Geospatially enhanced convolutional neural network for HY-2B scatterometer-based marine wind field retrieval

Area of Science:

• Earth and Space Sciences
• Oceanography
• Atmospheric Sciences

Background:

• Accurate sea surface wind field monitoring is crucial for ocean-atmosphere interaction and extreme weather event analysis.
• Spaceborne scatterometers offer unique data but face accuracy and efficiency challenges with traditional geophysical model functions (GMFs).

Purpose of the Study:

• To develop advanced convolutional neural network (CNN) models for improved sea surface wind speed and direction retrieval from satellite scatterometer data.
• To integrate geographical coordinates and positional features into CNN models for enhanced physical consistency in diverse marine environments.

Main Methods:

• Development of CNN-based models (CNN-WSPD and CNN-WDIR) integrating HY-2B scatterometer data and ECMWF reanalysis.
• Embedding positional features into multi-angle backscatter processing to improve physical consistency.
• Global dataset validation comparing CNN models against traditional GMF approaches.

Main Results:

• The proposed CNN models significantly reduced wind speed root mean square error (RMSE) by 9.7% and wind direction RMSE by 1.6%.
• Achieved 91.5% of samples with wind speed differences within 2 m/s and 90.5% with wind direction differences within 20°.
• Demonstrated superior global accuracy compared to traditional GMF methods.

Conclusions:

• The developed CNN models offer a significant advancement in satellite scatterometer data processing for sea surface wind retrieval.
• These models can revolutionize product development, enabling real-time high-resolution wind mapping for enhanced typhoon forecasting.
• The approach pioneers next-generation designs supporting climate resilience initiatives through improved weather prediction.