[Figure 6]
The spatio-temporal variability of rainfall using the PCA technique was based on climatological classification similar to Fig. 4. Based on the Bartlett’s test statistics (e.g., Snedecor and Cochran, 1989) more than five significant modes explaining non-random variations in rainfall at 95% confidence level were identified. But for purposes of physical interpretability, we focus on three of those rainfall modes. Leading rainfall modes in the Congo basin are characterised by annual, multi-annual and short-term seasonal signals (Figs. 5a-b and 6a-b). There is a strong dipole patterns in dominant mode where strong spatial patterns of rainfall on the opposite side of the equator are notable (EOF-1, Figs. 5a-b and 6a-b). These spatial patterns show the different wet seasons in the basin. From the time series associated with these spatial patterns, the wet season (positive phase) in northern section coincides with the dry season (negative phase) of the southern region (EOF-1/PC-1, Figs. 5a-b and 6a-b). The multi-annual signal, which corresponds to the bimodal rainfall patterns and short-term seasonal signals are relatively stronger on the West and in regions encompassing Congo, Equatorial Guinea, and Gabon (EOF-2 and EOF-3, Figs. 5a-b and 6a-b). Generally, there is significant difference in the observed spatial loadings (averaged spatial distribution) of the first orthogonal mode of rainfall over different climatology (EOF-1, Figs. 5a-b and 6a-b) and confirms the extensive impacts of extreme and severe droughts, which affected more than 50% of the basin between 1901 and 1930 (Ndehedehe et al., 2019). The total variability of the leading modes (annual variations) of rainfall increased during the 1931- 1960 (56.3%) and 1961- 1990 (57.3%) periods compared to the 1901- 1930 baseline period (51.3%). It varied less between 1991 and 2014 (55.4%) as opposed to the two climatological periods between 1931 and 1990. Furthermore, the total variability in the multi-annual rainfall signals declined from 16.5% at the start of the century (1901- 1930) to 13.6% in the 1991- 2014 period while the total variability accounted for by other short-term meteorological signals oscillated between 4.0% and 2.7% during the entire period.