On Tuesday morning, 21 August, Tropical Depression (TD) nine developed and was located around 500 miles east of the Lesser Antilles. This system developed into Tropical Storm Isaac late in the afternoon on Tuesday. This tropical cyclone provides an opportunity to see a contemporary example of some of the material presented in an early July webinar entitled “How are tropical cyclones represented in operational model initial conditions? And why does it matter?”
The July webinar presentation by Dr Gary Lackmann was inspired by research activities associated with the current CSTAR project that are in part aimed to better understand tropical cyclone (TC) initial conditions and which investigated the use of the GFDL bogus vortex to improve local modeling of TCs. The presentation noted some of the challenges of TC prediction and initialization. It featured background information on data assimilation for tropical cyclones and discussed tropical cyclone initialization for several operational models including the GFS, HWRF, GFDL, RAP, and NAM.
The GFS TC initialization process was described during the presentation and a portion of this process is summarized in the figure to the right. In short, once NHC declares a storm of TD strength or greater, the GFS will examine the first guess forecast from the 6 hour forecast from the previous cycle. The initialization process can either relocate the vortex to the NHC position (1a), or insert bogus winds into the background fields (1b). Regardless of which option is used (either 1a or 1b), a “TCvitals” files is created that contains important information (location, central pressure, etc.) used in the data assimilation. An example “TCvitals” file is shown below.
The relocation process generally moves the vortex a small amount (typically a few tenths of a degree) and this new location is then “absorbed” into the data assimilation process of the new model run. Inserting bogus winds can have a more significant impact as an artificial structure (sometimes of too great an intensity or structure) is inserted into the new model run.
An example of “bogusing” with Tropical Storm Bud in May 2012 using the GFS parallel forecasts before the recent GFS upgrade was shared in the presentation and is shown below. The top image is the 18Z GFS PARA 6-hour forecast of 850 mb winds and vorticity, valid at 00 UTC on 21 May 2012.
The 6-hour forecast showed a weak system, but satellite imagery and other information indicated a much stronger storm, and the GFS tracking algorithm was unable to find a strong or coherent enough circulation. So the GFS used option (1b) from the figure above and inserted a synthetic or bogus wind field into the GFS.
The bottom image is the 00Z GFS PARA analysis of 850 mb winds and vorticity, valid at 00 UTC on 21 May 2012 using the bogus winds. Note the radical change to Bud from the assimilation of synthetic wind observations. There was no relocation in this case, since tracker could not “find” the storm.
Tropical Depression nine (now Isaac) presents an example of GFS vortex relocation. The image to the right is the 6 hour GFS forecast from the 06Z cycle valid at 12Z. The minimum central pressure is forecast to be 1009 mb with a slightly elongated wind field. This 6 hour forecast is used as the first guess or background in the model initialization process for the 12Z cycle.
The National Hurricane Center (NHC) initialized TD 9 at 12Z at 15.1N 52.0W with an intensity of 30 knots and a central pressure of 1007 mb. The image to the right is the MSLP and 10-m wind from the 12Z GFS analysis on a 35-km grid. The minimum MSLP noted was 1008 mb and the wind center is pretty close to the NHC position and is displaced a bit to the southwest of the pressure minimum.
In the comparison image below, the 06Z GFS 6-hour forecast is shown in the left panel and the 12Z GFS 0-hour initialization is shown in the right panel. The white circle is the NHC defined center at 12Z (15.1N 52.0W). Note that westward adjustment of the 12Z analyzed position relative to the 6-h forecast from the 06Z cycle. The central pressure is also 1 mb lower in the image. It is unclear though, how much of the position change was from the vortex relocation and how much is from the normal data assimilation process associated with a subsequent model run.
So how can we determine whether and to what degree the GFS is relocating the vortex or if bogus winds are being used? After collaborating with Daryl Kleist at EMC, this information is now publicly available in real time via a inform.relocate file. The information is shown in a table that contains the observed information, model information, and relocation distance prescribed. You can get the information through ftp, nomads, or any other locations that have EMC operational output available. For the Tuesday, 21 August, 12Z GFS forecast, the information is available in a table format at the URLs below:
If the storm was bogused instead of relocated, the file will note it as shown below in the example below from the storm Kai-Tak.
I need to acknowledge the assistance of Daryl Kleist (EMC), Dr Michael Brennan (NHC), and Dr Gary Lackmann (NC State) with this post. They provided much of the imagery and information and their assistance is greatly appreciated.