6.6
Tier-one Seasonal Prediction with CES Coupled GCM

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 1 February 2006: 11:45 AM
Tier-one Seasonal Prediction with CES Coupled GCM
A314 (Georgia World Congress Center)
Jong-Seong Kug, KORDI, Ansan, South Korea; and I. S. Kang and D. H. Choi

A tier-one seasonal prediction system is developed at CES/SNU. The prediction system consists of three steps of the initialization, prediction and statistical downscaling. In the present system CES coupled GCM is used for dynamical prediction. It is verified from long-term simulation that the CES CGCM has an ability to reasonably simulate the observed ENSO characteristics in terms of amplitude, pattern and seasonal locking. For initialization, the ocean component of the CGCM is integrated by prescribing observed SST and wind stress as a surface boundary condition. To get different initial conditions for ensemble prediction, we used three kinds of the wind stress data. Given three different ocean initial conditions, we carried out 6-month lead prediction, starting from each May 1st during year of 1960-2001. For three ensemble members, predictive skill is evaluated. The results show that the present system has a predictive skill for SST and precipitation over the tropical Pacific, though some systematic biases are found over the Indian Ocean. In addition, the predictive skill of the present one-tier system is compared to those of DEMETER and NCEP seasonal predictions for recent 20 years. To reduce systematic bias, new statistical downscaling model is developed. The predictor of the model is dynamical model prediction, and the predictand is SST and precipitation at each grid. The model procedure is stepwise. That is, firstly the model select good predictors from the dynamical model output. Then, final forecast is performed based on the selected predictors. The predictive skill of the statistical downscaling model is examined based on the cross-validation technique. Because we use relatively long prediction results of 42 years, the model is less sensitive to the data sampling in the cross-validation technique. After correction, the prediction skill of the present system is significantly improved by reducing the systematic bias of the dynamical model. In particular, the improvement over the Indian Ocean is distinctive. The dynamical model has a problem that the ENSO-related pattern is shifted westward compared to the observed one. This deficiency can bring about the low skill over the Indian Ocean. However, the statistical model can reflect the systematic bias, so that the predictive skill can be improved over the Indian Ocean.