Enhancing early-season soybean identification through optical and SAR time-series integration

Area of Science:

• Agricultural Remote Sensing
• Geospatial Analysis
• Crop Monitoring

Background:

• Accurate soybean distribution mapping is vital for food security in China.
• Traditional surveys are labor-intensive and limited in scope.
• Satellite remote sensing offers large-scale, cost-effective crop monitoring but faces challenges with spectral similarity between crops.

Purpose of the Study:

• To develop and evaluate a multi-source remote sensing approach for accurate soybean classification.
• To overcome spectral similarity challenges between soybean and maize using integrated data.
• To improve the timeliness and accuracy of soybean mapping.

Main Methods:

• Integration of Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical time-series imagery.
• Utilizing statistical descriptors, harmonic fitting, phenological indicators, and radar-based features.
• Employing a random forest classifier for multi-source classification and temporal analysis.

Main Results:

• The multi-source fusion approach achieved high accuracy (OA: 96.85%, Kappa: 0.9493, F1-score: 95.84%).
• Synthetic Aperture Radar (SAR) data significantly improved classification during the flowering stage, increasing F1-score by up to 6.96%.
• Earliest Identifiable Time (EIT) was advanced to Day of Year (DOY) 210, approximately 20 days earlier than with optical data alone.

Conclusions:

• Multi-source remote sensing effectively enhances soybean classification accuracy and timeliness.
• The integrated approach provides reliable support for precise soybean mapping and in-season monitoring.
• This method offers a valuable solution for crop classification under complex climatic conditions.