ML for climate science questions: causal discovery, teleconnections, pattern detection, detection and attribution, climate model evaluation.
Search and filter this topic Plain list on GitHub
Machine learning that accounts for model uncertainty detects a physically interpretable anthropogenic signal in extreme precipitation across all global observational datasets.
Machine-learning-based detection indicates the climate change signal is identifiable from a single day of global weather; abstract unavailable, so classification is tentative.
Nature review on deep learning for data-driven Earth system science, covering weather, climate and process understanding; abstract absent, judged from title and venue.
Lawrence Berkeley National Laboratory / NERSCMax Planck Institute for BiogeochemistryMax Planck Institute for Meteorology
A CNN generates dynamically consistent counterfactual extreme heat events, estimating the 2023 southcentral North America heat was 1.18–1.42°C warmer due to warming.
Uses VAE latent spaces to visualize droplet size distributions from bin-microphysics cloud simulations, showing similar spectrum evolution across aerosol levels at different paces.
GeoCryoAI, a CNN-RNN hybrid, models Alaskan permafrost active layer thickness and CO2/CH4 fluxes, with a 1.327 cm ALT RMSE over 1969–2022.
Feature-importance methods for echo state networks identify climate pathways, applied to reanalysis data around the 1991 Pinatubo eruction as a proxy for stratospheric aerosol injection.
CNN-based backward optimization attributes 1900-2014 global temperature change to forcings without assuming additivity, with uncertainty estimates compared to regularized optimal fingerprinting.
Transfer learning merges Earth system model simulations with historical observations to cut the spread of 21st-century global temperature projections by more than 50%.
Machine learning on GCM and ERA5 environmental variables identifies humidity convergence and CAPE as key predictors of tropical cyclone intensification rates.
Self-organizing map identifies six subseasonal Indian monsoon stages and two wind-based indices for subseasonal monitoring and forecasting.
An unsupervised neural network manifold maps the continuum of mesoscale shallow cloud organisation, showing organisation alters albedo by about ±1.4% beyond cloud-fraction changes.
Unsupervised learning finds three propagation modes of Indo-Pacific rainfall extremes linked to Pacific SST, with early warning skill up to four weeks ahead.
Technical University of Munich (TUM)Potsdam Institute for Climate Impact Research (PIK)
Unsupervised Markov-state analysis of 500 hPa heights identifies Northern Hemisphere winter blocking and teleconnection-like regimes with stronger dynamical grounding than EOFs.
Clusters global cities into five surface urban heat island grades and uses an interpretable TabNet model to weight the drivers of each grade.
A simple ML model links PW, CIN and wind shear to tropical rain-regime transitions, with a PW above 60 mm threshold, proposed as a convection trigger.
Machine learning attributes the 2022 China compound drought-heatwave forest productivity decline mainly to soil moisture rather than VPD, identifying a tipping point.
A CNN trained on daily precipitation from climate models detects emerging anthropogenic signals in observed rainfall since the mid-2010s.
Nanjing University of Information Science & Technology (NUIST)University of Washington
Linear trends and XGBoost analyse three decades of POWER temperature data in four Ghanaian cities, finding local warming, with Accra warming most.
Applies machine learning to analyse complex climate change dynamics in south-west Australia; no abstract was available, so the classification is based on the title only.
Explainable CNN on 1200-year simulations attributes North China summer precipitation variability to PDO and AMO sea surface temperature patterns, with 96-98% accuracy.
Fudan University (FuXi)Institute of Atmospheric Physics, Chinese Academy of Sciences
Self-organizing maps link large-scale patterns to extratropical cyclones and show CMIP6 models produce too few cyclones and too zonal patterns over northeastern US.
Multi-proxy drought atlas reconstructs southwest Pacific drought back to 1640 and uses Isolation Forest to flag extreme El Niño-linked droughts.
Applies climate network analysis with manifold-learning embeddings to show the Amazon losing short-range and gaining long-range connectivity, with faster perturbation spread in recent decades.
Review of causal inference methods for time series, showing how causal discovery and attribution tools apply to Earth system and climate data.
CNN recognizes El Nino and La Nina phases with 86.4% validation accuracy, and weight analysis identifies the few active kernels driving the classification.
Data-driven pattern reduction plus fluctuation-response causal inference, with a new null model, maps interventional causal links between regions in global monthly sea surface temperature fields.
Neural network sensitivity analysis of observations separates aerosol primary convective invigoration (72% of cloud-top cooling) from meteorological effects in deep convection.
Neural networks classify US West Coast atmospheric river landfalls back to 1916, and tree-ring regression extends annual AR counts to 1400 CE.
Columbia University (LEAP, Lamont-Doherty)University of California, Los Angeles (UCLA)
Review of ML for climate variability modes, extreme weather precursors, observation-model integration, downscaling and bias correction, aimed at domain scientists.
NSF NCAR / UCARUniversity of California, BerkeleyLawrence Berkeley National Laboratory / NERSCOak Ridge National Laboratory
CNNs with partial dependence analysis reveal nonlinear soil moisture–temperature coupling across mid-latitude regions using reanalysis data.
DeepLCZChange links airborne LiDAR urban structure to Landsat 8 surface temperature in New York, showing a cooling effect of urban forests.
Explainable neural networks trained on Earth system model output quantify climate distinguishability after stratospheric aerosol injection, showing slower change under SAI.
A neural network with Shapley values on ERA5 predictors explains 70% of European heatwave variability, with upper-level geopotential the largest contributor at 27%.
Review of AI, statistical and GIS methods for climate change impact on the Gujarat coastline, with no quantitative results of its own.
Deep spatiotemporal clustering with a CNN-RNN autoencoder groups multivariate climate data, outperforming conventional and deep clustering baselines.
Hybrid ML and physics models attribute projected photosynthetically active radiation changes in China and India to climate change versus urban expansion, using CMIP6 models.
Uses permutation tests and machine learning attribution to show summer heat in western Sydney rose sharply after 1992, with 35% more days above the 95th percentile.
Gaussian mixture modelling classifies UKESM1 CMIP6 ozone profiles into six coherent global regimes, with future projections showing ozone hole recovery and lower tropospheric ozone.
Random forest, XGBoost and ANN with Shapley analysis identify latent heat flux, humidity and soil moisture as key drivers of US extreme precipitation, plus return-period emulators.
TCINet, a recurrent causal inference model with probabilistic balancing, quantifies atmospheric drivers of Arctic sea-ice melt under continuous treatment.
Adversarial generators find Earth system model configurations that trigger AMOC collapse, with a discriminator predicting collapse accurately without the oracle.
Hybrid neural and neuro-symbolic approach uses a surrogate ocean model to predict Atlantic Meridional Overturning Circulation collapse with high accuracy, adding explainability.
Multi-model machine learning estimates that fully developed Fennoscandian forests warm annual mean land surface temperature by 0.26 °C relative to present forest states.
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.
A neural network classifies satellite cloud morphology, showing that closed-cell cloud changes contribute about 0.06 W/m²/K to global shortwave cloud feedback under CO2 quadrupling.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCARUniversity of Washington
Explainable neural network on 100 MPI Grand Ensemble members identifies spring tripolar North Atlantic SST patterns as precursors of European summer heatwaves.
Max Planck Institute for MeteorologyColorado State University (incl. CIRA)
Logistic regression and neural networks show stratospheric aerosol injection signals in regional temperature emerge within a decade, while precipitation signals are harder to detect.
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.
Broad review of data science and machine learning for climate detection, attribution, impacts and mitigation, with limited technical novelty.
Machine learning analysis of changes in extreme precipitation over southern China; abstract unavailable, so judgement is based on title only.
Machine learning classifiers identify individual days of CMIP6 and ICON-Sapphire output versus observations, showing model differences appear at daily timescales.
Hybrid error-index and machine-learning method detects Arabian Sea monsoon onset vortices in ERA5, recovering all 27 observed vortices with 5 false positives.
Surrogate-based statistical framework constrains UKESM1 aerosol parameters against MODIS observations on a cloud platform, using 3-hourly outputs for tighter constraints.
GeoCryoAI, a hybrid CNN-LSTM ensemble, models permafrost degradation and soil carbon feedbacks in Alaska, aiming to fill a gap in Earth system models.
Pseudoproxy tests compare Bi-LSTM with linear PCR and CCA for millennial NH and NAE summer temperature reconstruction; Bi-LSTM is less skilful but captures extreme cooling.
CNNs predict daily maximum temperature from soil moisture and geopotential, and partial dependence reveals nonlinear soil moisture-temperature coupling across mid-latitude regions.
Introduces a distributionally robust supervised fingerprint method using anchor regression for climate change detection and attribution that stays robust under interventions on other drivers.
A neural network trained on CMIP6 global temperature and precipitation means predicts the year, and Alopex-derived patterns reveal how climate change signals evolve.
ResGraphNet, a GraphSAGE model with residual links, predicts HadCRUT5 global monthly mean temperature more accurately than 11 baselines across seven datasets.
Causal machine learning on satellite observations estimates aerosol-cloud interaction with uncertainty bounds, finding only one of three climate models reproduces the cooling trend.
Deep-learning weather typing analyses interdecadal and interannual variability of autumn extreme rainfall in Taiwan; abstract unavailable, scope inferred from title.
A machine learning framework trained on climate models attributes tropical tropospheric warming: forcing adds 0.25 K/decade, internal variability offsets 0.07, and an aerosol bias adds 0.04.
NSF NCAR / UCARUniversity of California, Los Angeles (UCLA)Colorado State University (incl. CIRA)
Doubly sparse spatiotemporal Gaussian process builds a probabilistic European climate-field reconstruction from the Last Glacial Maximum to mid-Holocene, combining simulations with pollen proxies.
Unsupervised clustering of storm-scale climate model output shows extreme precipitation changes driven mainly by shifts in storm regimes rather than precipitation efficiency.
Unsupervised ML on tree-ring PDSI drought atlases finds outlier drought patterns in 80% of 21st-century years, showing unusual hydroclimate states.
Deep learning detects weather fronts in high-resolution CESM simulations over CONUS, showing frontal extreme precipitation becomes less likely under climate change despite overall increases.
Applies standard ML models to historical global land temperature and CO2 records since 1750 to examine warming trends; methodologically basic.
Random forest and boosting models estimate pre-industrial cloudiness, suggesting aerosols increased global cloud fraction by 1.27% with about -0.42 W/m2 cooling.
Regression-based machine learning ranks 45 climate attributes, linking declining April–May rainfall in the northern Murray-Darling Basin to global warming, ENSO, SAM and IOD.
Explains a 42-year lengthening of European summer as early onset driven by Greenland topography and Rossby wave responses, building on CNN-detected trends.
Shows transfer-entropy causal graphs can be spurious in climate and ecohydrology data and proposes a subsample ensemble approach for more robust causality analysis.
Uses a VAE with knockoff interventions to estimate causal links under hidden confounding, applied to aerosol-cloud-climate time series with better accuracy than a deconfounding benchmark.
A neural network with Deep Taylor Decomposition attributes simulated soil moisture changes to forcing agents and compares the signals with SMAP observations.
Self-organizing maps identify Eastern US atmospheric river types and show rising trends for Nor'easter and Gulf of Mexico ARs.
Supervised classifier trained on 18 observation and reanalysis datasets labels Central and Eastern Pacific El Niño types, then shows the observed CP increase lies within internal variability in large ensembles.
MATLAB neural networks map daily ECMWF-derived rainfall across Nigerian locations, showing rainfall decreases with distance from 4°N; the ML contribution is minimal.
WISRnet combines a CNN autoencoder with k-means to identify Dunkelflaute periods from reanalysis-type wind and solar patterns, finding most events fall in 5 of 25 clusters.
Applies a deep neural network to estimate future rainfall trends and their uncertainty under climate change; abstract unavailable, so classification is based on title only.
Uncertainty-aware regression network in CESM2 finds low-uncertainty initial states for decadal sea surface temperature predictability linked to Atlantic and Pacific variability.
CNN predicts ENSO nine months ahead with about 0.82 correlation; contribution maps identify three precursors with asymmetric roles in El Nino and La Nina.
Links high-mountain in-situ observations to a subset of GCMs via machine learning weather types, showing mid-troposphere drying in the past and moistening in the future for East Africa.
Machine learning attributes aerosol climate forcing largely to increased cloud cover. Abstract unavailable; judged from the title only.
Nonlinear ML regression shows urbanisation and land-use change most strongly explain 1901–2020 Indian monsoon precipitation extreme trends.
Trains a shallow neural network to identify which climate model produced Arctic temperature maps, then applies it to reanalysis, using attribution to reveal model-specific regional patterns.
Explainable neural networks using layer-wise relevance propagation combine multivariate climate fields to detect forced climate change earlier than single fields.
Advanced machine learning and regression show the QBO matters more than ENSO for stratospheric water vapor; CMIP6 and CCMI models underrepresent the QBO link.
Hierarchical clustering separates northward-moving typhoon tracks affecting Northeast China and links each type to preceding Pacific SST anomalies, with numerical experiments confirming the mechanism.
A network trained only on 2D Ising phase transitions detects early warning signals in real and simulated climate tipping systems, often beating standard statistical indicators.
Progressive cascade networks start from linear models and add complexity only as data supports, then map global temperature and precipitation to years to reveal climate change patterns.
Neural network with explainability predicts decadal global warming slowdowns from ocean heat content in CESM2 large ensemble, linking them to Interdecadal Pacific Oscillation-like patterns and testing on observations.
Scalable sensitivity bounds with uncertainty-aware deep models for continuous-treatment causal effects, applied to how human emissions affect cloud properties in 15 years of satellite data.
Supervised classification of MODIS images builds the greenCuDb dataset, revealing that continental green cumulus form under similar large-scale conditions worldwide.
Analyzes seasonal temperature, precipitation and river flow features across Köppen-Geiger climates using explainable ML, finding autocorrelation and seasonality most informative.
Gaussian process regression fits an empirical stochastic model to stable boundary layer inversions at tower sites, revealing bifurcation-like dynamics similar to idealized models.
ML and causal inference reduce CMIP6 land model evaporative fraction biases at 64 FLUXNET sites, raising correlation from 0.47 to 0.80.
Explainable ML with Shapley values on satellite and reanalysis data shows liquid water path rises nonlinearly with droplet number, with sensitivity dependent on precipitation and SST.
A deep autoencoder detects anomalous heavy precipitation events in Japan and attributes their increase since 1977 partly to anthropogenic forcing, with more expected this century.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCARPrinceton UniversityNanjing University of Information Science & Technology (NUIST)
Applies time-evolving self-organizing maps to a CESM pre-industrial run, tracing a 10-year transition from positive NAO to AMV and weak negative NAO in North Atlantic SST.
Neural forward-intensity Poisson process models nonlinear excitation links between extreme climate events and political violence across countries over 20 years.
Unsupervised clustering and cyclone tracking identify four weather patterns behind middle tropospheric cyclones that drive extreme rainfall over western India.
A deep network with label deformation loss detects and classifies synoptic fronts in ERA5 over Europe and North America, with CSI above 62.9%.
Applies the CNN warm conveyor belt identifier to reanalyses and ECMWF ensembles, showing CNN and trajectory methods agree on North Atlantic forecast biases.
CNNs predict warm conveyor belt inflow, ascent and outflow footprints from gridded fields, reproducing trajectory-based climatology at much lower computational cost.
Applies linear, exponential and polynomial regression to 17 Indian climate parameters, projecting values to 2025–2035; a basic trend-extrapolation study.
Paleoclimate data assimilation with ANN proxy system models produces a millennial, spatially continuous Northern Hemisphere temperature reconstruction from tree rings.
Saliency-map explainable deep learning on global SST reveals ENSO-related information beyond indices and improves river flow prediction with uncertainty estimates.
Software applies ridge, PLS and neural-network regression, trained across leave-one-out CMIP model ensembles, to separate forced from unforced tropospheric warming using surface temperature maps.
A CNN emulator plus variational inversion attributes 1900–2014 warming to greenhouse gases (0.8–1.9 °C in 2014), aerosols and natural forcings.
Compares regression with NARX-based neural models to link climate variables to tree-ring growth in NE Korea and reconstruct the growth index for 1999–2019.
A multi-channel VAE jointly embeds vertical wind, temperature and water vapour from storm-resolving simulations, yielding unsupervised weather-pattern clusters with more interpretable latent structure.
Information-bottleneck method learns causal macrovariables from high-dimensional data and recovers both known El Niño variants in real climate data.
An artificial neural network links MJO indices to North Indian Ocean cyclone wind intensity over 1974–2019, with scatter index falling from 0.45 for depressions to 0.03 for super cyclones.
A smoothed CNN classifies six anticyclonic circulation types linked to European drought and heat, to help identify their drivers in climate simulations.
Non-linear causal machine learning reveals how aerosol effects on cloud droplet radius vary across meteorological regimes in the Southeast Pacific.
Neural network ensembles compare marine low cloud fraction relationships in E3SM at two vertical resolutions against ERA5-CERES, showing shifts in meteorological drivers dominate differences.
Manifold learning characterizes the tropical Pacific climate attractor, revealing model biases that diverge between historical and warming-scenario climate model simulations.
XGBoost predicts Swiss foehn from synoptic fields in reanalysis and CESM simulations, projecting more spring south-foehn in Altdorf and summer north-foehn in Lugano under warming.
A variational autoencoder with a linear latent penalty separates circulation-driven daily precipitation at 12 km over Europe, enabling dynamical adjustment of high-resolution fields.
A CNN identifies 91% of Midwest extreme precipitation events from circulation patterns and finds increasing frequency of the associated patterns over two decades.
TSInet deep learning reconstructs total solar irradiance over more than 9,000 years, matching physics-based models where data overlap.
A self-organizing-map blocking index detects European blocking more accurately than existing indices, with higher precision and recall against an expert-labelled ground truth.
Machine learning that accounts for model uncertainty detects a physically interpretable anthropogenic signal in extreme precipitation across all global observational datasets.
K-means clustering and C4.5 decision trees link Poyang Lake basin precipitation clusters to preceding spring large-scale circulation indices.
Nanjing University of Information Science & Technology (NUIST)
Layer-wise relevance propagation attributes regional temperature signals to aerosol versus greenhouse forcing in single-forcing large ensembles.
Explainable neural networks with layerwise relevance propagation identify North Atlantic and North Pacific SST regions that drive decadal predictability of continental temperature in an Earth-system model.
Information-theoretic causal analysis with neural networks quantifies direct solar influences on hydro-climatic variables at specific frequencies and lags.
A probabilistic graphical model evaluates how well CMIP6 models reproduce daily spatial monsoon rainfall patterns and clusters over India.
Applies machine learning feature analysis to E3SM climate model output to explain its disparity with observed climate change; abstract unavailable.
Uses machine learning to relate urban climate to built environment features; abstract unavailable, so scope is inferred from the title.
Segmentation neural network identifies dry intrusion origins from ERA-Interim geopotential height alone, enabling detection in climate datasets lacking wind fields.
6-km regional simulations show local SST warming shifts US West Coast snowpack, with neural networks attributing the response to precipitation amount and temperature.
General overview of AI and big data methods for analysing climate and environmental datasets; it is a broad survey with no new method or quantitative result.
Machine learning on 2001-2019 global NPP, heterotrophic respiration and climate data finds soil carbon losses likely as precipitation drives warming-related vulnerability.
A machine learning fit of a six-parameter Langevin model reproduces AMOC collapse trajectories from intermediate-complexity climate models under freshwater forcing.
A machine learning föhn detection algorithm trained on automatic weather stations maps 1979-2018 föhn-driven Antarctic Peninsula melt, finding 3.1% of total melt.
Random forest reconstructs global biome vegetation from pollen for past climates, giving robust preindustrial and mid-Holocene patterns but failing at the Last Glacial Maximum.
Thesis develops learned pattern recognition for atmospheric rivers (topological data analysis plus ML) and blocking (deep CNNs) in climate simulations and reanalysis.
Spectral analysis of proxy temperature records trains a neural network on pre-1880 oscillations to forecast 20th-century warming, estimating natural versus anthropogenic contributions.
A nonlinear Gaussian-process Granger causality method attributes global vegetation greenness to precipitation and soil moisture over 30 years of gridded satellite data.
Random forests with SMOTE oversampling identify geopotential height and sensible heat flux as key drivers of Indian heatwave days, reaching an F1 score of 0.85.
Random forest models using FLUXNET, reanalysis and LAI explain 53–90% of global latent and sensible heat flux variance, linking contrasting flux trends mainly to rising CO2.
Support vector regression and CNN reconstruct the RMM MJO index back to 1905, improving on prior regression by up to 4% and revealing a 22-27% rise in MJO intensity.
Combines an emergent constraint with machine learning to narrow CMIP5 end-of-century GPP to 171 ± 12 Gt C/yr and reduce gridded projection uncertainty.
Columbia University (LEAP, Lamont-Doherty)Max Planck Institute for Biogeochemistry
Combines quasipotential landscape theory and manifold learning to locate multistable climate states in an intermediate-complexity model, finding a third stable intermediate state.
Uses probabilistic classifiers, including tree-based models and neural networks, for a goodness-of-fit test checking CESM climate simulation ensemble consistency after code changes.
Applies machine learning clustering to delineate homogeneous climate zones in Pakistan's Indus sub-basins for selecting GCMs and projecting seasonal precipitation trends to 2099.
DISTANA with active tuning infers hidden causal factors, recovering Earth's land-sea mask from temperature dynamics alone while improving temperature prediction.
Reservoir computing and LSTM learn coupling relations among toy climate variables and reconstruct one series from another; convergent cross mapping guides input selection.
A quasi-supervised Mask R-CNN reidentifies Northern Hemisphere extratropical cyclones, finding 8.29% more cyclones than the traditional automatic method.
Nanjing University of Information Science & Technology (NUIST)
Machine learning identifies three-dimensional drought events in tree-ring reconstructions and CMIP simulations, suggesting models underestimate the range of future hydroclimate states.
Review of big data and machine learning for climate change modeling using satellite, sensor and climate model data, covering challenges and outlook without new methods.
Generic overview of AI-driven environmental simulation and climate pattern prediction; accuracy claims lack methodological detail, so significance is minimal.
Unsupervised clustering of Argo ocean temperature profiles detects El Niño events without geographic inputs, offering an alternative to SST-based ENSO indices.
Unsupervised clustering of 850-hPa circulation over the NE Pacific links synoptic regimes to marine boundary-layer cloud droplet properties and cloud radiative forcing.
Uses linear regression and support vector regression to attribute and predict Canberra's seasonal precipitation and temperature from ENSO, SAM and IOD drivers, with SVR reaching r=0.737 for precipitation.
Kernel k-means clustering of German station series yields robust regions, enabling a nonstationary model of summer temperature extremes above 35 °C.
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.
A neural network trained to predict the year from climate model temperature and precipitation maps, with relevance visualization, reveals indicator patterns of forced climate change.
Random-forest-based Granger causal framework isolates soil moisture to precipitation feedback over the US and uses it to benchmark Earth system models for land-atmosphere coupling.
Artificial neural networks classify 500 hPa circulation patterns, showing the 2019 Greenland melt season was about five standard deviations anomalous in atmospheric persistence.
Applies linear, quadratic and Gaussian process regression to monthly greenhouse gas data to extrapolate temperature anomalies, with results said to resemble IPCC projections.
Causal networks from sea level pressure fingerprints objectively evaluate climate models and better constrain precipitation projections than storm-track metrics.
Winning NeurIPS Causality 4 Climate algorithms combine linear methods for causal structure learning from Earth science time series, with open benchmark implementations.
Machine-learning-based detection indicates the climate change signal is identifiable from a single day of global weather; abstract unavailable, so classification is tentative.
Uses neural networks to identify heat waves in CMIP5 RCP8.5 projections for southern Iran, finding mostly decreasing frequency for 2040–2074; ML role is minor.
Partial directed coherence and a kernel nonlinear variant reveal linear and nonlinear causal links among ENSO, PDO, AMO and other teleconnection indices from 1950 to 2018.
Proposes Bayesian networks as a sparse, non-redundant alternative to correlation networks for building data-driven climate networks, illustrated on a global climate dataset.
A machine-learning föhn detector trained on Antarctic weather stations finds föhn-driven melt strongest near Larsen C, with no significant increase over 1979-2018.
Deep convolutional networks detect two anthropogenic aerosol-driven perturbations in satellite cloud imagery to assess their prevalence and climatic effects.
A neural network trained to predict year from temperature maps detects forced climate patterns without explicitly separating signal from noise.
A neural network filters a 50-member regional climate ensemble to show Vb cyclones shift from summer to spring under RCP8.5.
Perspective arguing that machine learning should mine existing climate data and simulations for teleconnections and warnings, not only improve Earth system models.
DL-FRONT, a 2-D CNN trained on five surface variables, detects about 90% of manually analyzed fronts over North America for 2008–2015.
Nonlinear causal discovery method estimates interdependency networks from large generic time series datasets.
The Alan Turing InstitutePotsdam Institute for Climate Impact Research (PIK)University of Oxford
Tests boosted regression trees and neural networks for palaeoclimate reconstruction from fossil pollen, with a spatially blocked cross-validation scheme.
Information-flow analysis of FLUXNET eddy-covariance sites with a neural network maps global land-atmosphere coupling, revealing hotspots and aridity dependence.
Encodes multi-variable monthly climate series as RGB images and uses CNNs to classify temperature and precipitation rises and falls over 1901-2016.
CNN with a novel probability-to-object step detects synoptic warm and cold fronts in reanalysis grids, achieving 0.73 probability of detection at 250 km.
Contour-based stable rank metrics from topological persistence produce descriptors that characterize spatial patterns of atmospheric cloud fields.
Climatology of Red River Valley blizzard patterns from 1979–2018 reanalysis, with a self-organizing map grouping events into an 8-class map.
Editorial introducing a special issue on statistics and machine learning for climate informatics, covering climate clustering, rainfall extremes and correlated extreme time series.
Nature review on deep learning for data-driven Earth system science, covering weather, climate and process understanding; abstract absent, judged from title and venue.
Lawrence Berkeley National Laboratory / NERSCMax Planck Institute for BiogeochemistryMax Planck Institute for Meteorology
Hilbert phase analysis with temporal averaging reveals hidden periodicities in surface air temperature, linked to El Niño, and yields features for time-series classification.
Talk overview of large-scale circulation patterns linked to local extremes, using machine learning among methods to assess climate model projections.
U-Net detects stationary fronts around Japan from multi-variable weather fields, reproducing seasonal front shapes on an untrained year except during typhoons.
Gaussian process clustering by covariance kernel groups British Columbia coastal rainfall records, linking clusters to El Niño and La Niña patterns.
Uses an artificial neural network with tree-ring and earlywood vessel features to simulate precipitation time series; abstract unavailable, classification based on title only.
Random forests show that climate seasonality metrics of precipitation and temperature predict 66-87% of global multi-year fire activity variability.
Uses unsupervised clustering to auto-label North American weather regimes in climate model output, then trains CNNs to re-identify them, reaching 90–94% accuracy.
Three classifiers, with SVM best, identify synoptic weather types over Taiwan from reanalysis with 52–83% hit rates.
A CNN stratifies convection-permitting climate runs and finds too few quasi-linear convective systems in the Midwest, with future declines not spatially significant.
Neural-network-based self-organizing map clustering of Indian Ocean SST reveals ten IOD patterns, showing pseudo-dipoles with distinct rainfall teleconnections over 130 years.
Uses time-delay neural networks with Bayesian regularization to estimate regional effects of climate change and land use on biosphere carbon fluxes.
Proposes combining manifold learning and persistent homology to detect and characterise weather and climate patterns in fluid and climate simulation data.
Proposes a Tollgate neural network with time-bound learning for climatic data analysis; the abstract is missing, so scope and results are unclear.
Applies density-based clustering to choose grid-cell model subsets for multi-model means of tropospheric ozone, improving agreement with satellite climatology in chemistry-climate models.
Boosted regression trees identify drivers of global soil organic carbon in observations and CMIP5 Earth system models, revealing key mismatches such as clay content and temperature.
Abstract missing; applies topological data analysis with machine learning to classify atmospheric river patterns in large climate datasets.
Random forest on 800,000 years of ice-core data finds CO2 the largest driver of temperature change, followed by CH4 and N2O.
Takagi-Sugeno fuzzy neural network detects failures ('crashes') in climate model simulations, reaching 97.96% accuracy versus 94.4% for SVM.
Causal effect networks identify Barents and Kara sea ice and Eurasian snow cover as drivers of winter Arctic Oscillation and midlatitude circulation.
Accessible introduction to constraint-based causal discovery for climate scientists, illustrated on boreal winter atmospheric low-frequency modes.
ClimAgent is an LLM agent framework that runs end-to-end climate science analyses, introducing ClimaBench and a 40% gain over baseline LLM solutions.
Nature Communications review of AI for modeling and understanding extreme weather and climate events; abstract absent, judged from title and venue.
Extremes & hazardsInriaMax Planck Institute for BiogeochemistryTechnical University of Munich (TUM)
Combining RCM dynamical downscaling with diffusion refinement lets large CMIP6 ensembles be downscaled cheaply, giving better uncertainty bounds than BCSD and smaller dynamical ensembles.
Downscaling & super-resolutionCalifornia Institute of Technology (Caltech)Google Research
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
ML-derived ventilation parameters in a simple dynamical model show deep-layer vertical wind shear dominates tropical cyclone weakening across basins during 1982–2021.
Genetic-fuzzy GENOFIS model beats ANFIS and SVM for Kigali rainfall prediction using 1991–2023 gauge data, combined with downscaled climate-change scenarios.
A CNN with cosine-distance loss predicts equatorial Pacific SST zonal patterns up to 16 months ahead, outperforming dynamical models, with physically interpreted ENSO precursors.
Sub-seasonal to seasonalNanjing University of Information Science & Technology (NUIST)
Interpretable machine learning projects how climate change under emission scenarios will degrade agricultural land suitability across Central Eurasia, highlighting key climate drivers.
A CNN trained on SEVIRI satellite imagery and DARDAR lidar retrievals provides a geostationary observational constraint on cirrus and mixed-phase ice cloud microphysics.
Physics-informed neural network uses climate projections and landscape features to map projected changes in intense rainfall, hail and strong-wind frequency for climate adaptation.
Review of AI methods for predicting climate extremes such as heatwaves, droughts and heavy rain, covering hybrid AI-climate-model approaches, challenges and future perspectives.
Transfer-learned CNN ensemble emulates seasonal tropical cyclone activity from large-scale fields, with skill on operational forecasts and projected global TC decline under warming.
Uses neural networks to correct GFDL-ESM2G CMIP5 surface air temperature over Andhra Pradesh against IMD observations, with RMSE of 0.97–1.33 °C.
A genetic-algorithm plus geographically weighted regression framework finds explicit terrain-precipitation equations that beat empirical models and support fine-scale precipitation downscaling under future climate.
Machine learning models forecast precipitation from teleconnection indices across different precipitation regimes, an approach for seasonal-scale rainfall prediction.
A stacking ensemble ML model projects heat and cold wave days in China to 2060, showing wind power output falls during heat waves and rises during cold waves.
MMST-ViT transformer predicts US county crop yields from satellite images and weather, capturing short-term weather and long-term climate change effects with three metrics improved.
Self-organizing maps classify synoptic circulation patterns behind Hyderabad urban flood rainfall, identifying 35 atmospheric river events and their moisture sources.
Bayesian deep learning forecasts long-term daily solar irradiance from climate simulation outputs for energy applications.
Applies machine learning with uncertainty quantification to climate precipitation predictions over South America; the abstract is missing, so details are limited.
Combines machine learning and wavelet simulations with reweighting to bias-correct CMIP6 NEX-GDDP precipitation for Cork, projecting a decline to about 116–119 mm/month by 2076–2100.
Fuses random forest, KNN, extra trees and GBDT into a CMIP6 multimodel ensemble for Hanjiang Basin precipitation, reaching correlation 0.88 and Taylor skill 0.764.
Graph convolutional networks with extended generalized Pareto distributions model monthly Australian wildfire spread, including extremes, for hazard assessment.
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.
Recurrent neural operator with conformal prediction forecasts tipping points, including a stratocumulus cloud cover climate tipping point, using physics-constraint deviations.
UNet-RegPre CNN predicts regional monthly precipitation in China with skill comparable to CFSv2 and identifies summer monsoon precursors.
Random forest analysis of 2000–2020 global human-perceived heatwave risk using heat index shows rising high-risk areas and identifies aspect and land use as main drivers.
Multi-input autoencoder index from tropical Pacific SST gives Southern California precipitation predictability up to 4 months, beating Niño-3.4.
Logistic regression matches machine learning for detecting extreme precipitation in Denmark from ERA5 predictors, with ROC AUC 0.86 versus 0.87 for the best ML model.
Adaptive recurrent graph networks infer historic snow accumulation from deep ice-layer thickness in airborne radar data, outperforming prior models in polar ice records.
Mixture density networks model monthly wildfire frequency and area burned across the western US, matching annual area burned with r = 0.85 observed.
Extremes & hazardsColumbia University (LEAP, Lamont-Doherty)University of California, Los Angeles (UCLA)
Neuro-symbolic encoder-decoder translates climate questions into programs to direct deep learning simulations of AMOC tipping, aiming at explainability.
Machine learning on Berkeley Earth temperatures and World Bank GDP finds a significant relationship between average temperature and long-term GDP growth.
An ensemble of 20 segmentation networks detects atmospheric rivers with 1.7–10.1% higher IoU; CMIP6 projections show AR-driven extreme precipitation rising under SSP5-8.5.
Extremes & hazardsNanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
Non-parametric trend tests and machine learning models analyze and forecast meteorological variables in the Thoubal river watershed, India; the abstract is missing.
Neural networks trained on A-Train radar-lidar data add cloud height, layering and rain intensity to tropical cloud systems, enabling convective organization analysis for model evaluation.
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.
Random forests combining natural and anthropogenic factors hindcast seasonal China temperature and precipitation with R² above 0.95, beating natural-only factors (0.86).
A neural-network precipitation parameterization fails at extremes unless informed by a novel sub-grid organization metric, which raises variance R² from about 0.4 to 0.95.
Hybrid physics-ML & parameterizationColumbia University (LEAP, Lamont-Doherty)
A neural network estimates GEV extreme-value parameters for CCSM3 annual maximum temperatures about as accurately as maximum likelihood, with large computational speedups.
An explainable ML wildfire model trained on 7 million satellite observations projects a 55% rise in fire exposure for forest carbon projects by 2080 under SSP2-4.5.
Multimodel ensemble machine learning for regional climate assessment and prediction; no abstract was supplied, so this classification is tentative.
Unsupervised local causal states detect hurricanes, atmospheric rivers and precipitation extremes in climate data, and track vortices in 2D turbulence.
Introduces an information-tracking calibrator that corrects error accumulation in decadal global land-surface temperature prediction, with predictions consistent with known teleconnections.
Transfer-learned deep networks estimate western North Pacific tropical cyclone size from infrared imagery, producing a 1981–2017 DeepTCSize record with R of 0.76–0.85.
A deep learning atmospheric river tracker is applied to a high-resolution large-ensemble climate dataset and checked for consistency against established algorithms over the US West Coast.
Extremes & hazardsNVIDIANSF NCAR / UCARETH ZurichUniversity of OxfordUniversity of California, Berkeley
3D-Geoformer, a self-attention transformer, predicts Niño3.4 SST anomalies 18 months ahead with high correlation skill for ENSO.
Sub-seasonal to seasonalNanjing University of Information Science & Technology (NUIST)
Particle filter with linear inverse model emulators of CMIP-class GCMs produces ensemble millennial temperature reconstructions and skillful one-year-ahead forecasts in a perfect-model test.
CNN ENSO forecasts from 5000 random-initialized members show western Pacific SST gives better Nino3.4 skill than eastern Pacific SST.
Machine learning examines basin-dependent tropical cyclone genesis environments and their projected future changes; the abstract is unavailable, so this is judged from the title.
Unsupervised clustering of ERA5 wind patterns drives a few WRF runs, reproducing a year of wind farm power and wake predictions at under 2% of the compute.
Interpretable attention-based fire model for African and South American burned area reveals lagged climate-wetness controls and projects South American burned-area recovery under high emissions.
Extremes & hazardsLawrence Berkeley National Laboratory / NERSC
Debiased machine learning estimates adaptation to damaging heat in US corn and soy, finding short-run heat losses are largely offset in the long run.
Deep learning reconstructs monthly surface air temperature over the Tibetan Plateau for 1901–2020, filling gaps in the historical station record.
A divided 3D CNN (TS-3DCNN) predicts the 2020–22 multi-year La Niña, comparable to a dynamical intermediate coupled model.
Sub-seasonal to seasonalNanjing University of Information Science & Technology (NUIST)
Random forest regionalizes India into seven lightning climatologies and predicts seasonal lightning density from CAPE and humidity, with R² of 0.81 in monsoon.
Open dataset of climate indices from 1000-year Earth system model control runs, with ML benchmarks predicting Sahel rainfall and ENSO.
PCMCI-based multidata causal discovery selects drivers for Western Pacific tropical cyclone intensity prediction, outperforming noncausal and correlation-based feature selection.
A 3D DBSCAN clustering workflow detects the 2022 Yangtze heatwave and drought from ERA5 and tests how clustering parameters change event tracking.
Extremes & hazardsFudan University (FuXi)China Meteorological Administration (CMA)
Ridge, lasso, random forest and neural models forecast Horn of Africa long-rains precipitation; preselecting predictors on full data inflates skill, and honest tests fall below dynamical hindcasts.
A knowledge-guided LSTM encoding diel hysteresis and Birch effect outperforms Q10 models and CLM in dryland soil respiration (correlation 0.7 vs under 0.4).
Causal discovery identifies recurrent European heatwave tracks and their seasonality in a single-model large ensemble, enabling a new model evaluation approach for extremes.
Feed-forward neural networks show heat stress and fertilizer drive maize and millet yields in the Sahel, explaining 59-79% of yield variance.
A hybrid statistical deep learning model links VPD and temperature to Mediterranean burnt area, projecting a 17.1% rise in expected August wildfire frequency under observed warming.
Random forest forecasts and non-parametric trend tests for rainfall and temperature at one Indian tiger reserve; no abstract provided, so scope is limited to local station climate.
Gradient boosted trees link California rangeland productivity to climate variables and project regional changes in gross primary production under downscaled future climate.
Machine learning on climate and deforestation data attributes Amazon fire counts to deforestation, estimating a 42% rise in 2020 and 32% fewer fires under the Brazilian target.
Rotation-invariant autoencoder clusters 22 years of MODIS imagery into 42 AI-generated cloud classes (AICCA), revealing cloud-type trends linked to stratocumulus decline.
Applies layer-wise relevance propagation to echo state networks, showing they detect ENSO from sea surface temperature anomalies and explaining the learned features.
Artificial neural networks and isolation forests on Kenyan station data predict drought and flood occurrences with 78-90% accuracy, supporting early warning.
Machine and deep learning back-forecasting models reconstruct multidecadal US electricity demand records to separate temperature-driven natural variability from climate change effects.
Knowledge-based deep learning trains on simulated and real data plus physics losses to forecast extreme events in chaotic systems, including El Niño SST and precipitation.
Convection-permitting simulations with ML damage models project 128% higher typhoon losses for Southeast China under RCP8.5 warming, with 50-year losses up 20–170%.
Applications & impactsNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)Princeton University
A CNN trained on CMIP5/6 models forecasts eastern Indian Ocean Dipole SST anomalies up to 6 months ahead, matching the ACCESS-S2 dynamical model.
Sub-seasonal to seasonalNanjing University of Information Science & Technology (NUIST)Bureau of Meteorology (Australia)
Shows XAI attributions for a climate-signal network change substantially with the chosen baseline, using large-ensemble SSP3-7.0 simulations, so baseline choice must match the science question.
Explainability, uncertainty & physicsColorado State University (incl. CIRA)
Applies regression, clustering and CNN analysis to 28,713 Australian tropical cyclone records, identifying drivers of low central pressure and anomaly root causes.
Bayesian neural network weights CMIP6 models per location and season, improving monthly US precipitation predictions over simple ensemble baselines with interpretable weights.
Neural network reconstructs a monthly 0.05° Antarctic air temperature product from MODIS and 119 stations, with MAE 2.23 °C and better agreement with in situ data than ERA5.
CNN trained on long climate model runs forecasts France 14-day heatwaves up to 15 days ahead, showing skill from 500 hPa heights and soil moisture.
Assesses machine learning models for seasonal precipitation simulation in Iran using teleconnection indices as predictors; no abstract available.
Uses tree-ring chronologies and machine learning to reconstruct precipitation and temperature back to 1886, then ranks CORDEX models and bias-corrects them in the Tien Shan.
ANN, random forest and SVM multimodel ensembles improve CMIP6 precipitation and temperature projections for an Indian river basin over simple averaging.
Uses regression, decision trees and deep networks to merge 16 CMIP6 models into a new ensemble; it outperforms single models (R=0.81 precipitation, 0.99 temperature) and projects 1.5 °C by 2024 under SSP5-8.5.
ANN statistical downscaling of five GCMs projects rising non-monsoon and falling monsoon rainfall for Pune, India.
Neural networks quantify how warming alters midlatitude subseasonal skill from tropical precipitation in CESM2 large ensemble, linked to seasonal variability changes.
Sub-seasonal to seasonalColorado State University (incl. CIRA)
Deep learning on Sentinel-1 satellite imagery yields the first multi-year global ship-track climatology, showing 2020 fuel rules cut ship-track frequency to decades-low levels.
LightGBM uses preceding-winter sea surface temperature, soil moisture and snow to simulate summer eastern European heatwave frequency, with SST contributing most.
Review of machine learning for analysing, predicting and attributing atmospheric extreme events, covering methods, examples and open perspectives.
Neural networks trained on SMAP soil moisture project Southeast Asian peat drying under RCP8.5, finding decreases about half as strong as recent El Niño droughts.
Applications & impactsMassachusetts Institute of Technology (MIT)NASAStanford University
First deep-learning semantic segmentation of scanned 1940-2000 US Navy Antarctic Peninsula photos, reaching 74% accuracy on eight test images despite analog and scanning artifacts.
Slide deck on machine learning classification of cloud types in satellite observations and climate models; no abstract or quantitative results provided.
Observations & remote sensingMET Norway (Norwegian Meteorological Institute)
Wavelet analysis and Bayesian ML project emergence windows for Atlantic hurricane categories 2–5, with category 5 storms expected between 2026 and 2031.
ENSO-driven machine learning predicts the NAO index about three months ahead at 68% accuracy, and a TrajGRU network improves short-term NAO forecasts.
Self-organizing maps link North Atlantic spring SST to summer jet-stream teleconnections, identifying windows of opportunity for seasonal European summer predictions with MPI-ESM hindcasts.
Sub-seasonal to seasonalMax Planck Institute for Meteorology
Statistical trend analysis and SMOreg regression of 1891–2020 Bay of Bengal cyclone records project cyclogenesis to 2021; ML role is minor.
Neural network retrieves 13 years of IASI temperature profiles trained on ERA5, agreeing within 0.5 K of radiosondes between 750 and 7 hPa.
Compares ML ensembling (RF, SVR, ETR, LSTM) of 34 GCMs for daily precipitation and temperature in an Indian basin; LSTM reaches R² of 0.9 for precipitation.
Neural network retrieves stratospheric temperature from single-footprint AIRS sounder data, cutting error versus operational products and resolving gravity waves.
Builds a regression surrogate of outdoor thermal comfort from WRF climate simulations and a portfolio-based allocation method for urban cooling resources in Singapore.
Hybrid neural network combined with principal oscillation pattern analysis for ENSO prediction; the abstract is missing, so the classification relies on the title.
Deformable CNNs with transfer learning forecast North Atlantic-European weather regimes 1 to 15 days ahead, outperforming CNNs beyond 5-6 days lead time.
Benchmarks XAI attribution methods on a CNN that counts atmospheric rivers in climate simulations, exposing fidelity issues such as gradient shattering and sign ambiguity.
Explainability, uncertainty & physicsColorado State University (incl. CIRA)
Uses a MaxEnt model on environmental variables to project a significant decline in future tropical cyclone genesis across CMIP6 SSP scenarios, with a nonlinear link to potential intensity.
Extremes & hazardsTsinghua University (NowcastNet)China Meteorological Administration (CMA)
Proposes MMFSDI, a multi-model drought index built with forecastable component analysis and Gaussian mixtures over 23 CMIP6 models on the Tibetan Plateau, consistent with SPI.
A Bayesian neural network ensemble weights CMIP6 models by location and season, calibrating projections to observations with interpretable weights and uncertainty growing away from constrained periods.
A stochastic vector-autoregressive climate model generates gridded teleconnection benchmarks for causal discovery, plus a new grid-level causal method outperforming existing approaches by orders of magnitude.
Machine learning links weather to household air conditioning use, projecting 8–13% higher US residential demand under 1.5–2.0°C warming.
Random forests with monthly soil moisture and extreme-weather features predict German winter wheat yield anomalies, reaching test R-squared of about 0.70 and explaining 2003, 2014 and 2018 extremes.
Uses a Bayesian neural network to fuse chemistry-climate model output with sparse satellite and sonde data, producing a continuous vertically resolved stratospheric ozone record with uncertainty.
Observations & remote sensingUniversity of Cambridge (Aardvark)The Alan Turing Institute
Applies neural networks and SHAP to predict European drought one month ahead from a 50-member regional climate large ensemble, with about 55–60% correct event classification.
Machine learning links daily climate and weather extremes to Ukrainian winter wheat yields, explaining 54% of variability and identifying heat, frost and drought during grain filling as key losses.
Generalised additive and machine learning models link North Indian Ocean tropical cyclone activity to boreal summer intraseasonal oscillation; no abstract was available, so this rests on the title.
Extremes & hazardsNanjing University of Information Science & Technology (NUIST)
CNN identifies long-duration freezing rain and ice pellet synoptic patterns in CRCM5 output over Montréal, reaching a Matthews correlation of 0.50.
UWILD random forest classifies ice versus liquid particles in Southern Ocean clouds from 2D-S probe images with 95% accuracy, beating two existing methods.
Neural network and random forest predict tropical Pacific rain type from MERRA-2 large-scale state, but neither beats a GLM and all miss extreme rates.
Multiple machine learning models predict Atlantic Multidecadal Variability from North Atlantic SST, salinity and sea level pressure maps, beating persistence in CESM-LE.
Sub-seasonal to seasonalMassachusetts Institute of Technology (MIT)
XGBoost with SHAP predicts and explains NASA GEOS lightning-occurrence errors, linking them to convection and land-surface characteristics.
Ensemble machine learning projects future South Asian drought and its effect on simulated crop yield under climate scenarios.
Ensemble machine learning downscales and projects climate variables across Iran's climate regions, using GCM output for regional climate projections.
ML diagnostic model explains over 80% of monthly burnt-area variance in southeast Australian forests, showing extreme fire weather, linked to SAM and IOD, as the primary driver.
Extremes & hazardsNanjing University of Information Science & Technology (NUIST)
GRU recurrent network statistically downscales WRF/MPI-ESM-LR RCP8.5 output to project Metro Vancouver daily temperature extremes to 2100, with warming near 3°C.
Path-signature machine learning model using climate indices predicts El Niño NINO3.4 SST six months ahead with MSE of 0.596 and identifies North Pacific precursors.
CNN trained on present-day WRF storms classifies strongly rotating thunderstorms in end-of-century climate simulations with skill, and explanations reveal learned physical features.
Random forest combines CMIP6 multi-model ensembles over China, projecting less warming and wetter conditions with better spatial skill than the arithmetic mean.
Post-processing & calibrationNanjing University of Information Science & Technology (NUIST)
Wavelet-ANN statistical downscaling of CanESM2 under RCP8.5 projects 17-25% wetter and 2-3 °C warmer conditions in the Jhelum basin by 2100.
Interpretable machine learning ranks 24 predictive indices for March–September Midwest US S2S precipitation and links them to three moisture and circulation mechanisms.
A random forest trained on ERA5 variables reproduces a station-based heatwave index for South Korea (1979-2020) with MAE 0.365 and CC 0.873, and yields gridded maps.
AdaBoost models using early-stage vortex and environment features classify western North Pacific storms as weak or strong at genesis (about 75% accuracy) and estimate lifetime maximum intensity (R2 0.30).
Compares decision trees, deep learning, GLM, gradient boosting and random forests to forecast Almus Dam hydropower from GCM projections; gradient boosting best with r=0.717.
Assesses whether deep CNN downscaling of GCM projections over Europe yields plausible end-of-century changes, finding smaller departures from raw GCM signals than GLMs.
A neural network on MODIS brightness temperatures detects Atacama Desert fog better than a cloud-height method (Heidke skill 0.56 vs 0.49), enabling a fog frequency climatology.
Observations & remote sensingMassachusetts Institute of Technology (MIT)
Unsupervised topological data analysis is used to identify atmospheric river patterns linked to flooding in Nigeria; no abstract is available.
Random forests show that climate sensitivity and aerosol forcing scaling dominate parametric uncertainty in 2020-2150 sea-level risk, with thermal expansion driving near-term hazards.
Data-driven MSSA matches CMIP5 GCM skill for SST and Indian summer monsoon rainfall decadal hindcasts, improving ISMR skill over GCMs.
Neural network emulates shortwave radiative fluxes trained on ERA5 to diagnose Arctic albedo, water vapour and cloud feedbacks, capturing albedo nonlinearity that linear kernels miss.
Spatio-temporal threshold clustering uses pattern recognition and clustering to define consistent extreme precipitation episodes over the contiguous US from gauge data.
Information-flow causality links sunspot numbers to El Niño Modoki, and a causal deep learning model reproduces events 12 years ahead, according to the authors.
Systematic literature review of machine learning for climate prediction and adaptation planning, with a Morocco focus; offers no new model or result.
A NARX neural network predicts the 12-month drought index SPI-12 in West Papua from the MEI ENSO index, capturing ENSO-linked rainfall patterns.
Supervised classifier on 20 years of MODIS imagery maps Swiss mountain lake ice, finding freeze duration falling about 0.8 days per year.
Random forest over a 99-year climatology identifies drivers of tropical cyclone precipitation in eastern Mexico, with distance to track the strongest predictor for events.
Applies support vector regression to model precipitation-driven home insurance claims in a Canadian city; low-significance sector application.
Rotation-invariant autoencoder (RICC) clusters satellite cloud imagery without predefined classes, with evaluation showing physically meaningful, spatially coherent and orientation-invariant cloud clusters.
Random forest and MARS models link TRMM, GPM and GLDAS hydro-climate drivers to Pakistani wheat yield, identifying six phenological phases with strong responses.
Graph-guided regularized regression on Pacific sea surface temperatures improves seasonal winter precipitation prediction in the southwestern US over benchmark regularizers and teleconnection models.
A CNN trained on current-climate storms skillfully classifies strongly rotating, potentially severe convection in end-of-century climate simulations, generalizing to future conditions.
EfficientNet encoder with UNet decoder classifies satellite shallow-cloud organization patterns, scoring a Dice of about 66% on a Kaggle benchmark.
Machine learning models explain up to 65% of historical crop yield anomalies in 17 European countries, linking 2018 extreme temperature and rainfall to losses.
Applications & impactsPotsdam Institute for Climate Impact Research (PIK)
Philosophical framework arguing data-driven machine learning climate models can provide scientific understanding, illustrated with an atmospheric research case study.
Nonhomogeneous hidden Markov model identifies five ENSO regimes from 1856 to 2019 and gives out-of-sample probabilistic Niño forecasts comparable to existing models.
Sub-seasonal to seasonalColumbia University (LEAP, Lamont-Doherty)
XGBoost and statistical analysis of ERA-Interim identify relative intensity, sea-surface temperature drop and wind shear as dominant drivers of western North Pacific typhoon weakening.
Gaussian process parameter estimation for the Lorenz-96 climate benchmark outperforms deep learning and linear regression under small-data constraints, with uncertainty quantification.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)
Gaussian process regression on water vapor isotopes estimates shallow convective moistening over the SE Pacific better than conventional fields alone.
Combines artificial bee colony optimization with a backpropagation network to forecast basin precipitation from 17 stations over two decades, linked to wavelet periodicity and drought/flood years.
Support vector machines use Pacific and Atlantic oscillation indices to extend precipitation and streamflow forecast lead times in the Colorado River Basin, with KNN for finer precipitation resolution.
Uses principal component analysis and artificial neural networks on 47 years of station data to regionalize rainfall and map extreme precipitation in western Morocco.
A deep neural network predicts Chinese rice yields from climate model data, improving correlation about 38.7% over multiple linear regression.
Machine learning links climate variables from regional and global climate models to European hydropower capacity factors under future scenarios.
Bayesian model averaging with SVR, MLP and random forest downscales eight CMIP5 models' precipitation in the Upper Han basin, with best PCC 0.82.
Spatial quantile regression forests map rainfall erosivity in Greece using RCM projections to 2100, with regional climate models validated against gauges.
Scoping review of 298 studies applying machine learning across wildfire science, including fire weather and climate change, and identifying gaps for deep learning.
Bayesian generative models relate US maize yield to monthly temperature and precipitation, projecting a 2°C warming to reduce yield by 8% and triple its variance.
Climate-network topology features from surface air temperature feed ten classifiers that predict next-season strong or weak ENSO phases with 80% AUC.
Logistic regression with linear baroclinic model simulations shows the past 15 days of MJO activity improve midlatitude circulation predictability beyond 15 days ahead.
Sub-seasonal to seasonalColorado State University (incl. CIRA)
An unsupervised neural network finds hierarchical trade-wind cloud organization regimes in 10,000 GOES-16 images, linking regimes to radiative properties for climate model evaluation.
Applies backwards optimization and layerwise relevance propagation to neural networks, recovering scientifically meaningful patterns in climate variability data.
Explainability, uncertainty & physicsColorado State University (incl. CIRA)
Integrated machine learning is proposed for detecting tropical cyclones in advanced climate models; no abstract is provided, so the method is not verifiable.
Committor functions are shown to quantify intrinsic predictability of ENSO events in the Jin-Timmerman model, and neural-network inference of committors from data is compared to direct estimation.
Physics-based unsupervised representation learning segments extreme weather structures in CAM5.1 water vapor climate model output, a route to identifying events without labels.
Classifies 887 northern Italian extreme precipitation events into three dynamical classes using K-means and random forests, with integrated water vapour transport as key predictor.
K-means clustering of upper-tropospheric troughs shows rapid intensification is more likely with shorter-wavelength troughs and cutoff lows 500–1000 km southwest of the storm.
Feed-forward and wavelet neural networks statistically downscale GCM predictors to Bagmati basin precipitation, projecting 2014–2036 decreases at the study sites.
Crowd-labelled about 50,000 shallow cloud clusters in satellite images to train deep learning that maps four organization patterns and builds global climatologies.
Observations & remote sensingMax Planck Institute for MeteorologyTechnical University of Munich (TUM)
Self-organizing maps applied to reanalysis identify four circulation pattern types behind extreme US floods, including tropical moisture exports and tropical cyclones.
Extremes & hazardsColumbia University (LEAP, Lamont-Doherty)
Uses machine learning on sub-national yield data to show climate extremes explain 18-43% of yield-anomaly variance, with temperature extremes dominating global crop losses.
Applications & impactsETH ZurichPotsdam Institute for Climate Impact Research (PIK)
Archetypal analysis plus a 2D CNN predicts eastern US regional precipitation and discharge extremes from synoptic-scale fields, with the most reliable skill from 300 hPa geopotential height.
Tests backward optimization and layerwise relevance propagation on a neural network for the Madden-Julian Oscillation, recovering known spatial structure and suggesting seasonality shifts.
Self-organizing maps applied to downscaled REGCM4 projections divide Iran into nine extreme-heat regions and project increases in mid-century heat.
Compares LSTM and linear regression for ENSO prediction from 1 to 11-month leads, finding LSTM matches LR on monthly data and gains skill with daily SST at longer leads.
Machine learning merges FLUXNET eddy-covariance towers with remote sensing and meteorological data into a 147-product global land-atmosphere energy flux database with uncertainties.
An ANN downscales GCM predictors to four Jhelum basin stations, projecting 1.4–1.6°C warming and 31–35% precipitation decline by 2100.
Combines GCMs, a weather generator and ANN/ANFIS models to project Qassim rainfall and temperature; ANNs outperformed ANFIS for long-term predictions.
A random forest moist convection parameterization trained on high-resolution output runs stably in a GCM, capturing precipitation extremes and climate change when trained on warm-climate samples.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)
Framework learns ice storm occurrence from atmospheric variables with weak labels, evaluated on reanalysis and GCM data under future climate scenarios.
Hybrid ARIMA and neural network using climate-network topological features extends El Niño prediction to a 12-month lead, matching CFSv2 skill up to six months.
ContrailMod, a CNN applied to Himawari-8 imagery, outperforms a conventional contrail detection algorithm and is used to evaluate potential contrail coverage from ERA-Interim over South China.
Observations & remote sensingNanjing University of Information Science & Technology (NUIST)
First deep CNN application to detect tropical cyclones, atmospheric rivers and weather fronts in climate datasets, reaching 89-99% accuracy.
Early review of neural network applications to meteorological and oceanographic prediction and data analysis, from the pre-deep-learning era.
