Hybrid Gain Data Assimilation in the Taiwan Global Forecast System (TGFS)

The Hybrid Gain Data Assimilation (HGDA) method presents a different flavor of hybrid data assimilation, distinct from the traditional covariance hybridization technique. HGDA can be implemented with two different update scenarios to combine the Ensemble Kalman Filter (EnKF) and the Variational (VAR) sub-systems. The HGDA scenario a (HGSA) is a two-step update scenario that allows VAR to directly correct the EnKF analysis ensemble mean state, which aims to complement the corrections not captured by EnKF in the 2^nd VAR step. The HGDA scenario b (HGSB) uses the same background ensemble mean state to derive the analysis increment of the two sub-systems, making itself more feasible in practice, eliminating the necessity of sequential execution for the sub-systems.

In this study, we implement HGDA method in Central Weather Administration (CWA)'s newly-launched Taiwan Global Forecast System (TGFS), which is developed based on the National Centers for Environmental Prediction (NCEP)’s FV3-based Global Forecast System (GFS) version 15. The performance of the two HGDA scenarios is evaluated and compared in a quasi-operational setup of the TGFS: The ensemble size of the EnKF is 32; The EnKF, VAR, and the hybridization are updated at a C192 resolution (about 50 km), followed by a 120-hour deterministic forecast conducted at the same resolution. Preliminary results from one-month cycling experiments show that the forecast performance of HGSB is less sensitive to the hybridization weight. In contrast, the optimal hybridization weight for HGSA depends more on the performance of the EnKF, suggesting that the potential performance of HGSA could be limited if using a static hybridization weight. In addition, non-affine hybridization weights (i.e., with a sum greater than one) are examined in HGSB to fully utilize the EnKF information. Results suggest clear benefits of employing non-affine weights in HGSB.