Influence of Assimilating Satellite-derived Atmospheric Motion Vectors (AMVs) on Analyses and Forecasts of Tropical Cyclone Track and Structure

Given that tropical cyclones (TCs) spend most of their lifetimes over the ocean, their forecasts are crucially dependent on the accurate assimilation of satellite data. In order to provide an accurate representation of the TC structure, assimilation of high-resolution satellite data into mesoscale models is necessary. One example of such data is cloud-derived Atmospheric Motion Vectors (AMVs) from geostationary satellites. Operationally, the AMVs are derived from sequential images every 30 minutes, and these are processed hourly by CIMSS. When a TC is a potential threat to society, a ‘rapid-scan' mode is activated, yielding even more frequent sequential images and therefore a higher volume of AMVs. In this study, the influence of assimilating hourly and rapid-scan AMVs into the Weather Research and Forecasting (WRF) model using the Ensemble Kalman Filter (EnKF) is investigated for the case of Typhoon Sinlaku (2008) during its period of intensification.

Three EnKF cycles, using 32 ensemble members with a resolution of 27 km on the analysis grid (and 9 km in the forward forecast model), are prepared. A ‘Control' EnKF cycle is first produced with the assimilation of conventional observations (without radiances), but with the experimental CIMSS AMV data excluded. Next, a parallel EnKF cycle (‘CIMSS hourly') that includes these AMVs is computed over the life cycle of Sinlaku. Finally, a third parallel EnKF cycle (‘CIMSS hourly+RS') that includes both the hourly and rapid-scan CIMSS AMVs is produced. Rapid-scan AMVs were available during the second half of Sinlaku's life cycle.

In comparison to the ‘Control', the ‘CIMSS hourly' ensemble analyses generally produce lower track errors and better capture the rapid intensification. The ‘CIMSS hourly' analyses exhibit a deeper warm core and stronger low-level vorticity and convergence than the ‘Control', with values of mean sea level pressure that are generally more consistent with the best track. Analyses with respect to QuikSCAT winds and NRL P3 ELDORA radar wind will also be presented. A comparison of the respective steering flows reveals that the ‘CIMSS hourly' analyses possess a generally more accurate northwestward flow, while the flow in the ‘Control' analyses is typically more westward. Forecasts are conducted by initializing with the three analyses with different initial times for the first 20 members. Three initial times chosen are 00 UTC 9 Sept (FT09), 00 UTC 10 Sept (FT10) and 00 UTC 11 Sept (FT11) at 2008. ‘CIMSS hourly' has lower track error below 200 km in the first 48 h for FT09 and FT10. In FT11, ‘CIMSS hourly+RS' exhibits the re-curvature toward Japan after making landfall at North Taiwan on 13 Sept 2008, while ‘Control' and ‘CIMSS hourly' keep moving westward and make erroneously landfall at southeast of China.