Classifying behavior from short-interval biologging data: An example with GPS tracking of birds

Abstract

Recent advances in digital data collection have spurred accumulation of immense quantities of data that have potential to lead to remarkable ecological insight, but that also present analytic challenges. In the case of biologging data from birds, common analytical approaches to classifying movement behaviors are largely inappropriate for these massive data sets.We apply a framework for using K-means clustering to classify bird behavior using points from short time interval GPS tracks. K-means clustering is a well-known and computationally efficient statistical tool that has been used in animal movement studies primarily for clustering segments of consecutive points. To illustrate the utility of our approach, we apply K-means clustering to six focal variables derived from GPS data collected at 1-11 s intervals from free-flying bald eagles (Haliaeetus leucocephalus) throughout the state of Iowa, USA. We illustrate how these data can be used to identify behaviors and life-stage- and age-related variation in behavior.After filtering for data quality, the K-means algorithm identified four clusters in >2 million GPS telemetry data points. These four clusters corresponded to three movement states: ascending, flapping, and gliding flight; and one non-moving state: perching. Mapping these states illustrated how they corresponded tightly to expectations derived from natural history observations; for example, long periods of ascending flight were often followed by long gliding descents, birds alternated between flapping and gliding flight.The K-means clustering approach we applied is both an efficient and effective mechanism to classify and interpret short-interval biologging data to understand movement behaviors. Furthermore, because it can apply to an abundance of very short, irregular, and high-dimensional movement data, it provides insight into small-scale variation in behavior that would not be possible with many other analytical approaches.

Keywords: GPS telemetry; K‐means clustering; bald eagle; behavioral classification; biologging data; path segmentation; short‐interval data; unsupervised learning.

FIGURE 1

Distributions of eight focal variables associated with GPS data collected from bald eagles in Iowa, USA. Focal variables were standardized (after square‐root transform when necessary to reduce skewness) then used as input into a K‐means cluster analysis to classify flight behavior of these birds. See text for details on analysis

FIGURE 2

Plots of (a) within‐cluster sum‐of‐squared distances between each point and the cluster centroid; and (b) bootstrapped average silhouette width as a function of number of specified clusters K. In (b), the gold line represents the mean across all 1000 bootstrap samples of the average silhouette widths. Cluster centroids were determined by a K‐means analysis of standardized GPS telemetry data collected from bald eagles in Iowa, USA. See main text for additional details on data collection and analysis

FIGURE 3

Examples of biplots of the first two principal components of raw GPS telemetry data collected from a single bald eagle color‐coded by cluster membership for K ∈ {2, 3, 4, 5, 6, 7}. Input variables are those used in the K‐means clustering as described in the main manuscript (i.e., the GPS data in Table 1, transformed and standardized). Making these plots for each value of K in addition to the bootstrapped average silhouette widths provides additional insight into which number of K is most appropriate. In each biplot, the number of colors present equals the value of K indicated

FIGURE 4

Loadings and biplot from principal component analysis of GPS telemetry data collected from bald eagles in Iowa, USA. Input data were the 6 focal variables shown in Table 1 in the main text. The first two components accounted for, respectively, 45% and 21% of the total variability in the covariates. Points on biplot are from a representative bird and are color‐coded by cluster membership assignment for K = 4 clusters

FIGURE 5

Boxplots of variables relevant to clustering short‐interval GPS data from bald eagles. Boxes visualize 25th, 50th, and 75th percentiles. Whiskers extend either to the most extreme data value or 1.5 * IQR (IQR = 75th–25th percentile) from the nearest quartile, whichever is closest. Variables chosen were those with high factor loadings from a principal components analysis of GPS variables used in K‐means clustering with K = 4, as described in the main text. AGL, altitude above ground level; KPH, kilometers per hour

FIGURE 6

Plots showing details of movements and behaviors of bald eagles. Movement data were collected by GPS telemetry collected at short time intervals. GPS data were then assigned to behavioral categories via K‐means clustering. In each panel of the figure, the left‐most plot shows the map of the bird's movement, plotted on a UTM scale. The top right plot shows the flight speed of each point plotted sequentially over time. The bottom right plot shows the altitude above ground level, also plotted sequentially over time. In all three panels, points are color‐coded based on the behavioral mode identified by the K‐means clustering

FIGURE 7

Relationship of behavioral classifications with life stages of eagles. Behavioral classifications were assigned by experts with a strong background in eagle ecology and behavior and associated with K‐means clusters of GPS telemetry data from bald eagles in Iowa, USA. Life stages of eagles were determined by gross movement characteristics of birds (i.e., were their movements migratory or local in nature). See main text for additional details on clustering and assignment to life stages. Bold numbers under bars indicate marginal percent of points in each flight stage

FIGURE 8

Relationship of behavioral classifications with age classes and life stages of eagles. Behavioral classifications were assigned by experts with a strong background in eagle ecology and behavior and associated with K‐means clusters of GPS telemetry data from bald eagles in Iowa, USA. Life stages of eagles were determined by gross movement characteristics of birds (i.e., were their movements migratory or local in nature). Ages were estimated when eagles were marked. See main text for additional details on clustering and assignment to age classes and life stages. Bold numbers under bars indicate marginal percent of points in each age class and of each flight stage

FIGURE 9

Relationship of length of behavioral subsegments identified in GPS telemetry data and the behavioral mode of the next subsegment. Relationships are shown by behavioral mode. Behavioral classifications were assigned by experts with a strong background in eagle ecology and behavior and associated with K‐means clusters of GPS telemetry data from bald eagles in Iowa, USA. See main text for additional details on clustering. Numbers under bars indicate marginal percent of points in each time bracket for the given behavioral mode