AIFS ENS

Version

Status

Summary

v1

AIFS ENS version 1 is the first operational version of the AIFS ensemble CRPS model. It was released on 27 August 2025, superseding the previous experimental diffusion model. The diffusion model used a different methodology to AIFS ENS v1, therefore version 1 should not be compared with the previous experimental version, but instead it should be seen as the first version of a new model. Full details can be found on the dedicated implementation page for AIFS ENS.

Resource	Link
Operational implementation page	Implementation of AIFS ENS v1
arXiv paper	AIFS-CRPS: Ensemble forecasting using a model trained with a loss function based on the continuous ranked probability score, Lang, S., Alexe, M., Clare, M., Roberts, C., Adewoyin, R., Ben Bouallègue, Z., Chantry, M., Dramsch, J., Dueben, P., Hahner, S., Maciel, P., Prieto-Nemesio, A., O’Brien, C., Pinault, F., Polster, J., Raoult, B., Tietsche, S., Leutbecher, M.
Model card	https://huggingface.co/ecmwf/aifs-ens-1.0
Demonstrative notebook	https://huggingface.co/ecmwf/aifs-ens-1.0/blob/main/run_AIFS_ENS_v1.ipynb
Model performance scorecard	Scorecard of AIFS ENS compared against IFS
Data access	https://www.ecmwf.int/en/forecasts/dataset/aifs-machine-learning-data

AIFS ENS-CRPS v1 (operational)

Model description

Version 1 of the AIFS ensemble CRPS model is a probabilistic weather forecasting system developed by ECMWF that uses machine learning to generate ensemble forecasts. It is implemented using the Anemoi framework, which consists of open source components, enabling the user community to run AIFS themselves. The model and an example notebook are freely available to the user community, demonstrating how to run AIFS ENS from ECMWF open data and the anemoi-inference package.

The model is trained via a probabilistic training approach based on the Continuous Ranked Probability Score (CRPS), a loss function that helps ensure the forecasts are both accurate and well-calibrated.

Model performance

Interactive scorecards presenting model performance across the period March 2024 to March 2025 are available:

Scorecard of AIFS ENS compared against IFS

Model resolution

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)*
1	~32 km	0.25°	13

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

Model output parameters

A list of parameters output by AIFS-ENS is provided on the implementation page. All parameters are output in GRIB2 format and available via dissemination, the MARS archive, open data platforms and OpenCharts. See https://www.ecmwf.int/en/forecasts/dataset/aifs-machine-learning-data for more information on how to access AIFS model output data.

v0.1

Experimental version 0.1 was the first version of the AIFS ensemble diffusion model. It was released on 21 June 2024 and superseded by ENS-CRPS operational version 1.

AIFS ENS-DIFF v0.1

This blog post introduces version 0.1 of the AIFS ensemble diffusion model. It comprised 51 members, using the same initial conditions as the IFS ensemble system.

Model resolution

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)*
0.1	~111 km	1°	13

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

Model output parameters

The parameters listed in the table below were output by AIFS ensemble version 0.1.

EXISTING PARAMETERS
Param ID	Short name	Name	Units	Level type
129	z	Geopotential	m² s^-2	Pressure
130	t	Temperature	K	Pressure
131	u	U component of wind	m s^-1	Pressure
132	v	V component of wind	m s^-1	Pressure
133	q	Specific humidity	kg kg^-1	Pressure
135	w	Vertical velocity	Pa s^-1	Pressure
151	msl	Mean sea level pressure	Pa	Surface
165	10u	10 metre U wind component	m s^-1	Surface
166	10v	10 metre V wind component	m s^-1	Surface
167	2t	2 metre temperature	K	Surface
228	tp	Total precipitation	m	Surface

AIFS Single

Version

Status

Summary

v1.1

Version 1.1 is the current operational version of the AIFS deterministic model. It was released on 27 August 2025. Full details can be found on the dedicated implementation page for AIFS Single.

Resource	Link
Operational implementation page	Implementation of AIFS Single v1
arXiv paper	An update to ECMWF’s machine-learned weather forecast model AIFS – Moldovan, G., Pinnington, E., Prieto Nemesio, A., Lang, S., Ben Bouallègue, Z., Dramsch, J., Alexe, M., Santa Cruz, M., Hahner, S., Cook, H., Theissen, H., Clare, M., O'Brien, C., Polster, J., Magnusson, L., Mertes, G., Pinault, F., Raoult, B., de Rosnay, P., Forbes, R., Chantry, M.
Model card	https://huggingface.co/ecmwf/aifs-single-1.1
Demonstrative notebook	https://huggingface.co/ecmwf/aifs-single-1.1/blob/main/run_AIFS_v1.1.ipynb
Model performance scorecard	Scorecard of AIFS Single v1.1 compared against IFS Scorecard of AIFS Single v1.1 compared against AIFS 0.2.1
Data access	https://www.ecmwf.int/en/forecasts/dataset/aifs-machine-learning-data

AIFS Single v1.1 (operational)

The upgrade of AIFS Single to version 1.1 represents a slight modification to the AIFS Single model. This new version addresses point rain artefacts identified in AIFS Single v1 (see Known AIFS Single Issues P3).

The underlying cause was identified as an unphysical relationship between soil moisture and precipitation, caused by an overemphasis of soil moisture in the training.

This new version of AIFS Single minimally changes the training configuration to a lower (by a factor 100) contribution from soil moisture. The outcome is a model with equivalent (but not identical) skill and bias.

No other architectural changes were introduced; further details about the model are provided in the description of AIFS Single v1 (below).

v1

Version 1 was the first operational version of the AIFS deterministic model. It was released on 25 February 2025 and superseded by operational version 1.1 on 27 August 2025. Full details can be found on the dedicated implementation page for AIFS Single.

AIFS Single v1

Model description

Version 1 of the AIFS deterministic model was implemented using the Anemoi framework. Anemoi consists of open source components, enabling the user community to run AIFS themselves. The model and an example notebook are freely available to the user community, demonstrating how to run the AIFS from ECMWF open data and the anemoi-inference package.

The model architecture is identical to the previous experimental versions (v0.2 and v0.2.1). However, the training regime and pressure level scaling is revised, and a new way to handle bounded variables is introduced. In addition, more data are used for the fine-tuning step towards operational IFS analyses. The changes result in improved forecast skill, particularly for precipitation.

Model performance

Scorecards presenting performance of AIFS Single v1 across 2024 are available on Hugging Face.

Model resolution

There were no changes to the resolution of version 1.

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)*
1	~32 km	0.25°	13
0.2.1	~32 km	0.25°	13
0.2	~32 km	0.25°	13
0.1	~111 km	1°	13

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

Model output parameters

AIFS Single v1 introduces new parameters, detailed below.

NEW PARAMETERS
Param ID	Short Name	Name	Units	Level Type	Comments
160	sdor	Standard deviation of sub-gridscale orography	m	sfc	Please note that sdor is not forecast by AIFS Single v1. However, this parameter exists in the GRIB 2 output to enable the autoregressive forecast for future time steps.
163	slor	Slope of sub-gridscale orography	Numeric	sfc	Please note that slor is not forecast by AIFS Single v1. However, this parameter exists in the GRIB 2 output to enable the autoregressive forecast for future time steps.
169	ssrd	Surface short-wave (solar) radiation downwards	J m^-2	sfc
175	strd	Surface long-wave (thermal) radiation downwards	J m^-2	sfc
3073	lcc	Low cloud cover	%	sfc
3074	mcc	Medium Cloud Cover	%	sfc
3075	hcc	High Cloud Cover	%	sfc
228144	sf	Snowfall water equivalent	kg m^-2	sfc
228164	tcc	Total cloud cover	%	sfc
228246	100u	100 metre U wind component	m s^-1	sfc
228247	100v	100 metre V wind component	m s^-1	sfc
231002	rowe	Runoff water equivalent (surface plus subsurface)	kg m^-2	sfc
260199	vsw	Volumetric soil moisture	m³ m^-3	sol
260360	sot	Soil temperature	K	sol

Two existing parameters already introduced with AIFS Single v0.2.1 see changes due to GRIB 2 formatting with the implementation of AIFS Single v1.

Short name	Name	Old Param ID	Old units	New Param ID	New units
cp	Convective precipitation	143	m	228143	kg m^-2
tp	Total precipitation	228	m	228228	kg m^-2

All other existing parameters, listed on the implementation page, remain unchanged.

All parameters are output in GRIB2 format and available via dissemination, the MARS archive, open data platforms and OpenCharts. See https://www.ecmwf.int/en/forecasts/dataset/aifs-machine-learning-data for more information on how to access AIFS model output data.

v0.2.1

Experimental version 0.2.1 of the AIFS single model was released on 21 February 2024. This was superseded by operational version 1 on 25 February 2025.

AIFS Single v0.2.1

This blog post introduces version 0.2.1 of the AIFS deterministic model. Full scientific details are available in the arXiv preprint.

Model description

The architecture and training regime stay much the same as version 0.2, but the big addition is precipitation, both total and the convective contribution. Further details about the model architecture are provided in AIFS Single v0.2.

A scorecard comparing performance between v0.2 and v0.2.1 is available.

The experimental AIFS Single v0.2.1 model is available on Hugging Face. An example notebook is available to the user community, demonstrating how to run the AIFS from ECMWF open data and the anemoi-inference package.

Model resolution

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)*
0.2.1	~32 km	0.25°	13
0.2	~32 km	0.25°	13
0.1	~111 km	1°	13

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

Model output parameters

AIFS Single v0.2.1 introduced three new surface parameters, detailed below.

NEW PARAMETERS
Param ID	Short name	Name	Units	Level type
136	tcw	Total column water	kg m^-2	Surface
143	cp	Convective precipitation	m	Surface
228	tp	Total precipitation	m	Surface
EXISTING PARAMETERS
Param ID	Short name	Name	Units	Level type
129	z	Geopotential	m² s^-2	Pressure
130	t	Temperature	K	Pressure
131	u	U component of wind	m s^-1	Pressure
132	v	V component of wind	m s^-1	Pressure
133	q	Specific humidity	kg kg^-1	Pressure
134	sp	Surface pressure	Pa	Surface
135	w	Vertical velocity	Pa s^-1	Pressure
151	msl	Mean sea level pressure	Pa	Surface
165	10u	10 metre U wind component	m s^-1	Surface
166	10v	10 metre V wind component	m s^-1	Surface
167	2t	2 metre temperature	K	Surface
168	2d	2 metre dewpoint temperature	K	Surface

v0.2

Experimental version 0.2 of the AIFS deterministic model was released on 10 January 2024. This was superseded by version 0.2.1 on 21 February 2024.

AIFS Single v0.2

This article provides an overview of changes introduced with AIFS Single v0.2.

Model description

Architectural changes were introduced with the upgrade from version 0.1 to 0.2. In version 0.2, the encoder and decoder use attention-based graph neural networks, very similar to a transformer (Vaswani et al., 2017) architecture.

The input and output grids for AIFS Single v0.2 are the native ERA5 reduced Gaussian grid, which provides near-constant resolution across the globe.

Model performance

This article compares performance of version 0.1 ('old AIFS') and version 0.2 ('new AIFS').

Model resolution

Improvements in model resolution were introduced with the upgrade from version 0.1 to 0.2.

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)*
0.2	~32 km	0.25°	13
0.1	~111 km	1°	13

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

Model output parameters

No new model output parameters were introduced when the experimental AIFS deterministic model upgraded from version 0.1 to 0.2. One parameter, sea surface temperature, was removed.

REMOVED PARAMETERS
Param ID	Short name	Name	Units	Level type
34	sst	Sea surface temperature	K	Surface
EXISTING PARAMETERS
Param ID	Short name	Name	Units	Level type
34	sst	Sea surface temperature	K	Surface
129	z	Geopotential	m² s^-2	Pressure
130	t	Temperature	K	Pressure
131	u	U component of wind	m s^-1	Pressure
132	v	V component of wind	m s^-1	Pressure
133	q	Specific humidity	kg kg^-1	Pressure
134	sp	Surface pressure	Pa	Surface
135	w	Vertical velocity	Pa s^-1	Pressure
151	msl	Mean sea level pressure	Pa	Surface
165	10u	10 metre U wind component	m s^-1	Surface
166	10v	10 metre V wind component	m s^-1	Surface
167	2t	2 metre temperature	K	Surface
168	2d	2 metre dewpoint temperature	K	Surface

v0.1

The first version (0.1) of the AIFS deterministic model was released on 13 October 2023. This was superseded by version 0.2 on 10 January 2024.

AIFS Single v0.1

This blog post and newsletter article introduce the first version of the AIFS deterministic model.

Model description

The first implementation of the AIFS was based on message-passing graph neural networks and with an internal icosahedral grid with multi-scale edges, similar to Deepmind’s GraphCast.

Graph Neural Networks allow us to move away from lat-lon grids, which have many points near the poles, and use reduced Gaussian grids which have near equal distance between grid-points no matter where you are on the globe. These are also the grids which are used by the IFS to create ERA5 and operational initial conditions.

Version 0.1 was trained on a subset of the ERA5 reanalysis for 1979–2018 and fine-tuned on operational IFS data from 2019 to 2020.

Model resolution

Model version	Horizontal grid resolution		Vertical resolution (pressure levels)
0.1	~111 km	1°	13 levels*

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

Model output parameters

Param ID	Short name	Name	Units	Level type
34	sst	Sea surface temperature	K	Surface
129	z	Geopotential	m² s^-2	Pressure
130	t	Temperature	K	Pressure
131	u	U component of wind	m s^-1	Pressure
132	v	V component of wind	m s^-1	Pressure
133	q	Specific humidity	kg kg^-1	Pressure
134	sp	Surface pressure	Pa	Surface
135	w	Vertical velocity	Pa s^-1	Pressure
151	msl	Mean sea level pressure	Pa	Surface
165	10u	10 metre U wind component	m s^-1	Surface
166	10v	10 metre V wind component	m s^-1	Surface
167	2t	2 metre temperature	K	Surface
168	2d	2 metre dewpoint temperature	K	Surface