Species Apportionment of Lake Superior Hydroacoustic Targets Using Midwater Trawl Catch Proportions and Classification and Regression Trees

The identification of methods for estimating species-specific fish densities and distributions is necessary to fully exploit information collected during hydroacoustic fisheries surveys.  Simultaneously collected midwater trawl samples are a commonly used approximation of species proportions in hydroacoustic survey areas, although the trawled water volume may not be fully representative of the acoustic volume sampled.  Classification and regression trees (CART) can make use of midwater trawl and ancillary information to determine a hierarchical model for estimating species proportions in volumes of water which are acoustically sampled, but not directly sampled using midwater trawls.  Hydroacoustic data were collected at night using Biosonics 120- and 70 kHz split-beam transducers along approximately 1500 km of transects in the open waters of Lake Superior during summer months from 2003-2006.  Constant-depth midwater trawl samples (n= 47) were collected during portions of acoustic data collection for the determination of pelagic fish species attributes including fish size, depth within the water column, and bottom depth at species occurrence.  A CART model constructed using midwater trawl information was applied to all single echo detections (i.e. individual fish) within the upper 100 m of the water column.  Species apportionment of hydroacoustic data using only species proportions observed in trawl would numerically be 45% Kiyi (Coregonus kiyi), 37% Cisco (Coregonus artedi) and 7% Bloater (Coregonus hoyi), while CART modeling resulted in overall numerical proportions of 42% Kiyi, 21% Cisco and 18% Bloater.  The CART method of species apportionment of hydroacoustic data improves on the trawl-only approach because it does not rely on the assumption that the species presence in trawl samples is directly proportional to species presence in all areas of the hydroacoustic survey. Because of this, the CART model can be used to help determine species proportions in volumes of water not directly sampled by midwater trawls.