Fig. 3 The proportion of variance (% variation) in compositional dissimilarity among plant communities (βturnover, based on Simpson index) in each subset (large/small genome size) as explained by environmental variables (including MAP, MAT, altitude, soil N, and soil P) and geographical distance using different genome size thresholds. Plant communities were separated into the large- and small- genome size (1C DNA content, pg) subsets by either the median (1.59 pg) or mean (3.19 pg) genome size for the 161 species of the present study or from the median (2.50 pg) or mean (5.90 pg) genome size of the global terrestrial plants in the Kew database (see Materials and Methods ). Generalized dissimilarity models (GDMs) were fitted using presence/absence data for the large- (blue) and small- (yellow) genome size subsets. (a) Pure environmental variation without a spatial component represents the effect of environmental filtering. (b) Pure distance variation without an environmental component can represent the effect of dispersal limitation if all important environmental drivers are included in the environmental component. (c) The ratio of variation explained by pure environmental variation versus geographic distance was calculated, with estimates also shown numerically on top of each bar. (d) Measured genome size for the 161 species ranked from small to large along the x-axis, with red dashed lines indicating thresholds used in the analyses; the numbers inside parentheses indicate the number of species in each group.