GribStream

GribStream Blog

NOAA proposes GFS v17 for October 2026: 9 km coupled upgrade and product removals

|
gfs gdas

NWS is seeking comments on a proposed GFS/GDAS v17 upgrade targeted for October 2026, with a 9 km coupled forecast system, major GDAS changes, and several legacy products slated for removal.

On April 15, 2026, NOAA/NWS published PNS 26-29, proposing an upgrade of the operational Global Forecast System (GFS) and Global Data Assimilation System (GDAS) from v16 to v17 in October 2026. One day later, on April 16, 2026, EMC published PNS 26-30, a separate notice covering the legacy products it plans to remove as part of the same cutover.

The important framing is that this is still a proposal, not a final implementation notice. NOAA is taking comments on the science changes through May 15, 2026, and on the product removals through May 16, 2026. If NOAA proceeds, it says a formal Service Change Notice will be issued 30 days before implementation.

If you are directly affected, NOAA lists two feedback paths: gfs.feedback@noaa.gov for the science proposal and emc.products.feedback@noaa.gov for the removal list.

The big shift in GFSv17

The headline change is that GFSv17 becomes a fully coupled Earth-system forecast instead of a mostly atmosphere-centric global model. NOAA says the operational system will couple:

  • the atmosphere
  • the land surface
  • the ocean
  • sea ice
  • waves

NOAA also says the deterministic atmospheric resolution will increase from C768 (~13 km) to C1152 (~9 km) on the FV3 dynamical core, with fractional grids along oceanic coastlines.

That is a meaningful structural change, not a minor tuning cycle. For users who compare long histories of GFS, this is the kind of version jump that can move thresholds, distributions, and error characteristics in visible ways.

Science changes NOAA has already spelled out

The April 15 notice is unusually specific about the science package.

For the atmosphere and land, NOAA lists:

  • convection updates aimed at reducing biases, improving CAPE forecasts, improving MJO prediction, and eliminating negative tracer mixing ratios
  • new handling of environmental wind shear plus revised PBL-convection interaction to improve hurricane intensity forecasts and better represent surface inversions
  • a new sea spray parameterization plus roughness and stability updates to improve surface fluxes
  • replacement of the one-moment GFDL cloud microphysics with Thompson-Eidhammer two-moment microphysics, plus semi-Lagrangian hydrometeor sedimentation
  • radiation updates including MERRA-2 aerosol climatology, updated cloud-radiation interaction, and fixes for excessive shortwave radiation into the ocean at low sun angles
  • replacement of Noah LSM with Noah-MP
  • updated gravity-wave drag and mountain-blocking parameterizations, including turbulent orographic form drag
  • fractional-grid compositing of albedo and emissivity to improve atmosphere-land coupling near coastlines

For the new coupled components, NOAA lists:

  • MOM6 on a 0.25 degree tripolar grid for the ocean
  • CICE6 on a 0.25 degree tripolar grid for sea ice
  • continued use of WAVEWATCH III, but with an unstructured grid and two-way coupling inputs/feedbacks across the system
  • continued use of the Unified Forecast System (UFS) framework
  • use of CMEPS and the NUOPC/ESMF coupling stack
  • continued advancement of the portable Global Workflow used to run the coupled UFS application

GDAS changes are just as important

The same proposal includes a substantial GDASv17 refresh, because the coupled forecast system needs new initial conditions for non-atmospheric components too.

NOAA says GDASv17 will introduce:

  • a multi-scale upgrade to the GSI Hybrid 4DEnVar atmospheric data assimilation system with scale-dependent localization
  • replacement of GLDAS with a soil temperature and soil moisture analysis embedded in the existing atmospheric LETKF
  • new snow data assimilation in the JEDI framework using a 2DVar algorithm and assimilating station snow depth plus IMS snow cover
  • new ocean and sea-ice data assimilation in the JEDI framework using 3DVar-FGAT

NOAA is also explicit about the marine observations it plans to assimilate, including:

  • sea-surface height from Jason-3, Sentinel-6 Michael Freilich, SARAL/AltiKa, and CryoSat-2
  • sea-surface temperature from VIIRS and AVHRR
  • sea-ice concentration from AMSR2
  • in-situ temperature and salinity observations from Argo floats and surface drifters

For GDAS users, this matters because the analysis baseline itself is changing, not just the downstream forecast model.

Products NOAA plans to remove

The April 16 removal notice is the part many downstream users will care about most, because it names concrete files and product families that are slated to disappear when GFSv17 goes operational.

NOAA says it plans to remove:

  • synthetic nadir ABI GOES-R GRIB2 files:
    • gfs.t{CC}z.special.grib2f{FFF}
    • gfs.t{CC}z.goessimpgrb2.0p25.f{FFF}
    • gfs.t{CC}z.goessimpgrb2f{FFF}.grd221
  • the time-averaged categorical precipitation-type flags in gfs.t{CC}z.pgrb2.LpLL.f{FFF} for GRIB2 IDs CRAIN, CSNOW, CFRZR, and CICEP
  • all post-processed GRIB1 products interpolated to CONUS Grid 211
  • several wave buoy point outputs for buoys 44040, 44043, 44061, 51207, and 51210
  • NetCDF-based model history and analysis files:
    • gfs.t{CC}z.atmf{FFF}.nc
    • gfs.t{CC}z.sfcf{FFF}.nc
    • gdas.t{CC}z.atmf{FFF}.nc
    • gdas.t{CC}z.sfcf{FFF}.nc
    • gfs.t{CC}z.atmanl.nc
    • gfs.t{CC}z.sfcanl.nc
    • gdas.t{CC}z.atmanl.nc
    • gdas.t{CC}z.sfcanl.nc
  • lower-resolution 0.50 degree and 1.00 degree GRIB2 products, including the pgrb2, pgrb2b, and pgrb2full file families and their .idx index files

NOAA gives concrete reasons for those removals:

  • the synthetic GOES-R files are being cut to reduce runtime and NOMADS bandwidth
  • time-averaged categorical precipitation-type flags are being removed because NOAA says they are not physically meaningful
  • GRIB1 is deprecated, so v17 will be GRIB2-only
  • the listed wave buoy points are outside the computational domain
  • the NetCDF, 0.50 degree, and 1.00 degree outputs are being removed to make room for new ocean and ice products and for an expanded set of atmospheric fields

NOAA also says the NOMADS GRIB filters for gfs_0p50 and gfs_1p00 will stop serving those lower-resolution files once they are removed.

What NOAA has not published yet

Just as important as the published details are the missing ones.

As of April 20, 2026, NOAA has not yet published:

  • the exact implementation day within October 2026
  • the final Service Change Notice
  • the promised folder directory structure and name changes
  • a full public list of the new file names and paths that will replace today's layout

That missing directory-layout notice matters operationally. The April 15 notice says significant folder and naming changes are planned, but the actual path map is still pending.

What this means for GribStream users

For GribStream, the main thing to watch is not whether GFS or GDAS continue to exist. They will. The bigger question is how NOAA reshapes the public delivery layout and which sidecar products disappear as the coupled system goes live.

If you rely on the core 0.25 degree GRIB2 forecast stream, this looks manageable. If you rely on NOAA's NetCDF, 0.50 degree, 1.00 degree, GRIB1, or specialty sidecar files outside GribStream, you should assume those integrations need attention before the cutover.

For validation and backtesting, the cleanest approach is to mark October 2026 as a likely version boundary now, then refine once NOAA publishes the final implementation date. A coupled 9 km GFS plus a materially different GDAS initialization stack is large enough that pre/post-v17 comparisons should be treated as separate model eras.

Sources

Related models

gfs gdas
Try GribStream