Background Material on Model Output Statistics
Background Material on MOS
The standard four-panel progs we introduced in Chapter 17 are generated using data from the GFS and NAM dynamic models. By "dynamic", we mean that these models solve mathematical equations in order to predict how the atmosphere will evolve with time (these equations include variables such as temperature, moisture and wind).
On average, the dynamic models don't predict surface temperatures very accurately (we can say the same thing about surface dew-point temperatures and surface winds). To give you a sense for why the NAM and GFS dynamic models don't predict surface temperatures very well, we point out that physical processes near the earth's surface (such as the cooling and heating of the ground by radiation gains and losses) are just too complicated to mathematically model very accurately. Instead, meteorologists use oversimplified mathematical parameters and schemes to model complicated physical and radiative processes near the ground. Not surprisingly, these "oversimplifications" lead to forecasting errors.
Why, then, are forecasts for daytime high and low temperatures generally pretty accurate? Good question. As it turns out, weather forecasters use statistical models to more accurately predict surface air temperatures, dew points and winds. Here's the scoop. Over a relatively long period of time, meteorologists kept statistics of observed surface temperatures (for example) and the corresponding data from the dynamic models that helped to predict surface temperatures. These data, called predictors, included 850-mb temperatures, wind direction, wind speed and a couple of other variables (these are some of the parameters that govern surface temperatures).
At each airport or weather station, meteorologists developed statistical equations that expressed surface temperatures as a function of these predictors. To ground these statistical equations in reality, meteorologists also incorporated climatological values into the statistical schemes. They also developed similar statistical equations for dew points, wind direction, wind speed, probability of precipitation, visibility, cloud coverage, cloud ceilings, etc. In the final analysis, forecasts based on statistical equations turned out to be a more accurate way to predict surface temperatures, surface dew points, surface winds, etc. at a given airport or weather station (compared to the raw output from the dynamic models).
Collectively, all the forecasts from the statistical equations are called Model Output Statistics (MOS, for short; pronounced "moss"). For the record, the GFS MOS is run four times per day (at 00Z, 06Z, 12Z and 18Z), while the NAM MOS is run twice daily at 00Z and 12Z. Check out the GFS MOS and the NAM MOS for University Park, PA (based on the 12Z runs on December 7, 2009). In case you're wondering, KUNV is the four-letter station identifier for University Park. By the way, we retrieved these MOS data from the interactive Web site at the University of Wyoming.
To interpret the NAM MOS and GFS MOS for University Park, please carefully read the Description of the NAM MOS and the Description of the GFS MOS. The row labeled "N/X" lists the miNimum and maXimum temperatures for the nighttime and daytime cycles during the specific forecast period. For example, check out the GFS MOS for University Park, PA (based on the 12Z run on December 7, 2009). Note that we added some explanation along the bottom to give you more insight into how to interpret MOS predictions for high and low temperatures.
Once you feel comfortable with how to interpret NAM MOS and GFS MOS, you'll be ready to tackle Laboratory Exercise #5.