HR: 15:10h
AN: H53K-07    [Abstracts]
TI: Remote sensing of river channel morphology with passive optical image data
AU: * Legleiter, C J
EM: carl@geog.ucsb.edu
AF: Department of Geography, UC Santa Barbara, Santa Barbara, CA 93106-4060, United States
AU: Roberts, D A
EM: dar@geog.ucsb.edu
AF: Department of Geography, UC Santa Barbara, Santa Barbara, CA 93106-4060, United States
AB: Although LiDAR provides precise topographic information for terrestrial surfaces, data from submerged areas are unreliable because near-infrared laser pulses are strongly absorbed by water. By retrieving water depth from passive optical image data, a more complete characterization of river morphology can be achieved. Radiative transfer simulations indicated a sound physical basis for estimating depth from measured spectral radiance: the log of the ratio of two bands yields an image-derived quantity linearly related to depth because variations in bottom albedo affect both bands similarly whereas light is much more strongly attenuated by the water column in one band than in the other. We corroborated these results by collecting field spectra along Soda Butte Creek, WY. To determine an optimal pair of bands for depth retrieval, all possible band ratios were regressed against measured depths. Maximum R² values ranged from 0.79 to 0.98 for seven independent data sets, including one (n = 55) obtained under turbid conditions. Aggregating all of our data (n = 199) resulted in an R² value of 0.80 for regression of ln(R₅₇₀/R₇₁₆) values against depth. Convolution to match the spectral response of various remote sensing instruments did not significantly degrade the ability to estimate depth; for a multispectral satellite, R² was only reduced to 0.70. Regression of log(green/red) values from digital aerial photography of our field area against point measurements of depth along 52 cross-sections yielded an average R² of 0.40; accuracy improved with stream size from third- to fifth-order. R² values for three sites surveyed in greater detail ranged from 0.23 to 0.37. This reduction in accuracy can be attributed to registration error, poor spatial and radiometric resolution, and color balancing and image compression of these publicly available data. Applying a log band ratio-depth relationship derived from our field spectra to a hyperspectral scene of the Lamar River, WY, produced a hydraulically reasonable depth map, but validation data are lacking. We are currently developing a forward image model for examining how channel morphology, imaging conditions, and sensor characteristics interact to determine the accuracy and precision with which depth can be mapped.
DE: 1825 Geomorphology: fluvial (1625)
DE: 1855 Remote sensing (1640)
DE: 1856 River channels (0483, 0744)
DE: 1894 Instruments and techniques: modeling
SC: Hydrology [H]
MN: 2007 Fall Meeting