Figure 6. Difference
between validation metrics for the standard 30-year average (STA)
against each of the three temporal resolutions (T01, one-year; T05,
five-year; T10, ten-year). Each panel shows the pairwise difference
between validation metrics of 1000 iterations (4 folds × 250 repeats)
based on repeated k -fold cross-validation. The histograms show
the distribution of the pairwise differences of the area under the curve
(AUC; shown in A), the Continuous Boyce Index (CBI; shown in B), the
10th percentile of training omission rate of the
validation folds (OR; shown in C), and the withheld data (OR-W; shown in
D) between the standard and each time-matched approach. The black dashed
lines represent the expected mean values if the standard and
time-matched matching approaches are the same. The red lines in the
histograms indicate the mean value of the difference. Mean values above
the dashed line indicate higher values for the standard 30-year average
approach. Higher values in AUC and CBI represent a better performance of
the standard approach. In contrast, for the omission rates (OR and
OR-W), lower values represent a better performance of the standard
approach. Asterisks in the top-right of each plot indicate significant
differences obtained by the correlated t-tests.