Sensitivity of Southern Hemisphere and Global Numerical Weather Prediction to Reanalysis Uncertainty

This study addresses optimal distribution of observations for short-to-B medium range weather prediction over South America. An uncertainty estimate is obtained from the difference of ECMWF and NCEP reanalyses of the same state. This uncertainty measure assumes that the differences of modern analyses would be negligibly small for a perfect observing system. One global prediction is made from each reanalysis. These diverge with time, and the forecast differences provide an estimate of forecast sensitivity to initial state uncertainty. On a global basis, the spatial scales responsible for most of the uncertainty growth are contained in global wavenumbers 1-15 in the initial state uncertainty. The present results suggest that an observing system that resolves these relatively large scales would provide most of the forecast benefits of a high resolution global observing system. Several experiments are done to determine the benefits associated with targetting specific local regions in reducing forecast errors. Targetted regions include the South American sector, as well as the South Pacific sector.