Bells rang and geeks chortled across the world on Wednesday morning as NOAA pulled the plug on the old American weather model, known as the GFS (Global Forecast Systems), and replaced it with an updated model we've creatively called the "GFS Upgrade." Behold! Slightly better forecasts.
The upgraded GFS became operational on Wednesday morning at 12z, or 7:00 AM EST, meaning that forecasters at the National Weather Service officially began using the upgrade in their forecasting process. The upgraded model features a multitude of algorithmic improvements, an extended short-range forecast, and the most notable change, a higher resolution. You can read a full list of changes at NCEP's website.
The most noticeable change for regular users is the bump in resolution. The old GFS model ran at a horizontal resolution of 27 kilometers (16 miles), meaning that there were 27 kilometers of horizontal distance between each of the model's forecast grid points. For a global model, a 27-kilometer resolution worked well enough for a while. The model's new resolution is more than doubled at just 13 kilometers (8 miles). That doesn't sound like much, but it's a huge improvement. It allows the model to generate forecast maps that more accurately display how variables like wind, temperatures, and precipitation interact with smaller features like mountain ranges and bodies of water.
Here's an example of old versus new using the coastal storm that made a mess of travel the day before Thanksgiving last year.
First, the old model:
And the new model, from the same run and valid for the same time:
The first thing you'll notice is that the forecasts are slightly different—a result of the improved algorithms—and the resolution is improved. The 32°F and 35°F lines aren't smoothed over in the new model like they are in the old, isobars are less smoothed out, and the areas of precipitation are all more refined, all thanks in large part to the improved resolution.
While many weather geeks keep saying that the GFS model now has a similar resolution as the NAM, it's like comparing apples to oranges. The NAM, or the North American Model, is a mesoscale model with a horizontal resolution of 12 kilometers (about 7 miles) that's designed to handle smaller systems across the United States and parts of Canada and Mexico. The GFS is designed to handle large-scale systems like nor'easters, hurricanes, the polar vortex, or the jet stream across the entire world. The 12-km NAM and its much higher resolution counterpart, the 4-km NAM, on the other hand, are great tools to use when you want to forecast convection (thunderstorms).
The upgrade is a great tool that's already had some real-world impact in its improved accuracy. Ian Livingston, a forecaster for the Washington Post's Capital Weather Gang, noted on Twitter a few days ago that the upgraded model "ended up not too bad" when it came to the fast-moving clipper systems that dumped snow on the Washington D.C. area. Not exactly a glowing endorsement (models rarely deserve them), but it's a good sign for the GFS upgrade's future that it picked up on two nebulous light snowfall events in a tricky part of the country.
While the GFS model is still inferior to the ECMWF (European) model in many respects, NOAA is actively working on more computing power to help improve our modelling capabilities. Using money appropriated to the agency in the aftermath of Hurricane Sandy, NOAA plans to upgrade its supercomputing capacity "nearly tenfold" by this October. Faster supercomputers allows for more calculations and better results from our weather models.
The future is now, and it's a little more accurate.