Fig. 5: Confirmatory path models for each local land-use type.

a, forest. b, meadow. c, arable field. d, settlement. Each confirmatory path model contains two components: one part using linear models to explain green-up variables by mean spring temperature. In forests, n = 55 per model, in meadows, n = 45 per model, in arable fields, n = 44 per model, in settlements, n = 35 per model; the other part using generalized additive models to explain BIN richness and biomass by the green-up variables, climate variables, regional land use and plant species richness as fixed linear effects. As in the initial model (Fig. 2, Table 1), ‘Day’ was used as smoothed effect and space as random effect. Instead of offsets of log(sampling duration), log(richness)/log(sampling duration) and log(biomass)/log(sampling duration) were used to control for sampling period differences. For biomass, family = gaussian(link = “log”) and for richness, family = negative binomial was used. In forest plots, n = 376 for richness and n = 395 for biomass, in meadows, n = 310 for richness and n = 333 for biomass, in arable fields, n = 298 for richness and n = 313 for biomass. In settlement plots, n = 230 for richness and n = 252 for biomass. Numbers indicate standardized partial effects. Dark grey thick arrows indicate significant paths (p < 0.05), light grey insignificant paths. Pictures show Malaise trap plots in each of the local land-use types.