Genome size and species distribution: HMSC model
The HMSC model showed a good fit to the data, with a mean Tjur R2 of 61% for explanatory power and 14% for predictive power. We observed both positive and negative associations with species occurrences for each variable (Fig. 4a). In our HMSC model, parameter ρ, the strength of the phylogenetic signal, was 0.1 in median, with the 95% confidence interval between 0.0 and 0.3 (overlap with zero), suggesting that species niches (i.e. their responses to the environmental covariates) showed no phylogenetic correlations. We then calculated the community weighted mean (CWM) genome size against environmental covariates by using predicted occurrences from the HMSC model (Fig. 4b). This is because using CWMs based on raw occurrence data in regressions can be subject to type I error (Miller et al.2019). The model approach to analyse trait–environment associations with community data maximizes power and information contained in the data (Brown et al. 2014; Miller et al. 2019). CWM of genome size was greater in communities from those sites with higher resource availability, supporting our hypothesis that alleviating resource (water and nutrient) limitation would favour large-GS species. Altitude was a poor predictor for CWM of genome size while MAT was negatively correlated with CWM of genome size. Previous studies also found that low temperature favours large-GS plants because the large-GS species gain an advantage in frost resistance (Grime & Mowforth 1982; MacGillivray & Grime 1995).