A scheme is presented to improve the representation of a tropical cyclone in the initial condition of a high-resolution mesoscale model. In the proposed method, a specified surface low, based on a few observed parameters, is incorporated into the model initial condition through a variational approach. The forecast model serves as a strong constraint during the process of incorporating the specified surface low into the model initial condition, allowing the other model variables (other than surface pressure) to be also adjusted. Other types of observations, when available, can be incorporated directly into such a procedure.
This variational initialization scheme was tested on a real data case using the Penn State/NCAR nonhydrostatic mesoscale model version 5 (MM5). The case tested is Hurricane Felix (1995) in the Atlantic ocean during its mature stage. Initialization experiments include assimilating the specified surface low data and the satellite-derived water vapor wind measurements. It was found that the initialization scheme produced an initial field which not only has a surface low much more realistic in size and intensity than the storm structure obtained from the NCEP global analysis, but also exhibits a robust vertical structure of divergence, heating, moistening, and vertical motion, which is absent or hard to obtain without variational initialization. In addition, the initial vortex was well-adapted to the forecast model. As a result, dramatic improvements occurred in the intensity forecast as well as the description of the inner-core structures of the predicted storm. Improvement in the track prediction, though not as significant as in the intensity forecast, was also observed. Using the new scheme, the simulated storm moved northwestward, northward, and northeastward, similar to the observed track. The 24-h, 48-h, and 72-h forecast error was 75 km, 80 km and 150 km, respectively, compared with 110 km, 220 km and 220 km for the non-initialized forecast starting from the global analysis. The averaged National Hurricane center operational forecast error at 24 h, 48 h, and 72 h was 75 km, 147 km and 227 km for the same case.
The model correctly forecasted the gradual weakening of Hurricane Felix during the initial 12 h. The simulated storm maintained 65-70 knot sustained winds and a central pressure near 970 mb during the period of 12-72 h, in good agreement with observations. As verified against cloud observations, the model captures reasonably well the inner-structures of the storm. In particular, the model reproduces the ring of maximum winds, the eye, the eye-wall, and the spiral cloud-bands. Since the initial vortex obtained by the proposed
method is well-adapted to the prediction model, problems of initial
adjustment and false spin-up of the model vortex are alleviated.