Will a New GFS Weather Model Upgrade Close the Gap with The European Model?

Image credit: NOAA

Are you with Team GFS or Team ECMWF, the “European Model”? I hate to pick sides, but as a meteorologist I defer to the weather model that, consistently, is most accurate. Of course I’m rooting for the “American Model”, the GFS or Global Forecast System, to win. But here’s the thing: if you’re sanding a table or building a deck you want to use the best tools at your disposal, right? So it goes with weather forecasting. Meteorologists examine scores of models, looking for consistency, continuity, and trends – ultimately choosing a blend of model solutions that has the highest probability of coming true. Believe it or not, we want to get the forecast right!

ECMWF vs. GFS Accuracy Since 2007. The data is the data. Graphic courtesy of blog.weather.us and meteorologist Ryan Maue.

Like most people I defer to what works, based on personal experience. And in recent years many meteorologists have reached the conclusion that I have over time: ECMWF, The European Model, is consistently more accurate. Not perfect, but stepping back and looking at the big picture…which prediction of future weather is better in most real-world scenarios? ECMWF wins most days.

At the risk of oversimplification, there are many reasons why ECMWF is better. The European model is run by The European Centre for Medium-Range Weather Forecasts in Reading, England. Unlike NOAA, the U.S. National Oceanic and Atmospheric Administration, which runs dozens of models, ECMWF runs one global model at high resolution. All efforts and resources have been focused into perfecting this one weather simulation. There are other factors involved, including demonstrably better data assimilation (getting the most recent observations into the ECMWF model faster) and a steady pipeline of meteorological research (and better physics) being applied to improve the ECMWF model at a quicker pace. The result: ECMWF is arguably the best weather model on the planet.

Hurricane Florence, NASA file image.

A metaphor for life, weather modeling is a work in progress – a journey, not a destination. And recently announced updates to NOAA’s flagship GFS model leave me hopeful that the accuracy gap may finally shrink over time. Resolution will increase from 64 to 127 vertical levels, with improved physics and a global wave model (“WaveWatchIII”). New data assimilation capabilities from satellites and aircraft should improve GFSv16 model performance significantly over time. “This substantial upgrade to the GFS, along with ongoing upgrades to our supercomputing capacity, demonstrates our commitment to advancing weather forecasting to fulfill our mission of protecting life and property,” said Louis W. Uccellini, Ph.D., director, NOAA’s National Weather Service, in a recent press release.

I am hopeful, but like every other meteorologist, I will withhold judgement until I see tangible proof that GFSv16 is more competitive. Can you prove this new iteration of GFS is, in fact, comparable or superior to ECMWF? Early results from NOAA’s Model Evaluation Group (MEG) are encouraging. “They have shown significantly improved frontal boundary and tropical system placements in the medium range. They have shown case after case of v16 pinning down the correct positioning as significantly as multiple days before v15. Extending our lead times tropical system detection is going to be a big help on closing the gap with ECMWF” said AerisWeather meteorologist Justin Deal. AerisWeather is offering the new improved GFS data – GFSv16 is one of numerous regional and global weather models now available for companies to access via API or graphics (AerisWeather AMP).

An Inflection Point for Hurricane Forecasts? ECMWF was first to predict that Hurricane Sandy would not sail out to sea, but hook inland, toward the New Jersey coast in late October, 2012. For many meteorologists that was one (of many) aha-moments in deciding which model to rely on most days. The graphic above is from a corporate weather briefing I issued for Superstorm Sandy. Image credit: WSI and Praedictix.

It is the “Black Swan” events that worry meteorologists most: tropical systems intensifying rapidly before landfall, tornadoes spinning up in areas not under a tornado watch or warning, flash floods that strike with little warning. Improvements in GFSv16 should help to better initialize hurricane models, including HMON and HWRF, resulting in higher-confidence predictions for hurricane track and intensity.

Will the improvements be enough to make NOAA’s GFS model truly competitive with ECMWF? Time will tell, and meteorologists will be comparing the models, side by side. One event, one storm does not tell the tale, but over time we should have an answer to that perpetually vexing question. “I think we’re going to see that the time and money spent by NOAA improving GFS (and the FV3 dynamic core) will once again have paid off and provide wide-reaching benefits to the weather industry” Deal added.

I share Deal’s enthusiasm, but I will go off results, not rhetoric or press releases. Like every other practicing meteorologist, I will be comparing GFSv16 and ECMWF back-to-back, over time, to see which one is the most accurate and reliable, day in and day out.

Team GFS? Yes, I hope so. But at the end of the day, the proof is in the (weather) pudding.

Share this post: