Prediction and detection of extremes: tropical cyclones, heatwaves, heavy rain, severe convection, hail, tornadoes, fire weather, atmospheric rivers, cold spells.
Search and filter this topic Plain list on GitHub
Operational AI ensemble WeatherNext Cyclones predicts tropical cyclone track, intensity and size to 15 days, gaining about 1 day of lead time over leading operational models.
Colorado State University (incl. CIRA)Google DeepMindGoogle ResearchMet OfficeNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
FourCastNet trained without Category 3-5 cyclones cannot forecast Category 5 storms, showing AI weather models fail to extrapolate to unseen extremes.
Operational AI ensemble WeatherNext Cyclones predicts tropical cyclone track, intensity and size to 15 days, gaining about 1 day of lead time over leading operational models.
Colorado State University (incl. CIRA)Google DeepMindGoogle ResearchMet OfficeNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Nature Communications review of AI for modeling and understanding extreme weather and climate events; abstract absent, judged from title and venue.
InriaMax Planck Institute for BiogeochemistryTechnical University of Munich (TUM)
Comparison of Pangu-Weather, GraphCast and IFS HRES finds data-driven models mostly beat IFS on global RMSE for 1-10 day forecasts and compete on temperature and wind extremes.
FourCastNet trained without Category 3-5 cyclones cannot forecast Category 5 storms, showing AI weather models fail to extrapolate to unseen extremes.
Compares GraphCast, Pangu-Weather, FourCastNet and HRES on three extreme events, finding ML models weaker on heatwave aggregates but better on a winter storm.
Reviews deep learning medium-range forecasts and shows leading models struggle with extremes, proposing a workflow for extreme-event forecasting models.
Case study of Storm Ciarán finds FourCastNet, Pangu, and GraphCast capture synoptic structure but underestimate peak winds and miss fine warning-relevant features.
ML-derived ventilation parameters in a simple dynamical model show deep-layer vertical wind shear dominates tropical cyclone weakening across basins during 1982–2021.
Graph Transformer exploits an emergent graph among geospatial points to improve six-hourly hurricane trajectory prediction on HURDAT over a standard Transformer baseline.
CNN-BiLSTM-attention model forecasts convective and heavy precipitation from GFS data, reaching threat score 0.336 versus 0.252 for forecasters.
Nanjing University of Information Science & Technology (NUIST)Institute of Atmospheric Physics, Chinese Academy of Sciences
A denoising diffusion model restores fine-scale detail in FuXi 5-day forecasts, improving extreme precipitation and wind over HRES, but not TC intensity.
A deep-learning segmentation model with transfer learning from synthetic fields detects and characterises trapped lee waves hazardous to aviation, faster than spectral methods.
TC-TrajGRU fuses track time series with meteorological fields for 12-hour tropical cyclone track forecasts that are comparable to official forecasts.
AdaBoost with MLP base predicts India's March–June Tmax anomalies 10 days ahead, matching CFS reforecast skill in April–May but not in March or June.
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.
Cascaded diffusion models forecast tropical cyclone tracks and precipitation up to 36 hours from satellite and atmospheric data, running in 30 minutes on a single GPU.
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.
Self-organizing maps classify synoptic circulation patterns behind Hyderabad urban flood rainfall, identifying 35 atmospheric river events and their moisture sources.
Hybrid ARIMA, K-means and autoencoder method generates synthetic North Atlantic tropical cyclone tracks and intensities from HURDAT2 for risk assessment.
A conditional Wasserstein GAN trained on WRF-SFIRE simulations infers wildfire arrival times from satellite detections, achieving a 0.81 Sorensen coefficient for fire perimeters.
LSTM predicts local heatwave occurrence from historical temperature, humidity and wind data, outperforming traditional statistical models in accuracy and F1.
An LSTM with wind speed and storm direction context predicts North Atlantic tropical cyclone tracks at 3 to 12 hours, cutting path error by a reported 1.9 to 4.5 km.
Graph convolutional networks with extended generalized Pareto distributions model monthly Australian wildfire spread, including extremes, for hazard assessment.
Vision Transformer and CNN regressors estimate tropical cyclone intensity from satellite cloud images, and a fused two-stage strategy gives the best accuracy.
Machine learning predicts maximum air temperature to define heatwaves in Rajasthan and Karnataka, India; the abstract is unavailable, so details are from the title only.
Review surveys deep learning for extreme weather, covering thunderstorms, precipitation, droughts, heatwaves, cold waves and tropical cyclones, and discusses open limitations.
Deep learning is applied to predict sea-surface temperature variability and tropical cyclone patterns linked to coral bleaching; classified from the title only, as the abstract is missing.
LSTM-based blizzard prediction on Antarctic polar station data; reported accuracies of 49.6% and 55.2% are low and the evaluation is weakly documented.
Proposes MPM-Net, combining dehazing and ensemble learning to forecast Indian heatwave and cold wave events; the abstract is missing, so details are limited.
WEY-TCNet combines wavelet fusion and YOLOX with channel attention to detect tropical cyclones in Fengyun-2D infrared images, reaching 91.5% mAP versus 86.7% baseline.
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.
A CNN plus random forest hybrid using brightness temperature features improves severe hurricane prediction by 12% up to 3 days ahead.
Surface wind asymmetry descriptors from 1-km tropical cyclone simulations, classified by machine learning, improve intensification prediction skill by 11% to about 75%.
An analog-based stochastic weather generator emulates prolonged European heatwaves and extends return-time estimates, though a CNN predicts better with long training data.
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.
CatBoost model estimates western North Pacific tropical cyclone intensity from FY-2 cloud-top brightness temperatures, achieving 3.74 m/s RMSE and beating prior CNN approaches.
Neural network with SHAP projects US fire PM2.5 emissions to rise 4–75% by mid-century under SSP5-8.5, driven by warming and drying.
Lightweight deep learning detects tropical cyclone presence during simulations for online data reduction, reaching 92% recall on held-out ERA-Interim years.
A deep neural network combining GOES cloud-top satellite data with reanalysis parameters estimates severe hail likelihood, reaching CSI 0.511 against radar MESH.
Mixture density networks model monthly wildfire frequency and area burned across the western US, matching annual area burned with r = 0.85 observed.
Columbia University (LEAP, Lamont-Doherty)University of California, Los Angeles (UCLA)
Proposes a context-aware hybrid deep learning model for tropical cyclone track prediction; the abstract is missing, so results are unverified.
Kazakhstan avalanche bulletin uses a trained neural network on mountain weather station data, reaching 84-90% danger-level classification accuracy.
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.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
Analyzes 1979–2022 Dongjiang basin extreme precipitation trends and uses a BP neural network to simulate and predict them.
Mesogeos is a 1 km daily Mediterranean wildfire datacube with meteorology and ignition records, plus ML tracks for danger forecasting and burned-area estimation.
Wildfire Assessment Model uses a self-supervised CNN on atmospheric and greenness variables to estimate resources and burned area in Spain, improving baselines by up to 21%.
An ensemble of ML models trained on ERA5 jointly classifies and localizes tropical cyclone centres, detecting lower categories than those in training with improved localization over single models.
Consensus of classical ML models on HWRF predictors raises rapid-intensification detection to 56% POD, versus 10–30% for operational models.
Dynamic-balance CNN estimates tropical cyclone intensity from satellite imagery with 8.32 kt RMSE, 35% lower than the Dvorak technique.
Nanjing University of Information Science & Technology (NUIST)
ML models using VIIRS retrievals and HRRR variables estimate dead fuel moisture content over CONUS, outperforming climatology by 24% in hourly RMSE and beating MODIS-based models.
A GAN converts parametric typhoon wind and pressure fields into NWP-like forcing, giving storm surge accuracy comparable to NWP and better than parametric models.
Latent-space digital twin of JULES-INFERNO for global wildfire prediction, with latent data assimilation, running 500 times faster than the original model.
S2S models overestimate heatwave frequency over northern Nigeria at 30-day lead; deep-learning bias correction improves forecast reliability.
Neural emulator estimates wildfire spread likelihood under perturbed weather forecasts, matching SPARK fire boundaries (Jaccard 67.4%) about ten times faster than ensembles.
Review of deep learning methods for extracting tropical cyclone intensity and wind radius from satellite infrared imagery, with an outlook on future directions.
A neural network estimates GEV extreme-value parameters for CCSM3 annual maximum temperatures about as accurately as maximum likelihood, with large computational speedups.
Review-style paper on machine learning for wildfires and climate change; with no abstract, scope and results cannot be verified.
Transformer model for joint tropical cyclone track and intensity forecasting; the abstract is missing, so results cannot be verified.
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.
Trs-net applies a 2-D CNN to multispectral satellite images to analyse and classify tropical revolving storms. No abstract was provided.
ML models predict India heatwave occurrence a month ahead for each 1° grid cell using six months of maximum temperature data.
Unsupervised local causal states detect hurricanes, atmospheric rivers and precipitation extremes in climate data, and track vortices in 2D turbulence.
Uses an AlexNet-based CNN to identify tropical cyclones from aerial imagery for an alert system; a basic classification application.
GRU_CNN forecasts northwestern Pacific tropical cyclone tracks for 6 to 72 hours using track history and reanalysis fields, beating RNN, LSTM and GRU.
Nanjing University of Information Science & Technology (NUIST)Institute of Atmospheric Physics, Chinese Academy of Sciences
Machine learning lightning prediction model trained on ten years of Lokoja station weather records to give advance lightning warnings.
Rotation-equivariant convolution combined with a Transformer estimates tropical cyclone intensity from multi-platform satellite and environmental data, with RMSE of 8.19 kt.
Nanjing University of Information Science & Technology (NUIST)
A convolutional autoencoder is proposed to predict atmospheric rivers; the abstract is missing, so accuracy and data cannot be confirmed.
Supplementary figures only for a machine learning study of tropical cyclone precipitation; the record gives no method or results.
A graph neural network predicts regional heatwaves from weather station data, enabling real-time warnings at lower computational and data cost.
Probabilistic structural forecasts simulate geostationary IR radial profiles with PixelSNAIL and feed a CNN, giving intensity errors near NHC official forecasts.
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.
NVIDIANSF NCAR / UCARETH ZurichUniversity of OxfordUniversity of California, Berkeley
Modified light-weight Context Guided CNN with weighted loss raises recall for tropical cyclone detection in climate simulation data.
Machine learning gives probabilistic tropical cyclone intensity forecasts with intervals, and its point forecasts outperform an operational statistical-dynamical model over 2010–2020.
Coupled machine learning trajectory and intensity models generate synthetic Atlantic hurricane tracks and winds consistent with HURDAT2 historical records.
Temporal convolutional networks project live fuel moisture content three months ahead across the US, with RMSE 27.5% close to nowcasting accuracy.
SALAMA, a feedforward network post-processing convection-resolving NWP fields, gives calibrated thunderstorm probabilities with skill beyond NWP reflectivity up to eleven hours lead.
A spiking neural network downscales coarse precipitation to map meteorological drought (SPI) in Osmanabad, India; reported efficiency is claimed without rigorous benchmarking.
Neural network applied to MOSDAC satellite imagery for cyclone identification and monitoring, reported only qualitatively without a quantitative benchmark.
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.
Hybrid learning techniques predict tropical cyclone intensity; the abstract is unavailable, so this is judged from the title.
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.
Transfer-learning framework predicts unbalanced severe typhoon formation from atmosphere and ocean variables, outperforming ECMWF IFS and prior ML methods across three regions.
Bi-LSTM tuned by nomadic people optimisation detects and tracks tropical cyclones; the abstract is missing, so performance cannot be confirmed.
Pre-training on a suitable pretask improves U-Net generalization for segmenting unseen natural hazards from satellite imagery across regions without target labels.
CNN-derived ERA-Interim features (humidity, vorticity, ozone) added to SHIPS predictors improve tropical cyclone rapid intensification prediction by 23–43%.
Random forest regionalizes India into seven lightning climatologies and predicts seasonal lightning density from CAPE and humidity, with R² of 0.81 in monsoon.
Builds a recurrent neural network fuel moisture model for WRF-SFIRE, initialized as a numerical ODE solver and trained on RAWS sensor and RTMA weather data to improve fire-weather inputs.
YOLOv7 detects smoke columns in camera imagery for early wildfire warning, reaching F1 of 0.74 at confidence 0.298.
Ensemble of 20 deep learning models combining satellite images and environmental predictors yields higher rapid-intensification detection than operational forecasts for western Pacific tropical cyclones.
Random forests linking direct upward lightning tower measurements to reanalysis meteorology identify drivers of lightning risk at wind turbines, with skill on independent data across Germany.
Machine learning combined with WRF improves western North Pacific tropical cyclone track simulation; the abstract is unavailable, so the gain is unquantified.
Hybrid LSTM driven by climate model outputs to predict natural drought; the abstract is missing, so details are limited.
Multi-input CNN with wavelet-decomposed satellite imagery estimates Atlantic hurricane maximum wind speed with 15-knot RMSE.
Artificial neural networks project future rainfall intensity-duration-frequency curves under climate change; the abstract is missing, so results are limited.
Deep learning network detects tropical cyclones; no abstract is available.
Neural network predicts lightning occurrence in South Africa from meteorological station data; no abstract is available.
PTCIF gives probabilistic 6-24 h tropical cyclone intensity forecasts from multimodal data, with prediction intervals comparable to forecast centres on deterministic scores.
Neural style transfer makes simulated cloud images resemble satellite data, raising tropical cyclone precursor classification performance by 40–90% at 3–7 days lead.
PCMCI-based multidata causal discovery selects drivers for Western Pacific tropical cyclone intensity prediction, outperforming noncausal and correlation-based feature selection.
SmokeyNet fuses wildland camera imagery with weather sensor data to detect smoke, with a few-minute time-to-detection for early wildfire alerts.
Decision-tree framework using GOES cloud-structure parameters predicts major hurricane intensification with 73% accuracy 6 to 54 hours ahead.
A random forest on a modified MOD14 scheme with Himawari-8 AHI data detects early-stage forest fires in Australia at about 90% precision and recall.
A 3D DBSCAN clustering workflow detects the 2022 Yangtze heatwave and drought from ERA5 and tests how clustering parameters change event tracking.
Fudan University (FuXi)China Meteorological Administration (CMA)
Applies machine learning to improve tropical cyclone intensity forecasts from the NCMRWF ensemble prediction system; abstract not provided, so results are unknown.
A 2D/3D CNN with location-aware adaptive normalization improves wildfire danger forecasting on the FireCube dataset over baselines, with the performance gain reported but not quantified in the abstract.
Affinity Propagation clustering on a 2009-2021 Sicilian rainfall dataset identifies extreme rain events, including anomalous recent events near Catania.
Graph convolutional network estimates tropical cyclone intensity from satellite imagery; the abstract is missing, so details are inferred from the title.
A deep learning model forecasts tropical cyclone formation up to 60 hours ahead from ERA5 and IBTrACS, reaching 86.9% to 92.9% accuracy across six ocean basins.
Causal discovery identifies recurrent European heatwave tracks and their seasonality in a single-model large ensemble, enabling a new model evaluation approach for extremes.
Pre_3D fuses tropical cyclone intensity series with environmental factors via separate sub-networks, beating official agency 24-hour intensity forecasts by over 15%.
Lightweight multitask CNN estimates tropical cyclone intensity and size from satellite data, with intensity RMSE of 8.40 kt, 33.5% below the ADT baseline.
Nanjing University of Information Science & Technology (NUIST)
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.
Bayesian physics-informed neural networks model wildfire fronts under changing wind, adding data assimilation and uncertainty quantification to level-set fire spread.
Case-based expert system predicts forest fire risk in the Irkutsk region from 2017–2020 fire records, weather and remote sensing data, with a web-service prototype.
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.
Pyrocast pipeline with a 148-event pyroCb database and CNN or random-forest models predicts pyroCb six hours ahead with AUC 0.90.
Compares GEV models with GLM or neural network spatial interpolation of extreme Mediterranean precipitation parameters, with the ANN hidden-unit complexity key to performance.
Reviews 2021 extreme events and barriers to deploying ML models on Earth observation imagery for disaster damage detection, highlighting InSAR as a tool.
Early wildfire alert system combining a 3D CNN on satellite data and a CNN bias correction of WRF, feeding a Fire Weather Index for Colombian regions.
YOLOv4 locates tropical cyclone centres in satellite cloud images with over 99% confidence, the best of six deep learning models tested.
Deployed camera-based system detects wildfire smoke with a high true-detection rate and alerts faster than human observers.
Global open datacube and segmentation deep learning model predicting burned areas 8 to 64 days ahead using climate and fire-driver variables.
Software accompanying a study that uses neural networks to predict hurricane storm surge and assess its sensitivity to storm characteristics.
Deep learning estimates 34, 50 and 64-kt tropical cyclone wind radii in four quadrants from geostationary IR images, cutting errors 12-36% over prior methods.
Review argues for bottom-up AI and hybrid physical-statistical drought forecasting, prioritising robustness and uncertainty stability over event-based verification.
Ensemble k-nearest-neighbour method in a GIS environment predicts regional typhoon tracks with fast simulation and high accuracy.
Scattering-transform features with channel attention from multi-band satellite imagery estimate tropical cyclone intensity and predict lightning occurrence.
A deep network on ERA5 vertical profiles identifies lightning-favourable atmosphere better than expert features, with SHAP pointing to cloud ice and snow content.
TITP-Net spatio-temporal CNN forecasts typhoon intensity at 18-48 h, improving MAE by 5-7% over deep learning and official dynamical forecasts.
Neural network with a spatiotemporal encoding module estimates tropical cyclone intensity from infrared satellite imagery.
ConvLSTM with an added convolutional layer predicts typhoon tracks from ERA5 reanalysis image sequences, reaching 54.69 km mean distance error.
Review of flash drought concept and the potential of machine learning and deep learning for monitoring and subseasonal-to-seasonal prediction of rapid drought onset.
Artificial neural networks and isolation forests on Kenyan station data predict drought and flood occurrences with 78-90% accuracy, supporting early warning.
A ConvGRU with CBAM attention and a Deep&Cross CNN fusion forecasts tropical cyclone tracks from CMA and ERA-Interim data, outperforming prior deep learning baselines.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
CFS-SR applies image super-resolution to CFSv2 to give 4 km county-level fire weather forecasts to 7 days, improving accuracy over the western US.
University of WashingtonMax Planck Institute for Biogeochemistry
CNN classifies ERA5 and MERRA-2 fields for tropical cyclone presence, and ERA5-trained models transfer better to MERRA-2 than the reverse.
Distributed online learning across ten western US locations predicts wildfire grid maps one to four weeks ahead from remote sensing data, handling incomplete observations.
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.
Hybrid CNN-RNN-MLP network post-processes GEFS western North Pacific tropical cyclone tracks at 2–6 day leads, reducing track error and correcting speed and direction bias.
Deep learning on Himawari-8 and FY-4 imagery identifies typhoon vortices (88.6% accuracy above strong tropical storm grade) and estimates intensity with 3.8 m/s MAE.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
A CNN ensemble (PRIME) predicts tropical cyclone rapid intensification from MIMIC-TC microwave imagery, but overfits and reaches only about 30% correct No-RI classifications.
Deep learning predicts next-day wildfire danger in the Eastern Mediterranean, beating the Fire Weather Index, with explainable AI attributing importance to wetness variables.
Combines convolutional and recurrent layers to classify hail-favourable atmospheric profiles and forecast changes in hail frequency over coming decades.
Spatio-temporal neural network predicts Palmer Drought Severity Index across regions, reaching R² 0.90 versus 0.85 for gradient boosting, and projects drought under climate scenarios.
Neural network regression extends the global PDSI drought index from GLDAS data, achieving RMSE below 1 over 99.5% of the globe and providing an updated dataset.
Compares threshold, neural network and belief-function methods for 0–24 h lightning strike risk over the sea for aviation, with WWLLN ground truth.
U-ConvLSTM wildfire prediction using environmental and meteorological data reaches over 97% accuracy and predicts 2018 California fires 5-14 days ahead, versus about 76% for CNNs.
Partially-interpretable neural networks for extreme quantile regression of U.S. wildfire risk, plus a point-process extreme value model avoiding finite lower-endpoint issues.
Machine learning trajectory and intensity models trained on HURDAT2 generate synthetic Atlantic hurricanes whose landfall locations and wind speeds match historical records.
A CNN-LSTM hybrid (C-LSTM) predicts South China Sea typhoon tracks using Granger-selected meteorological features, showing smaller errors than a plain LSTM.
Machine learning system predicts tornadoes up to five days ahead with up to 84% probability on a new dataset of over 5,000 events.
Applies regression, clustering and CNN analysis to 28,713 Australian tropical cyclone records, identifying drivers of low central pressure and anomaly root causes.
Detects tropical cyclones in South Pacific and Atlantic coastal imagery by combining optical flow estimation with a ResNet deep learning classifier. Abstract unavailable.
Random forests trained on GEFS reforecasts and severe weather reports give probabilistic US severe-weather forecasts for days 4-8, comparable to SPC outlooks at days 4-5.
Colorado State University (incl. CIRA)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Data-driven forecasting of drought indicators for Jaisalmer, India; the abstract is unavailable, so the method and results are unverified.
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.
Combines Twitter sentiment with Global Fire Atlas satellite data to predict wildfire attributes, treating social media as a human-sensor for wildfire nowcasting.
Systematic review of 51 studies on tropical cyclone track and intensity prediction using NWP, machine learning and satellite methods across basins.
Position review arguing ML weather and climate studies must evaluate and improve performance on rare, high-return-period extremes, which remain poorly assessed.
Bayesian-optimised gradient boosted trees predict western North Pacific tropical cyclone intensity change at 12–72 h, beating linear regression by 4–8% in R².
Cluster-wise typhoon rainfall model combines SVM, RF, AdaBoost, CNN and LSTM with grid-search tuning, claiming 45–90% forecast efficiency gains for 1–6 h lead times.
Nanjing University of Information Science & Technology (NUIST)
Applies machine learning models to forecast severe convective weather. The abstract is unavailable, so the classification relies on the title.
Reviews regression, KNN and deep learning approaches for predicting heat waves, comparing algorithms to guide future heatwave forecasting models.
Reviews machine learning methods for tropical cyclone forecasting using satellite, radar and surface predictors, and discusses challenges for interpretability and precision.
LightGBM uses preceding-winter sea surface temperature, soil moisture and snow to simulate summer eastern European heatwave frequency, with SST contributing most.
AutoML framework forecasts US droughts from meteorological variables, matching deep temporal models with less training data and time.
Location-Refining neural network framework for heavy rainfall forecasting. The abstract is unavailable, so the classification relies on the title.
Review of machine learning for analysing, predicting and attributing atmospheric extreme events, covering methods, examples and open perspectives.
LSTM recurrent networks give real-time predictions of hurricane evolution; no abstract was provided, so quantitative skill and data details are unknown.
Compares SVM, Naive Bayes and pretrained AlexNet on satellite images to flag areas likely hit by hurricanes; methods are standard and evaluation is thin.
Convolutional neural weather models trained with extreme-focused losses improve global heatwave forecasts 1 to 28 days ahead, beating persistence and ECMWF S2S after two weeks.
California Institute of Technology (Caltech)Google ResearchUniversity of Oklahoma (AI2ES)
A ConvLSTM encoder-decoder with spatial attention forecasts 24 h western North Pacific tropical cyclone tracks from satellite images, about 30 km better than ECMWF-HRES and GFS.
I-RI CNN uses satellite imagery and scalar features to predict tropical cyclone rapid intensification, beating operational guidance at 12- and 24-h leads in the Atlantic.
Graph convolutional network with learned spatio-temporal affinities and a class-imbalance loss predicts rare rainstorms better than competing methods on two datasets.
Simplified VGG-16 CNN predicts tropical cyclone intensity from reanalysis predictors, outperforming LeNet-5 and AlexNet; SST, 500 hPa temperature and vertical wind shear are the top drivers.
Physics-informed neural network estimates 3D boundary-layer winds of extratropical cyclones over eastern North America, coupled with a synthetic track database for wind risk.
Deep learning model using drought indices, humidity and wind projects South Korean summer forest-fire occurrence to rise, with July up 15% by 2071-2100 under SSPs.
Wavelet analysis and Bayesian ML project emergence windows for Atlantic hurricane categories 2–5, with category 5 storms expected between 2026 and 2031.
Deep neural network ensemble for tropical cyclone track forecasting; abstract unavailable, so no quantitative results can be reported.
Multimodal multitask learning model for typhoon trajectory prediction; abstract unavailable, so performance cannot be verified.
DeepFire neural emulator of fire spread turns high-resolution weather forecasts into next-day fire danger maps and ensembles for 13 Corsican fires.
Surrogate-based multiobjective calibration of WRF parameters improves Bay of Bengal cyclone 10-m wind speed by 17.6% and precipitation by 8.2%.
Statistical trend analysis and SMOreg regression of 1891–2020 Bay of Bengal cyclone records project cyclogenesis to 2021; ML role is minor.
Hybrid deep learning forecasts the western Pacific subtropical high index during typhoon activity; abstract unavailable, so no metrics are reported.
Makes the NFDRS fire danger index differentiable so its parameters can be tuned by gradient descent on observed US and European fire events.
Neural network system with time-frequency decomposition forecasts tropical cyclone wave height in real time; the abstract is missing, so the classification relies on the title.
Uses a machine learning framework to observationally constrain 13 CMIP6 Earth system models, projecting higher socioeconomic wildfire exposure in Africa despite lower fire carbon emissions.
Oak Ridge National LaboratoryColorado State University (incl. CIRA)
Predicts annual tropical cyclone frequency for Guangdong from winter precursors using LSTM-FC, which achieves lower error than MLR and GPR for 2010–2019.
Trains an LSTM on COAMPS ensemble output to predict Typhoon Mangkhut's intensity and identify drivers of rapid intensification from asymmetry and heat flux features.
A machine-learning tool diagnoses inland tropical cyclone maintenance or intensification events from MERRA-2 and IBTrACS, highlighting land-surface state importance.
Faster R-CNN recognizes hail areas in Doppler radar imagery with over 86% accuracy and 97% precision on a test set.
Ongoing project builds ML models for extreme urban rainfall forecasts in Rio de Janeiro, reporting early results and open challenges.
Random forest forecasts monthly extreme rainfall indices over Bangladesh 1–3 months ahead using CAPE, humidity, and ENSO/IOD/NAO teleconnections.
Nanjing University of Information Science & Technology (NUIST)
XGBoost reconstructs hurricane surface wind fields from storm and ERA-Interim features, with errors below climatological variance, for synthetic risk applications.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)Columbia University (LEAP, Lamont-Doherty)New York University (M2LInES)
Backpropagation neural network replaces regression in a typhoon full-track model to generate 1000 years of Northwest Pacific typhoon events for hazard risk.
Applies deep convolutional neural networks to improve tropical cyclone wind field estimates for extreme event analysis.
CNN with residual and attention blocks estimates tropical cyclone intensity from Himawari-8 cloud products, reaching RMSE 4.06 m/s on independent tests.
Uses CNN models on IMERG satellite precipitation to forecast near-real-time rainfall from hurricanes, targeting flood-relevant extreme rainfall.
A deep belief network driven by relative humidity, wind speed and drought index estimates monthly forest fire occurrence in South Korea with R² of 0.84.
Nonparametric neural-network test links 24-hour tropical cyclone infrared image sequences to rapid intensification events, with guaranteed type I error control.
A CNN detects bow echoes in French kilometre-scale model output with 86% hit rate and 39% false alarms, and forecasters found the products useful.
Two machine learning models forecast summer Eurasian heatwave frequency from preceding SST, snow, soil moisture and sea ice, outperforming linear regression in 2010–2019 forecasts.
A 3D CNN combined with a GRU forecasts tropical cyclone tracks, aiming to improve track prediction over conventional approaches.
Argues that data science approaches can improve hurricane forecasting beyond traditional statistical methods; methods and results are only vaguely described.
Self-attention augmented convolutions with a highway layer forecast extreme precipitation up to 12 hours ahead from pressure and wind, outperforming CNNs by 12%.
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.
Tsinghua University (NowcastNet)China Meteorological Administration (CMA)
A random forest on ERA5 variables shows self-initiated upward lightning at a tall tower depends mainly on the height of the -10 C isotherm.
Applies neural network models to predict extreme precipitation events in southeastern Brazil; the abstract is unavailable, so results are unknown.
Proposes a neural network framework for fine-grained tropical cyclone intensity prediction; the abstract is unavailable, so details are limited.
Applies artificial intelligence to hurricane storm surge hazard assessment; abstract unavailable, so significance and scope are low-confidence.
Uses a backpropagation neural network for typhoon warning relevant to historical buildings; the abstract is unavailable, so the weather component is unclear.
A parametric cyclone and neural network hybrid predicts typhoon storm surge at Longdong, Taiwan, with 33–40% lower RMSE than a pure network at 8 h lead.
A deep learning model trained on image-like wind fields around PIREP turbulence reports predicts Japanese airspace turbulence with accuracy comparable to existing grid-point indices.
Nanjing University of Information Science & Technology (NUIST)
Meta-analysis of machine learning for dust storm detection and prediction, finding SVMs dominate satellite detection and SVM/random forests dominate meteorological-data prediction.
DeepTyphoon couples hybrid dilated CNN and LSTM on satellite image sequences to predict typhoon tracks, achieving 64.17 km mean error at 6 hours.
Reviews a CNN and data-mining based real-time cloudburst prediction system motivated by the 2023 Sikkim flood; no quantitative results reported.
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.
Conference talk on integrating numerical simulation with machine learning for weather forecasting and improving typhoon forecasts through a data competition.
Encoder-decoder LSTM predicts Florida hurricane storm surge 1 to 12 hours ahead from observed surge and meteorological inputs, outperforming a CNN in all cases.
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.
Dilated GRU recurrent network with mutual-information feature selection predicts North Indian Ocean cyclone landfall location and intensity 12 to 36 hours ahead.
DBF-Net fuses TC state vectors with reanalysis pressure fields to forecast Northwest Pacific tropical cyclone tracks at multiple horizons, improving accuracy over single-modality methods.
Stacked LSTM-CNN ensembles with fully connected or SVM meta-classifiers improve sand-dust storm prediction over single LSTM or CNN models in Inner Mongolia.
Combines grey prediction with a BP neural network to forecast extreme precipitation events; the abstract is unavailable, so results are unknown.
Hybrid NOF-ELM classifier uses hurricane meteorological data to detect extreme wave height occurrence in the Gulf of Mexico with over 99% mean accuracy.
LSTM model using multi-factor typhoon time series predicts storm-surge water level at a Yangtze Estuary station, giving 1 h accuracy and 15 h predictions, 41% better than traditional ML.
3D-attention CNN estimates tropical cyclone intensity from multichannel IR, water vapour and microwave imagery, with RMSE 9.48 kt, 25% better than the Advanced Dvorak Technique.
Nanjing University of Information Science & Technology (NUIST)
Random forest on dynamically downscaled weather produces a daily gridded chamise fuel moisture product for California, reaching R2 0.79 for new growth, to support wildfire danger.
Curated nearly decade-long US wildfire spread dataset combining satellite fire data with weather and fuel variables, plus a neural network baseline for next-day spread.
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.
Multitask framework uses an improved GAN to predict tropical cyclone fields, then networks extract track and intensity, with 24-h track error 116 km and intensity error 13.06 kt.
Mask R-CNN detector, wind speed filter and CNN classifier detect tropical cyclones in satellite images with 97.10% precision on test images.
Competition-derived deep networks with oversampling, ensembles and cost-sensitive losses classify imbalanced cloud images for tropical cyclone precursors, improving performance by 65.4% over conventional methods.
Applies a Single Shot Detector to detect Northern Hemisphere extratropical cyclones at developing, mature and declining stages, reaching 86.64% mAP for mature cyclones.
Applies classical machine learning classifiers to predict heavy rainfall days at one peninsular Indian station; the abstract is missing, so results are unreported.
Deep neural network with online estimation predicts burned area at five western US locations from climate and fire history data.
Conference abstract proposing rare-event simulation and stochastic process-based machine learning for dynamical prediction of extreme heatwaves; no details are available.
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.
Nanjing 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.
Random forest classifies whether southwest Pacific tropical depressions develop into storms with 91% test accuracy, with aerosol optical depth among the top predictors.
A neural network using a microwave rain index and 24 h wind changes identifies tropical cyclone rapid intensification with near-perfect skill and reveals environmental drivers up to 1050 km away.
SAF-Net proposes a spatio-temporal deep learning model for typhoon intensity prediction; no abstract was available, so this rests on the title alone.
ConvLSTM on harmonized Earth observation and weather covariates forecasts next-day national fire danger with AUC 0.926, beating a random forest baseline.
A BiGRU with attention predicts Northwest Pacific tropical cyclone tracks at 6–72 h, outperforming RNN, LSTM and GRU baselines and beating agency forecasts at longer leads.
Many-to-many RNN models predict hurricane trajectories from HURDAT2 data, running faster than NHC ensembles with comparable errors for 12-hour forecasts.
A GRU recurrent network predicts heatwave onset probability for Seoul from daily weather, reproducing 86% of test-period heatwave days.
A simple CNN estimates tropical cyclone intensity for Indian-basin storms, claimed to beat state-of-the-art methods with lower computation, with few quantitative details in the abstract.
Random forest trained on 150,000 ERA5 proximity soundings best classifies significant tornadoes and hail, with SHAP-derived thresholds for operational nowcasting.
Simple machine learning models refine satellite-based advanced Dvorak technique features, cutting global tropical cyclone wind RMSE by 23–30% over 2017–2018 test years.
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.
Nanjing University of Information Science & Technology (NUIST)
Advanced machine learning models predict heatwaves across diverse Iranian regions from meteorological variables; the abstract is missing, so details are limited.
A lightweight deep learning filter detects tropical cyclones in running simulations with 92% recall and 36% precision, so models save only relevant output.
LightNet+ fuses radar, station and NWP data in a deep network for lightning forecasting, beating three established schemes and improving as data sources are added.
Fudan University (FuXi)China Meteorological Administration (CMA)
AI sequence models with statistical inference reveal interpretable convective structure patterns preceding tropical cyclone rapid intensification, aiding forecaster insight.
Urban energy balance-guided machine learning predicts nocturnal surface urban heat island during a Naples heatwave; abstract missing, judged from title.
Surrogate model for time-dependent metocean conditions during hurricanes. No abstract is available, so this classification is low-confidence.
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.
A VAE maps monsoon precipitation fields to a known distribution, enabling controlled synthesis of more extreme rainfall scenarios over southwest India.
Random forests trained on convection-allowing WRF output probabilistically forecast excessive rainfall over CONUS, but they consistently underperform WPC's day-1 Excessive Rainfall Outlooks.
Cloud-deployed machine learning compares classifiers for predicting hurricane and cyclone occurrence over Pacific and Atlantic coasts, with map visualization; the methodological novelty is limited.
Classical ML models estimate North Indian Ocean cyclone grade (88% accuracy) and maximum sustained winds (RMSE 2.3) from 28 years of best-track data.
Backpropagation neural network estimates the Holland B wind-field parameter for Northwest Pacific typhoons from pressure difference, improving super-typhoon wind speed estimates over statistical models.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
Benchmarks 16 classical and 16 deep learning models predicting drought levels from 90 days of 18 meteorological indicators, with no single model best across metrics.
DR-Transformer estimates tropical cyclone intensity from satellite images with rotation and distance-consistency features, reaching 7.76-knot RMSE on the TCIR dataset.
Integrates multi-sensor Earth observation data and machine learning to monitor meteorological drought in central India; the abstract is missing.
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).
VGG-16 with fisheye preprocessing classifies typhoon intensity from single satellite images more accurately than prior work, with Grad-CAM showing physically relevant eye features.
Unsupervised topological data analysis is used to identify atmospheric river patterns linked to flooding in Nigeria; no abstract is available.
Typhoon-CNNs classifies typhoon intensity from more than 10,000 satellite cloud images with 88.74% accuracy, beating ResNet50, InceptionV3 and VGG16.
An MLP using SHIPS predictors beats operational statistical-dynamical models by 5-22% for Atlantic tropical cyclone intensity in 2019-2020 real-time tests.
Compares CNNs with analogue methods for predicting US regional extreme precipitation from large-scale patterns, finding CNNs more skillful and integrated vapor transport the most predictive variable.
CSU-MLP random-forest excessive rainfall forecasts, built from ensemble reforecasts, were transitioned into operational use at NOAA's Weather Prediction Center.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)Colorado State University (incl. CIRA)
Interactive map system that uses a machine learning model to show real-time US wildfire risk from environmental data and historical fire events.
Survey of deep learning methods, including RNNs and CNNs, for predicting extreme weather events.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
A 1D CNN uses hydrometeorological precursors to assess basin-scale meteorological drought, with over 70% accuracy across seven drought categories.
Improved CNN estimates hurricane intensity from infrared satellite imagery with 7.6-knot RMSE and adds VGG19-based post-disaster damage assessment.
An artificial neural network identifies K-index, AOD and CAPE as key predictors of lightning over India using daily and monthly data.
GAN generates spatio-temporal weather fields conditioned on extreme-event segmentation masks, tested on surface radiation and zonal wind data.
Deep neural networks on UAS orthomosaics detect and classify tornado damage, producing georeferenced damage heatmaps for post-event assessment after the 2018 Eureka tornado.
Spatio-temporal threshold clustering uses pattern recognition and clustering to define consistent extreme precipitation episodes over the contiguous US from gauge data.
ML-optimized ensemble selection and weighting improves preseason Atlantic hurricane forecasts over simple averaging for major hurricane counts.
CNN-based fusion of satellite and ERA5 predictors detects 97.1% of developing tropical cyclones in the western North Pacific at a 20.3% false alarm rate.
Nanjing University of Information Science & Technology (NUIST)
CNN feature fusion of sea-level pressure, SST and wind predicts preseason Atlantic hurricane activity with skill up to 18 months ahead.
Machine learning classifiers categorize tropical cyclone intensity from infrared cloud images over the Bay of Bengal and Arabian Sea.
LSTM on best-track data predicts North Indian Ocean landfall intensity, location and time; 24-hour input gives 4.24-knot and 51.7 km errors.
CNN-LSTM predicts tropical cyclone landfall location and time from 9–21 hours of ERA5 data, with location errors of 66–159 km across six basins.
Interpretable machine learning methods inspired by the Dvorak technique estimate tropical cyclone intensity from satellite imagery; the abstract is unavailable, so details are limited.
CNNs estimate western North Pacific tropical cyclone intensity from Himawari-8 infrared images, reaching 88.9% accuracy and 4.62 m/s RMSE with focal loss and binary framing.
CNN classifiers (VGG, ResNet, Xception) on satellite imagery detect tropical cyclone rapid intensification, with Xception reaching ROC-AUC 0.911 and FAR 0.182.
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.
Tests 13 machine learning algorithms with pre- and post-integration of satellite precipitation products for drought indices over India; Bayesian-regularized neural networks perform best.
CNN trained on climate model output forecasts long-lasting extreme heatwaves up to 15 days ahead, using class-imbalance undersampling and transfer learning.
Video-based CNN with ConvLSTM locates tropical cyclone centers from hourly geostationary infrared sequences, giving a 19.3 km median error and 42% lower error than baseline.
An LSTM with Kalman filter smoothing predicts Atlantic hurricane tracks from 1851–2016 historical data with the lowest error among compared LSTM variants.
An LSTM with rolling forecasts predicts typhoon intensity up to 120 h from historical data and outperforms a feed-forward network on 2015 typhoons.
Institute of Atmospheric Physics, Chinese Academy of Sciences
ML-enhanced analogue method using LLE reduces NWP precipitation error for tropical cyclones at Shenzhen, improving on raw NWP and traditional analogue approaches.
Backpropagation neural network with iteratively selected predictors forecasts typhoon precipitation occurrence in Anhui with 100% hindcast and 90% validation accuracy.
Compares AI models with conventional methods for forecasting central pressure drop and maximum wind speed of Arabian Sea cyclones.
Deep learning model for predicting climate-induced natural disasters; abstract unavailable, so the approach and results are unverified.
Combines WUDAPT urban morphology data with machine learning to spatially estimate heatwave patterns; the abstract is unavailable, so the details are limited.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
A modular framework generates extreme precipitation scenarios from interchangeable ML components to support climate resilience planning, with initial comparisons of two approaches.
ClimateNet releases an expert-labeled dataset of tropical cyclones and atmospheric rivers in CAM5.1 output, and a deep learning segmentation model that enables pixel-level extreme event analysis.
Colorado State University (incl. CIRA)ETH ZurichLawrence Berkeley National Laboratory / NERSCNSF NCAR / UCARNVIDIAUniversity of California, Berkeley
A machine learning model built on remote sensing data predicts forest fires in Indonesia with AUC above 0.81, beating a 0.70 baseline.
Support vector machine model predicts heat waves over Pakistan under climate change; the abstract is unavailable, so details are limited.
RNN outperforms DNN for fast prediction of typhoon-driven storm surge time series at ten Tokyo Bay points, trained on 151 simulated typhoons.
Reviews machine learning methods for drought prediction, covering drought indices, datasets and performance metrics, with an India focus.
A distributed multitask deep network using climatic, geographic and physical features is proposed for typhoon track prediction; the abstract gives no quantitative gain.
A cascaded CNN classifies and estimates northwest Pacific tropical cyclone intensity from infrared satellite images, reaching 8.60 kt RMSE against best-track data.
Random forest on RUC pre-storm sounding data discriminates tornadic from nontornadic storms, with feature analysis showing pressure has little impact.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Object-based verification of storm-scale mesocyclone probabilities, plus evaluation of ML calibrations of Warn-on-Forecast severe weather guidance.
AGU 2020 session on deep learning for climate science and extreme weather prediction; a conference session listing with no technical content.
AGU 2020 poster session on deep learning for climate science and extreme weather prediction; a conference listing with no technical content.
Random forest, autoencoder and CNN models diagnose hailstorm occurrence globally from four ERA5 thermodynamic attributes, with random forest performing best.
Feasibility study replacing WRF-SFIRE numerical solvers with physics-informed machine learning in Julia to enable real-time wildfire spread simulation.
Physics-augmented deep learning estimates tropical cyclone intensity and size from satellite imagery; classified from title only, as no abstract was available.
Trains a ConvLSTM on simulated data to model wildland fire spread under dynamic wind and terrain, finding it outperforms other architectures in capturing spread rate.
Machine learning model for western North Pacific tropical cyclone track forecasts; abstract unavailable, so method details and skill cannot be verified.
Hurricast combines encoder-decoder features with gradient-boosted trees for 24-hour tropical cyclone track and intensity, matching operational skill in seconds and improving NHC consensus.
CNN on polar-coordinate satellite images profiles full tropical cyclone structure parameters and wind profiles, with a new labeled benchmark dataset as baseline.
GAN-CNN framework trains on visible and passive microwave data and estimates tropical cyclone intensity from IR and water vapor images every 15 minutes, versus 3-hourly before.
Modular fuzzy neural network with information-gain predictor selection improves 24 h heavy-rainfall forecasts from ECMWF output over Guangxi stations.
Compares four deep learning models on nearly a decade of remote sensing, weather and fire records, reaching 83% AUC for wildfire likelihood.
A CNN trained on current-climate storms skillfully classifies strongly rotating, potentially severe convection in end-of-century climate simulations, generalizing to future conditions.
Deep learning framework monitors physically relevant predictors and their relationships to aid human forecasters in tropical cyclone intensity change prediction.
Support vector regression on ATMS satellite warm-core temperature anomalies estimates Atlantic hurricane intensity, with Pmin error of 7.23 hPa when day of year is included.
TCLNet, a lightweight fully convolutional network with a new typhoon center dataset, locates typhoon centers with 14.4% higher accuracy and 92.7% fewer parameters.
CNNs pre-screen GCM circulation data so only a small subset needs kilometre-scale dynamical downscaling, retaining 92–98% of extreme precipitation events.
A 3D-CNN on surrounding atmosphere and ocean fields predicts 24 h tropical cyclone intensity change, improving MAE and intensification classification accuracy.
GAN combined with extreme value theory generates realistic extreme samples, evaluated on US precipitation, with constant-time generation at a given extremeness probability.
A tensor-algebra CNN with multitask learning estimates tropical cyclone wind speed and intensity class from FY-4 multispectral imagery, handling invalid bands.
A GCN-LSTM model classifies typhoon levels from Himawari-8 cloud images with 92.35% accuracy, and 95.12% for typhoons and super typhoons.
Deep CNN trained on remotely sensed atmospheric and environmental series predicts 24-hour wildfire perimeter evolution in the Sierra Nevada with 78-98% validation accuracy.
Generalized regression neural network with dual-factor feature extraction for rainstorm prediction; only the title is available, so the abstract-based assessment is limited.
Machine learning using satellite inner-core precipitation and outflow temperature beats the NHC operational rapid-intensification consensus, improving Peirce skill score by 12 to 138%.
IBM ResearchNASACalifornia Institute of Technology (Caltech)
Interpretable satellite-derived convective structure features with logistic lasso classify tropical cyclone rapid intensity change, matching the accuracy of environmental predictors.
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.
Proposes a neural conditional-density estimator with case-control sampling for tropical cyclone intensity forecasting, outperforming other ML conditional-distribution methods in simulations and Atlantic storm data.
Deep CNN real-time system for evaluating extreme rainfall to support construction-industry decisions; the abstract is missing, so details are limited.
Reviews machine learning for tropical cyclone genesis, track, intensity, rainfall, storm surge and numerical model improvement, highlighting opportunities and challenges.
CNNs using radar images and proximity soundings predict next-hour tornado occurrence with AUC above 0.9, with a GridRad critical success index of 0.31.
Machine-learning multi-model ensemble and bias correction of GCMs project Nigerian meteorological drought changes under RCP scenarios through 2099.
An autoencoder plus GRU model forecasts typhoon landing locations, beating NWP by 15-56% in absolute position error at 24-72 hours.
Applies machine learning to predict summer temperatures and heat wave occurrence in Korea at seasonal lead; abstract unavailable, so judged from the title only.
Four U-Net models detect tropical and extratropical cyclone regions from GFS precipitable water and GOES imagery, with Dice scores of 0.51 to 0.76 and 3x speedup over a heuristic.
Unsupervised representation learning over text and weather data improves clustering for detecting severe weather events such as windstorms, floods and tornado outbreaks.
CNNs predict next-hour tornado occurrence (AUC 0.94) and detect fronts in reanalysis, with interpretation methods explaining tornadogenesis failure.
Random forests trained on GEFS reforecasts give calibrated Day 1-3 severe weather outlooks that beat SPC at Days 2-3, and blending with SPC improves Day 1 outlooks.
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 k-nearest-neighbour model conditioned on reanalysis environment generates stochastic tropical cyclone event sets for the South West Pacific for cyclone risk assessment.
A 3D convolutional network using environmental fields and sea surface temperature predicts 24-hour tropical cyclone intensity change, with better strengthening/weakening classification accuracy.
Produces a gridded Canadian extreme precipitation IDF dataset by linking reanalysis predictors to extremes with a support vector machine, with 5–25% RMSE against observations.
Adds word-embedding features from disaster tweets to CNN and BiLSTM models to improve typhoon intensity prediction, gaining 12.1% and 3.1% F1 over those baselines.
Random forest, SVR, ANN and a multi-model ensemble correct next-day LDAPS extreme temperature forecasts in Seoul, cutting maximum-temperature RMSE from 2.08 to about 1.6 °C.
A CNN trained on 35 years of meteorological maps and visibility data forecasts severe haze occurrence in Singapore, beating a no-skill baseline but missing many haze events.
Uses an RNN with dropout to predict tropical storm trajectories with Bayesian intervals, giving evacuation-relevant uncertainty ranges from about 500 Atlantic storms.
Builds the first global nonlinear SVM and neural-network tropical cyclone intensity hindcast schemes; globally trained SVM improves skill by about 10% over regional linear schemes.
Scoping review of 298 studies applying machine learning across wildfire science, including fire weather and climate change, and identifying gaps for deep learning.
Stacked denoising autoencoder with logistic regression predicts thunderstorm occurrence up to 15 hours ahead over 400 km2 South Texas, beating shallow networks and forecasters.
Deep learning approach to predicting hurricane genesis and evolution; only the conference listing is available, so details are unverified.
Artificial neural network approach to predicting tropical cyclone storm surge; only the title is available, so details are unverified.
NASA report on a machine learning aid for predicting hurricane rapid intensification, giving communities more warning time; technical details are limited.
Machine learning on satellite imagery estimates high-resolution temperature during heat waves; abstract unavailable, so details are unverified.
Multi-task deep learning fuses COMS geostationary images with CFSv2 forecasts, cutting 6h and 12h tropical cyclone intensity forecast RMSE by 13% and 16%.
PSO-optimised neural network forecasts cloudburst rainfall in India from historical records; accuracy figures are not given in the abstract.
Random forest and CNN on RUC pre-storm soundings predict supercell tornadoes, reaching about 66% accuracy for tornadic cases after resampling.
CNN estimates FY-4A satellite tropical cyclone intensity off southeast China with over 97% class accuracy and 1.75 m/s mean absolute error.
CNN classifies storm versus non-storm satellite images for cyclone detection after augmentation, reporting 90% accuracy, with linear regression for track prediction.
CNN-GRU model with feature selection predicts tropical cyclone tracks from 1945-2017 data, outperforming NWP and statistical baselines; no numeric gain given.
Hybrid CNN with classification and BP network estimates tropical cyclone intensity from satellite images with RMSE of 8.91 kt.
Fudan University (FuXi)Nanjing University of Information Science & Technology (NUIST)
2D and 3D CNNs estimate tropical cyclone intensity from multispectral geostationary imagery, reaching 8.32 kt RMSE, about 35% better than single-channel CNNs.
Integrated machine learning is proposed for detecting tropical cyclones in advanced climate models; no abstract is provided, so the method is not verifiable.
Uses buoy wind and air temperature time series with Kalman-filter imputation and standard ML models to predict severity of tropical storms and hurricanes.
Deep neural network emulates WRF-simulated typhoon wind speeds at arbitrary locations in Northern Taiwan, matching WRF accuracy across 47 typhoon events.
Neural network optimisation for predicting similar typhoons; the abstract is only a generic AI sentence, so the contribution is unclear from the abstract.
A neural network fusing past storm trajectories with reanalysis wind and pressure fields forecasts 24 h tropical cyclone displacement in seconds, complementary to dynamical models.
Hybrid ANN and multiple regression forecasts typhoon total rainfall and groundwater-level change in Taiwan's Zhuoshui basin, raising accuracy to about 80% from station data.
Deep learning approach to forecasting different convective weather types; the abstract is missing, so details are limited.
Benchmark of satellite images for tropical cyclone rapid intensification, with an attention-based spatiotemporal model showing deep learning feasibility for RI prediction.
A two-branch CNN estimates northwest Pacific tropical cyclone intensity from infrared and water vapour images, reaching RMSE 5.13 m/s with 60×60 pixel inputs.
Nanjing University of Information Science & Technology (NUIST)
Knowledge-enhanced deep learning emulates tropical cyclone boundary-layer wind fields; abstract unavailable, so assessment is based on the title.
Physics-based unsupervised representation learning segments extreme weather structures in CAM5.1 water vapor climate model output, a route to identifying events without labels.
Describes a deep-learning portal for estimating tropical cyclone intensity; no abstract is available, so the detail is limited.
Thesis-style work applying machine learning to tropical cyclone impact assessment; the abstract is unavailable, so scope and results are unclear.
Pattern-recognition algorithm identifies mixed wind hazard climate types from station records, then recalculates extreme wind speeds and design return periods for southeast China.
Uses a CNN-LSTM on COMS satellite channels to estimate typhoon maximum wind speed, reaching 9.84 knot RMSE with the WV channel.
Deep-PHURIE uses deep learning to estimate hurricane intensity from infrared satellite imagery; the abstract is unavailable, so accuracy is unknown.
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.
Classifies tropical cyclone intensity from infrared satellite images using GLCM texture features and an SVM, reaching 88% accuracy against the Saffir-Simpson scale.
Random forests post-processing HREFv2 ensembles with isotonic calibration give severe hail probabilities comparable to SPC outlooks over CONUS.
Capsule networks on Z500 patterns predict North American heat and cold waves 1-5 days ahead, beating CNNs and logistic regression with about 80% accuracy when temperature is added.
Random forests, boosted trees and elastic nets predict storm longevity in NWS quasi-operational testing, with lower error than human forecasters.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
A CNN outperforms logistic regression at predicting severe hail from convection-allowing NWP fields, and interpretation reveals learned storm and environment features consistent with hail physics.
XGBoost predicts Western North Pacific tropical cyclone intensity changes at 6–24 h lead times, with lower MAE than a back-propagation network.
AdaBoost classifies whether mesoscale convective systems become tropical cyclones, reaching 97.2% F1 at 6 h lead and beating the genesis potential index.
Feed-forward network on HWRF output beats operational intensity guidance at 24–57 h in cross-validation and quadruples rapid-intensification hits.
DeepMicroNet, a CNN on passive microwave satellite imagery, estimates tropical cyclone intensity with 9.6-14.3 kt RMSE, matching known best-track biases.
Neural networks trained on past BRAMS forecasts and lightning density fields predict cloud-to-ground discharge occurrence as an ancillary convective-alert tool.
A three-layer backpropagation network predicts whether a West Pacific tropical cyclone will make Philippine landfall at genesis, with 91% accuracy at 54 hours.
Random forest, decision tree and SVM detect western North Pacific tropical cyclone formation from WindSat with 94–96% hit rates, 26–30 h early.
Applies machine learning with ERA5 convective parameters and radar reflectivity to predict large hail, highlighting which predictors matter most.
Hybrid CNN-LSTM model forecasts typhoon formation; abstract unavailable, so the approach and results are inferred from the title only.
Self-organizing maps applied to reanalysis identify four circulation pattern types behind extreme US floods, including tropical moisture exports and tropical cyclones.
Compares logistic regression, random forests, boosted trees and neural networks predicting ERA5 thunderstorm occurrence, with about twofold skill gain over CAPE at 0–45 min.
Builds a data-driven model linking SPEI drought indices to Catalan burned areas, forecasting summer burned area seasonally with stakeholder co-design and cross-validated skill.
Literature review of over 90 studies (2000–2017) on neural networks for drought and monsoon rainfall forecasting, mainly in India.
Interview on an RNN-based tropical cyclone intensity prediction model, arguing deep learning can address the black-box nature of typhoon forecasting.
Deep learning applied to phased-array weather radar volumes detects tornadic vortices in Typhoon Nanmadol (2017); the abstract gives no quantitative results.
A CNN trained on satellite infrared and microwave data estimates tropical cyclone intensity with 8.39 kt RMS error, offering an objective alternative to the Dvorak technique.
Random forest on Himawari-8 brightness temperatures and NWP water vapour classifies weak, medium and severe pre-convective storms over East Asia with 0.79 accuracy.
Gauge-equivariant CNNs on the icosahedron improve segmentation of global climate patterns, alongside omnidirectional images, with a single conv2d implementation.
Self-organizing maps applied to downscaled REGCM4 projections divide Iran into nine extreme-heat regions and project increases in mid-century heat.
RNN trained on Western North Pacific tropical cyclone records predicts 24-hour intensity with 5.1 m/s error, beating some dynamical models.
Neural network model using lagged precipitation and vegetation variables predicts drought one month ahead in Kenya, with R2 of 0.78 on out-of-sample data.
Updated machine learning ensemble for Atlantic tropical cyclone rapid intensification forecasting; no abstract is available, so details are limited.
Random forest predicts 3-month SPI drought in East Nusa Tenggara with about 0.8 AUC, with MERRA-2 inputs outperforming TRMM.
Compares ANFIS neuro-fuzzy and WRF models for South China Sea typhoon wind speed; ANFIS gives higher correlation and lower RMSE.
Deep learning model estimates tropical cyclone intensity from satellite data at higher temporal frequency, deployed in a production monitoring system with a portal.
Uses a GAN on satellite images plus reanalysis fields to predict typhoon centre tracks 6 hours ahead, with 67.2 km mean error versus 95.6 km for images alone.
A CNN ensemble detects western North Pacific tropical cyclones and their precursors 2–7 days ahead in cloud-resolving simulations, with 79.9–89.1% detection.
PHURIE uses machine learning on infrared satellite imagery to estimate hurricane intensity; abstract unavailable, classification based on title only.
MARS regression on Korean emergency visits and meteorological data yields heatwave thresholds, such as 32.58°C maximum temperature for two consecutive days.
Fused deep learning model forecasts hurricane tracks from reanalysis data; abstract unavailable, so classification is based on the title only.
Deep learning similarity search to retrieve historical tropical cyclones matching a query storm; abstract missing, so details are inferred from the title.
Applies deep learning to cloud-resolving simulations to detect tropical cyclone precursors; abstract missing, so details are inferred from the title.
A GAN predicts one-step-ahead typhoon centres and cloud structure from 6-hourly satellite images, with 42% of centre errors under 80 km.
Review of gust forecasting methods, comparing physical boundary-layer parametrizations on NWP fields with statistical and machine learning approaches trained on observations, and outlining challenges from higher resolution.
Self-organising incremental neural network predicts tropical storm tracks from centre latitude, longitude and pressure over East Asia, with low computational cost.
Framework learns ice storm occurrence from atmospheric variables with weak labels, evaluated on reanalysis and GCM data under future climate scenarios.
LSTM neural network model for typhoon track nowcasting; abstract not available, classified from title.
Deep CNNs estimate tropical cyclone intensity directly from infrared satellite imagery as an objective alternative to subjective Dvorak analysis.
Generalized regression neural network consensus of dynamical models improves Atlantic tropical storm intensity forecasts over 2012–16, outperforming individual models and the ensemble mean.
Conference abstract applying deep learning to detect tropical cyclone precursors in a global nonhydrostatic atmospheric model simulation.
A fully connected RNN predicts hurricane tracks from NHC best-track latitude, longitude, wind and pressure, forecasting about 120 hours ahead at 6-hour steps, competitive with NHC methods.
An AMS 2018 conference abstract on applying machine learning to reduce tropical cyclone prediction errors. No abstract is available, so the assessment is low confidence.
Machine-learning post-processing gives calibrated probabilistic forecasts of flash floods, extreme rain, tornadoes, hail and convective winds across the US.
Shows AI methods, combined with physical understanding, improve forecasts of storm duration, severe wind and hail, precipitation type, renewable energy and aviation turbulence.
NSF NCAR / UCARNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Storm-based machine learning produces calibrated probabilistic severe hail forecasts from convection-allowing ensembles, achieving higher critical success index than existing hail forecasting methods.
Introduces ExtremeWeather, a labeled climate dataset and semi-supervised spatiotemporal CNN for localizing hurricanes, extratropical cyclones and fronts in simulations.
First deep CNN application to detect tropical cyclones, atmospheric rivers and weather fronts in climate datasets, reaching 89-99% accuracy.
An early neural network predicts tornadoes from Doppler radar-derived storm attributes. Abstract not available; classification based on title.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Extreme Weather Bench is an open community benchmark of case studies, observations and impact metrics for verifying AI and NWP models on high-impact weather.
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.
Prithvi WxC is a 2.3-billion-parameter MERRA-2 foundation model that transfers to forecasting, downscaling, gravity-wave parameterization and extreme-event tasks.
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.
Climate analysis & attributionColorado State University (incl. CIRA)Stanford University
WeatherQA is a multimodal benchmark of 8,000+ image-text pairs for severe convective weather reasoning; GPT-4o shows a substantial gap to human forecasters.
LLMs & agentsAllen Institute for AI (Ai2, ACE)University of Washingtoncode
Hybrid ANN plus quantile-mapping post-processing of GEFSv12 reforecasts reduces warm bias and improves summer Tmax and extreme-day forecasts over Taiwan at extended ranges.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCAR
Light3DUnet, a 3D U-Net on radar reflectivity, nowcasts cloud-to-ground and intra-cloud lightning within one hour over central China.
U-Nets retrieve storm maximum vertical velocity from 3D radar reflectivity with probabilistic outputs, reaching R² above 0.65 on WoFS simulations.
Conditional GANs generate synthetic ensemble members from HRRR convection-allowing forecasts, and a CNN turns them into severe weather probabilities with up to 20% Brier skill score gain.
Neural networks emulate mean radiant temperature to guide an optimization of urban tree placement that reduces heat stress during heatwaves and over decades.
A capsule neural network estimates tropical cyclone intensity from INSAT-3D satellite images; the abstract reports no quantitative accuracy.
PostRainBench and CAMT post-process NWP for precipitation, with the first deep model beating NWP on heavy-rain CSI by 15.6–31.8%.
Compares deep learning architectures for European precipitation, including extremes; a two-level U-Net matches deeper versions at lower cost, with layer-wise relevance attribution.
Machine learning on GCM and ERA5 environmental variables identifies humidity convergence and CAPE as key predictors of tropical cyclone intensification rates.
Over-parameterized neural networks predict GPM radar tropical Pacific rain-rate distributions from MERRA-2 profiles, capturing heavy tails better than other machine learning methods.
A deep generative nowcasting model with a balanced loss and temperature input extends skillful heavy-rain nowcasts over the Netherlands to 60-minute lead times.
EarthFormer transformer forecasts Palmer Drought Severity Index up to six months ahead, while ConvLSTM performs better at longer lead times, using open monthly climate data.
Machine learning attributes the 2022 China compound drought-heatwave forest productivity decline mainly to soil moisture rather than VPD, identifying a tipping point.
Reviews seasonal and multi-annual tropical cyclone forecasting, including emerging ML/AI methods, ENSO-neutral predictability and landfall skill gaps relative to user needs.
Sub-seasonal to seasonalNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCARMet OfficeMétéo-France (incl. CNRM)Colorado State University (incl. CIRA)China Meteorological Administration (CMA)Japan Meteorological Agency (JMA)Bureau of Meteorology (Australia)
LightGBM with aerosol features and GLM satellite data nowcasts hourly CONUS lightning at 0.25°, reaching 94.3% accuracy and 75% probability of detection.
RRDBNet residual-in-residual dense network statistically downscales Yellow River precipitation and improves extreme precipitation over regression and deep baselines.
LightGBM with SHAP explanations predicts tropical cyclone disaster loss grades in China with 0.86 accuracy, with rainfall and wind as the main drivers.
Applications & impactsNanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
XGBoost trained on dual-polarized Sentinel-1 SAR collocated with SFMR and SMAP retrieves tropical cyclone surface winds with 2.53 m/s RMSE up to 70 m/s.
Machine learning seasonal prediction of summer extreme precipitation frequency over Southwest China; classified from the title only, as the abstract is missing.
Sub-seasonal to seasonalInstitute of Atmospheric Physics, Chinese Academy of Sciences
A deep neural network predicts typhoon-caused building loss ratios in southern Korea, outperforming multiple regression on MAE, RMSE and R2.
ML-surrogate Sobol analysis of WRF physics parameters for southeast Australian heat extremes finds only three of 24 parameters matter.
Compares ARIMA, exponential smoothing, MLP and Gaussian processes for extreme temperature in five US cities; MLP performs best and projects temperatures to 2030.
Multi-proxy drought atlas reconstructs southwest Pacific drought back to 1640 and uses Isolation Forest to flag extreme El Niño-linked droughts.
Few-shot transfer learning with a novel Hough-space SR loss segments flight contrails in satellite imagery for climate-impact monitoring.
A U-Net ConvLSTM nowcasts near-global IMERG precipitation every 30 minutes up to 4 hours; a classification-loss variant better predicts rates above 8 mm/hr.
Presents a RetinaNet detector, labeled dataset and annotation tool for mesoscale convective systems in Meteosat SEVIRI imagery, achieving mAP 0.75.
NowcastNet unifies advection physics with learned conditional generation for extreme precipitation nowcasts up to 3 hours, ranked first by meteorologists in 71% of cases.
NowcastingChina Meteorological Administration (CMA)Tsinghua University (NowcastNet)University of California, Berkeley
Physics-constrained deep learning (CGsNet) nowcasts convective wind gusts from radar at 0.01° and 6-minute resolution for 0-2 h, beating NWP-based nowcasting.
Neural networks classify US West Coast atmospheric river landfalls back to 1916, and tree-ring regression extends annual AR counts to 1400 CE.
Climate analysis & attributionColumbia University (LEAP, Lamont-Doherty)University of California, Los Angeles (UCLA)
Teleconnection-driven vision transformer forecasts global burned area up to four months ahead, with gains growing at longer lead times using climate indices.
Review of ML for climate variability modes, extreme weather precursors, observation-model integration, downscaling and bias correction, aimed at domain scientists.
Climate analysis & attributionNSF NCAR / UCARUniversity of California, BerkeleyLawrence Berkeley National Laboratory / NERSCOak Ridge National Laboratory
Backpropagation ANN statistically downscales ENSO and NOAA indices to monthly rainfall in Bondowoso, Indonesia, to drive EDI drought assessment with R2 of 0.97.
Transformer detects concentric eyewalls in passive microwave imagery, producing a 30-year global dataset of 377 events with 92.9% precision and recall.
Observations & remote sensingTsinghua University (NowcastNet)
A neural network with Shapley values on ERA5 predictors explains 70% of European heatwave variability, with upper-level geopotential the largest contributor at 27%.
SOM-CNN-LSTM post-processing of 15-day precipitation forecasts in the Huaihe basin improves correlation by up to 32.3% and reduces RMSE by 26.6%.
Machine-learning downscaling of CORDEX projections to hourly rainfall over Spain projects intensified sub-daily extremes and 100-year events, increasing flash flood risk.
A random forest driven by weather-station data predicts wet-snow avalanche activity in the Swiss Alps, with F1 near 0.8 in nowcast and NWP-driven 24-hour modes.
FlashBench combines a LightNet-style deep model with observed field evolution for regional lightning nowcasts over West India, delivered via a cloud forecast system.
A Wasserstein GAN downscales tropical cyclone rainfall from about 100 km to 10 km, reproducing spatial structure and spectra with mean bias within 5% and extrapolating to extreme storms.
An attention-based GAN reconstructs low-quality SAR wind speeds in tropical cyclones, halving RMSE versus the existing SAR product in affected regions relative to ECMWF.
Unpaired diffusion bridges downscale geophysical fluid simulations without paired data, correcting biases including in extremes and promising use for climate precipitation and temperature.
Adversarially trained 3D LSTM nowcasts sudden localized heavy rain from MP-PAWR radar with 10-min lead, outperforming an advection model and JMA's operational system at onset.
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.
Proposes compact U-Nets with Self-ONN layers for satellite active fire detection, reporting superior detection with reduced computational complexity.
A physics-informed recurrent deep learning model super-resolves regional climate precipitation from 2.5 km to 250 m, producing realistic intense short-duration events over Montreal.
Downscaling & super-resolutionEnvironment and Climate Change Canada (ECCC)
Random forest with SST-based predictors forecasts monthly summer extreme precipitation in the Yangtze basin with R2 above 0.81 and physical explanations.
Dual-polarization radar nowcasting of severe convective weather combines physical constraints with a deep neural network; abstract unavailable, so details are limited.
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.
Machine-learned correction operator, trained on nudged coarse E3SM runs against ERA5, makes free-running climate statistics and extreme-event frequencies resemble observations.
Climate emulationMassachusetts Institute of Technology (MIT)
SeaUnet, a fully convolutional network trained on CMIP TC tracks, gives skillful seasonal forecasts of Western North Pacific typhoon track density up to four months ahead.
Sub-seasonal to seasonalTsinghua University (NowcastNet)University of ReadingChina Meteorological Administration (CMA)
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.
Curates an hourly-interpolated satellite image and wind-speed dataset for tropical cyclones, plus a community competition whose winners improved on the benchmark by about 50%.
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.
Applies unpaired UNIT image-to-image translation, combined with quantile mapping, to bias-correct HadGEM3 monsoon simulations toward ERA5, preserving cross-variable correlations and spatial structure.
Tuned fusion deep learning model forecasts tropical cyclone passive microwave rainfall images from satellite data, with no quantitative results reported in the abstract.
CNN and ML models predict August fortnight air temperature in Paris and Córdoba months ahead, targeting heatwave-related signals; skill is reported as very good.
Weighted U-Net corrects ECMWF-IFS precipitation over China, improving heavy-rainfall forecast skill over the ordinary U-Net, with gains growing at longer lead times.
Post-processing & calibrationInstitute of Atmospheric Physics, Chinese Academy of SciencesChina Meteorological Administration (CMA)
TC-Resnet estimates tropical cyclone 34-kt wind radii from infrared and microwave satellite data with MAE of 11.3 nmi and r of 0.907.
YOLOv3 detects tropical cyclones in IASI infrared sounder radiances over the North Atlantic, a proof of concept with AP 78.3% at IoU 0.1.
Localized statistical retrievals from IASI sounder data reduce dampening and inflating errors in extreme land surface temperature retrievals, from 1.9 K to 1.6 K.
Explainable neural network on 100 MPI Grand Ensemble members identifies spring tripolar North Atlantic SST patterns as precursors of European summer heatwaves.
Climate analysis & attributionMax Planck Institute for MeteorologyColorado State University (incl. CIRA)
Enhanced multi-scale ConvLSTM network predicts typhoon cloud images three hours ahead from satellite sequences with sharper detail than prior methods.
Linear and random forest models forecast central European summer heatwaves 1–6 weeks ahead, matching ECMWF hindcast ensemble mean beyond two weeks lead.
Sub-seasonal to seasonalETH ZurichMassachusetts Institute of Technology (MIT)
CNN and GRU models forecast East China rainfall anomalies at 10–30 day leads, beating CMA and ECMWF dynamical models in skill and extreme-event scores.
Sub-seasonal to seasonalNanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
Compares regression models and ensembles on NASA POWER climate data for precipitation forecasts in southern India, with ensemble E-MLR-RFR performing best.
U-Net trained on WRF rainfall outputs cuts district-scale heavy rainfall MAE by 64.78% over Assam and lifts categorical accuracy to 77.9%.
TS-RainGAN splits heavy-rainfall nowcasting into coverage and intensity GAN modules, reducing blurriness and giving skillful district-level forecasts up to 2 hours ahead.
XGBoost used for both downscaling and calibration of satellite precipitation during typhoons, better reproducing extreme rainfall spatial variability than single-step alternatives.
Releases LSDSSIMR, a 2020–2022 dataset of FY-4A satellite dust images at 15-minute, 4 km resolution, paired with reanalysis for dust storm prediction.
DenseNet-based fusion of infrared and water vapor satellite images, with a tropical-cyclone-specific loss, improves TC center location accuracy.
Observations & remote sensingChina Meteorological Administration (CMA)
Attention network with Kalman filter estimates tropical cyclone intensity from multispectral infrared images, cutting RMSE from 9.79 to 7.82 knots.
A DyConvGRU and U-Net forecast-refinement network with WGAN training improves high-reflectivity radar echo extrapolation over Shanghai, raising POD and CSI.
Reviews feature selection methods such as LASSO and causal selection for climate-economy studies, and applies them to find an optimal heatwave definition for economic outcomes.
Super-resolution deconvolution network with orography downscales hourly precipitation from 100 to 12.5 km over Australia, improving fine-scale extremes over DeepSD.
Two-stage deep learning predicts typhoon satellite images and converts them to rainfall maps, reaching correlation 0.81 at 3 hours ahead.
Compiles and proposes a searchable web platform of ML benchmark datasets for natural disasters, organized by disaster management cycle. Mostly a catalogue, low novelty.
Ranks dual-polarimetric radar variables for predicting heavy rainfall from mesoscale convective systems over Korea and uses them in ML early-warning models.
Random forest fusing MODIS, GPM and GLDAS factors reproduces the SPEI drought index for Shandong, China, from 2002-2020, outperforming XGBoost and SVM.
Fine-tuned BERT recognizes seven drought impact types in tweets, reaching 0.89 macro-F1 on the Drought Impact Report test set.
An LSTM encoder-decoder nowcasts extreme rain and wind over Malpensa airport for one hour, reaching over 90% wind detection probability with station and GNSS inputs.
Dense neural networks retrain passive microwave rainfall and snowfall retrievals from GPM and CloudSat, detecting precipitation with over 0.95 probability and showing large global snowfall uncertainty.
A physics-statistical GAN downscales 0.25° climate model rainfall to 0.01° fields, quantifying uncertainty and matching observed extreme rainfall distributions.
Downscaling & super-resolutionMassachusetts Institute of Technology (MIT)
Diffusion models generate realistic high-resolution UK rainfall samples conditioned on low-resolution simulations, matching the high-resolution model's quantiles and time-mean.
A deep convolutional network classifies satellite cloud images from MOSDAC over India with 94% accuracy, aimed at tropical cyclone prediction.
Deep-learning weather typing analyses interdecadal and interannual variability of autumn extreme rainfall in Taiwan; abstract unavailable, scope inferred from title.
Hybrid dynamic model and LSTM aim to improve seasonal drought prediction over China; the abstract is missing, so scope is inferred from the title.
Transformer produces probabilistic sub-seasonal forecasts of extreme high temperatures over eastern China; abstract unavailable, so scope is inferred from the title.
A simple CNN with tailored loss functions outperforms regression for downscaling New Zealand rainfall, raising wet-day explained variance from 0.35 to 0.52.
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.
XGBoost with SHAP links SPI drought indices to text-based drought impacts in Texas across agriculture, fire, and public health categories.
Unsupervised clustering of storm-scale climate model output shows extreme precipitation changes driven mainly by shifts in storm regimes rather than precipitation efficiency.
Deep learning fuses radar, lightning, satellite, NWP and terrain data to probabilistically nowcast lightning, hail and heavy rain at 1 km and 5 min up to 60 min.
HFIT statistically fits NWP track forecasts per lead hour, cutting 2014–2021 Atlantic hurricane track errors 8% at 24 h versus the NHC official forecast.
A GAN-enhanced FourCastNet better captures fine-scale structure and extreme global precipitation percentiles, matching NWP skill at 1-2 day leads.
Global medium-range NWPUniversity of California, BerkeleyLawrence Berkeley National Laboratory / NERSC
Multi-resolution deep learning debiases GCM tail statistics against reanalysis and reconstructs finer scales, producing realistic Western European weather-extreme realizations for risk assessment.
Downscaling & super-resolutionMassachusetts Institute of Technology (MIT)
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.
A linear inverse model predicted the February 2021 cold air outbreak four weeks ahead, up to two weeks earlier than numerical models, and estimated it as a top-1% event.
Sub-seasonal to seasonalNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Evaluates the ANN-based PDIR-Now near-real-time satellite rainfall product during California atmospheric river events, validated against profilers and disdrometers.
Backpropagation neural network corrects underestimated high winds from HY-2B/C/D scatterometers in tropical cyclones, cutting RMSE by 20.4% to 3.40 m/s.
ML exospheric temperature model calibrates NRLMSIS 2.0 against satellite density, cutting model-satellite differences by 25% with calibrated uncertainty intervals.
Machine learning ensemble of 27 CMIP6 models improves China extreme precipitation simulation and projects increases of up to 13.7% by late century.
Post-processing & calibrationChina Meteorological Administration (CMA)
MESMER-X statistical emulator reproduces spatially resolved annual maximum temperatures of Earth system models using global mean temperature as the driver.
TSGAN uses spatio-temporal GAN with attention for radar nowcasting of strong convection, giving sharper forecasts than optical flow and deep baselines; it is reported as used in an operational system.
Random forest predicts typhoon economic losses for Korea by region, projecting 6.5–14.8 times higher losses by 2100 under IPCC scenarios.
TICAENet, a LeNet/VGG-style CNN, classifies typhoon intensity from geostationary satellite images at 97.12% accuracy and estimates wind speed with 4.78 m/s MAE.
Deep feedforward networks with LIME explanations assess building-level hurricane wind and surge risk using updated forecasts for a Louisiana case.
Random forest on three preceding atmosphere-ocean indices gives a seasonal early warning of Yangtze summer extreme precipitation, including the 2020 event.
Sub-seasonal to seasonalInstitute of Atmospheric Physics, Chinese Academy of Sciences
ANN-based postprocessing of GEFS forecasts with short 60-day training windows improves probabilistic 24-h precipitation forecasts over CONUS, especially for heavy rainfall thresholds.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Random forest classifiers predict Swiss avalanche danger levels from 20 years of station weather and snow simulations, reaching 72-78% accuracy.
Applications & impactsETH ZurichSwiss Data Science Center (SDSC)
Hybrid deep learning model gives multi-step-ahead forecasts of daily maximum temperature and heatwaves; no abstract was provided, so skill figures are unknown.
A Conv-LSTM encoder-decoder with rain-weighted MSE extrapolates radar echoes for 2 h, reaching 0.42 CSI at 30 dBZ for 1 h, above CTREC.
LightningCast uses a CNN on GOES-16 ABI imagery to predict next-hour lightning over CONUS, giving about 20 min or more of lead time and transferring to GOES-West.
A graph-guided ConvLSTM nowcasts precipitation from tropical cyclones, combining spatio-temporal graph structure with convolutional recurrent modelling.
GAN and VAE-GAN downscale low-resolution forecasts to radar-scale precipitation maps with CRPS matching state-of-the-art pointwise post-processing, including heavy rain.
PSO-optimized Bi-LSTM with VMD decomposition predicts typhoon wind speed at offshore wind farms; abstract unavailable, so accuracy is unreported.
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.
Climate analysis & attributionNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCARPrinceton UniversityNanjing University of Information Science & Technology (NUIST)
Deep learning detects polar lows in Sentinel-1 SAR images with F1 0.95, using a new public dataset labelled from ERA5; higher resolution improves detection.
A deep CNN identifies tropical cyclone fingerprints in satellite cloud images over the Northwest Pacific with 96% accuracy, with heat maps showing eye and rainband focus.
Spatially enhanced loss functions using neighbourhood and spectral filters improve neural network convection forecasts, with spectral-filter models giving better-calibrated probabilities than pixelwise training.
Benchmarks, datasets & evaluationColorado State University (incl. CIRA)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
A recurrent-convolutional network nowcasts lightning occurrence over Switzerland for 60 minutes, reaching CSI 0.75 at 5 minutes and 0.32 at 60 minutes.
Unsupervised clustering and cyclone tracking identify four weather patterns behind middle tropospheric cyclones that drive extreme rainfall over western India.
Tests seven optimizers and eight learning rates for a CNN forecasting hourly typhoon rainfall 1–6 h ahead; moderate rates (0.1–0.001) work best.
NowcastingNanjing University of Information Science & Technology (NUIST)
Gradient-boosted nowcasts of reflectivity, lightning and hail-related echo tops show radar is the most useful input and satellite helps across all hazards.
A deep network with label deformation loss detects and classifies synoptic fronts in ERA5 over Europe and North America, with CSI above 62.9%.
CLSTM-LFN video-prediction model fuses lightning history with NWP variables to forecast 0-3 h cloud-to-ground lightning areas, outperforming parameterization schemes.
Skill-score-weighted ensemble of MLR, SVR, GRU and CNN post-processors corrects LDAPS next-day maximum temperature, with CNN reaching 1.41 °C RMSE in South Korea.
A CNN with multi-task learning on geostationary IR images estimates western North Pacific tropical cyclone RMW and R34 with correlations of 0.95 and 0.93.
TCR-GAN converts geostationary infrared images of tropical cyclones into passive microwave rainfall fields and releases a new benchmark dataset (TCIRRP).
A neural network corrects empirical typhoon model wind fields toward JMA mesoscale model output, enabling more realistic storm surge simulation for Typhoon 10 (2020).
SVM and general regression neural network models estimate western North Pacific tropical cyclone size parameters (RMW, R34, R50, R64) from infrared satellite imagery.
Observations & remote sensingChina Meteorological Administration (CMA)
Applies machine learning to nowcast extreme weather using different input datasets; no abstract available, so details are inferred from the title.
Applies deep learning to post-process ECMWF forecasts of heavy rainfall, correcting NWP output for extreme precipitation events.
ANFIS models forecast typhoon rainfall at a Tamsui station up to three hours ahead and are packaged in a mobile app, with best RMSE of 7.9 mm.
Local & station forecastingMicrosoft Research (AI for Science)
ResDeepD, a residual super-resolution CNN, downscales 1° Indian monsoon rainfall to 0.25° and outperforms SRCNN, DeepSD and Nest-UNet, including during cyclonic storms.
A self-attention GRU nowcasts radar echoes for landfalling hurricanes and extreme convective rainfall in the southeast US, aiming to improve skill beyond the first hour.
NowcastingNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)Colorado State University (incl. CIRA)
DMSF-GAN extrapolates radar images for 0-6 h typhoon nowcasting and outperforms optical-flow and RNN baselines; the abstract gives no quantitative results.
A residual deep network fusing SAR backscatter, texture and morphology retrieves hurricane sea-surface winds up to 75 m/s with RMSE 1.72 m/s.
TCR-GAN generates passive microwave rainfall images from TC infrared imagery and releases the TCIRRP benchmark dataset for IR-to-rainfall prediction.
VIIRS nightlight data and simulated hurricane weather feed machine learning that predicts Hurricane Maria power outages in Puerto Rico where outage records are missing.
Machine learning post-processing of subseasonal-to-seasonal forecasts improves prediction of extreme precipitation occurrence over the contiguous US.
Sub-seasonal to seasonalNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Causal-predictive ML links tropical cyclone exposure to mortality and hospitalization in Medicare recipients, finding substantial respiratory hospitalization increases with wind speed as main driver.
Applications & impactsColumbia University (LEAP, Lamont-Doherty)Colorado State University (incl. CIRA)
Machine learning uses typhoon meteorological and grid data to quantify fault risk and generate N-m contingency plans for power grid dispatch, aimed at early warning.
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 multitask CNN estimates global tropical cyclone precipitation from HURSAT-B1 imagery, reaching 0.68 detection probability and beating MLR and random forest baselines.
Observations & remote sensingNanjing University of Information Science & Technology (NUIST)
A smoothed CNN classifies six anticyclonic circulation types linked to European drought and heat, to help identify their drivers in climate simulations.
Bias-corrected super-resolution CNN downscales six GCMs' daily precipitation and temperature from 1.25° to 0.25° over India, targeting extremes better than SRCNN and LSTM.
An XGBoost model post-processes ECMWF EPS outputs to correct systematic negative biases in tropical cyclone maximum wind forecasts over the western North Pacific, 2016–2019.
An artificial neural network disaggregates 3-hourly rainfall to hourly extreme rainfall for Cambodia using ERA5 covariates, reaching R2 of 0.93.
A hybrid MCEEMD decomposition and RBF neural network forecasts Taipei daily maximum temperature 7 days ahead, aimed at heatwave risk under warming.
InceptionV3, ResNet50 and VGG models classify 9081 optical satellite images into five weather events, with InceptionV3 reaching 92% average accuracy.
Deep ensemble learning on radar reflectivity videos issues severe thunderstorm warnings for Liguria, Italy, with a value-weighted skill score for probabilistic-to-binary alerts.
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.
Deep networks trained on an NWP ensemble with a frequency-weighted loss correct rainfall, improving extreme-event forecast skill by factors of two to above six.
Post-processing & calibrationPotsdam Institute for Climate Impact Research (PIK)Technical University of Munich (TUM)
Uses machine learning to model the economic impact of incoming tropical cyclones. The abstract is missing, so this classification is low-confidence.
A CNN identifies 91% of Midwest extreme precipitation events from circulation patterns and finds increasing frequency of the associated patterns over two decades.
Temporal Fusion Transformer forecasts weekly extreme precipitation quantiles up to six months ahead, beating calibrated SEAS5 at six-month lead and modestly beating climatology.
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.
Climate analysis & attributionUniversity of California, Los Angeles (UCLA)
Modular fully convolutional networks use radar echoes to forecast hourly typhoon rainfall 1 to 6 h ahead in southern Taiwan, outperforming an MLP baseline.
An end-to-end spatiotemporal scale-selection model improves ECMWF precipitation bias correction, especially for heavy rain, beating 11 published methods on two benchmarks.
Neural network uses satellite storm observations and ensemble NWP to predict convection 36 h ahead, yielding a probabilistic indicator that beats existing air traffic convection indicators.
FOCUSED corrects short-term NWP wind speed forecasts for dangerous bora gusts on the Croatian Krk bridge, outperforming ANN, random forest and SVM baselines.
SRDRN residual CNN downscales daily precipitation and temperature from 100 km to 4 km, reproduces extremes, and transfers across regions.
Random forest fills cloud-caused gaps in satellite land surface temperature during heatwaves using land cover, elevation and vegetation indices.
Regression and neural network models using Jason altimeter ocean parameters estimate tropical cyclone 10 m winds up to 35 m/s with 1 m/s accuracy.
Random forests combining GOES-16 satellite, lightning and radar fields improve severe storm warning prediction over radar alone, identifying cloud-top divergence as most informative.
NowcastingNASAColorado State University (incl. CIRA)University of Oklahoma (AI2ES)
Segmentation neural network identifies dry intrusion origins from ERA-Interim geopotential height alone, enabling detection in climate datasets lacking wind fields.
GRU models using radar reflectivity and station data forecast typhoon rainfall 1, 3 and 6 hours ahead in southeastern Taiwan, outperforming XGBoost and linear regression benchmarks.
Volume-to-point machine learning maps gridded QPESUMS radar sequences to Taiwan station rainfall, performing comparably to the operational scheme and excelling in heavy-rain detection.
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 forests generate gap-free daily 100 m and 500 m NDVI and land surface temperature products to support urban heatwave and heat-island assessment.
A modified fuzzy inference model gives probabilistic typhoon rainfall forecasts 1–3 hours ahead, with median-based deterministic forecasts performing satisfactorily.
Neural networks combine RAP forecasts, surface analyses and GOES-16 satellite data to correct temperature and moisture profiles for severe weather nowcasting.
Thesis develops learned pattern recognition for atmospheric rivers (topological data analysis plus ML) and blocking (deep CNNs) in climate simulations and reanalysis.
A machine-learning bias correction for a 25-km regional climate model raises simulated violent typhoon frequency from zero to match observations and projects increases under warming.
Post-processing & calibrationColumbia University (LEAP, Lamont-Doherty)
Quantile regression forest fuses 691 crowd-sourced citizen weather stations with open data to map Zurich urban temperature at 10 m resolution during a 2019 heatwave.
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.
A super-resolution CNN enhances fire risk maps from climate models, outperforming interpolation at 4x and 8x and projecting risk for 2040 and 2100.
Random forests, gradient-boosted trees and logistic regression calibrate NOAA Warn-on-Forecast ensemble guidance for tornado, hail and wind, beating updraft-helicity baselines.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Deep CNN approach detects polar mesocyclones and polar lows in satellite mosaics, addressing class imbalance and scale filtering for satellite data.
PERSIANN-CWB uses a deeper convolutional autoencoder CNN on Himawari-8 infrared and water vapor channels, improving rainfall retrieval over Taiwan relative to the prior CNN version.
Wavelet neural network forecasts drought indices for Fiji tropical stations, applying a hybrid wavelet-ANN approach to seasonal drought prediction; abstract not provided.
Applies GANs and deep belief networks to improve ECMWF short-term intense rainfall forecasts; the abstract is missing, so reported gains are unverified.
Post-processing & calibrationChina Meteorological Administration (CMA)
A Faster R-CNN detects global cyclones and anticyclones from ASCAT scatterometer wind fields, reducing false alarms and tracking Hurricane Irma.
Machine learning identifies three-dimensional drought events in tree-ring reconstructions and CMIP simulations, suggesting models underestimate the range of future hydroclimate states.
Climate analysis & attributionNSF NCAR / UCARColumbia University (LEAP, Lamont-Doherty)
Random forests and gradient-boosted regression trees using HRRR forecast features improve upper-level aviation turbulence (EDR) forecasts and reduce false alarms versus regression baselines.
TRU-NET, a convolutional-recurrent U-Net with cross attention, predicts high-resolution rainfall from climate model fields with lower RMSE than a dynamical model's rainfall output.
Uses deep neural networks and multiple linear regression on Spark-based big-data infrastructure to predict typhoon rainfall at four Taiwan stations, with faster computation than standalone systems.
Compares calibration of heat index ensembles, finding direct GEV-based EMOS gives the best skill up to day 6; machine learning performs comparably only for warning categories.
Kernel k-means clustering of German station series yields robust regions, enabling a nonstationary model of summer temperature extremes above 35 °C.
Random forests convert convection-allowing ensemble forecasts into day-1 severe weather probabilities, beating calibrated updraft helicity and SPC outlooks in Brier skill, especially for wind and hail.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Applies artificial neural networks to predict drought (SPI at 3 to 48 months) in Tabriz, Iran, using station precipitation and other hydro-meteorological variables from 1992 to 2010.
Neural networks project future hurricane parameters under RCP8.5 for southeastern US coast, showing economic losses rising sharply under climate change.
Grad-CAM CNN classifies typhoon-damaged areas in Chiba aerial photos, producing explainable damage maps to support faster disaster response.
Random forest and CNN framework corrects GPM DPR orographic precipitation retrievals using HRRR, cutting false alarms by 77% over the Southern Appalachians.
Stacking TrajGRU ensemble members with orographic features via a CNN roughly doubles skill on extreme rain rates, matching Lagrangian persistence on TAASRAD19 radar data.
Distills the analog ensemble post-processing into a small neural network, enabling fast on-the-fly ECMWF lower-stratosphere wind corrections with a model of only hundreds of kilobytes.
ConvLSTM with star-shape bridge and rain-oriented loss improves radar precipitation nowcasting over convolutional LSTM baselines and extrapolation.
DeepCTC neural network classifies cloud types from GOES-16 ABI imagery using CloudSat labels and evaluates PERSIANN-CCS precipitation by cloud type against MRMS.
XGBoost corrects KMA quantitative precipitation forecasts for heavy rain in Seoul, cutting MAE by up to about 30% at some sites.
Random forest automates quality control of tail Doppler radar data from Hurricane Michael's rapid intensification, classifying over 99% of radar gates correctly.
Two-step deep detection and segmentation locates tropical cyclone centers in satellite infrared images with about 0.237° average position error.
Clusters lightning-related utility outages and predicts daily outage groups from radar thunderstorm forecasts, handling class imbalance.
ML-tuned blending of multiple ensembles gives probabilistic thunderstorm forecasts from 9 hours to 4 days, clearly better than raw ensemble output.
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.
A UNet++/ResNet-inspired CNN on satellite images predicts lightning and thunderstorms 15 minutes ahead with over 94% probability of detection.
Random forest postprocesses HREFv2 ensembles for hail, with isotonic calibration matching SPC outlook reliability for operational day-1 forecasts.
Post-processing & calibrationNSF NCAR / UCARUniversity of Oklahoma (AI2ES)
A neural network filters a 50-member regional climate ensemble to show Vb cyclones shift from summer to spring under RCP8.5.
Evaluates a neural-network hurricane impact-level forecast over 2017 storms, finding best accuracy about 30 h before landfall and wind speed as the key driver.
Applications & impactsColorado State University (incl. CIRA)
Machine learning nowcasts lightning up to 30 minutes ahead from four surface variables at single stations, beating persistence, electric-field and CAPE baselines.
Synthesizes ML interpretation methods such as saliency maps and permutation importance, demonstrated on tornado, hail and convective-mode applications.
Explainability, uncertainty & physicsNSF NCAR / UCARNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Nowcasts convective initiation 17-40 minutes ahead from FY-4A satellite data using optical flow and BP-AdaBoost, reaching detection probability 0.80 in South China.
Deep recurrent networks in a two-step model predict typhoon wind speed and wave height at a Taiwan buoy, beating a one-step wave-height baseline.
A hybrid 3D CNN and ConvLSTM encoder-decoder fuses radar and 3D re-analysis fields to nowcast convective storms, outperforming extrapolation baselines.
Deep learning forecasts 300 m leaf area index 10 days ahead over Europe using weather inputs, aiming at heat-wave crop warnings; weather prediction itself is not ML.
A CNN postprocesses GFS integrated vapor transport forecasts, cutting RMSE by 9-17% from 3 hours to 7 days and improving atmospheric river forecasts over the western US.
Real-time deep learning detector locates typhoon eyes in images for meteorological information; the abstract is unavailable, so accuracy is unknown.
Observations & remote sensingNanjing University of Information Science & Technology (NUIST)
Two-stage GANs generate synthetic radar reflectivity sequences for extreme precipitation in Shenzhen to address class imbalance in radar QPE.
Tracks radar-identified convective storm cells and classifies their power-outage damage potential with random forests and deep networks on imbalanced data.
Regression and ANN models forecast Vietnam East Sea tropical cyclone counts six months ahead; multiple linear regression performed best on independent data.
Neural network selects physically constrained precipitation events to make dynamical downscaling of climate simulations more efficient; abstract unavailable.
Machine learning temporally downscales climate projections to estimate short-duration extreme precipitation IDF curves, trained on about 2,000 US stations.
NARX network fuses GNSS precipitable water and station data to forecast hourly intense rain in Lisbon, cutting false-positive rate to 21–36%.
An MLP neural network forecasts monthly SPEI drought index at 17 Pakistani climate stations, showing usable skill by MAE, RMSE and correlation.
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.
Combines optical flow with machine learning for nowcasting and early warning of strong convective weather; abstract not available.
Develops and interprets deep learning models for nowcasting convective hazards, with attention to model explainability.
Applies machine learning to an early warning system for very short-term heavy rainfall; abstract unavailable, classification based on title only.
Combines MOS downscaling of GCM rainfall with ANN temporal disaggregation to project IDF curves for Peninsular Malaysia under RCP4.5 and RCP8.5 with uncertainty bounds.
An ANN fuses HY-2A radiometer and scatterometer data to retrieve typhoon wind fields, outperforming operational HSCAT winds against GRAPES and H*Wind references.
Linear regression and a feedforward network predict Chilean agricultural productivity drought anomalies 1–6 months ahead using MODIS, rainfall and ENSO indices.
An ANN combined with a particle filter retrieves ocean surface wind speed from BeiDou GEO GNSS reflections, reaching about 1.9 m/s RMSE against buoys.
A CNN stratifies convection-permitting climate runs and finds too few quasi-linear convective systems in the Midwest, with future declines not spatially significant.
Zero-inflated neural networks predict tornado property damage occurrence (AUROC 0.872) and amount (R² 0.432), producing gridded monthly damage probability maps.
Compares ELR, ZAGA and quantile regression forests for postprocessing HARMONIE-AROME hourly precipitation; QRF is most skillful for high thresholds.
Post-processing & calibrationKNMI (Royal Netherlands Meteorological Institute)
Random forests and neural networks predict grid power outage damage from convective storm cells, using SMOTE to handle rare extreme events.
Random forests trained on 11 years of GEFS reforecasts give better CONUS extreme precipitation exceedance forecasts than raw ensembles and ECMWF.
Post-processing & calibrationColorado State University (incl. CIRA)
Backpropagation neural networks forecast typhoon wind speeds at three western Taiwan stations, outperforming linear regression across different typhoon tracks.
Local & station forecastingUniversity of California, Los Angeles (UCLA)
Abstract missing; applies topological data analysis with machine learning to classify atmospheric river patterns in large climate datasets.
Climate analysis & attributionLawrence Berkeley National Laboratory / NERSC
Neural network parameterizations of typhoon-induced sea surface cooling reduce maximum wind intensity errors by 60–70% in four atmosphere-only typhoon forecast cases.
Hybrid physics-ML & parameterizationMassachusetts Institute of Technology (MIT)China Meteorological Administration (CMA)
Shape and motion features with a trained classifier detect comma-shaped cyclone-associated cloud patterns in satellite imagery to support severe weather forecasting.
Machine learning applied to weather radar data for severe weather nowcasting; abstract unavailable, so details are inferred from the title only.
