Butterfly data
Data were collected with Butterfly Monitoring Schemes running in Finland, Northern Spain (Catalonia), and the UK. In these schemes, volunteers count butterflies along line transects following a standardised framework called the ’Pollard walk’ (Pollard & Yates 1993). The counts are then processed using generalized additive models (GAMs) to give an index of abundance per site and per year (Denniset al. 2013; Schmucki et al. 2016). Note that a single number is provided both for multivoltine and univoltine species, therefore it is the total abundance across the observation period. If there are missing counts at a site, the GAM model interpolates abundance based on counts made at other sites in the same bioclimatic zone (Metzger et al. 2013), producing unbiased estimates of abundance (Schmucki et al. 2016).
In the different countries, the number of sites and the duration of monitoring varies: Finland (1999, number of sites = 107), Spain (1994, n = 130) and the U.K. (1976, n= 2128). Therefore, the country with the scheme of shortest duration, Finland, set the time period for our study (1999-2017), and only sites with >= 10 years of data were retained, leaving 59 and 55 sites from Spain and Finland respectively. To maintain a balance in the sampling between the countries, 60 sites with >= 10 years of data were randomly sampled from the larger collection of UK sites under a condition that each was >20 km from the others. To assure robust estimates from the GAM, indices of abundance with >50% of missing observations were also removed. This retained the same number of transect sites (n=174) but removed some years at sites when the counts were made infrequently. A further 31 sites were removed prior to the final analysis, as they either contained only one species (n=3), there was an insufficient overlap of species occurrences to calculate asynchrony, or too few occurrences per species to calculate average population stability (n=28); note this occurred after species were removed due to data limitations when constructing the hypervolume (below). This left a final dataset of 143 sites.