★ An update to ECMWF's machine-learned weather forecast model AIFS
Updates ECMWF's AIFS with bounding layers for physical consistency, better precipitation and new variables; operational at ECMWF since February 2025.
ECMWF
47 papers in the archive · tier-1 watchlist.
★ AIFS-CRPS: Ensemble forecasting using a model trained with a loss function based on the Continuous Ranked Probability Score
AIFS-CRPS trains an ML ensemble model on a CRPS-based loss, outperforming the IFS ensemble for most medium-range variables and competing at subseasonal lead times.
★ ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting
ArchesWeather is a transformer deterministic model, and ArchesWeatherGen a flow-matching ensemble that beats IFS ENS and NeuralGCM on WeatherBench headline variables at academic compute cost.
★ Robustness of AI-based weather forecasts in a changing climate
Shows weather-trained ML models forecast skillfully across pre-industrial, present-day and +2.9K climates, though two models show global-mean drift biases toward their training climate.
★ Regional data-driven weather modeling with a global stretched-grid
Graph-based stretched-grid model refines a global AI model to 2.5 km over the Nordics, beating the MEPS ensemble mean for 2 m temperature but underestimating extremes.
Regional & km-scale modelsECMWFMET Norway (Norwegian Meteorological Institute)
Machine Learning for the Physics of Climate
Review of ML across climate physics, covering observation reconstruction, subgrid parameterization, causal inference and climate and weather prediction.
Hybrid physics-ML & parameterizationECMWFInriaLawrence Berkeley National Laboratory / NERSCUniversity of Chicago
★ On Some Limitations of Current Machine Learning Weather Prediction Models
Shows Pangu-Weather, FourCastNet and GraphCast fail to reproduce sub-synoptic and mesoscale phenomena, lacking physical consistency versus physics-based models.
★ Aardvark weather: end-to-end data-driven weather forecasting
End-to-end model ingests raw observations to produce global and station forecasts, beating operational NWP on several variables using 8% of input data.
Data assimilationThe Alan Turing InstituteECMWFGoogle DeepMindMicrosoft Research (AI for Science)University of Cambridge (Aardvark)
★ Neural General Circulation Models for Weather and Climate
NeuralGCM couples a differentiable dynamical solver with ML components, matching ML and ECMWF ensemble skill for 1-15 day forecasts and stable multi-decade climate runs.
Hybrid physics-ML & parameterizationECMWFGoogle DeepMindGoogle ResearchMassachusetts Institute of Technology (MIT)
Identifying and characterising trapped lee waves using deep learning techniques
A deep-learning segmentation model with transfer learning from synthetic fields detects and characterises trapped lee waves hazardous to aviation, faster than spectral methods.
On some limitations of data-driven weather forecasting models
Analysis of Pangu-Weather shows its forecasts lack the fidelity and physical consistency of physics-based models, suggesting part of its deterministic skill advantage comes from these peculiarities.
★ WeatherBench 2: A benchmark for the next generation of data-driven global weather models
WeatherBench 2 provides an open evaluation framework, training and ground-truth data, and baselines for data-driven global 1-14 day forecasts.
Benchmarks, datasets & evaluationECMWFGoogle DeepMindGoogle Research
Deep learning and a changing economy in weather and climate prediction
Nature Reviews review of deep learning's impact on weather and climate prediction; abstract absent, judged from title and venue.
Global medium-range NWPECMWFETH ZurichMax Planck Institute for MeteorologyUniversity of Oklahoma (AI2ES)
★ The rise of data-driven weather forecasting
Compares PanguWeather with ECMWF IFS and finds comparable skill for global and extreme metrics, with forecast smoothness and bias drift as drawbacks.
★ The EUPPBench postprocessing benchmark dataset v1.0
Releases EUPPBench, a time-aligned forecast-observation dataset with evaluation methods and a first benchmark of 2 m temperature postprocessing.
Benchmarks, datasets & evaluationDeutscher Wetterdienst (DWD)ECMWFKarlsruhe Institute of Technology (KIT)Met OfficeMétéo-France (incl. CNRM)MeteoSwisscode
Evaluation of Heat Wave Predictability Skills of Numerical Weather Models
S2S models overestimate heatwave frequency over northern Nigeria at 30-day lead; deep-learning bias correction improves forecast reliability.
Extremes & hazardsECMWFNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Statistical Modeling of 2-m Temperature and 10-m Wind Speed Forecast Errors
Linear regression, random forests and neural networks correct ECMWF IFS 2 m temperature and 10 m wind errors, improving RMSE by 10–15% with uncertainty estimates.
Outcomes of the WMO Prize Challenge to Improve Subseasonal to Seasonal Predictions Using Artificial Intelligence
WMO prize challenge winners used AI to post-process ECMWF S2S forecasts, beating bias-corrected references for weeks 3-6 precipitation and temperature.
Sub-seasonal to seasonalColumbia University (LEAP, Lamont-Doherty)Environment and Climate Change Canada (ECCC)ECMWFMax Planck Institute for MeteorologySwiss Data Science Center (SDSC)
Online model error correction with neural networks in the incremental 4D-Var framework
Online neural-network model error correction is added to incremental 4D-Var in ECMWF's OOPS, showing online learning beats offline learning on a quasi-geostrophic model.
Improving the prediction of the Madden–Julian Oscillation of the ECMWF model by post-processing
Machine learning post-processing of ECMWF forecasts improves MJO prediction, outperforming multiple linear regression in location and intensity skill.
Post-processing & calibrationECMWFKarlsruhe Institute of Technology (KIT)
Challenges and Benchmark Datasets for Machine Learning in the Atmospheric Sciences: Definition, Status, and Outlook
Provides a recipe for building atmospheric-science ML benchmark datasets and lists domain-specific challenges and needed benchmarks for weather, climate and air quality.
Benchmarks, datasets & evaluationECMWFForschungszentrum Jülich / Jülich Supercomputing CentreNSF NCAR / UCARNVIDIAUniversity of Oklahoma (AI2ES)
ESA-ECMWF Report on recent progress and research directions in machine learning for Earth System observation and prediction
Workshop report on machine learning for Earth system observation and prediction, outlining opportunities and challenges across satellite and forecasting applications.
Implementation of a machine-learned gas optics parameterization in the ECMWF Integrated Forecasting System
Implements a neural-network gas optics scheme in ECMWF's IFS, speeding radiation by about one third with a similar model climate in 1-year coupled runs.
Hybrid physics-ML & parameterizationECMWFUniversity of Reading
Deep Learning to Estimate Model Biases in an Operational NWP Assimilation System
Convolutional networks learn IFS model bias from radio occultation temperatures, pretrained on ERA5 and transfer-learned on one year of IFS data, as an alternative to weak-constraint 4D-Var.
A Generative Deep Learning Approach to Stochastic Downscaling of Precipitation Forecasts
GAN and VAE-GAN downscale low-resolution forecasts to radar-scale precipitation maps with CRPS matching state-of-the-art pointwise post-processing, including heavy rain.
Dynamical–statistical seasonal forecasts of winter and summer precipitation for the Island of Ireland
Hybrid neural-network and regression models using GloSea5 and SEAS5 pressure indices forecast Irish winter and summer rainfall up to 4 months ahead.
Machine Learning Emulation of 3D Cloud Radiative Effects
Neural networks correct the ecRad 1D radiation scheme for 3D cloud effects, recovering 3D signal with 20-30% error at about 1% extra runtime versus the 5x costlier SPARTACUS.
Hybrid physics-ML & parameterizationECMWFUniversity of Reading
Challenges when preparing machine learning tools for use in operational weather predictions
Talk slides on the challenges of preparing machine learning tools for operational weather prediction; abstract gives no quantitative results.
Building Tangent‐Linear and Adjoint Models for Data Assimilation With Neural Networks
Neural network emulator of gravity wave drag yields tangent-linear and adjoint models that succeed in 4D-Var assimilation, easing code maintenance.
Process‐Based Climate Model Development Harnessing Machine Learning: III. The Representation of Cumulus Geometry and Their 3D Radiative Effects
Calibrates cloud geometry parameters for 3D radiation using ML surrogate emulators, reducing top-of-atmosphere and surface flux errors from about 10 to 5 W/m2.
Hybrid physics-ML & parameterizationECMWFMétéo-France (incl. CNRM)The Alan Turing Institute
Learning earth system models from observations: machine learning or data assimilation?
Perspective framing data assimilation and machine learning as equivalent inverse methods in a Bayesian network, e.g. linking 4D-Var to recurrent neural networks, to learn earth system models from observations.
Statistical approaches to assimilate ASCAT soil moisture information—I. Methodologies and first assessment
Compares CDF-matching and neural network retrievals of ASCAT soil moisture for NWP assimilation, finding inversion better suited than forward modelling.
Machine learning emulation of gravity wave drag in numerical weather forecasting
Neural emulators of the non-orographic gravity wave drag scheme run stably to seasonal timescales, improve medium-range forecasts and run ten times faster on GPUs than the CPU scheme.
Hybrid physics-ML & parameterizationECMWFUniversity of Oxford
Machine Learning for Model Error Inference and Correction
Neural networks reproduce weak-constraint 4D-Var model error corrections in the operational ECMWF IFS and could extend correction to the whole atmospheric column.
Accelerating Radiation Computations for Dynamical Models With Targeted Machine Learning and Code Optimization
Neural networks replace gas optics in the RRTMGP radiation scheme, running 1–6 times faster with heating-rate errors below 0.1 K/day, giving 3.5x faster longwave fluxes.
Hybrid physics-ML & parameterizationECMWFNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Machine-Learned Preconditioners for Linear Solvers in Geophysical Fluid Flows
Machine-learned preconditioners for the semi-implicit linear solver in a shallow-water model are competitive with conventional preconditioners and generalize beyond the training range.
Hybrid physics-ML & parameterizationECMWFNSF NCAR / UCARUniversity of Oxford
TRU-NET: A Deep Learning Approach to High Resolution Prediction of Rainfall
TRU-NET, a convolutional-recurrent U-Net with cross attention, predicts high-resolution rainfall from climate model fields with lower RMSE than a dynamical model's rainfall output.
A comparative study of convolutional neural network models for wind field downscaling
Compares CNN architectures and proposes DeepRU, a U-Net variant, to downscale ERA5-driven 100 m wind forecasts from 31 km to 9 km over the Alps.
Downscaling & super-resolutionECMWFTechnical University of Munich (TUM)
Statistical post‐processing of heat index ensemble forecasts: Is there a royal road?
Compares calibration of heat index ensembles, finding direct GEV-based EMOS gives the best skill up to day 6; machine learning performs comparably only for warning categories.
Deep Learning for Post-Processing Ensemble Weather Forecasts
A deep neural network post-processes a subset of ensemble trajectories, improving CRPS by over 14% and allowing fewer members to match the full ensemble.
★ Statistical Postprocessing for Weather Forecasts -- Review, Challenges and Avenues in a Big Data World
Review of statistical postprocessing methods for NWP forecasts and their challenges in big-data settings, with little explicit machine learning focus.
Post-processing & calibrationDeutscher Wetterdienst (DWD)ECMWFKarlsruhe Institute of Technology (KIT)KNMI (Royal Netherlands Meteorological Institute)MET Norway (Norwegian Meteorological Institute)Met OfficeMétéo-France (incl. CNRM)MeteoSwiss
★ WeatherBench: A benchmark dataset for data-driven weather forecasting
WeatherBench introduces an ERA5-derived benchmark with standard metrics and baselines (linear regression, deep learning, physical models) for data-driven medium-range global weather forecasting.
Benchmarks, datasets & evaluationECMWFTechnical University of Munich (TUM)University of Washingtoncode
★ WeatherBench: A Benchmark Data Set for Data-Driven Weather Forecasting
WeatherBench provides an ERA5-derived benchmark with metrics and baselines for data-driven global medium-range forecasting, enabling direct model comparison.
Benchmarks, datasets & evaluationECMWFUniversity of WashingtonTechnical University of Munich (TUM)code
Predicting Weather Uncertainty with Deep Convnets
Deep 3D U-Net variants estimate weather forecast ensemble spread from a small set of numerical simulations, reducing the compute needed for uncertainty quantification.
SMOS Neural Network Soil Moisture Data Assimilation in a Land Surface Model and Atmospheric Impact
Neural-network soil moisture retrievals from SMOS are assimilated into ECMWF's land model, improving Southern Hemisphere and some Northern Hemisphere forecasts through atmospheric feedback.
★ Challenges and design choices for global weather and climate models based on machine learning
Discusses challenges and design choices for deep-learning global weather forecasting, illustrated with a toy global neural forecast model.
A Neural Network Approach for a Fast and Accurate Computation of a Longwave Radiative Budget
Early neural network approach computes longwave radiative budgets quickly and accurately for atmospheric models; abstract unavailable.