Emulators of climate models and long stable rollouts (decades to centuries), scenario emulation, emulators of the coupled climate system.
Search and filter this topic Plain list on GitHub
SamudrACE is a coupled atmosphere-ocean ML emulator giving stable centuries-long runs at 1 degree with realistic ENSO variability absent in uncoupled mode.
Allen Institute for AI (Ai2, ACE)Columbia University (LEAP, Lamont-Doherty)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)New York University (M2LInES)Princeton University
ACE2 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.
Allen Institute for AI (Ai2, ACE)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
ACE 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.
Allen Institute for AI (Ai2, ACE)NVIDIAUniversity of California, Berkeley
SamudrACE is a coupled atmosphere-ocean ML emulator giving stable centuries-long runs at 1 degree with realistic ENSO variability absent in uncoupled mode.
Allen Institute for AI (Ai2, ACE)Columbia University (LEAP, Lamont-Doherty)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)New York University (M2LInES)Princeton University
cBottle is a generative diffusion emulator of global 5 km climate simulations and reanalysis on HEALPix, avoiding autoregressive rollout and reaching 12.5M-pixel resolution.
Couples the ACE2 emulator to a slab ocean and trains it across CO2 levels, reproducing equilibrium warming but violating energy conservation in transient runs.
Allen Institute for AI (Ai2, ACE)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
ACE2 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.
Allen Institute for AI (Ai2, ACE)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
DLESyM, a deep learning Earth system model, simulates 1000-year observed climate without drift, matching CMIP6 variability at a fraction of the cost.
Institute of Atmospheric Physics, Chinese Academy of SciencesUniversity of Washington
Nature Climate Change perspective on machine learning in climate modelling and analysis; abstract absent, judged from title and venue.
Allen Institute for AI (Ai2, ACE)California Institute of Technology (Caltech)Colorado State University (incl. CIRA)Columbia University (LEAP, Lamont-Doherty)Google ResearchInriaLawrence Berkeley National Laboratory / NERSCNSF NCAR / UCARNVIDIANew York University (M2LInES)University of California, BerkeleyUniversity of Oxford
LUCIE, a lightweight ERA5-trained emulator trained on 2 years of data, stays stable and climatologically consistent for 100-year, 100-member simulations on one GPU.
DYffusion combined with a spherical Fourier neural operator emulates a coarse FV3GFS climate model, producing stable 100-year probabilistic ensemble simulations at low cost.
A stochastic latent transformer emulates a stochastically forced zonal jet model with CRPS training, giving a five-order-of-magnitude speedup for large ensemble generation.
QuickClim uses machine learning to cheaply emulate climate simulations across many net-zero concentration pathways, quantifying uncertainty and pathway sensitivity.
ACE 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.
Allen Institute for AI (Ai2, ACE)NVIDIAUniversity of California, Berkeley
Perspective on climate model hierarchies since 2016, highlighting the rapid growth of machine learning in climate modelling and model use for climate risk decisions.
Uses an interpretable neural network trained on Mars Planetary Climate Model outputs to predict Gale Crater relative humidity with R² of 0.92, emulating a local climate variable.
A Gaussian process surrogate emulates the NASA Mars Global Climate Model in minutes instead of weeks for sensitivity studies.
Commentary on reservoir-computing hybrid models coupling ML to a coarse climate model and SST predictor, arguing ML could disrupt conventional climate modeling.
Evaluates Gaussian process, SVR and kernel ridge regression emulators on ClimateBench; GP performs best but is computationally expensive.
Decomposes pattern scaling errors from the MESMER emulator into scenario pattern and timeseries components, showing timeseries error is about 20% for high emissions but comparable to pattern error for low emissions.
Uses machine-learning emulators to select climate model parameter configurations, showing that conventional tuning underestimates warming uncertainty while matching present-day observations.
Bayesian Gaussian-process emulator obeying energy-balance equations emulates global and spatial surface temperature across scenarios, with interpretable parameters and uncertainty quantification.
Introduces a neural-network stochastic precipitation generator for New Zealand that reproduces daily and hourly statistics and captures climate-change drift; runs orders of magnitude faster than an RCM.
A GAN conditioned on month, location and period generates hourly 2m air temperatures over North America from NLDAS data, reproducing diurnal cycles.
Hybrid AI emulators of cloud-climate response, with an interactive visual tool, enable what-if testing of marine cloud brightening scenarios without costly Earth system runs.
Shows temporal subsampling of training data adds small-scale bias in reservoir-computing emulators of surface quasi-geostrophic turbulence, with echo state networks more robust than NVAR.
Diffusion model conditioned on monthly temperature or precipitation generates realistic daily fields matching Earth system model heat waves and dry spells at a fraction of the cost.
Attributes instability in AI weather models to spectral bias and proposes a mitigation enabling hundreds of years of physically consistent integration.
Machine-learned correction operator, trained on nudged coarse E3SM runs against ERA5, makes free-running climate statistics and extreme-event frequencies resemble observations.
Reviews machine learning's role in Earth system modelling, covering components beyond the atmosphere such as biogeochemistry and the terrestrial biosphere.
Data-calibrated PREMU emulator predicts ESM precipitation from gridded temperature modes, reproducing global land precipitation interannual variability (R 0.82-0.93) across nine models.
AiBEDO uses the Fluctuation-Dissipation Theorem and 8250 years of Earth system model output to emulate regional climate responses to marine cloud brightening forcing.
AiBEDO, an MLP built on the Fluctuation-Dissipation Theorem, emulates climate responses to marine cloud brightening forcing and matches Earth system model simulations at far lower cost.
Neural advection-diffusion equation approach to climate modeling; the abstract is missing, so the scope and results cannot be verified.
Hybrid AI surrogate links cloud radiative effects to large-scale circulation and regional climate, enabling low-cost scenario runs for cloud-based climate intervention.
UNet-based CNNs emulate interannual oxygen isotope composition of precipitation from temperature and precipitation fields, explaining about 40% of temporal variance in a last-millennium run.
LSTM combined with an unscented Kalman filter to predict the global land-ocean temperature index; the abstract is missing, so the scope is inferred from the title.
Randomly wired dense layers in ClimateBench emulators improve performance in 14 of 24 cases, up to 30.4% for MLPs, with no slower prediction.
Builds a PCA plus LSTM probabilistic surrogate for the TIE-GCM thermosphere model, giving density distributions with under 5 km satellite position bias in the 2003 storm.
A statistical GAM emulator predicts local surface temperature responses to tree cover change from ESM output, offering a cheap complement to greenhouse-gas emulators.
ETH ZurichKNMI (Royal Netherlands Meteorological Institute)Max Planck Institute for BiogeochemistryMax Planck Institute for Meteorology
Transfer learning from unused high-resolution data improves machine-learnt emulators of chaotic systems, giving better stability, generalisation and forecast skill for climate-model emulation.
MESMER-X statistical emulator reproduces spatially resolved annual maximum temperatures of Earth system models using global mean temperature as the driver.
Generic overview of neural networks and deep learning for climate modeling and prediction, with no new method or quantitative result.
Bayesian inference on a stochastic energy balance model (ClimBayes) emulates global temperature variability and separates forced from internal variance across decadal to centennial scales.
A CNN emulates perturbed-parameter ensembles of a higher-resolution Earth system model from few high-resolution runs, making calibration 20-40% more efficient.
A convolutional VAE pretrained on imperfect quasi-geostrophic simulations and transfer-learned on noisy observations gives stable stochastic long-term rollouts with accurate climatology.
MESMER-M extends an Earth system model emulator to monthly local temperatures, emulating 38 CMIP6 models with skewed, spatially correlated variability.
Chains the MAGICC global emulator with the MESMER spatial emulator to produce millions of local temperature projections per emission pathway, validated against CMIP6 ensembles.
Perspective arguing that climate emulators grounded in statistical mechanics and machine learning can better capture physics, tipping points, and reduce computational cost.
Fourier reservoir computing with localization extends skillful forecast horizons for a coupled atmosphere-ocean quasi-geostrophic model beyond standard reservoir computing.
Proposes neural diffusion equations, combining neural ODEs with diffusion, for climate modeling; it beats eleven baselines on two real and one synthetic dataset.
A Gaussian process emulator couples the HadSM3 climate model to an ice sheet model, enabling efficient million-year ice sheet–climate simulations and sensitivity tests.
ESEm, an open-source Gaussian-process emulation and calibration toolkit, reduces parametric uncertainty in a GCM and explores CMIP6 scenario uncertainty.
Neural networks trained on CESM2 output emulate 21st-century Greenland ice sheet surface melt across CMIP6 models, avoiding costly regional climate model runs.
Compares GAN and VAE deep generative models against classical weather generators for multisite precipitation synthesis, offering guidance for future climate scenario generation.
Two loosely conditioned GANs emulate 32-day global daily precipitation from a coupled Earth system model, matching dry-day statistics at a fraction of the compute.
Shows echo state networks reduce to a polynomial regression (D2R2), a SINDy generalization that outperforms reservoir computers on Lorenz 63 and 96.
Ridge, Lasso, random forest and extra trees emulate UKESM1 tropical stratospheric ozone projections under SSPs, matching the model when trained on future simulations.
Wasserstein GAN with a CNN generator learns a 64-dimensional latent space of synthetic PLASIM climate states, reproducing leading EOF statistics and geostrophic balance.
Nearest-neighbour Gaussian process and PCA emulators reproduce gridded Earth system model temperature responses to emission scenarios, enabling cheap projections for impact assessment.
DeepClimGAN, a conditional GAN, emulates Earth system model daily weather under chosen scenarios to sample extremes far more cheaply than full ESM runs.
Machine learning maps short-term climate model responses to long-term temperature change under forcing scenarios, enabling cheaper climate projections.
Meta-modeling uses cheap low-fidelity models to skip costly high-fidelity evaluations when predicting temperature fields, giving an adaptive training strategy for geophysical emulation.
Overview of neural networks for climate change prediction, discussing opportunities and challenges without new methods or quantitative results.
Position review arguing climate emulation with machine learning differs from weather forecasting, since climate is a boundary-condition problem with steady-state emulation opportunities.
Gaussian process emulator of an isotope-enabled climate model constrains Last Interglacial Greenland ice sheet geometry against ice core isotope data, reducing uncertainty.
A CNN surrogate trained on CanESM2 monthly precipitation outperforms persistence and stays skillful over 120-month rollouts, with scheduled sampling and preforecasting documented.
MESMER statistically emulates grid-point land temperature from CMIP5 models, producing a 40-model superensemble at negligible computational cost.
Gaussian process emulators of PLASIM-GENIE ensembles produce a downscaled, high-resolution reconstruction of Pliocene-Pleistocene temperature and precipitation at 1,000-year intervals.
A multi-fidelity Bayesian treed co-kriging emulator handles non-nested designs for WRF model output, giving full predictive distributions at each fidelity level.
Statistical emulator GCMET reconstructs global climate over the last 800,000 years at high spatial detail, revealing stable habitat refugia.
Statistical emulators built from perturbed-physics ensembles iteratively tune weather@home parameters, cutting northwestern US warm biases by 74%.
Neural networks trained on simple GCMs (PUMA, PLASIM) forecast weather for a few days, with skill depending on model complexity and training length.
Compares echo state networks, feed-forward ANNs and LSTMs on a multi-scale Lorenz 96 system; reservoir computing forecasts hundreds of steps and matches PDF tails.
Low-dimensional statistical emulator of the MAGICC climate model, built from designed CO2 scenarios, gives out-of-sample temperature errors below 2%.
Frames stratospheric aerosol injection as deep reinforcement learning, using a deep-learning emulator of HadCM3 as the training environment; presented as a first DRL application in climate science.
LeNet CNN on images of 30 years of global monthly temperature predicts the next decade's trend with 97% reported accuracy, as a top-down alternative to physics models.
Fldgen emulates Earth system model temperature fields with internal variability and space-time correlation, producing large ensembles from global temperature input.
Differentiable physics-informed graph networks inject domain physics into latent space, improving climate prediction tasks and multistep forecasting; details are limited in the abstract.
SVD plus a Bayesian-optimized neural network emulates 42,660 carbon-flux outputs of an Earth system ecosystem model from only 20 simulations.
R package stilt provides simple AR(1) Gaussian process emulation of time series climate model output, with cross-validation and no MCMC needed.
Statistical emulators trained on 30 GCM runs map eruption parameters to volcanic aerosol forcing, producing response surfaces for thousands of eruptions.
Deep network emulates a simple general circulation model, forecasting its states and producing stable climate runs with matching statistics and no long-term drift.
Convolutional autoencoders compress CMIP5 pre-industrial surface temperature fields and beat PCA in reconstruction error at 40 dimensions, suggesting surrogate climate model use.
Random forests trained on a multiresolution perturbed-parameter GCM ensemble predict high-resolution TOA flux and precipitation using few high-resolution runs, cutting simulation cost.
Multi-level statistical emulation of climate model responses to boundary forcing; abstract missing, so ML role is inferred from emulator-style methods and is weakly evidenced.
Bayesian neural networks emulate GCM monthly temperature and precipitation for glacial cycles, adding calibrated noise to mimic internal variability from 22 to 120 ka.
ACE2 produces skilful winter seasonal forecasts from SST-persisted initialisation, reaching NAO correlation 0.47 over 1993–2016, comparable to a physics-based model.
Sub-seasonal to seasonalAllen Institute for AI (Ai2, ACE)Met Office
Five recommendations for standardized, independent evaluation of ML-based Earth system models, stressing physical consistency for climate projections without historical analogues.
Benchmarks, datasets & evaluationColorado State University (incl. CIRA)Lawrence Berkeley National Laboratory / NERSCNSF NCAR / UCAROak Ridge National LaboratoryUniversity of California, Berkeley
A diffusion model (CPMGEM) emulates 2.2 km UK convection-permitting precipitation from 60 km GCM inputs, reproducing extremes up to about 100-year return times at much lower cost.
A 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.
Review of ML for climate physics and climate simulations, arguing physics knowledge and interpretability are essential in the small-data, non-stationary regime of long-term climate projection.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)Stanford UniversityUniversity of Chicago
Chronological review of machine learning in weather and climate modelling, from solvers and parameterization emulation to full ML forecast models, with ethics and future directions.
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)
Neural network emulates the Met Office nonorographic gravity wave scheme using only two of six inputs, and reproduces QBO and ENSO-linked gravity wave variability in a climate model.
Ensemble of 32-layer residual CNNs emulates superparameterized convection and clouds, runs stably for 5 years in CAM5, and generalizes to a +4K warmer climate.
Hybrid physics-ML & parameterizationTsinghua University (NowcastNet)
Transfer learning merges Earth system model simulations with historical observations to cut the spread of 21st-century global temperature projections by more than 50%.
Multi-fidelity randomized prior networks combine physical parameterizations and storm-resolving runs to extrapolate to +4K warming, yielding more skillful climate projections with uncertainty.
Hybrid ARIMA, K-means and autoencoder method generates synthetic North Atlantic tropical cyclone tracks and intensities from HURDAT2 for risk assessment.
DLWP-HPX uses a HEALPix mesh and GRU-augmented U-Net with fewer variables at 110 km, reaching skill about one day behind Pangu-Weather and GraphCast at one week and stable one-year rollouts.
Nature Climate Change perspective on combining AI and computing for climate modelling and prediction; abstract absent, judged from title and venue.
Hybrid physics-ML & parameterizationCalifornia Institute of Technology (Caltech)Massachusetts Institute of Technology (MIT)NSF NCAR / UCARPrinceton University
An analog-based stochastic weather generator emulates prolonged European heatwaves and extends return-time estimates, though a CNN predicts better with long training data.
SEEDS uses diffusion models to emulate operational weather ensembles, generating thousands of realistic members cheaply with improved reliability and extreme-event probabilities after bias correction.
Position paper on Earth Virtualization Engines that pair high-resolution physics-based climate models with machine learning and federated access to exabyte-scale climate data.
Software, scaling & compressionCSCS - Swiss National Supercomputing CentreETH ZurichMeteoSwissMassachusetts Institute of Technology (MIT)Max Planck Institute for MeteorologyNSF NCAR / UCARNVIDIAUniversity of Reading
Machine learning emulators (MLP and error-corrected regression with random forest) downscale GCM evapotranspiration from 12.5 km to 1.5 km, reaching KGE of 0.86 and beating statistical downscaling.
A DeepONet with autoencoder replaces the nudging bias-correction module in the E3SMv2 climate model, matching nudging tendencies with a learned surrogate for online use.
ANNs trained on climate model output estimate observed timing of 1.5°C and 2°C warming, with a 2033–2035 central estimate for 1.5°C.
Climate analysis & attributionColorado State University (incl. CIRA)Stanford University
ClimaX is a transformer foundation model pretrained on CMIP6 that fine-tunes to many weather and climate tasks, outperforming data-driven baselines.
Foundation modelsMicrosoft Research (AI for Science)University of California, Los Angeles (UCLA)code
Trains a CNN on internal variability of a long dry-core control run to predict the nonlinear jet stream response to tropospheric forcing, extending fluctuation-dissipation ideas beyond linear response.
Surrogate-based statistical framework constrains UKESM1 aerosol parameters against MODIS observations on a cloud platform, using 3-hourly outputs for tighter constraints.
Climate analysis & attributionMet OfficeUniversity of Oxford
Physically constrained cGANs correct CMIP6-class Earth system model precipitation biases in both local distributions and spatial patterns, outperforming existing methods on extremes' spatial intermittency.
Post-processing & calibrationPotsdam Institute for Climate Impact Research (PIK)Technical University of Munich (TUM)
Introduces ClimateBench, a CMIP-based benchmark with baseline ML emulators predicting temperature and precipitation responses to emissions scenarios.
Benchmarks, datasets & evaluationETH ZurichMET Norway (Norwegian Meteorological Institute)University of Oxford
Benchmarks off-the-shelf AutoML libraries on climate modelling and wind power forecasting; they fail to beat human-designed models, motivating spatiotemporal AutoML search spaces.
Argues that making Earth system models automatically differentiable enables objective calibration and hybrid ML components, targeting climate sensitivity, spatial biases and abrupt transitions.
Hybrid physics-ML & parameterizationPotsdam Institute for Climate Impact Research (PIK)Technical University of Munich (TUM)
A physically constrained GAN corrects Earth system model precipitation, fixing local distributions and the double-peaked ITCZ while preserving global precipitation sums under future scenarios.
Downscaling & super-resolutionPotsdam Institute for Climate Impact Research (PIK)Technical University of Munich (TUM)
RICE-N integrated assessment model with multi-agent reinforcement learning simulates climate-economy negotiations to test cooperation frameworks for climate mitigation agreements.
A neural-network RCM emulator hybrid-downscales daily 12 km near-surface temperature from large-scale predictors, reproducing RCM variability with training in under 2 hours and inference in under a minute.
Compares INLA and varycoef for a spatio-temporal varying-coefficient statistical emulator that downscales NARCCAP regional climate model output.
Residual neural networks replace moist physics and radiation in a realistic GCM, giving stable simulations over 10+ years with improved tropical land precipitation.
Hybrid physics-ML & parameterizationTsinghua University (NowcastNet)
A DNN surrogate replaces the process-based wildfire model in E3SM, matching observed burned area with R2 0.93 and far lower parameterization cost.
Hybrid physics-ML & parameterizationLawrence Berkeley National Laboratory / NERSC
Proposes tunable neural-network climate parameterizations calibrated online against long-term statistics, reducing long-term biases of the ML model in a simple dynamical system.
Hybrid physics-ML & parameterizationMétéo-France (incl. CNRM)
Combines GANs with extreme value theory (evtGAN) to model spatial dependence of temperature and precipitation extremes over western Europe, outperforming classical GANs using a 2000-year simulation.
Extremes & hazardsKNMI (Royal Netherlands Meteorological Institute)
Climate-invariant ML embeds physical knowledge into parameterization emulators so they generalize across climate regimes, maintaining offline accuracy in three atmospheric models.
Explainability, uncertainty & physicsColumbia University (LEAP, Lamont-Doherty)Google ResearchMassachusetts Institute of Technology (MIT)NVIDIAUniversity of California, Los Angeles (UCLA)
Neural network emulates the M7 aerosol microphysics scheme in ECHAM-HAM, reaching R² of 89% and a 120x GPU speed-up for climate model runs.
Adversarial sub-pixel CNN super-resolves 2.5 km regional climate simulations of Montreal to 250 m, improving urban temperature errors and heat-stress indicators.
Downscaling & super-resolutionEnvironment and Climate Change Canada (ECCC)
Neural network combination of 15 GCMs for Tehran temperature and precipitation outperforms the best single model and multi-model mean in present-day simulations and future projections.
Multi-fidelity Gaussian processes fuse a coarse GCM and a regional climate model to predict Peruvian temperatures, evaluating only 6% of the high-fidelity region.
Physically regularized ML emulators of aerosol activation outperform many existing parameterizations, enabling hybrid models for Earth system models.
A single shallow neural network emulates the full atmospheric physics suite in NCEP GFS, remaining stable for medium-range forecasts and a year-long AMIP run with realistic output and speedup.
Surveys physics-informed ML approaches with ten case studies for emulating, downscaling and forecasting weather and climate, reporting greater consistency and data efficiency.
Explainability, uncertainty & physicsCalifornia Institute of Technology (Caltech)Lawrence Berkeley National Laboratory / NERSCNVIDIAStanford UniversityUniversity of California, Berkeley
Reservoir computing predicts North Atlantic SST in a CESM control run, beating linear inverse models with limited data and suggesting emulator potential.
Compares random forest and neural network moist-physics parameterizations trained on a 3 km global simulation; NN wins offline but crashes within 7 days when coupled.
Physically constrained neural network subgrid parameterization trained on high-resolution simulations stays stable in coupled runs and tolerates reduced numerical precision.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)
A conditional sampling method uses short partial observations to generate training data that reduces model error and improves machine learning forecasts of turbulent and climate-like systems.
ResCu, a residual CNN trained on SPCAM output, parameterizes moist convection and clouds with moist-static-energy conservation, outperforming conventional schemes in single-column tests.
Hybrid physics-ML & parameterizationTsinghua University (NowcastNet)
A VAE trained on about 6 million high-resolution vertical velocity samples reconstructs convection and clusters organization regimes, supporting stochastic convection parameterizations.
A Gaussian process emulator with an augmented bias term corrects Amazon climate biases in a land-surface model and identifies plausible parameter values despite structural error.
Random forests, linear regression and CNN-based networks emulate CAM6 physical forcings offline; tuning is needed to capture detailed patterns, with online coupling left for future work.
A random forest subgrid parameterization learned from 3D high-resolution output yields stable coarse-resolution climate simulations that reproduce precipitation extremes.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)
Iterating an ensemble Kalman filter with a neural network emulates the chaotic Lorenz-96 model from sparse noisy observations, giving forecast skill up to two Lyapunov times.
A neural network trained prognostically on ERA5 emulates northeast Pacific boundary-layer tendencies in single-column runs over 7-day periods, outperforming the underlying forecast model in tracking reanalysis.
Hybrid physics-ML & parameterizationUniversity of Washington
Deep neural networks emulate SP-E3SM shortwave and longwave radiation with 90–95% accuracy at 8–10 times lower cost than the original scheme.
Hybrid physics-ML & parameterizationOak Ridge National Laboratory
ML learns timestep-by-timestep error corrections to a physical model; tested on Lorenz-96, it gives stable runs, better initialised skill and improved long-term statistics.
Applies artificial neural networks for statistical downscaling of global climate model output to regional scales.
Linear ML regression predicts 3D ozone from temperature alone, improving stratospheric ozone variability over fixed climatology for preindustrial and 4xCO2 climate sensitivity runs.
Hybrid physics-ML & parameterizationUniversity of Cambridge (Aardvark)Karlsruhe Institute of Technology (KIT)
A neural network trained on cloud-resolving output replaces subgrid convection in a global climate model, giving stable multi-year runs that capture precipitation extremes and approximately conserve energy.
Hybrid physics-ML & parameterizationColumbia University (LEAP, Lamont-Doherty)
Eos news piece proposing that climate models learn and adapt on the fly with machine learning to reduce prediction uncertainty.
CBRAIN, a neural network trained on superparameterized aquaplanet data, replaces convection parameterization and reproduces its features with reduced stochasticity.
Hybrid physics-ML & parameterizationColumbia University (LEAP, Lamont-Doherty)
Neural network emulates CAM longwave radiation 50-80 times faster with near-identical 10-year climate simulations, an early ML physics emulation.
Hybrid physics-ML & parameterizationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
