Variables and data
We used climate data from the Koninklijk Nederlands Meteorologisch Instituut (KNMI 2016) from the period of 2000 to 2015 in bioclim format (Fick & Hijmans 2017). A Principal Component Analysis (PCA) was used to transform the 19 bioclim variables in five orthogonal PCA axes that explained more than 90% of the variation. The resolution of the climate variable was 100 m (100 m by 100 m) and all other variables were transformed to the same resolution. The land use data consisted of 15 land use categories from different sources (see table S1 in supporting information; Inter Provinciaal Overleg, 2016; Ministerie van Economische Zaken (EZK), 2015; Centraal Bureau voor de Statistiek (CBS), 2012). Together with the sum of the different land uses per grid cell as an indicator for landscape heterogeneity, this resulted in 16 land use variables. The 8 main aggregated soil types in the Netherlands (table S2; grondsoortenkaart 2006) and the land use data were transformed to a 100 m resolution raster format with percentage cover values and they were not strongly correlated (Spearman’s ρ < 0.7; Dormann et al. 2013).
The bee occurrence data used in this study consist of opportunistic observations of bees and flower visitation records from 2004 to 2019, obtained from the European Invertebrate Survey Netherlands (EIS; http://www.eis-nederland.nl/). Of the more than 300 bee species in the Netherlands we selected 194 species, discarding the species without flower visitation information and with less than fifteen observations. Using the flower visitation data, the most visited plant species and genus were determined. Bees were classified as either oligolectic (includes both monolectic species which visit a single plant taxon and oligolectic species which show a clear preference for a single plant family), polylectic (collect pollen and nectar from various plant taxa) (Rasmussen et al. 2020) and cleptoparasitic bees; these traits were based on a database created for the Status and Trends of European Pollinators (STEP) project and maintained by Stuart Roberts (Potts et al. 2015). For all the 45 modelled oligolectic bees the most visited plant was the plant that they are dependent on for their pollen in the bee trait database (table S3). Since no quantitative interaction data was available for cleptoparasitic bees and their host, we used the occurrences of all the host bee species and genera in the Netherlands listed in the literature (table S4; Peeters & Nieuwenhuijsen 2012). The plant occurrences for the period 2004-2019 were obtained from Dutch National Database of Flora and Fauna (NDFF 2021). The occurrences of the interacting species were kept at a resolution of 100 m or aggregated to a coarser resolution (500 m, 1 km, 5 km and 10 km) and they presented presence/absence, i.e. binary data, similar to Godsoe et al. (2009). Some of the cleptoparasitic bees had multiple hosts and the biotic interaction variable map had a presence if any of the host species or genera was present.