Application of breeding ensemble to tropical cyclone track forecasts using the Regional Atmospheric Modeling System (RAMS) model

In this study, a breeding ensemble technique that is designed specifically for tropical cyclone (TC) track forecasting is presented. By introducing two different scaling factors to represent two different growing modes for the storm-scale and the large-scale processes during the breeding cycling, it is shown that ensemble TC track forecasts with the Regional Atmospheric Modeling System (RAMS) model can be improved. Retrospective experiments with 14 TCs during the 2009-2011 seasons in the north Western Pacific basin show that the proposed TC-breeding (TCB) method could reduce the track forecast errors most significantly at 4-5 day lead times. Comparison of the track forecast bias between deterministic forecasts and TCB ensemble forecasts shows however that both the TCB ensemble and the deterministic forecasts possess similar pattern of the cross- and along-track forecast errors. This suggests that a significant component of the track bias in the RAMS model is determined by model inherent uncertainties that cannot be removed with the TCB method. Sensitivity experiments with different ensemble members show further that increasing the number of ensemble members could reduce the track forecast errors, but the rate of the track error reduction saturates when the number of ensemble members is larger than 30 due to the inefficiency of the TCB method in orthogonalizing bred vectors. While the TCB method cannot remove intrinsic model errors related to inadequate representation of model physics in the RAMS model or the model resolution, this method could optimize the use of the breeding ensemble for TC track forecasts in real-time forecasting systems with limited computational resources.