Multispectral red-edge indices accurately estimate nitrogen content in winter cereal cover crops

Abstract

Winter cover crops reduce erosion and nutrient runoff from agricultural systems. Although cereal cover crops can decrease field nitrate leaching by 50%–95%, the magnitude of this reduction varies within and between fields, making it challenging to monitor the impact of cover crops on nitrate leaching at large spatial extents. Satellite remote sensing using red-edge bands has been shown to effectively estimate crop nitrogen (N) content (kg ha⁻¹) in later growth-stage crops with a closed canopy. In this study, we evaluated 15 spectral indices derived from Sentinel-2 imagery to estimate N concentration (%) and content (kg ha⁻¹) of cereal cover crops, using 1627 destructive samples collected from 2018 to 2023 in Maryland. Observed N content ranged from 0.1 to 214.7 kg ha⁻¹, while N concentration ranged from 0.6% to 5.5%. The 15 indices considered were poor predictors of N concentration (adj. R² = 0.089, root mean squared error [RMSE] = 0.802%), but were more successful at measuring N content (biomass × N concentration). Delta red-edge (ΔRE) was the best predictor of N content (adj. R² = 0.748, RMSE = 13.10 kg ha⁻¹ from cross-validation with 80% train and 20% test splits iterated 100 times) using samples with imagery collected within ±4 days of destructive sampling (n = 1110). Our findings indicate that longer red-edge wavelengths (783 and 740 nm) are more suited for estimating N content in cereal cover crops compared to shorter red-edge wavelengths, which have been shown to be more sensitive to biomass. Leave-one-year-out cross-validation demonstrated that the relationship between ΔRE and N content was robust across all four cover crop sampling years included in the study (adj. R² = 0.700–0.769, RMSE = 10.70–15.40 kg ha⁻¹). Regression model performance improved with the addition of multiple predictors, including biomass (estimated from Normalized Difference Vegetation Index), weather variables (adj. R² = 0.765, RMSE = 12.37 kg ha⁻¹), management variables (species, season, adj. R² = 0.772, and RMSE = 12.13 kg ha⁻¹), and biophysical variables (height, fractional ground cover, adj. R² = 0.818, and RMSE = 10.29 kg ha⁻¹). These findings demonstrate the feasibility of quantifying N content in cereal cover crops using a red-edge-based spectral index across large geographic extents and indicate the inclusion of additional predictors, such as weather and management data, improves model accuracy. This work has implications for quantifying reductions in N leaching associated with cover crops, aiding in policymaking and evaluation of conservation programs that impact water bodies such as Chesapeake Bay.