Sturgeons are amongst the most threatened fish species. To make matters worse, bottom trawl fisheries are known for the high amount of bycatch they generate. Unfortunately, the trawl fishery that targets California halibut happens to catch quite a few green sturgeons. Of course, fishermen release them back into the water when this happens, but we need to know if it is safe for these fish to be caught and released. In this study, Dokakis et al. present a method for monitoring the survival of green sturgeons using PSATs by using machine learning to determine if the released sturgeons survived or not during a specific post-release time frame.
How can you reduce the costs of PSAT studies?
What Dokakis et al. demonstrate in this article is a simple yet very neat trick. To get the most out of your PSAT studies, reach for help! First of all, the public can play a big role in recovering tags. In this particular study, almost half of the pop-up tags (37 out of 76) were retrieved. This way, you can swap out the plug-in battery/release section and use the tag again! Remember that recovered tags let you access their whole dataset, too. Doukakis et al. were able to efficiently determine mortality at a 90% rate with recovered tags, compared to 70% with broadcast tags. A reward in cash is a good investment for securing additional data from a recovered tag. The Argos satellite constellation is very slow, and in areas like San Francisco where the weather is often cloudy, transmissions can be blocked, which leads to incomplete datasets that sometimes even need to be discarded. More complete data means fewer tags are released in the first place!
Another way to involve people in your studies is to collaborate with fishermen. Scientific vessels and crews can cost an incredible amount of money, and fishermen are already out there... with a vessel and a crew! Dokakis et al. boast the advantages of opting for this strategy. Cooperation can generate trust and helps fishermen learn about environmental issues they are facing in their industry. This in turn strengthens the impacts of your research.
How do you apply machine learning to PSAT studies?
Towards Data Science has a nice explanation of what AI is: AI lets computers “make decisions based on patterns hidden in the data”. As you know, satellite tags are great for building large datasets. This is why using them alongside machine learning can create a powerful tool. Doukakis et al. used a type of machine learning model called the support vector machine (SVM). Here is a short video on SVMs in case you wanted to learn more about them. These are made for classifying data into two groups, so an SVM is perfectly suited for identifying living and dead fish in this study.
The first thing Doukakis et al. did was to determine how data recovered from living sturgeons differs from dead fish data. For starters, they gathered some preliminary data to train the algorithm. As you can probably tell, Doukakis et al. have found that a dead fish generates more predictive data. Accelerometers would always indicate that the SeaTag was upright, meaning of course that the fish was not swimming or making any sharp moves. As a comparison, live fish would record higher temperatures with varying depths and accelerometer data. Once the algorithm has been fed this information, it can start classifying data from unknown fish, and tell if the individual ended up alive or dead.
Another important thing to know is how much data is enough? In this particular case, this means how many fish should be tagged? And how long after the release a sturgeon can die while it is still considered post-release mortality? The authors have determined it would be 21 days, based on literature. 76 fish were tagged during this study, and 51 of the datasets were usable, which is a good ratio given losses from poor data transmission in this area. It was enough to let the researchers use their algorithm to determine quite well if a fish had died following its release.
Does the release of bycatch have an effect on fish mortality?
The authors mention that this study by itself is not sufficient to draw conclusions on the exact effects of bycatch encounters on green sturgeons, but here are some interesting takeaways. Following their work, Doukakis et al. have concluded that there exists a post release mortality of green sturgeons captured as bycatch… mainly influenced by the amount of time elapsed from the end of the tow until the fish is finally released. It is therefore important for fishermen to spot any bycatch ASAP and limit their time spent out of the water.
They estimated that 82% of the fish have survived, which translates into 11 losses. This is sad news, as green sturgeons have a life expectancy of 70 years. According to NOAA, they can mate several times during their lifetime and only attain maturity at 15 years old. This is why every individual counts. Remember that post release mortality isn’t the whole story. The recovery of sturgeon populations will require that fish have better access to spawning areas, and that they stay alive long enough to spawn, too.
We really enjoyed reading this paper (don’t hesitate to visit NOAA’s page to consult the original article and learn more about the researcher’s work). We think Doukakis et al. really demonstrated the power of PSATs in this study, and have shown that it is feasible to gather enough data for machine learning, which is known to require lots of data. To conclude, we would like to notify you of our new PSAT newsletter page. We improved the looks and plan on releasing more reviews soon. If you enjoyed reading this article, let us know! And of course, we wish you a better second half of 2020, and we hope you can get back to what really matters to you… saving the ocean, one tag at a time!
