US Geological Survey

 

ORCID IDs

http://orcid.org/0000-0001-7360-4057

http://orcid.org/0000-0003-0940-8013

http://orcid.org/0000-0002-1602-482X

http://orcid.org/0000-0002-9683-4282

Date of this Version

2-21-2018

Citation

2018 John Wiley & Sons, Ltd.

Comments

DOI: 10.1002/rra.3270 wileyonlinelibrary.com/journal/rra River Res Applic. 2018;34:430–441.

Abstract

Methods for spectrally based mapping of river bathymetry have been developed and tested in clear‐flowing, gravel‐bed channels, with limited application to turbid, sandbed rivers. This study used hyperspectral images and field surveys from the dynamic, sandy Niobrara River to evaluate three depth retrieval methods. The first regressionbased approach, optimal band ratio analysis (OBRA), paired in situ depth measurements with image pixel values to estimate depth. The second approach used ground‐based field spectra to calibrate an OBRA relationship. The third technique, image‐to‐depth quantile transformation (IDQT), estimated depth by linking the cumulative distribution function (CDF) of depth to the CDF of an image‐derived variable. OBRA yielded the lowest depth retrieval mean error (0.005 m) and highest observed versus predicted R2 (0.817). Although misalignment between field and image data did not compromise the performance of OBRA in this study, poor georeferencing could limit regression‐based approaches such as OBRA in dynamic, sand‐bedded rivers. Field spectroscopy‐based depth maps exhibited a mean error with a slight shallow bias (0.068 m) but provided reliable estimates for most of the study reach. IDQT had a strong deep bias but provided informative relative depth maps. Overprediction of depth by IDQT highlights the need for an unbiased sampling strategy to define the depth CDF. Although each of the techniques we tested demonstrated potential to provide accurate depth estimates in sandbed rivers, each method also was subject to certain constraints and limitations.

Share

COinS