An ML model trained only on harmonized gridded observations achieves competitive medium-range weather forecast skill with the IFS for several upper-air and surface headline scores when verified against observations.
Canonical reference
Learning skillful medium-range global weather forecasting.Science, 382(6677):1416–1421, 2023
Canonical reference. 80% of citing Pith papers cite this work as background.
48
Pith papers citing it
1,014
external citations · Crossref
Background
80% of classified citations
citation-role summary
background 8
baseline 1
method 1
citation-polarity summary
representative citing papers
Observation-only machine learning can generate multi-decade global atmospheric reanalyses with large-scale skill near ERA5 and surface errors between ERA-Interim and ERA5, in a single day of compute.
Forward Flux Sampling applied to a 1-degree neural weather emulator resolves conditional tropical cyclogenesis rates spanning three orders of magnitude across 98 Atlantic initial conditions, with self-consistency ratio 1.03 to direct sampling and computational gains up to 140X.
RATrain introduces a resource-aware scheduler and MT-3000-specific backend for 1F1B LLM training that achieves 1.35x speedup and 97% scaling efficiency while preserving training correctness.
TianJi-Environ is a WRF-Chem-based multi-agent AI framework for autonomous validation of atmospheric chemistry mechanisms through executable experiments and evidence assessment.
Targeted perturbations in the Aurora AI model can steer Hurricane Sandy's trajectory by more than 500 km after seven days via amplification in sensitive regions identified by FTLE and wave activity diagnostics.
A GNN learns sparse corrections to AMG operators, cutting iteration counts on large PDE and graph-Laplacian systems while leaving solve-phase cost unchanged.
AI weather models may simulate the atmosphere via particle positions in latent space whose updates follow gradient flow on a learned free energy functional rather than conventional physical equations.
LUCAS-MEGA fuses 68 heterogeneous soil datasets into a 70k-sample multimodal collection and demonstrates its value by pretraining a tabular transformer whose representations align with established soil processes.
Cast3 translates NWP principles into a data-driven model using cubed-sphere grids, super-ensembles, and generative nudging to achieve state-of-the-art ensemble predictions that outperform baselines.
Influpaint uses generative diffusion models on image-encoded influenza data to produce realistic and diverse epidemic trajectories that match leading ensemble methods in accuracy.
Historically trained ML weather emulators quantify fast precipitation changes from CO2 perturbations and produce results that agree with Earth System Models.
AlphaEvolve is an LLM-orchestrated evolutionary coding agent that discovered a 4x4 complex matrix multiplication algorithm using 48 scalar multiplications, the first improvement over Strassen's algorithm in 56 years, plus optimizations for Google data centers and hardware.
Two new global-domain smoothing methods enable spatial verification scores like FSS on high-resolution global precipitation forecasts while handling grid area variability and missing data.
Fully trained BiLSTM outperforms zero-shot and LoRA-adapted TSFMs on California wildfire PM2.5 under leave-one-incident-out evaluation, especially at hazardous AQI thresholds.
Location embeddings from geographic INRs can be decomposed into sparse latent concepts, natural language concepts, and visual features while retaining high reconstruction capability.
Online conformal prediction post-processing guarantees calibrated uncertainty coverage for GenCast, NeuralGCM, and AIFS-ENS forecasts of temperature and precipitation including extremes.
Hybrid LSTM-ViT model using mesonet surface data and profiler vertical profiles improves HRRR forecast error prediction for precipitation, wind speed, and temperature, with roughly twofold skill gain for precipitation over baseline LSTM.
NTK-UQ produces 31-37% sharper 90% prediction intervals than split conformal prediction for extreme weather forecasts, with adaptive scaling via architecture-dependent eigenvalue truncation and ICA decomposition of last-layer features.
GNNs are shown to lack continuity under graph resolution changes due to message-passing schemes, with a derived modification enabling consistent multi-scale representations validated experimentally.
ARC-STAR reduces velocity rollout error by at least 36x over raw Poseidon across all tested regime cells via auditable global and local correction stages on five flow benchmarks.
Mechanism learning infers active local evolution rules via prototype-anchored descriptors to achieve more robust forecasting than direct state prediction on benchmarks like Burgers, WeatherBench2, and Lorenz96.
ShardTensor is a domain-parallelism system for SciML that enables flexible scaling of extreme-resolution spatial datasets by removing the constraint of batch size one per device.
AxiomOcean deploys a 3D encoder-backbone-decoder architecture that jointly predicts upper-ocean variables and outperforms prior AI models by 20-35% in day-1 RMSE while preserving eddy kinetic energy and vertical consistency.
citing papers explorer
-
AIFS-DOP: End-to-End Medium-Range Weather Prediction from Observations Alone with Machine Learning
An ML model trained only on harmonized gridded observations achieves competitive medium-range weather forecast skill with the IFS for several upper-air and surface headline scores when verified against observations.
-
Global reanalysis from observations alone with machine learning
Observation-only machine learning can generate multi-decade global atmospheric reanalyses with large-scale skill near ERA5 and surface errors between ERA-Interim and ERA5, in a single day of compute.
-
Conditional Tropical Cyclogenesis Rates via Rare-Event Sampling in a Neural Weather Emulator
Forward Flux Sampling applied to a 1-degree neural weather emulator resolves conditional tropical cyclogenesis rates spanning three orders of magnitude across 98 Atlantic initial conditions, with self-consistency ratio 1.03 to direct sampling and computational gains up to 140X.
-
RATrain: A Resource-Aware Training Runtime for Large Language Models on Bandwidth-Constrained Heterogeneous Supercomputing Platforms
RATrain introduces a resource-aware scheduler and MT-3000-specific backend for 1F1B LLM training that achieves 1.35x speedup and 97% scaling efficiency while preserving training correctness.
-
TianJi-Environ: An Autonomous AI Scientist for Atmospheric Environmental Research
TianJi-Environ is a WRF-Chem-based multi-agent AI framework for autonomous validation of atmospheric chemistry mechanisms through executable experiments and evidence assessment.
-
Steering Tropical Cyclones Using Small Perturbations in an AI Weather Model
Targeted perturbations in the Aurora AI model can steer Hurricane Sandy's trajectory by more than 500 km after seven days via amplification in sensitive regions identified by FTLE and wave activity diagnostics.
-
RAPNet: Accelerating Algebraic Multigrid with Learned Sparse Corrections
A GNN learns sparse corrections to AMG operators, cutting iteration counts on large PDE and graph-Laplacian systems while leaving solve-phase cost unchanged.
-
The physics of AI weather models
AI weather models may simulate the atmosphere via particle positions in latent space whose updates follow gradient flow on a learned free energy functional rather than conventional physical equations.
-
LUCAS-MEGA: A Large-Scale Multimodal Dataset for Representation Learning in Soil-Environment Systems
LUCAS-MEGA fuses 68 heterogeneous soil datasets into a 70k-sample multimodal collection and demonstrates its value by pretraining a tabular transformer whose representations align with established soil processes.
-
Cast3: Translating numerical weather prediction principles into data-driven forecasting
Cast3 translates NWP principles into a data-driven model using cubed-sphere grids, super-ensembles, and generative nudging to achieve state-of-the-art ensemble predictions that outperform baselines.
-
Generative diffusion models for spatiotemporal influenza forecasting
Influpaint uses generative diffusion models on image-encoded influenza data to produce realistic and diverse epidemic trajectories that match leading ensemble methods in accuracy.
-
Examining Fast Radiatively Driven Responses Using Machine-Learning Weather Emulators
Historically trained ML weather emulators quantify fast precipitation changes from CO2 perturbations and produce results that agree with Earth System Models.
-
AlphaEvolve: A coding agent for scientific and algorithmic discovery
AlphaEvolve is an LLM-orchestrated evolutionary coding agent that discovered a 4x4 complex matrix multiplication algorithm using 48 scalar multiplications, the first improvement over Strassen's algorithm in 56 years, plus optimizations for Google data centers and hardware.
-
Smoothing and spatial verification of global fields
Two new global-domain smoothing methods enable spatial verification scores like FSS on high-resolution global precipitation forecasts while handling grid area variability and missing data.
-
Evaluating the Generalizability of Foundation Models for Extreme Environmental Events: Case Study of California Wildfire PM2.5
Fully trained BiLSTM outperforms zero-shot and LoRA-adapted TSFMs on California wildfire PM2.5 under leave-one-incident-out evaluation, especially at hazardous AQI thresholds.
-
What's in an Earth Embedding? An Explainability Analysis of Location Encoders
Location embeddings from geographic INRs can be decomposed into sparse latent concepts, natural language concepts, and visual features while retaining high reconstruction capability.
-
Rigorous uncertainty quantification of probabilistic AI weather forecasts with conformal prediction
Online conformal prediction post-processing guarantees calibrated uncertainty coverage for GenCast, NeuralGCM, and AIFS-ENS forecasts of temperature and precipitation including extremes.
-
A Hybrid LSTM--Vision Transformer Architecture for Predicting HRRR Forecast Errors
Hybrid LSTM-ViT model using mesonet surface data and profiler vertical profiles improves HRRR forecast error prediction for precipitation, wind speed, and temperature, with roughly twofold skill gain for precipitation over baseline LSTM.
-
Scalable Uncertainty Quantification for Extreme Weather Forecasting via Empirical Neural Tangent Kernels
NTK-UQ produces 31-37% sharper 90% prediction intervals than split conformal prediction for extreme weather forecasts, with adaptive scaling via architecture-dependent eigenvalue truncation and ICA decomposition of last-layer features.
-
Graph Neural Networks Are Not Continuous Across Graph Resolutions
GNNs are shown to lack continuity under graph resolution changes due to message-passing schemes, with a derived modification enabling consistent multi-scale representations validated experimentally.
-
ARC-STAR: Auditable Post-Hoc Correction for PDE Foundation Models
ARC-STAR reduces velocity rollout error by at least 36x over raw Poseidon across all tested regime cells via auditable global and local correction stages on five flow benchmarks.
-
Mechanism Learning: Prototype-Anchored Mechanism Inference for Scientific Forecasting
Mechanism learning infers active local evolution rules via prototype-anchored descriptors to achieve more robust forecasting than direct state prediction on benchmarks like Burgers, WeatherBench2, and Lorenz96.
-
ShardTensor: Domain Parallelism for Scientific Machine Learning
ShardTensor is a domain-parallelism system for SciML that enables flexible scaling of extreme-resolution spatial datasets by removing the constraint of batch size one per device.
-
AxiomOcean: Forecasting the Three-Dimensional Structure of the Upper Ocean
AxiomOcean deploys a 3D encoder-backbone-decoder architecture that jointly predicts upper-ocean variables and outperforms prior AI models by 20-35% in day-1 RMSE while preserving eddy kinetic energy and vertical consistency.
-
Supercharging Bayesian Inference with Reliable AI-Informed Priors
Rectified AI priors, obtained by correcting AI-induced data laws before embedding them in techniques like Dirichlet process priors, reduce bias, improve credible interval coverage, and boost performance in tasks like skin disease classification.
-
Controlling Transient Amplification Improves Long-horizon Rollouts
Commutativity regularization mitigates transient error amplification in autoregressive neural simulators by penalizing non-normality and non-commutativity of Jacobians, yielding stable long-horizon rollouts.
-
Extreme Weather Bench: A framework and benchmark for evaluation of high-impact weather
Extreme Weather Bench supplies standardized case studies, observational data, impact metrics, and code to evaluate weather models on high-impact hazards.
-
StormNet: Improving storm surge predictions with a GNN-based spatio-temporal offset forecasting model
A spatio-temporal GNN model reduces storm surge water-level forecast RMSE by more than 70% for 48-hour horizons and over 50% for 72-hour horizons on U.S. Gulf Coast hurricane data.
-
Earth System Foundation Model (ESFM): A unified framework for heterogeneous data integration and forecasting
ESFM is a single open foundation model that unifies heterogeneous Earth data sources and forecasts missing regions while preserving inter-variable physical relationships.
-
Integrating Weather Foundation Model and Satellite to Enable Fine-Grained Solar Irradiance Forecasting
Baguan-solar integrates Baguan weather foundation model forecasts with geostationary satellite data via a decoupled two-stage multimodal framework to deliver kilometer-scale 24-hour solar irradiance predictions, cutting RMSE by 16% versus baselines over East Asia.
-
EnScale: Temporally-consistent multivariate generative downscaling via proper scoring rules
EnScale emulates high-resolution regional climate model outputs from global circulation models for multiple variables using a two-step generative process with sparse local stochastic layers and energy score optimization, including a temporally consistent variant.
-
Towards Fair Comparisons of AI- and Physics-Based Weather Models for Extreme Events via the Weighted Potential CRPS
Extends Potential CRPS with weights and IDR post-processing to enable fair comparisons of AIWP and NWP models on extreme weather, finding AI models more informative across most variables and thresholds.
-
Instrumented data for causal scientific machine learning
Instrumented data augments observations with mechanistic models, uncertainty, and counterfactuals to enable causal interventions via Pearl's do-operator in scientific machine learning.
-
DiffUNet^2: Bidirectional Prediction, Probabilistic Generation and Collaborative Visual Discovery for Scientific Data
DiffUNet^2 is a bidirectional conditional diffusion model integrated with visual tools for probabilistic exploration of scientific time series across five evaluated datasets.
-
Forecasting threshold exceedance of atmospheric variables at a specific location
Full conditional distribution modeling outperforms direct binary classification for rare threshold exceedances by learning bulk parameters from moderate events.
-
Striding Across Reynolds Numbers: Representation Geometry in Neural PDE Generalisation
ConvAE-Relay retrieval via source-trained autoencoder latent matching achieves 38.34+/-0.07% relative L2 error on 10x Re shift using only source database, with U-Net at 34.72% and matching quality identified as dominant factor.
-
Improving Ensemble CAPE Forecasts with a Diffusion Model Incorporating Aerosol Information
Diffusion model improves GFS/GEFS ensemble CAPE forecasts and incorporates aerosol optical depths for additional gains.
-
Discovery of Interpretable Surrogates via Agentic AI: Application to Gravitational Waves
GWAgent agentic workflow produces analytic surrogates for eccentric BBH waveforms with 6.9e-4 median mismatch and 8.4x speedup, outperforming baselines, and infers eccentricity for GW200129.
-
Mechanistic Interpretability Tool for AI Weather Models
An open-source tool is developed for mechanistic interpretability of AI weather models, demonstrated on GraphCast by identifying latent directions corresponding to interpretable weather features.
-
Enhancing AI and Dynamical Subseasonal Forecasts with Probabilistic Bias Correction
Probabilistic bias correction doubles ECMWF AI subseasonal skill and improves debiased dynamical forecasts for most pressure, temperature, and precipitation targets, winning ECMWF’s 2025 real-time competition.
-
CycloneMAE: A Scalable Multi-Task Learning Model for Global Tropical Cyclone Probabilistic Forecasting
CycloneMAE uses a TC structure-aware masked autoencoder with discrete probabilistic gridding and pre-train/fine-tune to deliver both deterministic and probabilistic forecasts, outperforming NWP systems in pressure and wind up to 120 hours and track up to 24 hours across five basins.
-
U-Cast: A Surprisingly Simple and Efficient Frontier Probabilistic AI Weather Forecaster
A standard U-Net with MAE pre-training plus short CRPS fine-tuning and MC Dropout matches GenCast and IFS ENS probabilistic skill at 1.5° while cutting training and inference cost by over 10×.
-
Towards Scaling Law Analysis For Spatiotemporal Weather Data
Scaling laws for weather models exhibit strong cross-channel and cross-horizon heterogeneity, where globally pooled metrics appear favorable while many individual channels degrade at longer leads.
-
Sampling Parallelism for Fast and Efficient Bayesian Learning
Sampling parallelism distributes Bayesian sample evaluations across GPUs for near-perfect scaling, lower memory use, and faster convergence via per-GPU data augmentations, outperforming pure data parallelism in diversity.
-
Event-Aware Loss Design for Forecasting of Convective Precipitation and Lightning
A multi-task Patch-cGAN with lightning-derived spatial loss weighting improves post-processed forecasts of intense precipitation and lightning occurrence over the Korean Peninsula in summer 2025.
-
Performance Evaluation of GraphCast for Medium-Range Weather Forecasting over Brazil
GraphCast shows regime-dependent skill versus ECMWF HRES in Brazil, underperforming on winter baroclinic systems in medium range but gaining in extended range and summer moisture transport.
-
Towards a Foundation Model for the Martian Atmosphere
The paper reviews data sources, physical models, downstream applications, and AI techniques to outline considerations for building a foundation model for the Martian atmosphere.
-
Toward Artificial Intelligence Enabled Earth System Coupling
AI methods can strengthen cross-domain interactions and support more coherent multi-component representations in Earth system models.