Night-based hyperspectral imaging to study association of horticultural crop leaf reflectance and nutrient status

Abstract

In the literature of hyperspectral remote sensing to assess nutrient status in crops, there is a general lack of studies conducted under greenhouse conditions. This may be attributed to technical issues associated with inconsistent lighting conditions during daytime data acquisitions due to shadows and spectral scattering inside greenhouse structures. In this proof-of-concept study, we developed a novel night-based hyperspectral remote sensing system with attached halogen lighting to study leaf reflectance of bok choy [Brassica rapa L. var Chinensis] and spinach [Spinacia oleracea L. ‘Correnta”] grown under high, medium and low fertilization regimes. The study objectives were to: 1) identify spectral regions in which average leaf reflectance values could be used accurately to characterize crop responses to overall fertilizer regimes, and 2) characterize consistency across crops of associations between crop leaf reflectance and levels of individual macronutrient elements. Our findings were: 1) leaf reflectance could be used to differentiate low versus medium/high fertilization regimes with 75% (bok choy) and 80% (spinach) accuracy, and 2) the following spectral regions: 700–709 nm, 780–787 nm and 817–821 nm were associated with N, K, Mg and Ca levels in bok choy and spinach. Based on comprehensive sensitivity analysis, we demonstrated that classification accuracy was highly similar when 50–80% of the crop reflectance data were used as training data, indicating robustness of the proposed linear discriminant classification models. We believe the proposed sensitivity analysis has broad relevance as a method to thoroughly examine the robustness of reflectance-based algorithms that are used to classify agricultural products.