An overview of crop nitrogen status assessment using hyperspectral remote sensing: Current status and perspectives

Abstract Nitrogen (N) is significantly related to crop photosynthetic capacity. Over-and-under-application of N fertilizers not only limits crop productivity but also leads to negative environment impacts. With such a dilemma, a feasible solution is to match N supply with crop needs across time and space. Hyperspectral remote sensing has been gradually regarded as a cost-effective alternative to traditional destructive field sampling and laboratory testing for crop N status determination. Hyperspectral vegetation indices (VIs) and linear nonparametric regression have been the dominant techniques used to estimate crop N status. Machine learning algorithms have gradually exerted advantages in modelling the non-linear relationships between spectral data and crop N. Physically-based methods were rarely used due to the lack of radiative transfer models directly involving N. The existing crop N retrieval methods rely heavily on the relationship between chlorophyll and N. The underlying mechanisms of using protein as a proxy of N and crop protein retrieval from canopy hyperspectral data need further exploration. A comprehensive survey of the existing N-related hyperspectral VIs was made with the aim to provide guidance in VI selection for practical application. The combined use of feature mining and machine learning algorithms was emphasized in the overview. Some feature mining methods applied in the field of classification and chemometrics might be adapted for extracting crop N-related features. The deep learning algorithms need further exploration in crop N status assessment from canopy hyperspectral data. Finally, the major challenges and further development direction in crop N status assessment were discussed. The overview could provide a theoretical and technical support to promote applications of hyperspectral remote sensing in crop N status assessment.