Relative habitat selection and resource selection functions in aquatic acoustic telemetry: theory, application, and process
Main-conveners:
Lucas Griffin (University of South Florida, USA)
Jonathan Rodemann (Florida International University, USA)
Co-conveners:
Robert Lennox (OTN, Canada)
About this workshop:
Welcome to our Resource Selection Function workshop! The goal of this workshop is to familiarize participants with the concept of RSFs, present case studies within aquatic acoustic telemetry, and how to run the analyses in R. We will start the workshop by providing background and theory on RSFs. We will then present multiple case studies using RSFs from our and other’s work in aquatic telemetry systems. These case studies will be followed a walkthrough of the data and R code used to conduct RSFs using Random Forest, Generalized Linear, and Generalized Additive Models. R scripts are provided to encourage participants to conduct Resource Selection Functions on their own acoustic data.
Resource selection functions (RSFs), defined as a function that produces values that are proportional to the probability of use by an animal, are a popular method to determine and predict relative habitat selection by animals. These functions evaluate the relationships between resource use (i.e., the units of area selected by an animal) and the environmental characteristics associated with each unit of area. Animal spatial data, from sources such as telemetry, can be incorporated into RSFs to define the relative habitat selection strengths among animal space use and a given set of environmental covariates, such as habitat type, substrate, elevation, or water depth. When the true absences are unknown, as generated by presence only data derived from sources such as telemetry approaches, RSFs are implemented within a use/availability framework where known presences (1) are compared with a random sample across ‘available’ resource units, also known as pseudo-absences or background points (0). Alternative to use/availability (e.g., from telemetry), data from observations collected from survey methods, often without timestamps, are typically referred to as presence-background and are fitted as species distribution models. Using RSFs to derive the relative probability of selection, rather than the absolute probability, telemetry data are then typically fitted using logistic regression models or, as of more recently, with machine learning algorithms [e.g., random forest (RF), boosted regression trees].
Resources
Workshop GitHub – Website that houses data and code for our RSF workshop
RSF publications:
Manly et al. 2007. Resource Selection by Animals: Statistical Design and Analysis for Field Studies
Boyce et al. 2002. Evaluating resource selection functions. Ecol. Model.
Boyce 2006. Scale for resource selection functions. Drivers. Distrib.
Boyce and McDonald 1999. Relating populations to habitats using resource selection functions. Trends Ecol. Evol.
Case study publications:
Sea turtle: Selby et al., 2019. Juvenile hawksbill residency and habitat use within a Caribbean marine protected area. End. Species Res.
Sharks: Griffin et al., 2021. A novel framework to predict relative habitat selection in aquatic systems: Applying machine learning and resource selection functions to acoustic telemetry data from multiple shark species. Front. Mar. Sci.
Seatrout: Rodeman et al. 2024. Finding a home in a fragmented world: Multi-scale habitat selection of Spotted Seatrout in an area of seagrass recovery. in prep.
Other RSF examples
Brownscombe et al., 2022. Applications of telemetry to fish habitat science and management. Canadian Journal of Fisheries and Aquatic Sciences.
Bangley et al., 2022. Modeling the probability of overlap between marine fish distributions and marine renewable energy infrastructure using acoustic telemetry data. Front. Mar. Sci.