Friday, March 8, 2013

Model Confusion

All models are wrong, but some are useful
- George Box

But not today
- Wasatch Weather Weenies

Numerical weather prediction models form the backbone of modern weather forecasting and will play an increasingly important role in the future.

Most of the time they are extremely helpful, but there are times when they seem to make the crystal ball even cloudier, and this is one such situation.  We've already seen situations this week where the NAM and GFS have produced dramatically different forecasts (e.g., Which Model Will Win), and we have another one tonight.  

In particular, compare the 6-hour accumulated precipitation in the 0600 UTC GFS forecast with that from the 1200 UTC NAM (sorry about the different initialization times, which is an artifact of availability as I write this, but this is still a reasonable comparison).  

0600 8 March 2013 GFS forecast of sea level pressure,
 surface wind, and 6-h accumulated precipitation valid
0600 UTC 9 March 2013. 
0600 8 March 2013 GFS forecast of sea level pressure,
 surface wind, and 6-h accumulated precipitation valid
0600 UTC 9 March 2013
In particular, look at northern Utah.  The NAM goes for a precipitation maximum that runs up the Wasatch Plateau and Wasatch Mountains to the area around the Cottonwoods, and then it tapers off further north.  The West Desert is dry.  The GFS goes for a gap in precipitation around the Cottonwoods, puts a precipitation maximum over the northern Wasatch, and goes berserk over the West Desert.  

As a forecaster, especially if you are trying to predict what will happen in the central Wasatch, this is sort of like having Clint Eastwood point his 44 Magnum at you and ask you if you feel lucky.  

Diversity in model forecasts can be an indication of forecast uncertainty, which is something we have discussed in previous posts and is important to consider for any forecast.  However, the precipitation these models are producing seems to reflect their physics (i.e., their formulation) more than that driving the real-world precipitation.  Further, the GFS has been overdoing precipitation bands the past couple of days, while the NAM overdid precipitation over the Wasatch Mountains during a similar pattern in early February.  As a result, it's unclear if these models are really all that useful for generating odds of snowfall amounts of various thresholds in this instance.  Sometimes I feel like I can throw the model precipitation out the window and come up with something better based on experience, but this is a weird pattern.  In other words, I'm largely clueless.

So, I'm going to stick with what I said yesterday.  Hope for a few inches today and tonight and remember that it is better to have low expectations and be surprised than to have high expectations and be disappointed.    


  1. That GFS precip. max over the west desert appears to be associated with a blob of vorticity at 500 mb that you can track back to originating near the base of the trough south of Vegas. I tend to not trust those vorticity lobes that shed off to the north because they don't seem to be very predictable, and they are nowhere near as organized as the GFS would have you believe. Just look at the higher resolution NAM or ECMWF 500 mb vorticity. It is much more variable and can change quite a bit between new runs. In generating a precip. forecast, I think it can be useful to look at why the model is producing precip. where it is, which can give you some hints as to how much uncertainty there is.

  2. Putting myself into the model :) ... If the model physics is sensitive to the vorticity blobs, it can be a mechanism for intensifying them, thus producing an unrealistic features in the forecast. For example, lets say you're trying to simulate the position of an intense mid-latitude convective complex (MCC) and the physics of your convective precipitation subroutine is hyper-sensitive and produces too much heating too quickly. The dynamics in the model reacts and produces an area of divergence aloft that can make the MCC blow up way too quickly and you end up with a precipitation maximum that is upstream from the observed max.

  3. I know what you're thinking: did he forecast six inches or only five? But being this is a 40-km model, the coarsest global model in the world, one the Europeans could blow the head clean off, you better be asking yourself, 'Do I feel lucky?' Well, do ya punk?

    1. Ha...good one. The NAM is actually 12 km (there is a 4 km version too, but it isn't used much) and the GFS about 27 km, but the point is well taken. What can I say, the EC is my daddy. I did look at it yesterday and thought it wasn't any more helpful than the others, but I can only access some crude graphics.