By Matt Gallagher
Most of you are probably familiar with weather models. For those of you who are not, weather models such as the GFS, NAM, Euro, CMC and others attempt to ‘solve’ for the state of the atmosphere by taking an assumed initial state of the atmosphere and then using computer power to simulate atmospheric processes, with the model outputting projected atmospheric conditions (weather) at a given time and location. Weather models have made tremendous advances in recent years, as both the power of the computers involved and the understanding of the physics of the atmosphere have greatly improved, and indeed, any forecaster worth his or her salt will make reference to weather models in forecasting. However, models are just one thing you should consult as part of your forecast. Future articles will talk more about models and model biases, but today I wanted to start with another tool you should have in your forecasting backpack as we start this challenge – MOS.
MOS is an acronym for “Model Output Statistics”. So what is MOS? How is it different than a computer prog (meteorologist shorthand for ‘prognosis’ or forecast -i.e. a model forecast)?
The GFS, NAM and the other models you are familiar with are computer-generated numerical models of the atmosphere. They’re also referred to as ‘dynamic models’ and use math and physics to solve for the state of the atmosphere. As a result, they’re pretty idealized and can often swerve far away from the actual state of the atmosphere at a single location. They’re vulnerable to two potential flaws: 1) imperfect data and thus an imperfect ‘initialization’ and 2) models that don’t perfectly simulate the processes in the atmosphere.
When we are forecasting for a specific point (such as the weather station at KORD or your house or a specific harbor), a model often varies from the ground truth (observation), even though it may very well be a good representation of the atmosphere over a larger area. There are a lot of reasons for this; some include issues of model grid resolution (the nearest grid point for the model could be miles away) and inability to model microscale features, like a nearby building causing a wind shadow or a black parking lot nearby affecting local wind currents and temperatures.
To attempt to address these inherent weaknesses in models when forecasting for important locations such as airports, meteorologists turn to MOS. At each specific location for which a MOS forecast is available, meteorologists keep detailed observational records over decades and compare how those observations vary from what the model suggests should be true. They then apply statistical relationships between the models’ forecast output and observational truth. The end result is that MOS can often provide a more accurate forecast of key variables such as temperature, humidity and wind speed than can a pure model output. However, MOS requires a lot of work both in terms of keeping accurate observational data and in developing the underlying statistical relationships. This is manual work and thus can only be done at certain locations, generally airports. MOS doesn’t tell us anything for forecasts between those points.
For this contest, MOS can be a great tool for you, as we are forecasting at airports, for which MOS is readily available (in the US). MOS is available based on both the GFS and NAM models, and the data is free.
Here are ways of accessing MOS forecasts:
- Graphically, on the Penn State e-wall
- In text form (more useful for this contest) – The MDL has MOS output based on several models. The most useful for the contest is their short-range MOS. The tables need some interpreting. Spend some time poking around on that site. A key for the tabular forecast data is available and you’ll need to spend a little time in understanding exactly what is what on those tables, but it’s worth it!
- The University of Wyoming also has a good MOS site – This includes both tabular and metogram formats (we’ll explain meteograms in a future post).
A few caveats:
- Dates and times are all in UTC (Zulu) so you will have to convert to local time. Be careful with dates. Our local midnight-midnight actually cover two “UTC days” – the end of one into the beginning of another.
- MOS Is only valid for the location of the forecast. So, for example, the KORD MOS may or may not tell you anything about conditions at Monroe Harbor.
- MOS is quite useful for temperature and wind forecasts, but using it for precipitation forecasting can be more challenging.
- MOS Wind speeds are for the exact time of the forecast. We, of course, are looking for something different – the maximum sustained wind at ANY time during the day. So in most circumstances, you want to take the highest forecast MOS wind and add a couple of knots to get a decent forecast for max sustained wind (but note that this is not ALWAYS true – keep the big picture and climatology in mind).
- The local noon wind speed and direction may or may not be available, as MOS is only displayed for certain times UTC which might not line up with local noon at the station. However, surrounding times might give you a hint of direction and speed at the time you care about.
- Even though MOS is based on models + statistics, it’s not the ‘holy grail.’ You still need to apply your judgment, information from all sources, and keep the big picture setup in mind. MOS is known to ‘bust’ a forecast just like computer progs do. You, as a forecaster, need to apply your judgment and experience to try to minimize the risk of that happening on any given day.
MOS is a very important tool in forecasting for point locations like our airports. However, it is not the beginning and the end, but rather it’s an important arrow in your forecasting quiver.