Mapping water surface roughness in a shallow, gravel-bed river using hyperspectral imagery
Thursday, 18 December 2014
Rapid advances in remote sensing are narrowing the gap between the data available for characterizing physical and biological processes in rivers and the information needed to guide river management decisions. The availability and quality of hyperspectral imagery have increased drastically over the past 20 years and hyperspectral data is now used in a number of different capacities that range from classifying riverine environments to measuring river bathymetry. A fundamental challenge in relating the spectral data from images to biophysical processes is the difficulty of isolating individual contributions to the at-sensor radiance, each associated with a different component of the fluvial environment. In this presentation we describe a method for isolating the contribution of light reflected from the water surface, or sun glint, from a hyperspectral image of a shallow gravel-bed river. We show that isolation and removal of sun glint can improve the accuracy of spectrally-based depth retrieval in cases where sun glint dominates the at-sensor radiance. Observed-vs.-predicted R2 values for depth retrieval improved from 0.56 to 0.68 following sun glint removal. In addition to clarifying the signal associated with the water column and bed, isolating sun glint could unlock important hydraulic information contained within the topography of the water surface. We present data from flume and field experiments suggesting that the intensity of sun glint is a function of water surface roughness. In rivers, water surface roughness depends on local flow hydraulics: depth, velocity, and bed material grain size. To explore this relationship, we coupled maps of image-derived sun glint with hydraulic measurements collected with a kayak-borne acoustic Doppler current profiler along 2 km of the Snake River in Grand Teton National Park. Spatial patterns of sun glint are spatially correlated with field observations of near-surface velocity and depth, suggesting that reach scale hydraulics could be mapped from hyperspectral images. These findings also suggest that aquatic habitats, which are often associated with specific hydraulic conditions and manifested as distinct surface textures, could be mapped quantitatively over large areas using hyperspectral imagery.