GC23K-1242
Constructing Virtual Forest Scenes for Assessment of Sub-pixel Vegetation Structure From Imaging Spectroscopy

Tuesday, 15 December 2015
Poster Hall (Moscone South)
Wei Yao1, Jan A van Aardt2, Paul Romanczyk1, David Kelbe3, Martin van Leeuwen1 and Thomas U Kampe4, (1)Rochester Institute of Technology, Rochester, NY, United States, (2)Rochester Institute of Tech., Rochester, NY, United States, (3)Oak Ridge National Laboratory, Oak Ridge, TN, United States, (4)NEON, Boulder, CO, United States
Abstract:
Assessment of vegetation structure via remote sensing modalities has a long history for a range of sensor platforms. Imaging spectroscopy, while often used for biochemical measurements, also applies to structural assessment in that the Hyperspectral Infrared Imager (HyspIRI), for instance, will provide an opportunity to monitor the global ecosystem. Establishing the linkage between HyspIRI data and sub-pixel vegetation structural variation therefore is of keen interest to the remote sensing and ecology communities. NASA's AVIRIS-C was used to collect airborne data during the 2013-2015 time frame, while ground truth data were limited to 2013 due to time-consuming and labor-intensive nature of field data collection. We augmented the available field data with a first-principles, physics-based simulation approach to refine our field efforts and to maintain larger control over within-pixel variation and associated assessments.

Three virtual scenes were constructed for the study, corresponding to the actual vegetation structure of the NEON’s Pacific Southwest site (Fresno, CA). They presented three typical forest types: oak savanna, dense coniferous forest, and conifer manzanita mixed forest. Airborne spectrometer and a field leaf area index sensor were simulated over these scenes using the Digital Imaging and Remote Sensing Image Generation (DIRSIG) Model, a synthetic image generation model. After verifying the geometrical parameters and physical model with those replicative senses, more scenes could be constructed by changing one or more vegetation structural parameters, such as forest density, tree species, size, location, and within-pixel distribution.

We constructed regression models of leaf area index (LAI, R2=0.92) and forest density(R2=0.97) with narrow-band vegetation indices through simulation. Those models can be used to improve the HyspIRI’s suitability for consistent global vegetation structural assessments. The virtual scene and model can also be used in other studies, for example, to investigate the impact of human activities on ecosystems using imaging spectroscopy. More detailed results will be presented at the conference.