Dynamic Downscaling Using the RegCM Model for Different Initializations Using CFSv2 Data

Abstract

The main objective of this study was to evaluate the regional climate forecasts of precipitation over Brazil during the winter season of 2018 through the RegCM4.7 model. It was used with different initializations, both spatial and in five specific areas. To emit alerts of possible below/above climate normal anomalies is necessary the verification these model abilities in predicting precipitation. The RegCM4.7 model was used with Global Climate Forecast System Version 2 global model data. The forecast quality was evaluated qualitatively and quantitatively, comparing its results with Climate Prediction Center (CPC) analysis data. The RegCM4.7 model was able to predict precipitation consistently a few months in advance for the june, july and august quarter (JJA), with minor mistakes over the northeastern and southeastern Brazil. However, the biggest errors were identified over the northern and southern regions. Prediction correlations were less than 0.8 during every experiments and subdomains, except for the Northeast region that presented the highest correlation values. In general, it stands out that RegCM4.7 was able to predict the spatial distribution of precipitation in advance over every domain, but with a tendency to underestimate what was observed.

Keywords:
RegCM; CFSv2; weather forecast; dynamic downscaling