Transformer based models with hierarchical graph representations for enhanced climate forecasting.

Area of Science:

• Climate Science
• Artificial Intelligence
• Machine Learning

Background:

• Traditional climate forecasting methods face challenges in regional accuracy, computational efficiency, and scalability.
• Accurate climate predictions are crucial for sectors like agriculture, urban planning, and disaster management.

Purpose of the Study:

• To develop and evaluate a novel Transformer-based deep learning model for precise daily temperature forecasting.
• To address limitations of existing methods by enhancing spatiotemporal dependency capture and computational efficiency.

Main Methods:

• A Transformer-based deep learning model incorporating Spatial-Temporal Fusion Module (STFM), Hierarchical Graph Representation and Analysis (HGRA), and Dynamic Temporal Graph Attention Mechanism (DT-GAM).
• A hybrid optimization approach (HWOA-TTA), combining Whale Optimization Algorithm (WOA) and Tiki-Taka Algorithm (TTA), for improved computational efficiency and feature selection.
• Utilized historical daily climate data from Delhi (2013-2017).

Main Results:

• The proposed model achieved 7.8% higher accuracy, 6.3% improvement in recall, and 8.1% enhancement in F1-score compared to baseline models.
• Demonstrated a 22.4% reduction in training time, indicating significant computational efficiency gains.
• Effectively captured complex spatiotemporal dependencies and structured climate relationships.

Conclusions:

• Hierarchical graph-based deep learning models offer a scalable and accurate solution for climate forecasting.
• The developed Transformer model shows significant improvements over conventional methods for daily temperature prediction.
• Future research should focus on broader validation across different climate zones and real-time deployment.