Ensemble-based data assimilation for cloud-resolving hurricane prediction: experiments with radar and dropsonde observations from RAINEX

A WRF-based ensemble Kalman filter (EnKF) data assimilation and forecast system is used for assimilating both simulated and field observations during RAINEX to explore the dynamics, observability and predictability of hurricanes. We focus on assimilating sounding and radar data from Hurricane Katrina during different stages of its life cycle when intense field missions were conducted. One stage of particular interest is when the storm strengthened from tropical a storm to a category 1-2 hurricane before and after its first landfall in southeast Florida on 25-26 August 2005; the other is on late 27 August just before the category 3 hurricane experienced rapid intensification. The NOAA 43 P3 aircraft (N43) took observations for a duration of 8 hours before Katrina made her first landfall. The aircraft collected well-positioned airborne Doppler and dropsonde observations and documented the broad circulation of the rapidly intensifying tropical cyclone. In the meantime, Katrina was within the range of the KAMX and KBYX WSR88D radars. Another two-aircraft mission was conducted during the asymmetric phase of Katrina before her rapid intensification. The NOAA N43 and the NRL P3 aircrafts simultaneously took Doppler radar measurements during the afternoon of August 27.

In addition to the data processing and quality control provided by the RAINEX science team, we have developed data processing and quality control tools capable of assimilating both ground-based and airborne radar raw radial velocity observations. It is found that assimilation of the KAMX and KBYX radar radial velocity observations significantly improves subsequent forecasts both in terms of initial and forecast track and intensity. Assimilation of the airborne Doppler velocity is equally promising: we were able to “hot-start” a category-3 hurricane at near the observed strength without any ad-hoc bogusing. We plan to perform both deterministic and ensemble forecasts using the ensemble analysis from EnKF for examining the hurricane predictability and observing system design. We will also be conducting observing system experiments with both simulated and real-data observations (OSEs or OSSEs) to provide guidance in assessing future observing needs, observing strategies and observing network design. We will further use OSSEs to examine the effectiveness of ensemble sensitivity analysis for targeting purposes. This will have strong implications on the observing strategies of the upcoming TCS08/TPAC field experiments.