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

While medium-range (day 3-5) track forecasts for North Atlantic tropical cyclones (TCs) have improved during the last three decades, numerical models occasionally have very large errors when forecasting certain TCs, such as Hurricane Sandy (2012) and Hurricane Joaquin (2015). Large cross-track errors can be especially problematic in that they can determine if or where the TC makes landfall. Understanding the commonalities of these events can help forecasters recognize situations when the model track guidance is prone to be significantly biased.

This study verifies the 2008-2016 North Atlantic TC track forecasts of the GEFS (21 members) and ECMWF (51 members) ensembles. The NHC’s best-track data is used as the verifying analysis. The day 3-5 forecasts are defined as “ET” (“non-ET”) based on the cyclone phase space diagram (Hart 2003) and whether the observed TC (never) crosses north of 30°N at any time in the forecast, thereby isolating cases not strongly interacting with mid-latitude baroclinic systems. The GEFS has a right-of-track bias for non-ET cases by 96 h. Focusing on the GEFS non-ET cases, the top 25% most negative and most positive cross-track errors are considered “Left” and “Right” cases, respectively. The relationship between model fields and cross-track errors in each of these cases are analyzed with various methods, including standardized differences between subsets of ensemble members.

Vortex-relative composites reveal that large environmental steering errors develop during the first 24 h. These steering errors are associated with a 700-hPa subtropical ridge north of the TC in more than half of the Right cases. The geopotential heights along the southwest extent of the ridge become more underpredicted with time. Variable resolution WRF ensembles were run for three of the Right cases involving ridges: Ike (2008), Isaac (2012), and Irma (2017). The WRF 36-km domain develops biases similar to the GEFS for these cases. The WRF runs reveal that the additional height falls along the ridge are driven by excessive upper-level divergence, which is associated with outer-core convection north of the TC. This extension in convection is triggered by a slight overprediction of initial conditional instability and moisture. The 4-km nests do not significantly reduce the growth of the errors and hence have little impact on the tracks of these cases. The range of processes for these relatively large cross track errors will be summarized in this presentation.