HRD seminar – Erin Munsell, PSU – 16 November 2015

Ms. Munsell presented a seminar “Dynamics and predictability of tropical cyclones evaluated through convection-permitting ensemble analysis and forecasts with airborne radar and sounding observations.


Dynamics and predictability of tropical cyclones evaluated through convection-permitting ensemble analysis and forecasts with airborne radar and sounding observations. The dynamics and predictability of tropical cyclones (TCs) are assessed through the initialization of convection-permitting ensemble forecasts generated by the Pennsylvania State University (PSU) real-time Atlantic hurricane analysis and forecast system, which employs the Weather and Research Forecasting (WRF) model coupled with an ensemble Kalman filter (EnKF) to assimilate airborne Doppler radar and sounding observations in near real-time. Divergent track and intensity forecasts amongst 60-member ensemble simulations of Hurricane Sandy (2012), Hurricane Nadine (2012) and Hurricane Edouard (2014) are analyzed to diagnose the sensitivity associated with the forecasts, as well as the causes of the significant divergence ensemble members are classified according to their forecast performance, creating composite groups whose different behaviors are explored through various compositing and ensemble sensitivity techniques.

Despite a complicated synoptic set-up, the majority of the members in the Hurricane Sandy (2012) ensemble correctly predicted Sandy’s eventual landfall in the Mid-Atlantic region of the United States, however, 10 members forecasted the TC to head out to sea (POOR). A comparison of the 10 most successful track forecasts (GOOD) to the POOR track forecasts revealed that the track differences arose from discrepancies in the strength of the westerly trade winds that the storms were initially embedded in. It was hypothesized that the stronger win the  GOOD members advected Sandy closer to the Florida coast before turning towards the and northeast  along the United States coastline. This displacement in vortex position allow Sandy to be close enough to an approaching mid-latitude trough in GOOD that a significant interaction between the two occurred, leading to a United States landfall. A similar synoptic set-up was present in the case of Hurricane Nadine (2012), which an extremely long-lived Atlantic TC. Track forecasts from this ensemble produced similar divergence to the Sandy ensemble; the majority of the members correctly forecasted that the southward traveling Nadine would turn towards the southwest ahead of a mid-latitude trough while 10 members were absorbed by the trough and carried eastward. Members were again classified according to their tracks (GOOD and POOR) and composite analyses demonstrate that the divergence in tracks could be attributed to differences associated with the strength position of the mid-latitude trough. In the POOR members, the mid-latitude trough was stronger and was more quickly advected towards the southwest than in GOOD, leading to the steering Nadine towards the east by the trough.

While the Sandy and Nadine ensembles focused on the predictability of TC track forecasts, the Hurricane Edouard (2014) ensemble was utilized to explore the predictability associated with TC intensity forecasts, particularly TC rapid intensification (RI). A significant percentage of ensemble members correctly underwent RI, however, there was significant variability in the RI onsets amongst the ensemble. Utilizing composite groups based on the onset times (or lack thereof), it was found that the members who undergo RI the earliest was able to do so because of initially stronger low-level vortices. In addition, the environment significant distance from the TC (up to 800-km to 900-km from the surface centers) is detrimental to development in the members who do not undergo RI, however, the extremely small differences between the developing and non-developing members are essentially impossible to diagnose. Although there is some potential for detecting whether a given member undergoes RI in the Edouard ensemble, the predictability of RI onset remains limited.

A copy of the presentation is available on the anonymous ftp site:

Munsell presentation
Munsell presentation