Let’s say you read this blog and watched the local 6 pm evening news/weather on TV. There’s some snow possible on Saturday, five days (120+ hours) in the future.
You’re thinking…I want more information about this possible storm. If I watch the 10 or 11 PM news/weather tonight, will I get more insight or an updated forecast for Saturday?
The answer is no.
Let me explain—
When it comes to forecasts for the 120+ hour time frame, there are really only three models that go that distance into the future: the GFS run four times a day, the Canadian Global CMC, and the European ECMWF. The Canadian and ECMWF are only run twice a day.
A new GFS 120 + hour forecast will become available about 11:10 PM, the new Canadian about 11:25 PM and the ECMWF around midnight.
Add some time to review the data and an updated 120 hour+ forecast simply isn’t available for the late 10 or 11 PM news/weather.
Yes, there are other important statistical “ensemble” models based on these three, but their data is available even later!
(Another good model, the German ICON model, just goes to 120 hours. The U.K.’S UKMET and the Navy NAVGEM are rarely worth looking at.)
So, when looking at long range forecasts (five days or more) where one wants to consider more than just the GFS, meaningful forecast updates more frequent than every 12 hours just aren’t possible.
Here is some basic weather information for understanding my blog:
GFS model is a global weather model and is the “Global Forecast System” model. (It used to be called the AVN/MRF model (AVN=aviation, MRF= medium range forecast model.)
The GFS is a global – it models the entire globe. It’s a relatively coarse resolution model, not as fine a resolution as the NAM described below, however it is very advanced. The GFS is run four times a day. While not as ‘fine’ a resolution as the NAM model, the GFS model has wonderful features and is the one I usually bet on. The GFS is constantly improved with regular ongoing updates.
Another subset of the model, the GFSX (X for extended) also predicts over longer ranges, as much as 144 hours to 384 hours into the future.
NAM stands for the North American Mesoscale model. Previously known as the Eta model, it was renamed on 1/25/05. It models North America only.
On 6/20/06, the NAM was converted from using older Eta physics to WRF (Weather Research Forecast) model physics. There are also variations of the NAM model which use different physics for their initial conditions. As of 2018, development of the NAM has been frozen.
As mentioned NAM has different physics than the GFS and is a higher resolution model. It attempts to predict in areas as small as 3 km in area. In meteorology, that scale is referred to as a “mesoscale” area. (Thunderstorms are mesoscale weather events.) Higher resolution doesn’t always mean more accurate.
The NAM is not a global model and only predicts for North America. It’s run four times daily and the usual form predicts 84 hours into the future. Experimental extended versions are being developed.
NAVGEM is the model developed and used by the US NAVY. A global model, somewhat similar in coverage as the GFS, but different physics. Run four times a day. I used to find it useful for hurricane prediction, but recent changes in the model haven’t been good.
GFS MOS and the NAM (Eta) MOS: MOS stands for Model Output Statistics. MOS data are statistically based forecasts for up to 84 hours (NAM and GFS) or 144 hours (GFSX or GFS eXtended).
The MOS predicts temperature, humidity and precipitation probabilities every 3, 6 and 12 hours, depending upon the specific MOS output . MOS forecasts use historical data and reinterpret the raw data for specific locations. However, MOS forecasts do not correct for model biases.
RAP: Rapid Update Cycle model. Based on the NAM model, this model is a short term model that is updated hourly.
HRRR: High Resolution Rapid Refresh Model. Hour by hour forecasts extending 18 hours into the future.
GFS LAMP Forecast: A short range hourly forecast, rerun every hour and based on the GFS, extending about 24 hours.
Ensemble Forecast Models: Statistical treatment of modeling, with 20-40 different versions (perturbations) run to see the range and spread of different possible outcomes.
National Blend of Models: A statistical post-treatment of several different models and their statistical versions. Under development.
Other country models include the Canadian GEMS, UKMET, ECMWF.
All models have their specific biases, inaccuracies, etc. New forms of each model are constantly being developed and tested.
Some acronyms and other information:
QPF is “quantity of precipitation falling”. Most of the models predict the amount of precip that will fall in a given period of time, based on the total amount of available moisture that can precipitate (PWAT = precipitable water) and the physical conditions (lift, convection, etc.) that will cause it to precipitate. The amount of snow (snow:water ratio) is usually calculated by multiplying the QPF in inches by a factor of 12-20, depending upon the temperature.
Atmospheric Thickness Levels: Heights in the atmosphere are often measured based on where the pressure is a constant value. The ‘thickness’ of the atmosphere is the three dimensional depth of the atmosphere between two pressure values. A useful thickness value is the thickness (or depth) of the atmosphere between the pressure of 500 millibars (mb) (about 18,000 feet) and 1000 mb (millibars) (about the earth surface near sea level). The thickness values become higher when the upper atmosphere is warmer and become lower when the upper atmosphere is colder. Thickness values are useful in predicting rain/snow or sleet. They correlate with temperatures at specific heights that are correlated with snow or sleet or rain.
Useful Temperature Levels: 800 millibars and 900 millibars- Both of these levels must be at or below 0 degrees C (freezing) for snow to form.