A dorsiventral leaf radiative transfer model: Development, validation and improved model inversion techniques

A Dorsiventral Leaf Model (DLM) is presented to simulate leaf radiative transfer. DLM was conceived as a plate model with a stochastic distribution of different groups of layers. Leaf asymmetry was modeled by assigning non-uniform distributions of pigments, water and dry matter to palisade and mesophyll layers and by simulating different amounts of light diffusion for adaxially and abaxially incident light. Surface reflections are based on micro-facets theory enabling the simulation of directional-hemispherical reflectance and a range of bidirectional reflectance factors. Adaxial and abaxial optical properties could be accurately simulated for a variety of leaf types with an overall error in reflectance and transmittance below 1.3%. Sensitivity analysis focused on optimizing model inversion schemes improves parameter estimation accuracy. Different inversion schemes were compared for two independent datasets. Results underpin most of the propositions of the sensitivity analysis: (i) masking the near-infrared wavelengths (band weighting) to account for variability in the dry matter composition consistently increased predicted accuracies for dry matter content, (ii) white reflectance measurements (reflectance with a 100% diffusely reflecting background) provided results superior to other optical measurements, making it a valuable and fast alternative and (iii) combining reflectance and transmittance into absorptance however did not result in improvements. Comparisons of DLM with the PROSPECT 5 model indicate an almost equal performance in content estimations. Improvements were thus not related to differences in model structure but to techniques that reduce the impact of leaf structure and compensate for sampling errors and variations in specific absorption spectra. DLM has important potential in the study of leaf radiative transfer and in the integration with canopy radiative transfer models. (C) 2009 Elsevier Inc. All rights reserved.