jetz, 2010; Stod-dart, 1992). Although such delineation of bioregionshas been based for a long time on expert knowledge ofqualitative data collection the increasing availabilityof species-level distribution data and recent techno-

∗ Corresponding authors: maxime.lenormand@irstea.fr & guil-laume.papuga@gmail.com who contributed equally to thiswork.

logical advances have allowed for the development ofmore rigorous frameworks (Kreft & Jetz, 2010). Mul-tivariate methods, such as hierarchical clustering algo-rithms, have thus been successfully applied in a widerange of studies focused on a variety of organisms, un-der very different spatial scale (from regional to world-wide perspective). Yet, the production of detailedcartographic outputs portraying the differentiation ofvegetation into distinct homogeneous bioregions re-mains difficult, especially where spatial heterogeneityof assemblages is associated with complex environ-mental gradients (Mikolajczak et al., 2015). Besides,the identification of meaningful and coherent biore-gions represents only one step of the biogeographicalregionalizations (Morrone, 2018). It is also crucial topropose new metrics to quantify the relationship be-tween bioregions and to analyze species and spatialrelationships.

Some regions of the world oppose inherent diffi-culties due to their highly diversified biota, reflect-ing complex eco-evolutionary processes. The Mediter-ranean basin is one of the largest and most impor-tant biodiversity hotspots in the world (Blondel et al.,2010; Myers et al., 2000). This region hosts about25,000 plant species representing 10% of the world’stotal floristic richness concentrated on only 1% of theworld’s surface (Greuter, 1991). Additionally, a highlevel of narrow endemism is a major feature of thisbiome (Thompson, 2005). Endemism and richness re-

arX

iv:1

803.

0527

5v3

[q-

bio.

PE]

7 J

an 2

019

2

sult in a very heterogeneous region, whose compre-hension of spatial patterns of plant distribution isclue to get better insights into past and actual pro-cesses shaping biodiversity (Quezel, 1999). The onsetof the Mediterranean climate during the Pliocene andthe diverse glacial periods of the Pleistocene (Quezel& Medail, 2004) have shaped the most importantphases of plant evolution since the Tertiary (Thomp-son, 2005). Additionally, due to a long history of hu-man presence, contemporary flora has been widely in-fluenced by human-mediated dispersal, land-use andother pressures (Dahlin et al., 2014; Fenu et al., 2014).The French Mediterranean area stretches from thePyrenees in the south-west to the slopes of the Mar-itime and Ligurian Alps in the east. It encompassesthree zones highlighted as glacial refugia (Medail &Diadema, 2009), and the eastern sector represents oneof the ten main biodiversity hotspots in the Mediter-ranean area (Medail & Quezel, 1997). This area rep-resents the northern limit of the Mediterranean cli-mate in the western basin, and thus constitutes a cli-matic transition from a Mediterranean zone that hasa summer drought to a temperate zone less prone tosummer drought (Walter & Breckle, 1991 1994). Ona finer scale, the climate is more complex with sev-eral subtypes and intricated boundaries (Joly et al.,2010; Tassin, 2017). Several works have tried to mapthe distribution of biogeographical entities. To date,no statistical analysis had been ran to tackle thoseexpert-based maps with up-to-date plant records, inorder to test their reliability.

In order to depict spatial structure in such a com-plex regional flora, a large dataset is required. Whilethe level of diversity and complexity of such datasetmay appear overwhelming at first glance, the emer-gence of network-based approaches has opened newpaths for identifying and delimiting bioregions wherethe presence-absence matrix is represented by a bi-partite network. For example, Kougioumoutzis et al.(2014) applied the NetCarto algorithm (Guimera &Nunes Amaral, 2005) in order to identify biogeograph-ical modules within the phytogeographical area of theCyclades. Similarly, Vilhena & Antonelli (2015) pro-posed a network approach for delimiting biogeograph-ical region based on the InfoMap algorithm (Rosvall& Bergstrom, 2008). By embedding species distri-butional data into complex networks, these methodshave the great advantage to be generic, flexible andto incorporate several scales in the analysis. Most im-portantly, these methods integrate species communityand spatial units within a single framework, whichallow to test the relative contribution of each taxato bioregions depicted, and to represent the relation-ship between those bioregions based on those contri-butions.

In this study, we present a biogeographical net-work analysis of plant species distribution in theFrench Mediterranean area at different scales. TheFrench Mediterranean territory represents an inter-esting study area to test new approaches, given theexcellent knowledge of the spatial distribution of the

plant species revealed by botanical inventories (Tison& Foucault, 2014; Tison et al., 2014) and the detaileddatabases compiled by the French National BotanicConservatory of Porquerolles and the Alpine NationalBotanic Conservatory. The objective of this work isto delineate bioregions, identify groups of species andanalyse the relationships between the two entities.

MATERIALS AND METHODS

Dataset and study area

The study area, situated in southern France,encompasses the former Languedoc-Roussillon re-gion (five departments of the current Occitanie re-gion: Pyrenees-Orientales, Aude, Herault, Gard andLozere) and the whole Provence-Alpes-Cote d’Azurregion. It extends around the entire Mediterraneancoastline of mainland France and inland, compris-ing almost all the Mediterranean hinterland, totalling558,776 km2 (Figure 1). The topography is struc-tured by three major mountain ranges, the Pyreneesin the southwest, the Massif central in the north-west and the Maritime Alps in the north-east. In-between, the landscape is mostly hilly with some low-lands around rivers that flow into lagoons or marshydeltas such as the Camargue. The Rhone is the mainstructuring river and delimitates western and easternsubregions. Acidic substrates and silicate soils aremainly found in the aforementioned mountain rangesand in the smaller Maures-Esterel range in southernProvence. The remaining part of the territory is dom-inated by calcareous or marly substrates (principallyCretaceous and Jurassic), with some significant allu-vial zones and small volcanic areas.

The SILENE database1, has been created in 2006,and is the reference botanical database in the studyarea. It contains historical data gathered from the sci-entific literature and herbaria along with more recentdata coming from public studies, partnerships, localamateur botanist networks and professional botanistsof the Botanical Conservatory. Our analysis is basedon a 5 × 5 km2 grid cells. We decided to only re-tain data whose georeferencement precision is below10 meters. While the SILENE database containednearly five million observations at the date of the ex-port (June 2016), we deleted several taxa whose dis-tribution is still insufficiently known and could distortthe results (e.g. apomictic taxa such as Rubus or Hi-eracium). For the same reason, we also aggregated allsub-taxa at the species level. The final dataset resultsin 4,263,734 vegetation plant samples correspondingto 3,697 plant species. We divided the study area us-ing a UTM grid composed of 2,607 squares of lateral

1 Conservatoire Botanique National Mediterraneen & Con-servatoire Botanique National Alpin (Admin.). AAAA.SILENE-Flore [online]. http://flore.silene.eu (accessedthe 16/03/2018)

3

0

200

400

600

800

1000

#Species

Figure 1. Distribution of the number of species per grid cell (l = 5 km). The inset shows a map of France including thestudied area colored in red. An altitude map of the studied area is available in Appendix.

size l = 5 km. In order to assess the impact of thespatial resolution on the results (Divısek et al., 2016;Lennon et al., 2001), we also applied the aforemen-tioned biogeographical network analysis with a gridcomposed of squares of lateral size l = 10 km (see Fig-ure S2 and Table S1 in Appendix for more details).

Biogeographical network analysis

1. Biogeographical bipartite network. Delin-eating bioregions requires a link between the speciesstudied and their spatial environment. This linkis usually identified with presence-absence matriceswhere each row represents a grid cell and each columna species. The region of interest is usually dividedinto grid cells, the resolution of which depends mostlyon the size of the study area, the taxonomic groupunder study and the accuracy of the data. Accordingto the type and quality of data, but also to the re-search question, the species releve can be aggregatedboth spatially or by group of species. Another wayof formalizing complex interactions between speciesand grid cells is to build a biogeographical bipartitenetwork. This bipartite network enables us to modelrelations between two disjoint sets of nodes, gridcells and species (in our case), which are linked bythe presence of a species (or a group of species)

in a given grid cell during a certain time window(Step 1 in Figure 2). This way of understandingcomplex interactions makes it possible to visualizeand analyze complex spatio-ecological systems as awhole from individual interactions to local and globalbiogeographical properties.

2. Delineating bioregions. To identify bioregionswe projected our biogeographical bipartite network ona spatial template (Step 2 in Figure 2), by defininga metric to measure the similarity of species com-position between grid cells. Several measures basedon beta diversity have been proposed to quantify thedegree of (dis)similarity between grid cells, typicallytaking into account the number of shared species be-tween grid cells (Koleff et al., 2003; Wilson & Shmida,1984). These measures are mostly based on presence-absence data and aim at quantifying species turnoverand species nestedness among grid cells, together orseparately (Baselga, 2012). Although this indicatormay be influenced by gradients in species richness(Baselga, 2012; Dapporto et al., 2015; Lennon et al.,2001), results obtained with the Jaccard index weremore spatially coherent in our case.

The resulting network is a weighted undirected spa-tial network whose intensity of links between grid cellsrange from 0, absence of a link (no species in common)to 1 (identical species composition). The detection

4

test-valueGrid cells Plant species

1

2

4

3 5

Figure 2. Steps of the biogeographical network analysis. 1. Biogeographical bipartite network where grid cells and speciesare linked by the presence of a species (or a group of species) in a given grid cell during a certain time window. Note thatthere is no link between nodes belonging to the same set. 2. The bipartite network is then spatially projected by using asimilarity measure of species composition between grid cells. Bioregions are then identified with a network community detectionalgorithm. 3. The test-value matrix based on the contribution of species to bioregions is computed. 4. Then, a network ofsimilarity between species is built, based on the test-value matrix. Groups of species sharing similar spatial features are identifiedusing a community detection algorithm. 5. Finally, a coarse-grained biogeographical network unveiling the biogeographicalstructure of the studied area and the relationship between bioregions is obtained.

of community structure in biogeographical networksis an interesting alternative approach to delineatingbioregions (Kougioumoutzis et al., 2014; Vilhena &Antonelli, 2015). Community structure is indeed animportant feature, revealing both the network inter-nal organization and similarity patterns among itsindividual elements. In this study we used the Or-der Statistics Local Optimization Method (OSLOM)(Lancichinetti et al., 2011). OSLOM uses an iterativeprocess to detect statistically significant communitieswith respect to a global null model (i.e. random graphwithout community structure). The main character-istic of OSLOM is that it is based on a score usedto quantify the statistical significance of a cluster inthe network (Lancichinetti et al., 2010). The score isdefined as the probability of finding the cluster in arandom null model. The random null model used inOSLOM is the configuration model (Molloy & Reed,1995) that generates random graphs while preservingan essential property of the network: the distributionof the number of neighbors of a node (i.e. the degree

distribution). Therefore, the output of OSLOM con-sists in a collection of clusters that are unlikely to befound in an equivalent random network with the samedegree sequence. This algorithm is nonparametric inthe sense that it identifies the statistically significantpartition, without defining the number of communi-ties a priori. However, the tolerance value that deter-mines whether a cluster is significant or not might playan important role for the determination of the clus-ters found by OSLOM. The influence of this value,fixed initially, is however relevant only when the com-munity structure of the network is not pronounced.When communities are well defined, as it is usuallythe case in biogeography, the results of OSLOM donot depend on the particular choice of tolerance value(Lancichinetti et al., 2011). See Lancichinetti et al.(2011) for a comparison between OSLOM and othercommunity detection algorithms.

3. Test-value matrix. To analyse the bioregionsand their species composition, we rely on test-valuesmeasuring the under- or over-representation of species

5

in a bioregion. Let us consider a studied area dividedinto n grid cells, a species i present in ni grid cellsand a biogeographical region j composed of nj gridcells. The test-value compares the actual number ofgrid cells nij , located in biogeographical region j andsupporting species i, with the average number ninj/nthat would be expected if the species were uniformlydistributed over the whole studied area. Since thisquantity depends on ni and nj it is normalized bythe standard deviation associated with the averageexpected number of grid cells (Lebart et al., 2000).The test value ρij is then defined as,

ρij =nij − ninj

n√n−nj

n−1(1 − nj

n) ninj

n

(1)

The test value ρij is negative if the species i isunder-represented in region j, equal to 0 if the speciesi is present in region j in the same proportion asin the whole study area or positive if the species iis over-represented in region j. In the latter casewe consider that the species i contribute positivelyto region j and the level of contribution dependsof the ρij value. Additionally, we consider that aplant species contribute positively and significantlyto a bioregion j if ρij is higher than a predeterminedsignificance threshold δ. Hence, The test-valuematrix ρ can be used to highlight set of species whichbetter characterize the bioregions. The test-valuesare easy to interpret by specialists and representan user-friendly way of ranking species according totheir relevance.

4. Groups of species. The next step is to iden-tify how similarities between species are spatially dis-tributed across the study area. Here also we builda network in which the similarity sii′ between twospecies i and i′ is equal to,

sii′ =1

1 +√∑j(ρij − ρi′j)2

(2)

This similarity metric is based on the Euclideandistance between test-values for each pair of species.Again, the community detection algorithm OSLOMis used to detect significant groups of species sharingthe same spatial features (Step 4 in Figure 2). Thisstep produces a preliminary delimitation of therelationships between bioregions by identifying howthe groups of species contributes to one or severalbioregions.

5. Coarse-grained biogeographical network.To quantitatively characterize relationships betweenbioregions, we retained only the positive and signif-icant species contributions by considering only test-values higher than δ = 1.96 (5% significance level of aGaussian distribution).

ρ+ij = ρij1ρij>1.96 (3)

Then, since we are interested in interactions be-tween bioregions we focused on the way species contri-butions are distributed among regions by normalizingρ+ by row (Equation 4).

ρ+ij = ρ+ij/∑i

ρ+ij (4)

We then determined for each bioregions j how theset of species Aj = {i ∣ρij > 1.96} that contributes tothis biogeographical region are specific to it or alsocontribute to other regions (Equation 5).

λjj′ =1

∣Aj ∣∑i∈Aj

ρ+ij′ (5)

λjj′ represents therefore the average fraction ofcontribution to cluster j′ of species that contributesignificantly to cluster j. The specificity of a biogeo-graphical region is therefore measured with λjj , whilethe relationships with other regions is given by λjj′ . Itis important to note that for a given region j the vec-tor λj. sum to one and can be expressed in percentage.

At the end of the process, we obtain a coarse-grained biogeographical network summarizing the bio-geographical structure of the study area. This net-work is composed of the bioregions and the speciesgroups (Step 5 in Figure 2). All the metrics usedto measure the similarity between the different biore-gions are derived from the matrix of test-value ρ.

0 500 1000 1500 2000

0.000

0.001

0.002

0.003

0.004

Degree

PD

F

#Plant species per grid cell#Grid cells per plant species

Figure 3. Degree distributions of the biogeographical bi-partite network. Probability density functions of the numberof plant species per grid cell (in blue) and the number of cellscovered per plant species (in red). Similar figures showinghistograms instead of densities are available in Figure S13 inAppendix.

RESULTS

Biogeographical bipartite network

The bipartite network extracted from the databaseis composed of 2,607 5 × 5 km2 grid cells and 3,697

6

Mediterranean border

Cork oak zone

Cévennes sensu latoSubatlantic mountains

Mediterranean lowlands

Pre-AlpsHigh mountains

Gulf of Lion coast1

2

3

45

6

7

8

Figure 4. Bioregions based on similarity in plant species (l = 5 km). Eight bioregions have been identified. 1. Gulf of Lioncoast in red. 2. Cork oak zone in orange. 3. Mediterranean lowlands in light green. 4. Mediterranean border in dark green.5. Cevennes sensu lato in purple. 6. Subatlantic mountains in pink. 7. Prealps and other medium mountains in yellow. 8.High mountains in brown. The inset shows a map of France including the studied areas colored in red. An altitude map ofthe studied area is available in Appendix (Figure S14).

plant species, where the links represent the occurrenceof plant species in the grid cells. Two network degreedistributions can be associated to this network: thenumber of species per grid cell and the number of cellscovered by each species. The probability density func-tions of these two distributions are displayed in Fig-ure 3. The spatial component of the network is verydense. Most of the grid cells host between 200 and500 plant species, with an average of 360 species percell (i.e. ∼ 15 species/km2). For species side, the sit-uation is different; the majority of plant species coverless than 10% of the study area, which highlight theimportance of range restricted taxa. Nevertheless, thedistribution exhibits a long tail with a non-negligiblenumber of widespread species.

Delineating bioregions

We identified eight statistically significant biore-gions reflecting the biogeographical structure of theFrench Mediterranean area based on plant species dis-tribution (Figure 4). Clusters size vary from 120 to807 square cells. Clusters are spatially coherent, ex-

hibiting a connectivity measure always higher than0.5 (i.e. ratio between the number of grid cells inthe largest patch and the total number of grid cells(Turner et al., 2001)). Results obtained are not scalesensitive, and the spatial coherence of each cluster ac-cording to the scale (l = 5 and 10 km) can be foundin Table S1 in Appendix. It also important to notethat this step can also be performed with standardhierarchical clustering methods. The results obtainedwith Ward’s clustering are available in Appendix.

Groups of plant species

The test-value matrix can be used to identify plantspecies that contribute positively and significantlyto one or more bioregions. It is worth noting thatthe number of contributions and their intensity varyamong species. Indeed, some species contribute verylittle to only one region while other species contributesignificantly to three or more regions. The numberof species contributing to a given number of regionsdepends on the significance threshold δ. A very smalland negative value of δ will imply that almost all plant

7

0 2 4 6 8 100.0

0.1

0.2

0.3

0.4

0.5

Significance threshold α

Fra

ctio

n of

spe

cies

012345+

Figure 5. Fraction of species contributing positively andsignificantly to a given number of bioregions (from 0 to 5or more) as a function of the significance threshold. Thevertical line represents the significance threshold δ = 1.96.

species contribute significantly to the 8 bioregions. Incontrast, a very high value of δ will result in all speciescontributing to no regions. In order to get a better un-derstanding of species contribution mechanisms andto assess the influence of δ, we plot in Figure 5 thefraction of species contributing positively to a givennumber of bioregions as a function of a significancethreshold value. If we consider the default thresholdδ = 1.96, that corresponds to a 2.5% significance levelof a Gaussian distribution, we observe that the vastmajority of plant species contributes positively to oneor two regions representing 35% and 45% of species,respectively. There is also 20% percent of plant speciesthat contribute to three or more bioregions. If we in-crease the minimum level of contribution necessaryto claim that a species contributes to a region, we seethat the fraction of species contributing to two or morebioregions dramatically decreases while the fraction ofspecies with no contribution increases. However, it isinteresting to note that the fraction of species con-tributing to one region to increases until reaching aplateau. This demonstrates that 50% of plant speciesare strongly connected to a single region.

The similarities between plant species’ contributionto the 8 regions allowed us to identify 20 groups ofspecies, and their contribution to each bioregion is dis-played in Figure 7. We observed different patterns ofcontributions in terms of shape and intensity. This al-lows for the identification of groups of species sharingsimilar spatial features and highlights relationshipsbetween bioregions through the way plant species con-tribute to different group of regions.

Relationships between bioregions

This leads us to the study of relationships betweenbioregions. The network of interactions λ derived

from the test-value matrix is plotted in Figure 6. Wefound that, globally, plant species contributing signif-icantly to a region contribute mostly to this region,with an average specificity of 51% across the eightbioregions. It must be pointed out however that someregions are more specific than others with λjj valuesranging from 40% to 65%.

Analysis of how bioregions connect with each othershowed that there is no isolated region in the sensethat every region is connected with at least one otherregion with a λjj′ value varying from 1 to 28%. More-over, for all regions, the maximal λjj′ value is alwayshigher than 10%. Although it is generally the case,it is also worth mentioning that the relationships arenot necessarily symmetric, which represents an inter-esting way of detecting hierarchical relationships. Atable displaying all λjj′ values is available in Table S4in Appendix.

55

6549

40

56

52

2415

13

15

18

11

11

14

11

13

2417

19

26

28

19 16

52

42

25

1

2

3

4

5

68

7

Figure 6. Network of interactions between bioregions.λjj′ , expressed here in percentage, represents the averagefraction of contribution to cluster j′ of species that contributesignificantly to cluster j. Only links with a value λjj′ higherthan 10% are shown.

DISCUSSION

In this study we delineate spatial bioregions insouthern France, a transition area between a mediter-ranean and temperate climate. The present analy-sis represents to our knowledge one of the largestnetwork-based studies published to date, relying ona database containing more than four million datapoints across a territory of about 558,776 km2. Whilethis territory has been divided into bioregions on ex-pert knowledge, we confront those approaches to data-driven classification, and discuss the coherence of the

8

−100

10203040 a

b

c

d

−100

10203040 e

f

g

h

−100

10203040

Test

−va

lue i

j

k

l

−100

10203040 m

n

o

p

1 2 3 4 5 6 7 8

−100

10203040 q

1 2 3 4 5 6 7 8

Biogeographical regions

r

1 2 3 4 5 6 7 8

s

1 2 3 4 5 6 7 8

t

Figure 7. Description of the groups of plant species. Boxplot of test-values according to the bioregions and the plantspecies groups. The horizontal line represents the significance threshold δ = 1.96. The number of plant species per group isavailable in Table S3 in Appendix.

different perspectives. We delineated eight statisti-cally significant bioregions, which we will first presentin relation to previously published work, and empha-size their specificity regarding associated groups ofspecies. We discuss the observed spatial patterns interms of ecological and historical drivers, to provideinsights into mechanisms driving the assemblage ofvegetation communities.

Bioregions

The clustering approach identified eight statis-tically significant spatial clusters, that representscoherent territories detailed below. Regions arepresented from Mediterranean toward temperate andmountainous climates.

1. Gulf of Lion coast is a bioregion that extendswest of the Rhone, penetrating more inland aroundthe wetland of the Rhone Delta. The latter, alongwith the Languedoc lagoons, is frequently used asan example of azonal vegetation (Ozenda, 1994),and the originality of the flora and the vegetationof these areas has long been recognized (Molinier &

Tallon, 1970). Some subdivisions have been suggestedseparating, even at a coarse scale, the sand-dunecomplex, the halophytic vegetation and the saltmeadows (Bohn et al., 2000), but were not foundhere probably due to the size of the cells we used.From a geological point of view, this bioregion isessentially made of sand dunes, lagoon sedimentsand modern alluvium. It is entirely situated undera Mediterranean climate, in the mesomediterraneanclimatic belt, with a dry season of two or threemonths in the summer (Rivas-Martınez et al., 2004a).Taxa specific to this cluster exhibit a distributionfollowing the Mediterranean coastal area, extendingin some cases towards other coastal areas or toarid inland zones. They are mostly encountered inhalophytic communities and surprisingly not thatmuch into dunes, suggesting that the key factordefining this bioregion might be the saline soilsrather than the coastal position alone. Severalnarrow-endemics rely on those habitats, especially inthe genus Limonium whose rapid radiation is typi-cal of mediterranean neoendemics (Lledo et al., 2005).

2. Cork oak zone encompasses the Maures-Esterelrange and neighbouring areas. West of the Rhone,

9

it is fragmented with cells in the eastern tip ofthe Pyrenees (low Alberes and the Roussillon low-lands), plus a few more sparsely dispersed zones inLanguedoc. The Provence and Alberes areas havebeen identified by phytogeographers (Ozenda, 1994;Ozenda & Lucas, 1987) as the “Cork oak zone”, asilicicolous warm mesomediterranean area. Indeed,climatic data show a clear summer dry period ofone to two months. Almost all of the cells containacidic soils over a variety of substrates (granites,gneiss, schists, sandstones, alluvial deposits, etc.).Species most linked to the “Cork oak zone” havea Mediterranean distribution, with some extendingtowards the Atlantic area. Characteristic species haveecological preferences for acid soils, and belong tovarious vegetation stages (forest, scrub or grasslandformations).

3. Mediterranean lowlands bioregion coversthe hinterland of the Gulf of Lion from the Rous-sillon to western Provence. Several authors haveindividualized an arc shaped mesomediterrean zone(Dupias & Rey, 1985; Ozenda, 1994) but their limitsdo not fit exactly ours. The closest match is thecatalonian-provencal mesomediterranean holm oakforests unit of the European natural vegetation map(Bohn et al., 2000). The area is principally com-posed of sedimentary rocks (mostly limestones andmarls) and alluvium. Its climate is Mediterranean(Rivas-Martınez et al., 2004a), with a summer dryperiod of one to three months. With few excep-tions, species most linked to this bioregion have adistribution included in the Mediterranean region(Rivas-Martınez et al., 2004b). Most of them belongsto communities of the Quercetea ilicis or of the formerThero-Brachypodietea, i.e. the matorral / forest andgrasslands communities making up the landscapelocally called “garrigues”. The other part of thesetaxa is usually found in disturbed communities,showing the strong incidence of human activities inthis area.

4. Mediterranean border is a bioregion whosenorthern edge roughly follows the limit of theMediterranean world as it is usually depicted (Dupias& Rey, 1985; Quezel & Medail, 2004). It broadly co-incide to what has been called a supramediterraneanbelt (Ozenda, 1994) or a submediterranean zone(Bolos, 1961), and fits quite well with four mappingunits of the Map of the natural vegetation of Europe(Bohn et al., 2000); namely the catalonian-provencalsupramediterranean holm oak forests and three typesof downy oak forests (ligurian- middle apennine,languedocian and those extending from the southernPyrenees to the southwest pre-Alps). The substratumof this area is mainly calcareous and marly. Thisarea has a short (one month) summer drought periodwith the exception of some Var and Alpes-Maritimesplaces where the summer drought is more pronounced(two months). Species most linked to this bioregionpresent a western eury-mediterranean distribution,

and share a common ecology, occurring frequently incommunities belonging to the Helianthemo italici-Aphyllanthion monspeliensis and to a lesser extentto the Ononidetalia striatae (Gaultier, 1989; Rivas-Martınez et al., 2002), i.e. dry dwarf scrubs and theirassociated grasslands on calcareous and marly erodedsoils (Mucina et al., 2016).

5. Cevennes sensu lato is a bioregion to whichmost of the cells are situated in the Cevennesareas, while the remainder is scattered over theeastern Pyrenees piedmont and the Montagne Noire(southern limit of the Massif Central). This spatialcluster overlays four zones of the phyto-ecologicalregions (Dupias & Rey, 1985): the lower Cevennes,the “warm” Cevennes valleys, the Aspres and thechestnut zone of the southern edge of the Mon-tagne Noire. The Cevennes proper part of thiscluster has also been identified by other authors(Braun-Blanquet, 1923; Ozenda, 1994) and putativeglacial refugia has been positioned there (Medail& Diadema, 2009). This area is not subject to asummer drought and covers siliceous substrata suchas schists, granites or gneiss. Taxa exhibiting thestrongest link to this biogeographical region areeither Cevennes endemics, subendemics (Dupont,2015; Lavergne et al., 2004) or plants with a moreor less Atlantic distribution (Dupont, 2015), but noclear ecological pattern is emerging among these taxa.

6. Subatlantic mountains The largest area coveredby cells of this biogeographical region is the northernpart of the Lozere department. The remaining cellsare mostly distributed in the Massif Central and inthe Pyrenees. These areas belong to the beech (Fagussylvatica L.) montane belt (Bohn et al., 2000; Ozenda,1994) with a few exceptions where Scots pines (Pi-nus sylvestris L.) dominate. It corresponds to thepredominantly siliceous subatlantic type (Ozenda &Lucas, 1987), where the climate is rather wet, withprecipitations frequently exceeding 1,000 mm peryear and no dry period. Thus, wetlands and bogs arenot rare, and the substratum is made of igneous rockswhich explain the acidic nature of the soils. Themajority of the taxa most linked to this spatial clusterare generally distributed all over the eurosiberian re-gion or the western part of this region, correspondingto a subatlantic distribution (Dupont, 2015; Rivas-Martınez et al., 2004b). Interestingly, most of thoseplants grow in wetlands habitats, a trend alreadynoticed in the Massif Central (Braun-Blanquet, 1923).

7. Pre-Alps and other medium mountainsrepresent a bioregion whose cells are disseminatedthrough the lower parts of the eastern Pyreneesincluding almost all the Pyrenean part of the Audedepartment, through the highest areas of the Causses,around the Mont Ventoux and through the mosteastern part of the Pre-Alps. This area has rarelybeen individualised in such a way even if at aEuropean scale it can be related to several more or

10

less calcicolous beech or fir-beech forest belts (Bohnet al., 2000) (Abies alba L. and Fagus sylvatica L.),or more specifically, for the Var department, to apre-alpine district (Lavagne, 2008). Most of the rockunderlying this area is calcareous. Climatically, weare outside of the Mediterranean climate as there isno dry period. The distribution of taxa most linkedto this biogeographical region is basically holarctic,avoiding the Mediterranean parts of Europe. Someof these taxa also avoid the most Atlantic part ofthe continent. Their ecology is varied, pertaining todifferent stages (grasslands, shrubs, forests) of moun-tain vegetation series, often (but not systematically)calcicolous.

8. High mountains This bioregion regroups thehighest part of the Alps and the Pyrenees. If mostauthors agree on individualizing the upper vegetationbelts of these mountain ranges, its unity and thecommon points are less often identified (Ozenda,2002). Both calcareous and acidic soils are to befound in this area. Cells of this region are thecoldest of our study area, and there is no dry period:the climate is relatively harsh and the vegetationperiod is reduced (Ozenda, 2002) compared to theother clusters. Taxa most linked to this region aremainly European mountains endemics, venturingalso in the Arctic. They belong to grasslands orsnowbeds communities, which is consistent with theiroccurrence on the highest ranges.

Species and spatial relationships amongbioregions

Defining the Mediterranean region

At a global scale, the delimitation of the Mediter-ranean border has been a long running question (La-tini et al., 2017), and the mismatch of the numerousattempts attest to the difficulties (Figures S5-12 inAppendix). In France, the first attempt goes backto third edition of the Flore Francaise by Lamarck &Candolle (1805), as shown in Ebach & Goujet (2006)followed by several other works such as Flahault &Durand (1887), who considered the distribution limitof the olive tree (Olea europaea L.) as a marker ofthe Mediterranean biome. This was later generalizedto the evergreen oak belt (Quezel, 1999), but it ap-peared that the situation was more complex (Quezel &Medail, 2004). Thus, variability in results has not leadto a comprehensive framework yet. This has severalimplications regarding conservation programs, as thedelimitation mentioned by European legislation hasbeen used as a reference to delimit the distribution ofseveral protected habitat2. In this study, the network

2 http://www.eea.europa.eu/data-and-maps/

F50D9CF8-FFEE-475F-8A65-4E095512CBB7 (accessed the

approach allowed to discriminate two “sub-networks”with little exchange regarding species composition anddifferent relative contribution to each area, whichglobally relate to a temperate and a Mediterraneansub-groups. Several earlier bioregionalizations in themediterranean basin have failed to separate mediter-ranean from eurosiberian ensembles, suggesting thisboundary would be highly permeable (Garcıa-Barroset al., 2002; Saiz et al., 1998) and easily crossed byspecies. Here, the use of a precise dataset coupledwith a network analyse has proven to be relevant todepict such spatial transition, which reinforce the needto gather coherent dataset to characterize complexand intricate spatial structures. This biogeographi-cal boundary has been linked to a change in the an-nual distribution of precipitation, which induces a pro-longed summer drought and a stronger climatic sea-sonality in the mediterranean (Antonelli, 2017). Ata finer scale, the three mediterranean clusters presenta high spatial coherence, and closely fit to the me-somediterranean thermoclimatic belt (Rivas-Martınezet al., 2004b) (see Figure S11 in Appendix). The highcongruence between climatic model (Rivas-Martınezet al., 2004b) and biogeographic entities has neverbeen pictured by previous bioregionalization works(see Appendix for maps), as most of them presented awider definition of the mediterranean biome, extend-ing northward. Then, the absence of orogenic bar-riers along this climate-based distinction is likely toproduce shallow boundaries typical of transition ar-eas (Antonelli, 2017; Ficetola et al., 2017) exemplifiedhere by the cluster “Mediterranean border” that con-tains all historical attempt to delimitate the mediter-ranean biome. West of the Rhone, this region is rel-atively thin and fence around the mesomediterraneanensemble; east of the Rhone, it occupies a wide areaon the Alpine piedmont. Thus, instead of drawing asingle line (Cox, 2001), we propose to identify a tran-sition area (Droissart et al., 2017; Latini et al., 2017)with an upper boundary as the limit of the Mediter-ranean biome (Antonelli, 2017).

Vicariance and fragmentation among bioregions

The relationship between bioregions can be seenthrough the understanding of species relative impor-tance in each area. First, the regions “Gulf of Lioncoast”, “Cork oak zone” and “Mediterranean low-lands”, all included within the same bioclimatic belt(Rivas-Martınez et al., 2004a), differ mostly on sub-stratum, i.e. calcareous (bioregion 3), siliceous (biore-gion 2) or quaternary deposits (bioregion 1). Thus,they are well defined and little uncertainty exists con-cerning their spatial configuration (Figure S15 in Ap-pendix); those three entities can be seen as climaticvicariant bioregions which have conjointly developed

04/07/2018)

11

on different geological substrates, or “islands”. Asa result, they share an important pool of species,and present the highest complementarity in the net-work, as they are the only three clusters all relatedto each other. In contrast, the relationship betweenthe “Cork oak zone” and the “Cevennes” exemplifythe opposite process: those two areas share a simi-lar bedrock (mainly acidic substrate) but are locatedat each extreme of the Mediterranean climatic gra-dient. While the “Cork oak zone” is present underhot and dry mesomediterranean climate (some coastalcells even belonging to a thermomediterranean belt),the “Cevennes” present a higher impluvium and a veryweak summer drought. Consequently, they share acommon set of species which interestingly are typicalof the “Cevennes” cluster, and extend into the “Corkoak zone”. Noteworthy point, those population canconstitute relictual rear edge populations, which of-ten retain particular interest for conservation (Hampe& Petit, 2005; Lavergne et al., 2006).

Finally, the “Pre-Alps” and “High mountains”bioregions are both present within the three moun-tain chains, and occupy climatic conditions with nodry period at all, and especially harsh prolonged win-ter for the second. Several species groups are highlyinformative for both of those bioregions, which signifythat they share an important group of species globallyadapted to mountain environment. “High mountains”present the highest percentage of typical species. Yet,within the numerous plant species groups characteriz-ing those entities (5 groups in Figure 7), the relativecontribution of each toward one or the other bioregionmight differ slightly, sometimes in association with an-other bioregion such as the “Mediterranean border”(Figure 7). This illustrates that groups of taxa areunevenly important across these two regions, proba-bly reflecting the complex geological substrate. Thus,while our analysis reflect an overall homogeneity ofmountain flora mainly driven by climate, it is likelythat finer divisions based on a more precise studycould be expected. This has been pinpointed by Bohnet al. (2000) who pictured a high local heterogeneitydue to steep altitudinal gradients and geological diver-sity, despite some vegetation groups shared betweenthe Alps and the Pyrenees. Therefore, a compara-tive analysis including a broader spatial perspectiveon those massif could improve our understanding ofthe spatial structure of mountain flora in western Eu-rope.

Eco-evolutionary factors driving the spatial organisationof plant diversity

The spatial distribution and species relative impor-tance for each bioregion can help us to better un-derstand processes that have shaped Mediterraneanbiota in the south of France. The regional speciespool results from several waves of colonization fol-lowing glacial cycles, constrained by ecological filtersthat allowed taxa to persist and ultimately shaped lo-

cal communities (Ricklefs, 1987). Indeed, our studyarea is at the crossroad of recolonization routes out oftwo major refugia, i.e. the Iberic and Italian peninsu-las (Hewitt, 2000), and represents an admixture zonefor several mediterranean taxa (Lumaret et al., 2002).Joint action of colonization-retraction sequences andlong term persistence within microrefugia has beensuspected to generate fragmented distribution. Thus,one particular feature of such climatic transition areais the high proportion of population isolated at the pe-riphery of their main range (Thompson, 2005), eitherat the rear or at the leading edge of their distribution(Hampe & Petit, 2005). However, spatial patternsalone do not inform on the evolutionary isolation ofsuch populations, could it be of recent dispersal fol-lowing Last Glacial Maximum (Lumaret et al., 2002),or long term persistence in a given refugia (Medail& Diadema, 2009; Papuga et al., 2015). Thus, in-tegrating phylogenies within bioregionalization wouldprove informative to analyse historical events thathave shaped current spatial patterns of biodiversity(Nieto Feliner, 2014), and capture the evolutionaryrelationship among bioregions (Holt et al., 2013).

Nevertheless, analysing the spatial organization offlora can help us to understand ecological factors thatshape such bioregions. Orographic barriers and pasttectonic movement are expected to have little im-pact on our study area, as no such events have oc-curred there since the onset of the Mediterranean cli-mate in the Pliocene (Rosenbaum et al., 2002). Inour analysis, spatial structuration relies principallyon two elements. On the one hand, a climatic gra-dient from Mediterranean to temperate climate cre-ates fuzzy spatial limits among adjacent groups, andincreases uncertainty when delimitating groups (Fig-ure S15 in Appendix). This is exemplified by thespatial imbrication of “Mediterranean lowlands” and“Mediterranean border”. On the other hand, geologi-cal variations can form sharp transitions creating im-portant species turnover between places close apart.This is exemplified by the “Cork oak zone” whose spa-tial delimitation is very clear, due to the presence ofan acidic substrate surrounded by places dominatedby calcareous-based rock. Interestingly, this area stillshares an important part of its biota with other placesin the Mediterranean basin probably inherited fromtimes where such geological islands formed a singleensemble, before the separation and later migrationof these islands (Medail & Quezel, 1997; Rosenbaumet al., 2002). The joint action of these two ecologicalfactors has already been highlighted in previous biore-gionalization of the Mediterranean basin (Buira et al.,2017). As a result, complex geo-climatic variationhave played a key role in shaping island-like territorieswhich have fragmented species distributions, a factorthat has strong influence on populations character-istics both genetically and demographically (Pirononet al., 2017).

The flora of the Mediterranean basin shows recur-rent patterns of narrow endemism, species turnoverand highly disjunct distributions (Thompson, 2005).

12

While allopatric isolation has been suspected to bethe main mechanism explaining the differentiationof taxa, the shared significance of different ecologi-cal variables (namely climate and geology) points outthe combined importance of spatial isolation and het-erogeneous selective pressures (Anacker & Strauss,2014; Thompson, 2005). Additionally, recent studieshave shown that this can be enhanced by small scalechanges of the ecological niche (Lavergne et al., 2004;Papuga et al., 2018; Thompson et al., 2005). Contraryto other mediterranean biomes (e.g. South-Africaand Australia), the mediterranean basin is markedby an active speciation, which has led to the highobserved proportion of neoendemic species (Rundelet al., 2016). If evidences have accumulated concern-ing cryptic microrefugia for temperate trees (Stew-art & Lister, 2001), little is known regarding mediter-ranean taxa, especially those that exhibit little disper-sal capacities, a shared trait among mediterranean en-demics (Lavergne et al., 2004). Thus, this bioregion-alization set the scene to investigate the shared phylo-geographic legacy of the Mediterranean biota (Puscas& Choler, 2012), and measure the evolutionary iso-lation of such communities that separate peripheralisolates from newly differentiated species (Crawford,2010).

CONCLUSION

The quality of a bioregionalization is dependent onthe data and the method used. To our knowledge,the present analysis constitutes the densest species-cells network analysed in a bioregionalization study,at such a high spatial resolution. Therefore, results ofthis study demonstrate that new statistical methodsbased on network analysis can bring solutions to man-age and analyse large databases, and provide efficientbioregionalization at different scales. New perspec-tives for bioregionalization will integrate communitystructure across different scales, in order to under-stand how deterministic (i.e. niche based) processesand stochastic events (dispersal, random extinction,ecological drift) interact to shape plant communities,from regional species pool to local assemblages (Chase& Myers, 2011).

ACKNOWLEDGEMENTS

This work was supported by a grant from the FrenchNational Research Agency (project NetCost, ANR-17-CE03-0003 grant). Partial financial support hasbeen received from the French Ministere de la Tran-sition Ecologique et Solidaire (MTES). We thank theAlpine National Botanic Conservatory for providingsome of the Provence-Alpes-Cote d’Azur data. Wewish to thank Christelle Hely-Alleaume, Virgile No-ble and James Molina for useful discussions. A spe-cial thank goes to John D. Thompson for correctingEnglish and interesting remarks.

DATA AVAILABILITY

Code and data are available at www.maximelenormand.com/Codes

REFERENCES

Anacker, B. L. & Strauss, S. Y. (2014) The geographyand ecology of plant speciation: range overlapand niche divergence in sister species. Proceed-ings of the Royal Society of London B: BiologicalSciences, 281, 20132980.

Antonelli, A. (2017) Biogeography: Drivers of biore-gionalization. Nature Ecology & Evolution, 1,0114.

Baselga, A. (2012) The relationship between speciesreplacement, dissimilarity derived from nested-ness, and nestedness. Global Ecology and Bio-geography, 21, 1223–1232.

Blondel, J., Aronson, J., Bodiou, J.-Y. & Boeuf, G.(2010) The Mediterranean Region: Biological Di-versity in Space and Time. Oxford UniversityPress, Oxford, New York, second edition edition.

Bohn, U., Gollub, G. & Hettwer, C. (2000) Karte dernaturlichen Vegetation Europas. Bundesamt furNaturschutz. Landwirtschaftsvlg Munster, Bonn,1., aufl. edition.

Bolos, O. (1961) La transicion entre la depresiondel ebro y los pirineos en el aspecto geobotanico.Anal. lnst. Bot. Cavanilles, 18, 99–254.

Braun-Blanquet, J. (1923) L’origine et ledeveloppement des flores dans le massif centralde France; avec aperu sur les migrations desflores dans l’Europe sud-occidentale. L. Lhomme,Paris.

Buira, A., Aedo, C. & Medina, L. (2017) Spatial pat-terns of the Iberian and Balearic endemic vas-cular flora. Biodiversity and Conservation, 26,479–508.

Chase, J. M. & Myers, J. A. (2011) Disentanglingthe importance of ecological niches from stochas-tic processes across scales. Philosophical Trans-actions of the Royal Society of London. Series B,Biological Sciences, 366, 2351–2363.

Cheruvelil, K. S., Yuan, S., Webster, K. E., Tan, P.-N., Lapierre, J.-F., Collins, S. M., Fergus, C. E.,Scott, C. E., Henry, E. N., Soranno, P. A., Fil-strup, C. T. & Wagner, T. (2017) Creating mul-tithemed ecological regions for macroscale ecol-ogy: Testing a flexible, repeatable, and accessi-ble clusteringmethod. Ecology and Evolution, 7,3046–3058.

13

Cox, B. (2001) The biogeographic regions reconsid-ered. Journal of Biogeography, 28, 511–523.

Crawford, D. J. (2010) Progenitor-derivative speciespairs and plant speciation. Taxon, 59, 1413–1423.

Dahlin, K. M., Asner, G. P. & Field, C. B. (2014)Linking vegetation patterns to environmentalgradients and human impacts in a mediterranean-type island ecosystem. Landscape Ecology, 29,1571–1585.

Dapporto, L., Ciolli, G., Dennis, R. L. H., Fox, R. &Shreeve, T. G. (2015) A new procedure for ex-trapolating turnover regionalization at mid-smallspatial scales, tested on British butterflies. Meth-ods in Ecology and Evolution, 6, 1287–1297.

Divısek, J., Storch, D., Zeleny, D. & Culek, M. (2016)Towards the spatial coherence of biogeographicalregionalizations at subcontinental and landscapescales. Journal of Biogeography, 43, 2489–2501.

Droissart, V., Dauby, G., Hardy, O. J., Deblauwe, V.,Harris, D. J., Janssens, S., Mackinder, B., Blach-Overgaard, A., Sonke, B., Sosef, M. M., Stevart,T., Svenning, J.-C., Wieringa, J. J. & Couvreur,T. L. P. (2017) Beyond trees: Biogeographicalregionalization of tropical Africa. Journal of Bio-geography.

Dupias, G. & Rey, P. (1985) Document pour un zonagedes regions phyto-ecologiques. Centre d’ecologiedes ressources renouvelables.

Dupont, P. (2015) Les plantes vasculaires atlantiques,les pyreneo-cantabriques et les elements floris-tiques voisins dans la peninsule iberique et enFrance. Societe botanique du Centre-Ouest.

Ebach, M. C. & Goujet, D. F. (2006) The first bio-geographical map. Journal of Biogeography, 33,761–769.

Fenu, G., Fois, M., Canadas, E. M. & Bacchetta, G.(2014) Using endemic-plant distribution, geologyand geomorphology in biogeography: the case ofSardinia (Mediterranean Basin). Systematics andBiodiversity, 12, 181–193.

Ficetola, G. F., Mazel, F. & Thuiller, W. (2017)Global determinants of zoogeographical bound-aries. Nature Ecology & Evolution, 1, 0089.

Flahault, C. & Durand, M. (1887) Limite de la regionmediterraneenne en France. Publications de laSociete Linneenne de Lyon, 5, 9–9.

Funk, V. A., Richardson, K. S. & Sakai, A. K. (2002)Systematic Data in Biodiversity Studies: Use Itor Lose It. Systematic Biology, 51, 303–316.

Garcıa-Barros, E., Gurrea, P., Lucianez, M. J., Cano,J. M., Munguira, M. L., Moreno, J. C., Sainz,H., Sanz, M. J. & Simon, J. C. (2002) Parsimony

analysis of endemicity and its application to ani-mal and plant geographical distributions in theIbero-Balearic region (western Mediterranean).Journal of Biogeography, 29, 109–124.

Gaultier, C. (1989) Relations entre pelouses eu-rosiberiennes (Festuco-Brometea Br. -Bl. Et Tx.43) et groupements mediterraneens (Ononido-Rosmarinetea Br. -Bl. 47) : etude regionale(Diois) et synthese sur le pourtour mediterraneenNord-occidental. PhD thesis, Universite de Paris-Sud.

Graham, C. H. & Hijmans, R. J. (2006) A compar-ison of methods for mapping species ranges andspecies richness. Global Ecology and Biogeogra-phy, 15, 578–587.

Greuter, W. (1991) Botanical diversity, endemism,rarity and extinction in the Mediterranean area:an analysis based on the published volumes ofMedChecklist. Botanika Chronika, 10, 63–79.

Guimera, R. & Nunes Amaral, L. A. (2005) Functionalcartography of complex metabolic networks. Na-ture, 433, 895–900.

Hampe, A. & Petit, R. J. (2005) Conserving biodiver-sity under climate change: the rear edge matters.Ecology Letters, 8, 461–467.

Hewitt, G. (2000) The genetic legacy of the Quater-nary ice ages. Nature, 405, 907–913.

Holt, B. G., Lessard, J.-P., Borregaard, M. K., Fritz,S. A., Arauo, M. B., Dimitrov, D.and Fabre, P.-H., Graham, C. H., Graves, G. R., Jønsson, K. A.,Nogues-Bravo, D., Wang, Z., Whittaker, R. J.,Fjeldsa, J. & Rahbek, C. (2013) An Update ofWallace’s Zoogeographic Regions of the World.Science, 339.

Joly, D., Brossard, T., Cardot, H., Cavailhes, J., Hilal,M. & Wavresky, P. (2010) Les types de climatsen France, une construction spatiale. Cybergeo :European Journal of Geography.

Koleff, P., Gaston, J. & Lennon, J. (2003) Measuringbeta diversity for presence-absence data. Journalof Animal Ecology, 72, 367–382.

Kougioumoutzis, K., Simaiakis, S. M. & Tiniakou, A.(2014) Network biogeographical analysis of thecentral Aegean archipelago. Journal of Biogeog-raphy, 41, 1848–1858.

Kreft, H. & Jetz, W. (2010) A framework for delin-eating biogeographical regions based on speciesdistributions. Journal of Biogeography, 37, 2029–2053.

Lamarck, J. d. M. d. & Candolle, A. (1805)Flore francaise, ou descriptions succinctes detoutes les plantes qui croissent naturellement enFrance, disposees selon une nouvelle methode

14

d’analyse, et precedees par un expose desprincipes elementaires de la botanique. Paris.

Lancichinetti, A., Radicchi, F. & Ramasco, J. J.(2010) Statistical significance of communities innetworks. Physical Review E, 81, 046110.

Lancichinetti, A., Radicchi, F., Ramasco, J. J. & For-tunato, S. (2011) Finding Statistically Signifi-cant Communities in Networks. PLOS ONE, 6,e18961.

Latini, M., Bartolucci, F., Conti, F., Iberite, M.,Nicolella, G., Scoppola, A. & Abbate, G. (2017)Detecting Phytogeographic Units Based on Na-tive Woody Flora: A Case Study in CentralPeninsular Italy. The Botanical Review, 83, 253–281.

Lavagne, A. (2008) Synthese phytogeographique dudepartement du Var. In Le Var et sa flore.Plantes rares ou protegees. Naturalia Publica-tions, Inflovar.

Lavergne, S., Molina, J. & Debussche, M. (2006) Fin-gerprints of environmental change on the raremediterranean flora: a 115-year study. GlobalChange Biology, 12, 1466–1478.

Lavergne, S., Thompson, J. D., Garnier, E. & De-bussche, M. (2004) The biology and ecology ofnarrow endemic and widespread plants: a com-parative study of trait variation in 20 congenericpairs. Oikos, 107, 505–518.

Lebart, L., Piron, M. & Morineau, A. (2000) Statis-tique exploratoire multidimensionnelle. Dunod,Paris.

Lennon, J. J., Koleff, P., Greenwood, J. J. D. & Gas-ton, K. J. (2001) The geographical structureof British bird distributions: Diversity, spatialturnover and scale. Journal of Animal Ecology,70, 966–979.

Lledo, M. D., Crespo, M. B., Fay, M. F. & Chase,M. W. (2005) Molecular phylogenetics of Limo-nium and related genera (Plumbaginaceae): bio-geographical and systematic implications. Amer-ican Journal of Botany, 92, 1189–1198.

Lumaret, R., Mir, C., Michaud, H. & Raynal, V.(2002) Phylogeographical variation of chloroplastDNA in holm oak (Quercus ilex L.). MolecularEcology, 11, 2327–2336.

Medail, F. & Diadema, K. (2009) Glacial refugia in-fluence plant diversity patterns in the Mediter-ranean Basin. Journal of Biogeography, 36, 1333–1345.

Medail, F. & Quezel, P. (1997) Hot-Spots Analy-sis for Conservation of Plant Biodiversity in theMediterranean Basin. Annals of the MissouriBotanical Garden, 84, 112–127.

Mikolajczak, A., Marchal, D., Sanz, T., Isenmann, M.,Thierion, V. & Luque, S. (2015) Modelling spatialdistributions of alpine vegetation: A graph the-ory approach to delineate ecologically-consistentspecies assemblages. Ecological Informatics, 30,196–202.

Molinier, R. & Tallon, G. (1970) Prodrome desunites phytosociologiques observees en Camargue.Molinier.

Molloy, M. & Reed, B. (1995) A critical point for ran-dom graphs with a given degree sequence. Ran-dom Structures & Algorithms, 6, 161–180.

Morrone, J. J. (2018) The spectre of biogeographi-cal regionalization. Journal of Biogeography, 105,1118–1123.

Mucina, L., Butmann, H., Dierßen, K., Theurillat,J.-P., Raus, T., Carni, A., Sumberova, K., Will-ner, W., Dengler, J., Garcı, R. G., Chytry,M., Hajek, M., Di Pietro, R., Iakushenko, D.,Pallas, J., Daniels, F. J., Bergmeier, E., San-tos Guerra, A., Ermakov, N., Valachovic, M.,Schaminee, J. H. J., Lysenko, T., Didukh, Y. P.,Pignatti, S., Rodwell, J. S., Capelo, J., Weber,H. E., Solomeshch, A., Dimopoulos, P., Aguiar,C., Hennekens, S. M. & Tichy, L. (2016) Vegeta-tion of Europe: hierarchical floristic classificationsystem of vascular plant, bryophyte, lichen, andalgal communities. Applied Vegetation Science,19, 3–264.

Murray, A. (1866) The geographical distribution ofmammals. Day and Son, limited,, London,.

Myers, N., Mittermeier, R. A., Mittermeier, C. G., daFonseca, G. A. B. & Kent, J. (2000) Biodiversityhotspots for conservation priorities. Nature, 403,853–858.

Nieto Feliner, G. (2014) Patterns and processesin plant phylogeography in the MediterraneanBasin. A review. Perspectives in Plant Ecology,Evolution and Systematics, 16, 265–278.

Ozenda, P. (1994) Vegetation du continent Europeen.Delachaux et Niestle, Lausanne.

Ozenda, P. (2002) Perspectives pour une geobiologiedes montagnes. Presses Polytechniques et Uni-versitaires Romandes, Lausanne.

Ozenda, P. & Lucas, M. J. (1987) Esquisse d’une cartede la vegetation potentielle de la france 1/1 500000. Documents de cartographie ecologique.

Papuga, G., Gauthier, P., Pons, V., Farris, E. &Thompson, J. (2018) Ecological niche differen-tiation in peripheral populations: a comparativeanalysis of eleven mediterranean plant species.Ecography, 176, 724–738.

15

Papuga, G., Gauthier, P., Ramos, J., Pons, V.,Pironon, S., Farris, E. & Thompson, J. D.(2015) Range-Wide Variation in the Ecologi-cal Niche and Floral Polymorphism of the West-ern Mediterranean Geophyte Narcissus dubiusGouan. International Journal of Plant Sciences,176, 724–738.

Pironon, S., Papuga, G., Villellas, J., Angert, A. L.,Garcıa, M. B. & Thompson, J. D. (2017) Geo-graphic variation in genetic and demographic per-formance: new insights from an old biogeographi-cal paradigm. Biological Reviews, 92, 1877–1909.

Puscas, M. & Choler, P. (2012) A biogeographic de-lineation of the European Alpine System basedon a cluster analysis of Carex curvula-dominatedgrasslands. Flora - Morphology, Distribution,Functional Ecology of Plants, 207, 168–178.

Quezel, P. (1999) Les grandes structures devegetation en region mediterraneenne: Facteursdeterminants dans leur mise en place post-glaciaire. Geobios, 32, 19–32.

Quezel, P. & Medail, F. (2004) Ecologie etbiogeographie des forets du bassin mediterraneen.Elsevier Masson, Paris.

Ricklefs, R. E. (1987) Community diversity: rela-tive roles of local and regional processes. Science(New York, N.Y.), 235, 167–171.

Ricklefs, R. E. (2004) A comprehensive framework forglobal patterns in biodiversity. Ecology Letters, 7,1–15.

Rivas-Martınez, S., Dıaz, T., Fernandez-Gonzalez, F.,Izco, J., Loidi, J., Lousa, M. & Penas, A. (2002)Vascular plant communities of Spain and Portu-gal: addenda to the syntaxonomical checklist of2001. Part II. Itinera Geobot., 15(2), 433–922.

Rivas-Martınez, S., Penas, A. & Dıaz, T. (2004a) Bio-climatic Map of Europe. University of Leon.

Rivas-Martınez, S., Penas, A. & Dıaz, T. (2004b) Bio-geographic Map of Europe. University of Leon.

Rosenbaum, G., Lister, G. S. & Duboz, C. (2002) Re-construction of the tectonic evolution of the west-ern Mediterranean since the Oligocene. Journalof the Virtual Explorer, 8, 107–130.

Rosvall, M. & Bergstrom, C. T. (2008) Maps ofrandom walks on complex networks reveal com-munity structure. Proceedings of the NationalAcademy of Sciences, 105, 1118–1123.

Rousseeuw, P. J. (1987) Silhouettes: A graphicalaid to the interpretation and validation of clusteranalysis. Journal of Computational and AppliedMathematics, 20, 53 – 65.

Rundel, P. W., Arroyo, M., Cowling, R. M., Keeley,J. E., Lamont, B. B. & Vargas, P. (2016) Mediter-ranean Biomes: Evolution of Their Vegetation,Floras, and Climate. Annual Review of Ecology,Evolution, and Systematics, 47, 383–407.

Rushton, S. P., Ormerod, S. J. & Kerby, G. (2004)New paradigms for modelling species distribu-tions? Journal of Applied Ecology, 41, 193–200.

Saiz, J. C. M., Parga, I. C. & Ollero, H. S. (1998) Nu-merical analyses of distributions of Iberian andBalearic endemic monocotyledons. Journal ofBiogeography, 25, 179–194.

Stewart, J. R. & Lister, A. M. (2001) Cryptic north-ern refugia and the origins of the modern biota.Trends in Ecology & Evolution, 16, 608–613.

Stoddart, D. R. (1992) Biogeography of the TropicalPacific. University of Hawaii Press.

Tassin, C. (2017) Paysages vegetaux du domainemediterraneen : Bassin mediterraneen, Cali-fornie, Chili central, Afrique du Sud, Australiemeridionale. Rfrence. IRD ditions, Marseille.

Thompson, J. D. (2005) Plant Evolution in theMediterranean. Oxford University Press, Oxford,New York.

Thompson, J. D., Lavergne, S., Affre, L., Gaudeul,M. & Debussche, M. (2005) Ecological Differen-tiation of Mediterranean Endemic Plants. Taxon,54, 967–976.

Tison, J.-M. & Foucault, B. d. (2014) Flora Gallica :Flore de France. Biotope Editions, Mze.

Tison, J.-M., Jauzein, P. & Michaud, H. (2014) Florede la France mediterranenne continentale. Natu-ralia Publications, Turriers, 1re dition edition.

Turner, M. G., Gardner, R. H. & O’Neill, R. V. (2001)Landscape Ecology in Theory and Practice: Pat-tern and Process. Springer-Verlag.

Vilhena, D. A. & Antonelli, A. (2015) A networkapproach for identifying and delimiting biogeo-graphical regions. Nature Communications, 6,6848.

Wahlenberg, G. (1812) Flora Lapponica. Taberna li-braria scholae realis, Berolini.

Wallace, A. R. (1876) The geographical distribution ofanimals : with a study of the relations of livingand extinct faunas as elucidating the past changesof the earth’s surface. Harper and Brothers, Pub-lishers, New York.

Walter, H. & Breckle, S.-W. (1991) Okologie der Erde.

Band 4: Spezielle Okologie der Gemaßigten undArktischen Zonen außerhalb Euro-Nordasiens.Gustav Fischer Verlag.

16

Walter, H. & Breckle, S.-W. (1994) Okologie der Erde.

Band 3: Spezielle Okologie der Gemaßigten undArktischen Zonen außerhalb Euro-Nordasiens.Gustav Fischer Verlag.

Wilson, M. V. & Shmida, A. (1984) Measuring BetaDiversity with Presence-Absence Data. Journalof Ecology, 72, 1055–1064.

APPENDIX

Interactive web application

An interactive web application has been designed to provide an easy-to-use interface to visualize the resultsand the maps of the main paper and the Appendix (Figure S1). The source code of the interactive webapplication1 can be downloaded from2.

Figure S1. Screenshot of the interactive web application.

1 https://maximelenormand.shinyapps.io/Biogeo/ 2 www.maximelenormand.com/Codes

17

Influence of scale on the biogeographical regions delineation

In order to assess the impact of the spatial resolution on the results, we also applied the analysis with a gridcomposed of squares of lateral size l = 10 km (Figure S2). The spatial coherence, defined as the ratio betweenthe number of grid cells in the largest patch and the total number of grid cells (Turner et al., 2001), is displayedfor both scale in Table S1.

Mediterranean border

Cork oak zone

Cévennes sensu latoSubatlantic mountains

Mediterranean lowlands

Pre-AlpsHigh mountains

Gulf of Lion coast1

2

3

45

6

7

8

Figure S2. Biogeographical regions based on similarity in plant species (l = 10 km). Eight biogeographical regions havebeen identified. 1. Gulf of Lion coast in red. 2. Cork oak zone in orange. 3. Mediterranean lowlands in light green. 4.Mediterranean border in dark green. 5. Cevennes sensu lato in purple. 6. Subatlantic mountains in pink. 7. Prealps andother medium mountains in yellow. 8. High mountains in brown.

Table S1. Spatial coherence of the biogeographical regions according to the scale.

Bioregion ni (l=5) SPi (l=5) ni (l=10) SPi (l=10)

1 170 0.63 63 0.75

2 183 0.73 47 0.87

3 529 0.86 124 0.83

4 807 0.57 164 0.51

5 120 0.50 45 0.31

6 152 0.70 48 0.79

7 400 0.67 114 0.75

8 246 0.78 110 0.77

18

Comparison of OSLOM to standard clustering methods

In this work, bioregions are delineating using the community detection algorithm OSLOM applied on aweighted undirected spatial network whose intensity of links between grid cells are measured with the Jaccardsimilarity coefficient. This algorithm is nonparametric in the sense that it identifies statistically significantcommunities with respect to a global null model, and therefore the number of communities does not need to bedefined a priori. In order to assess the accuracy of the method, we compared the results obtained with OSLOMwith the ones obtained with standard hierarchical clustering methods. Not that these standard methods cannotbe directly applied on the spatial network described above, we first need to transform the network into adissimilarity matrix. Three different agglomeration methods have been tested: average (UPGMA), mcquitty(WPGMA) and Ward1. To choose the number of clusters, we used the average silhouette index S (Rousseeuw,1987). For each cell g, we can compute a(g) the average dissimilarity of g (based on the Jaccard index in ourcase) with all the other cells in the cluster to which g belongs. In the same way, we can compute the averagedissimilarities of g to the other clusters and define b(g) as the lowest average dissimilarity among them. Usingthese two quantities, we compute the silhouette index s(g) defined as,

s(g) = b(g) − a(g)max{a(g), b(g)} (1)

which measures how well clustered g is. This measure is comprised between −1 for a very poor clustering qualityand 1 for an appropriately clustered g. We choose the number of clusters that maximize the average silhouetteindex over all the grid cells S = ∑ng=1 s(g)/n.

UPGMA and WPGMA failed to detect any coherent partitions, most of the grid cells were gathered in agiant cluster component even increasing significantly the number of clusters. Better results were obtained withWard’s method. The average Silhouette index as a function of the number of clusters is shown in Figure S3.

5 10 15 20

0.00

0.02

0.04

0.06

0.08

0.10

Number of Clusters

Ave

rage

Silh

ouet

te

HC WardOSLOM

Figure S3. Average Silhouette as a function of the number of clusters obtained with Ward’s clustering (in blue) andOSLOM (in red).

Two optimal partitions have been detected with the average Silhouette index. It is interesting to note thatthe number of clusters of the second partition is the same that the one automatically detected with OSLOM(Figure S3).

A map of the eight optimal bioregions obtained with Ward’s method is display in Figure S4. In order tocompare the two partitions a contingency between the partitions obtained with Ward’s method and OSLOM isshown in Table S2.

1 method=”average”, ”mcquitty” and ”ward.D2” with thehclust R function

19

1

2

3

45

6

7

8

Figure S4. Biogeographical regions based on similarity in plant species obtained with Ward’s clustering (l = 5 km).Eight biogeographical regions have been identified.

Table S2. Contingency tables between the partitions obtained with Ward (in row) and OSLOM (in column).

Bioregion 1 2 3 4 5 6 7 8

1 125 2 54 0 4 0 0 0

2 0 115 0 2 0 0 0 0

3 28 47 435 396 0 0 0 0

4 0 4 4 187 48 1 43 1

5 17 15 34 20 53 32 8 15

6 0 0 0 0 15 119 38 33

7 0 0 0 200 0 0 216 0

8 0 0 0 2 0 0 95 197

20

Comparison with other delineations

Lamarck (1805)

Figure S5. Comparison of the results obtained with OSLOM (l = 5 km) with Lamarck’s limit of the Mediterraneanlevel (Ebach & Goujet, 2006; Lamarck & Candolle, 1805).

Flahaut (1887)

Figure S6. Comparison of the results obtained with OSLOM (l = 5 km) with Flahaut limit of the olive tree distribution(Flahault & Durand, 1887).

21

Ozenda (1994)

Figure S7. Comparison of the results obtained with OSLOM (l = 5 km) with Ozenda’s mediter-ranean/supramediterranean limit (Ozenda, 1994).

Julve (1999)

Figure S8. Comparison of the results obtained with OSLOM (l = 5 km) with Julve’s mediterranean/supramediterraneanlimit (?).

22

Bohn (2000)

Figure S9. Comparison of the results obtained with OSLOM (l = 5 km) with Bohn’s mediterranean/supramediterraneanlimit (Bohn et al., 2000).

Rivas−Martinez (2004) − thermoclimat

Figure S10. Comparison of the results obtained with OSLOM (l = 5 km) with Rivas-Martınez’s thermoclimatic limit(Rivas-Martınez et al., 2004a).

23

Rivas−Martinez (2004) − biogeo

Figure S11. Comparison of the results obtained with OSLOM (l = 5 km) with Rivas-Martınez’s biogeographical limit(Rivas-Martınez et al., 2004b).

Natura 2000 (2006)

Figure S12. Comparison of the results obtained with OSLOM (l = 5 km) with the Natura 2000’s limit (?).

24

Supplementary Figures

0 200 400 600 800 1000

0

100

200

300

400

500

Plant species per grid cell

Fre

quen

cies

(a)

0 500 1000 1500 2000 2500

0

500

1000

1500

2000

2500

Grid cells covered per plant species

Fre

quen

cies

(b)

Figure S13. Histograms of the degree distributions of the biogeographical bipartite network. Histogram of the numberof plant species per grid cell (a) and the number of cells covered per plant species (b).

0

500

1000

1500

2000

2500

3000

3500

Figure S14. Altitude map of the studied area (in meters).

25

0.0

0.2

0.4

0.6

0.8

1.0

1 − J2/J1

Figure S15. Uncertainty map (l = 5 km). For a given cell, J1 represents the average Jaccard similarity index between thiscell and all the cells that belong to its cluster, and J2 represents the average Jaccard similarity index between this cell and allthe cells belonging to the second closest cluster (based on the Jaccard similarity).

26

Supplementary Tables

Table S3. Number of plant species per group.

Group Number of species

a 445

b 149

c 230

d 299

e 169

f 277

g 37

h 180

i 242

j 136

k 125

l 95

m 180

n 180

o 178

p 186

q 44

r 59

s 212

t 274

Table S4. Network of interactions between biogeographical regions.

Bioregion 1 2 3 4 5 6 7 8

1 0.52 0.24 0.15 0.07 0 0.01 0.01 0

2 0.15 0.56 0.11 0.14 0.01 0.01 0.02 0

3 0.13 0.18 0.42 0.25 0 0 0.01 0

4 0.03 0.11 0.1 0.55 0.01 0.01 0.19 0.01

5 0.01 0.13 0 0.07 0.4 0.24 0.11 0.05

6 0.01 0.03 0 0.02 0.1 0.49 0.17 0.19

7 0 0.01 0 0.16 0.01 0.04 0.52 0.26

8 0 0 0 0.01 0.01 0.05 0.28 0.65