Improved fish counting method accurately quantifies high‐density fish movement in dual‐frequency identification sonar data files from a coastal wetland environment

There are many ways to quantify fish movement through shallow‐water habitats, but most noninvasive methods (e.g., visual counts) are not effective in turbid coastal wetland waters of the Great Lakes. Dual‐frequency identification sonar (DIDSON) technology (Sound Metrics) offers a noninvasive, hydroacoustic‐based approach to characterize fish movement in wetlands and other habitats by collecting highly detailed fish movement data regardless of light and water quality conditions. High‐resolution data can be analyzed to estimate fish movement in areas where visual observations are difficult. However, enumerating a complex mix of fish sizes by manually counting fish visible in echogram files requires training and is very time consuming. Therefore, four counting techniques were tested to estimate fish abundance from DIDSON echograms that were collected at a hydrologically reconnected coastal wetland in the Great Lakes. Briefly, the four counting methods were (1) manually viewing the entire length of the echogram (full‐hour manual count), (2) manually viewing subsections of the echogram before generating fish estimates by per‐minute average (subsample manual count), (3) using Echoview automated software to generate automated estimates, and (4) using DIDSON viewer software to generate automated estimates. Over 800 echogram‐hours were recorded over a 9‐month period at an open‐flow water control structure connecting a coastal wetland to a tributary to Lake Erie. Commercial fish tracking software (Echoview) and custom software scripts from Milne Technologies were used to semi‐automate fish count estimates for a small subset of data. Semi‐automated software counts were compared to manual counts of identical data files to assess differences in accuracy, cost, processing time, and counter effort. Semi‐automated fish count estimates using Echoview and custom pre‐ and postprocessing software scripts did not differ from baseline manual counts, suggesting that the semi‐automated count process could be a reliable tool to increase efficiency when processing large DIDSON data sets.