Target Concept Tuning Improves Extreme Weather Forecasting

2026-03-17

AI Summary

This work addresses the poor performance of deep learning weather forecasting models on rare, high-impact extreme events such as typhoons, where conventional fine-tuning often degrades overall prediction accuracy. To overcome this limitation, the authors propose the TaCT framework, which employs a concept-gated mechanism to enable selective parameter updates—activating fine-tuning only when interpretable internal concepts associated with past failure cases are detected. By integrating sparse autoencoders with counterfactual analysis, TaCT automatically discovers physically meaningful atmospheric circulation–related concepts. Evaluated across multiple regions, the method substantially improves typhoon prediction without compromising accuracy on routine weather conditions or other meteorological variables, while also revealing interpretable sources of model bias.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Deep learning models for meteorological forecasting often fail in rare but high-impact events such as typhoons, where relevant data is scarce. Existing fine-tuning methods typically face a trade-off between overlooking these extreme events and overfitting them at the expense of overall performance. We propose TaCT, an interpretable concept-gated fine-tuning framework that solves the aforementioned issue by selective model improvement: models are adapted specifically for failure cases while preserving performance in common scenarios. To this end, TaCT automatically discovers failure-related internal concepts using Sparse Autoencoders and counterfactual analysis, and updates parameters only when the corresponding concepts are activated, rather than applying uniform adaptation. Experiments show consistent improvements in typhoon forecasting across different regions without degrading other meteorological variables. The identified concepts correspond to physically meaningful circulation patterns, revealing model biases and supporting trustworthy adaptation in scientific forecasting tasks. The code is available at https://anonymous.4open.science/r/Concept-Gated-Fine-tune-62AC.

Problem

• extreme weather forecasting
• typhoon prediction
• data scarcity
• model fine-tuning
• forecasting bias

Innovation

• concept-gated fine-tuning
• sparse autoencoders
• counterfactual analysis
• extreme weather forecasting
• selective adaptation