Validating nonlinear mixing models: Benchmark datasets from vegetated areas

Our understanding of nonlinear mixing events in vegetated areas is currently hampered by a pertinent lack of well-validated datasets. Most quantification and modeling efforts are based on theoretical assumptions or indirect empirical observations. In this study, a physically based ray tracer was used to create simulated hyperspectral datasets of vegetative systems. This model incorporates multiple scattering effects, and nonlinear mixing behavior can be observed in the rendered data. The main benefit of the ray-tracer is that we were able to demonstrate with in situ measurements that both the nature and the intensity of the nonlinear mixing events are realistically modeled. Different ray-tracer datasets will be made available to the wider scientific community as a benchmark dataset to test and validate new and existing unmixing methodologies. In this contribution, we would like to present the structure of these datasets, and show how they can be used to evaluate nonlinear mixing models. In addition, and maybe even more important, we would like to draw the attention to the limitations of the data, as well point out the assumptions made in the construction of the data.