2 General Methods
2.4 Track Analysis
The raw data from both types of tag need to be processed before they can be reconstructed into the animal tracks used for subsequent data analysis. The satellite data in particular needs to undergo a geolocation process to ensure the tracks are comparable between individuals, while the acoustic data need to be collated across the receivers into track files that can be exported for analysis. Network analysis, described below, was the primary track analysis technique used for the acoustic data.
2.4.1 Satellite Tag Geolocation
As the Argos positions produced by the satellite tags vary in frequency and quality it was necessary to process the data to obtain normalised positions that were comparable between individuals and over time. The raw Argos positions were processed in three steps, each different error field according to its Argos location class, it was necessary to decide the most
62
probable location for each point within its error field. This was achieved by using a Bayesian state-space model (SSM) that adjusted the filtered tracks by producing regular positions based on the Argos location class, mean turning angle, and autocorrelation in speed and direction, producing the most probable track through the error fields (Jonsen, Flemming & Myers 2005;
Jonsen, Myers & James 2007). The SSMs were applied to the tracks of each individual tiger shark using the R software package (R Foundation for Statistical Computing, Vienna, Austria), primarily using packages ‘bsam’, supported by ‘winBugs’, ‘snow’, ‘dclone’ and ‘rjags’ (Jonsen et al. 2005, 2007). Given that 80.1% of gaps between positions in the present tracks were under 12 hours (Figure 9), a time step of 12 hours was used in the SSM to produce two positions per day for each shark’s track. However, the SSM produces regular positions for the entire track, even on days where there were no raw positions. Consequently all positions for days on which there were no real Argos transmissions were deleted. This step resulted in the normalised track positions and formed the dataset used for the plotting of positions on maps by season and plotting latitude over time to display how the distribution of animals changes over time.
Figure 9: Frequency distribution of time between subsequent geolocations for all sharks.
63
Argos tracks only have locations for when the sharks were at the surface; consequently there is high variability in the number of locations in a given area, as a result of the shark’s varied surfacing behaviour rather than because of its actual location. This would introduce a bias into the analysis of time spent in different areas. To correct this bias, linear interpolation was used to normalise the transmission frequency by generating points at 12 hour intervals along track gaps of <20 days. Where gaps >20 days were encountered the track was split into sections to avoid spurious interpolation. Moreover, in order for space-use analyses to be as conservative as possible, all were conducted at a grid resolution of 0.5°×0.5°, greater than the reported errors of the worst location class (B, 10 km; (Hays et al. 2001; Hazel 2009)). Examples of how track positions varied between each processing step can be found in Figure 10.
Figure 10: Maps to show how the positions varied between each stage of track processing for four different sharks (S7, large male; S12 small female; S15 small male; S16 large female): a = raw Argos positions, b = speed filtered positions, c = SSM positions, d = SSM positions with interpolation on data-less days, e = SSM positions with linear interpolation across gaps <20 days. Maps created in ArcGIS, using GSHHG coastline data and ETOPO2v2 bathymetry data.
To determine track sections with higher turning frequency from those with more directed movement, the ‘straightness’ of individual trajectories was calculated for successive 12 day portions of each SSM processed, linearly interpolated track, where:
64
Straightness = displacement over 12 days / distance travelled over 12 days
Values closer to 1 indicate periods of straighter movement, and values closer to 0 indicate periods of higher turning frequency, providing a proxy for station-keeping or area-restricted searching (foraging) behaviour (Sims 2010). Straightness was calculated over 12 day periods as this was, on average, the time taken for the sharks to traverse a distance greater than the error of the worst location class (B, 10 km; (Hays et al. 2001; Hazel 2009)).
2.4.2 Acoustic Network Analysis
Acoustic arrays have the potential to provide vast quantities of data, however in turn this requires extensive database management (Lowe, Wetherbee & Meyer 2006). All downloaded detections were imported into a Microsoft Access (Microsoft Corporation, Redmond, USA) database, which assigned transmitter detections (pings) to the appropriate sharks and receiver locations, and filtered out any pings that did not match an active tag or receiver (i.e. false positives). Receiver clock-drift time corrections were also made during the import process, being calculated from the difference between the receiver and PC clock at the time of download, assuming linear drift. Tags were detected within 150 m of the receiver, as determined by range testing: mean range 165 m ± 33 (S.D.). This database could then be queried to extract track data under any selection criteria, e.g. by species, size, sex etc.
Network analysis was used to determine both where sharks spent more time and how they moved through the array (Jacoby et al. 2012). Each receiver location was treated as a node within the network, with node strength weighted according to the number of detections at that location. Any pair of subsequent pings that occurred between different nodes was treated as a connection between those nodes, with connection strength weighted by the number of
65
times that specific pairing occurred. In this way matrices were constructed that detailed the connections between receivers and the detections at each receiver, allowing networks to be constructed and graphed to visualise shark movements and occupancy throughout the array for each species.
Due to the different ping frequencies of the V13 and V16 tags (180 s vs. 120 s nominal delays), the node and connection strengths of V13 networks were increased by 50% to account for the decreased probably of detection compared to the V16 networks. All network maps were produced using ArcGIS (ESRI Inc., CA, USA), with bathymetry data obtained from the U.S.
Department of Commerce, National Oceanic and Atmospheric Administration (NOAA): 2-minute Gridded Global Relief Data (ETOPO2v2).
Several network metrics were used to describe each network: occupancy (or node strength) was computed from the number of detections occurring at each node and provided a measure of how much time individuals spent at each receiver location. Connectivity (or node centrality) is calculated from the total number of connections made to that node, i.e. the proportion of other nodes to which there is a connection. Transit (or node betweenness) represents the total number of paths to pass through that node and is computed by counting pings occurring at a receiver where the prior and subsequent pings for that individual occur at a different receiver.
Transit therefore measures the extent to which a node is part of a corridor of movement as opposed to an area of occupancy. Node density is the proportion of total available nodes actually used in the network, measuring the extent of the array occupied, and edge density is the proportion of total available connections actually formed within the network, providing a measure of mobility within the network, both ranging 0–1.
66