Tracking Marine Mammals Around Marine Renewable Energy Devices Using Active Sonar

Hastie, G. (2013). Tracking Marine Mammals Around Marine Renewable Energy Devices Using Active Sonar (Report No. SMRUL-DEC-2012-002.v2). Report by SMRU Consulting. Report for UK Department of Energy and Climate Change (DECC).

Download Citation File

Abstract

Currently, there is a high level of uncertainty surrounding the environmental impacts of marine renewable energy devices on marine wildlife (particularly seals, whales, and dolphins). The principal concerns derive from the potential of physical injury to marine mammals through direct contact with moving parts of marine energy devices.
In recent years, there has been accelerated development of active sonar systems for the defence sector for sub-sea monitoring of potential security threats, and for fisheries research and management. This may provide a basis for monitoring close range interactions between marine mammals and energy devices.
In this report, we describe a program to develop a sonar system to provide a behavioural monitoring tool for marine mammals around marine energy devices that is accurate, user-friendly, and data efficient.
A 4-phase program involved collaborations between marine mammal specialists, marine renewable energy developers, and sonar engineers to develop a sonar system for the marine renewable industry. The overall approach was one of caution with key tests at each phase which had to be satisfied in order to progress to the subsequent phases.
As part of Phase 1 of the project, a Request for Proposals (RFP) document was drafted and distributed to sonar manufacturers who were invited to provide a formal proposal for the development of a marine mammal sonar. A total of 5 proposals were received from sonar manufacturers and although all manufacturers offered solutions based on off-the-shelf systems, there appeared limited interest in investing internal R&D support to system development for this application; however, two manufacturers (BioSonics and Tritech) did provide a commitment to internal R&D to the project and these were included in subsequent testing.
It was important when developing the behavioural monitoring sonar that any observed behavioural responses could be attributed to the tidal turbine rather than to the sonar being used to measure it; there is the potential that low frequency components of the signals from sonar systems could be audible to animals and elicit behavioural reactions. Phase 2 of the project was therefore designed to measure potential behavioural responses by marine mammals to each sonar and interpret the significance of these.
The results of the behavioural response trials suggest that grey seals and harbour porpoises exhibit differing behavioural responses to the signals of each sonar system. Porpoises exhibited relatively subtle responses to the Tritech Gemini; in contrast, seals exhibited overt responses to the BioSonics DT-X by leaving the pool when the sonar was active.
In addition to the behavioural response trials, the range of audibility of the sonar signals was modelled to predict the ranges that different species would be likely to hear the signals in a tidal environment. Seals and harbour porpoises were predicted to be able to hear the signals of the Tritech Gemini and BioSonics DT-X at ranges of approximately 60 and 4,000 metres respectively.
Given the results of the marine mammal behavioural response trials and the predicted ranges of audibility, there was a clear need to modify the acoustic properties of the BioSonics DT-X signal prior to any further development. Although signal modification was carried out, analysis of the modified signal suggested these were unsuccessful in sufficiently reducing the lower frequency components likely to be responsible for eliciting the observed responses; the decision was therefore made not to continue with development of this system.
Improvements to the Gemini system focused on the development of efficient classification algorithms to reduce data volumes and provide a probabilistic indication of the identity of individual targets; these included variables such as size, shape and swimming characteristics to determine valid marine mammal targets.
In order to develop classification algorithms sonar image data for some of the more abundant marine mammal species around the UK (grey and harbour seals, harbour porpoises, and bottlenose dolphins) were collected. Using this, a series of detection and classification developments were implemented in the Gemini software.
Results of the analysis of the software ‘detection efficiency’ suggest that there is a significant negative relationship between range and probability of detection; the probability of the software automatically detecting a seal was greater than 0.9 for ranges up to around 37 metres and dropped to below 0.1 at ranges greater than 56 metres. In the context of using this sonar as a behavioural monitoring tool, this appears to limit analysis of small marine mammal behaviour to ranges of approximately 40-50m.
To provide an assessment of the software ‘classification capabilities’, the classification probabilities for each of the confirmed targets and the unidentified targets were analysed. The majority of classification probabilities for confirmed seals were ‘Probable’ and ‘Potential’. However, both the unidentified targets and confirmed debris also had a relatively high proportion of ‘Potential’ and ‘Probable’ classifications assigned to them. These results highlight the scope for behavioural monitoring using active sonar but also highlight the current limitations in terms of species ID.
To evaluate the long term reliability and detection capabilities of the sonar on an operational tidal turbine, a Gemini system was deployed on the SeaGen tidal turbine in Strangford Lough. A single Tritech Gemini sonar transducer was attached to a mounting plate and secured to the centre of the crossbeam of the turbine, facing south towards the seaward end the Narrows. Sonar images were collected on a total of 42 days between the 20th May and 29th July 2011. Only data collected during the flood tide (i.e. the sonar was facing the incoming tide) were used in analysis.
The data were then analysed post-hoc using the developed software to determine what proportion of targets were classified as marine mammals with a high probability and their proximity to the turbine. In addition, the temporal variation in the number of ‘marine mammal’ targets were analysed in a General Additive Modelling framework to assess how ‘time of day’, ‘tidal speed’, and turbine operation (ON/OFF) influenced the number of marine mammals around the turbine.
The results of the deployment suggested that there were 109 ‘high probability marine mammals’ in the data. A manual review of the data associated with these ‘marine mammal’ detections appeared to confirm that they were marine mammals with only 3 of the targets being obviously non-marine mammals. The detection rate of ‘high probability marine mammals’ was approximately 5.9 per day.
The ranges that ‘marine mammals’ were detected at Strangford Lough suggest that marine mammals do move in close proximity to the tidal turbine both when it was operational (minimum range=9.9m) and non-operational (minimum=8.4m).
The results of the modelling of ‘marine mammal’ detections with the temporal covariates suggested that the occurrence of ‘marine mammals’ changed with time of day. Detections generally decreased during early morning with a minimum at approximately 0500. In contrast, there was no significant variation in ‘marine mammal’ detections in relation to tidal speed and turbine operation (ON/OFF).
In terms of future sonar development work, the automated classification algorithms are currently highly conservative and the reduction of these to ‘high probability marine mammals’ requires post hoc analysis; this feature could either be incorporated into the existing software or an additional classification module could be developed to analyse the detection and track data. Furthermore, validation of marine mammal detections around a tidal turbine [through visual observations or by tagging seals with high resolution movement tags (e.g. Wilson et al., 2007b)] would be of clear benefit. In terms of measuring fine scale behaviour of animals around tidal turbines, the system developed here does not currently provide data on the depth of the targets and although there are deployment configurations (e.g. using more than one transducer in different orientations) that could address this to a certain extent, the development of a true 3D sonar system would be highly beneficial for measuring tracks of marine mammals around turbines.