1 INTRODUCTION

In 1963, William Niering noted something unusual when he was studying the plant diversity of the Kapingamarangi Atoll: the number of plant species was unrelated to island area for relatively small islands (Niering, 1963). This observation was intriguing because the species–area relationship is usually stated as one of the few general laws in ecology (Preston, 1960; Schoener, 1976; Triantis & Sfenthourakis, 2012), and Niering’s observation seemed a notable exception. The unusual pattern in which species richness varies independently of island area for islands below a certain size, or species richness increases at a different rate in small islands compared to larger ones, was later called the small-island effect (SIE; Heatwole & Levins, 1973). However, the SIE remained largely unexplored by most ecologists for a long time (Lomolino & Weiser, 2001), with only a few studies describing or discussing the phenomenon after it was first recorded (MacArthur & Wilson, 1967; Morrison, 1997; Rusterholz & Howe, 1979; Woodroffe, 1986; see Triantis et al., 2006 for a detailed list). A renewed interest in the SIE emerged when Lomolino (2000) suggested that the SIE may be a ubiquitous phenomenon that reflects the sigmoidal nature of the species–area relationship and provided analytical methods to model the pattern (Lomolino & Weiser, 2001). Since then, many studies have found evidence for a SIE in different archipelagos worldwide and for different taxonomic groups (Gao & Perry, 2016; Morrison, 2014; Qie, Lee, Sodhi, & Lim, 2011; Schrader, Moeljono, Keppel, & Kreft, 2019; Triantis et al., 2006; Wang, Chen, & Millien, 2018; Wang, Millien, & Ding, 2016).

However, few studies have proposed ecological mechanisms to explain why species richness may vary independently of area for small islands. The first two main propositions are related to disturbance or habitat quality. First, the disturbance hypothesis suggests that small islands may be more severely affected by occasional disturbance events in comparison to large islands (Whittaker, 1995), promoting extinction rates on small islands independently of area. Hence, an increase in island size would not necessarily result in an increase in number of species (MacArthur & Wilson, 1967). Second, the habitat hypothesis postulates that islands below a certain size may be ecologically restricted to one homogeneous stressful environment, tolerable to only a few species. As islands become large enough to allow the appearance of additional habitats, the typical relationship between area and species richness should emerge (Triantis, Mylonas, Lika, & Vardinoyannis, 2003; Triantis et al., 2006; Whitehead & Jones, 1969).

A third and relatively recent hypothesis is related to the productivity of the surrounding system, which, for marine islands, is marine productivity. Based on ideas about spatial subsidies proposed by Gary Polis and colleagues (Polis, Anderson, & Holt, 1997; Polis & Hurd, 1996), Anderson and Wait (2001) proposed the subsidized island biogeography hypothesis (SIB), which assumes that allochthonous resources derived from adjacent waters increase insular productivity in a way that may increase or decrease the expected species richness of subsidized islands. Because smaller islands have proportionally more area exposed to the ocean, the impact of marine subsidies should be substantial on small islands, but become negligible as island size increases (Anderson & Wait, 2001).

According to Anderson and Wait (2001), nutrients of marine origin may stimulate plant growth. Initially, the higher subsidized productivity would promote an increase in plant richness, therefore decreasing the slope (z) of the species–area relationship. However, after a certain level of subsidy input, diversity may decrease as productivity increases, and the z for those islands will become steeper (see figure 1 in Anderson & Wait, 2001). That is, the relationship between productivity and diversity may not be linear, but quadratic (Kassen, Buckling, Bell, & Ralney, 2000; Rosenzweig, 1992; Tilman, 1982). Thus, the direction and magnitude of how much residuals deviate from the traditional species–area relationship in small islands may depend on the amount of subsidy provided by adjacent marine productivity.

Marine subsidies are indeed an important component of island food webs (Barrett et al., 2005; Piovia-Scott, Spiller, & Schoener, 2011). Many studies have demonstrated that after entering the system by shore drift or seabird guano, marine-derived resources significantly increase primary productivity (Anderson & Polis, 1999; Sánchez-Piñero & Polis, 2000), as well as population density of primary and secondary consumers (Barrett et al., 2005; Polis & Hurd, 1995, 1996; Sánchez-Piñero & Polis, 2000). However, few studies have tested the SIB hypothesis and evaluated the effect of marine subsidy on insular species richness in the context of the SIE. To the best of our knowledge, only one regional study on lizard richness in the Gulf of California explicitly tested the SIB hypothesis (Barrett, Wait, & Anderson, 2003). It is therefore largely unknown whether marine subsidies contribute to the species richness of islands and might explain the SIE. At a global scale, for example, marine productivity is highly variable (see Figure 1). While in some waters life abounds, others are practically oceanic deserts (Lalli & Parsons, 1997). Given that the flux of subsidy depends on the productivity of the adjacent system (the donor habitat; Polis et al., 1997), some islands should be naturally more subsidized than others. However, no study has yet evaluated the importance of marine primary production for insular species richness considering the wide range of oceanic productivity, nor explored the relative importance of marine subsidy compared to area as island size increases.

In this study, we tested the SIB hypothesis by evaluating the effects of island area and marine productivity on angiosperm species richness of 790 islands distributed worldwide (Figure 1), including a large number of very small islands. Assuming that high marine productivity around the islands increases marine-derived nutrients on the islands, we used marine primary production as a surrogate for potential marine subsidy. Marine primary production is usually measured by estimating the biomass of benthic and pelagic primary producers (i.e., macrophytes and phytoplankton). Regarding the first one, previous studies have already demonstrated how environmental factors associated with macrophyte biomass production determine the amount of wrack deposited ashore (Liebowitz et al., 2016; Reimer, Hacker, Menge, & Ruggiero, 2018). However, the spatial distribution of these macrophytes is limited to shallow waters because of the rapid attenuation of light with depth, making phytoplankton the dominant primary producers in the ocean (Lalli & Parsons, 1997). Regions with high phytoplankton productivity are positively associated with high seabird density (Grecian et al., 2016; Grémillet et al., 2008; Sydeman et al., 2010), which act as an active subsidy vector. By using the sea to feed and marine islands to roost and nest, seabirds provide nutrients and organic matter to insular communities via guano and carrion (Polis, Sánchez-Piñero, Stapp, Anderson, & Rose, 2004; Sánchez-Piñero & Polis, 2000), potentially affecting plant species richness. For this reason, we used here phytoplankton biomass estimation to evaluate the effect of marine primary production on plant diversity.

2 METHODS

2.2 Data analyses

In order to test the SIB hypothesis, we first used multiple regression and variation partitioning to evaluate the linear influence of island area and marine NPP on the variation in angiosperm species richness of small islands. However, determining the size under which an island can be considered small is not straightforward. Thresholds to delimit the upper limit of the SIE and, consequently, the set of islands considered small, are highly variable in the literature (Anderson & Wait, 2001; Lomolino & Weiser, 2001; Matthews, Steinbauer, Tzirkalli, Triantis, & Whittaker, 2014). Likewise, the estimated threshold may vary greatly for the same dataset according to the function applied (e.g., left horizontal and two-slope breakpoint models; see Supporting Information Appendix S3). Another possibility could be to explore the interaction term between area and marine NPP. However, the interaction term was highly correlated with area (r = .89; Supporting Information Appendix S4), indicating that collinearity could influence the analysis. To overcome this problem, we conducted the variance partitioning analysis in an iterative manner, instead of using an arbitrary threshold (e.g., 3 km²). We selected the 40 smallest islands in our dataset (all smaller than 10^–4 km²) to start the variance partitioning. We then applied a sequential inclusion of islands, from the smallest to the largest (following Triantis et al., 2006), estimating at each step the variance component explained by each predictor variable. With this method, we could observe the estimated contribution of island area and marine NPP to explaining insular species richness with increasing island size. As the ‘hump-shaped’ productivity–diversity relationship is also commonly found for plants (Mittelbach et al., 2001), we also used our iterative procedure to test the hypothesis that species richness in small islands is unimodally related to marine subsidies (Anderson & Wait, 2001). However, here we used a quadratic regression to explore the relationship between marine NPP and the residuals of the fitted species–area relationship, modelling marine NPP as a quadratic function. In both analyses we expected that, as the average island size increases, the effect of subsidies becomes less important, so that marine NPP becomes a poor predictor of species richness (Figure 2). The successive inclusion of data not only increases the mean size of the islands but also the number of islands (sample size) at each step. To evaluate the consistency of results, we therefore also conducted the same procedure in the reverse order, from the largest to the smallest.

Another way to evaluate a potential quadratic relationship between productivity and diversity is to explore the effects of marine productivity on small island species richness at a regional scale. For example, if factors such as island type (e.g., continental, volcanic or coral islands) or macroclimatic conditions determine island in situ productivity, then species richness of different archipelagos could show opposing responses to the amount of marine resources input (Anderson & Wait, 2001). In this case, the subsidy effects at a global scale should not be linear but spatially heterogeneous (i.e., non-stationary). Therefore, we assessed the spatial heterogeneity in the relative contribution of area and marine NPP to insular species richness using variation partitioning integrated into geographically weighted regression (GWR; Fotheringham, Brunsdon, & Charlton, 2002). As in a two-dimensional spatial moving window, in GWR one defines the radius of a circle (bandwidth) around a focal point, and performs a regression using the subset of the sampling data that lies within the region, assigning weights to individual samples as a function of geographic distance to the focal point. The analysis then proceeds iteratively by moving the focal point to each sample (here island) in the dataset. We defined a region with a radius of 750 km around each island (our focal point) based on the mean of the maximum distance among islands within the archipelagos in our dataset (see the spatial distribution of the archipelagos in Supporting Information Appendix S1), assigning equal weight to all islands within the radius. We performed the explained variance partitioning for each regional regression, and mapped the components across the islands (Gouveia, Hortal, Cassemiro, Rangel, & Diniz-Filho, 2013).

Additionally, other island characteristics besides area, such as climate and isolation, are known to contribute to insular species richness (Kreft, Jetz, Mutke, Kier, & Barthlott, 2008), or may affect the subsidy dynamic, such as shape of the island and human population density (Ellis, 2005; Polis et al., 1997). To evaluate if the regional variation in these characteristics could increase the influence of marine NPP on species richness, we fitted a generalized linear mixed model (GLMM) with island area, distance to the nearest mainland (Weigelt & Kreft, 2013), connection to the mainland during the Last Glacial Maximum (Weigelt, Jetz, & Kreft, 2013), temperature, precipitation, shape of the island, human population density and marine NPP as the explanatory fixed variables and archipelago as the random variable, to account for the spatial organization of islands in archipelagos with distinct biogeographic histories (Bunnefeld & Phillimore, 2012). Distance to the coast and connection to the mainland during the Last Glacial Maximum were retrieved from the GIFT database (Weigelt et al., 2019). Mean annual temperature and annual precipitation were obtained from modern climate models (CSSM - Cloud Scene Simulation Model) retrieved from the ecoClimate database (Lima-Ribeiro et al., 2015). Shape of the island was calculated using the shape index (Patton, 1975), which is a good proxy for possible edge effects. Human population density was calculated based on the most recent census or demographic estimates publicly available online (mostly from 2010 and 2015; Supporting Information Appendix S5). Variation in adjacent marine productivity within archipelagos ranged from 0 to 1,367 mg C/m²/day (mean 173 mg C/m²/day).

To account for the dependence of the biological subsidy vector, we repeated all the above analyses considering only those islands with seabirds present (n = 377), based on more than 60 million records of seabird occurrence from the Global Biodiversity Information Facility (http://www.gbif.org; see details in Supporting Information Appendix S6). Lastly, we used estimates of ammonia emission from seabird colonies to evaluate directly the effects of marine-derived resources on angiosperm species richness. Ammonia emission was estimated based on the mass of nitrogen excreted by seabirds using an extensive inventory including 2,898 seabird colonies in conjunction with a bioenergetics model (Riddick et al., 2012a, 2012b). This estimation was calculated using population size and bird specific data such as adult mass, length of the breeding season, time spent at the colony during breeding, nitrogen content of the food and so forth. Therefore, it represents a good approximation of the transfer of nutrients from the marine environment to seabird colonies, where they may be used for plant growth. As most colonies for which data were available are present on continents or in small rocky-satellite islands, we could use data from only 136 colonies corresponding to 67 islands in our dataset. Estimates from multiple colonies in the same island were summed. We removed from this subset the Corsica island (8,700 km²), which is an outlier that disturbs the semi-log species–area relationship.

All analyses were conducted using two different transformations of species richness (S) and area (A): semi-log (S versus logA) and log–log (logS + 1 versus logA). Although the SIB hypothesis had been idealized considering a log–log transformation (Anderson & Wait, 2001), some studies have found evidence for the SIE using a semi-log function (Morrison, 2014; Wang et al., 2016). Therefore, if marine subsidy has any influence on species richness of small islands, our chances of detecting the pattern were not restricted to the kind of data transformation. Note that we used log(S + 1) to include those islands with S = 0. However, we also repeated the main analyses excluding all empty islands (i.e., without angiosperms; n = 740) and considering only those islands where both angiosperms and seabirds were present (n = 367). Importantly, preliminary tests with these subsets showed that using log(S) or log(S + 1) did not alter the results. All analyses were conducted in R version 3.5.0 (R Core Team, 2018) using the packages vegan (Oksanen et al., 2018), nlme (Pinheiro, Bates, DebRoy, & Sarkar, 2018), raster (Hijmans, 2018) and geosphere (Hijmans, 2017).

3 RESULTS

The relative importance of island area and adjacent marine productivity for angiosperm species richness, as measured by our iterative variation partitioning, varied widely with island size. For both data transformations (semi-log and log–log), the fraction of variance in angiosperm species richness explained exclusively by area was low and highly variable for very small islands (below 20% for islands smaller than 0.004 km²) but increased gradually with the addition of larger islands (maximum of 34 and 58% for semi-log and log–log transformations, respectively; Figure 3). Conversely, the fraction explained exclusively by marine NPP was generally low but highest for small islands (maximum 6%) and approached zero after inclusion of islands larger than 0.004 km². Note, however, that the fraction explained exclusively by marine NPP was always lower than the fraction explained by area (Figure 3). A similar pattern emerged when we repeated the same procedure in reverse order, especially for the log–log transformed data; the fraction explained by area was also small and highly variable when we started the analysis with few large islands. Likewise, the fraction explained exclusively by marine NPP was highest for the largest islands and decreased rapidly with the inclusion of other islands (Figure 3). We also found similar results when investigating the effect of the quadratic term of marine NPP on the residuals of the species–area relationship. Only a small fraction of the variation in species richness was explained by marine NPP, and this fraction was observed only when the number, not the size, of islands was small (Supporting Information Appendix S7).

Variation partitioning based on GWR provided a spatially explicit visualization of the relative contribution of island area and marine productivity for angiosperm species richness at a regional scale (Figure 4). Maps show that the fraction explained exclusively by area was consistently high across the globe, varying between 40 and 90%. In contrast, the fraction explained exclusively by marine NPP was consistently low, generally below 10%. Variability in the relative contribution of both area and marine NPP seemed to increase when the number of islands in the region, and consequently in the regression, was reduced (Figure 4; for semi-log results see Supporting Information Appendix S7). The GLMM results considering the variables associated with climate, isolation and human population also showed no significant effect of marine NPP (Supporting Information Appendix S8).

Results were qualitatively similar when we repeated the analyses considering only the data from those islands with confirmed presence of seabirds (Supporting Information Appendix S6), islands with at least one angiosperm species (Supporting Information Appendix S9) or islands with at least one angiosperm species and presence of seabirds (Supporting Information Appendix S10). The only change occurred when we evaluated directly the effects of guano deposition through ammonia estimation. We found that ammonia explained a large fraction of the angiosperm species richness variation for both semi-log and log–log transformations (semi-log: R²_(logA) = .27; R²_{adj (logA + logNH3)} = .52 | log–log: R²_(logA) = .30; R²_{adj (logA + logNH3)} = .60), contributing to an overall decrease in plant diversity (semi-log: r = −.46; log–log: r = −.49). However, the influence of ammonia on species richness was not restricted to small islands (Figure 5).

4 DISCUSSION

The subsidized island biogeography hypothesis, one of the few mechanisms proposed to explain why species richness on small islands may vary independently of area, predicts that resources derived from highly productive surrounding systems (i.e., the subsidies) should elevate insular productivity and thereby alter the expected species–area relationship (Anderson & Wait, 2001; Barrett et al., 2003). We evaluate this hypothesis by analysing angiosperm diversity of 790 islands worldwide. Our analyses involved data for 420 islands with an area <1 km², surrounded by waters varying in mean productivity by two orders of magnitude from 162 to 3,547 mg C/m²/day. Contrary to expectations, we did not find support for the hypothesis that adjacent marine productivity may influence insular plant species richness.

Previous estimates show that marine resources may indeed dominate the productivity of islands smaller than 1 km², and their relative importance tends to decrease as island size increases (Polis & Hurd, 1996). However, the subsidy input as estimated by donor productivity does not seem to have any effect on the number of species per island. Our results show that the variation in species richness explained by adjacent marine productivity was always very low (mostly near zero), and never higher than the variation explained by area. As in previous studies of island floras worldwide, area emerged as the strongest determinant of insular diversity (Kreft et al., 2008; Weigelt & Kreft, 2013). At first, our results suggest that marine productivity might have a small effect on species richness for very small islands, for which variation in species richness explained by area is low. However, our sensitivity analyses indicate that this result may be an effect of low sampling size. Alternatively, the poor predictive ability of NPP could result from the nonlinearity of the productivity–diversity relationship. For example, while high productivity could lead to higher population densities and lower extinction rates, potentially increasing diversity, a higher productivity might eventually lead to stronger dominance of a few species that outcompete weak competitors, decreasing diversity (Anderson & Wait, 2001; Mittelbach et al., 2001; Tilman, 1982). However, even assuming a quadratic relationship between productivity and diversity, marine NPP was not a good predictor as would be expected if marine subsidies were affecting small islands’ species richness.

Despite the large variation in marine productivity around the islands analysed here, we did not evaluate the total productivity of islands, that is, the allochthonous plus the autochthonous resources (which in this case should be the marine-derived nutrients plus the remineralization of in situ production). Assuming that on small islands the total productivity is dominated by marine-derived resources (Polis & Hurd, 1996), variation in autochthonous productivity should have minimal effect in our analyses. Nevertheless, to ensure that the potential spatial variation in island productivity among regions (difference in temperature and precipitation, for example; Wright, 1983) did not confound the interpretation of our global results, we applied methods that account for non-stationarity in the explored relationships. For example, following the SIB hypothesis, an increase in adjacent marine production could have a positive effect on small island species richness in regions where islands have low in situ productivity. Conversely, in regions where island productivity is high, an increase in marine resources could decrease insular diversity (Anderson & Wait, 2001; Mittelbach et al., 2001). However, even considering the possibility of non-stationarity in the process, the fraction of variation in insular species richness explained by marine productivity was very low. The relative importance of marine productivity only increased slightly when there were too few islands in the region, suggesting, once again, a statistical artefact. Accordingly, our GLMM results considering random variation among archipelagos showed no significant effect of marine NPP on insular species richness.

Many different strategies have been used for detecting the SIE, such as different transformations of area and species richness (Burns, McHardy, & Pledger, 2009; Gao & Perry, 2016; Morrison, 2014; Wang et al., 2016), different mathematical models (Dengler, 2010; Lomolino & Weiser, 2001) and inclusion or exclusion of empty islands (Dengler, 2010; Wang et al., 2016, 2015). Although evaluating the SIE was not our primary goal, we also used these methods here to explore its presence in our dataset. We found support for a SIE only when using the semi-log transformation (Supporting Information Appendix S3). Following recent SIE studies that use only the semi-log function because logarithmic transformation could bias the estimation of residuals when there are islands with zero species in the dataset (Wang et al., 2016, 2015), we can assume that the islands in our dataset present a clear SIE. However, one could also argue that the better fit of a model with no breakpoint in the log–log function indicates the opposite (see an interesting discussion about this topic in Dengler, 2010). Even assuming no SIE was present in our dataset, we believe that this issue is only of minor importance to our study. The detection of the SIE would indeed be important if marine productivity was affecting species richness of small islands. In this case we could evaluate the relationship of the detected process with the existence of the expected pattern. However, the absence of any clear effect of marine NPP on small islands’ species richness represents a lack of support for the basic mechanism underlying the SIB hypothesis, at least when considering only the estimated productivity of the donor habitat. A next logical step, therefore, is to evaluate possible variations in subsidy transport.

Our analyses also included the presence of a subsidy vector, which, although not related to the SIE, is intimately associated with subsidy dynamics (Polis et al., 1997). In island environments, marine resources may enter the system by shore drift or via seabird transport (Barrett et al., 2005; Polis & Hurd, 1995, 1996). For plants, specifically, seabirds are a key subsidy vector. Previous studies have shown that guano of seabirds, in fact, determines differences in plant growth and vegetal cover among islands (Anderson & Polis, 1999; Sánchez-Piñero & Polis, 2000). However, we also did not find evidence for marine productivity influencing angiosperm species richness when we restricted our analyses to those islands where seabirds are present. On the other hand, the amount of nutrient excreted by seabirds, estimated from ammonia emissions in breeding colonies, was a good predictor of species diversity, negatively affecting the number of species. This result is in accordance with previous studies carried out in the Gulf of California, which showed that although intense guano deposition stimulates plants biomass (Anderson & Polis, 1999; Polis et al., 1997; Sánchez-Piñero & Polis, 2000; Wait, Aubrey, & Anderson, 2005), islands with seabird colonies have a plant species richness 2.5 times lower than those where colonies are absent (Wait et al., 2005), supporting the predicted descending part of the hump-shaped productivity–diversity relationship. Interestingly, estimates of ammonia emissions were only weakly correlated with adjacent marine productivity (r = .08). Despite evidence of high seabird density in highly productive waters (Grecian et al., 2016; Grémillet et al., 2008; Sydeman et al., 2010), nutrient excretion may depend on other factors than just seabird density, such as body mass and length of the breeding season (Otero, De La Peña-Lastra, Pérez-Alberti, Ferreira, & Huerta-Diaz, 2018; Riddick et al., 2012a). These results suggest that although marine subsidies may indeed be a good predictor of angiosperm diversity on islands, their effect is more dependent on the subsidy transport than just variation in adjacent productivity, as already demonstrated by local-scale studies (Domingos & Lana, 2017; Marczak, Thompson, & Richardson, 2007). Even so, evidence is still lacking that marine subsidies may generate the SIE as ammonia effects were not restricted to smaller islands.

Overall, our study demonstrates that the variation in species richness on small islands is not strongly related to adjacent marine productivity, at least not for plants. Although marine subsidies may indeed influence insular species richness via guano deposition, our results suggest that this influence is independent of marine productivity and it does not seem restricted to small islands, therefore not supporting the SIB hypothesis. Future studies aiming to evaluate the effectiveness of the SIB hypothesis for plants should include nutrient estimation from more islands. Our analysis evaluating ammonia effects involved only a few relatively large islands, and the addition of data from smaller islands may provide further information to more rigorously test this hypothesis. Similarly, future studies could evaluate the importance of shore drift (a passive subsidy vector) to plant species richness. Although wrack deposition may potentially contribute to plant growth rate (Spiller et al., 2010), its effect on plant species richness and the relative contribution of this subsidy type compared to seabird guano remains largely unknown. Assuming that wrack deposition is determined mainly by donor productivity (Liebowitz et al., 2016; Reimer et al., 2018), we suspect that the results should be similar to those reported here given the positive correlation between macroalgae biomass and chlorophyll a estimates at regional scale, especially in coastal upwelling waters (Broitman & Kinlan, 2006). However, as passive subsidies may also be highly dependent on transport and permeability features such as topography, orientation of the shoreline and nearshore hydrodynamics (Barreiro, Gómez, Lastra, López, & de la Huz, 2011; Orr, Zimmer, Jelinski, & Mews, 2005), using wrack deposition estimation may be more appropriate. Such information could certainly improve our knowledge on the importance of marine subsidies for island species richness and reveal whether there are any conditions under which the SIB hypothesis may be supported.

Biosketch

André Menegotto is currently a postdoctoral fellow at the Federal University of Goiás. He has a broad interest in macroecology and marine biology, with special focus on understanding the factors that influence species distribution and determine global patterns of biodiversity.