There was (what some in the media may refer to as) a spectacularly inaccurate forecast this week from the Computer Forecast Models [JessePedia], and that was that Typhoon Fengshen (Frank) would miss the Phillipines. Thus far, no one in the media is really talking about this, but they should. To help them stay on task, I offer the following facts. Here is where we stand this morning, according to our World News Story (PREMIUM | PRO)the storm having made direct landfall on the Philippine islands.
As of Friday evening, local time, Typhoon Fengshen was over the central Philippines. Centered over Masbate Island and 250 miles to the southeast of Manila, Luzon Island, the typhoon wielded highest sustained winds of 90 mph; it was moving west-northwestward at 14 mph. Fengshen's forecast path is northwestward and northward over Luzon through the upcoming weekend, during which time it will unleash torrential, flooding rainfall even as the rugged island slowly weakens it. Indeed, the threat of deadly flash flooding and mudslides is quite high.
On Wednesday at 5 PM, we issued this map in an electronic press release and (I assume) on AccuWeather.com as a Weather Headlines (PREMIUM | PRO):
The press release stated:
This storm is in its initial stages but has a chance to become a typhoon over the weekend. The future track of the storm will take it away from the Philippines.
This was all based on what is generally referred to as the "Navy" or "JTWC (Joint Typhoon Warning Center" forecast, which (I think) usually does a good job forecaster international hurricanes, and which we would normally have no reason to not believe. Their map at 7 PM Wednesday still looked like this (presumably* it was further east before that).
The forecasts kept "trending westward" (as AccuWeather.com meteorologist Jim Andrews noted in his blog yesterday (PREMIUM | PRO)) until the storm was finally progged to hit land yesterday evening (though forecasts continued to trend westward overnight).
SO HOW FAR OFF WAS THE FORECAST?
Who among us knows the geography of the Philippines as much as the hurricane-prone regions of the United States. Certainly I do not. So I used Google Earth to help me find out. The Wednesday evening forecast appeared to be off by about 4 degrees of longitude.
Comparing the Philippines to Florida, that would be the equivalent of placing the storm well off the coast, versus onshore 36 hours in advance (or placing it in the Gulf vs. in the Atlantic at a particular time).
I don't recall off the top of my head how many U.S. hurricane forecasts have been that far off 36 hours in the future, but I'm sure it has happened. Probably not a lot - although models bounce hurricanes around quite a bit in the long-range forecasts, they usually have it more or less pinned down by 36 hours out (which is good because evacuations can take that long). It's possible that the lack of high-resolution models in the Pacific led to this bad forecast. It's also possible that the JTWC forecast favored a particular model, such as the GFS, or their own Navy model (the NOGAPS) and they stuck with it until the bitter end (both were still forecasting non-landfalls yesterday morning, click here to download movies of their forecasts from 12Z yesterday and 00Z today).
WHY WAS THE FORECAST SO FAR OFF?
Because I don't forecast for that part of the Earth, I don't have any specific insights as to what went wrong here. But as I've said before, hurricane/typhoon prediction (and really, the prediction of any low pressure system) depends on how it gets caught in between the cogs (circulations) of other high and low pressure systems. If one of those is weaker or stronger than the models thought, it can make a big difference in the track of the storm (ever seen "Plinko" on the Price Is Right? Try that, but have the size of the pegs dynamically change.)
WHAT DO YOU THINK?
Is this a huge bust or normal forecast variability? Would the forecast have been better with better models? Should the Navy have acted faster on the westward model swing? Should we have been more independent from their forecast? Post a Comment below to let me know. I hope you appreciate me bringing this forecast error into the light and not trying to hide it -- and I'm not blaming the Navy, they were using the same models that we all have.
SIDENOTES:
1. Conspiracy theorists rejoice. The JTWC issued two forecasts before 7 PM Wednesday but they (and also the 5th forecast) are marked "Forbidden; You Do Not Have Permission To Access" on the the Naval Research Laboratory tropical weather website's archive of forecast tracks. (Coming from an IT background, however, I can tell you that particular error message is generally a technical problem and not typically used to hide web files).
Because it has the official JTWC stamp, it is generally believed that the track map shown above is a human-produced forecast, but I have no way of confirming that as of this writing (it wasn't in their FAQ). It is not actually "The Navy" by the way; their site states that the JTWC is "the U.S. Department of Defense agency responsible for issuing tropical cyclone warnings for the Pacific and Indian Oceans" and is "manned by 26 U.S. Air Force and Navy personnel."
I find it curious that the Naval Research Laboratory tropical weather website labels the images from this forecast "ATCF," a conglomeration of models that I also call the Model Spread [JessePedia]. The JTWC site says nothing about the ATCF, saying that "As of December 2006, the multi-model ensemble called "Consensus-W" was comprised of 11 members: NOGAPS, GFDN, COAMPS, AFWA MM5, NCEP GFS, WBAR, UKMET, JGSM, JTYM, TCLAPS, and ECMWF." It could just be a naming misshap.
______________________________________________________________
This has to be one of the worse forecasting busts in years. If this had happened in the Atlantic, the media would be killing the NHC. I remember when Charley made landfall in Punta Gorda and not Tampa how angry a lot of people were.
