A vision transformer for lightning intensity estimation using 3D weather radar

Area of Science:

• Meteorology and Atmospheric Science
• Computer Science and Artificial Intelligence
• Geophysics and Earth Science

Background:

• Lightning poses significant threats due to its destructive power (blast wave, high temperature, high voltage).
• Estimating lightning intensity is vital for effective lightning protection and disaster risk assessment.
• Severe convective weather phenomena, like lightning, require advanced monitoring systems.

Purpose of the Study:

• To propose a novel Vision Transformer model for automatic lightning intensity estimation.
• To leverage 3D weather radar data for extracting features correlated with lightning intensity.
• To enhance lightning disaster risk assessment and protection strategies.

Main Methods:

• Utilized 3D weather radar data and lightning location data to create lightning feature samples.
• Transformed lightning intensity estimation into a multicategory classification task.
• Employed the Synthetic Minority Over-Sampling Technique (SMOTE) for sample balancing and optimization.
• Developed and evaluated a Vision Transformer model for lightning intensity estimation.

Main Results:

• The proposed Vision Transformer model demonstrated strong performance in lightning intensity estimation.
• The framework successfully extracted relevant 3D spatial features from weather radar data.
• The multicategory classification approach proved effective for the task.

Conclusions:

• Vision Transformer models are highly effective for lightning intensity estimation using 3D weather radar data.
• Automated estimation of lightning intensity can significantly aid in disaster preparedness.
• This approach offers a promising direction for advancing meteorological hazard monitoring.