Length measurement accuracy of adaptive resolution imaging sonar and a predictive model to assess adult Atlantic salmon (<scp> Salmo salar </scp>) into two size categories with long‐range data in a river
Abstract Imaging sonars are used around the world for fish population monitoring. The accuracy of the length measurements has been reported in multiple studies for relatively short (<15 m) ranges and high image resolution. However, imaging sonars are often used at longer ranges ( i.e. , &...
Published in: | Journal of Fish Biology |
---|---|
Main Authors: | , , |
Other Authors: | , , , |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
Wiley
2020
|
Subjects: | |
Online Access: | http://dx.doi.org/10.1111/jfb.14456 https://onlinelibrary.wiley.com/doi/pdf/10.1111/jfb.14456 https://onlinelibrary.wiley.com/doi/full-xml/10.1111/jfb.14456 |
Summary: | Abstract Imaging sonars are used around the world for fish population monitoring. The accuracy of the length measurements has been reported in multiple studies for relatively short (<15 m) ranges and high image resolution. However, imaging sonars are often used at longer ranges ( i.e. , >15 m) where the images produced from sonar returns become less detailed. The accuracy of the length measurements from the Adaptive Resolution Imaging Sonar (ARIS) was tested by releasing n = 69 known‐sized adult Atlantic salmon ( Salmo salar ) directly into the sonar field at ranges between 15 and 29 m, and measuring their echoes manually by four users and semi‐automatically using a computer workflow in Echoview software. Overall, the length measurements were very variable: compared to true (fork) lengths, the mean of differences varied between −9.9 cm and 7.8 cm in the human‐generated datasets, and between −42.8 cm and −20 cm in the computer‐generated dataset. In addition, the length measurements in different datasets were only in poor or moderate agreement with each other (intraclass correlation <0.61). Contrary to our expectations, the distance from the transducer or the subjectively assessed echo quality did not have an effect on the measurement accuracy in most of the datasets and when it did, the effect was not systematic between the datasets. Therefore, a size class and length prediction model was implemented in a Bayesian framework to group salmon into two size categories: One‐Sea‐Winter (<63 cm) and Multi‐Sea‐Winter (≥63 cm) groups. The model correctly predicted the size category in 83% of the fish in the computer‐generated dataset and ranged from 68% to 74% in the human‐generated datasets. We conclude that fish length measurements derived from long‐range imaging sonar data should be used with caution, but post‐processing can improve the usefulness of the data for specific purposes, such as adult Atlantic salmon population monitoring. |
---|