Biangular-Combined Vegetation Indices to Improve the Estimation of Canopy Chlorophyll Content in Wheat Using Multi-Angle Experimental and Simulated Spectral Data

Canopy chlorophyll content (CCC) indicates the photosynthetic functioning of a crop, which is essential for the growth and development and yield increasing. Accurate estimation of CCC from remote-sensing data benefits from including information on leaf chlorophyll and canopy structures. However, conventional nadir reflectance is usually subject to the lack of an adequate expression on the geometric structures and shaded parts of vegetation canopy, and the derived vegetation indices (VIs) are prone to be saturated at high CCC level. Using 3-year field experiments with different wheat cultivars, leaf colors, structural types, and growth stages, and integrated with PROSPECT+SAILh model simulation, we studied the potential of multi-angle reflectance data for the improved estimation of CCC. The characteristics of angular anisotropy in spectral reflectance were investigated. Analyses based on both simulated and experimental multi-angle hyperspectral data were carried out to compare performances of 20 existing VIs at different viewing angles, and to propose an algorithm to develop novel biangular-combined vegetation indices (BCVIs) for tracking CCC dynamics in wheat. The results indicated that spectral reflectance values, as well as the coefficient of determination (R(2)) between mono-angular VIs and CCC, at back-scattering directions, were mostly higher than those at forward-scattering directions. Mono-angular VIs at +30° angle, were closest to the hot-spot position in our case, achieved the highest R(2) among 13 viewing angles including the nadir observation. The general formulation for the newly developed BCVIs was BCVI(VI) = f × VI((θ1)) − (1 − f) × VI((θ2)), in which the VI was used to characterize chlorophyll status, while the subtraction of VI at θ1 and θ2 viewing angles in a proportion was used to highlight the canopy structural information. From our result, the values of the θ1 and θ2 around hot-spot and dark-spot positions, and the f of 0.6 or 0.7 were found as the optimized values. Through comparisons revealed that large improvements on CCC modeling could be obtained by the BCVIs, especially for the experimental data, indicated by the increase in R(2) by 25.1–51.4%, as compared to the corresponding mono-angular VIs at +30° angle. The BCVI(MCARI[705,750]) was proved to greatly undermine the saturation effect of mono-angular MCARI[705,750], expressing the best linearity and the most sensitive to CCC, with R(2) of 0.98 and 0.72 for simulated and experimental data, respectively. Our study will eventually have extensive prospects in monitoring crop phenotype dynamics in for example large breeding trials.