Enhancing Extended Weather Forecasts in the TCWAGFS Model Using Deep Learning Method for SST Bias Correction

Abstract. The extended weather (> 7 days) and the seasonal climate predictions are highly dependent on the status of Madden Julian Oscillation (MJO) and El Niño-Southern Oscillation (ENSO). Both the evolutions of MJO and ENSO are found to be correlated to the anomalies of the global sea surface temperature (SST). To decrease the predicting SST bias (BiasSST) in a coupled ocean-atmosphere global model, we evaluate nine well-developed machine learning algorithms. By using the Bi-directional Long Short-Term Memory (Bi-LSTM) algorithm, it is found the bias can be reduced significantly. For example, the Root Mean Squared Error on Day 10 forecast (denoted as D10) is reduced to 0.01 K by the Bi-LSTM algorithm while the original bias is 0.38 K by the Taiwan Central Weather Administration Global Forecast System (TCWAGFS), of which the error is reduced by 97 %.