Animal Movement: Advances in Movement Modeling and Their Applications

ROOM: HCCC, Room 25C
Animal behavior, population dynamics, and landscape ecological processes are all emergent properties of movement. Understanding and quantifying how movement patterns vary across scales and modes, and is influenced by landscape structure, is an important step towards refining ecological theory and has important implications for applied management and conservation of wildlife.This symposium will bring together experts in the fields of animal movement and movement modeling to discuss the progress and future of these ideas, and in particular, to demonstrate how they can be used to solve problems of applied importance.

8:10AM The causes and consequences of plasticity in elk migration timing across the Greater Yellowstone Ecosystem.
  Gregory J. Rickbeil; Jerod Merkle; Paul Atwood; Jon Beckmann; Erik K. Cole; Alyson B. Courtemanch; Sarah Dewey; David D. Gustine; Matthew J. Kauffman; Douglas E. McWhirter; Kelly Proffitt; P.J. White; Arthur D. Middleton
Migration – involving both perception and memory – is an effective behavioral strategy for prolonging access to seasonal resources and may be a resilient strategy for changing climatic conditions. In the Greater Yellowstone Ecosystem (GYE) elk are the primary ungulate, with approximately 20 000 individuals migrating to exploit seasonal gradients in forage while also avoiding energetically costly snow conditions. How climate induced changes in plant phenology and snow accumulation are influencing elk migration timing is unknown and critical for assessing how climate change may impact future elk migrations. Here, we present the most complete record of elk migration across the GYE, spanning nine herds and 414 individuals. When elk left winter range involved a trade-off between current and future forage conditions, while snowmelt governed summer range arrival date. Departure from summer range and arrival on winter range were both influenced by snow accumulation and exposure to hunting. At the GYE scale, migration timing changed in spring and fall, most notably with winter range arrival date becoming 55 days later since 2001. Similarly, snowmelt, snow accumulation, and spring green-up dates have all shifted through time with different herds experiencing different rates and directions of change. Plasticity in elk migration timing may indicate a resilience to future climate change; however, the demographic implications of such changes are unknown. Additionally, changes in elk migration timing will undoubtedly affect predator-prey dynamics, disease ecology, and game management across the GYE and may result in a divergence of migration timings between the western and eastern herds.
8:30AM Basis Functions for Continuous-Time Models of Animal Movement
  Frances E. Buderman; Mevin B. Hooten
Animal movement models typically fall within one of three broad categories: point process models, discrete-time models and continuous-time models. We developed a flexible, computationally efficient alternative to traditional continuous-time movement models (e.g., Brownian bridges, continuous-time correlated random walk) that uses basis functions to model the movement process. Basis functions are useful for modeling autocorrelation, and they form the foundation of a number of modeling approaches, including generalized additive models. They are particularly useful for modeling unknown, complex, potentially non-stationary, continuous processes. There are a number of benefits to using basis functions to model animal movement. First, they take an easily recognizable form and the basis functions can be generated by existing software. Second, they can be incorporated into hierarchical models to allow for inference on both the ecological process and the observation process (e.g., state-space models). Third, animal movement inference scales with the amount of available data and does not require observations to be collected at regular intervals. Fourth, they can be incorporated into two-stage models for inference on behavioral states, network associations, and movement drivers. I will discuss these benefits, when and when not to use a basis function approach for animal movement inference, and how these methods can be implemented using standard software.
8:50AM Animal movement models for migratory individuals and groups
  Mevin B. Hooten; Henry R. Scharf; Trevor J. Hefley; Aaron T. Pearse; Mitch D. Weegman
Animals often exhibit changes in their behavior during migration. Telemetry data provide a way to observe geographic position of animals over time, but not necessarily changes in the dynamics of the movement process. Continuous‐time models allow for statistical predictions of the trajectory in the presence of measurement error and during periods when the telemetry device did not record the animal’s position. However, continuous‐time models capable of mimicking realistic trajectories with sufficient detail are computationally challenging to fit to large datasets. Furthermore, basic continuous‐time model specifications (e.g. Brownian motion) lack realism in their ability to capture nonstationary dynamics.We present a unified class of animal movement models that are computationally efficient and provide a suite of approaches for accommodating nonstationarity in continuous trajectories due to migration and interactions among individuals. Our approach uses process convolutions to allow for flexibility in the movement process while facilitating implementation and incorporating location uncertainty. We show how to nest convolution models to incorporate interactions among migrating individuals to account for nonstationarity and provide inference about dynamic migratory networks.We demonstrate these approaches in two case studies involving migratory birds. Specifically, we used process convolution models with temporal deformation to account for heterogeneity in individual greater white‐fronted goose migrations in Europe and Iceland, and we used nested process convolutions to model dynamic migratory networks in sandhill cranes in North America.The approach we present accounts for various forms of temporal heterogeneity in animal movement and is not limited to migratory applications. Furthermore, our models rely on well‐established principles for modeling‐dependent data and leverage modern approaches for modeling dynamic networks to help explain animal movement and social interaction.
9:10AM Spatial Capture-Recapture for Migratory Populations with Application to North Atlantic Right Whales
  Daniel W. Linden
Spatial capture-recapture (SCR) models traditionally assume stationarity in the centers of spatial activity that represent the average location (i.e., home range) of individuals in a population. The stationarity allows spatial encounter data to inform capture probabilities across space during a given sampling period. Extensions to the standard SCR model have been developed to accommodate individuals that may be transient or dispersing. Migratory populations contain individuals exhibiting a predictable movement that might naively be interpreted as transience, but could be leveraged with reasonable constraints to inform the capture process for surveys conducted across large spatial and temporal extents. North Atlantic right whales were historically monitored using a near census of the population, as individuals had high annual capture probabilities (>0.90) during systematic aerial surveys conducted on both the breeding and non-breeding grounds. Capture probabilities have declined in recent years, potentially due to changes in movement across the species range. I propose a migratory SCR model that accommodates structured movement across space and time described at the population level, affording improved individual capture probabilities and the ability to estimate migratory parameters.
9:30AM Inferring Spatial Memory From Animal Movement Data
  Eliezer Gurarie; Chloe Bracis
The ability of wild animals to navigate and survive in complex and dynamic environments can be explained, in large part, by the ability to store relevant information and place it in a spatial context. Despite its self-evident centrality of spatial memory, and given our increasing ability to observe animal movements in the wild, it is perhaps surprising how steep a challenge it has proven to demonstrate spatial memory. On its face, the task seems straightforward: If a previous experience explains a future behavior, there is evidence of memory. The principle difficulties are the impossibility of observing cognition directly, and the number of confounding variables that influence movements in a natural environment. To infer a cognitive signal at least three ingredients are necessary: (a) a data set that includes events that are important enough to remember; (b) a plausible mechanism for a movement response to those events; and (c) a statistical framework to estimate the parameters from movement data. In this talk, I will review some empirical studies demonstrating spatial memory in wild animals. I then present a cognitive analysis of movements of several wolves (Canis lupus) in Finland during a summer period of intensive hunting and den-centered pup-rearing. Over two months, we collected movement data and identified locations and timing of kills of nearly all prey killed by tracked animals. We hypothesize an intrinsic, spatially explicit value assigned to hunting and territorial patrolling that is based on memory of predation success and territorial marking. The framework allows for estimation of multiple cognitive parameters, including temporal and spatial scales of memory. I show how memory-based models outperform other models in predicting movements and interpret these results with respect to differences among wolf packs. Finally, I discuss how incorporating spatial memory can improve movement modeling more broadly.
09:50AM Break
12:50PM Stochastic Differential Equation Models for Animal Movement
  Ephraim M. Hanks
We outline past and current approaches for modeling animal movement using stochastic processes, with a particular emphasis on stochastic differential equations (SDEs). We illustrate how SDEs, paired with state-switching processes, can capture a wide range of realistic movement behavior, including starting and stopping behavior, changes in directional bias and absolute movement rates, response to the local environment, attraction to or repulsion from other animals, and group movement behavior. We also show how numerical approximations to SDE models can be chosen to allow for computationally-efficient computing and statistical inference. We illustrate the use of these models to understand animal movement from multiple systems.
1:10PM A Comparison of Data Types and Connectivity Models for Capturing Dispersal Movement in Large Carnivores
  Kathy Zeller; Megan Jennings; T. Winston Vickers; Holly B. Ernest; Samuel Cushman; Walter Boyce
The underlying assumption of all connectivity models and wildlife corridors is that they promote movement of individuals across the landscape. However, due to the difficulty of obtaining movement data, particularly dispersal data, many connectivity models are not informed by movement directly, but instead use opportunistic detection data, location data from systematic surveys or telemetry studies, or genetic data as proxies. Others assume data from territorial movements, such as locations from when an individual is in a movement state or movement paths, will adequately estimate dispersal movement. Using a large empirical data set on puma (Puma concolor), we compared several of the most common approaches for modelling connectivity and validate them with dispersal data. We first estimated resistance to movement using opportunistic detection data, GPS telemetry data from puma home ranges, and genetic data with a variety of analytical methods. We then modeled connectivity with cost distance and circuit theory algorithms and measured the ability of each data type and connectivity algorithm to capture the locations of dispersing pumas. We found that connectivity models based on GPS telemetry points and paths outperformed those based on opportunistic detection data when applied using the cost distance connectivity algorithm. We also found these models outperformed those that used the circuit theory algorithm. We recommend the use of path or point selection functions derived from GPS telemetry data or landscape genetic models to estimate resistance to movement for wildlife. In cases where resource limitations prohibit the collection of GPS collar or genetic data, our results suggest that species distribution models, while weaker, may still be sufficient for resistance and connectivity modeling. We also recommend the use of cost distance-based approaches, such as least-cost corridors and resistant kernels, for estimating connectivity and identifying functional movement corridors for terrestrial wildlife.
1:30PM Statistical Inference about Landscape Connectivity from Animal Telemetry Data.
  Chris Sutherland; J Andy Royle; Angela Fuller
Landscape connectivity is an important tool for effective conservation and management of rare species and provides important information for designing corridors and reserve networks. Formal inference about functional connectivity requires direct observations of individual movement and an explicit model of connectivity that links observed movement outcomes to landscape structure. Despite considerable work on analytic methods for modeling animal movement and dispersal, less work has been done on formal estimation of connectivity using such models. Here we develop a statistical framework for formal inference about functional connectivity from telemetry data. We parameterize a Markovian movement model with an explicit least-cost path model in which “distance” between successive locations is a cost-weighted function of local habitat or other covariates, and also the distance from an individual’s home range center. We demonstrate the proposed method using data from a study of black bears (Ursus americanus) in New York to evaluate the effects of forest, agriculture, and elevation on bear movement.
1:50PM Movement-Assisted Localization From Acoustic Monitoring Data
  Andy Royle; Nathan J. Hostetter
Acoustic telemetry is widely used to study movement of a type ofwildlife known as “fish” and other aquatic species. Many large-scalecooperative telemetry networks have been deployed across North Americato study movement, habitat use and spatial ecology of a variety ofaquatic species. A key objective of acoustic monitoring studies,including both active acoustic methods such as telemetry studies or inpassive acoustic monitoring, is localization of the data from arraysof receivers, i.e., estimation of the location of an individual fromdetections at one or more sensors of an array. Localization isessentially statistical triangulation which can be done when signalsare obtained from an array of sensors so that potentially multipledetections of the same signal are possible. Localization may be basedon simple detection history information (the pattern of sensors atwhich detections occur) and also auxiliary information on time delayof arrival at different sensors, or signal strength. Localization ofsources is crucial for studies of spatial ecology, resource selection,density estimation, and individual behavior over time. Existingapproaches to localization from operational acoustic arrays lackgenerality and make inefficient use of the data obtained from acousticmonitoring. One important source of information lacking from currentlocalization methods is information derived from the underlyingmovement process. Intuitively the location of an individual at timet-1 and even at time t+1 should be informative about the location ofthe individual at time t, regardless of whether or not it wasobserved. We develop a method of localization from acoustic telemetrywhich integrates standard ideas of localization with an explicit modelof movement. This leads to improved localization, and provides deeperinsight into movement dynamics, resource selection and individualbehavior. Moreover, the formulation extends to a fullypopulation-level model, allowing for joint inference about populationdensity in addition to spatial population structure.
2:10PM Using Multiple Data Streams to Improve Our Understanding of Animal Movement
  Marie Auger-Méthé; Kristin Bøe; Joanna Mills Flemming
Movement data have become essential to understand the ecology of animals. However, tracking the movement of many species (e.g. fish and small birds) is still limited to inaccurate technology, such as light-based geolocation (mean error ≈ 200 km). Making behavioral and spatial inferences based on such inaccurate data is difficult, as the tracks they create are not good representations of the animals’ movement. Through a set of examples, I will demonstrate how incorporating auxiliary information into a state-space modeling framework can improve our understanding of animal movement. For example, using the movement data from Atlantic salmon (Salmo salar), I will show how acoustic detections, which provide a record of the time an individual comes close to a receiver, auxiliary behavioral information (e.g. diving data), and spatial barrier information (e.g. coastlines) can increase the accuracy of the predicted movement track.
2:30PM Refreshment Break
3:20PM Integrating Continuous-Time Animal Movement with Spatial Capture-Recapture: Inferring Habitat Use and Animal Interactions From Camera Trap Surveys
  Richard Glennie
In many camera-trap surveys, individual animals can be (partially) identified from their photographs and the exact time of each encounter between an individual and a camera is recorded. These records provide information on the population density and the movement of individuals over time. Existing continuous-time, spatial capture-recapture methods can be used to describe how the encounter rate between cameras and individuals changes in space and time; however, current methods assume encounters are independent given an individual’s activity centre, a single point in space that represents the average location of the individual over the entire survey. Repeated encounters with different cameras provides information on how animals move; this is ignored by the standard approach. Here, an explicit model for animal movement is incorporated with continuous-time spatial capture-recapture models. Each individual moves along an unobserved, continuous path such that encounters between nearby cameras are correlated in time. An individual’s movement path can depend on temporal and environmental covariates to account for landscape connectivity, movement corridors, and boundaries. Inference can be made on where individuals spend their time during the survey and on how individuals interact over time and space. This model makes more use of the temporal information collected during camera-trap surveys. The method is applied to a camera-trap survey of male jaguars (Panthera onca) in the Cockscomb Wildlife Sanctuary Basin, Belize. Jaguars are assumed to move around their activity centre according to a home-range movement process. Telemetry data collected on two jaguars in the survey region is also incorporated into the analysis. A model is considered where jaguar movement is related to the rivers and existing trail network.
3:40PM Animal movement and mortality risk
  Richard B. Chandler; Elina Garrison; Karl V. Miller; Mike Conner; Daniel Crawford; Brian Kelly; Kristin Engebretsen; Heather Abernathy-Conners; Lydia Stiffler; Hunter Ellsworth; Michael J. Cherry
An individual’s mortality risk is likely to change with its location as it moves throughout its home range. However, most survival analyses ignore this fundamental source of variation in the hazard rate, hindering efforts to understand how movement behavior and environmental variables influence mortality risk. One of the reasons why location effects are often ignored is that it is impossible to directly observe an individual’s entire movement path over the duration of a study. However, GPS telemetry and statistical movement models enable inference about the locations of individuals at any point in time by building probabilistic paths between known animal locations. As a result, spatio-temporal covariates associated with movement paths can be included in a survival analysis to evaluate hypotheses about the effects of environmental variables on mortality risk. We developed a joint movement and survival model, and we applied the model to GPS telemetry data collected on 241 collared white-tailed deer (Odocoileus virginianus) from January 2015 – May 2018 in South Florida where deer are the primary prey of the endangered Florida panther (Puma concolor coryi). The model provided insights about individual variation in mortality risk that would not have been revealed by standard survival models that ignore movement.
4:00PM Integrating Spatial Capture-Recapture and Telemetry Data to Jointly Estimate Abundance and Movement
  Nathan J. Hostetter; Sarah J. Converse; Eric V. Regehr; J. Andrew Royle; Ryan R. Wilson
Capture-recapture and telemetry methods are frequently used to evaluate demography, habitat use, and movement. In general, capture-recapture data inform demographic rates (e.g., abundance, survival), while telemetry methods provide information on individual-level movement and habitat use. Although capture-recapture and telemetry data each provide information on population dynamics, they are generally analyzed separately. Using both simulated and real data, we present a framework for integrating spatial capture-recapture (SCR) and telemetry data to evaluate movement dynamics and abundance for animals that range over large areas. We parameterize the movement process using step lengths and turning angles, while the observation process integrates SCR and telemetry data, and can accommodate challenges such as irregularly timed observations, unequal search effort, and movement on and off a study area. Ongoing polar bear research programs in the Chukchi Sea provide motivation for this work, where annual helicopter surveys collect SCR data in a circumscribed study area while GPS collar data inform daily movements across a much larger region. Polar bear movements were characterized by large step lengths and directional persistence, resulting in bears leaving the study area during the survey period, low numbers of recaptures, and little information to estimate abundance from SCR data alone. Integrating telemetry data allowed estimation of within-season movement dynamics without requiring the assumption of stationary activity centers common in SCR models. Simulation results suggest movement parameters may be poorly estimated when recapture rates are low and movements are large relative to the study area. Abundance parameters appear more robust, where even simple movement models (e.g., simple random walk) may be sufficient if abundance estimation is the primary objective. More broadly, this formulation extends SCR models by incorporating within-season movement processes allowing for combined population-level inference on movement, habitat use, and abundance. Our next steps will involve extending these models to multi-season contexts.
4:20PM Panel Discussion

Organizers: Chris Sutherland, University of Massachusetts, Amherst, MA; Andy Royle, USGS Patuxent Wildlife Research Center, Laurel, MD; Juan Manuel Morales, Inbioma-Conicet, Argentina
Supported by: Biometrics Working Group

Location: Huntington Convention Center of Cleveland Date: October 10, 2018 Time: 8:10 am - 5:00 pm