Underwater Challenge: Battelle Maps the Nearshore Environment

The data from the side-scan sonar image above are transformed into the 3-D colorized luminescence diagram below through the use of neural networks. The neural network is trained to recognize a wide range of spatial and acoustic features associated with specific targets, in this case, the strong acoustic returns from the edge of the eelgrass bed (represented by the yellow peaks).

Identifying areas that may be sensitive to coastal development, determining future restoration scenarios, and creating baseline maps are among the important applications of side-scan sonar and underwater videography. These tools acquire imagery and data for habitat characterization, and are particularly useful when combined to assess the environmental quality of aquatic resources.

Battelle recently used these methods to create high-resolution maps of approximately 22 km of the nearshore environment in Puget Sound, an area critical to salmonids and other aquatic species in the Pacific Northwest. The project was sponsored by King County, Washington. The maps delineated benthic surface sediment types, submerged aquatic vegetation, estimates of area covered by habitat and sediment types, and locations of fish and sessile (attached by the base) benthic invertebrates. Video surveys documented species population and distribution, and were also used to verify side-scan sonar data. Data were used to create geographical information system (GIS) polygons (Figures) of features that were overlaid on digital maps.

The original process is currently being improved. The U.S. Department of Energy and the National Imagery and Mapping Agency have recently sponsored Battelle—through the Pacific Northwest National Laboratory and Muse Applied Sciences, a Canadian firm—to develop neural-network-based pattern recognition capabilities to automate the process of mapping and characterizing structures, topology, and distribution of vegetation. Through an iterative training process, the neural network “learns” to differentiate targets in images based on acoustic and spatial information. The use of neural networks is a powerful tool for correctly identifying patterns in noisy data sets, such as those found in side-scan sonar data. Automated extraction techniques promise to reduce time and costs and improve accuracy associated with interpretation of side-scan sonar data.

For further information on side-scan sonar, contact Dana Woodruff at (360) 681-3608, Dana.Woodruff@pnl.gov, or for information on neural networks, contact Karen Steinmaus at (360) 681-3646, Karen.Steinmaus@pnl.gov.