ABSTRACT: The more frequent use of two-dimensional hydrodynamic river models also requires more detailed bathymetry surveys. For smooth bathymetries, there is little difficulty in developing accurate translations from survey data to model; however, in rivers with significant bottom structure (e.g., large woody debris), simple data averaging and interpolation methods may lead to misrepresentation of the bottom bathymetry. It is necessary to identify in the data set what is true bathymetry from what is caused by large woody debris. Two groups of methods are investigated to serve our objective: statistical techniques and filtering techniques. While the former are appealing for their simplicity and direct applicability in modeling, they fail at consistently treating spikes (hypothesized to be large woody debris signature) in the data set. Among filtering techniques, linear filters are turned down due to their inherent trade-off between edge retention and spike rejection. Two nonlinear filters are examined. Median and erosion filters are specifically designed to preserve sharp edges while eradicate spikes. Finally, median filtering is preferred to erosion filtering, for the former leaves large-scale bathymetric features virtually undisturbed.