Climate co-variability between South America and Southern Africa at interannual, intraseasonal and synoptic scales

Abstract

This paper investigates and quantifies co-variability between large-scale convection in the South American and Southern African sectors at different timescales (interannual, intraseasonal and synoptic), during the austral summer seasons (November–February) from 1979 to 2012. Multivariate analyses (Canonical Correlation Analysis and Principal Component Analysis) are applied to daily outgoing longwave radiation (OLR, used as a proxy for atmospheric convection) anomalies to extract the principal modes of variability and co-variability in each and between both regions, filtered to consider the appropriate time-scales. At the interannual timescale, results confirm the predominant role of El Niño Southern Oscillation (ENSO), favoring enhanced convection over both southeastern Brazil and northern Argentina on the one hand, and tropical Africa and the western Indian Ocean on the other hand. At the intraseasonal timescale, the leading mode of co-variability is related to modulations of large-scale atmospheric convection over most of South America, and 10 days later, tropical Southern Africa. This mode accounts for the impacts of the Madden–Julian-oscillation (MJO) over these regions: identifying robust co-variability at the intraseasonal timescale between both regions require thus to consider a temporal shift between the two sectors. At the synoptic scale, however, co-variability consists mostly of a synchronous modulation of the large-scale atmospheric convection over the South American and Southern African sectors. This results from the development of concomitant Rossby waves forming a continuous wave train over the South Atlantic in the mid-latitudes, affecting both the South Atlantic and South Indian Convergence Zones. Among the days when convection shows significant anomalies (30 % of the total days in each sector), this synchronous mode occurs about 25 % of the time, individual Rossby waves modulating convection over one single region only during the remaining 75 % events. Another mode of co-variability, involving a single Rossby wave modulating the convection first over the Americas, and 4 days later over Africa, appears as sensibly weaker than the synchronous mode, suggesting that the “wave train” mode occurs more frequently than the development and propagation of a single wave that could affect both regions.

Appendix: Dominant modes of large-scale convective variability in each region

The Canonical Correlation Analyses used in this work aim at maximizing the co-variability between two domains. The modes extracted could possibly differ from the leading modes of variability calculated for each region separately. This case would imply that the mode extracted probably accounts for a reduced fraction of the regional climate variability in each region and could thus correspond more to a statistical artifact than a physically robust mechanism.

Figures S1 and S2 present therefore the leading EOF of PCA applied to raw (unfiltered) OLR anomalies in each region. Both analyses extract the robust variability patterns, already presented in many previous papers for both regions (e.g. Carvalho et al. 2004; Pohl et al. 2009), which are associated with the spatio-temporal variability of both CZ. These patterns sensibly differ from those found at the interannual (Fig. 2) and intraseasonal (Fig. 5) timescales, which are discussed in this study. The latter are however in fair agreement with the spatial signature of ENSO and the MJO over these sectors, respectively, which confirms the hypothesis of significant co-variabilities caused by the common influence of a single, larger-scale atmospheric or climatic phenomenon encompassing both regions.

The spatial similarity is largest for the synoptic scale, counter-balancing a weaker and intermittent temporal co-variability. For this timescale, one can thus conclude that the mode of co-variability shown in Fig. 9 corresponds to the leading variability patterns influencing each region. Unlike the longer timescales, one can here conclude on co-variability between the “SACZ” and “SICZ”, strictly speaking, and not merely on the South American and Southern African regions.