Artificial Intelligence for Prediction of Climate Extremes: State of the art, challenges and future perspectives

AI shows promise for predicting extreme weather events that traditional climate models struggle with, particularly heat waves, droughts, and storms. The paper notes that "predictive skill of extremes can be increased when using more sophisticated approaches," suggesting untapped potential in current prediction systems.

Hybrid approaches combining AI with traditional climate models are most effective. Rather than replacing physics-based models entirely, the most promising results come from combining machine learning's ability to "reveal and exploit unknown spatio-temporal connections from data" with climate models that provide "theoretical foundation and interpretability of the physical world."

Explainability and interpretability are critical challenges for responsible AI deployment. The paper identifies "explainable AI" as one of the key approaches being tested, highlighting that for high-stakes decisions about extreme weather warnings, stakeholders need to understand why AI systems make their predictions, not just accept them as black boxes.

Data quality and curation present significant obstacles. The abstract notes that "challenges are multiple in many aspects, from data curation" onward, suggesting that responsible implementation requires careful attention to how training data is collected, validated, and maintained—not just algorithmic sophistication.

The rarity and chaotic nature of extreme events complicate AI development. Because extreme weather occurs infrequently and involves non-linear dynamics, AI systems may struggle with limited training examples and unpredictable patterns, requiring thoughtful validation before deployment in real-world warning systems.