| dc.description |
Multibeam sonar data was collected on Nehalem Bank, off of the Oregon Coast with the purpose of increasing the spatial resolution of our knowledge of the area and delineating seafloor habitats. The system used was the Kongsberg Simrad EM-300 operating at 30 kHz. The data set collected includes both topographic and backscatter information, both of which were processed and gridded at a 5m cell-size resolution. This thesis summarizes the methods used in the characterization of seafloor habitat using the Nehalem Bank data and the analysis tools incorporated within a geographic information system (GIS).
This work focuses on investigations into the analysis of benthic habitats through the use of algorithmic calculations at multiple scales to quantitatively delineate distinct
seafloor regions. A recently developed spatial analysis technique, the bathymetric
position index (BPI), was tested and used as part of the Benthic Terrain Modeler Tool to
classify the Nehalem Bank data. Those results were compared both qualitatively and
quantitatively to a newly developed method (developed for this project) entitled the
surface interpretation method (SIM). Both methods of classification use multiple scales
of BPI in conjunction with the topographic digital elevation model (DEM) to quantitatively delineate seafloor habitats. While accuracy assessments of remotely
sensed imagery on land are usually carried out using groundtruthing information
collected specifically for that purpose, sonar imagery is often assessed using any
information available. The groundtruth data available for Nehalem Bank is sparsely
located throughout the region and includes: submersible dives, sample data, and seismic
and sidescan imagery. In order to evaluate the fit of the classification attempts to the
DEM quantitatively, a set of reference points was created and interpreted using the three dimensional (3D) visualization capabilities of a GIS. The results were analyzed using error matrix statistical measures that provided a quantitative assessment of the classification results, the feedback from which guided a refinement of the classification methods. This was done in attempt to optimize “real-world” evaluation of a classification methods results, without an idealized level of seafloor groundtruth data.
The backscatter component of the data was processed to reduce the influence of
grazing angle on the intensity of surface returns, a common source of noise that
dominates many datasets. The processed image shows a reduced variability of the grey
level of the imagery in both along and across track directions. Submersible dive
groundtruth data were then co-located with the backscatter imagery using habitat
interpretations of the dives displayed upon the DEM. The backscatter data was then
classified in the regions of the dives using textural analysis with grey level co-occurrence matrix derived indices of homogeneity and entropy. The resulting substrate class divisions satisfactorily parsed the landscape into habitat regions, with the exceptions of the nadir and shadow regions. However, higher frequency deep-towed sidescan imagery in the same region shows that the delineation of substrate regions appears to be incorrect with the EM300 imagery likely imaging subsurface seafloor characteristics. |
|