How far out is too far out?

In response to a post on another thread by another forum member - I wanted to post the following question:

How far out is too far out to begin to forecast for an upcoming system?

I have watered things down here for the benefit of staying objective and perhaps learning something.............

Originally Posted by Sean McMullen

Karen, since when is less than a week away "so far out"? If this was the chase season and we were looking at a major trough sweeping through we wouldn't be talking about how far out it is.

I was surprised that Sean said this. Don't you think that if it were chase season and people were talking about a system that was 7+ days away and not even onshore and being sampled yet yet that there would be a few members of the discussion that would be pointing out the fact that it is hard to forecast what a system will do this far out???

It seems to me that in the past four years or so of forecasting posts being made on ST, I recall many, many assertions about systems being too far out to forecast accurately.

I was merely trying to point out that we will know much more nearer the time and it is - IN MY OPINION - very hard to forecast winter WX even one day in advance let alone one week.

For me, the 2005 and 2006 chase seasons were both heavy excercises in long-range forecasting, since I had to coordinate a long-distance chase trip with 2 to 4 people coming from different areas of the country. Plane, train and car travel was involved in our trips, so a last-minute departure was going to be impossible. Long-range models were all we had to go by to plan.

While it was a hair-pulling, stomach-flopping roller-coaster ride (Jason Persoff called it a 'stressfest', AKA long range planning for non-Plains chasers), we were able to work our way through the mess of model data to pick the right time to leave and the right times to hold off. During the death ridge in mid-May, a big trough showing up on day 7 was reason to get the 'crew' on alert. As the event drew closer, it wasn't quite as impressive as it was when it first showed up on the long-range radar, but it was still helpful.

When spring comes, the long-distance chasers need to have *some* idea of what to expect, when to buy plane tickets, when to take off work, etc. While the GFS certainly has let us down a few times, a large-scale pattern consistently showing up in the long range is enough to at least tell us we need to 'mobilize' for the possibility. So my answer would be that long-range models do have some limited practical use in chasing.
Last edited by a moderator:
How far out is too far out to begin to forecast for an upcoming system?

Seems the models are generally pretty accurate with development of a system to about 200 hrs out.

I recall many, many assertions about systems being too far out to forecast accurately.

When you bring up the question about an accurate forecast its hard to say, I guess it depends on how broad your definition of accurate is, but I would say you can accuratley forecast the impacts from a winter storm about 3 days out for a given region. After reviewing the thread you take these quotes from, I see no evidence of anyone becoming to particular with their forecasted impacts from next weeks system. I think Sean M was right less then a week is by no means to far out to forecast the impacts from a winter storm.
Last edited by a moderator:
In my own defense, I said less than a week, not 7+ days out. Look at the thread, if you can, look at the date we posted those comments they were clearly less than a week away not 7+. Anyways, it looks like we're going to get that weather, it may be a little farther north than I wanted but it looks like its going to happen.
My best advise when it come to reading long range models is watch for consistency. I remember back in 1991 what was called the MRF model was pretty accurate with bringing a big trough into the western U.S. more than a week in advance of the April 26th tornado outbreak. The models were so good I flew out from Florida for a quick chase with a three day advance purchase for the flight out. 2003 was another good long range forecast more than a week in advance. Some years the models are real accurate but there are some years they are terrible. You need to look at all extended models not just the gfs.
This system is so far out and winter wx gives us SO many notorious problems forecasting it that I think we would know a lot more if we waited until after this weekend. As of now - according to just the one run of the GFS - it looks like an all-rain event for anyone south of KS.


What I always find funny is that often those thinking it is too far out and say so, will many times comment on it anyway! I mean if it's too far out for folks to comment on/bring up why is ok for you to comment on it? It's either or it's not. How can you even say it's a system that will be here to rain, if others can't mention the possibility?

I think an ant can realize one will know more as time progresses. This is one of the most annoying things in chasing, imo. You always get to hear, "that's kind of far out there"....."the gfs fantasy progs". As if anyone is really being informed by that.
Last edited by a moderator:
How about this one - always a classic "We will know more after the next model run"

Models are not at all good with systems at 200 hours out... They can be good with patterns. ENSEMBLES. I beat them up hard. Anyone relying on just one specific run from one model will lose more often than not. When we're in a death ridge you can predict out to 7 days reliably. When we're in a pattern like this - it's tough to go more than two days.
Apparently 168-180 hours is close enough for the Weather Service to at least start briefly discussing the possibility of a winter weather event (see 168-180hr GFS).
They are forced to do a 7-day hazardous weather outlook, but they don't (can't) get too specific. I'm not sure what you mean by the GFS - that goes out to 360 hours.
They are forced to do a 7-day hazardous weather outlook, but they don't (can't) get too specific. I'm not sure what you mean by the GFS - that goes out to 360 hours.

It was in reference to the first sentence in my post.

OUN put out a SWS about the potential for freezing rain Monday night into Tuesday for the NW third of Oklahoma...and then mentioned the potential for a snow event for the days preceding Christmas (at the point it was the 168-180hr forecasts...right now, it is 144-180)
The question of the thread is how early is to early to begin forecasting for a system, not how early is too early to get specific!

When you are required to do a 7-day forecast, it doesn't matter if it's able to be done or not. You do a 7-day forecast, whether or not it can...
Too far out is probably about 10 days out for all dynamic models, at least statistically speaking you could look at climatology at that point and probably get an equivalent quality forecast of what to expect, or using teleconnection patterns might be a slight improvement over climatology as given in the climate forecasts (which don't use dynamic models, and have questionable skill except for under the strongest of patterns). As far as forecasting, what probably matters most is looking at a coarser scale as you look further out in time. I would not recommend using any one model to try and forecast beyond a few days, but instead look at several, or perhaps several different ensembles. NCEP gives a nice discussion on medium range forecast model differences and trends, and from that expected confidence of the forecast is often discussed. The language can be cryptic for the common reader depending on the forecaster writing the discussion, but read it for a while and you'll start to pick up on some of the common language.

Ensembles can be helpful in identifying the best fit forecast (ensemble mean, which gets progressively deamplified with time) and forecast confidence (as ensemble spread grows, ensemble mean forecast confidence declines). However, the reason spaghetti plots remain popular as well is it doesn't get as smoothed out and gives you some flavor for possible evolution scenarios.

Each global model is of course a gross simplification of what happens in the real atmosphere, and different models have different sets of simplifications that lead to certain models not doing well with some weather patterns, whereas another model might handle it better. Forecasters that look at these model outputs every day are quite skilled at recogninzing what is a believable model forecast from on that is likely junk. They know whether a Rossby wave pattern of three waves or six waves has greater predictability, or if there was a problem with the model ingest for a particular run, etc...

Thanks to climatology every once in a while a model forecast way out in the far distant future of a particular lee cyclone developing in the plains will seemingly pan out (though if you follow the finer details of the evolution to get there from the original forecast time of the model, it was probably not very accurate), and that probably gives prognosticators some confidence that they can trust model forecasts 10+ days out. My personal feeling is not to trust timing of a synoptic scale event to any particular day beyond 5 days out. Beyond this time, there may still be some skill in the Rossby wave patterns - so looking at the global weather pattern.

For severe weather events, one check you can make whether to believe the model forecast is to find the time the model is forecasting the event to develop and identify the triggering shortwave. Then, going backward in the forecast (toward the initial time) track the shortwave feature until you get back to the initial model time. Often, you will find it very difficult to confidently track such a small wave feature back more than a day or two. But, if you can get back to the initial time, you can then look at water vapor animations around the initialization time and try to get a feel for if the wave feature is 'real' or just a model artifiact, as well as whether the model has the feature in the right place, and if the forecast is short enough in length the wave might be in the CONUS observation network and you 'll then have other observations to check if the model initial condition had a good estimate of the wave intensity (too weak and the model wave will move to fast and vice versa).

With good model consistency and agreement you can get some idea as to what may happen in the 7 to 10 day range. ie "The southern plains looks promising eight days out." It's when people try to narrow their forecast down that it becomes a little absurd. "It's gonna go nuts somewhere between Elk City and Altus in eight days". Watch the CPC 6-10 day and 8-14 day forecasts for a few weeks. Sometimes they're good, but it can also flip flop back and forth over a few days time.

With that said, for those of you wanting to plan your chase vacations now, here's a helpful link. It's never too early to start looking at model data. :)

This will get you up to 277 days out, or about mid September, 2007.
IMO the obvious answer is there is no definitive answer. Meteorology is an inexact science to begin with. Seems the deciding factors on when is a good time to start getting serious would be personal, i.e. how far away you are from target, whether you're interested in long range forecasting or not, what time of year it is, etc etc.

My answer? 24hrs before the event. It's going to change constantly anyway, and because I have no interest in forecasting (beyond finding my target), there's no need to get worked up 3-4 days out. But I still do anyway sometimes.
Dont have too much to add; as Jim L. said consistency is one good thing to look for, also the particular pattern that is forecast: a giant death ridge is more likely to be a stable solution than trying to pin down a succession of progressive waves.
One thing to be very aware of though, is that as some have mentioned, even if you are able to accurately predict the pattern that may be useless toward pinning down the details: case in point, the blizzard that hit my area last couple days. The h5 pattern was accurately progged by the ECMWF (not the gfs tho) over a week ago, but 24 hours before the event, the NAM was forecasting significant snow on tuesday, not much on wednesday. We were dry slotted on tue and got next to nothing, Wed was the blizzard. If you can't trust a 24 h. prog, be wary of your 7 day forecasts!
Trends and evolution of the upper jet structures is what I look for the most when peeking way out into forecast space. Sometimes this gets hampered by the limited panels on the Euro and Canadian models that forecast out to 168 hrs. or less. GFS works for latching on to trends, but often seems to like to cutoff stuff at unusual times, and then the next run it shows an awesome progressive trough. As Shane said this is truly an inexact science. I am more often than not quite guilty of being too optimistic with things, but I've learned to watch ensembles closer and to fall back on the others (Euro & Canadian) to see if those patterns are also being signaled. It's all a matter of planning more than anything in my use of these extended models. Last spring was prime example.