UAV-Based Prediction of Cover Crop Biomass in Winter Peas Across Planting Dates and Cropping Systems

Winter pea grown as a cover crop in monoculture and mixture with grasses poses remote sensing challenges, including spectral mixing among peas, weeds and intercrop canopies and wide variation in canopy establishment and biomass across planting dates and years. The effect of these factors, cropping system, and model choice on UAV-based biomass prediction and whether such models transfer across years remain untested. Over two seasons (2022–2024) in Freeville, NY, we evaluated winter pea across four planting dates and two cropping systems, combining UAV multispectral imagery with four machine learning models to predict biomass from vegetation indices. Spring imagery achieved the highest cross-validation accuracy (R2 up to 0.894), with early plantings far more predictable than later ones. Crop growth stage at acquisition influenced accuracy more than model choice: a single well-timed spring flight matched or exceeded multi-date combinations. In the evaluation of an independent test set, models ranked plots reliably but did not generalize in absolute magnitude. Low-cost UAV imagery is thus practical for cover crop biomass monitoring, provided accuracy is interpreted relative to the biomass range and recalibrated across years.