Tuesday, January 8, 2013

The Ups and Downs (or Side to Sides) of Shark Research

If you’re reading this blog, chances are you’re well versed in the fate of the modern shark. Shark fin mafia, extinction, 38 million, ecosystem collapse, blah, blah, blah; there are a lot of buzzwords being thrown around by active and armchair conservationists alike. I don’t want to start any arguments so we’ll take a middle-of-the-road approach and just agree that these species are important, threatened, in need of much further research.

Historically we begin this research with direct (we say in situ) observations of the study species in their natural environment. Think Jane Goodall and her Gombe River Chimpanzees. From here we progress to higher order analyses with multidisciplinary syntheses and hypothesis testing (test tubes and bubbly liquids). The marine environment however, presents several unique challenges to this fundamental task. It’s bloody huge, often inaccessible, and the best visibility of 30 metres is but a fraction of some shark home ranges than can span thousands of kilometres. This is perhaps why we still know very little about many of the oceans inhabitants.

Fortunately though you are reading this from your shiny new iPad and we have made several advances in technology since Jane Goodall was out in the field. We can place tags on sharks that transmit their location to your very screen and find out just how they like to spend their weekends. As scientists, this allows us to shift our perspective from that of an outside observer and onto the animal itself. We remove the need to actually see the animal, and in turn remove many of those limitations to observation that I listed before.

This answers the ‘When’ and the ‘Where,’ but increasingly, scientists want to know the ‘Why,’ the ‘What,’ and the ‘How.’ So we put sensors on those tags. A current estimation identifies about 24 different types of available sensors measuring anything from temperature to heartbeat.

One such sensor, the accelerometer, allows for the detection of very fine scale movement. You’ve got one in your iPad and it is what detects the tilting and shaking of the screen when you play Temple Run. In the very same way, the accelerometers we place on sharks measure their tail movements and posture.

My work here at Oceans Research has been a ground truthing exercise whereby we test the ability of these accelerometers to measure shark tail beats, and from there extract specific swimming behaviours. Now that I’m into the final analyses before writing up the dissertation, I can show you how this is done.

After some manipulation, this is what I get from the sway axis of the accelerometer (the side to side movements). Essentially each peak and trough pair corresponds to one complete tail stroke cycle. So in the beginning we see some steady swimming, followed by a sweeping turn, some more steady swimming, and so on until the very end where we the shark stops moving al together before quickly resuming movement in  a quick burst of speed. And that’s pretty much it! By taking a little bit of physics into account, we can remodel the movement of the shark, even making some behavioural inferences, without ever having witnessed the animal.

Unfortunately science isn’t quite that simple and we have to ‘test’ hypotheses with ‘statistics,’ but that’s my problem, not yours. What’s that you say? My humour and brilliance has got you curious? Oh go on you! Fine, here’s a bit more…

Here is a comparison of several different methods for extracting those behaviours. At the bottom, marked ‘video’ is our direct observation, an ethogram, of the animal that will serve as the truth, against which we will compare all others. Each coloured bar indicates when a specific behaviour was observed. Marked ‘accel’ are the behaviours that I have manually observed from the accelerometer time series. The next one, marked ‘Eth’ are the behaviours extracted by an autonomous k-means clustering algorithm of the accelerometer record. Special software has analysed the data and grouped them together. Lastly, is the same analysis performed with some training. So in this case I have given the computer a set of criteria to use for each clustering before it automatically groups them together.

You can see clear differences in the efficacy of each method. Not immediately apparent however is the work involved in each method. Manually annotating the accelerometer record by hand is very time consuming compared with the split-second analysis performed by modern computers. So which method would you choose? Each has its own advantages and disadvantages, ups and down (side to sides). Leave a comment and tell me what you think! Here’s one thing to consider; we are comparing everything to a visual record, but who’s to say we can see everything? Perhaps the accelerometer can pick our subtleties not visible to the human observer. Food for thought!

For your viewing pleasure, here’s a short video demonstrating the ability of the accelerometer to identify specific tail beats. https://www.youtube.com/watch?v=zUfk3arpleU

No comments: