The winter precipitation difference (JJA97-JJA88) over southeastern SA for the observation, AGCM, and Eta were 1.90, 0.23, and 1.60 mm day-1, respectively. On the other hand, the summer precipitation differences (DJF97-DJF88) over southeastern Brazil were 4.22, 0.07, and 2.11 mm day-1 for the observation, AGCM, and Eta, respectively. Precipitation intra-ensemble noise maps indicated that the Eta exhibited higher predictability (less noise) than the AGCM over subtropical SA.
The spectral decomposition of kinetic energy (KE) and moisture flux convergence (MFC) spatial patterns indicated that the DDM's improvement resulted mostly from enhanced small-scale (500 to 1200 km) MFC associated with the topography. We conjecture that the improvement of DDM in this study is mainly due to proper representation of the Andes. Sensitivity tests were carried out to test this hypothesis. The lack of interannual variability in baroclinic wave activity, as indicated by the KE spectral analysis, suggested why the DDM's subtropical precipitation differences were not as large as observed.
This study also revealed a weakness in the simulation over the tropics. Similar to our investigation on seasonal downscaling (De Sales and Xue, 2006), the Eta simulations there were worse than the AGCM's alone. Other Eta-model DDM studies have also shown a deficiency in simulating intense summertime tropical rainfall over SA. Its possible causes will be addressed in this paper.
Supplementary URL: