Hyperspectral imaging predicts yield and nitrogen content in grass-legume polycultures

Successful use of hyperspectral imaging technology to progress precision agriculture is highly dependent on calibration on species of interest. To date, high-throughput hyperspectral imaging to predict plant growth and nutrient content has largely been limited to single-species cultivations. Therefore, this study aimed to calibrate a range of agronomic traits in mixed cultivations to hyperspectral image data. It successfully demonstrated that hyperspectral imaging can predict the plant traits biomass (g), foliar nitrogen (N) concentration (mg g(-1)) and N yield (mg), in grass and legume monocultures and polycultures in response to differential N and phosphorus (P) fertilization in a controlled greenhouse experiment. Visible light and near infrared (VNIR) and short wavelength infrared (SWIR) input resulted in only minor image misclassification (0.02%) for the green plants from the background regardless of species. The trained partial least square regression (PLSR) models VNIR-HH (hyper-hue) and SWIR had the lowest misclassification errors of 3.16% and 9.56% and were used for the grass-legume classification. For grass, there was good agreement between the mixed-effect models derived from the laboratory, and the PLSR models from hyperspectral measurements, except for the effect of N x P on N yield. Legume model agreement was not as precise as grass, likely because fertilizer-driven treatment effects on the measured traits were not as clear. Key wavelengths contributing to the strength of the PLSR models for predicting N content and biomass were identified from this study. Effective calibration of growth and nutrient uptake traits against hyperspectral data in mixed cultivations under controlled conditions is an important contribution towards improving remote sensing technologies for broader application in polyculture field cropping.

Automated summary

Successful calibration of agronomic traits in mixed cultivations using hyperspectral imaging data is important for advancing precision agriculture. This study demonstrated the capability of hyperspectral imaging to predict plant biomass, foliar nitrogen concentration, and nitrogen yield in grass and legume monocultures and polycultures under differential nitrogen and phosphorus fertilization. The study also identified key wavelengths contributing to the predictive power of the models. This research contributes to the improvement of remote sensing technologies for broader application in polyculture field cropping.