High-resolution climate reconstruction from historical Chinese weather records using optimized natural language...

Area of Science:

• Historical Climatology
• Natural Language Processing
• Data Science

Background:

• Reconstructing historical climate data is challenging due to subjectivity and lack of standardization.
• Existing methods struggle to capture the nuances of weather descriptions in historical documents.

Purpose of the Study:

• To develop and validate a novel framework for reconstructing high-resolution climate data from historical documents.
• To overcome subjectivity and standardization issues in historical climate data reconstruction.

Main Methods:

• Constructed a historical weather classification lexicon using optimized natural language processing (NLP) techniques, semantic clustering, and dynamic expansion.
• Developed a multi-dimensional index system to quantify historical weather, incorporating indicators like intensity, agricultural, economic, social impact, and casualties.
• Assigned objective weights using the entropy method and expert judgment.
• Validated the framework using low-temperature weather records from historical documents of Guangdong and Hebei provinces, China.

Main Results:

• The framework successfully converted qualitative historical narratives into quantitative climate data.
• Reconstructed low-temperature weather trends in Guangdong and Hebei align with existing Qing Dynasty climate change research.
• Provincial trend maps revealed synchronous change patterns and significant regional differences.
• A Random Forest model achieved 94.0% classification accuracy and AUC scores over 0.98 for low-grade events.

Conclusions:

• The data-driven methodology provides a replicable and scalable tool for historical climate data reconstruction.
• Enhances the understanding of past climate variability and its societal impacts.
• Offers a standardized approach to analyzing historical weather records.