Enhancing estimation of cover crop biomass using field-based high-throughput phenotyping and machine learning models

Area of Science:

• Agricultural Science
• Agronomy
• Plant Science

Background:

• Cover crops enhance agricultural sustainability by reducing soil erosion, suppressing weeds, and increasing soil carbon.
• Aboveground biomass (AGB) is a critical factor for cover crop efficacy, yet comprehensive field-based quantification is lacking.
• High-throughput phenotyping (HTP) and machine learning (ML) offer potential for accurate AGB estimation.

Purpose of the Study:

• To quantify the aboveground biomass (AGB) of five cover crop species using HTP and ML models.
• To compare the performance of four ML models (RFR, SVR, PLSR, ANN) for AGB estimation.
• To identify key spectral, morphological, and environmental features for predicting AGB.

Main Methods:

• A two-year field experiment was conducted in Eastern Nebraska, USA, with five cover crop species.
• Destructive AGB sampling was performed multiple times each spring season.
• Morphological, spectral, thermal, and environmental data were collected using sensors and used as input features for ML models.

Main Results:

• Partial Least Squares Regression (PLSR) achieved the highest prediction accuracy (R² = 0.84, NRMSE = 8.87%) for AGB estimation.
• Spectral features including Normalized Difference Red Edge (NDRE), Solar-induced Fluorescence (SIF), and Normalized Difference Vegetation Index (NDVI) were identified as important predictors.
• Artificial Neural Network (ANN) showed strong performance when using fewer input features, particularly morphological and spectral data.

Conclusions:

• HTP and ML techniques are feasible for accurate AGB estimation of multiple cover crop species.
• PLSR is a suitable model for predicting cover crop AGB, with spectral features being highly informative.
• Future research should involve multi-location, multi-year sampling to further enhance model robustness.