Hypothesis 1. Island trait spaces converge across islands
First, we explored the functional diversity patterns of the subterranean
communities in the Canary Islands relating the properties of each trait
space to the age and stage of the geological cycle for each island.
Since the subterranean environment imposes a similar ecological filter
across islands (i.e., permanent darkness and scarcity of food), we
expected a functional convergence of trait spaces across islands (Gibert
& Deharveng 2002; Trontelj et al. 2012; Mammola et al. 2022). Yet, we
expect some degree of polarisation of the trait space according to the
ontogenetic stage—young (La Palma and El Hierro), mature (La Gomera,
Tenerife, and Gran Canaria), and senescent (Fuerteventura and Lanzarote)
islands. We represented the subterranean trait space of each island
using geometrical n -dimensional hypervolumes (Blonder et al.,
2014). Since some of the traits were categorical, we applied a Gower
dissimilarity measure to the complete trait matrix and extracted
orthogonal morphological axes through principal coordinate analysis
(Carvalho & Cardoso 2020). We delineated the hypervolumes using a
gaussian kernel density estimate (Blonder et al., 2014, 2018), the first
four principal coordinate axes (cumulatively 89% variance explained),
and a default bandwidth for each axis as implemented in the functionkernel.build of the R package BAT version 2.7.0 (Cardoso et al.
2015, 2021). We opted for a gaussian kernel density estimation instead
of more frequently used techniques such as functional convex hull
(Mouillot et al. 2021) because we were interested in a probabilistic
characterisation of the trait space, allowing us to identify gaps and
areas of higher trait density (Blonder 2016; Mammola et al. 2021b).
Furthermore, hypervolumes are less sensitive to outliers than other
techniques for functional estimation (Mammola et al. 2021b), an
important property given the wide taxonomic coverage of our dataset. We
expressed the properties of the hypervolume within each island using
functional richness (kernel.alpha function) and regularity
(kernel.evenness ) (Mammola & Cardoso 2020).
We investigated the functional differences among islands compared to
their taxonomic differences using the BAT functions kernel.betaand beta , respectively. We expected the values of beta functional
diversity among islands to be low (around ca. 0.5), compared to the
values of beta taxonomic diversity, which we expected to tend to 1.