Automated detection and tracking of marine mammals: A novel sonar tool for monitoring effects of marine industry. (6th September 2019)

• Record Type: Journal Article
• Authors: Hastie, Gordon D.; Wu, Gi‐Mick; Moss, Simon; Jepp, Pauline; MacAulay, Jamie; Lee, Arthur; Sparling, Carol E.; Evers, Clair; Gillespie, Douglas
• Is Part Of: Aquatic conservation. Volume 29(2019)Supplement 1
• Page Start: 119
• Page End: 130
• Publication Date: 2019-09-06
• Subjects: behaviour -- mammals -- monitoring -- new techniques -- ocean -- renewable energy; Aquatic ecology -- Periodicals; Conservation of natural resources -- Periodicals; Aquatic resources -- Periodicals; 333.95216
• DOI: 10.1002/aqc.3103
• Languages: English
• ISSNs: 1052-7613

Abstract

Abstract: Many marine industries may pose acute risks to marine wildlife. For example, tidal turbines have the potential to injure or kill marine mammals through collisions with turbine blades. However, the quantification of collision risk is currently limited by a lack of suitable technologies to collect long‐term data on marine mammal behaviour around tidal turbines. Sonar provides a potential means of tracking marine mammals around tidal turbines. However, its effectiveness for long‐term data collection is hindered by the large data volumes and the need for manual validation of detections. Therefore, the aim here was to develop and test automated classification algorithms for marine mammals in sonar data. Data on the movements of harbour seals were collected in a tidally energetic environment using a high‐frequency multibeam sonar on a custom designed seabed‐mounted platform. The study area was monitored by observers to provide visual validation of seals and other targets detected by the sonar. Sixty‐five confirmed seals and 96 other targets were detected by the sonar. Movement and shape parameters associated with each target were extracted and used to develop a series of classification algorithms. Kernel support vector machines were used to classify targets (seal vs. nonseal) and cross‐validation analyses were carried out to quantify classifier efficiency. The best‐fit kernel support vector machine correctly classified all the confirmed seals but misclassified a small percentage of non‐seal targets (~8%) as seals. Shape and non‐spectral movement parameters were considered to be the most important in achieving successful classification. Results indicate that sonar is an effective method for detecting and tracking seals in tidal environments, and the automated classification approach developed here provides a key tool that could be applied to collecting long‐term behavioural data around anthropogenic activities such as tidal turbines.