Modelling biochemical processes in orchards at leaf- and canopy-level using hyperspectral data

Abstract

This research was conducted to evaluate the potential and limitations of hyperspectral remote sensing to detect iron deficiency in capital-intensive multi-annual crop systems, e.g. peach orchards. The noted deficiency can be regarded as a proxy for deviations from optimal plant functioning, while detection of such deviations is in turn of significant importance to monitoring and modelling efforts of orchards as production systems. Hyperspectral leaf, canopy, and airborne reflectance measurements were acquired in a peach (Prunus persica L.) orchard in Zaragoza, Spain. Leaf- and canopy-level data were collected with a handheld spectroradiometer (ASD, Inc.), while the AHS-160 hyperspectral sensor provided airborne data. Blocks of trees were treated with different amount of iron chelates (Sequestrene) which created a dynamic range of chlorophyll concentration as measured in leaves. Hyperspectral measurements at leaf-level were carried out to characterize the physiological aspects of nutrient stress, as opposed to the evaluation of plant nutrient status at the complete plant-level. Stressinduced physiological changes make stress detection at the leaf-level possible at an early stage of suboptimal photosynthetic functioning. Airborne imagery, however, is difficult to interpret due to altering illumination angles, BRDF effects, and intervening atmospheric light interactions resulting in an alteration of the vegetative reflectance spectrum. Although many studies have implemented hyperspectral analysis of nutrient status at large scales, this research field is still in its infancy phase, since the link between airborneand leaf-level measurements is lacking. This inevitably makes the physiological interpretation of existing hyperspectral research more complex. The multi-level (leaf, canopy, and airborne) approach presented here enabled the assessment of vegetation indices and their relationship with pigment concentration at each monitoring level. Pertinent classical chlorophyll-related vegetation indices were tested and new indices were extracted from the spectral profiles by means of band reduction techniques and narrow-waveband rationing, which involved all possible 2-band combinations. Robustness was evaluated by studying the index performance for datasets of increasing complexity, from leaf- to canopy- and airborne-level. Physiological interpretations extracted from leaf-level experiments were extrapolated to canopy- and airborne level. The measured spectra and estimated biochemical parameters were related via inversion of a linked directional homogeneous canopy reflectance model (ACRM) and the PROSPECT leaf model. Numerical model inversion was conducted by minimizing the difference between the measured reflectance samples and modelled reflectance values. An improved optimization method is presented. Results are compared with a simple linear regression analysis, linking chlorophyll to the reflectance measured at the leaf level and at the Top of Canopy (TOC), while optimal band regions and bandwidths also were analyzed.