Multi dimensional variable influence mechanism analysis for wheat biomass estimation using fused UAV spectral and...

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Accurate wheat biomass estimation is vital for crop monitoring and yield prediction.
• Unmanned aerial vehicle (UAV)-based remote sensing offers potential for non-destructive biomass assessment.
• Understanding variable contributions in machine learning models for biomass estimation is crucial.

Purpose of the Study:

• To estimate wheat biomass using fused spectral, vegetation index, and canopy height data from UAVs.
• To compare the performance of four machine learning algorithms (XGBoost, RFR, SVR, LASSO) for biomass estimation.
• To elucidate the contribution mechanisms of different variables in wheat biomass prediction.

Main Methods:

• UAV multispectral imagery was used to derive spectral reflectance and vegetation indices.
• Canopy height data was extracted from UAV-derived structural information.
• Wheat biomass was estimated across four growth stages using XGBoost, RFR, SVR, and LASSO models.
• SHAP analysis was employed to determine variable importance and contribution.

Main Results:

• XGBoost demonstrated the highest accuracy in wheat biomass estimation (R² = 0.919).
• Canopy height was identified as the most significant predictor, followed by spectral indices (R842, GNDVI).
• Structural and spectral variables were found to have complementary roles in biomass estimation.

Conclusions:

• UAV-based remote sensing combined with machine learning provides a robust framework for wheat biomass estimation.
• Canopy height and spectral information are critical for accurate biomass prediction.
• This study enhances mechanistic understanding of variable contributions in precision agriculture applications.