Isotopic Niche Structure of Communities
We tested for partitioning of isotope niches by comparing average
δ13C and δ15N values across
populations using linear models with random effects ‘Species’ and
‘Locality’ either on their own, or with the former nested within the
latter (lmerTest package: Kuznetsova et al. 2017). We used the
small sample-corrected Akaike’s Information Criterion, i.e.
AICc (Burnham & Anderson 2001), to compare
goodness-of-fit of each model combination, and Variance Components
Analysis (VCA) to partition proportions of isotopic variation explained
by each random term (as well as the within-population variance, i.e.
model residuals).
Niche breadths, as well as pairwise niche overlaps
(Oij ), were calculated based on ellipse areas in
isotopic bi-space (Fig. 1). Standard ellipse areas, corrected for small
sample sizes (SEAc ), were averaged over
104 iterations of a Bayesian model-fitting procedure
(SIBER: Jackson et al. 2011). We calculated asymmetric values forOij , dividing the area of intersection by the
total area of the focal population (i ), averaged over 100 random
draws of posterior SEAc distributions. Matrices
of all pairwise overlaps are presented in Table S3 of the supplementary
material.
The mean and standard deviation of Oij for each
community (μO and sO ,
respectively) was compared with the expected distribution derived from
null models in which species identity was disregarded in the dataset.
Numerical experiments with one community showed that 100 null model
permutations were sufficient to yield stable distributions ofμO . Deviations between observed and expected
values were used to infer the presence of isotopic niche structure
(μO ,observed <μO ,null in more than 95% of
iterations), and the type of structure if present
(sO ,observed <sO ,null for complete partitioning
of niches, and sO ,observed> null sO ,null for a
clumped distribution; see part 3a of supplementary information).