Abstract
The tidal energy industry requires effective marine life monitoring systems for characterising pre-deployment conditions and evaluating the environmental response to deployment, operation, and recovery of turbines. There is concensus that the continued development of in-stream tidal energy installations must go hand-in-hand with improved knowledge of the avoidance and evasion behaviours, strike detection, and species distribution of marine animals. Significant work focused on imaging sonars (active acoustics) and hydrophones (passive acoustics) is underway. However, active acoustic data that is regularly collected for measuring tidal flows is not currently utilised for marine life monitoring.
Acoustic Doppler current profilers (ADCPs) are the standard tool for monitoring ocean currents, and they are widely used by the tidal energy industry for physical site characterisations and monitoring. ADCPs also detect signals scattered by fish, marine mammals, and other discrete targets in the water column, however typical approaches to ADCP data processing require that such signals are rejected as noise. The rejected signals contain valuable information on marine life movement and hold the potential for a valuable new approach to fisheries acoustics. An alternative ADCP data processing method presented by Zedel and Cyr-Racine (2009) uses a least-squares based algorithm to compute both current velocities and velocities of fish and other discrete water column targets, even when fish are intermittently present. Even with a suitable processing algorithm, the challenge remains to verify the accuracy of fish detections in ADCP data.
A standard tool for detecting fish is the split-beam echosounder, which under ideal conditions has the ability to provide information on the speed and direction of discrete targets in the water column as well as backscattered signal amplitude. The ability to estimate both numbers and velocities of discrete targets in the water column using split-beam echosounders, combined with their established utility for biological surveying, makes them a sound choice for validating ADCP fish detection and velocity data.
This project advances research conducted by Dr. Len Zedel (Memorial University Newfoundland), with a focus on testing and validating the use of ADCPs for marine life detection. In this study, we make use of data collected using a co-located BioSonics DTX split-beam echosounder, ADCPs, and optical imaging systems to validate fish counts and velocities derived from ADCP data. The study area consists of Grand and Petit Passages in Digby County, Nova Scotia, as well as the surrounding areas of St. Mary’s Bay and the outer Bay of Fundy. Grand and Petit Passages are characterised by strong tidal currents, and are sites of interest for the development of in-stream tidal energy technology. The project objectives include: (1) validate the application of ADCPs to marine life monitoring; (2) gain information on marine life (swim velocity, counts, and identification) utilizing the split-beam sonar, optical images, and ADCP data; (3) advance methods for ocean bottom-mount and vessel-based data collection; and (4) develop algorithms for marine life detection from ADCP data, for academic use and commercial application.
The study consisted of multiple instrument deployments spanning from 2018 to 2020, using both fixed (sebead-mounted, with upward-oriented instruments) and mobile platforms (vesselmounted, downward-oriented instruments). The mobile platforms included the ‘Jetyak’, an inboard jet-powered autonomous surface vessel developed at Woods Hole Oceanographic Institution, and two research vessels operated by Luna Sea Solutions Inc. (Luna): the ‘Puffin’, an outboard-powered Rosborough RF-18; and the ‘Grand Adventure’, a Rosborough 28 with inboard diesel. The vesselmount infrastructure developed for use aboard the RV Puffin and the RV Grand Adventure has been demonstrated to be highly functional and robust for data collection in the high tidal flow environments of Grand and Petit Passages. The pole-mount systems employed on both vessels allowed us to rapidly deploy the co-located instruments. The vessel-based deployment platforms benefit from lower operational costs and complexity than autonomous bottom deployments. However, the shorter-term vessel deployments came with associated challenges of collecting sufficient and suitable validation data.
The Jetyak has been shown to be a uniquely capable deployment platform. Given its utility in shallow coastal waters, along with the decreased risk to survey operators, the Jetyak provides potential to address needs both for long term monitoring and for increased capacity for responsive (reactive) monitoring capability at the local level. Though data collected using the Jetyak was not suitable for use in the ADCP validation analyses, we were able to use data from a Jetyak-mounted sidescan sonar to visualise fish schools in some cases.
Fish detections in ADCP data were highly correlated to those from the BioSonics split-beam echosounder. The strong linear relationship is very encouraging and certainly met the expectations of the research team. Use of the ADCP for discrete target detection also provides an alternative processing approach for dealing with frequently-observed near surface bubble-plumes, eliminating the need for the more operator-intensive process of determining an exclusion line – required when using the split-beam echosounder in an upward-orientation in high energy tidal environments. The validated fish detection method gives us a basis to reanalyse earlier data sets to explore any biological signals of significance.
Validating ADCP-based fish swimming velocities remains a challenging task that is sensitive to properties of fish behaviour; namely, fish must be present, and in schools with sufficiently low density to (a) allow the computation of meaningful echosounder track data and (b) avoid oversaturation of the ADCP data by discrete targets. Data passing our selection criteria were not sufficient for a robust statistical validation of the ADCP-derived fish swimming velocities. However, the comparisons associated with the schools that met our selection criteria are favourable. We consider the sensitivity of the analysis to the fish swimming and schooling behaviour we observed an important result, and recommend that it be considered carefully in the event of similar future studies.
On the choice of instrumentation and processing software: The use of Sonar5-Pro versus Visual Aquatic for the echosounder data analysis represents a notable difference between the bottom-mount and vessel-mount experiments. The extraction of tracks from the echosounder data contains steps that are inherently subjective, though the quality of the fish track data were likely much more dependent on the properties of the fish schools. The fish tracking software packages known to us (i.e., Sonar5-Pro, Visual Aquatic, or Echoview) appear flexible enough to achieve comparable results that are sufficient for the analyses described in this report. Though direct comparisons to the echosounder data were only made with RDI ADCP data, we would expect comparable performance using a different ADCP (i.e., Nortek). Differences in sampling rates between instruments may have implications for the number of fish identified by the algorithm, due to differences in repeat counts of individual fish.
The data processing pipeline used to analyse the ADCP data presented in this document has been organized as a toolbox currently known as ADCPFish, which processes raw ADCP data into depth and time-averaged fish and water velocities. The toolbox is based on code written by Dr. Len Zedel, and was adapted for use as a toolbox by Muriel Dunn. The application has since been refactored and extended by Luna to encompass a broader range of instrument types, orientations, and functionality.