Utility of observational datasets to predict genetic differentiation
Biomass corrected reproductive traits, compared to vegetative traits, showed a better match between observational trait-environment relationships and response to source environmental gradients in the greenhouse (Table 1). For vegetative traits, there were two cases of predictable genetic differentiation out of eight. In both cases, field data correctly predicted not the presence but the absence of genetic differentiation. The low predictability in vegetative traits originated in some cases from interacting (e.g., Fig. 4b) or opposing (Fig. 4d,e) effects of source and exposure environments. In other cases, field patterns did not match with those expected from the combined source and exposure greenhouse effects (Fig. 4a). For biomass-corrected reproductive traits, observational data predicted the presence and direction of all seven source environment effects in the greenhouse. However, when reproductive traits were analysed without biomass as a covariate, observational data only predicted the presence and direction of three out of seven source environment effects (Table S10).