Promising EnKF analyses and Forecasts of Hurricane Humberto (2007)

Challenge: Hurricane Humberto, the first landfalling hurricane in the US since the end of the 2005 hurricane season, is the most rapidly intensifying near-landfall storm in US history. It strengthened from a 35-mph depression at 15Z September 12, 2007 to an 85-mph hurricane at 9Z September 13, a 50-mph increase in surface wind speed in only 18 hours. The storm caused extensive damage along the southeast Texas coast, and it was poorly predicted by operational models and forecasters. Post-event WRF simulations initialized with the NCEP GFS/FNL analyses running at cloud-resolving resolutions with lead times every 6 h from 6 to 48 h all failed to capture the rapid formation and intensification of the storm.

Findings: Ensemble-based data assimilation with an ensemble-Kalman filter (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. A WRF-based EnKF with model grid spacing down to 4.5 km is used to assimilate Doppler observations of three WSR-88D radars (KCRP, KHGX and KLCH) along the Gulf coast. Despite some sensitivity to the number of observations assimilated and to the radius of influence of a given observation, preliminary results show that assimilating the ground-based radar observations is promising for initializing the incipient convective activity associated with the developing cyclone. A deterministic forecast initialized with the EnKF analyses was fairly successful in predicting the rapid formation and intensification of the storm to a category 1 hurricane. On the other hand, a WRF ensemble forecast running at 4.5 km grid spacing and initialized with the EnKF analysis uncertainties shows significant uncertainty. The associated peak intensity forecast ranged from below tropical storm intensity to a category 2 hurricane with an ensemble spread of maximum surface winds of 10-15m/s, echoing the difficulties in the real-time forecast. Implications on our understanding of the dynamics and predictability of the storm using the cloud-resolving deterministic and probabilistic simulations will be discussed.