Geometric morphometric analysis
To compare the uniqueness or generality of functional traits of resident temperate and range-expanding tropical fish species, we modelled niche similarity based on body shape analysis of theoretically competing fishes within the same trophic guild. Body shape metrics can be used as a proxy for ecological niche in fishes given the significance of morphological traits in determining a species’ relative ecological niche breadth (Azzurro et al. 2014; Smith et al. 2016). The morphospace (2D relative trait space encompassing all species within the community) of each fish was modelled using an image of the left side of an adult individual (images sourced from (Allen 1999; Froese & Pauly 2000)). For each image, 27 points of ecological significance were defined following (Farré et al. 2013) then scaled and quantified using TPS DIG2 v. 2.31 (Rohlf 2004) following (Azzurro et al.2014). Relative warp analysis of images was conducted using tspRELW v.1.70 (Rohlf 2015) and coordinates generated from the first 2 warps (accounting for 52% variability) were used to plot relative community trait structure in two-dimensional space. Voronoi polygons were generated using R package ‘deldir’ Version 0.1-25 (Turner 2020) to define the theoretical niche space occupied by a species (i.e. the maximum polygon size around the species centroid). Voronoi area was retained as a proxy for relative niche space and used as a way of determining the potential for competitive interactions between neighbouring species, with larger Voronoi areas representing high niche dissimilarity (Du et al. 2012).
As a metric of morphological similarity between each species and the overall community mean, we generated a variable termed ‘distance’, calculated as the Gower distance between each point and the community weighted mean centroid (CWM). CWM was determined using the package ‘FD’ Version 1.0-12 (Laliberté & Legendre 2010; Laliberté et al.2014) and a lingeos correction to correct for negative values generated by a non-Euclidean matrix. This metric is used as a similarity proxy, where species with body shapes similar to the mean body shape will have small distance values and are therefore not be considered unique amongst the group, or do not show potential to uptake a different set of resources amongst the group.
Rare species that were not abundant enough to estimate probability of occurrence through time using the GLMMs described above were still included in the morphometric model but were given an occurrence change value of 0 to represent a null probability of occurrence relative to the theoretical niche size. We then ran two linear models where area and distance metrics generated for each species were used as predictor variables for mean change in probability of occurrence, which was calculated by subtracting the 2002 estimated probability of occurrence from the 2018 estimated probability of occurrence (SI, Table S1). If unique traits are important for determining competitiveness within the community, we would expect that species that show large positive changes in probability of occurrence would have high uniqueness metrics.