The effect of different initial conditions on short-term rainfall prediction

This study deals with the short term prediction of precipitation over the Southeast Region of Brazil during an extreme rainfall event. It will show the first attempts to obtain higher correlations between the 24-hr accumulated predicted precipitation fields and observations by means of a modified initial condition in a version of the Center for Ocean, Land and Atmosphere Studies (COLA) Global Spectral Model used by the Center for Weather Prediction and Climate Studies (CPTEC-Brazil). The modified initial conditions are provided by Physical Initialization (PI) reanalysis from the FSU Global Spectral Model (FSUGSM). The models have similar convection schemes. The PI procedure was developed at The Florida State University (FSU) by Krishnamurti et al. (1991) to improve the precipitation in the tropics. These reanalyses are then used to initialize the CPTEC/COLA model in order to reduce the "spin up" during the first 24 hours of integration. The National Centers for Environmental Prediction (NCEP) analyses are used for initializing the FSUGSM (Day -1) and for defining the prescribed value toward which vorticity, divergence and surface pressure are nudged (Day 0). Some additional initialization in the NCEP analyses is still required to increase the correlation at Day 0. Nevertheless, the correlation coefficients are better when using PI.