Host- and habitat-related variables
During sampling we recorded the locations of the plants with GPS and measured the size of the plants (Table 1). We recorded signs of herbivore damage, and surveyed surrounding vascular plant communities between 1930 June 2017 by counting the number of vascular plant species within multiple one-square-meter quadrats in each population. We used the vegetation data to calculate the Shannon diversity index (Shannon 1948) for each population (Table 1).
We estimated the proportion of agricultural area surrounding each study population from Corine Land Cover (CLC, version 2020_20u1) with QGIS (QGIS Development Team 2019) by creating a one-kilometre buffer zone around each study population following the patch borders and calculating the proportion of 20 m × 20 m pixels falling under agricultural land use category within this buffer zone (Table 1). We also estimated the coverage of P. lanceolata foliage in square meters in each population. We quantified the connectivity of a host population with respect to other populations by calculating the Euclidian distances between populations and calibrating these measures by the species dispersal capacity (Hanski 1999).
We obtained weather observations for the study populations from the Finnish Meteorological Institute (Aalto et al. 2016). We calculated the number of severe winter days during the winter before the sampling season (2016-2017), and the sum of temperatures of the effective summer days during previous summer (2016). See Table 1 and Supplementary Material and Methods for more details, including biological justification for the selected variables.