Analysis of a Multi-Model Ensemble to Assess the Predictability of Tropical Cyclone Tracks With Large Errors over the Western Atlantic

Analysis of a Multi-Model Ensemble to Assess the Predictability of Tropical Cyclone Tracks With Large Errors over the Western Atlantic

Nicholas Leonardo and Brian A. Colle

¹ School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, NY

            Track forecasts for North Atlantic tropical cyclones (TC) have significantly improved over the decades; however, some TC’s, such as Joaquin (2015) and several others over the last few years, have still proven to be significantly less predictable during the medium range (day 3-5).  These cases can be especially problematic given the growing confidence and expectations in model guidance. The question is whether some of these large track error events are associated with an identifiable error in the large-scale flow early in the forecast or a systematic bias in the models, both of which would suggest some practical predictability to be gained, or is there some intrinsic predictability limit for a certain subset of events. We hypothesize that there is still more practical predictability remaining for many of these large track error events.

            This study analyzes the common features among large track error forecasts to day 5 made from 2008-2015 using a multi-model ensemble (51 ECMWF members, 23 UKMET members, and 21 GEFS members) for the western North Atlantic.  The multi-model ensemble mean track forecasts are ranked from best to worst for each forecast lead-time using NHC’s best track data as a verifying analysis.  The top and bottom fifth percentiles are categorized as large and small track error cases, respectively.  The ensemble model fields on the large-scale are assessed and compared with the CFSR reanalysis in a vortex-centered framework (centered around the ensemble mean position).  The model fields from the 10 members with the largest track errors (BAD) are compared with the 10 members with the smallest track errors (GOOD) as well as the corresponding analysis.  Ensemble sensitivity analyses are also performed on the ensemble fields using the day 5 track error as the response function.

            The large track error events for days 3-5 are most commonly associated with recurving TC’s just east of the U.S. East Coast, with the initial points for these forecast tracks clustered over and east of Puerto Rico.  The large-scale flow with these cases is more amplified and sometimes more complex than small track error cases. Many of the large track error events grow exponentially beyond day 3 in response to differences in the phase and amplitude of an approaching upstream trough as well as the downstream ridge amplitude.  The difference between the BAD and GOOD members is often reflected in differences in the mid-latitude Rossby wave guide, leading to some of these trough/ridge phasing issues. Additional results will be presented separating the cyclone relative evolution for cases with well-defined large-scale errors versus smaller-scale errors, as well as quantifying how much the TC intensity in the various ensemble members (which can be dramatically underpredicted at times) relates to track error.