HR: 1340h
AN: B53B-1171    [Abstracts]
TI: A bottom-up geostatistical approach for quantifying landcover in Asian desert ecosystems and implications for global and climate models: a case study in Afghanistan utilizing a unique hyperspectral dataset
AU: * Shreve, C M
EM: cms9kq@virginia.edu
AF: University of Virginia, 291 McCormick Road PO Box 400123, Charlottesville, VA 22904-4123, United States
AU: Okin, G S
EM: okin@ucla.edu
AF: University of California, Los Angeles, 1255 Bunche Hall Box 951524, Los Angeles, CA 90095, United States
AU: Bowles, J
EM: Jeffrey.Bowles@nrl.navy.mil
AF: Naval Research Laboratory, 4555 Overlook Avenue, S.W., Washington, DC 20375-5337, United States
AU: Gardner, J
EM: Joan.Gardner@nrl.navy.mil
AF: Naval Research Laboratory, 4555 Overlook Avenue, S.W., Washington, DC 20375-5337, United States
AB: Political tensions, rough terrain, and remoteness have lead to a gap in the ecological understanding cold, mountainous deserts of Asia. Remote sensing is a time- and cost-efficient way to understand the spatial distribution and temporal dynamics of plant and snow cover in these regions. Here, a unique high-resolution hyperspectral dataset from Afghanistan is employed to classify ground cover at high resolution. The hyperspectral data was taken using a CASI-1500 Visible Near InfraRed (VNIR) spectrometer. The instrument was run in a mode with 1518 crosstrack pixels and 72 spectral bands between 380 and 1050 nm. The GSD was controlled by the altitude above ground level and aircraft speed, which varied resulting in GSD between 4 and 6 meters. Geolocation was provided by a CMIGITS II and the resulting accuracy will be better than 40 m. Atmospheric conditions were challenging and proper atmospheric compensation of the data remains a challenge. A bottom- up geostatistical approach for quantifying the coverage of vegetation and snow will be applied to establish the practical limits of coarse resolution MODIS data for classifying vegetation and snow cover, a scale suitable for monitoring large regions and for modeling. A Multiple Endmember Linear Spectral Mixture Algorithm (MESMA) will be applied to classify land cover. Semivariograms at the multispectral (30 m) and coarse resolution scale (1 km) will be compared with simulated variograms using hysperspectral data. Patches of vegetation and snow cover used for spatial comparison will be identified in the image and characterized using object-oriented image analysis software. The relative amount of cover will be determined using block-kriging and compared between scenes with statistical tests. Insight gained from this analysis can be applied to improve existing data products and can be applied for carbon budget and climate change models.
DE: 0428 Carbon cycling (4806)
DE: 1855 Remote sensing (1640)
SC: Biogeosciences [B]
MN: 2007 Fall Meeting