Estimating leaf and canopy nitrogen contents in major field crops across the growing season from hyperspectral images using nonparametric regression

Estimating leaf nitrogen (N) status is crucial for site-and time-specific crop N management, and can be accomplished more routinely than ever before with the advent of hyperspectral imaging techniques.We conducted field experiments with different nitrogen supply for rice, wheat and maize, in China, in which three types of hyperspectral features were extracted, including canopy reflectance (Ref), vegetation indices (VIs), and texture information (Tex).These features as well as crop development stage (DS) were applied to estimate crop N parameters, using five nonparametric regression algorithms: Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest Regression, Deep Neural Network, and Convolution Neural Network.The performance of PLSR and SVR models was more robust than that of the others and could be improved by incorporating the combined feature set RefVIsTex, although there was no further improvement when also incorporating DS.The prediction of the mass-based leaf N trait, leaf N concentration (LNC), was better than that of the area-based trait, specific leaf N (SLN).The models also predicted specific leaf area (SLA) better than its reciprocal, specific leaf weight.Values of SLN were better predicted via an indirect method (predicted via SLA; denoted as SLN sla ) than via the direct method (SLN dir ).However, when upscaled to canopy, the predicted canopy N content (N canopy ) using SLN dir agreed better with measuredN canopy than that using SLN sla , and even better than the direct predictionN canopy,dir in rice and maize.These results were discussed in view of coupling the predicted leaf and canopy N traits with dynamic crop growth models that can be used for optimizing field N management in sustainable agricultural production.