5A.6
A comparison of data assimilation methods for hurricane track prediction
Eugenia Kalnay, Univ. of Maryland, College Park, MD; and L. M. Leslie, M. S. Speer, L. Qi, S. K. Park, and Z. X. Pu
A number of previous studies by one of the authors (e.g., Leslie and LeMarshall, MWR98) have shown that dramatically reduced errors in hurricane track prediction can be obtained using a combination of high resolution data from in situ, aircraft and satellite derived observations; the use of 4D variational data (4DVAR) assimilation over a 24 hour period (from t=-24 to t=0) leading up to the start of the forecast period (t=0); and a state-of-the-art NWP model run at very high resolution (10 km or less). Promising results have also been obtained for hurricane intensity predictions and structure (for example, rainbands and eyeformation and behavior). However, a major problem facing this procedure is the prohibitive expense of the 4DVAR scheme.
In this study, the quasi-inverse 3D variational data assimilation scheme(I3DVAR, Pu et al., MWR, 1997, Kalnay et al., MWR, in press)is compared with the 4DVAR scheme for 40 hurricane forecasts out to 72 hours, following a 24 hour assimilation period. We found that the I3DVAR scheme yielded predictions that were not statistically significantly different from those obtained from 4DVAR. However, the I3DVAR assimilation was a factor of eight faster than the 4DVAR assimilation. These results have profound implications for real-time hurricane forecasting.
Session 5A, Adaptive Observing Systems and Data Assimilation II (Parallel with Sessions 5B and J2)
Wednesday, 24 May 2000, 10:15 AM-12:00 PM
Previous paper Next paper