Marine subsidies mediate patterns in avian island biogeography

The classical theory of island biogeography, which predicts species richness using island area and isolation, has been expanded to include contributions from marine subsidies, i.e. subsidized island biogeography (SIB) theory. We tested the effects of marine subsidies on species diversity and population density on productive temperate islands, evaluating SIB predictions previously untested at comparable scales and subsidy levels. We found that the diversity of terrestrial breeding bird communities on 91 small islands (approx. 0.0001–3 km2) along the Central Coast of British Columbia, Canada were correlated most strongly with island area, but also with marine subsidies. Species richness increased and population density decreased with island area, but isolation had no measurable influence. Species richness was negatively correlated with marine subsidy, measured as forest-edge soil δ15N. Density, however, was higher on islands with higher marine subsidy, and a negative interaction between area and subsidy indicates that this effect is stronger on smaller islands, offering some support for SIB. Our study emphasizes how subsidies from the sea can shape diversity patterns on islands and can even exceed the importance of isolation in determining species richness and densities of terrestrial biota.

Methods Figure 1S Temperate rainforest study islands, surrounded by a dynamic, productive matrix. Photo taken with a small Remotely Piloted Aerial System (sRPAS).

Island selection process:
Islands representative of the biogeographical and geomorphological variation in the region were chosen by using a two-step clustering method in SPSS statistical software (V23, IBM). This analysis considered 5 descriptors per island for the 1470 islands in the region: distance from mainland, area, exposure, normalized (size-independent) perimeter-to-area ratio, and percentage of area occupied by surrounding landmasses within 500 m of the island. For exposure, we used the British Columbia ShoreZone dataset which classifies a unit of shoreline with a given exposure classification from very exposed to very protected. These classifications are based on wave exposure categories derived from wind fetch distances. The analysis used these variables to identify 5 clusters of island types (Table 2S). To facilitate sampling in a remote location, we selected 9 physical groupings of islands with 6-17 islands per group.

Estimating isolation:
To create an isolation metric, we considered that the classical TIB prediction that species richness varies with distance to mainland does not apply to this system because birds are highly mobile, and numerous large islands serve as a functional "mainland" source population. Rather than using the classical "distance to mainland" metric, we predicted the minimum size of an island that acts as a functional "mainland", and then used distance to the nearest island of that size as a predictor for our models.
To do this, we used the 'nls' (non-linear least squares) function in R to fit a non-linear model to the unlogged, rarefied species richness data as a response to the unlogged area, using the Michaelis-Menten function for asymptotic data with a y-intercept of zero. The Michaelis-Menten formula is y = a*x / (x + b) (plus intercept which is 0), where a is the y value at highest rate of increase, and b is the x value at approximately 1/2 of the asymptote. We then extracted the area at which species richness reached 90% of the asymptotic value, and measured distance to the nearest island of that size. We compared these results with those obtained from extracting the area for 95% and 99% of the asymptote, and, although the size of island falling into these categories is widely variable, the number of species found on an island of 90% the asymptote is less than 1 fewer than on an island with area 99% of the asymptote, so we used the distance to the closest island able to hold 90% of the maximum number of observed species as a metric for isolation ( Figure 2S).

Analyses:
In a preliminary analysis, we evaluated the effects of isolation and habitat heterogeneity on bird species richness and population density. In the species richness analysis, there was no difference between the model including both area and isolation, and the model with area alone, even when considering Burnham and Anderson's least stringent cutoff of a difference of <2 ΔAICc units [1]. Considering habitat heterogeneity in the model also proved to be uninformative (ie. the parameter did not improve model fit). All four models (area, area + isolation, area + habitat heterogeneity, and area + isolation + habitat heterogeneity) were better than the null model, which carried zero weight. In the population density model, adding habitat heterogeneity to the area model did not improve model quality, but both the area only model and area and habitat heterogeneity models were better than any models containing the isolation parameter.

Relative Variable Importance (RVI):
To obtain meaningful RVIs, each parameter must occur in an equal number of models, so we considered all subsets of fixed effects and model-averaged across all outcomes to obtain coefficients and associated standard errors using the 'MuMIn' package in R [3]. Testing all possible combinations of parameters is not recommended when trying to determine "significance" or trying to isolate a top model, but is an effective technique to determine RVIs [4,5]. Figure 4S We fit separate global models for insectivore and "other guilds" species richness and total density to determine if any particular feeding guild was dropping out at higher levels of δ15N. We classified species based on the 5 diet categories described in the Elton Traits 1.0 database [2]. The "other" guild included "Omnivore", "FruiNect", "VertFishScav", and "PlantSeed" feeding categories. We combined these other guilds because they were poorly represented overall in our study with just a few species in each. The majority (~2700 out of 3600) of our observations were of insectivorous birds. Area had a strong positive effect on species richness of both invertebrates and other guilds; however, the effect of δ 15 N was only significant for insectivores. Neither area nor δ 15 N had a significant effect on insectivore density, but area had a strong negative effect and δ 15 N a positive albeit highly uncertain effect on the total bird density of other guilds. The interaction between area and δ 15 N was also positive for the density of individuals in other guilds.