Understanding The Weather Models

On a daily basis, you (the reader) scroll through our forecast posts, and gain insight into what to expect today, and what weather to expect in the days to come. But how do we as forecasters get to this point? Well, we use a combination of observations, meteorological knowledge and model guidance.

h/t nea.gov.sg
h/t nea.gov.sg

When I am referring to model guidance, I am referring to the different kinds of weather models that we go through to get a picture of how the atmosphere will behave. The different kinds of models, what they do and how they are biased can be confusing. This post will help sort that out, and make the models a bit easier to understand!

Wait, what are weather models?

Great question! Weather models are basically simulations of the atmosphere. They are developed and run by a series of meteorological formulas that are solved when observations are put into them. The model is only as good as the initial observations, so verifying initialization of a weather model is key.

h/t tenor.co
h/t tenor.co

Once these observations are input, the model is run using those formulas. The solution to those formulas are what we see in the weather model, and thus what we use to forecast the weather!

One of
One of the solutions that the NAM spit out over the weekend for the wintry event we saw. h/t pivotalweather.com

Most weather models are run at 00z, 06z, 12z and 18z. One of the major global models is run only at 12z and 00z, but I’ll get to that in a little bit. While these are mostly run at these times, the 06z and 18z don’t use fresh observations. They use the 06z and 18z forecasts from that model at 00z and 12z, respectively. Therefore, if the input to the 00z/12z runs was bad, then it likely won’t be better at 06z/18z.

All of the sites in the continental US that are used for upper air soundings. h/t SPC
All of the sites in the continental US that are used for upper air soundings. h/t SPC

Why are fresh observations plugged in twice a day, one may ask. Well, that is another great question. Upper air soundings are taken at 00z and 12z at 92 stations in North America, and 69 stations in the CONUS. Because these soundings give critical data about the upper levels of the atmosphere, the models can’t be run without them. That is why they are only run twice a day with fresh observations.

What kinds of models are used?

There are two major types of weather models: Global models and mesoscale/regional models.

Global models:

  • Use the Globe as their domain
  • Use global sounding data, and have a coarser resolution
  • They are mostly grid point models
  • Inclues: GFS, ECMWF

Mesoscale/Regional Models:

  • Are higher resolution
  • Are on the continental to subcontinental level
  • Their domains have edges
  • Grid point models
  • Includes: NAM, WRF, RAP

These models are vastly different in their biases, strengths and weaknesses. I’ll give you a bit of a guide into them, and how I prefer to use them in my forecasting.

Global Models


  • Global Forecasting System
  • American model; goes out to 384 hours
  • Biases: Tends to overhype systems in the long range, only for the system to be weaker; over-deepens troughs
  • May or may not be known as a hype machine
  • Verification is good, but not great


  • European Center for Midrange Weather Forecasting model
  • The primary European model; goes to 240 hours
  • Biases: it has too many cutoff lows in the southwestern US, high height biases in upper-troposphere, and over-amplifies upper level systems
  • Verification is very good
  • May or may not be considered the best model in the world

Global models are used in the short term, yes. However, the coarse resolution of global models makes them difficult to trust in short-term scenarios, especially with complex issues. Global models are used a lot to try and pick out trends int he upcoming pattern, and to trend systems in the 5-7 day range. In the 3-5 day timeframe, they’re the primary source of forecasting (as well as their ensemble forecasts).

Mesoscale/Regional Models


  • North American Model
  • The US mesoscale/regional model
  • Regular form is in 12 km resolution, but the Hi-Res version is now up to 3 km
  • Goes to 84 hours out; only for the US
  • Biases: Flip-flops in the 72-84 hour range, has near surface issues and can struggle with cyclogenesis
  • I love the 4 km NAM; the 3 km NAM is recent, and is still having kinks worked out


  • Rapid Refresh Model
  • Very Higher resolution US model
  • Run every hour; Goes to 18 hours
  • Biases: It is bad in the mountains


  • Weather Research and Forecasting Model
  • US model
  • Very high res; on the regional level often times
  • Goes out 36 hours in some models, in others it is shorter or longer
  • Biases: Too low on Southeastern temps, too much precip, spin-up issues with lows at 12 hours

Higher resolution models are where short term and mesoscale forecasting gets its bread and butter. Mesoscale forecasting is a combination of very high-res models and observations, but short term forecasting (1-3 days out) makes heavy use of these models. They are helpful in sorting difficult forecast details, and determining what may undercut, help or eliminate a certain forecast outcome.

Weather models are awesome, and are very helpful in determining what is going to happen in a given forecast. However, weather models aren’t everything. They are called guidance for a reason. Forecasters go to meteorology school to learn how the atmosphere behaves, and how things work within it. This training and education is most often used in combination with the model guidance, and not just on models alone. See, understanding weather models isn’t as complicated as you might think!

h/t giphy.com
h/t giphy.com