Measuring fish length and assessing behaviour in a high-biodiversity reach of the Upper Yangtze River using an acoustic camera and echo sounder

Abstract

Summary In the past decade improved acoustic hard‐ and software have enabled estimations of abundance and distribution patterns of aquatic organism, including non‐intrusive monitoring of fish migrations and behaviour. In this study, a high frequency acoustic camera (DIDSON‐LR, 1.2 MHz, 0.7 MHz) and a portable split‐beam scientific echo sounder (Simrad EY60, 200 kHz) collected acoustic data on 192 and 157 individuals within 24 hr (19–20 April 2011) in the Mituo reach of the Yangtze River, China. Mean fish length estimated from the acoustic camera data was 18.7 ± 5.6 cm, with an average swimming speed of 0.19 ± 0.13 m s−1. The mean fish target strength (TS) produced by the echo sounder was −43.8 ± 4.4 dB, which corresponded to 5.7–119.9 cm fish length when converted by three different TS‐length equations. Average swimming speed was 0.11 ± 0.06 m s−1 from the echo sounder. Compared with the actual fish catch by the three layers of drift gill net in the survey area, the target length indicated by DIDSON was more accurate than the EY60 results, which were highly affected by the choice of TS length equations. It was determined that the two devices used synchronously could estimate fish length effectively to investigate their behaviour and distribution.

Length measurement accuracy of adaptive resolution imaging sonar and a predictive model to assess adult Atlantic salmon (Salmo salar) into two size categories with long‐range data in a river

“…The measurement accuracy of the imaging sonars in these setups is not as well tested, but the measurements have been compared between low-frequency DIDSON and other methods. Lin et al (2016) used ranges up to 30 m and found that the average DIDSON-measured fish size was similar to the average length of fish in a drift gill net survey. Lilja, Romakkaniemi, et al (2010) compared the size distributions produced from long-range (up to 80 m) DIDSON and angling catch size distribution, and concluded that the DIDSON lengths were approximately 5-10 cm shorter than the true lengths of the fish.…”

“…Lin et al . (2016) used ranges up to 30 m and found that the average DIDSON‐measured fish size was similar to the average length of fish in a drift gill net survey. Lilja, Romakkaniemi, et al .…”

“…Although many studies have been previously conducted testing the measurement accuracy of imaging sonar using short ranges (Burwen et al ., 2007; Cook et al ., 2019; Daroux et al ., 2019; Grote et al ., 2014; Gurney et al ., 2014; Hightower et al ., 2013; Tušer et al ., 2014), there is often a need to use ARIS in longer (>15 m) ranges in fisheries management (Hakkola, 2011; Lilja, Orell, et al ., 2010; Lin et al ., 2016; Valjus et al ., 2017). To our knowledge, this study is the first to investigate the length measurement accuracy of an ARIS imaging sonar using 1.1 MHz frequency at ranges up to 29 m with free‐swimming fish.…”

“…As the frequency decreases, the length measurements are less accurate; using a 1.2 MHz frequency in a short-range setup (<13 m), Zhang et al (2014) found that the measurements from DIDSON were 35.6% shorter than in-hand measurements of fish lengths. In spite of the limitations in resolution, longer range setups are needed, and used, when the whole river width needs to be covered (due to the prohibitive costs of the sonar units, often exceeding $US100,000) or when the site location is best accessible from farther away (e.g., dams and fishways; Hakkola, 2011;Lilja, Orell, et al, 2010;Lin et al, 2016;Valjus et al, 2017). The measurement accuracy of the imaging sonars in these setups is not as well tested, but the measurements have been compared between low-frequency DIDSON and other methods.…”

FishSense: Underwater RGBD Imaging for Fish Measurement

“…Effective methods for underwater imaging are absolutely vital in retrieving valuable data. Typical methods include stereo vision [13], laser-based systems [15] [3], or acoustic-based systems [9]. These methods can be computationally complex and require intricate computer vision algorithms.…”