Above-Ground Biomass Estimation in Oats Using UAV Remote Sensing and Machine Learning.

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Traditional phenotyping for above-ground biomass is labor-intensive.
• Unmanned aerial vehicles (UAVs) offer high-throughput data collection for vegetation analysis.
• UAV-derived vegetation indices (VIs) show potential for accurate forage yield prediction.

Purpose of the Study:

• To assess the efficacy of UAV-based multispectral data and machine learning in estimating oat biomass.
• To compare the performance of different machine learning algorithms (PLS, SVM, ANN, RF) for biomass estimation.

Main Methods:

• Flown UAVs with multispectral sensors over oat fields in South Dakota during 2019.
• Derived various vegetation indices (VIs) from multispectral imagery.
• Developed and validated biomass estimation models using PLS, SVM, ANN, and RF algorithms.

Main Results:

• Significant positive correlations between VIs and dry biomass were observed in Volga and Beresford, but not in South Shore.
• Machine learning models (PLS, RF, SVM) explained up to 70% of biomass variance in Beresford during the post-heading phase.
• Model performance varied across locations, with lower accuracy and higher errors in Volga and South Shore compared to training data.

Conclusions:

• UAV-based remote sensing combined with machine learning shows promise for estimating oat biomass in breeding nurseries.
• Inconsistent prediction accuracy across different locations is a key limitation.
• Future research should integrate multi-year spectral data and textural features (e.g., crop surface model) for enhanced biophysical parameter estimation.