Comparing environmental values before model training
Following the temporal filtering, the analysis included 53 records
observed between January 1971 and December 2000 (see Supplemental
material, Table S1). The paired Wilcoxon signed-rank tests revealed
significant differences in the environmental values of the occurrences
when comparing the standard 30-year average approach (STA) to each of
the temporal resolutions (T01, T05, T10) for the maximum temperature of
the warmest month (bio05) and the precipitation of the driest month
(bio14). In addition, the precipitation of the wettest month (bio13)
showed a significant difference when comparing the 30-year standard
average (STA) against the one-year temporal resolution (T01). The
minimum temperature of the coldest month (bio06) did not present
significant differences between the approaches. The Bland-Altman plots
also confirmed the quantitative results, showing the difference between
the standard average and the temporal resolutions (Fig. 4).
Specifically, significant differences in environmental values were
observed when the mean values of the subtraction between the 30-year
average and the temporal resolution (represented by horizontal black
lines) deviated significantly from the y-axis zero. Bio06 indicated a
temporal pattern where older records (dark orange) had lower
temperatures when doing a time-matched approach, while recent
occurrences (light orange) had higher temperatures. This temporal
pattern was not evident for precipitation-based variables (bio13 and
bio14) and the other temperature variable (bio06). Moreover, the
magnitude of the difference decreases for all the variables as the
temporal resolution is lower (i.e., from one-year to ten-year). For
example, the difference between standard and time-matched approaches in
some occurrences was as high as 2°C when comparing a one-year temporal
resolution to the 30-year average but decreased to 0.5°C when comparing
against the ten-year temporal resolution.