Observer bias and subsampling efficiencies for estimating the number of migrating fish in rivers using Dual-frequency IDentification SONar (DIDSON)

摘要

Fixed-location, side-looking, multibeam, sonar techniques offer a practical approach to estimate the numbers of migrating fish in rivers that are too large or occluded for traditional sampling methods, such as weir trapping, visual observation techniques, and netting. While this technology has been used to enumerate salmonid escapement in coastal river systems of western North America, little use and evaluation has occurred in inland waters such as the Great Lakes, where rivers and runs of fish are considerably smaller than those along the Pacific coast. We use a "Dual-frequency IDentification SONar" ("DIDSON") imaging sonar system to investigate the error and variability among nine people performing fish counts. There was no significant difference found among observers' estimates of fish abundance per DIDSON file; however, the total count of all fish differed from the benchmark value by as much as 26%. Post-processing simple fish counts from DIDSON raw data is labour-intensive and costly. Three subsampling methods of fish passage estimations were developed and evaluated for their accuracy and precision for daily and seasonal time frames. The random and systematic subsampling methods had similar seasonal and daily accuracy and precision with few exceptions. Automation-assisted counting was much more accurate and efficient for seasonal estimates. A ratio of approximately 2:1 was found for the automated to manual fish counts and this varied little among years. The DIDSON multibeam sonar unit is useful in estimating potamodromous fish migrations for large tributaries of the Great Lakes. DIDSON image processing costs can be minimized through suitable subsampling approaches. The automation-assisted method is the most cost-effective means of estimating moderate levels of fish passage over longer study periods. Multiple individuals can be used interchangeably for the manual post-processing of DIDSON data