Identifying Movement States From Location Data Using Cluster Analysis

Abstract

Animal movement studies regularly use movement states (e.g., slow and fast) derived from remotely sensed locations to make inferences about strategies of resource use. However, the number of movement state categories used is often arbitrary and rarely inferred from the data. Identifying groups with similar movement characteristics is a statistical problem. We present a framework based on k‐means clustering and gap statistic for evaluating the number of movement states without making a priori assumptions about the number of clusters. This allowed us to distinguish 4 movement states using turning angle and step length derived from Global Positioning System locations and head movements derived from tip switches in a neck collar of free‐ranging elk (Cervus elaphus) in west central Alberta, Canada. Based on movement characteristics and on the linkage between each state and landscape features, we were able to identify inter‐patch movements, intra‐patch foraging, rest, and inter‐patch foraging movements. Linking behavior to environment (e.g., state‐dependent habitat use) can inform decisions on landscape management for wildlife.

Incorporating within- and between-patch resource selection in identification of critical habitat for brood-rearing greater sage-grouse

“…The estimated relative displacement index was one approach to incorporating information on behavior to strengthen inference on resource needs (Figures 1 and 2). Other approaches to identify behavioral modes or to assign data to defined modes have involved random walk models (Wu et al 2000), cluster analysis (Van Moorter et al 2010), state-space models (Jonsen et al 2005), fractals (Etzenhouser et al 1998;Webb et al 2009), and generalized additive models (Dzialak et al 2011a). Results of the resource Raster surfaces depicting encamped and traveling behavior of greater sage-grouse (Centrocercus urophasianus) in south-central Wyoming, USA, during 2009 to 2012 were developed with the predicted probability of occurrence classified into ten relative probability bins (1 = lowest, 10 = highest) that included 10% of the landscape area.…”

“…Efforts to identify and prioritize important wildlife habitat are particularly relevant in human-modified areas where wildlife conservation may be one of the several valued land uses or where protecting large contiguous areas of unmodified habitat is not an option (Moilanen et al 2005). Thus, incorporating information on behavior in predictive modeling can play an important role in developing accurate predicted occurrence maps that depict habitat in terms of its composition and configuration for different behaviors (Taylor et al 1993;Tischendorf and Fahrig 2000;Kindlmann and Burel 2008;Van Moorter et al 2010).…”

“…From a spatial perspective, a behavioral mode can be thought of as a manner of movement characterized by parameters such as movement rate, net displacement, or tortuosity (Patterson et al 2008). For example, in some species, behavior characterized by high directional persistence, a high rate of movement, and high displacement will reflect a between-patch mode such as traveling or migration, whereas high tortuosity, a low rate of movement, and low displacement will reflect a within-patch mode such as encampment, resting or foraging (Van Moorter et al 2010). From a management perspective, species-specific behavioral responses to the landscape establish a mechanistic basis for understanding how particular features function ecologically (Tischendorf and Fahrig 2000).…”

“…From a management perspective, species-specific behavioral responses to the landscape establish a mechanistic basis for understanding how particular features function ecologically (Tischendorf and Fahrig 2000). Behavioral mode is a variable that the investigator may wish to characterize before conducting further analysis (Van Moorter et al 2010) or incorporate as a latent process into structural equation or time-series models (Patterson et al 2008). Depending on the scale of investigation, dominant breeding season behavioral modes of sage-grouse might include pre-nesting behavior such as lek attendance, nest-site selection, nesting off-bout behavior, and between-versus within-patch movement of the brooded female (Aldridge and Brigham 2002;Aldridge and Boyce 2007;Doherty et al 2010;Dzialak et al 2011b).…”

Critical Considerations for an Urban Deer Collaring Program

“…GPS technology offers researchers the ability to monitor fine-scale movement with a high degree of accuracy [11]. Urban deer collaring programs, however, pose unique challenges compared to collaring programs designed for more remote locations.…”