Beginning in December of 2015, a new version of the National Centers for Environmental Prediction's (NCEP's) Global Forecast System ensemble model (GEFS) becomes operational. The GEFS, a 21 member ensemble, consists of a control run and 20 members using perturbed initial conditions, running at a lower spatial resolution than the deterministic version of the Global Forecast System. The new GEFS (NewGEFS) will run on a spectral grid that effectively doubles the spatial resolution of the previous version, going from a spectral resolution of T254 (about 55 km for the first 8 days of model integration, to TL574 (about 33 km). Vertical resolution increases from 42 levels to 64 levels. The physics scheme is improved and an Ensemble Kalman Filter scheme is used to initialize the ensemble. An overview of the changes is available online (http://www.nws.noaa.gov/os/notification/tin15-43gefsaad.htm). To allow an evaluation of the capabilities of the new model, retrospective model runs were performed for selected tropical storms from 2012 (6 storms), 2013 (7 storms), and 2014 (6 storms).
Output from these retrospective runs consists of the initial and forecast positions of a given tropical storm, and include the highest 10 meter wind speed in knots and the lowest central sea-level pressure. The output from the ensemble model includes the ensemble mean (AEMN), the interpolated ensemble mean (AEMI), and the forecasts for the individual ensemble members (AP01, AP02, etc.)
This paper presents an analysis of the track and intensity errors of the ensemble mean, as well as those of the individual ensemble members for the parallel ensemble model runs. As an example, statistics for Hurricane Cristobal (2014) are shown here. Cristobal became a tropical depression on August 23, 2014 at 18 UTC. The best track is shown in Figure 1. Cristobal became a hurricane by 00 UTC, August 26, and strengthened to 70 knots by 00 UTC on the 27th, weakened briefly to 65 knots and then strengthened again to 75 knots by 18 UTC, August 28. Cristobal underwent extra-tropical transition by 12 UTC, August 29, and weakened to below tropical storm strength after 06 UTC, September 2, 2014. Twelve parallel model runs were performed for Cristobal. Official National Hurricane Center storm locations and intensities ended at 06 UTC, September 9, which limited the error statistics for the longer range forecasts.
The number of forecast verifications is shown in the table.
Average track errors (nautical miles) were as follows:
# Forecasts GEFS NewGEFS NewGEFS GEFS
12 Hours 12 23 29 -6
24 Hours 12 44 53 -9
36 Hours 12 79 76 +3
48 Hours 11 108 98 +10
72 Hours 10 119 123 -4
96 Hours 9 141 162 -21
120 Hours 7 287 255 +32
Intensity errors (max 10 meter wind speed in knots) were as follows (Forecast Official NHC):
GEFS NewGEFS
12 Hours -17 -17
24 Hours -16 -17
36 Hours -17 -16
48 Hours -17 -17
72 Hours -13 -14
96 Hours -13 -14
120 Hours -13 -14
Date of Model Run Max GEFS Member Wind Max NewGEFS Member Wind
08/24/2014 00 UTC 76 60
08/24/2014 12 UTC 88 74
08/25/2014 00 UTC 80 80
08/25/2014 12 UTC 73 75
08/26/2014 00 UTC 75 80
08/26/2014 12 UTC 88 79
08/27/2014 00 UTC 81 73
08/27/2014 12 UTC 65 73
08/28/2014 00 UTC 80 64
08/28/2014 12 UTC 65 72
08/29/2014 00 UTC 57 58
08/29/2014 12 UTC 58 59
Average difference, NewGEFS GEFS: -3
Previous research suggests that high spatial resolution is needed to properly represent the intensity of a tropical storm, and research also shows that the intensity of a tropical storm can affect its track. Nagato et al. (2007), reporting on results from the Third Comparison of Mesocale Prediction and Research Experiment, noted that when the grid spacing was reduced from 50 km to 20 km, the corresponding tropical cyclone intensity increased significantly. Fierro (2009) conducted modeling experiments for Hurricane Rita (2005), and shows that as the horizontal grid spacing of an inner nest decreases from 30 km to 20 km, the minimum sea-level pressure was lower by 5 18 hPa at all lead times beyond 12 hours. Davis and Bosart (2002) note that intensity predictions from numerical models show notable improvement as grid spacing decreases to roughly the radius of maximum wind which averages around 45-50 km. Thus, with the higher spatial resolution, one could expect that the intensity errors for the NewGEFS would be smaller than those for the GEFS.
In this example, the track errors are worse at five of the seven times shown for the NewGEFS. The intensity errors are almost unchanged, and the NewGEFS is not producing more intense ensemble members, despite the much higher spatial resolution. Similar statistics for all of the parallel runs from 2012, 2013, and 2014 will provide useful guidance to forecasters who will be using the NewGEFS.
References:
Davis, Christopher and Lance F. Bosart, 2002: Numerical Simulations of the Genesis of Hurricane Diana (1984). Part II: Sensitivity of Track and Intensity Prediction. Mon. Wea. Rev., 130, 11001124.
Fierro, Alexandre O., Robert F. Rogers, Frank D. Marks, and David S. Nolan, 2009: The Impact of Horizontal Grid Spacing on the Microphysical and Kinematic Structures of Strong Tropical Cyclones Simulated with the WRF-ARW Model. Mon. Wea. Rev., 137, 37173743.
Nagata, Masashi, Lance Leslie, Yoshio Kurihara, Russell L. Elsberry, Masanori Yamasaki, Hirotaka Kamahori, Robert Abbey Jr., Kotaro Bessho, Javier Calvo, Johnny C. L. Chan, Peter Clark, Michel Desgagne, Song-You Hong, Detlev Majewski, Piero Malguzzi, John McGregor, Hiroshi Mino, Akihiko Murata, Jason Nachamkin, Michel Roch, and Clive Wilson, 2001: meeting summary: Third COMPARE Workshop: A Model Intercomparison Experiment of Tropical Cyclone Intensity and Track Prediction. Bull. Amer. Meteor. Soc., 82, 20072020.
Pasch, Richard J, 2014: Hurricane Cristobal (AL042014). National Hurricane Center Tropical Cyclone Report, February 11, 2015, 14 pages.
Figure 1. Best Track for Hurricane Cristobal, 2014. (Pasch, 2014).
- Indicates paper has been withdrawn from meeting
- Submission entered in competition