Evaluating Approaches for Apportioning Acoustic Density to Species Using Simulated Great Lakes Fish Communities

Standardization of acoustic (AC) data processing has advanced the field, but standard methods for incorporating mid-water trawl (MT) data when estimating species density are lacking. We developed a pelagic fish simulator that populated an artificial lake to mimic communities present in 3 Great Lakes. Virtual surveys mimicked data a real vessel could gather in 1, 3 and 5 nights. We investigated 5 apportionment methods based on (1) NEAR, the nearest trawl in 2-dimensions; (2) NEARD, having depth stratification; (3) HAM, decreasing levels of agreement in 3-dimensional location by transect, water layer, and bathymetric depth; (4) TREE, regression tree predicted composition from 3-dimensional location information; and (5) KISS, lake-wide averages uninformed by location. Performance was evaluated by comparing species density estimates to known values. The NEARD, HAM and TREE methods that combined MT results to AC information in layers performed similarly, and outperformed NEAR and KISS. The performance of the best methods improved with increasing effort. With 1 night of effort, there were instances when MT effort poorly characterized the community resulting in poor performance of all methods. Results suggest that decisions on the number of MT samples to collect and depths fished can profoundly affect accuracy of species density estimates.