Post provided by Kim Whoriskey

Early Career Researcher Kim Whoriskey takes us behind the Methods paper ‘Current and emerging statistical techniques for aquatic telemetry data: A guide to analysing spatially discrete animal detections’ which led to her being shortlisted for our Robert May Prize in 2019.

Understanding how aquatic animals move is becoming increasingly important for protecting them. Knowing where they migrate, how long they stay, and what they do when they travel through changing marine environments provides us with key information on movement corridors, habitat hotspots, and changing population distributions. This information can then be used to help manage and conserve many different aquatic species, from developing guidelines for recreational fishing practices to defining marine spatial planning measures.


Networks have emerged as a critical tool for conducting aquatic telemetry research. By sharing resources across project leaders, networks can enable researchers to develop a more holistic view of their study, for example by discovering long-distance migrations crossing geopolitical boundaries. The Ocean Tracking Network (OTN) consists of a global group of researchers that share data, equipment, and ideas to track more than 230 species worldwide. I work with them, and a sub-committee formed by early career researchers called ideasOTN, to create new ways of visualizing and analyzing the different kinds of data that we collect.

Types of telemetry for monitoring animal movement

I work with movement data collected from aquatic animals using two different kinds of electronic tags. Satellite telemetry tags sample an animal’s location wherever it is (conditional on tag programming), by communicating location to satellites in near-real time and producing data that look like an animal’s path. On the other hand, acoustic, radio, and PIT tags are based on a two-part system; uniquely coded tags emit ultrasonic pings, which are picked up by receivers placed in strategic locations across the globe, resulting in detections of animal presence. Because their structures are so different, the statistical techniques that you can apply to these two data types can be quite varied.

Satellite-tagged grey seal pup. Sable Island, NS. Photo credit: Shelley Lang

Advancements in acoustic telemetry

Acoustic telemetry tracking has become extremely popular over the last several decades, and through networks like OTN, researchers have been able to collect massive multi-species datasets. The questions that we’re asking with detection data are also getting more complicated. We’re no longer just describing where an animal went and how long it stayed there – we’re also asking, “Are animals more likely to survive in particular areas/habitats?”, “What factors can affect events like migrations?”, and “Are certain areas/habitats more important because they serve as connection hubs to lots of other areas?” Furthermore, although detection data themselves consist of timestamps associated with locations, they can be summarized into movements, residency, presence/absence, and the list goes on.

Acoustic receivers and moorings in Norway. Photo credit: Rob Lennox

All of these factors have sparked a rapid evolution of statistical methods for analyzing the resulting data, to the point where it can be pretty daunting for new or even seasoned telemetry users to figure out which method might be the best for them. Although it was evident that there were lots of powerful statistical tools available, it was clear that there was no general guide available to help users choose from among them.

Researchers in Norway acoustically tag and release fish. Screen shots from video footage by Rob Lennox

Choosing the best statistical method for your telemetry research

We created a decision tree that researchers can use to select the best statistical method for their study from among the most ubiquitous methods currently available. The four main methods that we identified were: generalized linear/additive modelling techniques, mark-recapture methods, survival (or time-to-event) modelling, and network analysis. Using this tree, you can start to eliminate some options based on a few criteria:

Your habitat: Whether you’re in a lake, river, estuary, or the open ocean.

Your study design: How many receivers and tags do you have? Are your receivers arranged in a grid that might give you movements on a finer scale, are they placed strategically along chokepoints down a river to monitor migration, or have they been deployed as gates in the open ocean to track large-scale movements?

Your question of interest: Do you want to model demographics like population size and survival of individuals? Perhaps you want to understand when and why a particular event occurs, like the causes of being accidentally swept through a dam? Or are you more interested in looking at the effects of habitat removal on the movement of the animal?

Decision tree for picking among statistical methods. Original tree and article can be found here.

I have often found that different branches of statistics are interrelated. This posed a challenge when creating our decision tree because under certain conditions, some of the different methods can be viewed as one in the same. Another challenge was the realization that multiple statistical methods can be used to ask the same question – I could probably use all three of mark-recapture, generalized linear mixed models, and survival analysis to estimate the probability of animal survival. So, when designing the tree, we separated each of the methods by highlighting the scenarios under which we thought their usage would be the most direct and appropriate.

New frontiers in aquatic telemetry

There is an incredible amount of research to look forward to in the aquatic telemetry world. New technologies are constantly being developed to broaden the types of data that ecologists can collect on their study animals – one group has even developed a method for remotely measuring the volume of blood and level of oxygen in the brains of diving seals! Plus, the enormous datasets that are being collated by networks are being synthesized to show broader movement patterns. This paper analyzed data from 92 species and discovered that they share enough characteristics to be grouped into four unique movement classes.

Determining how to analyze large amounts of data from increasingly varied sources with hugely ranging temporal and spatial scales is a really exciting frontier in our field. And we’re not going to be able to do it alone – it’s going to be critical to look outside our field for solutions. Some related areas we can look at are the terrestrial telemetry world and the study of human mobility. Other areas to look into include sociology and psychology (for network analysis), mining, epidemiology, and biostatistics (for spatial and longitudinal analyses), and business applications like recommendation algorithms or computational photography (for machine learning). With almost 8 billion people in the world now, it’s likely that somebody has already solved at least part of our problem. Solving the rest of it by tweaking and figuring out how to apply these new and exciting methods to our field is going to be the fun part!

Grey seals, including one tagged with a satellite tag, on Sable Island, NS. Photo credit: Damian Lidgard

To read the full paper: Current and emerging statistical techniques for aquatic telemetry data: A guide to analysing spatially discrete animal detections

Read more papers from our Robert May Prize 2019 here.