There’s no “Google Maps” for finding fish.  The radio signals that are the backbone of traditional GPS cannot pass through water.  But sound travels remarkably well, so scientists often use acoustic telemetry to estimate an individual fish’s location. That means attaching an acoustic transmitter to the fish and then using a network of stationary underwater listening stations to monitor for the short clicking sounds that these tags emit. When a fish swims near to a receiver, its click is heard, and its individual code number is recorded.

To date, most researchers have used ad hoc methods to analyze their data, and typically have not quantified uncertainty. A fish is generally assigned the position of the receiver that detected it, even though it might be anywhere in the detection range. And if it is not heard from for a while, no positions are assigned, even though we know where it isn't, and can estimate how far it could travel.

We developed a state-space model for estimating individual fish movement – one part that models the fish behavior, and one that models the observation of that behavior.  It can model a species’ preference for returning to a specific home base—like that protective coral cave—as well as a more nomadic movement style that does not return to a specific ‘home’.

The observation model uses a detection function, the probability of accurately logging animal presences. The detection function varies with environmental noise, and in the case of nearby receivers with overlapping detection funtions, can be used for triangulation. You know how the cell tower network allows your mobile phone's position to be triangulated? So that Big Brother knows where you are at every moment? This is similar to how acoustic telemetry works. The distance from your phone (or the fish tag) to several cell towers (or acoustic receivers) is measured, and circles of that radius are drawn around each tower (receiver). Where the circles intersect -- that's you (or the fish).

The observation model also uses negative data, or the lack of detections, in combination with the behavior model to estimate how far the fish may have traveled while undetected – knowing where the fish was not located can  tell you a lot about where it was located.

Paper at MEE     PDF      Supplement

R and ADMB code (right click and save link as)

Martin’s web page

We field tested the model at Palmyra Atoll in the central Pacific Ocean – home to myriad fish, sharks, mantas, whales and turtles. 

With monitoring data collected for coral reef fish from 51 underwater observer stations at Palmyra Atoll, Pedersen and Weng used their state-space model to develop a contour map that provided a visual representation of the confidence regions for the locations of the fish over time, along with a home range estimate. 

During daylight hours, fish locations were estimated with a 95% confidence region radius of 50 meters, at their most accurate. 

By reducing the uncertainties associated with underwater location tracking, Pedersen and Weng hope to provide researchers and marine managers with better information to help support marine conservation activities for reef fish and other threatened species