Katrina, Rita and Wilma: Met Office model forecasts

In 2005 the Atlantic spawned an unprecedented three category 5 hurricanes, all of which struck the USA whilst still classified as major hurricanes. This paper reviews the forecast tracks produced by the Met Office global model during the crucial stages of these hurricanes leading up to landfall. It also reviews the performance of the experimental Met Office North American model which was run during Hurricanes Katrina and Rita.

Prior to landfall over southern Florida, numerical models were predicting that Hurricane Katrina would undergo sharp recurvature and a second US landfall over north-west Florida. However, the first forecast from the Met Office model after Katrina re-emerged over the Gulf of Mexico made a dramatic shift and predicted landfall in three days near New Orleans to an accuracy of 59km. Subsequent forecasts consistently predicted landfall over or very close to New Orleans.

Initially, model forecasts for Hurricane Rita predicted landfall over Texas. However, the Met Office global model made an eastward shift in its predicted track some 54 hours before landfall, correctly predicting the location of landfall over the Texas-Louisiana border with just a few hours error in timing. Subsequent forecasts maintained this more eastward track.

Hurricane Wilma proved to be a difficult forecasting problem with some wildly varying model guidance on possible landfall over Mexico and timing of recurvature and landfall over southern Florida. The majority of forecasts from the Met Office global model predicted the landfall and slow movement over north-eastern Yucatan peninsula very well. In addition, the north-eastward acceleration and landfall over southern Florida was also well predicted several days in advance.

The combined track forecast errors from the Met Office global model for the three hurricanes were 65, 121, 223, 409 and 674km at 24, 48, 72, 96 and 120 hours respectively. Mean Atlantic track forecast errors from the Met Office global model for the 5-year period 2000-2004 were 133, 231, 319, 412 and 577km. Whilst the 120-hour errors were disappointingly large (due to a few poor predictions of recurvature of Katrina and Wilma), the short range forecast errors (0-72 hours) were remarkably low.

The experimental Met Office North American model produced forecast tracks of Hurricanes Rita and Wilma which, on average, were not greatly different to the global model. However, the higher resolution of the North American model gave a much better representation of peak wind speed, central pressure and eye and rain band structure of these hurricanes.