Environmental Dataset
As temperature influences intertidal organisms at all life-history stages, we calculated minimum, maximum, and average SST and AT in winter (December-February) and summer (June-September) months between 2008-2018. We included September in the summer calculations to encompass the entire potential larval period of S. umbilicalis . We used seasonal averages for the decade prior to and including the year of sampling so that the temporal resolution of the environmental data encompassed multiple generations of selective pressure (Flanagan et al. 2018). SST data was obtained as monthly means per year at a resolution of ~11 x 7.5 km2 from the EU Copernicus Marine Service Information Atlantic-European North West Shelf Ocean Physics Reanalysis dataset (REF: CMEMS-NWS-PUM-004-009). AT data was obtained as temperature 2 m from the sea surface, at a resolution of 9 km2 from the EU Copernicus Climate Change Service ERA5-Land dataset (DOI: 10.24381/cds.68d2bb30).
We also incorporated descriptors of historical climate using five bioclimatic variables obtained from the WorldClim dataset (Fick & Hijmans 2017): maximum AT of the warmest month, minimum AT of the coldest month, AT annual range, mean AT of the warmest quarter and mean AT of the coldest quarter. These variables represent historic 30-year monthly averages from 1970-2000 at a resolution of 5 km2.
Wave exposure data was taken from the UK dataset of Burrows et al. (2008), who developed a grid-based model to calculate indices of wave fetch, a parameter that describes the distance to the nearest land cell.
Most available records of environmental variables were either “land-based” or “sea-based” with few encompassing the intertidal environment. To address this, we took averages of all recorded temperature values within a 10 km radius of each sampling site. For site/environmental variable combinations where no data was available within this radius, we used the nearest data cell to characterize the site. As the exposure dataset is high-resolution coastal data, we were able to obtain more precise estimates for each site and retained the average wave fetch value of the nearest cell only. We used average wave fetch (representing the average of the summed fetch values for the cell and its immediate neighbours) to incorporate some variability in the local degree of exposure at each site. All environmental data manipulations were conducted in ArcGIS (v10.7.1; ESRI 2011).
We conducted a principal component analysis (PCA) in R (v4.0.3; R Core Team 2020) to reduce the set of 18 environmental variables into fewer, orthogonal PCs that describe environmental variation at our study sites. We retained PCs that had eigenvalues >1 and assessed the loadings of environmental variables onto each retained component. We used a cut-off value of 0.32 to attribute variables to each PC, as this threshold represents environmental variables that have at least 10% of their variance explained by the PC (Dormann et al. 2013). We then calculated the standardized scores of each retained PC for subsequent analyses.