Satellite-based monitoring of forage quality in grasslands of the United Kingdom using sentinel-2 data and random forest regression

Summary

This study presents a remote sensing approach combining Sentinel-2 optical imagery with random forest regression to predict forage quality across UK grasslands, addressing the limitations of destructive field sampling. Calibrated on over 9,500 georeferenced observations collected between 2020 and 2022 at the North Wyke Farm Platform, the models achieved strong predictive accuracy (R² 0.77–0.86) for crude protein, water-soluble carbohydrates, and fibre fractions. The findings demonstrate that spatial variation between paddocks was more pronounced than seasonal variation, and that improved pastures exhibited superior forage quality compared with permanent pastures.

Regional applicability

This study was conducted directly within United Kingdom grassland systems (southwest England), making the findings directly applicable to UK pasture management and forage monitoring. The authors note that the calibrated models are suitable for forage systems with species composition and quality ranges similar to the North Wyke dataset but should not be directly transferred to other forage systems without recalibration.

Key measures

Crude protein (CP), water-soluble carbohydrates (WSC), neutral detergent fibre (NDF), acid detergent fibre (ADF); R² and RMSE model performance metrics; Sentinel-2 spectral reflectance data (visible, near-infrared, and red-edge bands)

Outcomes reported

The study developed and validated random forest regression models using Sentinel-2 satellite data to predict four forage quality attributes (crude protein, water-soluble carbohydrates, neutral detergent fibre, and acid detergent fibre) in UK grasslands. Model performance showed R² values between 0.77 and 0.86 with low RMSE values, demonstrating high predictive accuracy for scalable forage monitoring.

Topic tags