Monday, September 08, 2008

Hurricane Ike's Pasta Problem

Ike is pounding Cuba and Cuba is counter-attacking by weakening Ike to Cat-2 status. Meanwhile, the National Hurricane Center says early this morning "there has been no significant change to the forecast track."

That announcement, however, comes with a great big asterisk: there is no change in the forecast because the forecaster chose not to make one.

This offers the opportunity to remind everyone that the popular NHC 5-day forecast cone track (shown above), and the computer spaghetti tracking models on which it is based (shown below), are heavily dependent on human interpretations and just plain guessing. With computers, of course, it's always "garbage in, garbage out." With hurricane forecasting by humans, it also can be "spaghetti tracks out -- and then ignored."

That's what has happened this morning. NHC forecaster "Franklin" notes that --
Ike's trajectory is expected to bend gradually to the right as the storm nears a weakness in the subtropical ridge over the next couple of days... and the guidance models are tightly clustered through 72 hours.
Then he candidly discloses that three spaghetti models -- the GFS, GFDL, and HWRF -- "show a bend to the right at the end of the forecast period." Other models do not.

What to do in constructing the 5-day graphic forecast cone above? Confesses Franklin:
Even though the GFDL has performed very well with Ike so far... I've chosen not to adjust the track eastward given that the large-scale fields in the GFS have no support from the other models.
Data in, choices must be made. Some data is accepted, some is ignored -- like three of the spaghetti models shown here:

Franklin may be proved a genius or a fool, but at least he's honest. He ends with this warning: "It is still too soon to know what portion of the Gulf Coast will ultimately be affected by Ike."


Anonymous said...

Last year, someone (an expert in tropical weather systems) explained when certain computer models are to be trusted over others.
Time of year, origin of storm and other such factors figured in the equation. Can anyone point me to that information?

Beach Blogger said...

You may be thinking of the National Hurricane Center's 9-part "Forecast Verification" article, avaiable on the NHC web site here: Several subsections of that article include links to other studies and analyses comparing computer forecast models, including James Franklin's semi-famous study of the relative performance of various models in the 2004 storm season. It is titled 2004 Hurricane Center Forecast Verification Report [pdf format].