Analysis and visualisation of movement: an interdisciplinary review

Abstract

The processes that cause and influence movement are one of the main points of enquiry in movement ecology. However, ecology is not the only discipline interested in movement: a number of information sciences are specialising in analysis and visualisation of movement data. The recent explosion in availability and complexity of movement data has resulted in a call in ecology for new appropriate methods that would be able to take full advantage of the increasingly complex and growing data volume. One way in which this could be done is to form interdisciplinary collaborations between ecologists and experts from information sciences that analyse movement. In this paper we present an overview of new movement analysis and visualisation methodologies resulting from such an interdisciplinary research network: the European COST Action "MOVE - Knowledge Discovery from Moving Objects" (http://www.move-cost.info). This international network evolved over four years and brought together some 140 researchers from different disciplines: those that collect movement data (out of which the movement ecology was the largest represented group) and those that specialise in developing methods for analysis and visualisation of such data (represented in MOVE by computational geometry, geographic information science, visualisation and visual analytics). We present MOVE achievements and at the same time put them in ecological context by exploring relevant ecological themes to which MOVE studies do or potentially could contribute.

Keywords: Animal movement; Computational geometry; Geographic information science; Interdisciplinary developments; Movement ecology; Spatio-temporal analysis; Spatio-temporal visualisation; Trajectories; Visual analytics; Visualisation.

Figure 1

(after [ 103 ] ). A geometric segmentation of a trajectory. Red/pink segments are migration flight, yellow segments are stopovers (one stopover is shown in more detail in the lower right corner). Blue markers indicate end of a stopover. Stopovers are described by staying within a bounded activity radius for at least 2 days. This is a conjunction of a monotone-decreasing criterion (if the sequence (B, C,…, G) stays in a disk of small radius then so does every subsequence, e.g. (D, E, F)) and a monotone-increasing criterion (if the sequence (B,…, G) corresponds to a duration of at least two days then so does every sequence which includes it, e.g., (A, B,…,G, H)).

Figure 2

(after [ 106 ]). a) three trajectories with a common start and end point and b) a median trajectory (bold) representing these three trajectories. The median trajectory is built of segments of the original trajectories and switches the original underlying trajectory at each intersection.

Figure 3

(after [ 127 ]). Progression of a set of trajectories through time (represented as horizontal axis and progression is from left to right). Colours indicate groupings of various sizes based on location and proximity of moving objects at that moment. A beige group has three objects, a yellow one four, an orange one five, a grey one six. At each moment in time the grouping is maximal.

Figure 4

(after [ 135 ]). Eye and mouse trajectories in a visual search task: the participant was asked to identify the target (green square) on a map (not shown) and click on it. a) Eye and mouse trajectories generated in this task. b) Time series plot of distances from eye & mouse to target vs. time. In both charts, eye is in red, mouse is in blue.

Figure 5

(after [ 138 ]). Tree classifying individuals based on spatio-temporal sequential habitat use during May-June. An extract of the habitat use sequences for the first five days (01/05 – 05/05) is shown. Covariates can be associated to each individual and help to identify relations between identified clusters of similar sequential habitat use and animal characteristics.

Figure 6

(after [ 190 ]). Two-dimensional kernels for trajectories that produce two-dimensional density surfaces. The point-based kernels in panel a) do not consider the temporal dimension of trajectory points, but treat them as independent observations in a point data set. Panels b), c) and d) show line-segment kernels, where sequentiality of two consecutive trajectory points is taken into account in kernel construction.

Figure 7

(after [ 182 ]). Two trajectories (blue/purple) with a potential encounter (red/green) computed based on the Brownian bridge movement model.

Figure 8

(after [ 190 ]). Three-dimensional kernels for trajectories that produce volumes in geo-time space. Panel a shows a) point space-time density with cylindrical kernels that do not take into account the temporal sequence of points in a trajectory. Panels b), c) and d) show polyline kernels, where there is one kernel for the entire trajectory (and not a separate kernel for each line segment). Distance from each voxel to trajectory in panel b) (shown in kernel with a dashed grey line) is calculated in 3-D, perpendicularly to the trajectory. Distance from each voxel to trajectory in panels c)-d) is measured at a constant moment in time (i.e. horizontally) and is calculated as 2-D distance. Panel d) shows the Brownian bridges version of the stacked 3-D kernel, where the width of the kernel at each moment in time depends on the position on the trajectory between each two consecutive points.

Figure 9

(after [ 190 ]. Stacked space-time density of animal trajectories. a) Space-time cube representation of one month of trajectories of one individual bird. The x-y plane represents geographic space and the z-axis is time (0-24 hrs). b) Brownian stacked space-time density of the trajectories from the space-time cube. c) Gaussian stacked space-time density of the same data and d) isosurface of the highest values in the Gaussian density, indicating a temporal column and a space-time hotspot.

Figure 10

Eye tracking data shown in the visual analytics tool ISeeCube [ 183 , 197 ] with multiple linked views. The data shows gaze information from multiple participants watching the same video [189]. The visual workspace is separated into several regions that can be freely adjusted by the user. Region a) displays the scanpaths of selected participants in an STC (coloured lines), along with clustered gaze point data (coloured regions on the two grey walls). The STC also contains a snapshot of the video at a time frame that can be adjusted by the user. At the top-left of region a), the same video replay is shown with two areas of interest marked by blue boxes (person and kite). Region b) provides a hierarchical clustering of the trajectories according to the similarity of their distribution of attention to areas of interest. Region c) shows the detailed information of one of the areas of interest (kite), including overall distribution of attention, as well as size and position of the area over time.