A “How-to” Guide for Interpreting Parameters in Habitat-Selection Analyses

This blog post is provided by John Fieberg, Johannes Signer, Brian Smith and Tal Avgar and tells the “StoryBehindthePaper for the article “A ‘How-to’ Guide for Interpreting Parameters in Habitat-Selection Analyses”, which was recently published in Journal of Animal Ecology.
John Fieberg (University of Minnesota), Johannes Signer (University of Goettingen), and Tal Avgar (Utah State University) teach quantitative courses at their institutions and have also offered a series of workshops related to habitat-selection and animal-movement modeling.  They have also collaborated on the development of the amt (animal movement tools) package in R for handling and analyzing animal telemetry data. Brian Smith, a graduate student in Tal Avgar’s lab, joined the team last fall to help with a workshop at the annual meeting of The Wildlife Society and is now a regular contributor to the amt package.  

Although habitat-selection functions (HSFs, also known as resource-selection functions or RSFs) were developed decades ago and are still widely used today to study environmental drivers of animal space-use patterns, they are often misunderstood. Recent papers that highlight connections between HSFs and other popular modeling frameworks offer new ways to interpret HSF parameters and clarify many previous misconceptions regarding appropriate use of these methods. Concurrently, advances in animal tracking technologies have led to an enormous increase in positional data for a wide range of taxa and have fueled the development of new quantitative methods for studying the behavior and ecology of wild animals. Step-selection functions (SSFs) and integrated step-selection functions (iSSFs) were developed to take advantage of increased sampling frequencies associated with newer Global Positioning System Collars and to address inherent statistical challenges resulting from autocorrelation in positional data. Analyses using HSFs/RSFs and SSFs/iSSFs often have similar aims, and therefore, we will use the term habitat-selection analyses to refer to analyses using this suite of approaches.

Flock of trumpeter swans in Minnesota, USA. The large female spreading its wings is equipped with a GPS/GSM collar and is providing tracking data for a movement ecology study focused on the Interior Population of Trumpeter Swans in North America.

So, what are the differences between habitat-selection and step-selection functions? Most importantly, they differ in how they quantify habitat that is available or accessible to the animal (i.e., “habitat availability”). Habitat-selection functions assume the distribution of available habitat is constant through time, and thus, the animal has equal access to all areas within a user-defined availability domain (e.g., a study area or an estimated home range) at all points in time. This assumption may hold when animal positions are observed infrequently enough that the animal could have traveled anywhere in-between observations, but it is clearly violated with many modern telemetry data sets. Step-selection functions relax this assumption and allow for time-specific habitat-availability distributions. These availability distributions are formulated by considering the animal’s previous location and its general movement tendencies. Thus, SSFs and iSSFs more explicitly consider the role animal movement plays in determining habitat-selection patterns.

A collared cow elk with a bull elk and several cow elk and calf elk in the background. Rich County, Utah, March 2020.

Both habitat-selection and step-selection functions contrast locations where an animal was observed with locations that could be used by the animal (i.e., they are parameterized using ‘use-availability’ data). Although use-availability data are often analyzed similar to how one would analyze ‘used-unused’ or ‘presence-absence’ data, there are important differences related to data preparation, model fitting, and parameter interpretation that need to be considered when conducting habitat-selection analyses. In teaching several graduate-level courses and workshops on these topics, we observed that many participants struggled with preparing data for habitat-selection analyses and interpreting analytical results.

While we have discussed workflows for habitat-selection analyses elsewhere, we focus in this paper on how to correctly interpret parameters in habitat-selection and step-selection functions. We start by explaining how using a logistic regression in a habitat-selection analysis (the common practice) is equivalent to estimating the parameters of an inhomogeneous Poisson point process or a weighted distribution, and then we explain how these equivalences facilitate interpretation of model results. We then walk the reader through a series of examples where we show how to interpret model coefficients for categorical and continuous covariates, as well as for interactions between covariates (a particularly confusing topic). Furthermore, we demonstrate the steps required to implement an integrated step-selection analysis, and how this approach can be used to simultaneously model habitat selection and animal movement. In addition to the main text, we provide extensive appendices that give detailed step-by-step explanations for how to use the amt package in R to prepare tracking data, fit a habitat-selection or step-selection function, and interpret and visualize the results.

Collared elk in Yellowstone National Park, USA, March 2020.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s