Retrieving Biophysical And Biochemical Crop Traits Using Continuum-Removed Absorption Features From Hyperspectral Proximal Sensing

In recent years, hyperspectral sensors, thanks to their very high spectral resolution, has attracted increasing attention for monitoring crop traits across the precision agricultural settings. In this paper, we evaluated the potential of hyperspectral spectroscopic indicators, by analyzing several properties of continuum-removed absorption features retrieved based on segmented upper hull and inflection points of the red-edge region, to model various biophysical and biochemical traits of alfalfa and rice crops. The studied crop traits were leaf area index (LAI), biomass, canopy water content, plant/leaf nitrogen concentration (N), and chlorophyll content. We denoted significant relationships between each examined crop trait and a particular property of the absorption features. We also underlined that the depth of absorption features is not the sole important element in describing crop traits. This approach allowed us, as a secondary goal, to test a proof of concept to effectively perform temporal monitoring of crop traits with continuous automatic proximal sensing. In particular, we estimated the temporal trends of LAI and N concentration, and derived the nitrogen nutritional index (NNI), throughout the cropping season for a rice field. Exploiting such properties of continuum-removed absorption features as input in machine learning techniques may help to establish in the future more robust predictive models. It would also be advisable to evaluate the methodology on other feature properties such as asymmetry to fully exploit the modelling of physically based diagnostic spectral region.