Sources of Subseasonal Predictability for Precipitation in South America Based on Model Experiments

Abstract This study investigates the sources of predictability underlying subseasonal precipitation skill over South America in existing subseasonal prediction systems. Using subseasonal re‐forecasts from the NCAR‐CESM2 model, we demonstrate that significant skill persists even when interannual variability is removed. The highest skill occurs during austral summer and spring. To isolate sources of predictability, we analyze a novel set of re‐forecast experiments initialized with climatological atmosphere, land and ocean states. Results indicate that atmospheric initial conditions are essential for achieving skill, while ocean initializations contribute to skill containing interannual variability and land initializations contribute minimally to skill. Canonical correlation analysis is used to identify the most skillful spatial patterns and associated time series during the high‐skill seasons. The most skillful patterns in December–February are not independent of the South American dipole but also contain additional skill beyond the dipole. This additional skill is not associated with propagating tropical convection or sea surface temperature anomalies indicating that the Madden–Julian Oscillation, El Niño–Southern Oscillation, or South Atlantic SST variability are not the sources of predictability. Instead, a wave‐train‐like structure extends across the South Pacific, resembling a Rossby wave response. Idealized experiments using a simplified atmospheric general circulation model show that this pattern can be reproduced by stationary tropical heating over the Maritime Continent, suggesting a dynamical link between tropical heating and South American precipitation variability on subseasonal timescales independent of ENSO and the MJO.