A REGRESSION SCHEME FOR PREDICTING MONTHLY PRECIPITATION IN SOUTHERN BRAZIL

R. M. Vil'fand and V. M. Khan

A relationship between rainfall in Southern Brazil and sea surface temperature, air temperature, global precipitation in the Southern Hemisphere, and Southern Oscillation Index is investigated by using linear correlation analysis with time lags from 0 to 12 months. Areas of significant correlations (synchronous and asynchronous) between the first three parameters and precipitation are found. Particularly, significant correlations between precipitation in the study region and SST anomalies in the area of the Antarctic Circumpolar Current in some months were detected. For selected prognostic zones of SST and air temperature data the EOF analysis was applied as a reduction method. Stepwise multiregressional linear models were then constructed for forecasting monthly precipitation. The principal predictor was SST, which describes large-scale pro¬cesses. The evaluation of the quality of the models based on hindcast and crossvalidation analysis has demonstrated good agreement between predicted and observed precipitation.