Forecasts from 2 weeks to about a year: sub-seasonal, seasonal and ENSO/MJO prediction.
Search and filter this topic Plain list on GitHub
ACE2 produces skilful winter seasonal forecasts from SST-persisted initialisation, reaching NAO correlation 0.47 over 1993–2016, comparable to a physics-based model.
Machine learning S2S model outperforms ECMWF's S2S system for precipitation and OLR up to 42 days, extending MJO prediction skill from 30 to 36 days.
China Meteorological Administration (CMA)Fudan University (FuXi)
3D-Geoformer, a self-attention transformer, predicts Niño3.4 SST anomalies 18 months ahead with high correlation skill for ENSO.
Nanjing University of Information Science & Technology (NUIST)
A CNN-based cubed-sphere ensemble of 32 deep-learning weather models gives 2-6 week forecasts that are only modestly worse than the ECMWF S2S ensemble over land.
Deep learning model trained on climate model simulations produces multi-year ENSO forecasts, a landmark result for long-lead ENSO prediction.
Nanjing University of Information Science & Technology (NUIST)
ACE2 produces skilful winter seasonal forecasts from SST-persisted initialisation, reaching NAO correlation 0.47 over 1993–2016, comparable to a physics-based model.
Machine learning S2S model outperforms ECMWF's S2S system for precipitation and OLR up to 42 days, extending MJO prediction skill from 30 to 36 days.
China Meteorological Administration (CMA)Fudan University (FuXi)
A CNN with cosine-distance loss predicts equatorial Pacific SST zonal patterns up to 16 months ahead, outperforming dynamical models, with physically interpreted ENSO precursors.
Nanjing University of Information Science & Technology (NUIST)
Machine learning models forecast precipitation from teleconnection indices across different precipitation regimes, an approach for seasonal-scale rainfall prediction.
Wavelet-decomposed deep learning forecasts monthly Indian precipitation from lagged rainfall and teleconnections; beats single-scale models.
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.
An LSTM forecasts the Oceanic Niño Index 18 months ahead, with the extratropical western North Pacific predictor giving the best skill and beating linear regression.
Random Forest and BP networks with causally screened circulation predictors forecast western China precipitation 1, 3 and 6 months ahead, improving 1- and 3-month skill over an institute's forecasts.
Reviews seasonal and multi-annual tropical cyclone forecasting, including emerging ML/AI methods, ENSO-neutral predictability and landfall skill gaps relative to user needs.
NOAA (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)
Sampling-based CNOP method finds ENSO optimal precursors in the Zebiak-Cane model without an adjoint, matching adjoint results with parallelizable samples.
Deep echo state networks with penalized quantile regression forecast quasi-periodic ocean-driven multi-year climate events, with calibrated uncertainty from fast ensembles.
Machine learning seasonal prediction of summer extreme precipitation frequency over Southwest China; classified from the title only, as the abstract is missing.
Institute of Atmospheric Physics, Chinese Academy of Sciences
Neural networks with explainable AI identify decadal variability in tropically driven subseasonal precipitation skill in CESM2 and confirm the relationships in ERA5 reanalysis.
UNet-RegPre CNN predicts regional monthly precipitation in China with skill comparable to CFSv2 and identifies summer monsoon precursors.
STL decomposition with temporal convolutional networks forecasts the Niño index over multiple steps, capturing ENSO trend and peak intensity for 1992–2022.
A hybrid ODE and memory-kernel model using backtracked initial values reaches 24-month ENSO index lead predictions and about 18 months of verified skill.
Quantile regression forests and random forest classifiers give skillful 14–28 day probabilistic forecasts of central European winter 2-m temperatures and cold waves, beating climatology.
Multi-input autoencoder index from tropical Pacific SST gives Southern California precipitation predictability up to 4 months, beating Niño-3.4.
A convolutional neural network is reported to outperform dynamical models in predicting Central Pacific ENSO events.
Nanjing University of Information Science & Technology (NUIST)Institute of Atmospheric Physics, Chinese Academy of Sciences
Regression trees predict ENSO phase over a 12-month window using 1950–2022 data, reporting RMSE of 0.13 for 2022; evaluation design is unclear.
Teleconnection-driven vision transformer forecasts global burned area up to four months ahead, with gains growing at longer lead times using climate indices.
A convolutional GRU encoder-decoder forecasts ENSO-region SST maps and the Niño 3.4 index, beating LIM, analog and RNN baselines in correlation, RMSE and useful prediction range on climate model data.
Deep learning model combining tendency and attention mechanisms for ENSO forecasts; the abstract is missing, so skill figures are unavailable.
Logistic regression on 14 large-scale climate predictors gives tercile seasonal precipitation forecasts for the Kosi Basin at 1–3 month leads.
Uses atmospheric circulation patterns as inputs to ANN models (NARX, PLANN) for monthly to 9-month sea level anomaly forecasts, outperforming post-processed NWP and climatology at US sites.
LSTM forecasts wintertime air stagnation index in southern China from autumn Niño indices and past ASI, reaching correlation 0.778 at long lead.
Proposes an attention-enhanced ConvLSTM with genetic-algorithm model recombination to forecast ENSO using CMIP5/6 simulations. No abstract was provided.
Random forests combining natural and anthropogenic factors hindcast seasonal China temperature and precipitation with R² above 0.95, beating natural-only factors (0.86).
Autoencoder models predict US subseasonal temperature anomalies using CESM2 S2S variables to prepare transfer learning; skill varies by region and season.
NAO-MCD couples causal discovery with a deep learning air-sea model, outperforming numerical models for 1–6 month NAO seasonal forecasts, especially in winter.
Introduces an information-tracking calibrator that corrects error accumulation in decadal global land-surface temperature prediction, with predictions consistent with known teleconnections.
Random forest with SST-based predictors forecasts monthly summer extreme precipitation in the Yangtze basin with R2 above 0.81 and physical explanations.
Data-driven ML models forecast stratospheric sudden warmings better than ECMWF S2S beyond 25 days and boost ensemble skill by up to 20%.
A two-step U-Net (TU-Net) predicts 4-month-lead summer precipitation over the Yangtze and beats most NMME dynamical models on correlation and RMSE.
Nanjing University of Information Science & Technology (NUIST)
Probabilistic Output Extreme Learning Machine builds multi-model ensembles for Indian summer monsoon rainfall, aiming to avoid overconfidence and Gaussian assumptions. No quantitative skill score given.
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.
Tsinghua University (NowcastNet)University of ReadingChina Meteorological Administration (CMA)
XGBoost corrects FIO-CPS ENSO forecasts, raising Niño3.4 anomaly correlation at 7 to 13 month leads from 0.40 to 0.63 and cutting RMSE by 25%.
Zenodo dataset release accompanying an adaptive graph convolutional recurrent network for ENSO forecasting; the abstract only describes the data.
A spatio-temporal transformer with recurrent multi-step prediction uses SST and heat content anomalies to forecast ENSO 18 months ahead.
Physics-informed neural networks predict regional ENSO variables and aim to add physical interpretability; the abstract is unavailable, so results cannot be verified.
LSTM post-processing uses MJO and jet-stream quasi-periodic signals to weight operational S2S ensembles for Asian minimum 2 m temperature.
Nanjing University of Information Science & Technology (NUIST)
A residual network with automatically selected precursors predicts ENSO up to 8 months ahead, with a 6-month correlation of 0.82 beating the multi-model mean of 0.70.
Tsinghua University (NowcastNet)Nanjing University of Information Science & Technology (NUIST)Institute of Atmospheric Physics, Chinese Academy of Sciences
EWT-MF-LSTM avoids decomposition data leakage and forecasts Indian Summer Monsoon Rainfall with performance parameter 0.86 to 0.95 across phases.
3D-Geoformer, a self-attention transformer, predicts Niño3.4 SST anomalies 18 months ahead with high correlation skill for ENSO.
Nanjing University of Information Science & Technology (NUIST)
Ensembles of deep convolutional networks extend ENSO forecasts to 36 months ahead; the abstract is unavailable, so the claim rests on the title.
CNN ENSO forecasts from 5000 random-initialized members show western Pacific SST gives better Nino3.4 skill than eastern Pacific SST.
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.
A divided 3D CNN (TS-3DCNN) predicts the 2020–22 multi-year La Niña, comparable to a dynamical intermediate coupled model.
Nanjing University of Information Science & Technology (NUIST)
Linear and random forest models forecast central European summer heatwaves 1–6 weeks ahead, matching ECMWF hindcast ensemble mean beyond two weeks lead.
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.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
Seasonal CNN ENSO predictor with a modified loss function forecasts ENSO 18–24 months ahead, overcoming the spring barrier better than SINTEX-F2.
Interpretable deep learning ENSO model uses global ocean heat signatures to reach 22-month effective forecast length, with skill maintained using SST alone.
ANN and BRNN models with 12-month lagged climate drivers forecast monthly precipitation in Sindh, Pakistan, reaching correlations of 0.57 to 0.83.
RNN outperforms LSTM and GRU for S2S wet-season precipitation in Sichuan, reaching 88.95% trend sign accuracy and highlighting Niño index sensitivity.
Applies machine learning models to seasonal forecasting of winter precipitation over China; the abstract is missing, so skill results cannot be confirmed.
CNN segmentation models (UNet++, ResNet, PSPNet, DeepLabv3) trained on CMIP6 and fine-tuned on ERA5 forecast global temperature a month ahead, with MAE of 0.7.
A reservoir computing model forecasts the MJO real-time index for over a month, with the best cases reaching two months, beyond the expected predictability limit.
A U-Net corrects CFSv2 one-month-lead summer precipitation forecasts over China, cutting RMSE by about 43-50% on test data while leaving correlation skill largely unchanged.
MTL-NET multi-task deep learning predicts the Indian Ocean Dipole up to 7 months ahead, outperforming most dynamical models in hindcasts.
Nanjing University of Information Science & Technology (NUIST)Shanghai Artificial Intelligence Laboratory
Ridge, lasso, random forest and neural models forecast Horn of Africa long-rains precipitation; preselecting predictors on full data inflates skill, and honest tests fall below dynamical hindcasts.
WMO prize challenge winners used AI to post-process ECMWF S2S forecasts, beating bias-corrected references for weeks 3-6 precipitation and temperature.
Columbia University (LEAP, Lamont-Doherty)Environment and Climate Change Canada (ECCC)ECMWFMax Planck Institute for MeteorologySwiss Data Science Center (SDSC)
ENSO-GTC graph-based model with a teleconnection coupler reaches Niño3.4 correlation 0.51 at 18-month lead and above 0.6 out to 20 months with direct strategy.
Hybrid machine learning model that jointly performs classification and regression for short-term climate prediction; the abstract is missing, so details are inferred from the title.
Review of deep learning methods for ENSO analysis and prediction, covering evaluation metrics, the spring predictability barrier and prediction uncertainty.
CNN using ECMWF SEAS5 pressure and temperature forecasts beats the raw ECMWF output for monthly regional rainfall in Great Britain at 1 to 6 month leads.
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.
Survey of AI methods for ENSO forecasting, reviewing deep learning approaches since Ham et al. (2019) and comparing them with traditional statistical and dynamical forecasts.
Nanjing University of Information Science & Technology (NUIST)China Meteorological Administration (CMA)
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.
SHAP-explained SVM, RF, CNN and LSTM forecast climatic water balance 1-3 months ahead over Bangladesh; SVM performs best with antecedent CWB inputs.
Nanjing University of Information Science & Technology (NUIST)
Deep learning and classical ML with data reduction predict August fortnight air temperatures in Paris and Córdoba months ahead, using reanalysis inputs.
An artificial neural network with ENSO indices skilfully forecasts April–June hot-day counts in northern Vietnam but performs poorly in southern regions.
Tree-based ML (decision tree, random forest, AdaBoost) combines four seasonal forecast systems for China summer rainfall, reaching ACC 0.30 versus 0.21 for simple averaging in 2019–2021.
A CNN trained on CMIP5/6 models forecasts eastern Indian Ocean Dipole SST anomalies up to 6 months ahead, matching the ACCESS-S2 dynamical model.
Nanjing University of Information Science & Technology (NUIST)Bureau of Meteorology (Australia)
A CNN gives skilful probabilistic MJO forecasts whose uncertainty tracks instantaneous predictability, using predicted standard deviation and Monte Carlo dropout.
Assesses machine learning models for seasonal precipitation simulation in Iran using teleconnection indices as predictors; no abstract available.
Neural network using global climate indices forecasts monthly and seasonal precipitation in Crimea up to six months ahead, with about 55–62% forecast quality.
GSFM uses SE-ResNet to predict intraseasonal Z500 and T850 components, substantially outperforming CFSv2 at 10–30 day leads.
Nanjing University of Information Science & Technology (NUIST)
Linear regression with lagged ENSO and IOD indices matches or beats an MLP for seasonal spring rainfall in Western Australia.
Random forest on three preceding atmosphere-ocean indices gives a seasonal early warning of Yangtze summer extreme precipitation, including the 2020 event.
Institute of Atmospheric Physics, Chinese Academy of Sciences
Echo state network trained on low-frequency ENSO indices skilfully predicts El Niño up to 21 months ahead, with correlation above 0.5.
Technical University of Munich (TUM)Potsdam Institute for Climate Impact Research (PIK)
LSTM-parameterised stochastic differential equations for probabilistic Indian monsoon rainfall forecasting from 40 years of station data.
Neural networks quantify how warming alters midlatitude subseasonal skill from tropical precipitation in CESM2 large ensemble, linked to seasonal variability changes.
Genetic-algorithm-optimized backpropagation network forecasts summer precipitation in the Yangtze-Huaihe basin, reaching MAPE near 20% and beating SVM and linear regression.
ENSO-MC multichannel deep network forecasts spatial SST anomalies for eastern and central Pacific El Niño up to one year ahead; saliency maps locate precursors.
Shallow neural networks predict the real-time MJO index out to about 18 days in October–March, outperforming linear models and explained via attribution techniques.
A CNN using Northeast Pacific SST predicts North China winter temperature with 0.65 correlation at 5-month lead, beating a state-of-the-art dynamical seasonal system.
Nanjing University of Information Science & Technology (NUIST)
ENSO-driven machine learning predicts the NAO index about three months ahead at 68% accuracy, and a TrajGRU network improves short-term NAO forecasts.
Self-organizing maps link North Atlantic spring SST to summer jet-stream teleconnections, identifying windows of opportunity for seasonal European summer predictions with MPI-ESM hindcasts.
A deep learning model pre-trained on GCM hindcasts and transfer-learned on ERA5 raises seasonal China precipitation correlation by 0.10-0.13 over raw GCM output; used operationally for guidance.
Tsinghua University (NowcastNet)China Meteorological Administration (CMA)
Hybrid neural network combined with principal oscillation pattern analysis for ENSO prediction; the abstract is missing, so the classification relies on the title.
Unet-LSTM model forecasts global monthly SST anomalies up to 24 months ahead, capturing ENSO events such as the 2015-16 El Niño with RMSE below 0.75 °C.
Random forests with explainable AI on ERA5 identify drivers of subseasonal high summer temperatures in Europe, with useful skill for warm anomalies at 15-day lead.
KNMI (Royal Netherlands Meteorological Institute)Potsdam Institute for Climate Impact Research (PIK)
Hybrid neural-network and regression models using GloSea5 and SEAS5 pressure indices forecast Irish winter and summer rainfall up to 4 months ahead.
Applies gradient-boosted trees with Optuna tuning to seasonal precipitation over South America, outperforming INPE numerical models in some cases.
Deep learning with optimized meteorological inputs predicts Thailand's seasonal precipitation two months ahead, with El Niño monitoring regions identified as key predictors.
A convolutional LSTM with backward stepwise feature selection forecasts the Oceanic Niño Index 1 to 12 months ahead, with SST the best predictor.
K-means plus SMOTE augmentation improves random forest and XGBoost monthly grid precipitation forecasts for the Danjiangkou basin.
Machine learning post-processing of subseasonal-to-seasonal forecasts improves prediction of extreme precipitation occurrence over the contiguous US.
NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)University of Oklahoma (AI2ES)
Late fusion of multiple neural networks with a denormalization layer improves long-range 2m temperature forecasts, beating 30-year climate normals; no headline number given.
Retracted article: feed-forward and recurrent networks predict the Madden–Julian oscillation with about 26–27 days skill, reaching 60 days for some initial phases.
A deep residual CNN with dropout and transfer learning extends ENSO Niño3.4 forecast skill to a 20-month lead, three months beyond the previous optimal model.
Random forest predicts Yangtze summer precipitation with 0.473 skill from December, outperforming eight climate models in several respects.
Nanjing University of Information Science & Technology (NUIST)
A graph-based air-sea coupler inside an ENSO deep learning model reaches Niño3.4 correlation of 0.78 at 6-month lead and outperforms other ENSO forecast models.
Multiple machine learning models predict Atlantic Multidecadal Variability from North Atlantic SST, salinity and sea level pressure maps, beating persistence in CESM-LE.
A wavelet neural network uses climate indices to predict seasonal rainfall with a one-year lead time.
Soft periodic convolutional recurrent network adds periodic-depth representations to forecast ERA-Interim Z300 and SST, improving accuracy and capturing an ENSO-related four-year periodicity.
Comparison of modern ML models against SubX dynamical models for US sub-seasonal forecasting finds ML wins on average, and feeding dynamical forecasts into ML improves skill.
Path-signature machine learning model using climate indices predicts El Niño NINO3.4 SST six months ahead with MSE of 0.596 and identifies North Pacific precursors.
Parallel CNNs with lead-time-specific architectures extend ENSO forecasts multiple years ahead with better skill than prior deep learning baselines.
ML trained on a large climate model ensemble yields seasonal precipitation forecasts over the western US competitive with NMME dynamical models.
Ridge and lasso models using Niño3.4 give modest week 3-4 skill for US winter temperature, only slightly beating CFSv2 and climatology-based baselines.
An ENSO Transformer (ENSOTR) predicts the Niño3.4 index skillfully up to 18 months ahead and outperforms CNN baselines, including the 1997–98 El Niño.
Deep learning using Indian Ocean sea surface temperatures and forecast ENSO hindcasts predicts spring rainfall in SE Australia, with skill for September/October from end of June.
Interpretable machine learning ranks 24 predictive indices for March–September Midwest US S2S precipitation and links them to three moisture and circulation mechanisms.
Temporal Fusion Transformer forecasts weekly extreme precipitation quantiles up to six months ahead, beating calibrated SEAS5 at six-month lead and modestly beating climatology.
Neural network uses SST and SLP indices to forecast monthly precipitation for the Ai-Petri region up to 6 months ahead, with about 23–28% mean absolute deviation.
DC-LSTM, a dense-convolution LSTM, forecasts ENSO Nino3.4 up to 20 months ahead with higher skill than a dynamical model over 1994-2010.
Nanjing University of Information Science & Technology (NUIST)
Parameter-free optimistic online learning with delay, motivated by climate forecasting, achieving low regret on four subseasonal forecasting tasks.
Gradient-boosting and related ML models forecast Eurasian winter precipitation seasonally, outperforming linear regression and matching or beating dynamical models; Arctic autumn sea ice is a key predictor.
Deep learning bias correction of multi-model dynamical forecasts cuts MJO amplitude error by about 90% and phase error by 77% over four-week forecasts.
Neural network identifies subseasonal forecasts of opportunity over the North Atlantic at 22-day lead, with layer-wise relevance propagation revealing tropical predictors.
Multitask local linear regression predicts China precipitation 2–6 weeks ahead with skill 0.11, rising to 0.13 when the Arctic Oscillation index is added.
Nanjing University of Information Science & Technology (NUIST)Institute of Atmospheric Physics, Chinese Academy of SciencesBaidu
Data-driven MSSA matches CMIP5 GCM skill for SST and Indian summer monsoon rainfall decadal hindcasts, improving ISMR skill over GCMs.
Information-flow causality links sunspot numbers to El Niño Modoki, and a causal deep learning model reproduces events 12 years ahead, according to the authors.
Graphino, a graph neural network with a learned connectivity module, outperforms prior deep learning models for ENSO forecasts up to six months ahead and yields interpretable connections.
A NARX neural network predicts the 12-month drought index SPI-12 in West Papua from the MEI ENSO index, capturing ENSO-linked rainfall patterns.
A Bayesian neural network trained on only 20 years of data predicts Niño 3 SST skillfully for 9.5 months, overcoming the spring predictability barrier.
Artificial neural networks predict the Madden-Julian Oscillation with about 26-27 days average skill, reaching 60+ days for some initial phases, competitive with dynamical models.
Improves annual precipitation prediction by combining wavelet packet decomposition with extreme learning machines; abstract unavailable, so details are limited.
An all-season CNN for ENSO forecasts improves boreal spring skill and keeps predictions consistent across seasons and lead times.
Tutorial on machine learning basics, with a CNN applied to East Asian winter monthly temperature and compared against a dynamical model; no headline numeric result.
Compares 1D CNN, MLP and Wide-Deep networks for summer monsoon rainfall in Rajasthan using IMD gridded and gauge data; the Wide-Deep model performs best.
Interpretable neural networks forecast MJO convection several weeks ahead, using layer-wise relevance propagation to explain sources of predictability compared with dynamical models.
A CNN-based cubed-sphere ensemble of 32 deep-learning weather models gives 2-6 week forecasts that are only modestly worse than the ECMWF S2S ensemble over land.
Reservoir computing predicts North Atlantic SST in a CESM control run, beating linear inverse models with limited data and suggesting emulator potential.
CNN and RNN models forecast 2 m temperature out to 52 weeks at six locations, beating climatology up to 30 weeks for correlation, but only at select sites.
A CNN using spatially restructured SST, sea-level pressure and wind predictors forecasts El Niño and La Niña types one year ahead with high precision.
Convolutional Gaussian processes are applied to improve seasonal forecasts of precipitation and temperature, targeting meteorological rather than hydrological variables.
Spatiotemporal graph neural networks forecast ENSO at 1 and 3 months lead, outperforming state-of-the-art systems and modeling teleconnections more interpretably.
ConvLSTM predicts the Niño3.4 index up to one year ahead, overcoming the spring predictability barrier and capturing strong El Niño events.
A probabilistic deep network trained on climate simulations beats a dynamical seasonal system for precipitation and temperature, with saliency analysis of ENSO-driven predictability.
Graph-guided regularized regression on Pacific sea surface temperatures improves seasonal winter precipitation prediction in the southwestern US over benchmark regularizers and teleconnection models.
Neural network models with ENSO indices forecast monthly rainfall anomalies in Southwestern Colombia five months ahead, with the data-driven approach showing suitable skill.
Machine learning models predict precipitation from large-scale climate indices in an arid-land region; abstract not available, so classification is low confidence.
Nonhomogeneous hidden Markov model identifies five ENSO regimes from 1856 to 2019 and gives out-of-sample probabilistic Niño forecasts comparable to existing models.
SVR and XGBoost seasonal winter surface air temperature forecasts over North America outperform two dynamical models in 2011–2017 real-time tests.
Random forest using Pacific, Indian and Southern Ocean climate indices beats the Queensland SOI phase model for Australian seasonal rainfall probability, improving accuracy by 28-219%.
Benchmarks ten ML methods for US sub-seasonal forecasting; gradient boosting performs best and beats climatology, with ENSO and soil moisture as key predictors.
Support vector machines use Pacific and Atlantic oscillation indices to extend precipitation and streamflow forecast lead times in the Colorado River Basin, with KNN for finer precipitation resolution.
Deep learning model predicts seasonal wet and dry season precipitation from global monthly surface temperature fields (abstract not available).
Applies deep learning to seasonal and sub-seasonal rainfall prediction in the Chao Phraya river basin in Thailand; conference abstract not available.
Compares curve-fitting and soft computing models to predict India's annual average temperature; ML is only loosely central and the study is weak.
Data-driven multi-model ensemble combining seasonal forecasts for deterministic and probabilistic precipitation prediction; abstract missing, so classification is based on the title only.
Support vector regression and random forest forecast the ENSO Modoki index 6 and 12 months ahead, with SVR performing better than RF.
Artificial neural network ensemble predicts the Indian Ocean Dipole from February-April predictors for May-November, outperforming NMME models in correlation and RMSE.
LSTM and CNN predict monthly mean temperature and precipitation from station and sea-surface temperature data, with mixed skill.
Artificial neural networks using ENSO and IOD indices forecast seasonal spring rainfall at three Western Australian stations, with testing correlations up to 0.96.
Climate-network topology features from surface air temperature feed ten classifiers that predict next-season strong or weak ENSO phases with 80% AUC.
Logistic regression with linear baroclinic model simulations shows the past 15 days of MJO activity improve midlatitude circulation predictability beyond 15 days ahead.
Machine learning corrects a coupled climate model's seasonal forecasts of daily air temperature for field-scale agricultural management.
Committor functions are shown to quantify intrinsic predictability of ENSO events in the Jin-Timmerman model, and neural-network inference of committors from data is compared to direct estimation.
Hybrid clustering, ARIMA and neural network model forecasts subseasonal station precipitation, reducing RMSE by 24–48% versus seasonal ARIMA in Iran.
Deep learning model trained on climate model simulations produces multi-year ENSO forecasts, a landmark result for long-lead ENSO prediction.
Nanjing University of Information Science & Technology (NUIST)
Autoencoder features with regression trees predict the EQUINOO index 7 months ahead (r=0.88) and ENSO 1 month ahead (r=0.87).
Quadratic discriminant analysis and neural networks classify Vaigai catchment rainfall classes from ENSO indices over 1959–2016 at seven rain gauges.
Regression and ANN models forecast Vietnam East Sea tropical cyclone counts six months ahead; multiple linear regression performed best on independent data.
Kernel analog forecasting, a Koopman-based nonparametric method, predicts Nino 3.4 out to 10 months, 3-7 months beyond a linear inverse model.
LSTM networks analysed with complex-network methods are applied to ENSO forecasting; the abstract is missing, so results cannot be assessed.
EEMD-decomposed ConvLSTM forecasts the NAO index and outperforms six benchmarks and NOAA GFS and MRF ensemble-mean forecasts in the reported tests.
Neural network predicts monthly rainfall in South Kalimantan from monthly SST EOF predictors; accurate in dry season, weakest during the wet-season transition.
Compares LSTM and linear regression for ENSO prediction from 1 to 11-month leads, finding LSTM matches LR on monthly data and gains skill with daily SST at longer leads.
CNN on global surface temperature images forecasts Chao Phraya basin precipitation two months ahead, performing better in the rainy season than the dry season.
Opinion article arguing machine learning should be adopted for S2S forecasting, showing statistical methods beat dynamical models on a polar vortex case.
Microsoft Research (AI for Science)Massachusetts Institute of Technology (MIT)Potsdam Institute for Climate Impact Research (PIK)Stanford University
Neural networks post-process PNU CGCM seasonal hindcasts with sea-ice and SST indices to downscale winter 2-m temperature over Mongolia to 20 km, with improved skill.
Ensemble of regression models with the SubseasonalRodeo dataset beats the top Rodeo entry and improves debiased CFSv2 precipitation skill by 129-169%.
A neural network tuned by a meta-heuristic predicts monthly and seasonal precipitation in southern Brazil; the abstract does not match the title.
Artificial neural networks replace expert steps in a seasonal rainfall forecasting workflow for Zambia, with results reported only qualitatively.
LSTM networks compared with linear regression for ENSO prediction across 11 leads; LSTM gains skill over regression only with daily SST predictors at longer leads.
Hybrid ARIMA and neural network using climate-network topological features extends El Niño prediction to a 12-month lead, matching CFSv2 skill up to six months.
ACE2 is a 450M-parameter emulator that conserves dry air and moisture, runs about 1500 simulated years per day, and reproduces El Niño responses over 80 years.
Climate emulationAllen Institute for AI (Ai2, ACE)NOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
DLESyM, a deep learning Earth system model, simulates 1000-year observed climate without drift, matching CMIP6 variability at a fraction of the cost.
Climate emulationInstitute of Atmospheric Physics, Chinese Academy of SciencesUniversity of Washington
ChaosBench is a physics-based S2S benchmark showing weather-scale AI models collapse to climatology, with ensembles and physics constraints suggested to extend skill.
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 map identifies six subseasonal Indian monsoon stages and two wind-based indices for subseasonal monitoring and forecasting.
Unsupervised learning finds three propagation modes of Indo-Pacific rainfall extremes linked to Pacific SST, with early warning skill up to four weeks ahead.
Climate analysis & attributionTechnical University of Munich (TUM)Potsdam Institute for Climate Impact Research (PIK)
Modular neural network encodes uncertain seasonal climate forecasts for supply-chain demand forecasting, reducing error by 13% to 17% versus existing methods.
CycleGAN bias correction of NUIST-CFS summer precipitation forecasts in southeastern China outperforms quantile mapping and better captures extreme precipitation frequency and intensity.
Post-processing & calibrationShanghai Artificial Intelligence LaboratoryNanjing University of Information Science & Technology (NUIST)
Hybrid model uses an ANN trained on analysis increments to correct errors in a coupled ocean-atmosphere model (MAOOAM), improving prediction skill across lead times.
Hybrid physics-ML & parameterizationLawrence Berkeley National Laboratory / NERSC
CNN post-processing of ECMWF ensemble means yields calibrated global tercile forecasts for weeks 3-6, outperforming recalibrated ECMWF sub-seasonal forecasts.
Post-processing & calibrationKarlsruhe Institute of Technology (KIT)
Ensemble neural networks correct nonstationary sea surface temperature bias in NCEP CFSv2 seasonal forecasts, reducing RMSE by 0.35 °C, with a deeper network giving further gains.
Post-processing & calibrationInstitute of Atmospheric Physics, Chinese Academy of Sciences
ML models predict India heatwave occurrence a month ahead for each 1° grid cell using six months of maximum temperature data.
Uses machine-learning combinations of ENSO, IOD and MJO indices to predict Medan rainfall, finding the SOI plus MJO combination gives the best verification, with MAE of 15.0 mm.
Builds a deep-learning westerly wind burst parameterization intended to improve ENSO prediction; the abstract is missing, so gains are unknown.
An explainable ANN corrects ECMWF extended-range European summer temperature exceedance probabilities and traces errors to western Pacific SST teleconnections, improving skill.
Post-processing & calibrationKNMI (Royal Netherlands Meteorological Institute)
Machine learning emulates the systematic error of a reduced-order coupled ocean-atmosphere model, and the resulting hybrid model beats the dynamical model alone in twin experiments, best when correcting both components.
Hybrid physics-ML & parameterizationLawrence Berkeley National Laboratory / NERSC
A U-Net replaces EOF models for precipitation response to SST anomalies in hybrid ENSO modelling, outperforming EOF over the western-central tropical Pacific.
Hybrid physics-ML & parameterizationNanjing University of Information Science & Technology (NUIST)
Hybrid LSTM driven by climate model outputs to predict natural drought; the abstract is missing, so details are limited.
Attention U-Net with multi-scale residual blocks bias-corrects CFSv2 sub-seasonal rainfall over Thailand, improving RMSE by 8.65% over linear downscaling.
Two deep-learning bias-correction pathways improve 4-month-lead summer precipitation forecasts from a Chinese GCM, with the deterministic pathway better for extremes.
Post-processing & calibrationNanjing University of Information Science & Technology (NUIST)Alibaba (DAMO Academy)
ML post-processing of lagged ensemble forecasts improves two-week-ahead monthly precipitation and temperature forecasts over the continental US, beating climatology and ensemble means.
Statistical ML seasonal forecasts of poor US soybean harvests from Pacific SST and soil moisture precursors reach 67% correct at three months before sowing.
Applications & impactsKNMI (Royal Netherlands Meteorological Institute)
Convolutional autoencoder bias-corrects seven SubX S2S rainfall forecasts over Java, raising weekly correlations (e.g. 0.76 to 0.81) and CRPSS by 20–30%.
Machine learning classifiers and committee models correct ECMWF S2S precipitation category forecasts over three regions, improving on the raw ECMWF baseline.
Global open datacube and segmentation deep learning model predicting burned areas 8 to 64 days ahead using climate and fire-driver variables.
Review argues for bottom-up AI and hybrid physical-statistical drought forecasting, prioritising robustness and uncertainty stability over event-based verification.
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.
Machine-learning adaptive bias correction of ECMWF subseasonal forecasts raises contiguous-US temperature skill 60-90% and precipitation skill 40-69%.
Post-processing & calibrationMicrosoft Research (AI for Science)Massachusetts Institute of Technology (MIT)
Random forest and ANN post-processing of four NMME models gives better monthly and seasonal precipitation forecasts over Southwest Asia than individual models.
Machine learning post-processing of ECMWF forecasts improves MJO prediction, outperforming multiple linear regression in location and intensity skill.
Post-processing & calibrationECMWFKarlsruhe Institute of Technology (KIT)
CNN predicts ENSO nine months ahead with about 0.82 correlation; contribution maps identify three precursors with asymmetric roles in El Nino and La Nina.
Compares six AI approaches across four climate tasks, claiming hybrid physics-ML best at 6-month temperature RMSE; claims are unverified and methodology thin.
LightGBM uses preceding-winter sea surface temperature, soil moisture and snow to simulate summer eastern European heatwave frequency, with SST contributing most.
Distorted-physics experiments show a CNN El Niño predictor corrects ocean adjustment distortions well but struggles with changes in upwelling feedback strength.
Hierarchical clustering separates northward-moving typhoon tracks affecting Northeast China and links each type to preceding Pacific SST anomalies, with numerical experiments confirming the mechanism.
Climate analysis & attributionChina Meteorological Administration (CMA)
Kalman filter with a trained linear inverse model performs strongly coupled assimilation, cutting SST analysis error over 20% and extending forecast skill to 50 days.
Data assimilationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)NSF NCAR / UCARUniversity of Washington
Random forest quantile regression and Bayesian LSTM surrogates emulate a malaria transmission model driven by seasonal temperature and rainfall ensembles, cutting computation while capturing forecast uncertainty.
Hybrid XGBoost models using S2S dynamical forecasts predict northern China winter wheat yield 3–4 months ahead with R² 0.85, beating observed climate inputs.
Applications & impactsInstitute of Atmospheric Physics, Chinese Academy of SciencesChina Meteorological Administration (CMA)
Predicts annual tropical cyclone frequency for Guangdong from winter precursors using LSTM-FC, which achieves lower error than MLR and GPR for 2010–2019.
Random forest forecasts monthly extreme rainfall indices over Bangladesh 1–3 months ahead using CAPE, humidity, and ENSO/IOD/NAO teleconnections.
Extremes & hazardsNanjing University of Information Science & Technology (NUIST)
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.
Hybrid machine learning and S2S dynamical forecasts improve northern China winter wheat yield prediction over observation-based inputs.
Applications & impactsInstitute of Atmospheric Physics, Chinese Academy of SciencesChina Meteorological Administration (CMA)
Introduces a skill-decomposition method for ML week 3–4 forecasts, finding western US temperature patterns predictable mainly through Pacific ENSO-related SST variability.
Video CNNs using GloSEA5 pressure and temperature fields forecast monthly regional rainfall in Great Britain; an ensemble with GloSEA5 reduces RMSE to 31-32 mm.
Dense neural networks predict daily max and min 2 m temperatures from radiosonde profiles up to 500 days ahead, beating persistence but not climatology beyond day 2-3.
SubseasonalClimateUSA is a curated dataset and benchmark for 2-6 week US forecasts, comparing dynamical models with ten ML and deep learning methods.
Machine learning models with satellite drought indices, weather and NAO predict Moroccan cereal yields up to 4 months before harvest, outperforming single-source models.
K-means clustering and C4.5 decision trees link Poyang Lake basin precipitation clusters to preceding spring large-scale circulation indices.
Climate analysis & attributionNanjing University of Information Science & Technology (NUIST)
Bayesian joint probability calibration of ML and GCM monthly precipitation forecasts over China yields reliable ensembles, though ML outputs lack skill beyond climatology.
Machine learning links cacao yields in Sulawesi to ENSO patterns 8–24 months ahead, explaining 75% of yield variation and timing fertilizer use.
CNN feature fusion of sea-level pressure, SST and wind predicts preseason Atlantic hurricane activity with skill up to 18 months ahead.
Wavelet neural networks forecast monthly rainfall at Queensland stations up to 12 months ahead, reducing three-month RMSE to 38.6 mm versus 72.2 mm for plain ANN.
Neural networks post-process NCEP CFS week-3–4 precipitation and 2-m temperature forecasts, improving skill over multiple linear regression correction.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Layer-wise relevance propagation applied to neural networks identifies atmospheric states associated with enhanced weather predictability, giving physical insight into predictability mechanisms.
Explainability, uncertainty & physicsColorado State University (incl. CIRA)
Neural networks and NGBoost post-process SubX GEFS 2-m maximum temperature forecasts, beating EMOS by about 0.2 in CRPS over lead times of 1 to 35 days.
Neural networks postprocess ECMWF reforecasts into probabilistic 7-day California precipitation outlooks out to week 4, beating a state-of-the-art method, though week 3-4 skill stays low.
Post-processing & calibrationNOAA (incl. GFDL, PSL, GSL, NCEP/EMC)
Uses machine learning to quantify how pre-occurring ENSO signals drive Australian wheat yield variation, linking seasonal climate drivers to agricultural impacts; abstract unavailable.
Applies machine learning to predict summer temperatures and heat wave occurrence in Korea at seasonal lead; abstract unavailable, so judged from the title only.
Neural networks post-process the North American multi-model ensemble to improve monthly precipitation forecasts; abstract unavailable.
Quantum-mechanics-inspired, kernel-based data assimilation using Koopman operators, applied to Indo-Pacific SST and Niño 3.4 probabilistic forecasts. Novel but early-stage.
A genetic algorithm learns optimal weights for 24 SINTEX-F2 ensemble members, significantly improving boreal summer 2m air temperature anomaly forecasts over several regions versus the unweighted mean.
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.
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.
Neural network and linear models predict Chilean agricultural drought from ENSO, PDO and precipitation indices at 1-6 month lead times.
Combines wavelet decomposition and machine learning to improve local precipitation forecasts from the NMME multi-model seasonal ensemble.
Post-processing & calibrationChina Meteorological Administration (CMA)
Artificial neural network predicts monthly rainfall in Kano, Nigeria from lagged ENSO indices, achieving correlation 0.73 versus 0.70 for a linear model.
Pre-season soybean and maize yield forecasts use satellite precipitation and seasonal climate forecasts instead of NDVI, with errors comparable to in-season systems.
Extra-Trees and AdaBoost correct seasonal forecasts and WRF downscaling to predict 6-month drought (SPI) in data-sparse Fiji, outperforming bias-corrected forecasts.
