Evaluating visible derivative spectroscopy by varimax-rotated, principal component analysis of aerial hyperspectral images from the western basin of Lake Erie

Abstract

The Kent State University (KSU) spectral decomposition method provides information about the spectral signals present in multispectral and hyperspectral images. Pre-processing steps that enhance signal to noise ratio (SNR) by 7.37–19.04 times, enables extraction of the environmental signals captured by the National Aeronautics and Space Administration (NASA) Glenn Research Center's, second generation, Hyperspectral imager (HSI2) into multiple, independent components. We have accomplished this by pre-processing of Level 1 HSI2 data to remove stripes from the scene, followed by a combination of spectral and spatial smoothing to further increase the SNR and remove non-Lambertian features, such as waves. On average, the residual stochastic noise removed from the HSI2 images by this method is 5.43 ± 1.42%. The method also enables removal of a spectrally coherent residual atmospheric bias of 4.28 ± 0.48%, ascribed to incomplete atmospheric correction. The total noise isolated from signal by the method is thus <±7% based on analysis of multiple swaths. The method is essentially independent of the order of operations, extracting the same spectral components within error in all cases, spatial patterns that are very similar and explaining nearly identical amounts of information from each image. Based on comparison between multiple swaths the uncertainty in the variance associated with each component averages ±1.69% and is as low as ±0.08% and in all cases <±3.15%.