If you want to know how a hurricane will behave near the U.S., it's sometimes best to ask the Europeans.
The European Center for Medium-Range Weather Forecast (ECMWF) has a reputation for being, on average, the most accurate weather forecasting model in the world, far outpacing the American Global Forecasting System (GFS). When it comes to forecasting the movements of hurricanes, ECMWF tends to make GFS look like the kid who's late to class. In 2012, for instance, it predicted (correctly) that Hurricane Sandy would end up walloping the East Coast a week before it actually did, while the GFS predicted it would simply stay in the Atlantic and lose power. The GFS did reach the same conclusion, albeit two days later.
The National Weather Service (NWS) has been working to increase its forecasting abilities since Hurricane Sandy, and it announced the development of a new core for its weather modeling, the Finite Volume Cubed-Sphere, or FV3, in 2016. This new model is currently running in parallel with the old GFS model, making Hurricane Florence a testing ground for FV3's abilities.
Weather arms race
The differences between the current GFS and the ECMWF mostly amount to computing power, the physics used by each system and, of course, a willingness to invest money in the system, according to a 2015 Scientific American article.
Running these models requires massive amounts of information, be it from weather balloons to ships to satellites. This information is turned into data points that are plotted onto grids to create a simulation of the atmosphere. The ECMWF excels at this, creating better grids at higher resolutions to achieve better forecasts. Additionally, the ECMWF is able to check its work more by running competing predictions at lower resolutions to figure out where gaps in the forecast might be. Basically, the ECMWF accounts for big changes in the forecast by creating hypothetical situations in which the weather is altered by small changes.
The ECMWF will run 50 of these cycles, according to the Scientific American article, while the GFS only does 20.
"The [Europeans] are better at this because their simulation system is better, and they use more data than we do," Cliff Mass, a professor of atmospheric sciences at the University of Washington, told Scientific American.
Even the NWS acknowledged the gaps between the U.S. and European models as it announced the development of FV3. For instance, the current GFS models have been around for more than 30 years, and they were designed in a time before we had the computers we do now, leaving the model unable to do the type of detailed calculations possible today.
"The current GFS, developed before the age of high speed computers," the NWS writes, "is not able to provide such highly detailed informations [sic]. Even if it ran on a computer with more processing power, it would not work faster."
By switching to FV3 for its new dynamic core, the NWS hopes this more robust version of the GFS will offer higher resolution images (the video above showcases FV3's resolution), the ability to zoom in on local weather events while generating a new global forecast every six hours and changing the representation of weather from waves to connected points on a grid to better accommodate weather that's irregularly shaped. Additionally, FV3 is expected to improve overall forecast accuracy beyond eight to 10 days and provide better hurricane models.
"Using our powerful supercomputers, our new dynamic core which drives the model, and the newest modeling techniques, we are poised to develop and run a more accurate and reliable global model that is used as a basis for all weather forecasts in the U.S.," Louis W. Uccellini, the director of the NWS, said in 2016 statement about FV3.
Florence and the weather-forecasting machine
FV3 is expected to officially become the new model for the GFS sometime in 2019, but that doesn't mean it isn't already at work, trying to account for the weather.
As Hurricane Lane headed for Hawaii in August, FV3 was working on predicting the storm's movements. According to data gathered by The Washington Post, The FV3 version of the GFS ended up producing the most accurate forecasts regarding Hurricane Lane four and five days into the future, making fewer errors than ECMWF in that same range. It more or less matched forecasts for three days out made by both the ECMWF and the National Hurricane Center.
As The Post explained, this represented a promising but limited sampling of FV3's capabilities. Hurricane Florence represents another storm with which to test FV3's forecasting chops.
This infrared satellite image of Hurricane Florence taken on Sept. 11 shows cloud top temperatures in Kelvin. Black indicates the coldest cloud tops and the location of the most powerful storms. (Photo: NASA Goddard Space Flight Center/Wikimedia Commons)
In a Sept. 10 blog post, Mass explains how both FV3 and the current GFS model are matched with regards to Hurricane Florence's arrival along the Carolina coast, but have very different forecasts regarding the storm's central pressure levels. The current model is predicting 913 hectopascals (hPa) of pressure on Sept. 14 while VF3 is calling for 979 hPa. The higher the pressure, the stronger the storm.
The two models diverge even more significantly by Sept. 16. The current model maintains the same amount of pressure, but it predicts that Hurricane Florence will remain along the coast, battering the states from there. FV3, meanwhile, expects the storm to move inland and weaken.
Which one will be right? "You will know in a week," Mass writes.
As for why figuring out better weather models matters so much to all of us, Mass explained it best to CityLab this week: "Better weather prediction saves lives and property, and has immense economic implications. In a period when we are worrying about more extreme weather impacts, the first line of defense is better weather prediction."