Resumo:
This work evaluated the quality of ECMWF-SEAS5 seasonal precipitation and 2 m temperature
predictions for South America. For this purpose, datasets of hindcasts from 1993 to 2016 and
forecasts from 2017 to 2020 were used. The predictions were validated against CPC precipitation
and temperature analyses. The average seasonal fields indicated that the model has a good
representation of the seasonal rainfall patterns in South America, adequately simulating the wet
and dry phases of the monsoon. However, the hindcasts present systematic overestimation of rain
in the Amazon, South Brazil, Southeast Brazil, and northern South America sectors. In addition,
the model also presents an underestimation of rain in Northeast Brazil and southeastern South
America. Regarding the temperature results, the model showed a systematic cold bias over most
of the continent, except for portions of Northeast Brazil and southeastern South America. The
skill score evaluation showed that the main correlations of precipitation and temperature anomaly
occur in regions of high climate predictability, such as the tropical latitudes of the continent. The
regionalized mean anomalies indicated that ECMWF-SEAS5 has a good performance to simulate
the interannual variability of rainfall and temperature, especially in transition seasons. However,
hindcasts were not efficient for predicting anomalous events such as the 2014/2015 drought in
Southeast Brazil and the 2015 drought in the east of the Amazon. The analysis of the forecasts
from 2017 to 2020 showed that systematic errors of overestimation (underestimation) of rainfall
persist in regions such as the Amazon, Southeast Brazil, South Brazil, and northern South
America (Northeast Brazil and southeastern South America). Similarly, temperature
underestimation (overestimation) errors in most of the continent (Northeast Brazil and
southeastern South America) remain in the real-time forecasts. Overall, it is concluded that the
ECMWF-SEAS5 model performs seasonal rainfall and temperature predictions for South
America with considerable dexterity and potential for diverse applications. However, its
limitations and errors must be considered for the best use of its predictions.