October 07, 2016

Hurricane Matthew: Which Forecast Model is Best?

Normally, this blog is dedicated to West Coast meteorology, but I can't help myself today with Category 3 Hurricane Matthew now raking the Florida coast.

The radar image at 7:20 PM PDT Thursday from the NWS Melbourne radar is impressive, with the eye clearly evident just offshore and intense spiral rainbands circling around the system.

The HRRR model pressure analysis shows a deep low center just east of Florida, which huge pressure gradients (differences in pressure).  The National Hurricane Center estimates max sustained winds of 130 mph and a central pressure of 939 hPa (mb).  For comparison, the great Columbus Day Storm of 1962 only dropped to 955 hPa.

This storm has savaged the Bahamas.  Here are the gusts from a buoy SPGF1, which reached 91 knots (105 mph)  before contact was lost.

At 3 AM PDT Friday a very well developed eye was evident offshore, with the eye wall not quite reaching shore.

Over Florida,the maximum gusts (last 24h), have reached into the 70s (mph) along the coast, so winds along the coast are barely making it to hurricane levels (sustained winds of 74 mph or more)

This storm is taking a very unusual track--forecast to parallel the Atlantic coast before circling back on itself to hit Florida again as a weakened storm)--see predicted track below by the National Hurricane Center.

This track will result in heavy precipitation  and hurricane-forced winds along a vast stretch of the Atlantic coast, with a storm surge north of the hurricane center.

But what about the forecasts?  Meteorologists have a wide variety of forecast models to look at including:
  • The European Center For Medium Range Weather Forecasting (ECMWF) model
  • The UKMET Office model
  • The U.S. GFS global model
  • The U.S. high-resolution HWRF model
  • ...and many more.
We can evaluate their skill in two ways:
  • the accuracy of the position over time or track error
  • the accuracy of the forecast of storm intensity (how deep the low center is, the maximum wind speed, etc.
Let's face it--the track forecast is the most important.  Having a correct intensity forecast is not very useful if the storm is in the wrong place!

It turns out that there are consistent skill differences and one model is clearly superior for Hurricane track forecasting:  the European Center model, with the UKMET right behind.

Here is a plot of the average position errors of various modeling systems for roughly the last week regarding Matthew (produced by an individual in NOAA).  AVNO (green) is the US GFS,  HWRF (light purple is HWRF), TECM (yellow) is ECMWF, and dark purple is UKMET.  The errors are divided into length of forecast.

The results are STUNNING.  UKMET and ECMWF are clearly the best.   Next is GFS and HWRF is in the rear.  U.S. loses big time.

OK.  This is one storm. The National Hurricane Center has done a similar analysis for many storms over the past several years, with similar results. This graph shows the change in 48h track errors for many modeling and forecasting systems for the Atlantic Basin.  Track e have gotten much better in time for all systems--a real advance.  But the best is clearly the European Center (light blue, with the GFS (dark blue) and HWRF (HWFI, reddish color) substantially behind.

HWRF  tracks have been particularly problematic for Matthew and has held on to a clearly problematic forecast (with Matthew moving toward New England) far too long (see example below for forecast from October 5 at 11 AM PDT).  HWRF heads to the NE (purple), while the GFS and official forecasts swing around in circle.

The inferiority of HWRF tracks is frequently observed, as is the superiority of the ECMWF tropical cyclone tracks.   The latter reflects the excellence of the ECMWF global forecast compared to the US global model (GFS), something that I have talked about in this blog previously.    My colleagues who are hurricane modeling experts tell me that HWRF has generally been a poor investment and should be terminated, a recommendation supported by objective information such as above.  Instead, the U.S. needs to invest in improving its global modeling system, which is now in transition from the GFS to a new system called FV-3.

It is time for US weather modeling to become world class again.  

And while national attention is directed toward Florida, the Northwest is getting its first blow of the season, with our winds comparable to much of Florida (see max gusts over the last 24 h below).  Over 60 mph on the coast and NW Washington and even 73 mph at (where else?) Hurricane Ridge.  Why does Weather Channel ignore us?

(PS:   the National Hurricane Center's website WENT DOWN as the Hurricane approached the Florida Coast on Thursday evening.   Not good)


  1. Nice analysis... I saw a LOT of people showing HRRR landfall, even though HRRR did bad with our last hurricane. Any way of incorporating that into your plot?

  2. I wish this analysis was more widely known, because I remember well when the only forecasting model that was correct for a huge winter storm that dumped over twenty inches on Chicago (when I was living there) was...the European one. Unfortunately, the local forecasters relied on the US model, thereby indirectly causing a huge disaster when tens of thousands motorists were stranded when the blizzard hit.

  3. Whatever the models may say, the NHC did a first-class job of calling this track. Even when nearly every model was shifting the track further west over land, they stuck to their guns and predicted a slightly offshore track. And, huzzah, it verified! The slightly more eastward track has saved E Florida homeowners (including my dad and stepmom in Port St. Lucie) billions.

    This goes to show that there is still an art to weather forecasting. The NHC are top-notch artisans in this regard. Now let's hope that GA, SC and NC similarly escape massive damage as well. And, let's not forget the loss of life and destruction in Haiti, Cuba and The Bahamas - they were not so lucky.

  4. Looking at reports from stations along the mid-Florida coast early this morning (Oct. 7), it looks like the Washington coast was windier than much of Florida's. Did Matthew rapidly weaken, or did it stay further out to sea than forecasted?

  5. I hope to never see the Weather Channel in the PNW. Just a circus. And correspondents need to know wind is invisible. No I didn't just see that wind gust.

  6. The Northwest (and especially the Northern Rockies) is the most ignored part of the country in media and climate outlooks. It is absolutely ridiculous.

  7. I also thought it was ironic that we were getting as much wind last night as many folks in east Florida. I have family in West Palm and also about 20 miles further inland, and they didn't fair much worse than my house (45mph gusts).

    The difference was that there was a kick-ass hurricane fifty miles away that could potentially have gotten them.

    But I really like the fact that we are almost NEVER in the national weather news, even when we do get hurricane-grade cyclones in winter. Not much in any news, in fact.

    I often tell people I live in the "coffee cup corner of the country". As in, if you unroll a map of the US on the table, you're are likely to put your coffee cup down on the corner to keep it from rolling back up as you study the map. We live under the coffee cup.

    This might someday be a virtue if things get weird.

  8. Youse guys need to stop complaining about the PNW being off the national radar. It's a terrible place, rains all the time, Seattle Freeze, and all that. Keep that coffee cup firmly on us.

  9. FYI - the National Weather Service has just issued a flash flood watch for NW Oregon and SW Washington. Rain expected to be over three inches by nightfall Sunday, with over five inches possible in the mountains.

  10. Well, what do you know. Turns out the US path prediction was actually closest to how it turned out.

  11. I was in Miami prior to the storm and flew out Thursday morning (day it "hit"). NOAA publishes 3 diff "max windspeed" maps. Hurricane, then 50knot (58mph) then Trop storm (39mph) maps. Those seemed pretty accurate. That is, for about 3 days prior and on the morning of the storm, they gave odds of suburban Miami ever getting hurricane force winds at zero. And then the 58mph winds at 5%, and the lower 39mph winds at about 60 %. I relied on those; both the constancy of their forecast plus the huge disparity in probability between the 39mph and the 58mph.

    MIA at the peak of the storm had about 20 mph sustained winds. So I guess the prediction was 39 mph and it ended lower.

    The part that might be more prone to complaint was all of Miami had a tropical stomr warning (which is "certain" and I guess matches the 60% likelihood I saw on the maps). But south Broward (county to north) had Hurricane warnings (so 73mph) imminent. So they closed FLL airport. FLL never saw hurricane force winds; not even tropical storm force winds.

    Again the consistency of the windspeed maps (to me) seemed to show they were confident in the odds at various locations.

  12. How can the general public in Haiti and the U.S. access the Hurricane and Tropical Storm forecasts from the UKMET and the ECMWF on the Internet or a free or paid smartphone app?

    Captians of the small local cargo boats that are now deliverying food along the southwest coast are now concerned about leaving port and are confused by false warnings.

    Can the US NHC be relied upon to predict the beginning of a new severe storm in that area?

    Should one follow blogs such as Fox News and look for low pressure areas on their seven day forecast map of pressure based on the ECMWF model?


Please make sure your comments are civil. Name calling and personal attacks are not appropriate.

Winter-Like System in May

Tomorrow will not feel like late May, with a winter-like Pacific weather system making landfall, bringing lots of rain, wind, and cool tempe...