https://ml4wx.github.io/Machine learning for weather & climate, tracked every weekhttps://ml4wx.github.io/papers/doi-10.1038-s41586-026-10953-2/doi-10.1038-s41586-026-10953-2Thu, 22 Oct 2026 08:00:00 +0000Operational AI ensemble WeatherNext Cyclones predicts tropical cyclone track, intensity and size to 15 days, gaining about 1 day of lead time over leading operational models.https://ml4wx.github.io/papers/arxiv-2509.18994/arxiv-2509.18994Thu, 22 Oct 2026 08:00:00 +0000Updates ECMWF's AIFS with bounding layers for physical consistency, better precipitation and new variables; operational at ECMWF since February 2025.https://ml4wx.github.io/papers/arxiv-2509.13523/arxiv-2509.13523Thu, 22 Oct 2026 08:00:00 +0000AERIS is a 1.3 to 80B parameter Swin diffusion transformer reaching 10.2 ExaFLOPS on Aurora, stable to 90-day forecasts and outperforming IFS ENS.https://ml4wx.github.io/papers/arxiv-2509.12490/arxiv-2509.12490Thu, 22 Oct 2026 08:00:00 +0000SamudrACE is a coupled atmosphere-ocean ML emulator giving stable centuries-long runs at 1 degree with realistic ENSO variability absent in uncoupled mode.https://ml4wx.github.io/papers/arxiv-2503.23953/arxiv-2503.23953Thu, 22 Oct 2026 08:00:00 +0000ACE2 produces skilful winter seasonal forecasts from SST-persisted initialisation, reaching NAO correlation 0.47 over 1993–2016, comparable to a physics-based model.https://ml4wx.github.io/papers/arxiv-2412.15832/arxiv-2412.15832Thu, 22 Oct 2026 08:00:00 +0000AIFS-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.https://ml4wx.github.io/papers/arxiv-2412.12971/arxiv-2412.12971Thu, 22 Oct 2026 08:00:00 +0000ArchesWeather 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.https://ml4wx.github.io/papers/arxiv-2412.11973/arxiv-2412.11973Thu, 22 Oct 2026 08:00:00 +0000A NeuralGCM-based hybrid model trained on satellite precipitation cuts climate-model precipitation biases and beats the ECMWF ensemble at mid-range forecasts.https://ml4wx.github.io/papers/arxiv-2411.11268/arxiv-2411.11268Thu, 22 Oct 2026 08:00:00 +0000ACE2 is a 450M-parameter emulator that conserves dry air and moisture, runs about 1500 simulated years per day, and reproduces El Niño responses over 80 years.https://ml4wx.github.io/papers/arxiv-2410.14932/arxiv-2410.14932Thu, 22 Oct 2026 08:00:00 +0000FourCastNet trained without Category 3-5 cyclones cannot forecast Category 5 storms, showing AI weather models fail to extrapolate to unseen extremes.https://ml4wx.github.io/papers/arxiv-2410.01776/arxiv-2410.01776Thu, 22 Oct 2026 08:00:00 +0000Combining RCM dynamical downscaling with diffusion refinement lets large CMIP6 ensembles be downscaled cheaply, giving better uncertainty bounds than BCSD and smaller dynamical ensembles.https://ml4wx.github.io/papers/arxiv-2409.18529/arxiv-2409.18529Thu, 22 Oct 2026 08:00:00 +0000Shows 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.https://ml4wx.github.io/papers/arxiv-2409.02891/arxiv-2409.02891Thu, 22 Oct 2026 08:00:00 +0000Graph-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.https://ml4wx.github.io/papers/arxiv-2408.10958/arxiv-2408.10958Thu, 22 Oct 2026 08:00:00 +0000StormCast, a generative diffusion emulator of the 3 km HRRR convection-allowing model, gives competitive 1-6 hour radar reflectivity skill with realistic cold pools and power spectra.https://ml4wx.github.io/papers/arxiv-2408.05472/arxiv-2408.05472Thu, 22 Oct 2026 08:00:00 +0000FuXi Weather is an ML system that assimilates multi-satellite observations and produces 10-day global forecasts, outperforming ECMWF HRES in observation-sparse regions like central Africa.https://ml4wx.github.io/papers/arxiv-2408.03100/arxiv-2408.03100Thu, 22 Oct 2026 08:00:00 +0000A 1.1-billion-parameter SFNO ensemble with bred vectors and perturbed parameters passes spectral tests and is calibrated, setting up ML huge ensembles of 1,000+ members.https://ml4wx.github.io/papers/arxiv-2407.00124/arxiv-2407.00124Thu, 22 Oct 2026 08:00:00 +0000A U-Net emulator of embedded cloud-resolving models gives a stable 5-year hybrid climate run with realistic cloud condensate and a 0.96 mm/day precipitation RMSE.https://ml4wx.github.io/papers/doi-10.1029-2023gl107377/doi-10.1029-2023gl107377Thu, 22 Oct 2026 08:00:00 +0000Shows Pangu-Weather, FourCastNet and GraphCast fail to reproduce sub-synoptic and mesoscale phenomena, lacking physical consistency versus physics-based models.https://ml4wx.github.io/papers/arxiv-2406.01465/arxiv-2406.01465Thu, 22 Oct 2026 08:00:00 +0000AIFS is ECMWF's operational GNN-transformer forecast system, trained on ERA5 and operational analyses, with skilled upper-air, surface and tropical cyclone forecasts run publicly.https://ml4wx.github.io/papers/arxiv-2405.13063/arxiv-2405.13063Thu, 22 Oct 2026 08:00:00 +0000Aurora is a large Earth-system foundation model trained on over a million hours of data that beats operational systems for air quality, waves, cyclone tracks and high-resolution weather.https://ml4wx.github.io/papers/arxiv-2405.05925/arxiv-2405.05925Thu, 22 Oct 2026 08:00:00 +0000FuXi-ENS, a VAE-based ensemble trained with CRPS and KL loss at 0.25°, beats ECMWF ENS in CRPS on 98.1% of 360 variable and lead time combinations.https://ml4wx.github.io/papers/arxiv-2404.00411/arxiv-2404.00411Thu, 22 Oct 2026 08:00:00 +0000End-to-end model ingests raw observations to produce global and station forecasts, beating operational NWP on several variables using 8% of input data.https://ml4wx.github.io/papers/arxiv-2402.00059/arxiv-2402.00059Thu, 22 Oct 2026 08:00:00 +0000First data-driven global model at 0.09 degree resolution, initialized by transfer from a low-resolution model, outperforming IFS-HRES in 2022 hindcasts.https://ml4wx.github.io/papers/arxiv-2401.15305/arxiv-2401.15305Thu, 22 Oct 2026 08:00:00 +0000Lagged ensembles give a parameter-free probabilistic comparison showing GraphCast and Pangu tie on CRPS, and that multi-step losses hurt probabilistic skill.https://ml4wx.github.io/papers/arxiv-2312.15796/arxiv-2312.15796Thu, 22 Oct 2026 08:00:00 +0000GenCast, a diffusion ensemble trained on reanalysis, beats ECMWF ENS on 97.4% of 1320 targets, with better tropical cyclone and wind power prediction, in 8 minutes.https://ml4wx.github.io/papers/arxiv-2312.12462/arxiv-2312.12462Thu, 22 Oct 2026 08:00:00 +0000Adas is an AI data assimilation model that, combined with FengWu, forms an end-to-end AI forecast system with skill comparable to IFS.https://ml4wx.github.io/papers/arxiv-2312.12455/arxiv-2312.12455Thu, 22 Oct 2026 08:00:00 +0000Couples the FengWu forecast model with 4D-Var using autodifferentiation instead of adjoints, producing cyclic analyses from simulated observations.https://ml4wx.github.io/papers/arxiv-2312.09926/arxiv-2312.09926Thu, 22 Oct 2026 08:00:00 +0000Machine learning S2S model outperforms ECMWF's S2S system for precipitation and OLR up to 42 days, extending MJO prediction skill from 30 to 36 days.https://ml4wx.github.io/papers/arxiv-2311.07222/arxiv-2311.07222Thu, 22 Oct 2026 08:00:00 +0000NeuralGCM 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.https://ml4wx.github.io/papers/arxiv-2310.02074/arxiv-2310.02074Thu, 22 Oct 2026 08:00:00 +0000ACE is a 200M-parameter autoregressive emulator that stays stable for 100 years, reproduces a 100 km reference model's climate, and runs about 100x faster.https://ml4wx.github.io/papers/arxiv-2309.15214/arxiv-2309.15214Thu, 22 Oct 2026 08:00:00 +0000Two-step UNet plus diffusion residual model downscales 25 km reanalysis to 2 km over Taiwan, recovering spectra and sharpening fronts and typhoon structures.https://ml4wx.github.io/papers/arxiv-2308.15560/arxiv-2308.15560Thu, 22 Oct 2026 08:00:00 +0000WeatherBench 2 provides an open evaluation framework, training and ground-truth data, and baselines for data-driven global 1-14 day forecasts.https://ml4wx.github.io/papers/arxiv-2307.10128/arxiv-2307.10128Thu, 22 Oct 2026 08:00:00 +0000Compares PanguWeather with ECMWF IFS and finds comparable skill for global and extreme metrics, with forecast smoothness and bias drift as drawbacks.https://ml4wx.github.io/papers/doi-10.1038-s41586-023-06184-4/doi-10.1038-s41586-023-06184-4Thu, 22 Oct 2026 08:00:00 +0000NowcastNet unifies advection physics with learned conditional generation for extreme precipitation nowcasts up to 3 hours, ranked first by meteorologists in 71% of cases.https://ml4wx.github.io/papers/doi-10.5194-essd-15-2635-2023/doi-10.5194-essd-15-2635-2023Thu, 22 Oct 2026 08:00:00 +0000Releases EUPPBench, a time-aligned forecast-observation dataset with evaluation methods and a first benchmark of 2 m temperature postprocessing.https://ml4wx.github.io/papers/arxiv-2306.14066/arxiv-2306.14066Thu, 22 Oct 2026 08:00:00 +0000SEEDS uses diffusion models to emulate operational weather ensembles, generating thousands of realistic members cheaply with improved reliability and extreme-event probabilities after bias correction.https://ml4wx.github.io/papers/arxiv-2306.12873/arxiv-2306.12873Thu, 22 Oct 2026 08:00:00 +0000FuXi is a cascaded ML system giving 15-day global forecasts at 0.25 degrees, the first ML model to match the ECMWF ensemble mean at 15 days.https://ml4wx.github.io/papers/arxiv-2306.06079/arxiv-2306.06079Thu, 22 Oct 2026 08:00:00 +0000MetNet-3 learns from sparse and dense sensors to forecast precipitation, wind, temperature and dew point up to 24 hours ahead, beating HRRR and ENS over CONUS and running in Google Search.https://ml4wx.github.io/papers/arxiv-2304.02948/arxiv-2304.02948Thu, 22 Oct 2026 08:00:00 +0000FengWu is a multi-modal transformer forecasting system that extends skillful forecasts to 10.75 days, beating GraphCast on 80% of 880 predictands.https://ml4wx.github.io/papers/doi-10.1126-sciadv.adf2827/doi-10.1126-sciadv.adf2827Thu, 22 Oct 2026 08:00:00 +00003D-Geoformer, a self-attention transformer, predicts Niño3.4 SST anomalies 18 months ahead with high correlation skill for ENSO.https://ml4wx.github.io/papers/arxiv-2301.10343/arxiv-2301.10343Thu, 22 Oct 2026 08:00:00 +0000ClimaX is a transformer foundation model pretrained on CMIP6 that fine-tunes to many weather and climate tasks, outperforming data-driven baselines.https://ml4wx.github.io/papers/arxiv-2212.12794/arxiv-2212.12794Thu, 22 Oct 2026 08:00:00 +0000GraphCast, a graph neural network trained on ERA5, makes 10-day global forecasts at 0.25 degrees in under a minute, beating operational systems on 90% of targets.https://ml4wx.github.io/papers/arxiv-2211.02556/arxiv-2211.02556Thu, 22 Oct 2026 08:00:00 +00003D Earth-Specific Transformer trained on ERA5 is the first AI model to beat IFS across variables and lead times from 1 hour to one week.https://ml4wx.github.io/papers/doi-10.1029-2021ms002954/doi-10.1029-2021ms002954Thu, 22 Oct 2026 08:00:00 +0000Introduces ClimateBench, a CMIP-based benchmark with baseline ML emulators predicting temperature and precipitation responses to emissions scenarios.https://ml4wx.github.io/papers/arxiv-2208.05419/arxiv-2208.05419Thu, 22 Oct 2026 08:00:00 +0000FourCastNet uses adaptive Fourier neural operators to produce global medium-range forecasts about 80,000 times faster than NWP, scaling to thousands of GPUs.https://ml4wx.github.io/papers/arxiv-2202.11214/arxiv-2202.11214Thu, 22 Oct 2026 08:00:00 +0000FourCastNet uses adaptive Fourier neural operators to forecast global 0.25 degree weather in under 2 seconds, matching IFS at short leads and beating it for precipitation.https://ml4wx.github.io/papers/doi-10.1029-2021ms002794/doi-10.1029-2021ms002794Thu, 22 Oct 2026 08:00:00 +0000Machine-learned nudging tendencies correct a 200-km FV3GFS model toward 3-km X-SHiELD, gaining 18 hours of 3-7 day skill and 30% better precipitation.https://ml4wx.github.io/papers/arxiv-2112.08440/arxiv-2112.08440Thu, 22 Oct 2026 08:00:00 +0000Climate-invariant ML embeds physical knowledge into parameterization emulators so they generalize across climate regimes, maintaining offline accuracy in three atmospheric models.https://ml4wx.github.io/papers/doi-10.1038-s41467-021-24262-x/doi-10.1038-s41467-021-24262-xThu, 22 Oct 2026 08:00:00 +0000Machine learning that accounts for model uncertainty detects a physically interpretable anthropogenic signal in extreme precipitation across all global observational datasets.https://ml4wx.github.io/papers/arxiv-2104.00954/arxiv-2104.00954Thu, 22 Oct 2026 08:00:00 +0000DGMR, a deep generative model of radar, produces skillful 5-90 minute precipitation nowcasts that Met Office experts ranked first in 88% of cases.