Why do I get videos with zero fish? I thought the recording was initiated by the camera detecting a fish.
The camera is “motion activated,” but multiple things can cause the motion detection software to trigger. Floating debris, light reflections (which can change during the day), turbulent water or a dislodged camera box can all create moving images that the software can’t distinguish from a fish passing by. If we adjust the sensitivity of the system to screen out these cases, we risk also filtering out actual river herring passing by. So we err on the side of including more zero counts, so we don’t miss fish.
Sometimes what triggers the video is a river herring going the wrong way, or a fish that isn’t a river herring. Here’s an especially fun example.
Videos appeared today from a previous day, even though they were not in the system yesterday. Can that be right?
Yes! In 2021, the internet connection we had at the fish ladder site has been down. So we have had to manually retrieve and then upload thousands of videos every couple of days. This process takes time, and sometimes happens out of order. So it can seem as if the number of videos from an earlier day has increased. This is not a mistake.
I put in a positive number of fish but the estimated total went down. Can this be right?
Yes! The way the model that estimates the total run is by partitioning each day into different time segments and then estimating the rate of fish passage in those periods, based on the counts that viewers make. But viewers are served up videos at random. So if someone counted many more fish representing a minute immediately following the video you counted, for instance, the system will revise its estimate for that overall period of time downward, based on the additional information you provided. And thus the total estimate can also go down.
This morning the estimated total number of fish was higher than it is now. Can this be right?
Yes! The working estimate of the total is based on the data that happens to have been input up to that moment. So especially early in a day’s count, the estimate of the total number of fish can bounce around, high and then lower and then back again. This variation tends to settle down as more data comes in. You will also notice that the error bars on each day’s estimate become smaller as more videos are counted. With only a few data points in, the uncertainty is high. As more counts come in, the estimates become more confident.
I put in a large number of fish, but the total didn’t change at all. Can this be right? Yes! If the number you input “confirms” the working estimate of the rate of fish passage that the model has for the period of time your video occurred in, the total estimate might be unaffected.
The number of fish counted on a given exceeds the estimated total number of fish in the fish run estimate. Can this be right?
Yes! Some videos are counted more than once, in order to acquire data on how consistent counts are across different users. So two people may count the same 100 fish in a given minute on the same day. The “fish counted” metric will be 200. But the system understands that only 100 fish could have passed in that minute.