Empirical findings from fitting the ATT2 models
The foregoing analyses demonstrate that the ATT2 model provides a generally good fit with data from a range of plant and animal taxa from five oceanic island archipelagos, both for numbers of native species (SR) and for metrics more directly indicative of evolutionary dynamics (nSIE, pSIE, DI). The effectiveness of the ATT2 model in fitting data for particular archipelagos is expected to depend on the extent to which the archipelago provides a full range of island developmental stages. So, for example, for archipelagos providing only young islands, it would be consistent with the GDM for a simpler ‘area + time’ model to provide a better fit than the full ATT2 model. However, across the data sets evaluated herein, comparison with the alternative models provides confirmation that the ATT2 model, although not the simplest model (and not necessary in all cases), has greater generality than the traditional diversity–area models, or time-only models.
There are, as outlined earlier, several important limitations to these tests: first, the effective ages of the islands cannot be determined with certainty, given the complexities of island histories and the limitations of the dating methods; second, the quality of the taxonomic and distributional data suffers from the same limitations as all such data sets (in particular, see Cowie, 1995); and third, the small number of islands per archipelago means that we are close to the threshold at which the fit of a three-parameter model in any particular case might be deemed trivial. We have minimized the taxonomic problem by ignoring subspecies designations and counting only endemic species in the DI and SIE metrics. We have also addressed the dating and survey quality problem by running a number of alternative models (above, Table S3). Overall, we take encouragement from the fact that the ATT2 models appear able to capture a significant proportion of the variance in the data from these rather different archipelagos, and that the general form of the relationships is, as predicted by the GDM (Table 2, for example predictions 1, 6, and 7), a linear relationship with log(Area), combined with a humped relationship with island age.
General evaluation of the general dynamic model
Consideration of the literature for the study systems we have examined herein provides further support for the GDM. For example, Silvertown (2004) notes that large endemic taxa within the Canarian endemic flora are typically monophyletic (e.g. 63 species of Crassulaceae, and 37 species of Echium); that is, they typically derive from single colonization events. Silvertown suggests that this may be indicative of the operation of niche pre-emption by early colonizing lineages, which, once having colonized, may have both inhibited the success of later-arriving mainland relatives and spread out across the archipelago as new islands formed, frequently radiating into new habitats. This interpretation has been the subject of some debate (Saunders & Gibson, 2005; Silvertown et al., 2005), but has been supported by further work (Carine et al., 2004) and is consistent with the GDM, and particularly the prediction of most rapid lineage radiation occurring on relatively young islands. An analogous mechanism is tentatively supported by Cowie (1995) in his analysis of Hawaiian land snails.
Phylogenetic analyses also provide support for the notion that younger islands are more active arenas for speciation. On the Hawaiian Islands, the estimated speciation rate for plants is a negative function of island age, varying from 0.20 species per lineage per million years on Kauai (island age: 5.7 Myr) to 2.1 species per lineage per million years on Hawaii (island age: 0.5 Myr) (Levin, 2004). Perhaps the best-resolved example of this within a taxon is that of the Hawaiian silverswords, for which Levin (2004) argues that the majority of the speciation events occurred early in the history of the group when the flora was not saturated (for phylogenetic analyses, see Baldwin & Robichaux, 1995; Baldwin & Sanderson, 1998). Studies from the flora of the Juan Fernández Islands also support the idea of high initial rates of radiation, with faster rates evident on the younger island: 0.33 species per lineage per million years on Mastierra (island age: 4 Myr) vs. 0.96 species per lineage per million years on Masafuera (island age: 1–2 Myr) (Levin, 2000; cited in Levin, 2004). Electrophoretic data for the endemic genus Robinsonia (Asteraceae) suggest that the founding population arrived early in the 4.0-million-year history of Masatierra Island, radiating and speciating rapidly after colonization (Crawford et al., 1992). Similarly, Kaneshiro et al. (1995, p. 71), in their analysis of species groups within the picture-wing Drosophila of Hawaii, make the following observation: ‘Most of these species, like many other extant terrestrial endemic fauna, show a very strong but by no means exclusive tendency to single-island endemism. Most species thus appear to evolve on an island early in its history and thereafter remain confined to that island…’.
Evidence for the loss of species as a result of island erosion and subsidence (prediction 5, Table 2) can be invoked from first principles but is likely to be extremely hard to demonstrate unequivocally as the likelihood of finding fossil evidence is negligible in such circumstances. However, there are numerous cases in which island phylogenies point to the existence of earlier forms on islands (or in upland habitats) that no longer exist (e.g. Wagner & Funk, 1995; Keast & Miller, 1996; Butaud et al., 2005; Pulvers & Colgan, 2007). Similarly, Emerson & Oromí (2005), in their discussion of the distribution of Canarian Tarphius, explain the absence of these beetles from the two oldest Canarian Islands (the eastern islands of Fuerteventura, and Lanzarote) as a result of the loss of their cloud forest habitat, prior to which presumably Tarphius did occur. Arnedo et al. (2000) use similar logic to explain a hump-shaped relationship between island age and number of endemic species of the aranid genus Dysdera in the Canaries, arguing that erosion of the mountains on these islands to below 800 m prevents the formation of the trade-wind inversion that supplies moisture to the mesic habitats found on the younger islands: ‘The low number of endemics in the eastern islands could therefore be explained by extinction mainly related to the major environmental change that took place on these islands. The distribution of the eastern endemic specimens seems to support this hypothesis. Most of the specimens were collected from sites located on the northern slopes of massifs over 400 m high. These places represent the wettest parts of these islands’ (Arnedo et al., 2000, p. 289).
Although there is therefore some support for increased rates of speciation and adaptive radiation on younger islands, these findings may at least in part be the outcome of the effects of erosion and subsidence on older islands reducing the persistence of neo-endemic lineages within these islands (Stuessy, 2007). As Peck et al. (1999, p. 535) write in relation to Hawaiian insect taxa: ‘Thus, it appears that the younger main islands display higher species/genus ratios, not just as a result of increased rates of speciation or explosive radiation but also because their varied habitats and relaxed competition allow for continued speciation as well as the retention of the products of such speciation.’ [our emphasis]. It is intrinsically difficult to obtain evidence of changes in rates of the vital processes (migration/immigration, extinction, and speciation) through time and in relation to other island attributes (spacing, overall archipelago isolation, Quaternary climate change), and this is especially the case for the biotas of remote oceanic islands, many of which can be attributed to natural colonization rates expressed in units of time of thousands of years (e.g. Wagner & Funk, 1995; Peck et al., 1999). Similarly, attributing evolutionary outcomes to ‘non-adaptive’ vs. ‘adaptive’ process is challenging (but see Price & Wagner, 2004). Claims for non-adaptive radiation have been most convincingly made for land snails on deeply dissected oceanic islands such as Porto Santo (Madeira) (Cameron et al., 1996), although, for example, Barrett (1996) has suggested that the term may be applied to particular plant lineages within Aegean islands.
The three indices of evolutionary dynamics that we present in Tables 4 and 5 are each based (wholly or partly) on the number of single-island endemic species, and it might be argued that they therefore only capture a small part of the outcome of evolutionary dynamics ongoing not only across whole oceanic island archipelagos such as the Galápagos, but also across island regions such as Macaronesia (comprising the Canaries, Azores, Madeira, Salvage Islands, Cape Verde Islands, and source areas in Africa and southern Europe). However, we follow other recent authors (e.g. Peck et al., 1999; Emerson & Kolm, 2005a) in adopting the rationale that SIE data are likely to be indicative of trends and patterns in other metrics of evolutionary dynamics. For example, tallies of data for the overall number of Canarian endemic plants across the seven main islands of the archipelago (reproduced in Whittaker & Fernández-Palacios, 2007) show that, at least in this case, the pattern for the number of Canarian endemics is strongly correlated with the nSIE and again shows a humped relationship with island age. Moreover, those endemics found on two islands (only) are all shared by island pairs in close proximity to one another, such that species are shared between Lanzarote and Fuerteventura; Gran Canaria and Tenerife; Tenerife and La Palma; Tenerife and La Gomera; and each pairing of La Gomera, La Palma and El Hierro. These data are consistent with the idea that, once a species has formed, it may in time colonize another island within the archipelago, with the probability of colonization being an inverse function of distance. In the Canarian archipelago, this pattern of shared endemism between adjacent islands also largely conforms to the developmental history of the archipelago (reviewed in Whittaker & Fernández-Palacios, 2007), as does the pattern in Hawaii (e.g. Wagner & Funk, 1995). However, not all patterns of shared endemics necessarily relate to a pattern of speciation followed by over-water colonization. In the case of Lanzarote and Fuerteventura, the islands were fused prior to the Holocene transgression, and so it cannot be determined where within the two islands the approximately 17 plant species currently shared only by them originated. Indeed, more generally, we have to allow that species may have occurred in the past on one or more other islands from which they have subsequently gone extinct by natural (e.g. Butaud et al., 2005) or indeed by anthropogenic (e.g. Paulay, 1994; Borges & Brown, 1999; Whittaker & Fernández-Palacios, 2007; Blondel, 2008) drivers.
Phylogeographic analyses of island lineages provide further lines of evidence for the pattern of movement and evolution across archipelagos. One commonly supported pattern involves taxa showing a pattern of dispersion from older to younger islands within an archipelago, with speciation occurring on newly colonized islands. This progression-rule pattern (Funk & Wagner, 1995) is particularly evident in archipelagos showing a clear linear age sequence of islands. Examples drawn from the many that provide support for this rule include: from Hawaii, Drosophila, Hesperomannia, Hibiscadelphus, Kokia, Remya, Tetragnatha (Funk & Wagner, 1995; Gillespie, 2004; Pons & Gillespie, 2004), and Orsonwelles spiders (Hormiga et al., 2003); from Macaronesia, Olea, Gallotia, Hegeter, Gonopteryx (cited in Whittaker & Fernández-Palacios, 2007), Pimellia and Hegeter (cited in Gillespie & Roderick, 2002) and possibly Dysdera (Arnedo et al., 2000); from Galápagos, scarabs and weevils (cited in Gillespie & Roderick, 2002); and from the Austral Islands, Misumenops rapaensis (Garb & Gillespie, 2006). Various other patterns (or no resolvable pattern) have been detected from these and other oceanic archipelagos. In some cases, for example Galápagos birds, evolutionary scenarios involve multiple phases of island-hopping and of alternating periods of allopatry and sympatry within a single radiation (Lack, 1947; Grant & Grant, 1996). In contradiction to the progression rule, data for some taxa are most parsimoniously explained by a sequence of colonization from a young to an older island. For example, mtDNA data suggest that the Canarian blue tit first colonized Tenerife (in the middle of the island-age gradient), and that this was followed by subsequent spread both to older and to younger islands (Kvist et al., 2005); and for Canarian plants, see Sanmartín et al. (2008). It should therefore be understood that the progression rule is merely a general tendency (as Funk & Wagner, 1995; Gillespie & Roderick, 2002), but it is one that based on the GDM should be expected to be a dominant pattern, followed by many taxa in archipelagos that show a strong island age sequence, and especially so in taxa that happen to colonize early in the developmental history of an archipelago, but that also exhibit sufficient dispersal limitation to speciate within the islands of that archipelago.
We are under no illusions that the GDM, as described herein, provides a complete model of island biogeography and evolution. Modification will be necessary for those classes of island that conform poorly to the ontogenetic model proposed here, including many island arc archipelagos and islands of continental origins showing complex histories of horizontal and vertical movement, erosion and re-building (e.g. Buskirk, 1985; Keast & Miller, 1996; McDowall, 2008). For those oceanic islands that do conform to the simple ontogenetic model, perhaps one of the most important omissions from the framework is the role of Quaternary climate change and the accompanying variation in the configuration of islands (e.g. Williams, 1972; Peck, 1990; Nunn, 1994; Carine, 2005; Whittaker & Fernández-Palacios, 2007; Ávila et al., 2008). This not only altered the number, area and elevational range of islands in these archipelagos, but also their relationship with source pools. For instance, Carine (2005) argues that the evolutionary pattern in Macaronesian Convolvulus is suggestive of there having been discrete waves of colonization of this island region, which he explains through a mechanism he labels the ‘colonization window’ hypothesis. This postulates that colonization opportunities vary through time in relation both to the geo-tectonic mechanisms discussed herein (island formation, island sterilization/disturbance) and to periods of climate change. Linked to the latter, low sea-level stands during the Pleistocene saw the emergence of stepping-stone islands, aiding dispersal among the more persistent islands of Macaronesia, and between them and the mainland (Whittaker & Fernández-Palacios, 2007). Similar arguments have been invoked elsewhere, and the notions that dispersal distances and directionality of dispersal related to major current systems can change through time provide additional components that require integration into a comprehensive general theory of oceanic island biogeography (Cook & Crisp, 2005; Cowie & Holland, 2006).
Finally, no consideration of oceanic island biogeography is complete without some reference to the impact of humans on the biodiversity of oceanic island archipelagos (e.g. Gillespie, 2007; Whittaker & Fernández-Palacios, 2007). In the historical period of island exploration (roughly the last 400 years), approximately 60% of recorded extinctions have been of island species, with Pacific island birds particularly badly depleted (e.g. Steadman, 2006). The application of the ideas discussed herein to particular oceanic islands and taxa is therefore subject to the potential disruption of biogeographical patterns following from the impact of anthropogenic habitat conversion, introductions, and extinctions.