The MASTS conference took place on 18th and 19th November at Strathclyde University, followed by a full day of workshops on the 20th. While I was primarily interested in the Technological innovation in environmental monitoring session, I also caught sessions on marine mammals and bioacoustics. There were many interesting talks, but if you are looking for a quick summary, scroll to the bottom of the page.
Although this was a two-day conference, I attended on the 19th only. The first session of the day for me was the Marine mammal special session. Comprised of nine talks focused on a variety of topics such as detecting collisions with underwater structures, noise reduction, and even teeth analysis.
The session opened with a presentation on how stable isotopes in cetacean teeth can be used to estimate age. For example, radiocarbon extracted from the teeth can be linked to nuclear weapon testing in the 60s and 70s, as well as nuclear plant discharge. The next talk evaluated the efficiency of acoustic deterrent devices and found that they were effective in reducing porpoise bycatch but did not work well with seals.
The following two talks discussed how we can better estimate the range of species. In the ‘Mobile species move’ presentation, we learned that adding a temporal component to distribution modelling helps to capture mobile species better. This research was focused on minke whales, meaning that with a better distribution model, we can determine where the area overlaps with creel fishing spots and can make modifications to reduce the bycatch. The second talk made use of citizen science to discover that bottlenose dolphins from Scotland increasingly visit the north east of England. People could submit their photos of dorsal fins through a program called Citizen fins, and these photos helped to establish that there were multiple sightings of the same dolphins in the Tay area and in the north east England.
The next set of talks explored the effects of noise and vessel traffic on marine mammals. There was an interesting discussion on whether noise reduction equipment used for building offshore wind farms is truly useful. The cost and length of building increase when the devices are in use, as an extra vessel needs to be used to deploy a bubble curtain. And while the talk agreed that the noise reduction is beneficial, the prolonged noise, even at lower strength, might be more harmful than a short and loud period.
The final two talks evaluated the risk of collisions between seals and tidal turbines. The data came from a movement tracker installed on the tidal turbine. 347 seal tracks were detected during the turbine operation, and 77% of these were classified as low risk, and the number where the collision might have occurred was very small. The research shows that there is a clear avoidance response, but highlights that it is important to bring developers, regulators and advisers together to keep the marine environments as risk-free as possible.
The second session I attended was on marine bioacoustics. Bioacoustics were dominant throughout the day, with elements present in other sessions as well. The majority of the talks in this session were on dolphins. The first two presentations showed how a full repertoire of dolphin whistles can be captured and analysed. Machine learning was used to divide 1271 whistle contours of Scottish west coast common dolphins into different categories, resulting in 79 different groups. This is a larger set compared to other species, and it also doesn’t change much over time. Bottlenose dolphins show geographic variation in whistles, and they use them to broadcast their identity. Strong inter-site variation of whistles was detected, and this effect was even stronger in adjacent populations.
More machine learning applications from this session include using CNNs to classify calls from seven different species to help us see which days and times they occur. There was also a research project on how military sonars can be detected using gradient boosted random forests.
This session revisited noise effects evaluation. First, for detecting if porpoises are disturbed by it and how we can reduce these disruptions. Followed by examining the effects of brief startle sounds on dolphins. The study showed that a startle sound resulted in heading changes and swim speed increase, suggesting that this might be an effective way to manage marine mammal movement around underwater structures.
The technological innovation session was fully focused on the use of AI. There was a big range of talks, opening with the use of UAVs to help monitor hand gathering. The goal was to detect human tracks on the beach using drone images. Good results were achieved using pixel-wise classification. The next talk was on using multiple data sources (image, bathymetry and side scan) to classify seafloor types. This method presented a robust way for classification, and having multiple data sources provided much-needed information to help with the task.
Moving on to what was probably my favourite talk of the day: a combination of AI and laser scanning for crustacean stock assessment. This research featured a special device that could scan lobsters and crabs very quickly. The output of this scanner is a point cloud for each animal that can then be used to segment different body parts and get measurements automatically. The new method was as good as human measurements. The data could potentially be used further to predict if the animal is male or female.
The variety of machine learning applications doesn’t stop there. Vision transformers were used to monitor plankton using images from the flow cytobot, easily producing 15k-30k images from each sample. This can help to detect harmful blooms early and allow for a quick response. Another research used AI to detect seals on the coast from drone images and classify their age and gender. The use of AI also appeared in a seal deterrent operation to protect salmon. Here, movement was detected with high-frequency imaging sonar that allows for the detection of non-vocalizing animals as well.
The session featured eDNA analysis for assessing fish and invertebrate communities around offshore wind farms. Finally, seabirds made an appearance in an aerial surveying project that aimed to determine if kittiwakes change their flight height around wind farms. However, the height of the flight seems to be linked more to the sea depth rather than the farm itself.
I chose to attend the full-day QGIS workshop from the workshop selection. The workshop was led by Prof. Rob Briers from Edinburgh Napier University. We received a quick overview of the software, and then it was time to tackle the five exercises. Each exercise introduced a new functionality and a set of steps to complete. Everyone completed the tasks at their own pace, which worked great for this type of workshop. There were extra challenges to work through to further test the newly learned skills. Both of the instructors were very attentive and helped with any software issues and explained any extra questions really well. QGIS is an incredibly large software, and it is impossible to cover all of it in one day, but these tasks were a great mix to cover the basics and understand some of the most commonly used functions.
The second day of the MASTS conference was dominated by bioacoustics and underwater noise monitoring. There was a lot of emphasis on studying how underwater noise influences mammals in the area. Machine learning is becoming increasingly more common as nearly every project is facing some sort of data bottleneck. There were many applications for both images and audio. It also shows that ML has become more accessible to non-programmers as it is being increasingly packaged as a tool inside various software. However, some of the talks also highlighted the generalisation issue that comes with deep learning. Achieving great results on the test dataset doesn’t guarantee the models will work well on any new data. The day featured a great range of talks and plenty of interesting posters in between. There was a lot of new information to take in.
The QGIS workshop was a fantastic introduction to the software and helped me build some foundations on how to use it and plan future projects. There were 5 different exercises covering the most commonly used functionalities.
All of the speakers did a great job making the talks very accessible to people from various backgrounds. Overall, this was a great learning experience, and I would recommend it to anyone looking to learn about the latest research in marine science.
Stock images taken from https://www.pexels.com