B31F-0087:
Characterization of Permafrost Degradation and Plant Communities Using Hyperspectral Reflectance

Wednesday, 17 December 2014
AJ Garnello1, Daniel Finnell2, Michael W Palace3, Jin Wu1, Lucie C Lepine4, Patrick M Crill5 and Ruth K Varner6, (1)University of Arizona, Tucson, AZ, United States, (2)Virginia Commonwealth University, Center for Environmental Studies, Richmond, VA, United States, (3)Complex System Research Center, Durham, NH, United States, (4)University of New Hampshire, Durham, NH, United States, (5)Stockholm University, Stockholm, Sweden, (6)Univ New Hampshire, Durham, NH, United States
ePoster
Abstract:
Annual temperatures in subarctic regions are increasing, resulting in rapid disappearance of long-standing permafrost. This directly affects plant community structure, increases soil active layer thickness, and changes greenhouse gas emissions. The change in carbon cycling alters climate feedback cycles, calling for accurate, efficient, permafrost degradation monitoring techniques in Earth Systems models.

At Stordalen Mire, 68°21’N and 19°02’E, 50 randomized one-square meter plots were measured for vegetation composition and hyperspectral reflectance using an ASDinc FieldSpec4. Plots were classified into one of five site-types based on vegetation composition and microtopography, with each site type representing differing stages of permafrost degradation.

A discrete forward multivariate model with stepwise selection successfully paired 49 of 49 plots with their site types using the reflectance values for the wavelengths centered at 1115nm, 1190nm, 1334nm, 1340nm, and 1813nm. These wavelengths correspond to reflectance features resulting from varying plant intercellular structure (1115nm, 1190nm), and particular in-cell air-water interactions (1334nm, and 1340nm 1813nm). A decision tree partitioning the reflectance feature at 1334nm across all site types resulted in a 4-split tree with an R2 of .793, and a 10-fold cross-validation R2 of .712.

A discrete forward multivariate analysis model successfully paired 84% of plots with their correct site types using the percent cover of five dominant species (Empetrum nigrum, Eriophorum vaginatum, Rubus chamaemorus, Sphagnum spp, Open Water). A five-split partition of site type with these five dominant species returned an R2 of .893, and a five-fold cross-validation R2 of .859.

A least squares regression model used 5 species-specific spectral bands distinguishing content of plant pigment, cell-water, and physical structure to predict percent cover of four primary species: E. nigrum (R2 .82), R. chamaemorus (R2 .64), Sphagnum spp. (R2 .92), and Open Water (R2 .79).

Results indicate the usefulness of hyperspectral measurements for estimating vegetative composition. Continued work will examine field spectrometer measurements with those from the EO-1 Hyperion sensor in an effort to scale-up permafrost degradation monitoring.

Back to: Remote Sensing of Vegetation Function and Traits I Posters
<< Previous Abstract | Next Abstract >>