Here's an example. Yesterday's GFS was calling for major upper-level trough to move over the western U.S. and bring an extended period of cold weather and snow to northern Utah. Below are the 156 and 180 hour forecasts valid 6 PM MDT next Tuesday and Wednesday. The trough is right over Utah and we are very cold and snowy. Release the hounds, let's go skiing!
Now let's look at the more recent GFS forecast valid at the same times. The trough structure and position are significantly different. It is more compacted (i.e., narrower) on Tuesday afternoon and more progressive, so that it is located further east by Wednesday afternoon. This forecast would still be a snowfall producer, even for the valley floor, but accumulations would be smaller and the event less long lived.
How about we take a different perspective. Let's look at a bunch of forecasts, in this case produced by the Global Ensemble Forecast System (GEFS), which is based on the GFS but run at a lower resolution and with slightly different configurations, and the Canadian Ensemble Forecast System (CMCE) and see what we find. Given that these are lower resolution models, we'll use some trickery to build back in some of the terrain effects that occur in northern Utah (this trickery is known as downscaling). The GEFS and CMCE ensemble combo is known as the North American Ensemble Forecast System or NAEFS.
The accumulated precipitation produced by these downscaled NAEFS forecasts at Alta is shown below. You can see last nights' minor event, which is followed by about a 3 day period during which none of the models produce precipitation. After 0000 UTC April 12th (6 PM Saturday), a few of the forecasts generate some precipitation, but most keep us dry. There is a system that passes to our north and those model runs that keep it a bit further south give us some brush by precipitation. Finally, we have the storm for next week. Here you see huge model spread. Everything from a major storm to practically nothing.
The reason that these forecasts are so divergent is that the intensity, speed, and position of the upper-level trough and its associated precipitation are very sensitive to the initial conditions given the model and the physics of the modeling system. Further, one can see a common characteristic of multi-model ensembles - the tendency for solutions to cluster by modeling system. Note that the CMCE members are mostly dry and the GEFS members are mostly wet. This is a bit unusual for this winter when the Canadian was often more bullish on snow.
If one looks at long-term statistics, the mean of an ensemble performs better than any individual ensemble member. In addition, the mean of a multi-model ensemble does better than the mean of any single model ensemble system. The thick black line above shows the ensemble mean, which for the storm next week gives us about an inch of water and 15 inches of snow. That would be nice! On the other hand, in any individual storm, any one member could come through and, in some instances, what verifies is actually outside the envelope of what all the members produce.
So, if you want a specific forecast for next Tuesday and Wednesday, no can do. There is pretty good agreement that we'll get a cold trough passage with the potential for a good storm, but ultimately there's a wide range of possibilities for precipitation and snowfall amounts. My momma always said snowstorms are like a box of chocolates. You never know what you're gonna get, at least in the extended range.
Addendum at 11:55 AM:
A humors look at extended day forecasts passed along by one of our readers:
Personally, I think contemporary extended forecasts have utility and are likely to continue to improve in the future. The devil, however, is in the details.