Thursday, March 10, 2011

Humble Words From Paul Douglas

Many readers of TMF likely follow Paul Douglas's weather blog on the Star Tribune. Paul has received a great deal of heat lately from his forecast for the recent storm as well as for some controversial long-range forecasts. In what we consider to be a humble, honest effort to gather the thoughts of his readers, Paul's current blog entry solicits their input. In addition to posting any comments directly on his blog, we offer this space to collect the thoughts of our valued TMF readers. We will be sure he is made aware of any comments you may leave.

The following is the full text of Paul's self-described mea culpa taken directly from the blog:

Mea Culpa. In my highly unscientific and statistically-meaningless poll of a few dozen people yesterday, fully 78% were "relieved" that the storm never materialized. The other 22% (apparent snow-lovers, still not quite content with the 79+" snow we've received) gave me an earful. "What happened to the 12" you PROMISED me Paul?" Huh?

On this blog I've been posting the latest model predictions looking out a week or more, knowing full well that some of these long-range guestimates are iffy. I went out of my way to remind (everyone) that confidence levels were low, I tried to explain, "here's what the models are saying." "Here's a look into the future, warts and all." Maybe that's the wrong approach. I'm starting to wonder if I should do a better job of editing the models and not putting everything out there. Long range forecasting is the meteorological equivalent of making sausage - you probably don't want to see the process, if you can help it. Dozens of models to choose from, some work better than others in certain scenarios, sometimes they all disagree. On occasion they're all wrong - which was pretty much the case with this last "storm".

I suspect that the physics in the models (especially the GFS, which is the only American model that goes out beyond 84 hours) has a tougher time dealing with the transition from a wintry regime to a springlike pattern, which is what's underway right now. I can't prove that, but I've heard enough complaining (from other meteorologists) that leads me to believe that a March model run is probably more unreliable than a January model solution. I did try to give a blow by blow account of what the models were predicting (for better or worse), and a few days ago it really did look like we could be in for a major snowfall. But since Monday the models (including the GFS and the NAM/WRF) began to pull back, pushing the track farther east, and we gradually dropped our predictions, to the point we were calling for "an inch or two" as of Tuesday. No excuses - this last system was not handled well. It's easy to blame the models, but it's our responsibilty as meteorologists to monitor and vet those models, throw out the dubious model runs - and through a combination of gut feel, historical context and experience zero in on the right solution, weeding out the "correct" model.

The GFS model nailed the major snowstorms back in December and February (which probably gave me a false sense of security and confidence). This time around the models were almost worthless. So I'm wrestling with this: do I continue to post what the models are predicting (come what may), or do more judicious editing, and only post what I think may REALLY unfold looking out 5-15 days? My gut is telling me to still share as much (raw data) as possible - get it out there and let you see what the guidance is saying, but do a better job of attaching a "confidence level" to those models. When all the models agree our confidence tends to go up, and for a time late last week all the models were hinting at over 6", maybe a foot for parts of southeastern MN. They were (all) wrong, in 20/20 hindsight. That's what's so frustrating about this last system. That's the thing about weather forecasting. You nail a couple of major storms, your confidence level goes up, and then Mother Nature kicks you in the Doppler. A painful learning curve.

A humbling profession, yes. Mea culpa and all due apologies to snow-lovers. Commuters caught a break this time around. Time to go back to eating my tasty crow souffle.

23 comments:

  1. 50 inches Paul, Not even the hypers believed that one!! He looked very foolish to the common Minnesotan. I had a co-worker comment that no wonder TV doesn't want him!

    ReplyDelete
  2. The company I work for performs scientific for clients that hire us. The worst thing we can do is give clients raw data. There are three possible outcomes: 1) they have no idea what it means, 2) they misinterpret it, or 3) they interpret it correctly. Let's just say option #3 happens very seldom. It's not that our clients are stupid, it's that they are looking to us for professional guidance in understanding what the data means.

    How's a model-based forecast any different?

    ReplyDelete
  3. This comment has been removed by the author.

    ReplyDelete
  4. You know it really bugs me how much people will gripe and complain when a forecast isn't exactly correct, or when a storm doesn't exactly pan out the way they said it would. Personally, I think that if someone has a gripe, then they should go to the station, look at the forecast models, and try it themselves. As someone who works in the picture tube world, I hear day in and day out about the phone calls and emails from people outside the meteorology profession trying to tell them how to do their jobs. The world of meteorology is still being perfected, and I highly doubt it ever will be an exact science, but maybe those people who can't tell a hail stone from a sleet pellet should keep their comments to themselves, unless they want to go ahead and give it a shot. Since a lot of us in here follow the weather so much, and go as far as to try to figure out the forecast models ourselves (some are quite good at it), their initial thoughts on the system were accurate. Two of the major models, both with a similar solution, both bringing a ton of snow. But anyone living here for any amount of time has to realize that these systems are constantly changing, and this last system was a prime example of that. Paul typically does a great job of telling people what the models show AT THAT POINT IN TIME, but also warn people that there is a lot that can change. Personally, if anyone hears a forecast that is more than a week away, with snow predictions, and takes that as the final predicted snow amount, then they need to wake up and realize where they live.

    ReplyDelete
  5. Duane,

    completely off base. Most folks that watch/read the forecasters looking for information so they can schedule there day or how it might effect there day.

    Are you kidding me? Paul Douglas is a highly educated, seasoned, and experienced meteorologist. Most folks are not, and therefore are looking to him/others for the correct, or something similar to to the truth on weather. As it will effect us as individuals. Telling us 50 inches of snow or severe weather is going to happen is out of line!

    If we all could do that we would not need Dahls, Douglass, or NOAA!

    Get a grip. Are you Paul's little brother or what?

    ReplyDelete
  6. Duane....unuf said, couldn't have said it better. On a side note I sure hope the Good For Speculation Model (oops I mean the GFS) does it's usual thing and makes a 180° turn for the last week of it's current run....

    ReplyDelete
  7. I don't know that I was really bugged or upset by the presentation of the models - but what really annoyed me was the Editor at the Strib who declared in a breaking news headline last Friday that the Storm was coming. My concern with PD and others is that they consistently predict the worst and we get much better most of the time. My simple fix for the weather folks is this: 1) don't provide amounts more than 48 hours in advance 2) only provide amounts if all models point to the same or similar results. 3) If the models are forecasting an event a week out - then say there is a possiblity of rain, slow or whatever that date but no numbers or details. Problems occur when you give specific times and amounts when you're that far out. PD stated in his blog - last week - 20" of snow starting in the morning Tuesday lasting until noon Wed. That is a degree of precision that should not be provided - instead - say something like - models are predicting a snow event next Tuesday or Wednesday.

    ReplyDelete
  8. As a reader, I like to see confidence levels reflected in a forecast, whether it's one that's 2 days out, a week out or a month out. This is an example that TMF has used previously, but I like it:

    Here are the current accumulation possibilities that will certainly evolve in the next several days:
    30% chance: A dusting or less
    30% chance: A dusting to 1"
    20% chance: 1-5"
    20% chance: 5"+

    ReplyDelete
  9. Anonymous:

    I had take take a step back and calm down before I posted a response to your post, while i did that I took time to page back at Paul's post about that model run. After reading this again can you please tell us how you extrapolated the fact the Paul said we were going to get 50"??? Here is what he said from that blog.

    Two "Waves" Of Snow? The latest models are looking more impressive for a "plowable" snowfall Monday, a surge of moisture and energy spinning off ahead of the main storm on Wednesday - possibly as much as 2-4" Monday. A lull Tuesday gives way to the main storm Wednesday, but the latest GFS models drop amounts on Wednesday into the 5-7" range. Between both snow events it could still wind up being 5-10" from Monday through Wednesday, nothing to sneeze at, but (hopefully) not the 10-15"+ that earlier models were hinting at.

    Say What? I know - I had to triple check and make sure this wasn't in metric. No - that's INCHES of snow between today and noon on March 19, factoring in next week's storm, and a predicted second major storm around March 18-19. A lot of factors to consider - this is a (very shaky) long-range outlook which is almost certainly going to change - but the trends are troubling for spring flooding. If we even pick up HALF this amount (15", instead of the predicted 31" for the Twin Cities, and only 26" across southwestern MN, instead of the 53" predicted for Windom) we will experience severe flooding, possibly the worst on many Minnesota tributaries since 1965. Thanks to Andy Revering over at F5Data.com for passing this along.

    (Again - the latest 12z GFS is pulling back on snowfall amounts - the first time we've seen less than a foot. I want to see if this is a true trend, or an aberration. We'll update you on what the 18z and 00z GFS numbers look like). 8" would still be a pretty big deal, but obviously not as bad as 10-15". I'm hoping this map doesn't verify - latest runs shift the heaviest snow bands south/east - we'll see if that trend continues).

    ReplyDelete
  10. haha...no I'm not his brother, although that would be pretty cool. I am just a weather enthusiast who knows enough about forecast models to understand that they change every day, every run, every hour of the day. I do agree with what Dave had said, in that mentioning that a large snow event, or in the summer a potential severe weather event, could potentially unfold during a certain time frame, instead of using words that would imply certainty. And in reality it just doesn't matter how much training or schooling a person has. Forecasters read and interpret what the models tell them, and give that information to the public. But mother nature doesn't look at what the GFS or NAM say and just think, ok that is what it says, so that is what I'm going to do. Models are not perfect, forecasters are not perfect, but they do their best day in and day out to try to figure out what is going to happen so that the public can plan their day. Personally, if someone told me today that next Friday we are going to, or could have 20 inches of snow, it is simply going to alert me that something could potentially be on the way, but I am going to continue to watch for updates. If we don't get that much, it doesn't bother me because I am aware enough to know how much these forecasts change, especially during transition times (Fall to winter, winter to spring). I'm sure they will adjust their headlines to not imply certainty so much, but P.D. does at least post what he is seeing in the models, so that others can see it too. Can't blame the forecaster if mother nature doesn't cooperate.

    ReplyDelete
  11. Anyone having trouble leaving comments here?

    ReplyDelete
  12. I am guessing most of us are at this site because we:
    (1) Really enjoy following the weather.
    (2) Are professional or backyard meteorologists.
    (3) Want to learn something new.
    99% of the rest of the population probably doesn't care to know or follow weather as closely as us.
    If Paul Douglas/Star Tribune had memberships/subscribers that were strictly meteorologists, than he could make certain claims or presentations on his blog that a meteorologist would fully understand and take with a grain of salt. The problem is most people (at least that I have dealt with outside of meteorology) couldn't tell you which way a low pressure rotates. When PD (and others) put out 10-15" numbers a week before a storm arrives, there is almost no where to go but down.
    We all have to realize how infrequent a 15" snowfall occurrence is within the Metro.
    I definitely give Paul credit for his mea culpa. I think he has been humbled. Personally I happen to like PD and the plethora of information he has to provide. But, he can't be surprised if there is backlash when he puts 15" on the table and it turns out to be 1".
    Models will be wrong, and this will happen again. But it is all in the (perceived) presentation.

    ReplyDelete
  13. Just my own two cents:

    I like to see the thought processes behind Paul Douglas' storm predictions. But I differ from most I guess in that I'm not looking to plan my day around the forecasts, more that I just like to learn how storms work. His blog is almost stream-of-consciousness, and that why I dig it, as it goes through progressions when there is a possible event on the horizon.

    So I hope he continues to blog as he has been, and I'll look to the NWS if I need to see if my picnic will be rained upon.

    ReplyDelete
  14. This is from poster DDwx:

    For some reason in the past 1-2 years many meteorologists began to post up model images on their blog pages, telecasts, etc. When various models are shown it adds even more to points #1 and #2 above -- I hear that all the time from readers, viewers, users as they will ask what in the world is it? Lets not forget today's models are tweaked more often than ever before as biases, trends, etc of models today are tough to get a handle on and that makes forecasting more difficult. Posting and talking about the models also gives a crutch to meteorologists when a forecast goes bad. Its the meteorologists job to decipher and figure out the rights and wrongs of models runs comparing to real, current data, climatology, experience and even hunches. Knowing the true physics and mechanisms of the different models DOES make a difference.

    What you see on a website, well, anyone can make a forecast from what they see but actually understanding and knowing the models is another story and always shows during tough forecasts. The biggest mistake made, and it did surprise me, was to throw out big snow amounts so incredibly far out in time when they know the true uncertainty of weather forecasting, the models variances and the late winter time table along with climatology and the overall atmospheric patterns.

    In reality if you do a dt/dp of the models from 7 days before the event and each day closer to it you could easily see the trends and changes taking place thus the forecasts could have easily reflected that sooner than they did. But, give them credit at days 2-3 most meteorologists did make the adjustment; the signal, however, was already there by day 4.

    Storm is over. Take the bitter pill and wash it down with castor oil and the burnt crow and move on. We have all been there before.

    ReplyDelete
  15. DDwx I agree with most of what you said, it appears that most of our mets went with the GFS/Nam solution when the Euro was the right one..Why did they ignore its solution? Why didn't they pick up on that as early as Saturday night??? This was my forecast on 2:13pm Firday, I had it weighted 70-30% Euro vs GFS..(I should have thrown the GFS under the bus at this point but didn't) ...."based on new guidance hot off the press have lowered snow fall totals to 5-8" ". I followed it up Friday night strongly hinting at the final outcome..."as of now accumulations are still possible for the metro, at this time of the year it is not uncommon to see more than one change in the track, stay tuned". Followed by a couple of forecast hinting that the heaviest snow's would be in far se MN. Until I got to my final forecast that was issued at 7:09 Sunday night.....northern metro 0-2", central metro including the airport 1-3" south metro 2-4, confidence level 5 out of 10 since I think there is a 50/50 chance that this system will drop se, there is a chance that the metro could see very little snow fall.

    So I want to ask our local meteorologist's a question?? Do you relay on what is more comfortable for you?? ie: gfs/nam, gfs/nam moss guidance , in house models etc....or do you broaden your horizon's and go with the model that has been the most accurate the last two weeks???

    ReplyDelete
  16. PS I can understand PD's post the night before the storm hit, with the Nam showing 6"...he is after all under a deadline with the star trib, and the Euro comes out later

    ReplyDelete
  17. Sorry he acted as a weather terrorist/nazi. Basically they can't get it right 48hrs plus out. 48 hrs in they have a shot but they still can get it wrong. Saying 20"s out a week is just throwing darts at a board.

    He mentions the dec storm, but everyone got that wrong also. They under calc that snowfall and kept raising the numbers when it was snowing.

    Why leave models out a week for others to see if your the expert getting paid to interpret them?

    ReplyDelete
  18. Anonymous:

    Well I really can't remember what the local mets were saying about the Dec storm, all I know is this, all the models were right with somewhere getting up to 20" of snow, but guessing where the deformation area would set up is like betting on the Vikings to win the Super Bowl

    ReplyDelete
  19. Yea, but Paul can post to his blog whenever he wishes... so that deadline issues shouldn't effect that.

    ReplyDelete
  20. Bill.....true, my bad

    ReplyDelete
  21. so that kind of reinforces my thought, relaying to much on the model your comfortable with?? or waiting till all of the suite of models come out??

    ReplyDelete
  22. I don't like models that much. They get everyone worked up about a big storm, only to have it change course and get only a dusting. Beyond 84hrs, they can change suddenly and randomly.

    ReplyDelete
  23. As a professional met, I say it is disingenuous for a forecaster to lob model data at a reader and say "Hey! Look at this!" Your job is to interpret as much data as you reasonably can (a met doesn't need to look at all models, there are just too many nowadays), including observed data, and in combination with your knowledge/experience, provide a forecast. Throwing out wild, distant, erratic possibilities is just a dangerous and losing "scoop" game. Yea, sometimes models can be "right" 6-10 days in advance, but usually they are "wrong".

    ReplyDelete