Medium-Range tropical cyclone track prediction as a data assimilation problem

Medium-range tropical cyclone (TC) track prediction (72-h position) is the linchpin of the operational forecast product. It is the target point towards which current TC knowledge and observations are advanced in time, and the starting point for the extended-range outlook (day + 4/5). Moreover, the 72-h forecast is challenging because the synoptic-scale environment can undergo large changes in three days and sensitive environment-TC interactions can lead to a wide variety of tracks. More importantly, the primary scientific basis of the medium-range TC track forecast is numerical weather prediction and it is the skill of the models that largely determines the quality of the official track forecast. Thus, detailed verification of the 72-h track forecast is a key diagnostic in assessments of model and official skill.

In this paper, medium-range TC track prediction is viewed as a data assimilation (DA) problem, i.e., a process that corrects a background (first-guess) using observations and modeling to form the analysis. For medium-range track forecasting, the background is the previous official forecast (6-h forward-in-time interpolation), the observations are the model/consensus tracks, and the analysis is the official track forecast. The formalisms of the DA problem are not addressed; rather, the errors are evaluated in the context of DA to give a different perspective on forecast error. The main finding is that mean forecast error, a measure of net skill, is often sensitive to a small number of critical cases where the model guidance fails/succeeds in unexpected ways. Identification and anticipation of these critical cases can be helpful to both forecasters and modelers and the DA-based diagnostic will be demonstrated for cases in the 2006/07 seasons.