Probability of Aviation Weather

sarangan

Pattern Altitude
Joined
Jun 7, 2008
Messages
1,896
Display Name

Display name:
Andrew, CFI-I
One thing I noticed in aviation weather products is the lack of confidence figures. I know TAF has PROB in some forecasts, but the longer range products such as Prog Charts do not. What probability can we attribute to the predicted frontal positions on a 7-day prog chart?

I am wondering if anyone has done a study by comparing 7-day prog chart to the 3-day and 3-hour chart to see how well they overlap with each other. I was going to do this study myself by downloading historic charts, but would like to know if such a study already exists.
 
I am not a meteorologist, but there are ensemble forecasts that can give you a feel for the uncertainty/sensitivity to initial conditions for a given forecast product. There are also tons of studies on forecast skill that compare forecasts to what actually happened, but those will probably put you in NWP research territory rather than operational forecast products.
 
Here’s an exercise, look at a TAF at the end of its forecast period, print it, then compare the same hour or two as you get closer to the target time. Just about always, the details are fine-tuned as the hour in question moves closer. It’s even more pronounced with a sorta large system nearby.

If that happens at 18 hours, what can one expect 5 days out?

Some sites show 10 day forecasts, not to be taken seriously. I think it’s best to just look at high & low pressure systems days out, even they will be adjusted. This is why alternate fuel & a plan, is always a good idea.
 
Hmmmm... a weather scientist has been known to wander into these forums every now and again.... Might be a good question for him.
 
Hopefully talking about NWP will summon him to shed some light, but I'll try my best as an interested amateur.

With a regular forecast, you start with your best estimate of the current state of the atmosphere, press go on a weather model, and get an estimate of a future state of the atmosphere. Inside those models are differential equations, which may yield a different answer if you change the initial conditions a little.

An ensemble forecast is where you take, say, 10 copies of those initial conditions, perturb each of them a little, and then run 10 copies of your weather model, giving 10 answers. Now you can get a feeling for what the "average" solution looks like as well as what the spread looks like.

Example: GEFS
 
Last edited:
Hopefully talking about NWP will summon him to shed some light, but I'll try my best as an interested amateur.

With a regular forecast, you start with your best estimate of the current state of the atmosphere, press go on a weather model, and get an estimate of a future state of the atmosphere. Inside those models are differential equations, which may yield a different answer if you change the initial conditions a little.

An ensemble forecast is where you take, say, 10 copies of those initial conditions, perturb each of them a little, and then run 10 copies of your weather model, giving 10 answers. Now you can get a feeling for what the "average" solution looks like as well as what the spread looks like.

Example: GEFS

Correct. And the spread in the ensemble gets larger and larger as you extend out in time, as small differences in the input data result in potentially large differences in output when extended over longer time periods. At some point the spread gets to the point it is meaningless for forecasting local conditions. The accuracy of such projections in the long term are very sensitive to small variations in the input data, which is not necessarily very precise, nor it is finely spatially gradated. This is not a flaw, but feature of the mathematics involved. The bottom line is that a 7 day prog chart is unlikely (near zero percent probability) to resemble in detail the actual prog chart when the time arrives. The general trends and large-scale features are instructive and potentially useful that far out, but not the fine details. The current limits of reasonable atmospheric forecasting, given the quality and spatial distribution of input data and the ability to run computer models in sufficient time, is about 72 hours or so. You can stare at the atmospheric models 7-10 days out to plan your VFR trip, but it will never work out that way. Something will be off...often the timing but also the details. Sometimes the input data is not even sufficient to successfully model even 72 hours out, e.g. forecasting the exact track of a imminent nor'easter, where 50 miles E/W difference means everything to a local or regional WX forecast.
 
Last edited:
I've never tracked that specifically but frontal position errors several days out are going to be on the scale of +/- the fuel range of your average bugsmasher.

NHC does track hurricane positioning errors out to 5 days and even with those it's a couple hundred miles usually.

Let's put it this way, 5 days out a typical low in the Midwest is often still out in the Pacific and hasn't been sampled by anything other than a satellite. Gotta be over land to get sampled by radiosondes (wx balloons), surface observations for temp/winds/pressure, radar, etc.
 
I can predict that 100% of the time, there will aviation weather that someone does not like.
 
Maybe so, but the three day forecasts are now about as good as 36 hour forecasts were 30 years ago. NWS recently upgraded their supercomputers and satellite data continues to improve, so I'm optimistic that the longer term forecasts will improve as well.
 
Weather forcasting..... You can be 100% wrong all the time, and still get paid.
 
We don't call them "weather guessers" for nothing.

I watched the weather forecast for today starting early in the week (neighborhood clean-up day). Early in the week the TV weather guessers were sure it was going to rain. The forecast kept getting better as the week progressed and we did not cancel today. The weather this morning was just fine. A bit cool, but sunny and dry. We had a squall come through late this afternoon, but we were all finished by then. Calm and dry this evening.

In short, don't worry about the forecasts a few days out. Odds are, they'll change by the time the actual date arrives. And, probably for the better.
 
One thing I noticed in aviation weather products is the lack of confidence figures. I know TAF has PROB in some forecasts, but the longer range products such as Prog Charts do not. What probability can we attribute to the predicted frontal positions on a 7-day prog chart?

I am wondering if anyone has done a study by comparing 7-day prog chart to the 3-day and 3-hour chart to see how well they overlap with each other. I was going to do this study myself by downloading historic charts, but would like to know if such a study already exists.

Call flight service and ask them to give you a 7 day briefing and see what they tell you.
 
Weather forcasting..... You can be 100% wrong all the time, and still get paid.
In Juneau, they upgraded the forecast office and put a window in. What you see is what you get...:rolleyes:
 
Aviation forecast discussion of Aviationweather.gov is a good source for the forecasters confidence level.
Agreed - glad these didn’t go away when Area Forecasts did......BUT, I notice that they vary widely from center to center in their depth of discussion....western Virginia (Blacksburg?) has generally been quite thorough.
 
Agreed - glad these didn’t go away when Area Forecasts did......BUT, I notice that they vary widely from center to center in their depth of discussion....western Virginia (Blacksburg?) has generally been quite thorough.
I’ve found the discussions to be one of the most useful things I read.
 
Back
Top