Prediction skill and potential predictability of subseasonal South American summer precipitation reconstructed from leading modes of variability

Abstract

Improving subseasonal precipitation forecasts requires better comprehending how models represent subseasonal variations in precipitation and what affects models’ predictability and prediction skill. Here, empirical orthogonal function (EOF) analysis is utilised to identify the leading modes of weekly precipitation variability over South America (SA) during the monsoon season, which are identified as (i) a dipole-like precipitation pattern affecting central-eastern and southern Brazil, as well as Paraguay, Uruguay, and northeastern Argentina; (ii) a precipitation pattern with opposite anomalies between northern and southeastern SA; and (iii) a pair of degenerated EOFs with large precipitation variability in eastern SA, including northeastern Argentina, southeastern SA, and part of northeastern Brazil. Evaluations of the links between the leading modes and climate drivers show that the first and third modes are associated with the Madden-Julian Oscillation, whilst the second mode is more closely related to El Niño-Southern Oscillation. Teleconnections, global wavenumber-1oscillation, and coupling ocean-atmosphere dynamics are further factors linked to the leading modes. Assessments of the ability of distinct subseasonal-to-seasonal prediction project models in predicting the leading modes one to four weeks ahead indicate that the prediction quality reduces when the lead time increases, with the multimodel typically overcoming individual models. A large fraction of prediction skill and predictability resulting from the leading modes reveals the importance of adequately representing these modes within models for improving the quality of subseasonal SA precipitation forecasts. Our findings have the potential to support model developments and enhance planning initiatives across multiple sectors, such as agriculture and energy.