Environmental conditions in source populations
To analyse the effects of environmental conditions of source populations
on traits, we collected information on climate, land-use and vegetation
for each location (Table S2). Mean annual values and seasonality
(coefficient of variation in monthly values) for temperature,
precipitation and moisture index were obtained from the BioClim database
(Kriticos et al . 2012, Fick & Hijmans 2017). We used the highest
resolution available for temperature and precipitation (30 s) and for
moisture index (10 min). In the field, we recorded whether populations
were subject to mowing or not, and estimated the percentage of
vegetation cover and bare ground for four random plots per population.
In two opposite corners of the plots, we quantified community vegetation
height as the height at which a pole was completely obscured by
vegetation, looking from a distance of ca. 4 m.
To avoid collinearity in environmental predictor variables (climate,
land-use and vegetation data), we performed a Principal Component
Analysis (psych package in R; R Core Team 2017, Revelle 2018). We
performed a second, orthogonal rotation that improved the interpretation
of the components (Quinn & Keough 2002). The first three rotated
components explained 70.4% of the variance (Fig. S1, Table S1). The
first component (hereafter “Aridity”) was positively associated with
low mean and high seasonality in precipitation and mean moisture index.
The second component (“Temperature”) was positively associated with
high mean and low seasonality in temperature. The third component
(“Vegetation cover”) was positively associated with high percent
vegetation cover, greater height of vegetation and low percent bare
ground cover. We used these rotated components and the binary factor
Mowing to test the effects of source climate (Aridity, Temperature),
vegetation (Vegetation cover) and land-use (Mowing) on trait variation.
We used t-tests to analyse differences between native and non-native
populations in the rotated components and the underlying variables
(effects of native/non-native range on Mowing were tested with a
Generalized Linear Model using Binomial errors; stats package, R
Core Team 2017).