Learning to accurately read the ensembles for winter weather prediction

[Jeff Miller]

First off, can a mod PLEASE take the J out of the "THE" in the title? MOD: Fixed!

It's a long ways out so I'll put this in here as I wasn't quite sure where to go, since forecasts aren't allowed here- as this is an education thread for me and hopefully anyone else attempting to perfect the use of ensembles.

I have been looking at the GFS long range models and have noticed extremely cold air massing up in central and northern Canada. Lows below -20F look to be common in those areas next week, but it looks like the true mass of arctic cold will remain bottled up there from what I can deduce now. The just of this is I'm trying to accurately read the ensembles for winter weather prediction, I have avoided it, but I'm stabbing at it, and here's what I have right now:

Looking at the spaghetti ensembles, and it *appears* to be stating that in the 192-384 run for temps below freezing at the 850, 06Z Nov 18, that the cold air is predicted, by this ensemble, to spill well over the northeast rounding a ridge in the central plains. Am I correct here? Looking at the precipitation spaghetti snowfall chances that seems to support a good swath of lake effect snows too, later in the forecast period.

What I am deducing from this ensemble package is that the cold air is predicted to be held at bay for the plains and spill over to the midwest and northeast with a cold wave.

So here then are my questions:

1. How much should I rely on ensembles (spec. spaghetti) for winter weather prediction.

2. How many runs should I observe before beginning to trust a particular ensemble model? Are flip-flops common with ensemble forecasting,and are there any biases I should extrapolate into the equation?

3. Am I close with my prediction or way off into left field?

4. What key points would you give someone attempting to get into the use of the spaghetti ensembles for winter weather forecasting?

 

[rdale]

1) MUCH more than you should any individual model. Even better - ensembles composed of different models. BUT remember that ensembles get rid of outliers, and often the biggest storms come from outliers.

2) There's no "xx runs in a row = likely outcome" any more than you can say "00Z and 06Z runs both have tornadoes tomorrow so I'm packing"

3) Not bad, but again you're using the GFS which by default brings down colder temperatures than verifying quite often

4) Use ECMWF or the Canadian

 

[Paul Knightley]

Ensemble means can be very useful, and indeed should be used (IMO) beyond about day 5 or 6 as opposed to a deterministic run. I use them operationally here in the UK when speaking to energy traders, and consistently find that using a blend of the GFS and ECMWF ensemble means, alongside trends in the deterministic models, is by far the best ploy.