A modeling approach to assess the robustness of spectrometric predictive equations for canopy chemistry

Abstract We present a 3D modeling approach to assess the robustness of remotely derived spectrometric equations predictive of forest chemistry (cellulose, lignin, proteins) to structural variables (tree ground cover, leaf area index: LAI, understory) and to the view direction. Our methodology uses two radiative transfer models that operate at leaf (PROSPECT) and canopy (DART) levels. It includes three stages: (1) simulation of short wave infrared bidirectional reflectances of a “reference scene” with constant architecture and variable chemistry; (2) establishment of predictive relationships of chemicals with stepwise regressions; (3) assessment of the robustness of these relationships for scenes with variable structures. Results stressed that predictive relationships are influenced by canopy structure and view direction. Their reliability decreases with increasing heterogeneity of the understory and also if tree cover or LAI decreases. On the other hand, they tend to remain robust if tree cover and LAI increase, that is if the influence of the understory on canopy reflectance decreases. Their reliability increases when the view direction becomes more oblique, except for the hotspot and specular directions. Finally, this work stresses factors that can explain the difficulty to establish robust predictive relationships with remote sensing data. It shows also that 3D modeling approaches can be an efficient tool for studying forest chemistry from space, for instance, in order to assess the domain of validity of predictive relationships.