February 21, 2009

Uncertainty II

Several of you have asked about forecast uncertainty...something I started to discuss yesterday.

The truth is that all forecasts are uncertain. That uncertainty increases in time. The amount of uncertainty varies by weather situation. Sometimes even a 3-4 day forecast is highly uncertain (like yesterday),while other times there is substantial forecast skill out 7-8 days. The fact that TV stations (and even the NWS) usually give single numbers for forecasts...particularly long-range ones is really crazy...we just don't have that kind of precision and skill. Forecasts are essentially probabilistic...and should be communicated that way. We should not forecast the temperature will be 64 in three days, but rather give the probabilities of various ranges.

This is a very active area of research and development right now...and the key tools are ensembles of forecasts. The old way is to run one computer simulation of the future and use that to provide a single potential weather evolution and single forecast numbers. Now we have much more computer power we can run many forecasts, each with a slightly different (but reasonable) starting condition or with somewhat different model physics (e.g., how clouds form). The forecasts will be different and the differences will generally increase in time. When the forecasts are very similar...then uncertainty is low, and vice versa. You can get probabilities from them as well. If half the simulations give rain and the rest do not...then a 50% chance of rain is reasonable (we have much more sophisticated ways of doing this, but it will serve as an example).

The NWS has several ensemble systems, as do we at the UW.
They are actively used to determine how uncertain the forecasts are and to understand the range of possibilities. We are now trying to develop the techniques to provide high-quality uncertainty/probabilistic information from the ensembles. In five years we should have reliable forecasts of this kind. But then there is the next problem....how do we communicate the information? Are people ready for such guidance? Are you ready for it?

Regarding today's forecast...lots of low clouds in eastern WA, but generaly sunny elsewhere. High clouds are moving in overhead as a system approaches from from the SW (see images). Tomorrow we will have considerable clouds and showers. And that will be the character of the upcoming week...with the mountains finally getting a modest amount of snow (particularly midweek). From my viewing of the ensembles, there appears to be very little likelihood of lowland snow...one of the reasons I was discouraging such talk from those looking at long-term forecasts.


I will be at the NW Flower and Garden Show in Seattle tomorrow at noon if anyone want to talk in person. And don't forget the KIRO weather special at 7 PM on Monday...it should be good.

28 comments:

  1. This comment has been removed by the author.

    ReplyDelete
  2. Cliff Said...We are now trying to develop the techniques to provide high-quality uncertainty/probabilistic information from the ensembles.

    "because as we know, there are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns -- the ones we don't know we don't know." Donald Rumsfeld

    ReplyDelete
  3. w\What are chances of getting the needed off-shore radar coverage and how will it enhance the accuracy of weather prediction?

    ReplyDelete
  4. Think as far as forecast go...they have come along ways in better forecasting. Back "in the day", we did`nt have such high tech computers that could spit out info a full 7 days or probably even 5 days for that matter. The information and the powerful technology just wasn`t around back in the early days of when computers first got started. Would guess it was more of like the weather spotters/RE-CON aircraft of today. Having them go and do daily weather checks to see what surface and upper levels were doing for that day and they try and predict the forecast a couple days out.

    So as Cliff said, forecast technology may very well become more 'high tech' with better ways to predict or give a summary of what the short or mid-long term weather may do.

    ReplyDelete
  5. I don't want to sound like a jerk, but I think that is the sort of musing that belongs over on the blog you created for yourself Andy. It isn't very specific to the topic and I think you are just speculating on the state of the science years ago. I do hope we hear from the professor about where the state of the science and computing is headed.

    Hopefully more on topic, Probcast is giving me 54 for a high temp Sunday, with a range of as high as 57 and as low as 50, with a 50/50 chance of rain. The likely amount is 0.01" but as much as 0.15"--now as a consumer of weather information, how does that forecast work for people, or you for that matter Andy working at a gardening center. Thoughts? And I hope you do not take offense--those of us who post often have to resist the urge to merely chat, and I'm as guilty as anybody--we all love to hear ourselves think out loud. But once again, don't let my comment put you off too much. I have a blog with thousands of posts, sort of an online diary, truth be told 95% of it is crap and only a few times have I had something insightful to archive--less is more.

    ReplyDelete
  6. Wow, I never would have thought meteorology was such a cutthroat field. I wonder if Steve Pool and Jeff Renner have ever slugged it out? Way more discussing what should and shouldn't be said rather than actual comments on the weather.

    ReplyDelete
  7. Cliff

    I'm attending the lecture on Wednesday night with a friend, looking forward to it. I missed you recent Bellingham talk by a day when I realized I got my days mixed up. :)

    ReplyDelete
  8. Andy..that message is nearly incoherent. Please only comment when you have something useful to say! We would all appreciate that...Jim R.

    ReplyDelete
  9. Cliff I have to commend you on being so accessible and open to rigorous inquiry with your posts. Thanks for sharing your knowledge concerning atmospheric sciences and explaining complex weather terminology in a manner that we cana all relate. Cheers!

    ReplyDelete
  10. Jim... I was on topic and it was relating to what Cliff had mentioned to the problems that can come with forecasting the short term period.

    ReplyDelete
  11. Weathernerd--I don't think it is at all cutthroat. Take the teevee personalities--mainly there to put a pretty face on it and use up a few minutes. Honestly, how often do you hear genuinely useful information "...the forecast high will be 47 here and 48 there and the warm spot will be 49 over there" (even though we know the forecast will miss everywhere by a few degrees).

    At the NWS we are throwing out a lot more information with the point and click forecasts and such--but is that much more skillful and useful? It is only a start. Like the UW's Probcast. The services are changing and we are sending out feelers to see what works and what does not.

    Andy--your enthusiasm is top notch but your feel for how the forecast models and computing have evolved is a little off the mark. That is why your post seemed off topic.
    I'll answer that question for you if you don't mind--I did work for a landscaping outfit a couple of summers. Suppose a forecast gives an 80% chance of a marine push and a temperature drop of 15 degrees. You might lay sod ahead of the push, right? Or if there was an 80%chance of a half inch of rain--you might not have to water ahead of that front. Of course, there is a 20% chance you are screwed and all that sod is on the back of the truck under a tarp, etc etc. Obviously cost benefit analysis comes into it for some users.

    ReplyDelete
  12. Hi Mike. Your answer is pretty good as I`m not sure how good mine would of been.

    Anyway, the probability of lets say rain vs. hot and sunny really depends on the percent and or chance you need to water your lawn for that day. If there`s a high chance of rain, then the chance you will be watering is low. But if the forecast calls for a high probability of it being hot/sunny, the probability of you watering is high and thus, you`ll need to water.

    ReplyDelete
  13. I'm old enough to remember when they still called the upcoming weather a "prediction"...it's weather in the making up here, folks...no better than reading tea leaves.

    ReplyDelete
  14. And not just watering Andy--there is difficulty in relying on a sky cover forecast. I had a big old homade polyethylene and pvc pipe greenhouse growing up. More than once I left for school in the morning under low thick stratus only to come home to a greenhouse full of wilting plants because I had no idea it would scatter out midday. Well, now it is me posting more often that I am being useful, heh, we could go back and forth forever with good examples I am sure. Maybe I will join you on your thread if you don't mind. I have plenty of off topic weather stories for anyone who will listen. My Hat (Gedney) Island story is a perfect example of needing a procast--prob of small craft winds in that case. A hairy afternoon for teenagers in an underpowerd 10ft boat.

    ReplyDelete
  15. Sure Mike of MLT... feel free to join my blog. :o)

    ReplyDelete
  16. I am definitely ready for probabilistic weather forecasts. But I think it is a question of keeping things simple for those who want it simple and yet offering more for people who want more. It seems like this is the same as anything else in that if you have more capability and information you make it more complicated. I'm not sure about a good analogy here but, you know, the more customizable something is the more complicated it usually is. Back to the topic of the weather, as a hiker/surfer/windsurfer I am keenly interested in "how much" and "when" things are going to happen. Even now the NWS uses probabilities for precipitation but that's about it. If you want more information you can dig around for specific (e.g. boater/aviation) forecasts and/or read the discussion. You can also try to interpret the models yourself (which I'm not too good at). I think this is going in the right direction: Don't force people who aren't that interested in the weather to read a bunch of jargon or learn how to interpret weather maps, but make it easier for people who want to know more to dig deeper.

    ReplyDelete
  17. Cliff - this is a little (okay, a lot) off-topic, but I was wondering... was looking at the UW Atmos Sci website and it looks like to apply for graduate school at the UW in meteorology, I have to wait to apply for Fall 2010 - too late to apply for Fall 2009 and I can't apply for any other time of year. Can you confirm this?

    ReplyDelete
  18. Answers to several questions:

    A little late to apply to UW atmos grad school, but still possible if you can have the application in during the next week...and assuming you have already have your GRE score.

    Regarding cutthroat meteorologists...no way, nicest people you would ever hope to meet.

    Chances of getting the radar...maybe 50-50 now...so all of you need to contact your congressman and Senator asap. There is stimulus money that could be used for it.

    ReplyDelete
  19. Cliff - I found your book at the Gig Harbor Costco; looking forward to reading it. Maybe in the next couple of days when all the sunlight doesn't inveigle me outside I'll start!

    - Pete from Olalla

    ReplyDelete
  20. Cliff-
    Can't wait to see your lecture when you come to Bainbridge. I am a junior in high school and my dream is to become a meteorologist. I spend hours a day studying the weather and hope to someday have you as a professor at the UW.

    ReplyDelete
  21. After my cheap shot about teevee forecasts (48 49 50...the old precision versus accuracy debate) I ought to conceed that we do exactly the same thing here with our so-called digital forecast, --only we give all the pixels, argh!

    The teevee mets have a limited amount of time and their delivery has to be smooth as silk, so I take back what I said if I can. The temp forecasts and the sun/cloud/raindrop icons day after day are simplistic, but pumping more information out in that format is probably a hard task. And I'll echo the professor's comment: I was in Cliff's synoptic lab some twenty years ago, Jeff Renner happned to be picking up the class at the time to get his met degree, and yes, the TV mets are certainly as pleasant and good natured as any people you are likely to meet. We get calls occasionally from the TV mets, and they are always a pleasure to talk to. Now gotta go offline for the day--I can really only post on my own time before and after my shift.

    I hope we see a lot more comments about probcast and such, that is mostly why I keep lurking. And I'm looking forward to seeing the UW spring forecast league start up too-, it is always fun to see the latest batch of students forecast.

    ReplyDelete
  22. So here is an insight I would love. So I live in Sammamish. I was watching the radar about 30 mins ago, and a GIANT blob of yellow was headed right for us from the due south. JUST as it reached where it should have hit Sammamish, it completely changed to light green and nothing fell at all where I live.

    What causes the breakup? All of the weather is coming straight from the south, and yet nothing is falling at all where I'm at. Is there some landscape that is blocking from the south?

    ReplyDelete
  23. SnowLover
    I took a look to see what you are talking about. It looked like there was more or less an east-west line of yellow and red precip which then lost some of its oomph as it headed north. The western part of the line I observed ran into Tiger Mountain (elev 3000 feet) and possibly Squak Mtn, just south of Issaquah. The mountains appear to have eaten up a good chunk of what you were anticipating.

    ReplyDelete
  24. regarding the radar question...high reflectivities (the yellow colors) in the radar are generally associated with convective cells, which have short (.5-1 hr)lifetimes...thus it is natural for there to be declines. Also, there is sinking on the northern (lee) side of the Issaquah alps...which also causes drying...cliff

    ReplyDelete
  25. Cliff said: In five years we should have reliable forecasts of this kind. But then there is the next problem....how do we communicate the information? Are people ready for such guidance? Are you ready for it?

    I'm ready! I would like to see forecasts of individual weather variables (e.g. temperature, % precip, wind speed/direction, cloud cover) each reported graphically as a probability distribution. So, there will always be the most likely value at the apex of the curve, but the breadth of the curve will indicate the range and relative confidences associated with the forecast. Is this asking too much of the modeling software? Of the meteorologists?

    This may be a bit too cumbersome for the 'just-tell-me-whether-it-will-rain' crowd, but, to me, this would be a relatively simple way to communicate confidences associated with forecasts.

    see http://en.wikipedia.org/wiki/Probability_distribution for examples of the types of graphs I am attempting to describe.

    ReplyDelete
  26. Gator--I don't think the UW has done probability histograms but from the ensemble page you can select spatial presentations of, for instance, wind speed. The two thresholds are set at 18 and 34kt to indicated chance of small craft advisory or gale force winds. It does not seem to be tuned right now--for instance the 4km will often give an area of 30kt winds with a system--and verify really well, but if you check the chance of gale from that page it will be zero and the chance of small craft will not be nearly high enough IMO. --MM

    ReplyDelete
  27. Cliff,

    Thanks for the extended discussion of uncertainty. For my part, I think the approach used in the probabilistic forecast is overly numerical. I think I would be satisfied with a qualitative scale. Something like "Confident", "Somewhat Confident" and "Uncertain". I would associate them with colors like blue, purple and red... then put the graph across the bottom of the days.
    As an applied mathematics major turned economist (I likely took the Amath 400 series with some of your students while at UW) I tend to prefer numbers. My sense is that an algorithmic approach is just going to introduce its own error when a skilled forecaster can better sense the certainty. Some phenomena are just better "sensed" by a pro than computed.
    Thanks again.

    ReplyDelete

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

Should I Let AI Write My Blogs?

I spend a lot of time thinking about and writing my blogs.   Could AI apps do the work for me?   Could any of you tell the difference? Well,...