Further Work on the Prediction of Northeast Brazil Rainfall Anomalies

View More View Less
  • 1 Department of Meteorology, University of Wisconsin—Madison. Madison, Wisconsin
© Get Permissions
Full access

Abstract

This study expands our earlier climate prediction work for Brazil's Nordeste to develop methods of forecasting the March–June precipitation with differing lead times by exploring the potential of various data sources and options of information extraction. Observations include indices of Nordeste rainfall, an index of sea surface temperature (SST) in the equatorial Pacific, and the fields of meridional wind component and SST in the tropical Atlantic. Empirical orthogonal function (EOF) analysis was applied to construct indices of the meridional wind component and SST. These series formed the input to stepwise multiple regression models, an experimental neural network model, as well as to linear discriminant analysis. The dependent dataset 1921–57 (excluding 1943–47) was used for the method development, while the independent dataset 1958–89 was reserved for prediction.

Of primary interest is the prediction of March–June rainfall from information through January. A new SST dataset with improved quality control proved useful, especially with the EOF analysis confined to the more sensitive portion of the tropical Atlantic. The cardinal predictor is the preseason rainfall. Using indices of the Atlantic meridional wind and SST fields in conjunction with this allows one to predict half to three-fourths of the interannual rainfall variability in the independent dataset. Regarding predictions with greater lead times, about one-fourth of the variance of March–June precipitation can be forecast from the Atlantic SST field in December. Prediction from the February meridional wind and SST fields and preseason rainfall yields no improvement over the forecasts based on information through January. With similar skill, the April–June rainfall is predictable from the end of March. Experiments with neural networking revealed no advantage over regression. Linear discriminant analysis performed best in forecasting extremes.

The essential input information for Nordeste climate prediction consists of accumulated regional rainfall and quality-controlled databases of the Atlantic meridional wind and SST fields, as well as equatorial Pacific SST. For an operational prediction system three phases are found realistic: an early warning from the December SST field; the main forecast of March–June precipitation from rainfall, meridional wind, and SST information by the end of January; and a prediction for the April–June tail of the rainy season based on corresponding information through March. The remarkable recent communal effort in updating datasets was crucial for a real-time forecast of the 1992 Nordeste rainy season.

Abstract

This study expands our earlier climate prediction work for Brazil's Nordeste to develop methods of forecasting the March–June precipitation with differing lead times by exploring the potential of various data sources and options of information extraction. Observations include indices of Nordeste rainfall, an index of sea surface temperature (SST) in the equatorial Pacific, and the fields of meridional wind component and SST in the tropical Atlantic. Empirical orthogonal function (EOF) analysis was applied to construct indices of the meridional wind component and SST. These series formed the input to stepwise multiple regression models, an experimental neural network model, as well as to linear discriminant analysis. The dependent dataset 1921–57 (excluding 1943–47) was used for the method development, while the independent dataset 1958–89 was reserved for prediction.

Of primary interest is the prediction of March–June rainfall from information through January. A new SST dataset with improved quality control proved useful, especially with the EOF analysis confined to the more sensitive portion of the tropical Atlantic. The cardinal predictor is the preseason rainfall. Using indices of the Atlantic meridional wind and SST fields in conjunction with this allows one to predict half to three-fourths of the interannual rainfall variability in the independent dataset. Regarding predictions with greater lead times, about one-fourth of the variance of March–June precipitation can be forecast from the Atlantic SST field in December. Prediction from the February meridional wind and SST fields and preseason rainfall yields no improvement over the forecasts based on information through January. With similar skill, the April–June rainfall is predictable from the end of March. Experiments with neural networking revealed no advantage over regression. Linear discriminant analysis performed best in forecasting extremes.

The essential input information for Nordeste climate prediction consists of accumulated regional rainfall and quality-controlled databases of the Atlantic meridional wind and SST fields, as well as equatorial Pacific SST. For an operational prediction system three phases are found realistic: an early warning from the December SST field; the main forecast of March–June precipitation from rainfall, meridional wind, and SST information by the end of January; and a prediction for the April–June tail of the rainy season based on corresponding information through March. The remarkable recent communal effort in updating datasets was crucial for a real-time forecast of the 1992 Nordeste rainy season.

Save