Hierarchical prediction of landfalling tropical cyclone tracks

Forecasting of tropical cyclone tracks is important yet difficult. Recent studies have focused on better assimilation of TC track and structure using the ensemble Kalman filter (EnKF), better TC initialization using vortex bogusing, bogus data assimilation, and relocation. There is good reason to think that with better assimilation of initial and boundary conditions and improvement in model architecture, the errors in TC predictions are systematic rather than purely random. In fact, it is realistic to theorize that better assimilation of initial and boundary conditions and improvement in model architecture will make the “total” error more systematic, albeit not necessary smaller. This perception has indeed been corroborated by our experiment using the NCEP GFS as the initial and lateral boundary conditions, where we examined the effectiveness of a physics-based Advanced Research WRF (ARW) ensemble run at 36- and 12-km resolution on the predictability of Hurricane Ike (2008). Even though the ARW ensemble yielded improved cyclone intensity, the tracks retained similar bias and spread as those of the NCEP ensemble forecast.

In this talk, we propose a hierarchical scheme to improve the prediction of TC tracks. The scheme works by first making the track prediction errors more systematic and then fully off-setting the systematic errors with a transformation. If the transformation only contains a few simple parameters, then a simulated track can be used to reliably predict the observed track. Such a scheme will be valuable for combining multiple predictions of a single model, as well as combining ensemble forecasting of different models.