If you have ever forecasted the weather, you probably know about the various computer model simulations which help to guide meteorologists in as they make their predictions. However, choosing the most accurate one to aid in your prediction can be a difficult task.
First, let’s take a look at the long-range models, which include the GFS, ECMWF, GEM, and their Ensembles, as well as the UKMET, NOGAPS and DGEX.
Among these, my personal favorite is the ECMWF, or European model. Run on a supercomputer based in the United Kingdom, this model does a superior job of simulating the atmosphere worldwide up to 10 days out. Even when other models fall short, this one will often nail the forecast. It initializes once every 12 hours, at 0:00 and 12:00 UTC.
So, while in my opinion it is the most accurate, one of the major drawbacks is that most of its data is not readily available for free. In order to access the full model run, one must pay a fair amount of money. For example, AccuWeather’s professional site has the full model available, along with many other features, for about $30/month or $250/year.
As for free access to the European model, my favorite site is called “Raleigh WX” and is maintained by a man by the name of Alan Huffman. There, you can access some of the most basic data for the surface, 850 mb, and 500 mb levels in 24 hour intervals, which can only help a little bit when making predictions.
So, if you are not willing to pay such a large sum to access the full ECMWF, the next best thing is the GFS, or Global Forecast System. Produced by the NOAA in the United States, this model initializes once every 6 hours and will simulate the global atmosphere up to 16 days out.
However, its predictions beyond 10 days from the initialization time are usually wildly inaccurate and should generally be ignored. Within 10 days, it is generally good at getting a grasp on the overall pattern, but many of the specifics are usually off. The more exact details can’t really be trusted until about 4 or 5 days out.
But, unlike the Euro, all of the GFS data is available on the Model Analyses and Forecasts page at the National Center for Environmental Prediction (NCEP) website, which can help an experienced forecaster to make the most of all of the available data.
Another thing to know about the GFS is that some of the runs are more accurate than others. The 0:00 and 12:00 UTC initializations factor in much more observed data than the 6:00 and 18:00 UTC runs, thus allowing the NOAA supercomputer to better simulate what the atmosphere is going to look like in the future. Those so-called “off” runs can’t be trusted until you get within the 72 hour range.
Besides the GFS and Euro, not to be forgotten is the Canadian Meteorological Centre’s GEM model. Officially known as the “Global Environmental Multiscale” model, the GEM may also be referred to as the CMC, or simply the Canadian model. Like the European model, it initializes once every 12 hours and goes out to 6 days and can be accessed at the Raleigh WX page.
While the GFS or European models are generally better, it is good to look at to get a second or third opinion on what is going to happen. Often, the GEM will project a slightly different scenario than the other two, which allows you to see some of the other possibilities in the forecast.
The last of the long-range operational models are the NOGAPS, operated by the United States Navy, UKMET, run by the United Kingdom’s meteorological office, and the DGEX, another NOAA model. In my experience, all three of these usually deliver extremely inaccurate forecasts, and are generally not worth taking a close look at.
What you should take a look at instead are the ensemble predictions of the GFS, European, and GEM models. Ensemble predictions are essentially a reprocessing of the data used for the operational. They can provide a different take on what is going to happen. You can also look at probabilities of certain things happening, a very helpful feature which the operational runs lack.
The GFS Ensembles can be accessed at the NCEP site, while you can European and GEM Ensembles on the Raleigh WX page that I mentioned earlier.
So, make sure that you take a look at the different models and their ensembles when making long-range predictions and you should continue to use them even as you enter the short range. But, once you get inside 4 days, a variety of other models come into play, including the NAM, SREF, RUC, and WRF.
Of these, my favorite is the NAM. The NAM is also based at the NOAA, and is officially called the North American Model. All of its data can be accessed at the NCEP website. It initializes once every 6 hours, like the GFS, and simulates the atmosphere out to 84 hours for North America only.
Generally, its predictions beyond 60 hours should be taken with a grain of salt, but otherwise it does an excellent job. Within the 48 hour range, it is generally my overall favorite, even more so than the European.
However, there are a couple of things to watch for when using the NAM. First, sometimes there will be computer issues which will result in a bad or even useless run. The biggest warning sign of this is if a run displays data that is very different from that of previous runs.
Secondly, the NAM will sometimes overdo precipitation forecasts, so it is usually best to cut them down when you issue your actual forecast if they seem unusually high. Personally, I have seen it overestimate how much liquid water will fall by as much as half an inch, which can be especially problematic in winter forecasting.
But other than those minor two hiccups, the NAM is usually the model of record for short term forecasting. However, as I have continually stressed, it is good to look at multiple sources of data. Another one of my favorites for the short term is the SREF, or Short Range Ensemble Forecast model.
The SREF is another sister model of the NAM and GFS, based at the NOAA, so it can be accessed at the NOAA site. Like the NAM and GFS, it initializes every 6 hours at 3:00, 9:00, 15:00, and 21:00 UTC, instead of the more common times that the others use. Its simulations will go to 87 hours in the future.
Generally, the SREF is the best for short-range ensemble forecasting. Typically, the best way to use it is to clarify the NAM’s output through the probability forecasts, or as an alternative should the NAM data be strange or unreliable.
Moving on, the last of the major short range models is the RUC, or Rapid Update Cycle. This model is only used in the extreme short range as it initializes once an hour and only goes out to 18 hours. It also features very high resolution data, and can be very good for short range storm prediction. Like the other NOAA models, its data can be obtained at the NCEP website.
Finally, there is the WRF, or Weather Research and Forecasting model. This model is part of collaboration between a variety of United States governmental agencies, and can be accessed at its own dedicated website.
The WRF works best for getting the finest details of a storm due to its high resolution, and is generally most accurate within 36 hours. It will often predict mesoscale features that the lower resolution models often miss out on, making it a valuable tool for forecasting.
So, hopefully this will help you determine the difference between a good and bad forecast simulation. Also keep in mind that while forecast models are a great tool to use, they shouldn’t be relied upon. Use your intuition as a forecaster to interpret the data. Good luck!
