Linear and nonlinear imaging spectrometer denoising algorithms assessed through chemistry estimation

Hyperspectral sensing of forest chemistry can provide indicators of forest health. Foliar pigments are directly involved with the photo synthetic process and, therefore, are intimately tied to vegetation vigor [1]. Data from the University of Victoria´s Airborne Imaging Spectrometer for Applications (AISA) were acquired in 2006 for the Greater Victoria Watershed District (GVWD). Minimum Noise Fraction (MNF) [2], an algorithm based on linear principal components, and a nonlinear local geometric projection algorithm (NL-LGP) [3] were used to denoise this hyperspectral dataset. The initial reflectance and the two denoised datasets were used to generate estimates of foliar chemistry, which were evaluated with field measurements taken prior to the AISA acquisition. Initial analysis was performed at the plot level. Data that were denoised produced marginally less accurate chlorophyll-a estimates than the original data set. The NL-LGP denoising algorithm provided better chemistry mapping at finer spatial resolutions than MNF or the original dataset.