This page tracks major lifecycle events for models documented in this repository — upcoming implementations, scheduled retirements, experimental systems, and version upgrades.
The goal is to give users a single place to check "what is changing" without having to browse individual model entries. Each item links to the relevant entry where detailed information lives.
For UFS-related transitions specifically, see UFS.md, which provides the full programme context. UFS items are listed here briefly; UFS.md has the detailed narrative.
Last updated: July 2026.
Major upgrade to the National Blend of Models with longer hourly guidance (36 h → 48 h), new products, removed Haines Index, and new inputs including ECAIFS and AIGFS.
- Entry: NBM
- Authority: NWS SCN 26-24 (AAC revision, April 28, 2026)
- Verification note: Originally targeted April 15, 2026; rescheduled to April 23, then April 30. The April 30 attempt fell within a Critical Weather Day / Enhanced Caution Event window, triggering the SCN's contingency provision and pushing actual cutover to May 5, 2026.
The next National Blend of Models upgrade after v5.0. Planned scope: 12-hour winter products, fixes for "blocky" percentile and precipitation-type (ptype) fields, remaining post-v5.0 winter-weather guidance work, and a new Southwest Pacific domain (a coverage expansion rather than a methodology change). No NWS Service Change Notice or PNS has been issued and it does not appear on the NBM versions page — this is a planning signal only, and no implementation date is recorded here pending a formal notice.
- Entry: NBM
- Source: NBM user webinar (April 15, 2026) and WPC/HMT "NBMv5 Winter Overview" (February 10, 2026) — presentation decks, not authoritative notices
ECMWF's physics-based and AI forecast lines upgraded together on the same day, as planned. Cycle 50r1 brings fully coupled atmosphere–ocean–sea-ice data assimilation, the new NEMO4-SI3 ocean/sea-ice core, revised convection and cloud-microphysics (addressing stationary heavy rainfall), and over 40 new ocean and sea-ice variables — all with no change in horizontal or vertical resolution. AIFS Single v2 adds a 10 hPa pressure level, a new data-driven wave stream, and a snow cover parameter. AIFS ENS v2 introduces a new probabilistic wave ensemble stream and a multi-scale loss function. The separate HRES stream is discontinued: the ex-HRES data stream becomes the ENS Control (MARS users migrate stream=enfo, type=cf → stream=oper, type=fc and stream=scda, type=fc for 06/18 UTC). ECMWF's experimental external ML model suite (Aurora, FourCastNet, GraphCast, Pangu-Weather) has been discontinued in operations as of the same day.
- Entries: IFS Open Data · ECMWF EPS · AIFS Single · AIFS ENS
- Authority: ECMWF news release, IFS Cycle 50r1 and AIFS v2 go live (May 12, 2026); Implementation of IFS Cycle 50r1; ECMWF Newsletter 185, Upgrade of IFS Cycle 50r1
- Downstream note: Individual model entries (IFS, IFS ENS, AIFS Single, AIFS ENS) still describe these changes under "Upcoming changes" — they should be migrated to "Version history" / "Recent version history" sections in a follow-up pass.
ECCC's operational global deterministic system becomes a hybrid physics–AI system: GEM's large-scale temperature and horizontal wind fields (250–850 hPa) are spectrally nudged toward the GEML v1.1 AI weather model. This is the operationalization of the configuration previously distributed as the experimental GDPS, which is now retired as a separate experimental product. Replaces v9.1.0.
- Entry: GDPS
- Authority: ECCC technical note CMC-GDPS-EXP-10.0.0-2026; GDPS 10.0.0 fact sheet and technical specifications (May 2026)
NOAA's next-generation convection-allowing system for North America is now formally scheduled. RRFS and REFS implement together, with legacy NAM, HREF, SREF, and HiresW (all domains except Guam) retiring on the same day. A pre-implementation real-time feed is expected on NOMADS on or about August 11, 2026. See UFS.md for the full UFS context including the wave of retirements that occurs on the same day.
- Entries: RRFS · REFS
- Authority: NWS SCN 26-48 (RRFS/REFS implementation) + companion SCN 26-47 (terminations), both updated July 6, 2026 (originally May 12, 2026)
- Verification note: Originally targeted early 2026, then August 31, 2026; slipped again to October 6, 2026 in the July 6, 2026 update (AAB), which also decoupled the real-time parallel feed to on or about August 11, 2026. The October 6 date is subject to the standard CWD/ECE contingency — if the implementation date is declared a Critical Weather Day, an Enhanced Caution Event, or other significant weather is occurring or anticipated, implementation moves to 12 UTC on the next eligible weekday. NBM v5.0 was deferred under exactly this provision earlier in 2026, so the contingency is not theoretical.
NOAA's three AI-based global weather systems became operational on the same date, replacing EAGLE SOLO (deterministic) and EAGLE Ensemble. AIGFS and AIGEFS are standalone AI systems based on the GraphCast architecture; HGEFS pools the 31 GEFS physics members with the 31 AIGEFS AI members into a 62-member grand ensemble, though it distributes only the pooled mean and spread rather than the members themselves. None of the three replaces an operational physics-based system — they run alongside GFS and GEFS. All three are distributed on NOMADS only, with no FTP and no NODD/AWS presence.
- Entries: AIGFS · AIGEFS · HGEFS
- Authority: NWS SCN 25-89 (implementation of AIGFS, AIGEFS and HGEFS), December 2025
ECMWF's AI deterministic and ensemble forecast systems both upgraded to v2, jointly with IFS Cycle 50r1 (see above). v2 was fine-tuned specifically on Cycle 50r1 esuite analyses, adds a 10 hPa pressure level, and introduces ECMWF's first data-driven wave forecasts (deterministic and ensemble) alongside a new snow cover parameter. AIFS ENS v2 replaces v1's afCRPS loss with a multi-scale loss function.
- Entries: AIFS Single · AIFS ENS
- Authority: ECMWF news release, IFS Cycle 50r1 and AIFS v2 go live (May 12, 2026)
NOAA upgraded the deterministic AIGFS to v1.1 at the 12 UTC cycle. Three training changes: the grid-point MSE loss function was replaced with a spherical harmonic loss using wind speed and direction rather than u/v components; the model was fine-tuned on four years of GDAS analysis; and fine-tuning was extended to train autoregressively to a 72-hour lead time. Reported improvements to tropical cyclone intensity, forecast precipitation, and the blurring of fields at long lead times. This is the AIGFSdev2.1 spectral-loss line of work reaching operations.
- Entry: AIGFS
- Authority: NWS SCN 26-68 (AIGFS v1.1 implementation), July 27, 2026
- Verification note: Product content was unchanged — grid, parameter set, cycle schedule, and forecast length all match v1.0, so this is a model-quality upgrade with no downstream ingest impact. AIGEFS and HGEFS were not included; the NOMADS paths still resolve to
aigefs/v1.0/andhgefs/v1.0/as of 2026-08-06, so the deterministic and ensemble AI systems are no longer at matched model versions.
These systems are scheduled for retirement on October 6, 2026 at 12 UTC, the same cycle that brings RRFSv1 and REFS into operations. The retirement set was originally proposed in NWS Public Information Statement 25-41 (June 26, 2025); the retirements are scheduled by NWS Service Change Notice 26-47 (termination) and RRFS/REFS implementation by companion SCN 26-48, both updated July 6, 2026 (second slip, from August 31 to October 6). Subject to the standard CWD/ECE contingency. See UFS.md for the consolidation context.
- NAM — full retirement (12 km parent and all 3 km nests)
- NAM Nest — all convection-allowing nests
- HREF — replaced by REFS (extends 48 h → 60 h)
- HiresW — CONUS, Alaska, Hawaii, and Puerto Rico domains; see Status and retirement for the surviving Guam exception
- SREF (not in repo) — replaced by REFS; the SCN-confirmed retirement supersedes the earlier expectation that SREF would persist into the second wave under RRFSv2
- NARRE (not in repo) — replaced by REFS
- NAM MOS (not in repo) — retired alongside NAM
These systems are publicly signaled but have no formal retirement notification as of May 2026. They are expected to be retired with RRFSv2 (MPAS-based). See UFS.md for the second-wave retirement context.
Operational standalone AI global deterministic weather model — the AI component of GDPS v10.0.0 (operational May 26, 2026). Graph neural network derived from Google DeepMind's GraphCast architecture, retrained by ECCC on ERA5 reanalysis (1979–2016) and ECMWF HRES analyses (2016–2021). 0.25° (~28 km) resolution, 13 pressure levels, 10-day forecast 2× daily, initialized from GDPS analysis. Serves a dual role as both a standalone forecast product (distributed at https://dd.weather.gc.ca/today/model_gdps-geml/25km/) and the spectral nudging target for the operational GDPS (since v10.0.0). Trained model weights are publicly released via HuggingFace at https://huggingface.co/ECCC-ASTD-MRD/geml — unusual openness compared to most operational AI weather systems.
Experimental coupled atmosphere-ocean-sea ice prediction system at ~3 km resolution covering a large pan-Arctic domain (northern Canada, Alaska, Greenland, Iceland, Scandinavia, Baltic countries, eastern Russia). GEM atmospheric component (configuration inherited from the decommissioned HRDPS-North) two-way coupled to NEMO v3.6 ocean and CICE v6.2.0 sea ice via the GOSSIP coupler at 5-minute exchange frequency. Provides forecast guidance to Canadian Storm Prediction Centres, DND, CCG, and DFO. Distributed via the standard MSC datamart but explicitly designated as experimental. Current version 3.0.0 implemented June 18, 2025.
Experimental kilometric-scale (~1 km) deterministic NWP system covering Southern British Columbia. GEM v5.2.0 atmospheric model piloted by HRDPS national v7.0.0, with surface initial conditions from CaLDAS. Distributed via the MSC alpha datamart (separate from the standard operational MSC datamart). Intended to better resolve mesoscale features in BC's complex topography than the operational 2.5 km HRDPS. Current version 1.5.0 released November 6, 2024.
Experimental productionization of Google DeepMind's GraphCast architecture by NOAA, fine-tuned on GDAS+ERA5 data. Predecessor of the operational AIGFS, but continues to run experimentally alongside its operational descendant.
Experimental productionization of NVIDIA's FourCastNet architecture using Spherical Fourier Neural Operators. No operational descendant announced as of April 2026.
Complete migration of ECMWF IFS to GRIB2-only parameter representation. Affects Open Data users directly. Legacy GRIB1-style parameter references move to GRIB2-native identifiers; CCSDS compression required.
- Entries: IFS Open Data · ECMWF EPS
- Authority: ECMWF Migration to GRIB edition 2 page
Extended forecast length from 198 h to 246 h; added parameters and levels; renamed height_asl_on_pressure_levels to geopotential_height_on_pressure_levels. Already effective.
- Entry: MOGREPS-G
Imminent or recent changes are events that have happened in the last few months or will happen in the next few months. These need the most active attention from catalog users.
Recently operational (AI weather systems) highlights AI-based forecast systems that have transitioned from experimental to operational status. Listed separately because the AI lifecycle is moving fast and is worth tracking distinctly.
Scheduled retirements are events with authoritative NWS/ECMWF/ECCC notifications confirming the change. These are locked in pending operational implementation.
Expected future retirements are events that are publicly signaled by operators (through roadmap documents, press releases, or downstream product announcements) but have not yet received formal retirement notifications. These are directionally certain but lack specific dates.
Experimental and pre-operational systems exist today but are not the operational primary system for their role. They may become operational, be superseded, or remain experimental indefinitely.
Format and distribution changes are events that don't change the underlying model but do change how users consume it.
If you notice a missing status change — a new SCN, a retirement notification, an operational implementation date — please add it to the appropriate section above with a link to the authoritative source. For UFS-related changes, please also update UFS.md so the two files stay synchronized.