Spectral Mixture Analysis of Amazon Floodplain Water Surface Reflectance Using Hyperion/EO-1 for the Comprehension of Temporal Variability of Water Compositon
Rudorff, INPE, email@example.com
M. L. M.
Novo, INPE, firstname.lastname@example.org
Galvão, INPE, email@example.com
The Amazon floodplain water composition undergoes intense variations along the year as a response to the annual flood pulse. The present study analyses the spectral mixtures of the optically active substances (OAS) in the Amazon floodplain waters by using spaceborne hyperspectral images. The test site was located upstream the confluence of Amazon (white water) and Tapajós (clear-water) rivers, where two EO-1 Hyperion sensor images were acquired. The first image was acquired September 16, 2001, during the period of the outflow of water from the floodplain lakes to the Amazon River, due to its receding level. The second image was acquired June 23, 2005, at the end of the high water period. A field campaign was carried out between June 23 and 29, 2005 to collect radiometric and limnological data almost simultaneously to the acquisition of this image. The images were pre-processed to remove stripes of abnormal pixels and were converted from radiance to surface reflectance values, thus, correcting the effects of atmospheric absorption and scattering. A sequential procedure with the techniques of Minimum Noise Fraction (MNF), Pixel Purity Index (PPI) and n-dimensional visualization of the MNF feature space was employed, within the spectral range of 457-885 nm, to select end-members from both images. A single set of end-members was gathered to represent the following spectrally unique water masses: clear-water; dissolved organic matter; suspended sediments; and phytoplankton. The Linear Spectral Unmixing (LSU) algorithm was applied to the images to map the spatial distribution of the four types of water masses, in terms of sub-pixels fractional abundances. The suspended sediment and phytoplankton concentrations in floodplain lakes showed a general tendency to increase towards the receding period, which was clearly evidenced by the results. Important non-linear spectral mixture effects were observed in the complex Amazon floodplain waters, which should be accounted for to achieve better estimates.