Comparing environmental values before model training
Before building models or comparing their performance, it was essential
to evaluate whether the environmental values between the standard
approach and the temporal resolutions differed. Only if the
environmental values are different is it expected that the models will
also differ. Here, time-matched bioclimatic values of the occurrences
(i.e., T01, T05, T10) were compared against those of the standard
30-year average values (STA). The differences between values were
evaluated qualitatively with Bland-Altman plots (Bland and Altman 1986,
1995) and quantitatively with paired Wilcoxon signed-rank tests
(Wilcoxon 1945). The Bland-Altman plots help to visualize the
differences in environmental values between the 30-year average and the
respective temporal resolutions per occurrence. In the X-axis, each
occurrence is represented by the mean of two values (30-year and one
temporal resolution), and the Y-axis denotes the difference between the
same values. If the difference between the two bioclimatic variables at
a particular site of occurrence is larger than zero, it means that the
30-year average value is higher than the average value obtained by the
time-matched approach; on the other hand, a negative value means that
the 30-year average value is lower than the time-matched approach. For
example, if a reported difference is -20 mm for precipitation, the
standard approach was lower by 20 mm compared to the time-matched
approach. For the Wilcoxon signed-rank tests (12 total), a Bonferroni
adjustment was applied separately to each of the three temporal
resolution comparisons against the standard approach (i.e., STA vs. T01,
STA vs. T05, STA vs. T10 each were considered a family of comparisons)
to reduce the probability of obtaining false positives when performing
multiple comparisons. Each comparison group (e.g., STA vs. T01)
consisted of the four paired Wilcoxon signed-rank tests corresponding to
each of the four bioclimatic variables.