Target Concept Tuning Improves Extreme Weather Forecasting

TL;DR

TaCT is a novel interpretable fine-tuning framework that enhances extreme weather forecasting by selectively adapting models to failure cases, improving typhoon predictions without sacrificing overall accuracy.

Contribution

Introduces TaCT, a concept-gated fine-tuning method that automatically discovers failure-related concepts and selectively updates models, addressing the trade-off in extreme weather forecasting.

Findings

Improved typhoon forecasting accuracy across regions

Discovered physically meaningful circulation concepts

Enhanced model trustworthiness and interpretability

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.