By Annie Mitchell, Fall 2020.
No one can read the future. That is, except for meteorologists. Thanks to these scientists of weather, the human race can predict what Mother Nature plans to offer the world with astounding accuracy. Current 10-day weather forecasts are 50 percent accurate, seven-day forecasts 70 percent, and five-day forecasts 90 percent. [1]
Three parts of weather prediction methods and technology make these crystal ball-like predictions possible: an understanding of physics equations, data describing atmospheric conditions, and a dynamical core. These three facets construct the Global Forecast System (GFS), the model most often used in the U.S. since 1980. [2]
The GFS began using its current dynamical core in June 2020. [3] As a computer algorithm, a dynamical core translates equations of movement into language that a computer can read and solve. This software, called the Finite-Volume Cubed-Sphere dynamical core (FV3 for short), was the first substantive implementation of new GFS software in 40 years. [1] According to Lucas Harris, a physical scientist involved in FV3’s development, “the GFS’s old spectral dynamical core was very well designed for its day and so could hang on for many years” — until FV3 this year.
FV3 offers a new method of atmospheric representation— and is aptly named due to the “finite volumes” that it uses to build a simulated atmosphere. “Instead of representing the atmosphere as a mesh of isolated points, or as a global set of waves, it slices the atmosphere into a set of blocks,” Harris said. After slicing the atmosphere, meteorologists use physical data and equations to describe the interaction of each finite volume’s energy, momentum, and mass with those of its neighbors.
Harris works at the birthplace of FV3, the Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, NJ. He focuses on grid refinement techniques in FV3, which uses Gnomonic grids to organize the finite volumes. [3] This type of grid displays all of the globe’s “great circles,” which are circles made by planes that pass through the center of a sphere, as straight lines. Thus, the shortest route between two points on a Gnomonic grid is a straight line.
FV3 applies two-way nesting and continuous stretching to its grids, adding complexity to the dynamical core but improving the forecast.
Two-way nesting involves “nested” grids, which are smaller than their parent or “coarse” grids. The smaller nested grids produce a sharp, narrow perspective of weather, while the coarse grids offer a blurrier, yet broader view of weather. Two-way nesting is the interplay between these two grid types, allowing for accuracy at both local and global scales.
Continuous grid stretching, on the other hand, involves only one grid at a time. This technique smoothly stretches the grid cells in one area and contracts them in another. The contracted area ends up with a higher resolution with cells closer together, allowing meteorologists to focus on a specific region where there might be a hurricane, for example.
When combined, the accuracy benefits of two-way nesting and grid stretching multiply, according to the GFDL webpage for FV3. [3]
FV3 is even more than its grids— as the webpage exclaims, “a fast model is a good model!” The dynamical core’s computational efficiency and scaling made it a prime candidate for the update that the National Oceanic and Atmospheric Association (NOAA) demanded after the devastation of Hurricane Katrina in 2005. [4] FV3 was the second-fastest software in a 2015 NOAA performance study. [5] In a follow-up study with a higher-resolution grid (3 km rather than 13 km resolution) and a more powerful supercomputer, FV3 emerged as the fastest and best, according to Harris.
A unique aspect allows FV3 to be 30 percent faster than other models. This feature is the software’s Lagrangian vertical coordinates. These coordinates orient the finite volumes such that the flow of weather components only occurs in the horizontal direction. “Vertical motion and transport is computed ‘for free,’ so that we only need to compute motion in the horizontal and not also in the vertical,” Harris said. The FV3 expert adds that the horizontal focus of this coordinate system minimizes the “noise” in the vertical direction, clearing the view for weather near the atmosphere’s surface.
FV3 combats noise, or “spurious oscillation,” in the horizontal direction, too. This means that abrupt changes in the atmosphere don’t make it falter. “One of the biggest problems is that many models cannot handle jumps in a variable, such as a cold front or the edge of a cloud or a pollutant plume, without creating a lot of noise or by smoothing out the jump,” Harris said. FV3 can illustrate these fluffy weather objects without this distortion.
Since the dynamical core’s implementation, the U.S. has more accurately predicted hurricane behavior. The Atlantic Oceanographic and Meteorological Laboratory, a partner to Harris’ lab at GFDL, accurately predicted that Hurricane Dorian would miss Florida in 2019. “We have developed the first global model that has been able to skillfully predict changes in hurricane intensity, something that has been a challenge even for specialized hurricane models,” Harris said.
All in all, FV3’s improvements to forecasting software through its grids, computational efficiency, and noise reduction promise to save lives by quickly and accurately informing the U.S. of dangerous future weather.
Our vision of the future is sharper than ever.
[1] Klesman, Allison. “How Weather Forecasts Are Made.” Discover Magazine. August 13, 2019. https://www.discovermagazine.com/planet-earth/how-weather-forecasts-are-made
[2] “NCEP Central Operations.” National Weather Service. https://nomads.ncep.noaa.gov/txt_descriptions/GFS_doc.shtml
[3] “FV3: Finite-Volume Cubed-Sphere Dynamical Core.” Geophysical Fluid Dynamics Laboratory. https://www.gfdl.noaa.gov/fv3/
[4] “New engine is driving NOAA’s flagship weather forecast model.” NOAA Research News. June 12, 2019
[6] Image: “NOAA to develop new global weather model.” National Oceanic and Atmospheric Association. July 27, 2016. https://www.noaa.gov/media-release/noaa-to-develop-new-global-weather-model