An Atlantic influence on Amazon rainfall

Abstract

Rainfall variability over the Amazon basin has often been linked to variations in Pacific sea surface temperature (SST), and in particular, to the El Niño/Southern Oscillation (ENSO). However, only a fraction of Amazon rainfall variability can be explained by ENSO. Building upon the recent work of Zeng (Environ Res Lett 3:014002, 2008), here we provide further evidence for an influence on Amazon rainfall from the tropical Atlantic Ocean. The strength of the North Atlantic influence is found to be comparable to the better-known Pacific ENSO connection. The tropical South Atlantic Ocean also shows some influence during the wet-to-dry season transition period. The Atlantic influence is through changes in the north-south divergent circulation and the movement of the ITCZ following warm SST. Therefore, it is strongest in the southern part of the Amazon basin during the Amazon’s dry season (July–October). In contrast, the ENSO related teleconnection is through anomalous east-west Walker circulation with largely concentrated in the eastern (lower) Amazon. This ENSO connection is seasonally locked to boreal winter. A complication due to the influence of ENSO on Atlantic SST causes an apparent North Atlantic SST lag of Amazon rainfall. Removing ENSO from North Atlantic SST via linear regression resolves this causality problem in that the residual Atlantic variability correlates well and is in phase with the Amazon rainfall. A strong Atlantic influence during boreal summer and autumn is particularly significant in terms of the impact on the hydro-ecosystem which is most vulnerable during the dry season, as highlighted by the severe 2005 Amazon drought. Such findings have implications for both seasonal-interannual climate prediction and understanding the longer-term changes of the Amazon rainforest.

Appendix 1: Emperical orthogonal function (EOF) analysis

Principal component analysis has been used to identify coherent interannual variability patterns over the Atlantic Ocean (e.g., Ruiz-Barradas et al. 2000). Here, rotated multivariate (rainfall and SST) EOF is adapted to summarize our findings. Conventional approaches are taken, such as normalization by individual variables’ standard deviation, and the first five eigenmodes meeting a criterion of explaining >5% of the variance each (together they explain more than 68%) were then rotated, while preserving the orthogonality in time, but relaxing it in space. In particular, the total number of grid points for both variables have been adjusted using interpolation, to reduce the possibility that one variable dominates the other. By relaxing the spatial orthogonality constraint, the spatial patterns of the rotated eigenvectors should be more directly related to the natural patterns of variability in the physical system (Richman 1986), and becomes more useful in identifying localized variability in Atlantic Ocean (Dommenget and Latif 2002). A number of algorithm choices are possible for rotation. Here the varimax method is used following Kaiser’s (1958) approach.

To emphasize the relative roles of different oceanic SSTs on Amazon rainfall, we rank our eigenmodes in decreasing order of local variance over the Amazon basin rather than following a conventional decreasing order of global variance of each mode. To achieve this, each reconstructed eigenmode [Z(x,y,t)] is obtained by multiplying its eigencoefficient [C(t)] and eigenvector [E(x,y)], in other words projecting these modes back to the time-space domain. The standard deviation of Amazon basin averaged rainfall of each eigenmode is computed. The first three modes, using this local variance ranking, are displayed in Fig. 11.

Fig. 11

Rotated Multivariate EOF (RMEOF) of rainfall (Xie and Arkin 1996) over tropical South America (north of 22°S) and SSTs (Rayner et al. 2003) over the tropical Pacific and Atlantic Oceans. The contour interval is 0.2 (°C and day⁻¹). The variance of precipitation over the Amazon explained by these three modes is 57.9% for North Atlantic SST, 30.4% for tropical Pacific SST, and 7.8% for South Atlantic SST

One of the advantages of the RMEOF is isolation of the temporally orthogonal variability. It is well known that both tropical Atlantic oceanic SSTs and Amazon rainfall are affected by the interannual variability of the tropical Pacific Ocean (e.g., Enfield 1996; Zeng 1999). By employing this RMEOF, we can present three different patterns of interannual variability, connecting Amazon rainfall and the surrounding oceanic SST’s. Tropical North Atlantic and Pacific influence on Amazon rainfall variability in interannual timescales are actually separated in RMEOF analysis. These patterns are arranged in the decreasing order of its influence on Amazon rainfall variability. The largest pattern is associated with the North Atlantic Ocean (58%). The second and third ones are related to the tropical Pacific Ocean (30%), and the South Atlantic Ocean (8%), respectively. These magnitudes of variance are slightly different from those obtained from multiple linear regression. This could be due to numerical relaxation of orthogonality for the spatial pattern. For example, the western side of the tropical Atlantic Ocean is more closely correlated with tropical ENSO (Enfield and Mayer 1997). Yet, this analysis separates the western Atlantic variability from the Pacific variability, so that some portion of variance in the first pattern (Fig. 11) might not be pure Atlantic influence on Amazon rainfall. Consistent patterns of interannual variability with previously identified ones are found in Fig. 11. Also, this is consistent with observations and model results which show that the SST over the tropical Atlantic affects the interannual variability of climate over the Nordeste and the eastern Amazon (Hastenrath and Heller 1977; Marengo 1992; Moura and Shukla 1981; Ronchail et al. 2002). The influence of Pacific SSTs on the interannual variation of South American rainfall is concentrated over the northern part of the Amazon and northeastern Brazil (Nordeste).