Effects of individual collection biases on model accuracy
Phenoclimate models derived from sample data that exhibited biases towards collection of specimens proximate to the beginning or end of their individual flowering periods exhibited higher MAE than models constructed from specimens collected with no inherent bias, with mean MAE among species exceeding 8 days in all cases (Fig. 5). However, models constructed from specimens collected with a bias towards collection of specimens proximate to their peak flowering DOY consistently exhibited lower MAE than models derived from unbiased data (Fig. 5). However, the effects of this form of bias are intrinsically linked to individual flowering duration (with longer durations associated with higher MAE); models produced from specimens exhibiting bias towards collection of specimens shortly after onset or shortly before termination exhibited greater accuracy and lower MAE among species with 15- or 30-day flowering durations (Table S4) than among species with 60-day durations.