Thursday, August 10, 2017

A Look at How Bad Hurricane Forecasting Is

Six days ago, Watts Up With That had post with a fascinating title:  "Hurricane drought to end? Models show Hurricane on track for East Coast".  Naturally, I had to look.  To my surprise, the intro to the post was a Tweet from Dr. Ryan Maue, a hurricane researcher I respect and have been following on these pages, since he was a student at Florida State University (earliest post?).  Naturally, I had to read it.
Hurricane season may ramp up a bit over the next 7-10 days w/action in southern Gulf of Mexico and in the far Atlantic w/Cape Verde system.

A 10-12 day forecast of a developing tropical storm off the coast of Africa is the next frontier of tropical weather forecasting in 2020s.

Both mesoscale hurricane models HMON and HWRF develop wave off Africa (Invest 99L) into a powerful hurricane in 5-days in open Atlantic.
The 8/5 WUWT post includes impressive simulated pictures of this tropical wave as a monster hurricane.  That peaked my interest, so I've been keeping on eye on it.   Here's the 2:00 PM update of the National Hurricane Center's Tropical Weather Outlook.  This storm is the yellow X on the right - the notation "1 (20%)" refers to this storm. 
What I find interesting here is just how spectacularly wrong the model was.  Dr. Maue’s August 4th tweet said that two of the leading edge models, "HMON and HWRF develop wave off Africa (Invest 99L) into a powerful hurricane in 5-days in open Atlantic."  How well did they predict?  It’s six days later and 99L never became a powerful hurricane; it never even became a tropical storm.  It’s still a disorganized tropical wave with the NHC giving it a 20% chance of development in the next 48 hours, up to 40% chance within 5 days. The spaghetti runs show it re-curving out to the North Atlantic, staying a few hundred miles offshore.

It's hard to imagine how the models could be more wrong.  I suppose it could have dissipated, but that’s not much worse. 

Remember Dr. Maue said A 10-12 day forecast of a developing tropical storm off the coast of Africa is the next frontier of tropical weather forecasting in 2020s.  I suppose this means we have to wait for the 2020s – maybe the late 2020s – because these results sure aren’t there, yet.  This is not to imply the models are hopeless, only that they're not done.  The only way they'll get better is if the model writers keep relentlessly looking at why they got things wrong and trying to improve them. 
Long time readers may recall that last October, within 24 hours of closest approach, the NHC forecast Hurricane Matthew to be over my head as a Cat IV storm. Actual closest approach was about 50 miles away and a much weaker cat II. We didn’t get hurricane force winds. That’s an enormous difference in the risk from the storm, since wind damage scales as velocity squared.  I'd like to see them more accurate at 24 hours, let alone at 10 days. 


  1. The Eco Warriors and Social Justice Warriors (though how hurricanes impact social justice eludes me) have been predicting the end of the gulf coast as-we-know-it with something on the order of Category 5+ storms. Imagine their profound disappointment when the weather failed to produce....anything more than the occasional cloudy day and rain squalls.

    In fact the hurricane cycle much as with other weather events, run on cycles that don't have anything to do with national politics or whether or not somebody BBQ's steaks in the back yard.

    Don't get me wrong. There will be hurricanes large and small and tropical depressions will continue as well, but it's a lot more difficult to predict the future weather based on the models that currently exist than the mandarins and progressives apparatchiks would have us believe.

    I had business with the entire planetary sciences department at Cal Tech within the past five years. Part of what they were doing is trying to predict weather on Mars, and Titan, building models, etc. Long discussions with them over the weather models for the Earth as well as weather predictions for tidally locked planets around distant suns with bulk rotating atmospheres led me to the conclusion set forth above. I know very little about the specific details of building climate models that are accurate three or four years out. Some events like an El Nino are easier to predict and the models do ok. But the granularity that they hope to predict only delivers results if they're lucky. And that's not science.

    1. Hurricanes became political when Gore showed one coming out of a smokestack on his first propaganda film. People who actually study hurricanes for a living dismissed it at the time and said he was wrong. The idea stuck, though.

      If we had real, functioning climate science, they might be studying the long term trend to fewer tropical storms and trying to understand it. There's two or three cycles that affect tropical storms, and Atlantic storms are supposed to be going into the inactive half of a 40 year cycle.

      This is more about the real modeling, though. There's a handful of universities in the world that write these forecast models. I'd love to see them get more accurate, even at 3 days, but they're essentially just a rough guide. I told the story of how bad they were less than 24 hours out.

  2. My understanding of it is that anything beyond 72 hours is mostly guesswork. Some guessers are better (or luckier) than others.

    It's not that forecasting is bad, so much as the variables would crash a Cray.

    The fact that we have no frickin' wild idea how weather works, and that modeling it beyond 3 days is akin to trying to make a supercomputer out of seaweed, rocks, and seashells, is the entire reason "climate science", particularly with regard to global/warming/cooling/whateverinhell they're pimping today, makes deacdes and centuries in the future prognostication so much voodoo witchcraft b.s.

    It tells you everything about the purveyor, and nothing about the phenomenon it purports to describe.

    I consider everything beyond that 72-hour window of relative reliability to be written in pencil, not carved in stone.

  3. They used to publish all of the individual models for each tropical storm/wave that was of interest. (If they still do, they have moved them somewhere and I don't have time to look.)

    As someone who used to live on the water in hurricane country, it was kind of important to know what to expect. But it is clear you couldn't get a straight answer from the weather experts.

    The one storm that stands out vividly is Charley. The predictions were all for it hit Tampa Bay head on, but it took an early turn to the east and hit about 100 miles south near Punta Gorda. While they were technically in the cone of concern for the forecast, they had virtually no preparation. For the most part.

    Though if you really want to see how unpredictable hurricanes are look into the 1998 sinking of the Windjammer Cruise Lines ship the Fantome by Hurricane Mitch. When it was predicted to go north, it dipped south. When it predicted to head west it went east.

    1. Deb - the "spaghetti plots" of all the models on top of each other are in a few places. My favorite hurricane haunt lately is The Central Florida Hurricane Center -

      The South Florida Water Management District hosts some of those model plots. They're down the page on CFl Hurricane. The plot for Invest 99 is

      I remember Charley well. We got nothing but TS winds out of it, but after Punta Gorda, it picked up I4 in Orlando and followed it to Daytona and offshore.

  4. Many years ago I used to create computer models. Nothing as dramatic as predicting weather but it was still interesting. I always asked the client "what do you want it to predict"? A serious question because I had to make it do something and I would rather lean it towards what the client wanted than lean it away from their expectations. I also used to write games on the computer before all the wunderkind began writing computer games (I did say "many years ago". The games were for my enjoyment and I always included biases and back doors. It is the nature of creating a computer model or a computer game that you have to put in parameters and assumptions and you get to choose what they are. Now you don't have to put in "triggers", "drivers" and "multipliers" that are triggered by a specific word or value or drive one result over another or multiply one input effect vs another, BUT they are fun to do.

    Another way to look at this is:
    Is the deck stacked against you? Think about computer solitaire. Most of us have played it and generally it’s just a mindless time waster not given to deep thinking. Sometimes, if you’re competitive, you are really into beating the game. Maybe you have a method you follow to gain an advantage. It’s not complicated but there are a couple of options with every move and which option you take can determine the game. Perhaps you don’t even have a play plan and you just play the cards as they pop up. But think about it. Someone programmed this game and they put some tricks and traps into it. Don’t believe it? Random deal you think?

    Think about this. When computer solitaire first came out it had one version. Then it got an option to set the skill level. How did they do that? Skill level? Take a deck of 52 cards and shuffle them and think now what can I do to make a different skill level? Well, the only way you can do it is to stack the deck. That’s right your computer solitaire has been stacking the deck even before it ever offered different skill levels. Don’t believe it? Think about the times the deal would show all red cards but maybe one and the turnover would show mostly red cards but the black ones just wouldn’t play. Sure that can happen by random but think about how often this or a similar dead end game would be dealt. Now suppose you had a play technique to give you an edge. You turned over the cards without playing any as they came up just to see what would come up and what might come up if you were careful on which playable cards you selected. Kinda counting cards. It works, try it. A king buried in the middle of the three turned over cards can be exposed by playing two cards before you reach it and going around the second time. But what if there are three cards that will play? Well than, you miss playing the king. Maybe it is a crucial card.

    In other words all computer games and models are "stacked" towards or against something.

  5. The one weatherman I actually follows is Joe Bastardi, who's paid by the oil companie to forecast hurricane in the Gulf. As his forecast can determine whether oil platform gets shut down or continued to operate, there's a lot more at stake than just being. He is far more right than pretty much everyone else.

    1. I like Joe, too, and pay more attention to his AccuWeather than the Weather Channel, but he screwed the pooch on last year's Hurricane Matthew, too. 24 hours out, when the NHC was calling for it to be overhead here as a Cat 4, he was saying it might become Cat 5. It ended up Cat 2.

      OTOH, everyone's allowed to make a mistake. I can't "fire him" for blowing one call.