Using electronic monitoring imagery to characterise protected species interactions with commercial fisheries: A primer and review
Monitoring commercial fisheries provides essential information to enable effective fisheries management. Traditionally, human observers have provided the majority of fisheries monitoring services, with other methods used including position monitoring (e.g. using Vessel Monitoring Systems), at-sea boarding and aerial surveillance. While monitoring using human observers can work well in some cases, challenges such as occupational safety, space constraints on smaller vessels, representativeness of data collected, and cost, have catalysed the exploration of other methods. In this context, electronic monitoring (EM) using on-vessel cameras has developed through extensive trial, pilot and operational programmes in the last 15 years. Amongst other objectives, EM has been used to monitor interactions between threatened, endangered and protected species (TEPS) and commercial fisheries.
This report presents the findings of an extensive review investigating the types of TEPS interactions that EM has been used to explore, and training given to analysts to detect and describe those interactions. The review encompassed published and unpublished reports, social media posts, and the websites of practitioners, companies, agencies, and multilateral bodies known to use or promote EM. Experts were also consulted directly to collect information on work that is underway but not yet publicly available.
The majority of EM programmes to date that have focused on TEPS interactions were trials or pilots, with a smaller number of operational programmes underway. Information reviewed showed that EM has been widely tested and proven effective in monitoring captures of a range of TEPS in fishing gears. When EM imagery captures these interactions, species identification is possible in most cases. Life status can also be determined when animals are vigorous, especially when brought on deck prior to release. Detection of unusual or unexplained behaviour, that may result from crews wishing to avoid a TEPS capture being recorded by EM, is also possible. EM has been explored (but found less effective) for monitoring seabird interactions with trawl warps and third wires.
Other effective applications of EM that are relevant to the impacts of fishing on TEPS include monitoring handling of these species after capture, deployment of mitigation devices (e.g. tori lines, pingers, turtle exclusion devices), and detecting the presence of fish waste discharge within camera views. Collecting robust quantitative information on the abundance of TEPS present in the air or in the water around vessels and fishing gear is difficult using EM.
Species identification using EM imagery has been approached by practitioners using a number of methods, e.g. employing analysts who are trained and work as at-sea observers or who have received observer training, using field guides, species lists, and images of species of interest. Characteristics such as body size, morphology, colour, and distinctive markings are all important to facilitate identifications. Based on the findings of work reviewed, it is recommended that analyst training materials for species identification include actual EM imagery (or sources as close to this as possible, e.g. observer photos) when available. This is because animals seen in imagery may be wet, incomplete, covered in fish slime or scales, or not visible from an angle that optimises identification. Documenting how identifications were made is also important (i.e. using which of the animal's visible characteristics), to add rigour to EM datasets.
In many reports and published papers, the training provided to EM analysts is not described. However, because EM analysts are the source of data, the training process has a strong bearing on data quality and therefore end-user confidence in datasets produced. In studies where training is described, it routinely incorporates elements such as core instruction, self-tests and practice runs after which feedback is provided, and a formal assessment that documents analyst competence. When a particular level of competence is reached in the formal assessment, this provides an assurance of a commensurate level of data quality.
The development of automated review methods for EM imagery is accelerating. However, currently, there are no routine deployments of automated review in place for any species. The majority of work on automated EM imagery review has focused on fish to date. Four projects dedicated to automated review of imagery recording TEPS interactions were identified during this review. Deployment of automated review will change, rather than eliminate, the role of humans in EM review. For example, automated review algorithms must be written, trained, and tested, and the need for processes that provide assurance of data quality remain.
The detection of interactions between TEPS and fishing gear occurs after a number of other steps in the EM process chain. For example, monitoring objectives and business requirements must be clearly defined, EM cameras must be deployed in appropriate positions on-vessel, system specifications (e.g. frame rate) must be optimised to record interactions, and crew activities onboard must occur such that these objectives and requirements can be addressed. Once EM imagery is captured, the review process provides for the extraction of data on TEPS. EM imagery review may be undertaken using a census or a sample approach, depending on monitoring objectives and time and resourcing available. Business requirements must identify the data elements that analysts are instructed to extract from imagery during review. As part of a rigorous monitoring process, data extraction must be repeatable and auditable. The need for standards for data collection from EM, review processes, and quality assurance of review is well recognised. However, the development of standards to underpin EM is in its early days.
To support the continued exploration and adoption of EM to monitor TEPS interactions with New Zealand commercial fisheries, it is recommended that:
• Data standards are developed and documented to specify the information that EM analysts are tasked with extracting from imagery,
• Quality assurance standards are developed for EM review,
• Training materials and programmes are prepared to enable EM analysts to populate data fields and to document their findings,
• The development of training materials is initiated where requirements are already understood,
• Photos and videos taken by fisheries observers are catalogued and stored for use as part of EM training materials, and potentially to contribute to the development of machine learning approaches over time,
• New Zealand remains abreast of the regional development of EM process and data standards, and,
• Practitioners in New Zealand and internationally are encouraged to make available EM process and data standards, review protocols and training materials.