Poster Session 3, Wednesday, October 5, 16:00–18:00
Discrimination of river runoff over ice from melt ponds by k-mean clustering of Landsat 8 and MODIS derived remote sensing reflectance spectra
A k-mean clustering approach was used to characterize unique spectra groups in remote sensing reflectance scenes from Landsat 8 OLI and MODIS, for a high-latitude coastal region (Stefansson Sound, Alaska) in late spring when riverine freshet waters discharge onto adjacent landfast sea ice. For each sensor, Calinski-Harabasz clustering, evaluated using squared Euclidian distances, identified 5 significant spectral clusters whose spectral shapes resemble those of reference spectra from pixels known to contain dry ice, wet snow, melt ponds, open water, and freshet overflow. Transects taken across known melt pond and freshet flood pixel regions show discrepancies in the reflectance values of the blue bands, with freshet pixels containing much lower Landsat OLI B1/B2 ratios than are found in reference scenes containing pixels with melt ponds. These lower values likely reflect the presence of higher concentrations of colored dissolved organic matter (CDOM) in the freshet waters which absorbs more strongly in the shorter wavelengths. A principal component analysis of each scene indicated that 97% and 80% of the spectral variability is described by the first component in Landsat OLI and MODIS, respectively. This dominant component likely represents the proportion of water-like absorbing constituents in each pixel. For the MODIS sensor, the second component (accounting for 18% variance) improves the distinction between the melt-pond like pixels and the freshet pixels, suggesting a potential approach to better spatially differentiate these two regimes in complex, ice-covered high latitude regions using remote sensing reflectance.
Sam Laney, Woods Hole Oceanographic Institution, [email protected]