Predictability of the NCEP-NCAR reanalysis model during austral summer

Ten years of forecasts during Northern Hemisphere winter produced by the NCEP-NCAR reanalysis project are analyzed and their errors are documented with a focus on forecasts over South America. Previous studies have documented a seesaw pattern in the South Atlantic Convergence Zone (SACZ). Events associated with a strong SACZ are periods when the subtropical plains of South America exhibit precipitation deficits. When the SACZ weakens, precipitation in the plains is abundant. The forecast errors during these periods are examined separately. An error climatology is also obtained based on all available forecasts for January and February. It is found that cases with weak SACZ are characterized by relatively smaller forecast errors in the Americas and neighboring oceans during the first three days than are strong SACZ cases. The error growth for these weak SACZ cases is larger and the forecast errors exceed those in the strong SACZ cases after about 5 days. Upper-tropospheric divergence is systematically underpredicted over tropical and subtropical South America in the 117 studied forecasts, and it is overpredicted over the Caribbean. Similar features are also found in the 6-h forecast precipitation verified against global precipitation index estimates. The 3-8-day model forecasts retain some characteristics of the weak and strong SACZ events, but underpredict their amplitudes. Eight-day predicted upper-tropospheric zonal winds have substantial errors over North America. Integrations with a simple global numerical model suggest that the predicted wind error over North America is due to latent heating errors associated with precipitation forecast errors located nearby. Central and western Pacific errors do not appear to be as important for wind forecast errors in this region. Weak SACZ events are characterized by stronger low-level jets and more poleward moisture transport east of the Andes than are strong SACZ events in the NCEP-NCAR reanalysis. This distinction is substantially weakened in the 8-day forecasts. Integrations with a global model suggest that inadequate simulation of the distribution of latent heating and of radiative heating may have contributed to the inability of the NCEP forecast model to adequately distinguish low-level flow structures during positive and negative events.