ClimateLLM presents a novel approach to weather forecasting that addresses key limitations of current deep learning methods while maintaining computational efficiency.

• Frequency-aware architecture that captures both long-term patterns and short-term weather changes simultaneously
• Multi-frequency modeling that significantly improves prediction accuracy for extreme weather events
• Lower computational costs compared to traditional deep learning methods, enabling more sustainable deployment
• Enhanced security applications through better disaster prediction and mitigation capabilities for vulnerable communities

This research directly enhances security capabilities by providing more reliable early warnings for extreme weather events, enabling better emergency planning and response systems, and reducing infrastructure vulnerabilities.

ClimateLLM: Efficient Weather Forecasting via Frequency-Aware Large Language Models