Forecasting weather relies on a combination of pattern recognition, computer model guidance and intuition. However, when it comes to computer models there is a lot to choose from and several models that don’t perform very well.

A computer model is a computer program that is made to take the seven main equations that measure the atmosphere and turn it into an algorithm that can project weather variables into the future. Computing weather equations involve a tremendous amount of computation time and the size of the domain (the map on which you’re forecasting the weather) is also large enough that it compounds that time into a tremendous drain on super-computing resources.

If we were to use the full equations and write a computer model using no approximations, we’d end up with a program that takes several months or years to run for a short-term forecast. Obviously, that’s too late. So, we force the equations to run faster by approximating terms and simplifying the equations where we hope to get a very close answer, while significantly reducing the time needed to run the program.

Since there are several ways of approximating the equations, we end up with several computer models. Some are better at resolving winter’s shallow Arctic airmasses, like the NAM model. However, that model has a tendency to be less reliable when it comes to precipitation forecasts.

In the very short-term, the RUC model (the Rapid Update Cycle model) is good for predicting when severe weather development may occur that afternoon, but since it only forecasts out 12 hours, it’s not much use day-in and day-out.

Longer range models like the European (ECMWF) model can often be a very reliable model in parts of the country, but we’ve found that it has a harder time accurately predicting how a storm system develops over the Rockies as compared to the GFS (Global Forecasting System) model. However, the GFS is not as good as the ECMWF when it comes to longer range precipitation forecasting.

Then there’s the question of whether or not we use the raw model data or run a statistical analysis of the output and use that data to create our forecast. Either way, we know that models are good and bad, but can be the one tool that gives us some longer range insight into the development of certain patterns in weather that we could never compute by hand.