Non-destructive estimation of crop nitrogen using Proximal Hyperspectral Remote Sensing

Hyperspectral remote sensing made a significant breakthrough in precision agriculture by providing accurate estimates of important crop health indicators. Canopy nitrogen content (CNC) is one of the major crop traits that need to be monitored in a timely and accurate manner for making timely decisions regarding input applications. In the present study, a hybrid retrieval model was developed by combining the physically based PROSAIL-PRO radiative transfer model (RTM) and Gaussian process regression (GPR) for the non-destructive estimation of CNC in wheat crops in the experimental fields of ICAR—Indian Agricultural Research Institute, New Delhi, India. The proximal non-imaging hyperspectral data in the wavelength region of 400-2500nm and spectral resolution of 1nm were collected in situ using an ASD FieldSpec Spectroradiometer. Principal component analysis (PCA) and active learning (AL) techniques were employed to reduce the dimensionality in the spectral and sampling domains, respectively. The validation results showed that this hybrid approach successfully retrieved CNC with an R² of 0.71 and NRMSE of 14.36%. Moreover, a low relative uncertainty below 35% supports the broad applicability and portability of the retrieval model. Upscaling this hybrid retrieval workflow to farmers’ fields can facilitate operational delivery of important crop traits, enabling effective monitoring of crop growth and production.