NOAA's CFS replacement: the science behind SFS Beta

NOAA is developing the Seasonal Forecast System (SFS) to replace CFSv2, its operational seasonal forecasting system. SFS Beta is the experimental step toward that successor. The September 2026 v2.0 update addresses how the model represents sea ice, ocean–ice interactions and uncertainty in the tropical Pacific.

For users of seasonal forecasts, the important questions are scientific: where does predictability months ahead come from, how is the new system using it, and how should its predictions be interpreted? NOAA's SFS project overview sets out the replacement programme.

What does a seasonal forecast tell us?

A seasonal forecast helps assess the likelihood of conditions such as a warmer or wetter season. It does not make the weather on a particular day many months from now reliably predictable. Energy demand, agricultural planning and water management can benefit from those shifts in risk, while still requiring shorter-range forecasts as decisions approach.

The scientific opportunity comes partly from the slower evolution of the ocean, soil moisture, snow and sea ice. These components retain information about earlier conditions and exchange heat and moisture with the atmosphere. NOAA's SFS development team identifies that long-term memory and its initialization as central to seasonal prediction.

A coupled Earth system, built for longer forecasts

CFSv2 already couples the atmosphere, ocean, land and sea ice. SFS modernizes those components within the Unified Forecast System rather than introducing coupling for the first time. Its main components are:

• FV3 and CCPP for the atmosphere, representing atmospheric motion and physical processes such as clouds and convection.
• MOM6 for the ocean, allowing ocean conditions to evolve and interact with the atmosphere.
• CICE6 for sea ice, representing changes in the ice cover and its interaction with the ocean and air.
• Noah-MP for land, representing the land surface, soil and snow processes that influence exchanges with the atmosphere.

NOAA describes these components in its SFS Beta launch article. The objective is a more consistent account of the processes that influence seasonal conditions, not simply a finer map.

Physics tuned to climate variability

SFS atmospheric physics draws largely on GFSv17, with changes for longer time scales. NOAA describes a hydrostatic FV3 configuration to reduce computational cost, revised gravity-wave drag to improve stratospheric variability, a turbulence scheme, evolving aerosol climatology and changes to deep convection intended to improve Madden–Julian Oscillation prediction. NOAA's SFS Beta science overview.

The Madden–Julian Oscillation is a moving pattern of tropical rainfall and circulation whose influence extends beyond the tropics. Representing such large-scale variability is part of the challenge of connecting weather processes to longer-range predictability. NOAA's development plan also prioritizes tropical air–sea interactions, land–atmosphere coupling and snow-related feedbacks.

Why the September v2.0 update matters

The latest update tunes sea-ice physics to reduce excessive summer melt and adds a snow climatology to initial ice conditions that previously contained too little snow. It also adjusts ocean–ice coupling, atmospheric gravity-wave drag and stochastic physics to reduce excessive ensemble spread in the ENSO region. These are targeted corrections to interacting physical processes, not evidence that every forecast is now more accurate. NOAA's September update.

NOAA warns that v2.0 can change model climatology and forecast characteristics relative to v1.2. Consequently, comparisons and anomaly calculations need a historical reference consistent with the forecast version.

Ensembles and reforecasts: uncertainty needs a reference

The Beta design has 31 forecast members, initialized monthly for 12 months, and 11-member historical reforecasts beginning in 1991. An ensemble represents a range of possible outcomes; the members are not 31 independent guarantees of what will happen. Reforecasts rerun the model for past dates, providing a basis for assessing skill and correcting systematic biases. SFS Beta specifications.

NOAA reports improved performance over CFSv2 for selected temperature, precipitation and sea-surface-temperature evaluations using March–May initializations from 1991–2020. Those are encouraging results for the earlier Beta system, not universal proof of v2.0 skill across all regions and seasons. See the NOAA evaluation summary.

Experimental today; GribStream support is planned

CPC has used SFS Beta as experimental guidance since March 2026. NOAA describes operational SFSv1 implementation as a target around the end of 2027; that is not a confirmed CFSv2 retirement date. The configuration will continue to evolve. NOAA's programme update.

GribStream intends to support SFS once a suitable public feed has verified timing and coverage. Preparatory work is underway, but SFS is not yet available through our API.

As of September 25, 2026, NOAA publicly distributes monthly atmospheric and oceanic products. The model's 12-month forecast horizon does not mean that daily or six-hourly values are publicly available over that whole period. NOAA's public-data overview.

Our evaluation has not yet established reliable valid times and coverage for the daily atmospheric feed, or identified a public six-hourly feed across the seasonal horizon. We are therefore holding integration until these details are clarified or a suitable feed becomes available. We will update this article when support launches, with the supported products, coverage and examples.

Meanwhile, existing CFS pressure-level, surface and flux, ocean and isentropic products remain available. Experimental SFS support will complement those datasets as we prepare for NOAA's eventual operational transition.