Mixture Model Analysis for Species Apportionment of Dual-Frequency Identification Sonar Data

Fixed hydroacoustic technology can provide accurate estimates of upstream fish passage in streams, yet estimating abundance by species is often far more difficult.  When species differ in size, and a hydroacoustic correlate of fish size is available, it may be possible to discern information about species composition from the frequency distribution of the hydroacoustic variate.  Under some configurations, dual-frequency identification sonar (DIDSON) can provide reasonably precise estimates of fish length.  We show how a species / age mixture model can be fit to observed DIDSON length measurements to provide useful estimates of species composition, even when species partially overlap in size and length measurements are subject to error.  Auxiliary information about the length distribution of each species is required.  Bayesian statistical methods are employed, implemented with Markov-Chain Monte Carlo.  The approach is demonstrated with data from a fixed DIDSON monitoring migrating Chinook (Oncorhynchus tshawytscha) and sockeye salmon (O. nerka) in the Kenai River, Alaska.