Real-time Cloud-resolving Ensemble Analysis and Forecast Assimilating Airborne Doppler Radar Observations during the 2008/2009 Atlantic Hurricane Seasons

The ensemble-based data assimilation, commonly known as ensemble-Kalman filter or EnKF, has recently been demonstrated to be an effective and maturing assimilation technique with simulated and real observations for NWP across a range of scales. Building on our recent success in assimilating both ground-based and airborne Doppler radar observations for hurricane initialization and prediction, and under the auspices of Hurricane Forecast Improvement Program (HFIP) along with support from ONR and NSF, we developed a cloud-resolving ensemble analysis and forecast system that is capable to assimilate the NOAA P3 airborne Doppler radar observations and to provide both deterministic and probabilistic hurricane forecasts in real time.

This prototype future hurricane prediction system is implemented in massively parallelized high-performance computing facilities at the Texas Advanced Computing Center (TACC). It was first tested in real time for selected storms during the 2008 Atlantic hurricane season than include Tropical Storm Fay and Hurricane Gustav and Ike that showed great promise in forecasting performance of both track and intensity. As part of the real-time demonstration project of HFIP, this cloud-resolving ensemble analysis and forecast system, with the finest grid spacing of 4.5 km, was performed quasi-operationally for all 2009 Atlantic storms during which at least one airborne Doppler mission was conducted.

The encouraging real-time performance of this prototype future ensemble analysis and prediction system will be presented. It represents the first time that airborne Doppler radar observations are assimilated into hurricane prediction models, and that the cloud-resolving ensemble analyses and forecasts for hurricanes were done in real-time. Also unprecedented are the real-time coordination, parallelization, and on-demand usage of more than 23,000 computer cluster cores simultaneously. Moreover, besides providing flow-dependent analysis and forecast uncertainty, this prototype system showed significant advantages other deterministic operational forecasting models.