Operational ExtraTropical Cyclone Tracks Verification System Tracking sorting

Operational Extra-Tropical Cyclone Tracks Verification System Tracking, sorting and matching model analysis and its corresponding forecast cyclone tracks for verification Geoff Di. Mego, Guang Ping Lou, Tim Marchok NCEP/EMC/MMB

1. The ideas Problems: • NCEP/EMC existing operational tracker systems put out tracks in ATCF format. In order to do tracks verification, the “truth” is needed. In the tropical case, HPC puts out storm vitals, the best observed tracks as the “truth”. However, there is noneexistence in the extra-tropical regions. Operationally, it is laborious at best, and almost impossible to manually identify all cyclone tracks in the middle latitudes. Solutions: • Utilization of existing resources to create quasi-truth for verification. That is, model analysis tracks are used as the “truth”, which are embedded to the tracks output files. Stringing them together will form the analysis tracks. Schematic Diagram:

00 hr Forecast 00 hr 00 hr 00 Z 24 hrs 12 Z 00 Z 48 hr 12 Z 00 Z 72 hrs 12 Z 00 Z 96 hrs 12 Z 00 Z 120 hrs 12 Z 00 Z 24 hr 48 hrs 72 hrs 96 hr 120 hrs Verification Extra-Cyclone Tracks Verification: 1. Extra-Tropical cyclone verification will use analysis as the truth since observed cyclone tracks are not available. 2 Therefore the tcvitals are already contained in the output of the forecast tracker after it completes the tracker processing. The first three records in the output file are the analysis position and intensity of the cyclone. 3 Position, center pressure, and/or direction, and wind strength for each quadrant are possible variables for verification every 12(24) hours. Mean errors of these variables can be calculated in the interval of 12(24) hours on a monthly bases and/or continuously (moving average).

2. The Steps a) Tracking: Operational tracker is invoked whenever model forecast data arrive at NCO. The forecast models include GFS, NAM, CMC, UKMO, ECMWF, NOGAPS, and ensemble forecasts such as NCEP ensemble and SREF, CMC and ECMWF. Cyclone tracks are plotted and graphics are pushed to a web server. They are also archived in ATCF format. http: //www. emc. ncep. noaa. gov/mmb/gplou/mchurr/nwprod/

b) Sorting: 1) 2) 3) 4) Model analysis ATCF data are picked out from the forecast tracks. Search nearby cyclone center in the next synoptic time. The criteria for recognizing as the same cyclone are: 3. 5 by 4. 5 lat and lon in middle to low latitudes; 4. 5 by 5. 5 lat and lon in high latitudes. Set aside preliminary storm tracks. Search for cyclones that are weakened and then re-appeared within 12 hours in the vicinity areas. Another iteration that string fragmented analysis storm tracks to form a unique track that will span the cyclogenesis to die out.

c) Matching: 1) 2) 3) Separate each forecast track into independent files. Search for analysis cyclone track that matches the forecast track. Combine the forecast and analysis tracks into one file. Operationally, previous 10 day’s cyclones are lined up for verification.

3. Examples From 12 Z March 26 to 12 Z April 5, there a total of 24 traceable extra-tropical cyclones that are automatically searched out within the operational tracking domains. Of these 24 cyclones, 86 forecasts were made. Top panel: GFS forecast tracks Bottom panel: analysis tracks in Alaska and vicinity

GFS forecast (left) and analysis (right) tracks in 10 day span in North Atlantic region (12 Z March 26 – 12 Z April 5, 2008)

4. Verification Following conventional tropical storm verification procedure, cyclone tracks are arranged in two separate decks: a) b) c) Adeck, contains past 10 day’s forecast tracks. Bdeck, holds analysis tracks that match the forecast’s. Verification is done for both track distance errors and intensity errors amongst other stats. In the examples shown previous slides, track errors are calculated as displayed next.

Average forecast distance errors for tracks in 10 day span

Verification output table Average track errors (NM) for homogeneous sample fcst hrs 00 12 24 36 48 72 GFS 0. 0 24. 2 54. 5 70. 6 96. 8 0. 0 #Cases 89 60 36 22 10 0 96 0. 0 0 120 0. 0 0 Error standard deviation fcst hrs 00 12 GFS 0. 0 24. 7 24 59. 6 36 53. 2 48 76. 6 72 0. 0 96 0. 0 120 0. 0 Average Xbias (NM) fcst hrs 00 12 GFS 0. 0 -1. 0 24 -10. 7 36 -6. 8 48 28. 4 72 0. 0 96 0. 0 120 0. 0 Average Ybias (NM) fcst hrs 00 12 GFS 0. 0 -4. 8 24 -10. 3 36 -26. 7 48 -64. 8 72 0. 0 96 0. 0 120 0. 0

5. Discussion • • • In the first stage, the operational track verification will be performed on those tracks that are initiated on the analysis cyclones for past 10 days. Due to model difference, tracks from forecast model will only be verified against its own analysis tracks for now. Matched tracks will be archived for further statistical calculations (such as monthly, seasonally and regionally, etc. ) The plots will be pushed to a server. Forecast and analysis tracks are not well matched even in the same model. This is a challenge in verifying all models against one analysis (“truth”). The "unified" sfc analyses produced every 6 hours amongst OPC, HPC and TPC would soon be put into gridded form, providing realistic “observed” cyclone tracks. This may be the “best” hope for middle latitude cyclones, but limited regions?