7 Day Outlooks & Computer Models
Some of you may have noticed this past week that my outlook for the Thanksgiving Holiday was frequently changing. 7 Day Outlooks are a relatively new product in meteorology (one private company even attempts to give a 15 day outlook, which has a terrible verification score!).
Many of us who are operational forecasting meteorologists have mixed feelings about a 7 Day, as opposed to a 5 Day Outlook. Accuracy verification drifts downward the further out in time we go. However, there are several extended range models dealing with global forecasting which now go out to 7 days (one from the European Center for Medium Range Weather Forecasting) and one NWS model which goes out 384 hours, or 16 days! The latter model, called the Global Forecast System (GFS), runs on supercomputers every 6 hours. Last week, the GFS seemed to "flip-flop" on timing for a cold front and its precipitation between late Wednesday, the 21st, and early Friday, the 23rd. With each new model run came a new outlook for those days. Very frustrating, folks.
These extended range models from Europe, the United Kingdom, the NWS, and Canada are frequently good at capturing major storm systems and jetstream pattern changes well in advance. But it's very tricky business using them to forecast timing and intensity for smaller, fast-moving minor storm systems 7 days out. That is, they can be good for overall trends beyond 7 days, but not for day-to-day forecasts of precipitation. Personally, I hold the idea of a 15 Day Outlook which details daily forecasts--being offered by only one private company (who shall remain nameless)--as being an absurdity. It's bad science, and it doesn't work well.
When there is a well-established, strong pattern in place over North America, we can sometimes forecast with reasonable confidence for temperatures and, less so, for precipitation out to 7-10 days. Last week happened to bring us an unusually unstable set of computer model guidance forecasts for the Holiday, so it was very difficult to have a consistent outlook for Wed-Fri. On the other hand, there have been occasions in past autumns where I was able to forecast a drastic change to a more wintry pattern about 2 weeks in advance. That wasn't because of some brilliant intellectual exercise on my part--it was because of an increasingly more reliable set of extended range model tools meteorologists now have at their fingertips.
But as the timespan covered for a detailed forecast shrinks, a good meteorologist will still always look to the models, but also conduct more individual analyses of the atmosphere based on meteorological education and experience. Ignoring models and doing all things the old fashioned way is a bad idea, but blind reliance on models with no real independent analysis isn't a good idea either. The human factor in forecasting weather isn't about to disappear (lucky for folks like me!).
Many of us who are operational forecasting meteorologists have mixed feelings about a 7 Day, as opposed to a 5 Day Outlook. Accuracy verification drifts downward the further out in time we go. However, there are several extended range models dealing with global forecasting which now go out to 7 days (one from the European Center for Medium Range Weather Forecasting) and one NWS model which goes out 384 hours, or 16 days! The latter model, called the Global Forecast System (GFS), runs on supercomputers every 6 hours. Last week, the GFS seemed to "flip-flop" on timing for a cold front and its precipitation between late Wednesday, the 21st, and early Friday, the 23rd. With each new model run came a new outlook for those days. Very frustrating, folks.
These extended range models from Europe, the United Kingdom, the NWS, and Canada are frequently good at capturing major storm systems and jetstream pattern changes well in advance. But it's very tricky business using them to forecast timing and intensity for smaller, fast-moving minor storm systems 7 days out. That is, they can be good for overall trends beyond 7 days, but not for day-to-day forecasts of precipitation. Personally, I hold the idea of a 15 Day Outlook which details daily forecasts--being offered by only one private company (who shall remain nameless)--as being an absurdity. It's bad science, and it doesn't work well.
When there is a well-established, strong pattern in place over North America, we can sometimes forecast with reasonable confidence for temperatures and, less so, for precipitation out to 7-10 days. Last week happened to bring us an unusually unstable set of computer model guidance forecasts for the Holiday, so it was very difficult to have a consistent outlook for Wed-Fri. On the other hand, there have been occasions in past autumns where I was able to forecast a drastic change to a more wintry pattern about 2 weeks in advance. That wasn't because of some brilliant intellectual exercise on my part--it was because of an increasingly more reliable set of extended range model tools meteorologists now have at their fingertips.
But as the timespan covered for a detailed forecast shrinks, a good meteorologist will still always look to the models, but also conduct more individual analyses of the atmosphere based on meteorological education and experience. Ignoring models and doing all things the old fashioned way is a bad idea, but blind reliance on models with no real independent analysis isn't a good idea either. The human factor in forecasting weather isn't about to disappear (lucky for folks like me!).
posted by Don Paul at
7:48 PM
7 Comments:
I have been very impressed with technology now days where us non professional weather enthusiast are able to look at some of the models via the internet and make our own guesses on what might happen! But as you stated . . . weather is not predictable and can easily change. I personally prefer the 7 day forecast but don’t count on anything until that day!
Is it possible to put a link to the blog in the regular "Your WIVB Daily Weather Report" email messages that go out? The blog is mentioned but a link to it would be really handy : )
I agree with mon@rch. I like checking out the NWS website and making my own judgements based on what I see on radar and in the Buffalo Forecast Discussion page.
Of course, watching Channel 4 helps, too. When they talk about all the technical stuff, it makes sense.
Anonymous: The NWS Forecast Discussion is an excellent product.
We also offer a less technical discussion for our audience on wivb.com on what's coming up over the next 5-7 days, along with a detailed forecast prepared by the 4 professional mets here at the station from shift to shift. In other words, you won't be reading about Omega and Theta-e advection in these discussions, but then ours are offered for the public with a general interest. The NWS discussions are designed for degreed meteorologists. One helpful feature the NWS had added in recent months is to have vocabulary links to some of the technical terms for non-mets with a strong interest and good instincts.
Interpreting radar precipitation imagery is sometimes fairly straightforward, but sometimes far more complex than meets the eye. I'll have more on that in a future post.
Yeah, I do get lost in some of the jargon in their discussion, but I still find it interesting reading. I've always wondered how much interaction you guys have with the NWS. Do you work (wholly or partially) from their models and data, or do you have the means to gather that info on your own?
Hey Anonymous:
For short-term forecasts, like "Today" "Tonight" & "Tomorrow," there are a few other methods to help Meteorologists make a forecast in addition to the models. One way is to look at the current conditions upstream. For example, lets say your forecasting tonight's low temperature and you know that the same cold air mass that was over Michigan last night will be over Buffalo tonight. Just knowing this information can help you decide if you want to agree or disagree with the current model output, which can help you come up with a more realistic temperature forecast. I've also seen panels that show how many degrees MOS was off by (in terms of temp forecasts...which can also help you see if the model is handling the situation well or not so you can better decide if you want to stick with the model or disregard it. (MOS stands for Model Output Statistics, which is a tool many Meteorologists work with that combines climatology & and some of the physical equations other dynamical models are run off of. MOS is a great tool because it eliminates some of the biases built into some of the atmospheric equations, but if something "unclimatological" is forecasted, many times MOS won't pick up on it as well because it is partially based on climatology--- Sorry about the tangent...)
Here's another example to help Meteorologists see if the models are on the right track in a short term forecast. If you look at current radar image of the Northeast for instance...you can then look at what the forecast models were forecasting for the current time period to see how they handled the positioning of the precipitation. If the model had a good placement of the precip its SAFER (but still not guaranteed) to assume that the model has the right idea for the short-term forecast. If the model has the precip in the wrong position from the start, its safe to assume some forecast uncertainty. Also, as a general rule...the further out in the forecast period the less reliable. In other words if you had to make a bet on whether day 1 or day 10's forecast would verify- I'd bet day 1.
To the anonymous poster who asked about our interaction with the NWS:
WeatherWatch 4 has always had a fine relationship with the NWS Buffalo Forecast Office. We don't chat with them that often, but we will occasionally discuss potential major upcoming events. I know what their forecasters are thinking from the Area Forecast Discussion, and since we prepare our own forecasts here, there's no need to take up their time to ask for a spoon-feeding. In fact, my training and experience (and probably that of my colleagues here) was to try to do my own analysis first and prepare a forecast. Then, after that work is complete, if my forecast is miles apart from the NWS forecast, I'll probably go back over the data and models again to see what, if anything, I've missed.
If a meteorologist habitually looks at the NWS forecast first, however good it may be, it tends to make you lazy and bends your thinking from the outset. Ideally, it's best to begin a work week with a clean slate.
Forecasts from one meteorologist to another can occasionally differ greatly, even though we're working with some of the same models. Much of our data is subject to interpretation, and differing levels of experience and knowledge of local climatology. I suppose it's roughly analogous to 2 different radiologists looking at the same X-ray and coming up with 2 different diagnoses. But in tranquil and stable weather conditions, forecasts shouldn't differ drastically.
As for the models, the majority are generated at NWS HQ outside of Washington on supercomputers, as well as a local workstation model generated at the NWS Buffalo office, and another regional model originally developed at the Buffalo office and now used across the continent called BUFKIT. But we also have a brand new version of a workstation high resolution model called Super Microcast which is run here, and another generated by our data vendor in Madison, WI. In addition, we look daily at global weather models generated by the British Meteorological Office and the European Center for Medium Range Weather Forecasting, and a Canadian daily model. An operational meteorologist should be seeking to use an ensemble of model runs along with realtime data, satellite and radar imagery to formulate a forecast. We also try to have some continuity between shifts.
Post a Comment
Links to this post:
Create a Link
<< Home