| dc.description.abstract | Taiwan is located at north-western Pacific and is affected by typhoons often in the summer season. Accurate predictions of tropical cyclogenesis and track will be helpful for the disaster preparedness. Since tropical cyclones form over oceans, it is important to make use of the observations over oceans. FORMOSAT-7 radio occultation provides high-resolution observations over lands and oceans, which can improve the accuracy of initial conditions, and is helpful for predicting tropical cyclogenesis and track forecast. In this thesis, WRF hybrid data assimilation system has been used to assimilate seven typhoon cases in 2019 and 2020 (Lekima, Mitag, Hagibis, Hagupit, Halong, Hagupit, Maysak and Haishen). The experiments start 72 hours before observed cyclogenesis time, with the experiment that assimilates both GTS and RO data (WR) and the other that only assimilates GTS data (NR), and both experiments are conducted for 120-h forecast. The experiment is expected to find out if the assimilation of GNSS RO reflectivity data will give positive impacts on cyclogenesis prediction, and also analyze the specific cases with larger positive impact, to understand the dynamic processes leading to the differences in forecasted cyclogenesis.
The model results show that, although the average time error is not improved significantly in the WR experiment, the average location error is reduced by 8.9 km, and the number of successful cyclogenesis detection is more than that from the NR experiment. Most of the improvement by RO assimilation is present for the time error between 24 to 48 hours that is smaller and grows slower. For sensitivity tests, the experiments with the operational ensemble prediction from CWB for the initial background has significantly larger improvement than that from the control experiment (CTL), in particular when performing the blending with the large-scale analysis, leading to a reduction on the time error by 3.6 h and location error by 93.9 km. Other sensitivity tests for Lekima, Hagibis and Maysak show that use of the updated background errors has increased the prediction error. On the other hand, the prediction is improved in time error than the CTL experiment when the default of data thinning is deactivated.
For the two specific cases (Mitag and Hagupit), water vapor mixing ratio increases after 12-h assimilation, especially in lower and middle levels helpful for latent heat release. Although RO data assimilation only changes the water vapor mixing ratio and temperature, it will also modify the wind and vorticity after the model integration. This makes the WR experiment improve the cyclogenesis prediction one day earlier than the NR experiment. Through analysis of potential vorticity budget, the increase of diabatic heating indeed enhances the positive PV tendency, and vertical motions, and thus facilitates the tropical cyclogenesis with the stronger PV vertical advection. | en_US |