2.2.3. Phenological Metrics

The phenological metrics represent the characteristics of the vegetation within its phenological cycle, or phenophases, corresponding to dimensionless output parameters and can be calculated based on the EVI time series. In this study, the TIMESAT software [50,51] was used to analyse 20 years of EVI time series (from 2000 to 2019) and to compute 4 phenological metrics: Start of Season (SOS), End Of Season (EOS), Length of the season (LOS), and Amplitude (AMPL) difference between the peak and the base level value. After applying the Savitzky-Golay filter [52] in TIMESAT, a seasonality parameter per year was chosen, representing a phenological cycle with a start and end level of 20% of the seasonal amplitude. This threshold value was used in several studies and is known to be accurate in registering the plant's phenological transitions [53–58]. The four phenological metrics are shown in boxplot graphics for each ecoregion studied and used for correlations with environmental drivers.

#### 2.2.4. Seasonal Variability Analysis

Interannual variability in phenological metrics across sites—To test if the interannual variability in SOS and EOS (i.e., the variances in phenological transition dates) differs among the three sites, we used the non-parametric test Fligner–Killeen [59]. The Fligner–Killeen test compares the homogeneity of variances among samples [59]. The same analysis was performed to test if the variability of SOS differs from the EOS within each site. To test if the LOS and the AMPL of EVI differ among sites, we performed a one-way analysis of variance (ANOVA) followed by the Tukey post-hoc test. In addition, we used box plots to compare the three experimental sites' seasonal data on EVI, phenological metrics, and environmental drivers for 20 years.

Influence of environmental drivers on phenological metrics—The environmental drivers were also observed monthly to assess their influence on phenological parameters. This study analysed these relationships through scatter plots with smoothed (r) correlation curves from Pearson's classification (P) between the EVI and environmental drivers. Pearson's correlation between EVI time series and monthly environmental drivers was submitted to different monthly lag periods (0–3). The lag monthly of each environmental driver with the highest Pearson's correlation was applied to the partial correlation method. Then the partial correlation method was used to analyse the environmental drivers in the monthly EVI time series [60]. Data analysis was carried out in R [61].
