Behavioral Modes Segmentation: GIS Visualization Movement Ecology Lab @ CEAB 15 June, 2012 GIS visualization for segmentation and annotation of animal movement trajectories (for single or few trajectories at a time) Part A: Visualization of unique behaviours by time spent: Part B: A short note on 3D visualization A: Visualization of unique behaviours by time spent: We used geographic Information Systems (GIS) to properly visualize path segmentation and annotation in different trajectories using the same methodology. Tracks used have different spatial characteristics in terms of its resolution and space/area used (Fig. 1). Tracks can be either high resolution (e.g. tracks from GPS technology) or low resolution tracks (e.g. Argos) and can be located at very small or large areas. For instance, migration paths typically move along far away areas in the globe (e.g. from the poles to the equator), whereas foraging paths are typically found within smaller areas (larger scale). This causes visualization problems if using directly each single pair of coordinates (all data in the dataset) for the purpose of representing different tracks in the same way.
Fig 1. Illustration of different spatial track characteristics (resolution of the dataset and space used): A: osprey migration track with around 600 pairs of coordinates (represented as bullet points) for a very large area; B: migration albatross track with around 3000 pairs of coordinates (represented as bullet points) for a relatively small area. As an example of visualization problems, for very high resolution tracks that move large distances, behaviours associated to each pair of coordinates are difficult to be seen when plotting the whole trajectory at a time (small scale), either using points or lines. This is especially problematic when it comes to visualize behaviours that occupy a relative small space (e.g. foraging or resting) in comparison to behaviours that occupy a larger space (e.g. relocating). Fig 2. Illustration of behaviour visualization problems, using lines, for a: A) large area track (osprey migration track); B) small area but high resolution track; C) a zoom of a particular area in B. To overcome this and be able to map behaviours of all trajectories in the same way, we have developed a visualization methodology and protocol that is briefly defined below: Step 1; transforming pairs of coordinates into rhumb lines (Fig 3; A): locations (a pair of latitude and longitude coordinates) are transformed into lines (XY to Line, Data Management, ArcGis 10, ESRI). A line is formed by point i and point i+1, therefore the number of resulting lines is the number of pair of coordinates n-1, being the last point of the track the one deleted. Line type used is the rhumb line or line of constant bearing (direction), as recommended for bird trajectories. Each line is associated to the attributes of its initial pair of coordinates. This is consistent with the methodology used to associate behaviours to each
pair of coordinates, where physical attributes are attributed to each location using the next point (speed) or the surrounding points (turning angle). Step 2; dissolving continuous behaviours into single lines (Fig 3; B): individual small lines joining each coordinate s pair are dissolved into larger lines using the associated behaviour (Dissolve, Data Management, ArcGis 10, ESRI). Larger lines are formed by dissolving smaller and continuous lines associated to the same behaviour. This way each line represents a unique spatial and temporal continuous behaviour within each track. Step 3; converting lines into points (Fig 3; C): lines are converted into points (Points to Lines, Data Management, ArcGis 10, ESRI) using the inside option, so that points are located at the central position of the line forcing the point to be within the line (not the geographic centroid which can result in points found outside the line). Attributes are associated to each point for further qualitative and quantitative visualization. The attribute aggregation depends on the characteristics being evaluated. For instance, time spent in that particular behaviour is the sum of the time spent at each individual line, whereas for instance distance from home is the average. Step 4; final visualization (Fig 3; D): tracks are visualized as coloured points (each colour represents a different behaviour) where the size of the point represents the time spent within that particular behaviour (visualized in three categories as natural jenks). Moreover, we use the aggregated lines as the background layer using a different colour for each behaviour. Fig 3. Illustration of described steps (1-4) for behaviour visualization This methodology has proven to be particularly useful to solve the previous pointed problems and for representing behaviours in a spatio-temporal context (Fig 4; compare with figures 1 and 2).
Fig 4. Comparison of behaviour visualization using a simple line and point method (A) versus the methodology described above (B) However, some potential minor problems have been detected. Further research is needed to improve them: - Overlapping points: in some very specific cases, having overlapping points (points sharing exactly the same coordinates, which means that the animal has not moved or the animal has moved but the coordinate resolution does not capture the movement) may results in length zero lines. Zero length lines do not produce centroid points and therefore, that particular behaviour is not visualized. This only occurs when a behaviour is isolated in-between other behaviours and found only within overlapping points. Potential solutions may be: i) modify the original coordinates so that points never overlap, by adding a very small and unnoticeable visually amount of movement in them; ii) forcing the algorithm to modify those particular situations so that a particular behaviour is never isolated within a group of overlapping points. Apart from this, we have checked that overlapping points that are the continuation or the starting location of a movement (with the same behaviour) do produce lines and therefore, particular zero length lines attributes (those attributes from the part of the overlapping points) do compute for the attribute aggregation (their attributes are not lost).
3D visualization: 3D visualizations allow the visualization and navigation of several parameters at a time, thanks to the 3D capabilities. It is possible to represent values as colours/magnitude through a baseline line, an elevated line (or even extruded), points and extruded points. This allows, for example, the representation of several parameters using these possibilities. For instance, it is possible to represent the turning angle (values 0 and 1) as point values using extruded points (points converted to vertical lines), the speed as lines (baseline lines or elevated lines or both), as well as the annotation (behaviours) as elevated points. The extrusion/elevation values can be expressed as the flying altitude, or as another parameter (e.g. magnitude of the speed, etc.). It is a powerful tool if used as a 3D tool, since it allows interaction and navigation within features. The export as a 2D image is relatively poor in comparison to the in-situ capabilities of the 3D visualization. Fig 5 and 6. Illustration of potential 3D capabilities
Examples of behavioural segmentations obtained with the qsum algorithm
Examples of behavioural segmentations obtained with the tlma algorithm VELOCITY TURNING
BEHAVIOURAL MODES BEHAVIOURAL MODES (short time modes highlighted with large circles)
Behavioural Modes Comparison (zoom in) qsum Algorithm tlma Algorithm