Evaluation of Eta Model seasonal precipitation forecasts over South America
Abstract. Seasonal forecasts run by the Eta Model over South America were evaluated with respect to precipitation predictability at different time scales, seasonal, monthly and weekly for one-year period runs. The model domain was configured over most of South America in 40km horizontal resolution and 38 layers. The lateral boundary conditions were taken from CPTEC GCM forecasts at T62L28. The sea surface temperature was updated daily with persisted anomaly during the integrations. The total time integration length was 4.5 months. The Eta seasonal forecasts represented reasonably well the large scale precipitation systems over South America such as the Intertropical Convergence Zone and the South Atlantic Convergence Zone. The total amounts were comparable to observations. The season total precipitation forecasts from the driver model exhibited large overestimate. In general, the largest precipitation errors were found in ASON season and the smallest in FMAM. The major error areas were located along the northern and northeastern coast and over the Andes. These areas were present in both models. The monthly precipitation totals indicated that the intra-seasonal variability, such as the monsoonal onset, was reasonably captured by the model. The equitable threat score and the bias score showed that the Eta Model forecasts had higher precipitation predictability over the Amazon Region and lower over Northeast Brazil. The evaluation of the precipitation forecast range showed that at the fourth month the forecast skill was still comparable to the first month of integration. Comparisons with the CPTEC GCM forecasts showed that the Eta improved considerably the forecasts from the driver model. Five-member ensemble runs were produced for the NDJF rainy season. Both driver model and Eta Model forecasts showed some internal variability in the SACZ and over the Andes regions. Comparison of the Eta Model seasonal forecasts against climatology showed that in general the model produced additional useful information over the climatology. Transient variability was evaluated by tracking the frontal passages along the eastern coast. The frontal timing was no longer captured by the model but some indication of the frequency and of the northward movement was given by the model forecast. Weekly precipitation totals were evaluated for the São Francisco Basin. Some parameters, such as the mean and the standard deviation of the 7-day total precipitation, were comparable to observations. The correlations between the forecast and the observed 7-day series were positive, but low.