Session 18A.2 Prediction of Atlantic tropical cyclones with the Advanced Hurricane WRF (AHW) model

Friday, 2 May 2008: 10:30 AM
Palms GF (Wyndham Orlando Resort)
Jimy Dudhia, NCAR, Boulder, CO; and J. Done, W. Wang, Y. Chen, Q. Xiao, C. Davis, G. Holland, R. Rotunno, and R. Torn

Presentation PDF (799.1 kB)

The Advanced Hurricane WRF (AHW) is a derivative of the Advanced Research WRF model. The use of a moving, 2-way nested grid system allows local resolution of roughly 1 km, making it ideal for the prediction of the multiple length scales present in hurricanes ranging from the scale of outflow (1000 km or more) to the sharp gradients inside the eye wall (5-10 km). During the past four Atlantic hurricane seasons, the AHW model was run in real time and in retrospective mode to produce forecasts of hurricane track, intensity and structure out to several days lead time. During the 2004 and 2005 hurricane seasons, AHW performed comparably to operational models using an innermost nest of 4 km grid spacing, with evidence of improved intensity forecasts beyond 1.5 days during the 2005 season. During 2006, a second nest of 1.33 km grid spacing was added to resolve the eye wall of storms. In 2007, a mixed-layer ocean model was added to provide a feedback of mixing-induced sea-surface cooling to the atmosphere and forecasts were extended to five days.

The focus of this talk will be on the prediction of the magnitude and structure of the wind field for Dean, Felix and Karen, Noel and other storms from the 2007 hurricane season. The sensitivity of maximum wind forecasts will be investigated for changes in air-sea exchange coefficients and upper-ocean feedback. Other metrics of the wind distribution, such as the radius of maximum wind and integrated kinetic energy, are also evaluated. In the case of hurricane Dean, these metrics illustrate that the AHW realistically changes the vortex size and handles the collapse of the core at landfall, but does not replicate the fluctuations associated with eye-wall replacement cycles. Large initial transients generated through initialization with the GFDL model are shown to be substantially reduced through variational and ensemble-based initialization methods. The simulations from 2007 also indicate a relatively lower predictability of cyclones initialized in their earliest stages of formation compared with more mature depressions.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner