Org Divers Evol (2014) 14:203–214 DOI 10.1007/s13127-013-0161-3

ORIGINAL ARTICLE

Coupling impoverishment analysis and partitioning of beta diversity allows a comprehensive description of biogeography in the Western Mediterranean

Markus Heiser & Leonardo Dapporto & Thomas Schmitt

Received: 31 March 2013 /Accepted: 24 October 2013 /Published online: 14 November 2013 # Gesellschaft für Biologische Systematik 2013

Abstract Islands host a subset of organisms occurring at their Simpson compared to Sørensen index, and indicator sources, and these assemblages are usually dominated by the from islands where unpredicted to occur by impoverishment most generalistic and dispersive species. In this study, we aim analysis. This suggests that island species predicted absent to identify which species are missing on islands and which determine most of an island’s turnover pattern, thus ecological traits are responsible for differential occurrence. encompassing fundamental biogeographic information. Due Then, we apply this information to beta diversity analyses. to their absence on nearest sources, they are also at higher risk As a study group and area, we selected the Odonata in the of extinction, and deserving of special conservation effort. Western Mediterranean. Based on the presence/absence of 109 species, we applied a series of analyses at both commu- Keywords Dragonflies . Island biogeography . Community nity and individual species level. The islands of the Balearics, level . Species assemblages . Species impoverishment . Corsica, Sardinia and Malta are highly impoverished, but Dispersal capacity . Species-area relationship Sicily is not. Non-parametric multivariate adaptive regression splines predicted the occurrence of individual species on each island. Principal component analysis recognised differences Introduction between (Zygoptera) and dragonflies (Anisoptera), but members of the two suborders have similar The equilibrium theory of island biogeography (MacArthur occurrences on islands, and island occurrence is determined and Wilson 1967) is one of the most challenging theories in mostly by species’ frequencies at source and by their degree of ecology. One of this theory’s most important assumptions generalism. Island species predicted correctly to occur on predicts that immigration and extinction reach an equilibrium islands showed opposite characteristics to species unpredicted after an island-specific time of constant environmental condi- to occur and being present. The similarity pattern highlighted tions, thus stabilising species richness. Consequently, mature by turnover (Simpson index) is clearer than that obtained by island communities should host a defined number of species non-partitioned beta diversity (Sørensen index). In fact, indi- as a function of their area and isolation (MacArthur and cator value analyses revealed more indicator species for the Wilson 1967; Emerson and Gillespie 2008; Lomolino et al. 2010; Losos and Ricklefs 2010). Furthermore, the biotic com- position of the neighbouring areas is of major importance Electronic supplementary material The online version of this article because it represents the source of potential immigrants (doi:10.1007/s13127-013-0161-3) contains supplementary material, which is available to authorized users. (Williamson 1981; Dennis and Shreeve 1996; Santos et al. : 2011). M. Heiser T. Schmitt This line of reasoning introduces a corollary of the equilib- Department of Biogeography, Faculty of Geography / Geosciences, Trier University, 54286 Trier, Germany rium theory: due to their isolation, and the limited variety of ecological niches and geographic extent, islands usually host a L. Dapporto (*) limited fraction of their source fauna and flora and are thus Centre for Ecology, Environment and Conservation, Department of “impoverished” compared to their adjoining mainland regions Biological and Medical Sciences, Oxford Brookes University, Headington, Oxford OX3 0BP, UK (Williamson 1981; Whittaker and Fernández-Palacios 2007). e-mail: [email protected] The relationships between numbers of species occurring on 204 M. Heiser et al. islands and island characteristics have been tested in a plethora should be strong relationships between species member- of studies (see Whittaker and Fernández-Palacios 2007; Losos ships and their ecological traits; in such cases, SpaP should and Ricklefs 2010 for recent reviews). However, the “hidden” represent a very small fraction of species. Nevertheless, side of richness (i.e. the quantity and identity of species they usually occur, and have been mostly explained, as lacking on an island as a consequence of its geographic and relict island populations whose occurrence cannot be ex- ecological characteristics) has so far been mostly neglected. plainedonthebasisofmainlanddistribution(Dapportoand The amount of missing species can be assessed with a rela- Dennis 2009, 2010; Dapporto 2011). tively simple comparison of richness between islands and The results obtained for three groups of Lepidoptera neighbouring mainland areas, which are assumed not to be (Papilionidea, Hesperidae and Zygaena moths, Dapporto impoverished (Rosenzweig 1995; Whittaker and Fernández- and Dennis 2009, 2010;Dapporto2011)showedsomesimi- Palacios 2007; Dapporto and Dennis 2009). larities among these groups, but also effects deriving from Source species do not have the same probability to enter an different dispersal ability. Thus, the Zygaena moths, with their island fauna since they differ in dispersal and colonization poor dispersal, showed a more impoverished pattern and a abilities. In this context, larger and less isolated islands offer higher frequency of relict populations than members of other chances also for less opportunistic species (Diamond 1975; groups. Santos et al. 2011;Dennisetal.2012). Deviations from this Moreover, a recent debate on partitioning beta diversity expected pattern occur when an island has been colonised provided new insights into the possibility of separating the from different source assemblages. Such island faunas are effects of differences in richness/nestedness patterns and dif- largely the consequence of stochastic events of immigration ferences due to species replacement (turnover) in determining and extinction, and the spatial complexity deriving from these inter-area dissimilarities (Koleff et al. 2003;Baselga2010; historic phenomena. Long-term processes can also produce Carvalho et al. 2012). The turnover component of diversity unexpected patterns, and species may survive on islands lon- can represent a fundamental and unbiased component for ger than on the mainland, thus producing unexpected distri- identifying biogeographic regionalisation (Kreft and Jetz butions (Masini et al. 2008; Médail and Diadema 2009; 2010); indeed, it encompasses the decisive role of endemic Dapporto et al. 2011, 2012; Dapporto and Bruschini 2012), and non-widespread species (those replacing others in restrict- or they can evolve into endemic taxa (Whittaker and ed island areas). Fernández-Palacios 2007;Kudrnaetal.2011). Endemic and From this perspective, there is strong potential that SpaP relict species represent primary targets for biodiversity con- species, occurring on islands but most likely absent in the servation, and their identification represents an important nearest and most similar mainland units, account for most challenge. However, biogeographic information retained by island turnover patterns. In this study, we couple, for the first the reduced set of endemic or subendemic species can be time, the Dapporto and Dennis (2009) method with hidden by the high frequency of widespread species when partitioning of biodiversity as introduced by Baselga (2010). diversity is examined at the community level. We aim to identify specific island and mainland species re- Therefore, Dapporto and Dennis (2009) provided a sponsible for the turnover component and to recognise their method for (1) the evaluation of island impoverishment, putative dynamics responsible for island occurrence by im- (2) the identification of species predicted to be missing on poverishment analysis. islands, (3) the detection of ecological traits of species In doing so, we selected Odonata, an order responsible for differential occurrence and, finally, (4) the characterised by two suborders: the well dispersing dragon- demarcation of potential relict populations. In practice, flies (Anisoptera) and the poorly dispersing damselflies impoverishment analysis distinguishes between species (Zygoptera) (Corbet 2004). As study region, we selected the predicted to be present (Spp) and species predicted to be Western Mediterranean, which was also used in previous absent (Spa) at each island. Spp are subsequently classified studies, so as to allow a direct comparison with Lepidoptera. in those actually present (SppP) and those unexpectedly The distribution of the Odonata is well assessed at the regional absent (SppA), while Spa can be absent (SPAA) or unex- scale in the study area, thus allowing almost unbiased pectedly present (SpaP). SPAA represents species whose modelling. distributions do not fit the geographic characteristics of islands, and actually are not found there. From this perspec- tive, they have no biogeographic relevance. Distinctions Material and methods among the remaining three groups (SppP, SppA, SpaP) are fundamental keys to the recognition of deterministic Study area and study group factors in island occurrences. SppA represents determinis- tic island impoverishment and, if occurrence at different The Western Mediterranean comprises some islands or islands is driven strictly by particular species traits, there island groups (Balearics, Corsica, Sardinia, Sicily and Impoverishment analysis and partitioning of beta-diversity 205

Malta) characterised by relatively large areas. We divided Following the nomenclature of Dijkstra and Lewington the study area into geographic regions based on their (2006), the Odonata embrace 109 species in the study area. respective ecological and orographic structures (Dennis Of these species, 38 belong to the damselflies (Zygoptera) and et al. 1991; Heiser and Schmitt 2010). Although grid- 71 to the dragonflies (Anisoptera). The rather different dis- based data are available for the distributions of Odonata persal capacities of these two groups (Corbet 2004)allow in the Western Mediterranean (Boudot et al. 2009), we based testing the impact of this factor on the equilibrium species our analyses on regions as we intend to compare the fauna of number on islands. entire islands and not of their fractions (i.e. grids). Conse- Presence data were derived from numerous distribution quently, comparing entire island faunas only to comparable atlases and checklists of odonates throughout the Western regions on the mainland allows a comprehensive understand- Mediterranean region (for details see Heiser and Schmitt ing of the impoverishment and restructuring processes at these 2010). For each of the 24 regions, the presence was deter- islands (see below). Moreover, the use of such natural regions mined for each of the 109 species existing in the entire study is very useful to depict area-relationships and species move- area (Appendix S1). Presence data of Odonata at the regional ments, because it emphasises the influence of true natural scale were assessed by means of hundreds of studies over the barriers/connections among areas in determining species dis- last two centuries; thus, species absences at this scale are more tributions more than arbitrarily defined squares. This likely attributable to true gaps in the distribution of species regionalisation results in 24 geographic units (Fig. 1). We rather than to sampling incompleteness. Although absences determined the following characteristics for all island and cannot be easily proved, we can be fairly confident that mainland areas: area (A, km2), maximum altitude (AL, m), methods designed for presence/absence data can be applied geographic location at a western Mediterranean coastline (CR, to this dataset. We assessed the following traits for each as binary variable), mean latitude and mean longitude (LTand species: suborder membership (Zygoptera or Anisoptera), LG, decimal degrees) as well as latitudinal and longitudinal voltinism (V, generations per years), wingspan (WS, mm), ranges (LTR, LGR, decimal degrees). flight period (FP, months), maximum altitude at which the In principle, latitude and longitude are not the proximate species is found (ALmx, m), duration of the larval period (LP, drivers in shaping faunas determined by complex interactions years or fractions), occurrence in the Western Palaearctic between ecological (temperature, precipitation, productivity, regions (RS, scores derived from Heiser and Schmitt 2010) habitat) and historical (glacial refugia, physical barriers) de- and occurrence in regions with a western Mediterranean coast- terminants. However, latitude and longitude are important line (SR, scores derived from our data) as continuous variables proxies encompassing most of these determinants in the case (Appendix S1). of the Mediterranean. The main aim of the spatial analysis of this study is to predict distributions of species on islands based Community level analysis on the knowledge of their mainland distributions. From this perspective, we are not interested in finding the determinants Determinants for overall mainland richness were tested in for mainland occurrence (ecological or historical), but in preliminary univariate regressions and t-test (for CR). Then, recognising whether a given species occurring on certain the overall impoverishment of the five islands was tested by mainland regions (for a combination of such determinants), stepwise multiple linear regressions using the Akaike criterion also occurs on neighbouring islands, which are predicted to be to select the best model as implemented in the R package likely subjected to similar ecological and historical “MASS”. We used log-transformed species richness per re- determinants. gion as the dependent variable and transformed geographic characteristics for predictors when necessary to improve nor- mality. In particular, we log-transformed area (A), maximum altitude (AL) and latitude and longitude ranges (LTR, LGR). We first performed the model only on mainland data; then, we predicted the theoretical numbers of species for islands based on this model. To test whether the possible impoverishment of the entire dataset can be explained by different effects in the two suborders, we also analysed Anisoptera and Zygoptera separately.

Individual species level analyses

Fig. 1 Study area and its biogeographic subdivision based on orographic In previous papers applying impoverishment analyses, a series elements sensu Illies (1976) and modified by Heiser and Schmitt (2010) of logit GLZ was used towards this aim (Dapporto and Dennis 206 M. Heiser et al.

2009, 2010;Dapporto2011). For butterflies, presence data are among them. To analyse differences in frequencies between even more reliable than for Odonata and this allowed the use suborders in SppP, SppA, SpaP, we performed a G test for of a higher spatial resolution with more cases (regions) in the each island. analyses. Preliminary tests with logit GLZ in the Odonata dataset yielded unreliable results. These problems can be Partitioning of beta diversity solved by using multivariate adaptive regression splines (MARS) (Friedman 1991), which allow non-parametric esti- Two main phenomena interact in the evolution of community mation of both linear and non-linear relationships. In MARS, patterns in species composition: nestedness and spatial turn- relationships between dependent and independent variables over (Baselga 2010;Carvalhoetal.2012). The first phenom- are fitted using piecewise linear segments defined by pairwise enon reflects an ordered loss of species as faunas become basis functions (Friedman 1991; Elith and Leathwick 2007). poorer in species. On the other hand, spatial turnover implies The basis functions are selected in a forward stepwise entering the replacement of a species with another species in different of functions, until a model of maximum complexity is obtain- areas. From this perspective, unpartitioned diversity captures ed. Then, a backward procedure removes not contributing all the components of faunas, while turnover removes the predictors. We used the R package “mda” in combination with effect of systematic island impoverishment. Turnover is thus functions provided by Elith and Leathwick (2007)tofita predicted to give a special importance to vicariant species with generalised linear model with binomial error distribution, respect to the contribution of many widespread species. using the basis functions obtained by MARS. The presence- Baselga (2010) proposed three indices to partition diversity absence data of each species for the 19 mainland regions into these components: (1) the Sørensen dissimilarity index represented the dependent variable, and AL, CR, LT and LG between two areas [βsor=(b +c)/(2a +b +c)wherea is the regional characteristics were used as predictors. After deter- number of shared species and b and c the species exclusive of mining MARS models on mainland, they were applied to the first and the second site] comprises both, turnover and geographic traits of islands to predict presence or absence of nestedness, but it can be portioned as the sum of (2) the the individual species. turnover Simpson dissimilarity index [Simpson 1943, We excluded species absent in less than four mainland βsimp=min(b,c)/(a+min(b,c)] and (3) the nestedness index, regions and species present in less than four regions as they created by Baselga (2010, 2012)asβsor – βsimp. are unsuitable for any regression model. The former were To highlight and compare the patterns of faunal dissimilar- classified as ubiquitous, the latter as rare. Ubiquitous species ity obtained by Sørensen and Simpson indices, we performed were predicted present at all islands, rare species as absent at cluster analyses with Ward’s method. When dissimilarity ma- all islands (Dapporto and Dennis 2009). The resulting analysis trices of beta diversity turnover indices are projected in den- distinguished between species predicted to be present (Spp) drograms, a high frequency of ties and zero values produces and species predicted to be absent (Spa) at each island. Spp are trees whose topology and bootstrap supports are affected by subsequently classified in those actually present (SppP) and the order of areas in the original presence-absence matrix those unexpectedly absent (SppA), while Spa can be absent (Dapporto et al. 2013). In order to avoid such a bias, we used (SPAA) or unexpectedly present (SpaP). the “recluster” R package (available at http://cran.r-project. If occurrence at different islands is determined by specific org/), which re-samples row orders of the original matrix and species traits, there should be a strict correlation between creates consensus trees in bootstrap analysis (Dapporto et al. species memberships and their ecological traits. To highlight 2013). The recluster package allows diagnostic analyses for such complex correlations, we performed dudi.mix PCA as the identification of row order bias. According to diagnostic implemented by the R package “ade4”.Thisanalysisgener- results, we assessed the optimal parameters (number of trees alises the Hill and Smith (1976) method, thus allowing the and consensus rule) to be used in multilevel bootstrap. Due to inclusion of factorial and count variables. We included species the highly nested faunal structure of islands, turnover is usu- membership to Zygoptera or Anisoptera as categorical factor, ally determined by a small subset of species. The bootstrap voltinism (V) as a count variable and wingspan (WS), flight procedure relying on a subset of species is likely to remove a period (FP), maximum altitude (ALmx), duration of larval large part of such a small subset from most bootstrap trees and period (LP), occurrence in West Palaearctic regions (RS) and to assign low support to most nodes (Dapporto et al. 2013). In occurrence in regions with western Mediterranean coastline multiscale bootstraps, an increasing number of species is used (SR) as continuous variables. We log-transformed WS, FP and for bootstrapping at each step, thus allowing an increasing LP and sqrt-transformed RS and SR to improve normality. probability for the occurrence of species responsible for turn- The three groups of species (SppP, SppA and SpaP) were over in the analysis. For this reason, a node having increas- finally compared for their ecological traits by Kruskal-Wallis ingly strong support in a multiscale bootstrap means that the non-parametric ANOVA in order to test whether the relation- areas connected by this node host only a few species respon- ships highlighted in the PCA also show significant differences sible for turnover, but that the biogeographic pattern with Impoverishment analysis and partitioning of beta-diversity 207 respect to other areas is clear. Conversely, nodes showing constantly low support represent links among areas that are actually uncertain (Dapporto et al. 2013). We thus computed five levels of bootstrap (x1 to x5) using from one- to five-fold numbers of species in the original data frame. Finally, in order to identify species most involved in cluster identification, we performed an indicator species analysis (Dufrêne and Legendre 1997). This method provides an indicator value (IndVal) for each species, combining its frequency within and among clus- ters. We identified a series of clusters from recluster analysis and applied the IndVal analysis. The significance of the IndVal was assessed by permutation test using 999 permutations. Species with significant P values and IndVals higher than 0. 50 for a cluster were then indicated as indicator species. We Fig. 2 Observed versus predicted values of species’ richness (LogR) used the R package “labdsv” for IndVal analysis. based on multiple regression analysis for all species. See Fig. 1 for regional key; abbreviations in figures: A Balearics; B Corsica; C Sardin- ia; D Sicily; E Malta Results longitude entered in all these models. Latitude revealed the Overall impoverishment strongest power, being selected as predictor in 47 species; longitude was selected in 25 species. In addition, five models Preliminary univariate tests revealed that only latitude has a included the existence of a Mediterranean coastline and three direct effect on richness of mainland areas (Appendix S2). By models entered maximum altitude as additional variable to applying a multiple regression with Akaike selection for all latitude and longitude. Odonata species (F =11.63, R2=0.769, P <0.001), regional The numbers of species predicted to be present by the species richness in the Western Mediterranean region was MARS were quite similar to those found by the multiple linear confirmed to be largely predicted by increasing latitude (LT; regression model (Balearics: 48 species, Corsica: 55, Sardinia: β=1.044, P <0.001); moreover, decreasing longitude (LG; 50, Sicily: 45, Malta: 50). Furthermore, the existence of some β=–0.449, P =0.024), the occurrence of a Mediterranean species predicted to be absent on the islands was surprising. coastline (CR; β=0.238, P =0.090) and latitudinal extension So, Corsica had four SpaP species, Sardinia and Sicily eight (LTR; β=0.297,P =0.144) also entered the model, despite not and Malta three SpaP. Only Balearics showed no SpaP. all these variables showing a significant effect. The plot of observed versus predicted richness identifies the Balearics Ecological factors of species occurrence (observed=24, predicted=53), Corsica (observed=42, pre- dicted=57), Sardinia (observed=40, predicted=54) and Malta Principal component analysis (PCA) for the five islands (observed=13, predicted=33) as highly impoverished due to identified a rather common pattern for the associations lying below the 95 % confidence interval. Only Sicily did not between species characteristics and SppP, SppA and SpaP show any signal of impoverishment (observed=45, predict- membership (Fig. 3). SppP were strictly associated with ed=44) (Fig. 2). Individual regressions for Zypotera and wider distributions in Europe (RS) and in the Mediterra- Anisoptera resulted in similar results (see Appendix S2). nean area (SR) as well as with the maximum altitude where a species was found (ALmx). These relationships were Predicted species’ occurrences on islands significant in Kruskal-Wallis tests for all islands except for the Balearics (Fig. 4). SpaP showed opposite charac- Among the 109 Odonata species examined here, 26 were teristicswhencomparedtoSppPasrevealedbyPCA assigned as ubiquitous and were thus predicted to occur on scatterplots and by boxplots (Figs. 3, 4). Mann-Whitney all islands or island groups (hereafter simply called islands), tests also supported the highly differing median values for while 18 species were assigned as rare and were predicted to most traits (Fig. 4). SppA showed less contrasting traits in be absent on all islands. For eight species having high or low PCA except for Balearics where SpaP did not occur and, mainland frequencies, MARS did not highlight any effect for actually, they tended to show intermediate trait values be- geographic variables and predicted these species to occur or to tween SppP and SpaP (Fig. 4). be absent in all regions, respectively. For the remaining 57 Zygoptera and Anisoptera membership tended to be species, MARS modelled species occurrence on the basis of polarised with respect to membership for occurrence. This geographic predictors (Appendix S1). Latitude and/or suggested that Zygoptera and Anisoptera have similar 208 M. Heiser et al.

Fig. 3 a–e Scatterplots from dudi.mix principal component analysis (PCA) representing axes coordinates of species traits and species membership. V. L .and V.Q . represent high and low values of voltinism (V), respectively (see methods for other abbreviations). a Balearics, b Corsica, c Sardinia, d Sicily, e Malta

probabilities of occurring on islands when predicted to be S2 for details). Conversely, it has a great effect on trees present or to occur when unpredicted to be so. The only constructed by Simpson dissimilarity. For this reason, we exception is represented by Malta, where SppP were associ- carried out classical bootstrapping for Sørensen index (i.e. a ated with Anisoptera in the first component (Fig. 3e). These series of single trees with re-sampled species) and a consensus results were confirmed by G tests revealing no significant bootstrap for Simpson index (a bootstrap made of 1,000 trees, effects in any island except for Malta (Fig. 4). each based on 100 row-re-sampled trees with a consensus rule of 50 %) (see Appendix S2 for details). Partitioning of beta diversity Cluster analysis with Sørensen index revealed that richness has a greater power to determine the grouping of areas The diagnostic analyses performed by recluster revealed that (Fig. 5a). Actually, when examined in detail, the richest areas row order has a low biasing effect on the topology of trees of the Alps and France clustered together with the Pyrenees in constructed over the Sørensen dissimilarity matrix (Appendix a first cluster; in a second cluster, the Italian regions were Impoverishment analysis and partitioning of beta-diversity 209

Fig. 4 Boxplot illustrating median differences in species traits among SppA (black boxes), SppP (white boxes)andSpaP (grey boxes). Asterisks indicate significant effects for non- parametric Kruskal-Wallis ANOVA. The last row shows differences among the same classes of species in relative frequency of Anisoptera (black) and Zygoptera (white); asterisks indicate significant effects for G tests

grouped with the exception of Calabria (15). In the second identified, a fourth insular cluster grouped Sardinia, Sicily and group of clusters, the Iberian regions represented the third Malta and, finally, a fifth grouped Iberia and the Balearics. cluster and the Maghreb the fourth. All islands and area 15 Support for nodes seemed to be in general similar to that comprised the fifth cluster. obtained by the Sørensen index with higher nodes receiving The examination of indicator species revealed that the Alps- generalised low support. The main difference is represented France-Pyrenees cluster had eight species satisfying the selection by the island group that received an increasing support criterion (Appendix S1 for species identification). The Maghreb reaching 70 % at the x5 level. This suggested that there was was also well identified by six species. Iberia had a single a small subset of species responsible for this pattern. Finally, indicator species, while Italy and islands with Calabria did not we found more indicator species for the Simpson index com- show any indicator species (Fig. 5a). Support for most of the five pared to the Sørensen index: 14 indicator species were found groups tended to be acceptable mostly at the x5 level, while for the Alps, Pyrenees and Massif Central, 10 for the Ma- higher nodes and the island group showed a very low support ghreb, 2 for Italy-southern France-Corsica, 1 for Iberia and among the 1,000 bootstrap replicates at both x1 and x5 levels. Balearics and 2 for Sardinia, Sicily and Malta. The indicator This suggests that faunal relationships among islands and among species for the island cluster are two SpaP for these islands higher order clusters and islands are actually uncertain. (Orthetrum trinacria, genei). The analysis of species turnover resulted in a rather different pattern (Fig. 5b). Among the five evident clusters (Fig. 5b), the first was still represented by Alps, Pyrenees and the Massif Discussion Central, but the other species-rich Southern France area was grouped in the second cluster with all Italian areas including the The analyses presented here were performed at both the island of Corsica. In the third cluster, the Maghreb was still community and the individual species level, on an insect 210 M. Heiser et al.

Fig. 5 Cluster analysis with multiscale bootstrap values for a the Sørensen index and b the Simpson index. The number of IndVal significant species is also reported for each cluster group (Odonata) dominated systematically and ecologically revealed a similar pattern, with Sørensen dissimilarity being by the existence of two well distinct suborders (Zygoptera and linked to a large number of species determining overall impov- Anisoptera). These suborders show great differences in many erishment of areas while the Simpson index of turnover was morpho-functional and ecological traits, but relatively minor determined mostly by a reduced set of species with restricted differences among species of the same group (Corbet 2004). distribution. For islands, these species are SpaP as highlighted Nevertheless, analyses at both community and individual by impoverishment analyses for individual species. species level revealed a weak influence of this dichotomy in determining island distributions, except for the very small and Community level isolated islands of Malta. Occurrence on islands is most often determined by the A first important consideration needs to be made regarding the frequency in Mediterranean sources and by the degree of lack of expected occurrence differences between the two subor- species’ generalism, measured as a higher tolerance to altitudi- ders showing striking differences in dispersal capacities. Only in nal ranges (Fattorini et al. 2013). However, a set of species that the most extreme case of the Maltese islands, showing a com- is unexpectedly present on islands showed the opposite trend to bination of small area, high isolation and reduced resource such traits, suggesting the existence of complex colonization availability, were Zygoptera more impoverished than and persistence histories. Partitioning of community diversity Anisoptera. This confirms that area and isolation interact Impoverishment analysis and partitioning of beta-diversity 211 strongly with dispersal capacity in determining occurrence Northern Europe (Dennis et al. 1991; Varga and Schmitt (Dennis et al. 2012), but allow a strong filtering only for the 2008; Hawkins 2010;HeiserandSchmitt2010); the Maghreb smallest and most isolated islands. Heiser and Schmitt (2010) collects many sub-eremic and eremic species absent elsewhere found a similar pattern in the examination of the entire Western in the Mediterranean (Tennent 1996;DijkstraandLewington Palaearctic region and concluded that the lack of strong differ- 2006; Heiser and Schmitt 2010). These strong faunistic dif- ences in distribution between the two suborders could be due to ferences among different areas are due not only to climatic an ancient history of colonisation and long-term persistence. variation driven by latitudinal effects, but important historical Mediterranean islands represent important glacial refugia and events shaped the distributions of most Mediterranean organ- several endemic species and lineages evolved there (e.g. isms (Schmitt 2007; Médail and Diadema 2009; Fattorini and Cesaroni et al. 1994; Fattorini 2002; Dapporto and Dennis Ulrich 2012). In the Western Mediterranean, at least three 2009). Accordingly, the extinction risk for Odonata can be areashavebeenrecognisedaseffectiveanddistinctglacial relatively low and long-term persistence would allow accumu- refugia during Pleistocene glaciations: Iberia, peninsular Italy lation of species. and the Maghreb (Hewitt 1999; Schmitt 2007; Habel et al. Multiple regressions on the mainland revealed that rich- 2011;Husemannetal2013). Islands also should be consid- ness of Odonata in the Western Mediterranean can be ered together with these areas, as recently demonstrated by modeled at the community level on the basis of geographic several studies (Médail and Diadema 2009; Dapporto et al. characteristics, mostly latitude. Conversely, the large dis- 2012; Dapporto and Bruschini 2012). crepancy between predicted and observed richness for Despite the relative ease of predicting species occur- islands is likely due to their isolation, filtering source spe- rences on the mainland, the same models often failed to cies and resulting in impoverished faunas. Actually, our correctly predict occurrences on islands. The existence of analyses confirm that the larger and the less isolated an several SppA was predicted by analysis at the community island, the less impoverished its fauna. Indeed, Sicily did level that revealed highly impoverished assemblages on not harbour an impoverished fauna, while the smallest and islands except Sicily. However, membership of these spe- most isolated Maltese islands had the most impoverished cies on the basis of their ecological traits is not highly assemblage. These results represent a sort of preliminary predictable. Actually, if species that are more widespread test revealing that islands actually host impoverished and tolerant have a higher probability to occur on islands, faunas. stochastic events also must play a fundamental role, and Accordingly, in cluster analyses obtained with non- several species with suitable traits did not colonise islands. partitioned community diversity islands are not clustered with An alternative, but not mutually exclusive, hypothesis is their nearest mainland (Italy for Sicily, Corsica, Sardinia, and that we did not measure some key traits determining island Malta; Iberia for Balearics), but are grouped together in a occupancy. single cluster without showing any indicator species for such Surprisingly, islands (with the sole exception of Malta) a linkage. Actually, when the effect of nestedness/ are not differentially impoverished in Zygoptera and impoverishment is strong, even distant islands and mainland Anisoptera. This result was unexpected since the difficulty areas are linked together unexpectedly, mainly on the basis of to reach an island is a main determinant for impoverish- their species richness (Fattorini 2009; Heiser and Schmitt ment, and the dispersal dichotomy between Zygoptera and 2010). A similar effect was observed for small and isolated Anisoptera affects the filtering of the Straits of Gibraltar river systems and their freshwater fish faunas (Leprieur et al. barrier, the northwards expansion and the influence of wide- 2009). spread elements (Heiser and Schmitt 2010). Furthermore, previous assessments on Lepidoptera suggest that different Individual species level patterns among taxa can be explained easily by differences in dispersal capacities (Dapporto and Dennis 2009, 2010; Impoverishment analyses performed on the taxa themselves Dapporto 2011; Dapporto et al. 2012). revealed a clear distribution pattern of most of them in the The ecological traits assumed for the expected species Western Mediterranean. As it occurred for Lepidoptera being absent (SppA) are conversely observed in species pre- (Dapporto and Dennis 2009, 2010;Dapporto2011), latitude dicted to be absent but being present on islands (SpaP). The and, to a lesser extent, longitude explain the distributions of unexpected presence of a species on an island, assessed on the most species. These results confirm that the Mediterranean basis of mainland occurrence, means that this species is absent area retains highly contrasting faunas among different areas. or highly uncommon in the nearest continental areas. Unex- These patterns are due mainly to the presence of the Alpine pected presence can emerge as a consequence of several region at its northern side and to the presence of the Maghreb reasons, and four main hypotheses can be made: (1) islands in the South. The Alps retain a high biodiversity, also represent the extremes of the geographic distribution for a collecting many temperate species living in Central and species, thus escaping the model prediction; (2) exceptional 212 M. Heiser et al. and stochastic dispersal events may have produced unexpect- belonging to Italy. However, when examined at the individual ed colonisations of islands; (3) isolation on islands may have species level, Sicily has seven SppA and eight SpaP. These produced island endemics or sub-endemics that are absent on species represent a relatively large fraction of the Sicilian the mainland or occur only in reduced areas following retro- fauna composed of 45 species and together may well account dispersal; (4) species with once larger distributions survived for a large turnover with respect to peninsular Italy. Many on islands, but have gone extinct or are considerably reduced SpaP from Sicily link this island with the Maghreb and Iberia, on the mainland, thus producing the observed discrepancy. thus explaining the observed similarity pattern. Additionally, Hypothesis (2) is not supported by our data, since SpaP are the IndVal analysis identified two SpaP species as highly expected to show rather high dispersal capabilities—afactnot representative for the island cluster. In sum, by partitioning confirmed here. Hypothesis (3) is also not supported since no community diversity, the dominant ecological effect of spe- strictly endemic Odonata exist on Western Mediterranean cies impoverishment on communities was removed, thus islands. The only species restricted mostly to islands is allowing to recognize historic information encompassed by Ischnura genei, but it occurs on most circum Italian islands the turnover component. from the Tuscan to the Maltese Archipelago. It is nearly impossible that this species evolved over such a large and Integrating impoverishment analysis and partitioning fragmented area; most likely, this taxon conforms to hypoth- of beta-diversity esis (4); thus island populations may represent remnants of a once more widespread distribution. Several indications affirm that the observed Odonata assem- Separating the effect of hypotheses (1) and (4) is difficult, blages result from a long and unrepeatable sequence of phe- and the former could be the result of the latter. However, when nomena. Such processes operated differently on different a species has a shared distribution on islands and in relatively islands and have been driven, as in other organisms, by (1) distant areas, it could be predicted to have undergone a pro- past climate changes (Hewitt 1999;Schmitt2007;Fattorini cess similar to that described in hypothesis (4). In this respect, 2009), (2) long-term persistence favoured by more stable hypothesis (4) could fit for the subset of SpaP having a climatic conditions over larger areas and isolation from scattered distribution involving a few areas of the Maghreb competing taxa (Emerson et al. 1999, 2000a, b;Masini and of southern Italy as well as species occurring in Iberia and et al. 2008) and, finally, (3) contemporary immigrations the Maghreb. and extinctions involving both deterministic and stochas- tic events (Wiemers 1995; Dennis et al. 2000; Dapporto Partitioning of biodiversity and Dennis 2009;Fattorini2009). The distribution of a species, therefore, could have several explanations and The pattern shown by the Sørensen index revealed a strong the most obvious is not necessarily true (e.g. Emerson dependency on richness as confirmed by a recent review by 2002;Dincă et al. 2011;Dapportoetal.2012 for com- Baselga (2010). Nevertheless, a weak linkage (as revealed by parative analyses). low bootstrap support) of the islands with Iberia and the In this study, we tried to unravel these determinants starting Maghreb obtained by the Sørensen index is confirmed by from the species absent on islands. A high degree of deter- using the turnover information provided by the Simpson minism is encompassed in species occurrence on islands, and index. However, Corsica, clustering with Italy, was excluded most species predicted to occur on islands and actually living from the island group as well as the Balearics clustering with there represent faunal elements with a very large distribution. Iberia. However, these species enclose limited information for overall The exclusion of Corsica from the group of other islands is diversity since they occur on most source areas. On the other not surprising since there is evidence that this island retains a hand, species predicted to occur, but being absent, have a high more similar insect fauna to the European mainland when importance in the nestedness component, but tend to produce compared to Sardinia and Sicily, probably as a consequence a pattern grouping islands mostly on the basis of their physical of stepping stone dispersal from Italy through the Tuscan characteristics (e.g. area and isolation). Such an ecological Archipelago (Dincă et al. 2011; Dapporto et al. 2012). Ac- influence can be reduced by partioning diversity and selecting cordingly, Corsica is not strongly impoverished in its an index linked to species turnover (Baselga 2010, 2012). Zygoptera fauna, indicating that a larger fraction of the species Here, species having restricted distributions on the mainland living in neighbouring areas has been able to establish popu- disclose peculiar colonisation histories of different islands. lations there. They can be both SppP and SpaP. The first should uniform In this respect, it is surprising that Sicily clustered together islands with their nearest mainland areas, the latter should with Sardinia and Malta in the Maghreb-Iberia group. Indeed, create unexpected similarity patterns. Different relative fre- it could be hypothesised that Sicily, due to its large area and quencies of these entities are predicted to be related to more reduced isolation, should host most of the faunal elements “simple” or “complex” colonisation histories. Appendix S1 Impoverishment analysis and partitioning of beta-diversity 213 illustrates that species not being ubiquitous on the mainland Cesaroni, D., Lucarelli, M., Allori, P., Russo, F., & Sbordoni, V. (1994). show different frequencies on islands with respect to SppP or Patterns of evolution and multidimensional systematics in graylings (Lepidoptera: Hipparchia). Biological Journal of the Linnean SpaP membership (Balearics 5:0; Corsica 12:4; Sardinia 7:8; Society, 52,101–119. Sicily 10:8; Malta 3:1). Actually, the first two islands showing Corbet, P. S. (2004). Dragonflies: Behaviour and ecology of Odonata. the highest ratio clustered together with their nearest main- Colchester: Harley. land, while the latter three formed an individual cluster unex- Dapporto, L. (2011). Predicting distribution of Zygaena moths on West Mediterranean islands. Implications for biogeography and conser- pectedly more closely linked to the Maghreb and Iberia. vation (Lepidoptera Zygaenidae). Journal of Insect Conservation, As interpreted for Lepidoptera, such elements determining 15,445–454. unexpected links between Western Mediterranean islands and Dapporto, L., & Bruschini, C. (2012). Invading a refugium: post southern-western mainland areas are probably ancient relics glacial replacement of the ancestral lineage of a Nymphalid butterfly in the West Mediterranean. Organism Diversity and once distributed widely and then replaced by other entities on Evolution, 12,39–49. most of the mainland (Dapporto et al. 2011;Kudrnaetal. Dapporto, L., & Dennis, R. L. H. (2009). Conservation biogeography of 2011; Dapporto and Bruschini 2012). Their occurrence on large Mediterranean islands. Butterfly impoverishment, conserva- “ ” islands thus seems to represent the more or less ephemeral tion priorities and inferences for an ecological island paradigm . Ecography, 32,169–179. remnant of their ancient distribution. When organisms belong- Dapporto, L., Dennis, R.L.H. (2010). Skipper impoverishment on large ing to rather different taxa are analysed in depth, the distribu- West Mediterranean islands: deterministic, historical and stochastic tions of cryptic species and genetic lineages seem to fit with factors. Biodiversity and Conservation, 19: 2637–2649 ă this pattern in many cases (Lepidoptera: Cesaroni et al. 1994; Dapporto, L., Schmitt, T., Vila, R., Scalercio, S., Biermann, H., Dinc ,V., ă et al. (2011). Phylogenetic island disequilibrium: evidence for on- Dapporto et al. 2011, 2012;Dinc et al. 2011; Habel et al. going long-term population dynamics in two Mediterranean butter- 2011;Hymenoptera:Francketal.2000; Reptiles: Giovannotti flies. JournalofBiogeography,38,854–867. et al. 2007; Amphibians: Sara and Vogel 1996; Stöck et al. Dapporto, L., Bruschini, C., Dincă, V., Vila, R., & Dennis, R. L. H. 2008). This suggests that SpaP represent a “special treasure” (2012). Identifying zones of phenetic compression in West Mediterranean butterflies (Satyrinae): refugia, invasion and hybrid- among species occurring on islands, responsible for turnover ization. Diversity and Distributions, 18,1066–1076. and thus for a large part of the biogeographic pattern. On the Dapporto, L., Ramazzotti, M., Fattorini, S., Talavera, G., Vila, R., & other hand, they are also at a higher risk of extinction than Dennis, R. L. H. (2013). Recluster: an unbiased clustering procedure – SppP. Indeed, their absence or limited distribution in the for beta-diversity turnover. Ecography, 36,1070 1075. Dennis, R. L. H., & Shreeve, T. G. (1996). Butterflies on British and Irish nearby mainland make it highly unlikely that, after a possible offshore islands: Ecology and biogeography. Oxford: Gem. extinction event, a population could be re-established by long Dennis, R. L. H., Williams, W. R., & Shreeve, T. G. (1991). A multivar- distance colonisation from source areas far away. As sug- iate approach to the determination of faunal structures among gested by Dapporto (2011), these entities should be regarded European butterfly species (Lepidoptera: Rhopalocera). Zoological Journal of the Linnean Society, 101,1–49. with particular attention in conservation plans, especially as Dennis, R. L. H., Shreeve, T. G., Olivier, A., & Coutsis, J. G. (2000). these island populations might harbour still unknown genetic Contemporary geography dominates butterfly diversity gradients endemisms. within the Aegean archipelago (Lepidoptera: Papilionoidea, Hesperioidea). Journal of Biogeography, 27,1365–1383. Dennis, R. L. H., Hardy, P. B., & Dapporto, L. (2012). Nestedness in Acknowledgement We thank the Deutsche Forschungs Gemeinschaft island faunas: novel insights into island biogeography through but- for a scholarship (M.H.) in the graduate school “Verbesserung von terfly community profiles of colonization ability and migration Normsetzung und Normanwendung im integrierten Umweltschutz durch capacity. Journal of Biogeography, 39,1412–1426. rechts- und naturwissenschaftliche Kooperation”. Diamond, J. M. (1975). Assembly of species communities. In M. L. Cody & J. M. Diamond (Eds.), Ecology and evolution of communities (pp. 342–444). Cambridge: Belknap. Dijkstra,K.D.B.,&Lewington,R.(2006).Field guide to the References dragonflies of Britain and Europe. Milton on Stour: British Wildlife Publishing. Dincă, V., Dapporto, L., & Vila, R. (2011). A combined genetic- Baselga, A. (2010). Partitioning the turnover and nestedness components morphometric analysis unravels the complex biogeographical histo- of beta diversity. Global Ecology and Biogeography, 19,134–143. ry of Polyommatus icarus and Polyommatus celina Common Blue Baselga, A. (2012). The relationship between species replacement, dis- butterflies. Molecular Ecology, 20,3921–3935. similarity derived from nestedness, and nestedness. Global Ecology Dufrêne, M., & Legendre, P. (1997). Species assemblages and indicator and Biogeography, 21,1223–1232. species definition: the need of an asymmetrical and flexible ap- Boudot, J.-P., Kalkman, V. J., Azpilicueta Amorin, M., Bogdanovic, T., proach. Ecological Monographs, 67,345–366. Cordero Rivera, A., Degabriele, G., et al. (2009). Atlas of the Elith, J., & Leathwick, J. (2007). Predicting species distribution from Odonata of the Mediterranean and North Africa. Libellula museum and herbarium records using multiresponse models fitted Supplement, 9,1–256. with multivariate adaptive regression splines. Diversity and Carvalho, J. C., Cardoso, P., & Gomes, P. (2012). Determining the Distributions, 13,265–275. relative roles of species replacement and species richness differences Emerson, B. C. (2002). Evolution on oceanic islands: molecular phylo- in generating beta-diversity patterns. Global Ecology and genetic approaches to understanding pattern and process. Molecular Biogeography, 21,760–771. Ecology, 11,951–966. 214 M. Heiser et al.

Emerson, B. C., & Gillespie, R. G. (2008). Phylogenetic analysis of Koleff, P., Gaston, K. J., & Lennon, J. J. (2003). Measuring beta diversity community assembly and structure over space and time. Trends in for presence-absence data. Journal of Ecology, 72,367–382. Ecology and Evolution, 23,619–630. Kreft, H. & Jetz, W. (2010). A framework for delineating biogeo- Emerson, B. C., Oromí, P., & Hewitt, G. M. (1999). MtDNA graphical regions based on species distributions. Journal of phylogeography and recent intra-island diversification of Canary Biogeography, 37: 2029–2053. Island Calathus beetles (Carabidae). Molecular Phylogenetics and Kudrna, O., Harpke, A., Lux, K., Pennerstorfer, J., Schweiger, O., Settele, Evolution, 13,149–158. J., et al. (2011). Distribution atlas of butterflies in Europe. Halle: Emerson, B. C., Oromí, P., & Hewitt, G. M. (2000a). Colonisation and Gesellschaft für Schmetterlingsschutz. diversification of the species Brachyderes rugatus (Coleoptera) on Leprieur, F., Olden, J. D., Lek, S., & Brosse, S. (2009). Contrasting the Canary Islands: evidence from mtDNA COII gene sequences. patterns and mechanisms of spatial turnover for native and exotic Evolution, 54,911–923. freshwater fish in Europe. Journal of Biogeography, 36, 1899– Emerson, B. C., Oromí, P., & Hewitt, G. M. (2000b). Tracking colonisa- 1912. tion and diversification of insect lineages on islands: MtDNA Lomolino, M. V.,Brown, J. H., & Sax, D. F. (2010). Island biogeography phylogeography of Tarphius canariensis (Coleoptera: Colydiidae) theory: Reticulations and reintegration of a biogeography of the on the Canary Islands. Proceedings of the Royal Society of London species. In J. B. Losos & R. E. Ricklefs (Eds.), The theory of island B, 267,2199–2205. biogeography revisited (pp. 13–51). Princeton: Princeton University Fattorini, S. (2002). Biogeography of the tenebrionid bettles (Coleoptera, Press. Tenebrionidae) on the Aegean Islands (Greece). Journal of Losos, J. B., & Ricklefs, R. E. (2010). The theory of island biogeography Biogeography, 29,49–67. revisited. Princeton: Princeton University Press. Fattorini, S. (2009), Both Recent and Pleistocene geography determine MacArthur, R. H., & Wilson, E. O. (1967). The theory of island bioge- animaldistributional patterns in the Tuscan Archipelago. Journal of ography. Princeton: Princeton University Press. Zoology, 277: 291–301. Masini, F., Petruso, D., Bonfigliom, L., & Mangano, G. (2008). Fattorini, S., & Ulrich, W. (2012). Spatial distributions of European Origination and extinction patterns of mammals in three central Tenebrionidae point to multiple postglacial colonization trajectories. Western Mediterranean islands from the Late Miocene to Biological Journal of the Linnean Society, 105,318–329. Quaternary. Quaternary International, 182,63–79. Fattorini, S., Sciotti, A., Tratzi, P., & Di Giulio, A. (2013). Species Médail, F., & Diadema, K. (2009). Glacial refugia influence plant diver- distribution, ecology, abundance, body size and phylogeny orig- sity patterns in the Mediterranean Basin. Journal of Biogeography, inate interrelated rarity patterns at regional scale. Journal of 36,1333–1345. Zoological Systematics and Evolutionary Research.doi:10. Rosenzweig, M.L. (1995). Species Diversity in Space and Time. 1111/jzs.12026. Cambridge: Cambridge University Press. Franck, P., Garnery, L., Celebrano, G., Solignac, M., & Cornuet, J. M. Santos, A. M. C., Quicke, D. L. J., Borges, P. A. V. & Hortal, J. (2011). (2000). Hybrid origins of honeybees from Italy (Apis mellifera Species pool structure determines the level of generalism of island ligustica) and Sicily (A. m. sicula). Molecular Ecology, 9,907–921. parasitoid faunas. Journal of Biogeography, 38: 1657–1667. Friedman, J. H. (1991). Multivariate adaptive regression splines. Annals Sara, M., & Vogel, P. (1996). Geographic variation of the Greater white- of Statistics, 19,1–141. toothed shrew (Crocidura russula Hermann, 1780 Mammalia, Giovannotti, M., Cerioni, P. N., Kalboussi, M., Aprea, G., & Caputo, V. Soricidae). Zoological Journal of the Linnean Society, 116,377–392. (2007). Phylogeographic inferences from the mtDNAvariation of Schmitt, T. (2007). Molecular biogeography of Europe: Pleistocene cy- the three-toed skink, Chalcides chalcides (Reptilia: Scincidae). cles and postglacial trends. Frontiers in Zoology, 4,11. Journal of Experimental Zoology (Molecular and Developmental Simpson, G. G. (1943). Mammals and the nature of continents. American Evolution), 308B,297–307. JournalofScience,241,1–31. Habel, J. C., Lens, L., Rödder, D., & Schmitt, T. (2011). From Africa to Stöck, M., Sicilia, A., Belfiore, N. M., Buckley, D., Lo Brutto, S., Lo Europe and back: refugia and range shifts cause high genetic differ- Valvo, M., et al. (2008). Post-Messinian evolutionary relationships entiation in the Marbled White butterfly Melanargia galathea. BMC across the Sicilian channel: mitochondrial and nuclear markers link a Evolutionary Biology, 11,215. new green toad from Sicily to African relatives. BMC Evolutionary Hawkins, B. A. (2010). Multiregional comparison of the ecological and Biology, 8,56–74. phylogenetic structure of butterfly species richness gradients. Tennent, J. (1996). The butterflies of Marocco, Algeria and Tunesia. Journal of Biogeography, 37,647–656. Oxford: Gem. Heiser, M., & Schmitt, T. (2010). Do different dispersal capacities influ- Varga, Z., & Schmitt, T. (2008). Types of oreal and oreotundral disjunc- ence the biogeography of the western Palearctic dragonflies tion in the western Palaearctic. Biological Journal of the Linnean (Odonata)? Biological Journal of the Linnean Society, 99,177–195. Society, 93,415–430. Hewitt, G. M. (1999). Post-glacial re-colonization of European biota. Whittaker, R. J., & Fernández-Palacios, J. M. (2007). Island biogeogra- Biological Journal of the Linnean Society, 68,87–112. phy. Ecology, evolution and conservation. Oxford: Oxford Hill, M. O., & Smith, A. J. E. (1976). Principal component analysis of University Press. taxonomic data with multi-state discrete characters. Taxon, 25 ,249–255. Wiemers, M. (1995). The butterflies of the Canary Islands—a Husemann, M., Schmitt, T., Zachos, F. E., Ulrich, W., & Habel, J. C. survey on their distribution, biology and ecology (Lepidoptera: (2013). Palaearctic biogeography revisited: evidence for the exis- Papilionoidea and Hesperioidea). Linneana Belgica, 15,63–84 tence of a North African refugium for Western Palaearctic biota. and 87–118. Journal of Biogeography.doi:10.1111/jbi.12180. Williamson, M. (1981). Island populations. Oxford: Oxford University Illies,J.(1976).Limnofauna Europaea. Jena: Fischer. Press.