3D-Vegetation Laboratory: Science and Modeling Support for Accuracy Assessment and Prototyping of EO Data and Products
Morsdorf, Felix1; Leiterer, Reik1; Schaepman, Michael E.1; Brazile, Jason2; Pfeifer, Nobert3; Hollaus, Markus3; Disney, Mat4; Lewis, Phil4; Gastellu-Etchegorry, Jean-Phillipe5; Koetz, Benjamin6
1University of Zürich, SWITZERLAND; 2netcetera, SWITZERLAND; 3Vienna University of Technology, AUSTRIA; 4University College London, UNITED KINGDOM; 5CESBIO Toulouse, FRANCE; 6European Space Agency, ITALY
Upcoming satellite missions will routinely deliver high-resolution images globally at unprecedented accuracy and resolution. They will provide end-users and scientific communities alike with data and products that may contribute to improving observational based approaches such as forest and vegetation monitoring in the context carbon cycle and land-atmosphere processes assessment.
Conventional vegetation parameter retrieval approaches include validation concepts based on 'homogenous' areas, while homogeneity criteria are mostly being discussed at a semantic and not physical level. This is due to the complex interrelated nature of multiple properties originating from combined biochemical and biophysical composition of vegetation. Particularly structural properties (as an important aspect to describe vegetation) are known to have a significant impact on radiometric measurements, while being relevant itself for a number of ecosystem functions and variables.
We are establishing a novel validation site concept, accounting for site deficiencies in terms of homogeneity by providing an exhaustive description of the site using structural, physical and biochemical measurements simultaneously. This in combination with two different radiative transfer models. This effort will provide scientific support to bridge the gap between EO data and advanced vegetation products. We will achieve this by fully parameterizing radiative transfer models based on terrestrial and airborne laser scanning, biochemical and structural sampling and scaling from leaf/needle to the stand level. The reconstruction approach is based on a combination of automatic segmentation and manual digitization (e.g. automatic tree locations from airborne laser scanning and manually derived branching topology from terrestrial laser scanning) and makes extensive use of replication of basic 3D objects (e.g. trees, shoots) in order to populate the full 300 m by 300 m scenes.
All these efforts are seamlessly integrated in a software toolbox, allowing expert and end-users alike to model parameter variations and simulation of air- and satellite-borne optical instruments of these sites at stand level. The toolbox will be in particular capable to simulate and validate advanced (Level 2) Sentinel-2 vegetation products. Since the forward modeled data represent a physically existing ecosystem (current locations are at 2 different FLUXNET sites), retrieval algorithms can be tested using satellite measurements as well as modeled stands using exactly the same approach (algorithm).
The toolbox comprises of three main components, i) radiative transfer models (librat and DART), ii) a 3D scene reconstruction of two forested sites in central Europe (one deciduous and one coniferous) and iii) an exhaustive EO data set acquired over these sites covering a large range of spectral and spatial resolutions from existing EO instruments. The latter potentially enables the testing of retrieval algorithms across missions and platforms. The toolbox will be accompanied with documentation regarding the interfaces for 3D parameterization, so that interested users may provide additional sites.