Effects of species characteristics on model accuracy
We obtained five general results from our comparisons of model accuracy
when using simulated data sets characterized by different combinations
of phenological sensitivity to temperature, phenological parameters
(e.g., duration and standard deviation of flowering times), and sample
size. First, the magnitude of species’ responsiveness to climate had no
significant effect on model accuracy when predicting DOYs of flowering
onset, median, or termination using unbiased collections (F ≤ 0.3, p ≥
0.13, Table 1). The magnitude of intrapopulation phenological variation,
individual flowering duration, sample size, and collection bias did
exhibit significant effects on the accuracy of predicted flowering
onset, median, and termination DOYs (p<0.01 in all cases,
Table 1). However, mean MAE of predicted median (or peak) flowering DOY
remained both low and consistent across all categories of taxa (ranging
from 0.6 days at minimum, to 1.9 days at maximum, Fig. 2, Table 2).
Predictions of flowering onset and termination DOYs exhibited higher MAE
than estimates of median flowering DOYs across all categories of species
(p < 0.001 in all cases, Table 2), but also remained under 5
days unless individual flowering duration was long (60 days; Fig. 2,
Table 2). The mean MAE of predicted median flowering remained under 2.1
days among species exhibiting long individual flowering durations.
However, estimation errors for onset and termination DOYs were quite
high, with MAEs reaching 14 days when individual flowering durations
were long and intrapopulation variation was low (Fig. 2, Table 2).