Dynamic Downscaling of Seasonal Climate Predictions over Brazil

Abstract

Climate projections for March?April?May (MAM) 1985 and 1997 made with the NASA Goddard Institute for Space Studies (GISS) GCM over South America on a 4° latitude by 5° longitude grid are ?downscaled? to 0.5° grid spacing. This is accomplished by interpolating the GCM simulation product in time and space to create lateral boundary conditions (LBCs) for synchronous nested simulations by the regional climate model (RCM) of the GISS/Columbia University Center for Climate Systems Research. Both the GCM and the RCM simulations use sea surface temperature (SST) predictions based on persisted February SST anomalies. Each downscaled prediction is evaluated from an ensemble of five simulations and each is compared to a control ensemble of five RCM simulations driven by synchronous NCEP reanalysis data. An additional five-run control ensemble for MAM 1997 tests the impact of ?perfect? SST predictions on the RCM forecast. Results are compared to observational evidence that includes NCEP reanalysis data, Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) gridded fields, some rain gauge observations, and satellite measurements of monthly mean outgoing longwave radiation. The downscaled predictions and the downscaled analyses both capture the meridional displacement of the intertropical convergence (ITC) precipitation maximum over northern Brazil between the two seasons. The simulation of this feature for MAM 1997 is improved by using actual SST, but the correction of underestimates of eastern Brazil precipitation requires analyzed LBC in place of GCM forcing. The realism of spatial patterns and area averages of precipitation neither improves nor deteriorates with elapsed time, but the variability between individual runs forced by the same LBC decreases with time. The RCM shows a positive bias in surface temperature over central and southeastern Brazil and a positive bias in temperature at 850 mb over the Tropics. Results imply that improvements in seasonal climate prediction at the 0.5° spatial scale over Brazil could be realized by more realistic GCM forcing, accurate SST predictions, and improvements in the RCM.