Interannual-to-interdecadal Variability of river flows in southeastern South America: Description and predictability

The Paraná, Paraguay, Uruguay and Negro are major rivers of the La Plata basin in southeastern South America. They all exhibit interannual variations associated with ENSO. The two northernmost-the Paraná and Paraguay-also contain a 9-year component associated with sea surface temperature (SST) changes over the tropical North Atlantic (Robertson and Mechoso, 1998, J. Climate), together with an upward trend. This paper describes another oscillation of all the rivers that we have recently identified. This oscillation is associated with variations of the South Atlantic Convergence Zone (SACZ), and has a 15-17 year period. When the SACZ is intensified, the Paraná and Paraguay rivers in the vicinity tend to swell while the Uruguay and Negro rivers to the south tend to shrivel. This behavior is consistent with enhanced precipitation in the Paraná and Paraguay basins as the SACZ intensifies, and decreased precipitation in the Uruguay and Negro basins as a region of anomalous descent intensifies south of the enhanced SACZ.

The variance accounted for by these variations is about 26% for ENSO, 15% each for the 9-year and 15-17 year oscillations, and about 25% for the trend. While not strictly additive because each occurs to a varying degree in the four rivers, these oscillatory components indicate a potentially useful degree of regional climate predictability. In this paper we use the 1911-93 river flow records to build and test an autoregressive statistical prediction model for these rivers, based on the above oscillatory components. The model uses a standard approach based on singular spectrum analysis (SSA) combined with the maximum entropy method (SSA-MEM; Keppenne and Ghil, 1992, J. Geophys. Res.). SSA is firstly used to identify the predictable components. Low-order autoregressive models are then fitted to the filtered components, which are then extrapolated forward in time to make a prediction. Using contingency tables, we demonstrate marked changes in the conditional probability of above and below normal monthly streamflow, according to the phase of the slow oscillatory components. We report estimates of hindcast skill of these oscillatory components, and make some preliminary probabilistic predictions.