REPURPOSING NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION (NOAA) HYDROGRAPHIC DATA FOR SEA FLOOR SEDIMENT MAPPING ALONG THE CONTINENTAL SHELF OF THE DELMARVA PENINSULA

The U.S. Geological Survey (USGS) conducted a feasibility investigation to determine if automated image classification techniques performed on a set of National Oceanic and Atmospheric Administration (NOAA) hydrographic survey data could accurately differentiate seafloor bottom types for an 800 km² area of the inner continental shelf of the Delmarva Peninsula. Between 2006 and 2013, NOAA conducted 30 hydrographic surveys along the Delmarva Peninsula, in water depths generally less than 40 meters, using multibeam echosounder systems (MBES) and sidescan sonars for the mission of updating nautical chart. The acoustic backscatter data from the MBES data were obtained and reprocessed by the U.S. Geological Survey to create backscatter mosaics to aid the Delmarva regional geologic framework investigation. The hydrographic survey data were not collected with the intent to produce or publish any multibeam backscatter products and as such these data contain numerous acquisition-related artifacts nonetheless, data from four adjacent surveys were repurposed to test the performance of auto classification for seabed characterization. The four test hydrographic surveys were processed individually in QPS Fledermaus FM Geocoder toolbox (FMGT) to extract beam-averaged backscatter values and create backscatter mosaics at a 2 meter-per-pixel resolution for each survey. Both non-supervised (Isodata method) and supervised (Maximum Likelihood method) digital image classification algorithms were applied using the Image Classification Toolbox in ArcGIS 10.2 on two-banded acoustic images that were created by combining bathymetric gradient and backscatter intensity. Fifty seafloor sediment grab samples collected by the USGS in 2014, from a proximal area were used to ground-truth and evaluate the initial performance of the supervised classification technique. Results for the overall accuracy and Kappa statistic show that the supervised classification technique has greater accuracy than the unsupervised classification, 94.9% and 65.5% respectively, as well as more validity, 0.92 and 0.47, respectively. This investigation demonstrates the efficacy of using existing hydrographic survey data to produce a reliable automated sea floor sediment type map and greatly increase our knowledge of sea floor geology.