Preliminary hard and soft bottom seafloor substrate map derived from an unsupervised classification of gridded backscatter and bathymetry derivatives at French Frigate Shoals, Northwest Hawaiian Islands, USA

Processing steps used to generate the backscatter, bathymetry, and bathymetric rugosity grids used in the unsupervised classification process are described in their own metadata that can be found at www.soest.hawaii.edu/pibhmc. The most map contains three separated datasets including Simrad em3002d and Reson 8101 data collected on cruises in 2005 and Reson 8101 data collected in 2008. Backscatter from the three datasets were processed separately and resampled to 5 m grid cell size using the Raster Resample Tool in the ArcMap Version 9.3 Toolbox. Resampling of the backscatter was performed after charted data in the GeoTIFF format for the individual sonar systems were mosaicked using the ArcMap Version 9.3 Toolbox Mosaic to New Raster function. Mosaics were created in the ESRI GRID format. Bathymetric variance was calculated from the 5 m bathymetry in a two-step process using ArcMap Version 9.3. The first step involved calculating the standard deviation of the bathymetry layer using the Neighborhood Statistic tool from the Spatial Analyst pull-down menu in ArcMap Version 9.3. Rectangular neighborhood settings were set to a height and width of 2 grid cells to create a small-scale variance layer. The standard deviation output was then squared using the ArcMap Raster Calculater tool. A large-scale variance layer was also generated using a rectangular neighborhood height and width of 5 grid cells. The fine-scale variance is useful for defining small features that may indicate the presence of sediment ripples or possible coral-rich areas whereas the large-scale variance is useful for defining seafloor ridges, pinnacles, and significant changes in slope. Bathymetric variance proved to be a useful tool in previous efforts to map seafloor habitats by Dartnell and Gardner (2004). Their publication can be downloaded from www.soest.hawaii.edu/pibhmc/pibhmc_documentation.htm. Two separate sets of variance and rugosity layers were then created with the same pixel dimensions as the the Reson 8101 and Simrad em3002d backscatter datasets. The datasets were handled separately because the difference in sonar frequencies between the two backscatter prevents them from being directly comparable. They should not be merged into one backscatter layer prior to classification. The layers were then opened as ESRI GRIDS in the image processing software package ENVI (Environment for Visualizing Images) Version 4.3. Two multi-band images were then created with the backscatter, small- and large-scale variance, and rugosity. A mask was also created in ENVI for each backscatter dataset. Image classification was performed separately on the two multi-band images using the K-Means unsupervised classification method, which calculates initial class means evenly distributed in the data space and then iteratively clusters the pixels into the nearest class using a minimum distance technique. More information about K-Means algorithm can be found in the ENVI documentation available at www.ittvis.com/envi/. The number of classes was set to two and all other options were set to default. Class color mapping was performed on the resulting classified images. The substrate types of hard and soft were assigned based on results from supervised classification of the backscatter data (also performed in ENVI) using hard and soft bottom classes from classified seafloor photographs. The photographs, classifications, and descriptions of the classification process are available for download at www.soest.hawaii.edu/pibhmc. Three resulting classified images, two for the Reson 8101 datasets and one for the Simrad em3002d dataset, were mosaicked in ArcMap using the Mosaic to New Raster in the ArcMap Version 9.3 Toolbox and exported from ArcGIS as a GeoTIFF file to create one image containing three classes; sand, hard bottom, and unclassified (background).