Dangers of Forecasting with MOS

In a nutshell, MOS is a model that gives at-a-glance forecasts for specific locations.
Although it’s an easy way to make a forecast, MOS has many known biases and issues.

Bullet points…

• MOS has trouble with extreme weather events.
• MOS underestimates cold air under numerous scenarios.
• MOS has problems with forecasting wind speeds.

Model Output Statistics, commonly referred to as MOS, is a computer model “technique” of weather forecasting. It basically combines numerical weather data with statistics, to forecast weather conditions near the ground. (Numerical weather models focus most of their forecasts from what’s happening up in the atmosphere)

A MOS forecast looks something like this:

While MOS is a great tool for making a quick forecast, since everything is right at your fingertips, there are many “dangers” of forecasting with MOS.

Forecasting temperature:
Today is a great example. MOS performed very poorly with short-range forecasts.

1. The last three runs of the MAV MOS all predicted a high of 36F at IJD/Windham Airport for today. Actual high: 41F+. (error of at least 5 degrees)
2. Same as above, but for BDR/Sikorksy Airport. Last three forecast highs for today: 41, 40 and 40. Actual high: 44F+. (error of about 4 dergees)

We’re talking about forecasts only 12 to 36 hours out, so you’d expect a model to not be 4 or 5 degrees off.

Problems with climatology (climo):
MOS is skewed to “average” weather conditions. This doesn’t mean that it’s always too close to normal, but there are countless examples of this happening. With long-range MOS, the climo-bias gets far worse. As MOS forecasts further out, like day 4-5 and beyond, the data is even more skewed to climo. Although it may be good in some circumstances to forecast closer to climo for, let’s say a 7-day forecast, there are other times when it can really hurt you.
If a major warm-up or cold outbreak is expected, MOS is almost always going to be skewed too close to climo.
Does MOS ever predict record high/low temperatures 7 days out? Almost never. If it does, it’s usually an extreme signal that temperatures could be way above/below average.
Think of it as a conservative friend that usually leans towards the middle. If even they stray far from the “norm,” then you know it’s a big deal.

Problems with warmer and cooler air-masses:
It’s documented that MOS has a bias to underestimate the strength of approaching cold and warm air-masses. Part of the cause comes back to the skew towards climo. Also, the GFS MOS, for example, is not an extremely high resolution forecast model. It has a tendency to “broad-brush” air-masses, often leading to skewed temperature forecasts.

Prime example – Cold Air Damming (CAD):
MOS is notorious for being terrible at forecasting CAD.
I have seen times where MOS guidance, especially the lower resolution MAV MOS, has been 5+ degrees too warm during CAD events. This could be the difference between 30 degrees and a major ice storm, and mid-30’s with non-freezing rain. I’ve also seen forecasts must in that case.
One of the problems is that cold air near the surface may be relatively shallow and MOS is not all that good at resolving finer details like that.

Precipitation causes temperature problems:
In almost every snow storm I’ve ever tracked, MOS has been too warm with temperatures.
Why does this happen?
Well, if I see heavy snow forecast on a model, but MOS has a temperature of 34 with a dew-point of 25, there’s a huge red flag right there. MOS is not always good at evaporational cooling. Although MOS may accurately predict dew-points in the mid-20’s, for the example above, it does not cool off actual air temperatures enough. The end result is that during most significant snow events, you will notice that temperatures end up colder than predicted by MOS.
Even in a case where a rain event is expected, the same happens. A good example is a cold-season rain. The MOS may show a large difference in temperature compared to dew-point (dew-point depression) and the end result is an actual temperature that is at least a few degrees cooler than predicted by MOS.

Other problems with cold air:
Low temperatures – MOS is often too conservative with radiational cooling. Again, this may have to deal with a shallow layer of cold air or a small-scale event, but with that said, if strong radiational cooling is expected, make sure your forecast is colder than MOS.
Snow-cover – Snow on the ground, especially a fresh snow-pack, can also cause lower than “predicted” low temperatures. Studies show that temperatures often do trend colder when there’s snow on the ground and MOS is observed to also have trouble with this.

Wind speed:
Although temperatures have been the main focus here so far, MOS also has trouble with wind speeds. For whatever the reason may be,
MOS is almost ALWAYS too low with predicted wind speeds.
This becomes most evident when a strong wind event is predicted. It’s not uncommon to see MOS predict wind speeds of 15-20 MPH, when actual wind speeds end up being over 30 MPH sustained. Gusts can be much higher than MOS, since MOS predicts sustained wind speeds)
However, even on relatively calm days, MOS is generally too light with wind winds.
I participated in the WXChallenge national forecasting contest for three semesters. In the vast majority of cases, observed wind speeds were higher than predicted by MOS. Knowing this helped give me an advantage when it came to forecasting winds – go higher than predicted by MOS!

Beating MOS:
One of the goals of my ongoing research project (that I started in college) was to come up with a forecasting scheme that could literally beat MOS. Essentially, my research used 850mb temperatures (above the ground) to estimate temperatures near the ground. Although the scheme is not perfect, it has often beaten MOS and sometimes significantly.
When I get time in the future, I’d like to make a more detailed article about this forecasting technique.

What about LAMP MOS?
That MOS tool is near-term forecast model that provides hourly forecasts for 24 hours. Although LAMP sometimes catches onto trends better than the “regular” MOS, it is also prone to large errors.

In the end, MOS should be used as “guidance.” It’s not perfect and no computer model is. MOS has its benefits and while it can help make a forecast, it should not be used as a “plug and chug” tool. If it is, be prepared to have many busted forecasts!