An Investigation of North Atlantic TC Ensemble Forecasts with Large Cross-track Errors

Hurricane Irma caused catastrophic damage to the eastern Caribbean islands and later made landfall over western Florida on 1935 UTC 10 September 2017. While the GEFS track errors within 5 days of landfall were smaller than climatology, the model consistently placed the TC center over eastern Florida. On 20 September 2017, Hurricane Maria devastated Puerto Rico, bringing island-wide power outages. The GEFS accurately predicted Maria’s landfall, but had right-of-track biases when forecasting the day 3-5 positions afterwards. These two cases raise questions of whether the GEFS has a right-of-track bias for strong TC’s in deep tropical environments. If so, what are the sources of this bias and can they explain past cases with large cross-track errors?

This study verifies the 2008-2016 TC track forecasts of global ensembles, focusing on the GEFS (21 members) ensemble for the North Atlantic. The NHC’s best-track data is used as the verifying analysis. The largest ensemble mean cross-track errors from each day 3-5 forecast are analyzed. The forecasts are defined as “north” (“south”) if the verifying best-track (never) crosses north of 30N, thereby separating TC’s potentially undergoing extratropical transition. For each set separately, the top 20% most negative and most positive cross-track errors are considered “left” and “right” cases, respectively. The CFSR gridded fields are used to identify common flow patterns in these cases and to verify the model forecasts. The relationship between model fields and cross-track errors are further assessed using ensemble sensitivity analysis.

The day 3-5 GEFS right-of-track bias is significantly different from 0 for south TC’s and not for north TC’s. There is a statistically significant correlation between the cross-track biases and initial intensities among cases, with major hurricanes tending to have right-of-track errors. Preliminary analysis of Irma suggests that the initially strong TC in the GEFS rapidly decayed within the first day. During this time, the GEFS excessively weakened the western edge of a downstream ridge, which then incorrectly steered the TC further northeastward.

Additional work will determine how many past right-of-track cases are also associated with the model weakening the steering ridge due to feedbacks from the TC’s diabatic circulation. In both left and right cases, the upper-level vorticity will reveal if the model TC’s are also being excessively sheared, which can affect the track by either displacing the upper-level anti-cyclone or by lowering the TC’s depth. The impact of errors in the propagation of Rossby waves on the steering environment will also be assessed. It will then be determined how many cases are in inherently complex flow regimes, such that a slight initial drift in the TC will greatly impact its subsequent trajectory.