[UBC Climate Prediction Group Home Page]

Neural network is a fast-changing technique. Since January 1996, our neural network model has evolved through many versions, each with improved skill over the previous versions.

Version 1 used the FSU wind data as predictor

• Version 1.1: a "clearning" neural network model; used in the forecasts with data up to January 1996 - July 1996.

• Version 1.2: a neural network model with a single hidden unit; used in the forecast with data up to October 1996.

• Version 1.3: an ensemble neural network model using the "bagging" technique; used in the forecasts with data up to January 1997 - August 1997.

Version 2 uses the COADS SLP and NOAA's SST data as predictors

• Version 2.1: an ensemble neural network model using the "bagging" technique; used in the forecast with data up to October 1997.

• Version 2.2: an ensemble neural network model using a simple random initialization procedure; used in the forecasts with data up to December 1997, January 1998 and March 1998.

• Version 2.3: an ensemble neural network model using a simple random initialization procedure with EEOF predictors; used in the forecast with data up to May 1998.