Introduction to FV3 GFS
NOAA’s Global Forecast System, GFS, is perhaps the most popular global weather model in use worldwide. GFS is freely available and has undergone years of continual refinement.
Many people think of the GFS weather model as a static thing - something they may have first had experience with 5 or 10 years ago while making the assumption that todays version of this model is the same as it was back then. This is a mistake.
NOAA’s National Weather Service Environmental Modeling Center is responsible for the GFS model. As of the time this article was written (July 2018) the most recent update to GFS was July 19, 2017, when version v14 of GFS was put into production. There has been a continual series of improvements to the GFS, roughly once per year.
The next generation of GFS, FV3 GFS, is currently undergoing evaluation and the plan is to put it into production in Q2FY19 (January to March, 2019) as GFS v15.0. This next version of GFS represents a major change from the existing version.
The public information statement
NOAA publishes details on every upgrade for every model it supports. These statements can be interesting - mainly I do not understand the details (changes to the microphysics for example, what’s that?) but it is interesting to see the progress being made.
The following is from the NOAA public information statement on the GFS upgrade:
NOAA/NWS selected the Geophysical Fluid Dynamics Laboratory (GFDL) finite-volume cubed-sphere (FV3) dynamical core as the NWS Next Generation Global Prediction System (NGGPS). The Environmental Modeling Center (EMC) is seeking comments on the proposed changes to the Global Forecast System (GFS) running with FV3 through July 15, 2018.
The current operational GFS, which has a spectral dynamical core, will be replaced by the proposed GFS with FV3 dynamical core and improved physics parameterizations in Quarter 2 of Fiscal Year 2019 (Q2FY19). We are seeking feedback on the performance of the GFS with FV3 and the proposed product changes.
The proposed GFS version maintains a horizontal resolution of 13km and has 64 levels in the vertical extending up to 0.2 hPa. It uses the same physics package as the current operational GFS except for:
- Replacing Zhao-Carr microphysics with the more advanced GFDL microphysics
- Updating parameterization of ozone photochemistry with additional production and loss terms
- Introducing parameterization of middle atmospheric water vapor photochemistry
- Revising bare soil evaporation scheme
The data assimilation system will be updated to include:
- Infrared Atmospheric Sounding Interferometer (IASI) moisture channels
- Advanced Technology Microwave Sounder (ATMS) all-sky radiances
- Fix for an issue with the Near Sea Surface Temperature (NSST) in the Florida Strait
- Upgrade to the use of Cross-track Infrared Sounder (CrIS) radiances
- NOAA-20 CrIS and ATMS data
- Megha-Tropiques SAPHIR data
- Advanced Scatterometer (ASCAT) data from MetOp-B
A few details
(Note that this post is written by a layperson and is greatly simplified, and perhaps incorrect. If you are an expert in this area, sorry for all of the errors you may identify. I’ve tried to greatly simplify this presentation to aid general understanding.)
Without going into great detail, the next generation of GFS is based on a new core piece of software called FV3. Numerical weather models are large suites of software composed of many components. I will briefly focus on two components here: the physics module and the transport module. The physics module is responsible for incrementing a time step in the weather simulation based on the current conditions and the laws of physics (conservation of energy, momentum and so on.) The transport module is responsible for migrating weather conditions between cells, as appropriate, so that, for example, wind moves from cell to cell in the simulation.
The FV3 upgrade of GFS is a complete replacement of the transport module. There appear to be many advantages in doing this.
1) FV3 is a nonhydrostatic model. What this means is that the transport mechanism is able to more accurately model the vertical movement of weather. This allows, for example, for much better representation of convection (rain, storms, etc.) The existing GFS (and ECMWF) are hydrostatic models, where the weather simulation models horizontal movement within layers of the atmosphere. In a hydrostatic model, vertical movement of weather is modeled much differently than horizontal movement. It appears that there is a consensus that nonhydrostatic models are better in some important ways, and that this benefit increases as the resolution of the models improve. Do an internet search for “nonhydrostatic weather model” for more information
2) FV3 has the ability to run at different resolutions, including supporting nests of different resolutions for different regions. For example, some of the supporting documents in the Model Evaluation Group site (linked below) describe running the FV3 GFS globally at a 1km resolution! Hopefully this capability will result in higher resolution versions of GFS being made available to the community, over time
3) FV3 is able to generate additional weather parameters, such as Composite Reflectivity, which is very useful when studying convection
I believe the original goal in producing this next generation of GFS was to both improve the transport module (with FV3) as well as to improve the physics module. There appears to have been a decision to split this work into two distinct upgrades in order to be more timely in making the improvements available. For more information on this, see the last slide in this presentation.
This first upgrade is the upgrade of the core to FV3 and the current plan appears to be to move to an advanced physics module in the following upgrade.
NOAA’s goal in these improvements is to make GFS the best numerical weather model available worldwide. Whether or not they are able to achieve this goal remains to be seen, but it is a goal we all should support.
There is some additional information which may be of interest.