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

Area of Science:

• Earth and Space Sciences
• Oceanography
• Atmospheric Science

Background:

• Accurate sea surface wind monitoring is vital for ocean-atmosphere studies and extreme weather prediction.
• Spaceborne scatterometers offer unique data but face accuracy limitations 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 using satellite scatterometer data.
• To enhance the physical consistency and accuracy of wind field products across diverse marine environments.

Main Methods:

• Developed CNN-based models (CNN-WSPD and CNN-WDIR) integrating HY-2B scatterometer data and ECMWF reanalysis.
• Incorporated geographical coordinates and embedded positional features into multi-angle backscatter processing for enhanced physical consistency.

Main Results:

• Reduced wind speed root mean square error (RMSE) by 9.7%, with 91.5% of samples within 2 m/s difference.
• Reduced wind direction RMSE by 1.6%, increasing the proportion of samples within 20° difference to 90.5%.
• Demonstrated superior global accuracy compared to traditional GMF approaches.

Conclusions:

• The developed CNN models significantly improve the accuracy of sea surface wind retrievals from satellite scatterometers.
• These advancements enable real-time, high-resolution wind mapping, enhancing typhoon forecasting and supporting climate resilience initiatives.