| dc.description.abstract | This study focuses on two cases. The first case is Typhoon Haiyan on November 3, 2013, the second one is Typhoon Megi on October 12, 2010. These two typhoons lead to the significant disaster situation, so it is an important subject for the typhoon forecast. Because the traditional observations (Global Telecommunication System) mainly centralize on land, they are not numerous upon the sea. The observations upon the sea is usually obtained by using telemetering. The Global Positioning System Radio Occultation (GPSRO) observation is one of the examples. If we can combine the GPSRO observation with the numerical weather prediction (NWP), we will get the more accurate typhoon track forecast and the intensity, and might reduce more damage. In this study, WRF Hybrid Data Assimilation will be used to combine observations with the numerical weather prediction (NWP). We will use WRF Hybrid Data Assimilation to test GPSRO observation sensitivity and hybrid weighting sensitivity to the typhoon forecasting influence.
The result shows that assimilating GPSRO observation is better results than no assimilating GPSRO observation for these two cases during assimilation. Typhoon track forecast will be available within 96 hrs or so. On verification of the RMSE (Root mean squared error) and SCC (spatial correlation coefficients), the result shows that GPSRO observation significantly influence the water vapor field, the thermodynamic field and the dynamic field about 72 hrs, specially in the upper troposphere. WRF Hybrid Data Assimilation is also significant improvement in these two cases. When we add in ensemble background error covariance, typhoon track forecast will be closer to the JTWC best track, and using 50% ensemble background error covariance is the best for these two cases. After verifying the Rmse (Root mean squared error) and Scc (spatial correlation coefficients) and comparing with and without the ensemble background error covariance, we found that the thermodynamic field with te ensemble background error covariance is better. The result shows that 50% to 75% ensemble background error covariance is better. After the spin-up time testing of other two cases, we found that results are not satisfactory. Last, we did the localization experiment that 800 km radius case showed the best result in forecast performance.
| en_US |