What does this research mean for the field?

Artificial intelligence in weather and climate science requires architectures with specific physical inductive biases and must overcome unique evaluation pathologies, such as RMSE smoothness bias, and open challenges like climate non-stationarity. Novelty: ClaimNovelty.SYNTHESIS. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To review the application of artificial intelligence architectures to weather and climate science and identify foundational scientific machine learning challenges.

September 14, 2026Artificial Intelligence ReviewOpen Access

Artificial intelligence for weather and climate: a survey of methods, benchmarks, and scientific machine learning challenges

View Full Paper

Ask AI

Bookmark

Authors

AHAndreas HolzingerUniversity of Vienna SFSandro FioreUniversity of Trento TDTullio DegiacomiUniversity of Padua

Discussion

Member takes

Implication

Methodological review categorizes artificial intelligence architectures and evaluation pathologies in weather forecasting, highlighting key scientific machine learning challenges.

Key Points

To review the application of artificial intelligence architectures to weather and climate science and identify foundational scientific machine learning challenges.
Taxonomic classification of artificial intelligence architectures—including convolutional networks, graph neural networks, transformers, neural operators, and generative models—based on physical inductive biases.
Systematic evaluation of representation learning, foundation models, uncertainty quantification, and causal discovery frameworks applied to global observational and reanalysis archives.
Organized deep learning paradigms under geometric deep learning and operator learning frameworks suited for continuous spatiotemporal dynamics on spherical domains.
Identified critical evaluation pathologies, notably the RMSE smoothness bias that favors blurred predictions over sharp extreme forecasts, alongside ERA5 circularity and benchmark overfitting.
Framed unresolved challenges including climate distribution shift, physical consistency of learned operators, and sample efficiency as fundamental scientific machine learning problems.

Cite This Study

Holzinger et al. (2026) studied this question.

synapsesocial.com/papers/6aa7b2e10926e14a848b17e6 https://doi.org/10.1007/s10462-026-11690-8

View Full Paper

Ask AI

Bookmark

Implication

Methodological review categorizes artificial intelligence architectures and evaluation pathologies in weather forecasting, highlighting key scientific machine learning challenges.

Artificial intelligence for weather and climate: a survey of methods, benchmarks, and scientific machine learning challenges

Discussion

Member takes

Implication

Key Points

Cite This Study

Artificial intelligence for weather and climate: a survey of methods, benchmarks, and scientific machine learning challenges

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider