Description of the upgrade

The operational release of AIFS Single v2will introduce scientific changes to the AIFS deterministic model.

Changes include:

• The introduction of a new wave component, including 10 wave variables, marking ECMWF's first operational data-driven wave forecasts.
• The addition of 2 new snow variables to the existing land component.
• Improved vertical velocities, by changing parameter W from a prognostic to a diagnostic field. 
• An improved representation of the stratosphere, with the addition of pressure level fields at 10hPa.
• An improved training regime, using 2 years of additional training data compared to AIFS Single v1.1.

Description of the model

The upgrade of AIFS Single to version 2 introduces changes to the training regime. 

The full training procedure for AIFS Single v2 consists of:

• Pre-training, which remains the same as for AIFS Single v1.1.
  • Pre-training is performed on ERA5 data covering the years 1979–2022 over 260,000 training steps.
• Fine-tuning, which has been updated for AIFS Single v2. 
  • Fine-tuning is performed on 2 years of additional data compared to AIFS Single v1.1. In particular, fine-tuning is performed on ECMWF operational analysis data and available IFS 50r1 esuite analysis data, covering the years 2018-2024 over 7,900 training steps.
    
  • As in previous versions of AIFS Single, IFS fields are interpolated from their native O1280 resolution (approximately 0.1°) using MARS default interpolation tools down to N320 (approximately 0.25°) for fine-tuning and initialisation of the model during inference. 

For both phases of training, a cosine learning rate schedule and a batch size of 16 is used. 

This upgrade does not introduce any changes to the model architecture; the model remains the same as the existing operational model (AIFS Single v1.1). Technical details about the model architecture are detailed in the arXiv preprints here and here.

AIFS Single v2 is implemented using the Anemoi framework. Anemoi consists of open source components, enabling the user community to run AIFS themselves.

Timeline of the implementation 

Datasets affected

This upgrade impacts the deterministic AIFS forecast dataset (Set IX - AIFS).

Resolution

AIFS Single v2 introduces a new pressure level to the stratospheric component.

Meteorological content 

Initial conditions

• AIFS Single v2 uses the operational IFS control initial condition, regridded to an N320 grid.

Observations

• No observations are used to train AIFS Single v2.

Meteorological impact

Medium range

Forecast Accuracy and Activity

We observe a general improvement in forecast accuracy due to including more recent training data in the fine-tuning of AIFS v2.

Additionally, AIFS v2 forecasts are less smooth than AIFS v1.1 forecasts. This increased forecast activity leads to small "skill" degradations at longer lead times for some variables. 

Upper-air parameters

Stratosphere

The skill for all variables at 50hPa and 100hPa in the northern and southern hemisphere has significantly improved. The presence of 10hPa also means sudden stratospheric warmings are present in the AIFS v2 forecast. The skill at 50hPa and 100hPa in the tropics remains largely the same but will be targetted in upcoming cycles.

Vertical Velocities

The vertical velocities in AIFS v2 are much more physically consistent than in AIFS v1.1. The Hadley Circulation convection cells are now properly formed and the strength of these cells agrees with physical reasoning. It is likely other vertical flows have also improved.

Weather parameters 

Data-driven waves

ECMWF will issue data-driven wave forecasts for the first time, leading to a substantial improvement in medium-range skill of wave forecasts. Significant wave height forecast performance improves by around 10% compared to the current operational IFS wave forecasts and also relative to the IFS Cycle 50r1 wave model version. 

However, some degradation and a negative bias is observed in the Antarctic during the summer in comparison to the IFS Cycle 50r1 wave model version, linked to AIFS v2 forecasts smoothing along the sea-ice edge. This will be addressed in future developments.

Snow cover

ECMWF will issue data-driven snow depth and snow cover forecasts for the first time. The AIFS forecasts show improved performance compared to 50r1 forecasts, with the predicted snow cover fraction closer to observations. This indicates that AIFS v2 has a more accurate representation of snow distribution across the model domain.

2m temperature

For 2m temperature, degradations are observed in the Arctic, primarily linked to changes in sea-ice–atmosphere coupling introduced in 50r1, which impacts the initial conditions given to AIFS v2 and the limited amount of 50r1 data means the model has not yet fully learnt this new coupling. In particular, AIFS v2 has seen very limited examples of the new wintertime sea-ice coupling in the Arctic, meaning these impacts are most pronounced during Arctic winter conditions. This degradation is expected to disappear when additional 50r1 data is included in future training cycles.

Evaluation

Interactive scorecards presenting the new cycle performance are coming soon.

Known issues

Known issues with the AIFS Single forecast can be found on the following page: Known AIFS Single Forecasting Issues. This page will be updated as issues are investigated and resolved.

Key configuration values

*The MARS keyword stream=oper will be used for the new wave parameters introduced in this upgrade.

**Levels (hPa): 10 (new), 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000

Availability of AIFS Single v2 test data

Test data in MARS 

Coming soon!

Test data in dissemination

Coming soon!

Test data on ECMWF's Open Data platform

Coming soon!