Background
In 2017, the Mystic River Watershed Association (MyRWA) launched our on-line herring counting platform. Funded by an Environmental Education Grant from US EPA, the herring on-line portal has been a great educational success story. The platform has allowed thousands of people — including hundreds of local students in classrooms — to witness a remarkable wildlife migration, and to learn about the ecology and importance of river herring and about the conservation challenges the species face.
In 2019, we commissioned a new feature to the video system: an automated estimate of the total number of fish in the Mystic herring population, based on counts submitted by users. Our brilliant collaborator Jeff Walker developed a statistical model that estimates the total number of fish to date, each time a new count is submitted. The idea was to record videos whenever a motion detector detected fish passing (which ends up being tens of thousands of videos a year), but not have to count every video. Building on existing statistical models, Jeff developed a system whereby we can estimate on the fly the number of fish you would see if you watched all the videos, even though only a subset of randomly selected videos had been watched. Each video count adds to a set of random samples, and from relatively few samples, one can estimate the total.
Two counting methods
For the past few years we have had two side-by-side methods for estimating the size of the population of river herring returning to the Mystic River: this video system, and the in-person counting method used to generate the official count by the Massachusetts Division of Marine Fisheries (DMF). (The in-person estimate is also based on samples: but the samples in question are 10-minute counts every hour by volunteers, who stand at the dam and note the number of fish entering Upper Mystic Lake.)
It turns out the two methods have generated total numbers that have differed by 20-35%. In some years, the video estimate has been higher, in some years lower. Each year has to be diagnosed separately, but we think we have a pretty good handle on the shortcomings and strengths of the two methods and why these numbers have diverged in each year.
What have we learned?
Differences in visibility: Some years, cloudy water from relatively high algae populations prevents the camera system from detecting fish. Either the passing fish don’t reliably trigger the motion detector, or the animals can’t be made out confidently in the murky images by observers. Water that is murky to the underwater camera can look clear from above, so in-person counts are not affected in these years. In years of poor water clarity, the video system will therefore tend to underestimate the total, no matter how many videos get watched. (We did an intensive study one year, and found consistent relationships between lower video and higher in-person counts in murky conditions.)
On the other hand, there have been years (like 2018) when the water has been clear, and we have (through happenstance) not had staffing for in-person counts on what turned out to be days of high rates of fish passage. In those years, we think we have strong evidence that the in-person count probably underestimated the total, and the video counts may more accurately point to the true total.
Here you can see the difference in visibility through the video camera on a clear-water day vs. a murky-water day:
Equipment failure: Another factor that can lead to the video system leading to lower estimates is equipment failure. The whole system of camera and computer and cloud storage of videos is quite robust and reliable, but not infallible. We have missed some busy days of fish passage because the camera was down for technical reasons. Again, no matter how many videos get counted, when this happens, the video system will tend to underestimate the total.
This was the case in 2021, when the camera was out of commission for 3 days during the height of the migration. This equipment failure combined with high turbidity (murky conditions) were likely the reasons why 2021 saw a higher than average difference: 282,889 vs 550,000 in video vs. in-person counts. (Read more about 2021 in-person counting results here)
Day & night monitoring: The video system has some super powers that have led to new insights into the behavior of river herring in the Mystic system. For example, We can’t see the fish at night, but our infrared camera can. In 2018, we used the video system to investigate what percentage of fish were estimated to pass through the Mystic Lakes dam outside our normal monitoring hours of 7AM-7PM. The answer turned out to be about 20%: 10% at dusk, and 10% around dawn (more info here). This is genuinely new knowledge about the behavior of herring in our system that couldn’t have been generated without this video setup.
Sharing innovation: We plan to continue both methods, and to continue to refine our methods and understanding. For the foreseeable future, the in-person count will generate the official estimate at the state. But the Division of Marine Fisheries follows our innovative video work with interest, and our platform is the brains behind video fish counts now on Town Brook in Plymouth and on the North and South Rivers.
Conclusion: Both counting methods yield important information
Both our counting programs received high praise from the state. We continue to have one of the single most robust and well-supported programs in the state for fish counting. Because of this, the confidence in detecting trends over time (a slightly different statistical question) is extremely strong for the Mystic. This all is knowledge that only exists because of in-person and online volunteer effort. That’s a remarkable story in itself.
Even more importantly, our community scientists have helped MyRWA learn more about the behaviors of herring—information that is being shared with the Division of Marine Fisheries (DMF). During the lifetime of the grant, MyRWA’s Watershed Scientist has studied:
Behavior of herring at night; yes, they do still migrate during the night–but they really like to migrate between 7 am and 7 pm!
Modeling — this helped test the statistical model of both in-person and on-line monitoring so we understand how many herring are migrating.
You can help contribute to this story and our understanding of the herring migration in the Mystic! Click the link below to count 2022 herring online:
Have more questions about our online counting system? Check out the FAQ here.
Special shoutout to our 2021 Online Herring Counting Leaderboard volunteers! Thank you for volunteering your time to help us document the herring migration online. User kaagamc was our 2021 leaderboard winner with OVER 10,000 VIDEOS counted! Sign up for an account to be a part of 2022’s leaderboard competition.
