Chapter 7: General Discussion 130
7.4 Future Directions 134
The algorithms and analysis presented in this thesis have provided working solutions to a number of issues that arise from the analysis of Argos satellite
location data. This work was undertaken with the intention of making a contribution to the animal tracking and spatial analysis literature. While an
extensive body of work already exists in these fields, it is still a very active area of research. In this section I will describe some issues that have arisen in the
development of this work and ideas for future improvements and ongoing research.
Spatial utilisation distribution (UD) maps were created using various forms of kernel density estimation (KDE) both with and without location interpolation. Validation of each of these techniques was achieved using multiple comparisons of randomised location track subsets (Chapter 3). Statistically, the results of these tests (Chapters 3 and 4), provided convincing evidence in favour of the use of model interpolated kernel smoothing (MIKS) over other simpler techniques. The use of MIKS also produced unexpected deviations away from a more intuitively expected linear travel path. Regardless of the strength of the statistically
significant result in this analysis, the fact remains that the only true test of the applicability and accuracy of these methods is with knowledge of the actual movements of the animal in question (Matthiopoulos 2003). The addition of global positioning system (GPS) technology to the Argos satellite tracking system is now possible through the advent of Fastloc GPS (Rutz & Hays 2009). Recent studies have used this system to assess the error levels of the Argos location classes (Costa et al. 2010; Patterson et al. 2010). Similarly, this technology could be used to give valuable insight into the effectiveness of interpolation and
smoothing. Such a study could provide definitive data on the ability to predict movement from sparse location data sets using techniques such as MIKS (Chapter 4) and state space modelling (Patterson et al. 2008).
The MIKS algorithm (Chapter 4) and its associated kernel smoothing methods (Chapter 3) all make use of the McConnell et al. (1992) speed filter to pre-process the data. The reason that the more efficient optimisation method for speed filtering (Chapter 5) was not used was due to the adjustments that this technique makes to the retained Argos locations. In shifting the actual location of the data, the distribution of error as defined by the location class is assumed to have also changed. In its current form, the use of this filter is restricted to studies that do not require quantitative estimates of location error. In a future study to
further develop this technique, a methodology to redefine location error estimates will need to be conducted. It is expected that such a study will require data from the recently developed Fastloc GPS tags. Once this is done, incorporation of the optimising speed filter into methodologies that utilise the location error
distribution will become possible.
The concept behind the development of MIKS was to provide a configurable feedback mechanism that would guide the operation of an interpolation algorithm. The implementation of MIKS in Chapter 4 utilised a probability distribution function (PDF) of inter-location travel speeds as a measure of population behaviour. A goodness-of-fit statistic was used to assess the ongoing state of the interpolated data against the speed PDF. This process produced an interpolated UD with a structure that closely matched the measured population behaviour. In future applications of the MIKS method it is expected that other measures of behaviour may be incorporated. For instance, a common practice when tracking marine diving predators is to tag the same individual with an archival time depth recorder (TDR) (Boveng et al. 1996; Burns & Castellini 1998). The TDR is a small, battery powered instrument that uses a pressure sensor to store regular measurements of depth on a computerised memory chip (Burns & Castellini 1998). These tags are used to provide high temporal and spatial resolution data of an animal’s vertical displacement in the water column (Chapter 2). When Argos location data and TDR dive data are collected in unison, the potential exists for integration of the two data sets in order to gain a three-dimensional perspective of the animal’s movements (Georges et al. 2000; Hays 2004). Using MIKS, it would be possible to add an objective that restricts the surface level probability distribution based on an index of diving activity. The rationale for this approach being that if an animal is spending its time
continuously diving, it would most likely remain in the same general location during this period. Of course, the validity and strength of this assumption would be species dependant so the design of this objective would include an adjustable parameter that could be set using knowledge of the species diving behaviour. An additional outcome could occur as a result of having more refined location estimates. The geo-referencing of diving activity which is achieved by
accurately defined. In Chapter 2, some preliminary location, dive integration functionality was described. Owing to the complexity of the location analysis that was presented in subsequent chapters, a thorough exploration of the integration of diving and location data was determined to be beyond the scope of this thesis. However, the existing data set from the Heard Island expedition that was used throughout this study does have a complete set of paired Argos locations and TDR dive records. It is therefore envisaged that the development of a technique for the incorporation of diving activity into the MIKS algorithm will be undertaken in the near future.
The aims of this thesis were necessarily focussed on the application of analysis solutions to Argos supplied data. However, the issues encountered when working with an inaccurate and irregular data source are not limited to satellite location tracking. The analysis techniques presented in this study all come down to the measurement, interpretation and manipulation of probability distributions. Given the generic nature of this statement, it follows that methodologies like MIKS may be applicable to many areas of research. Some possibilities to consider might be; light level geo-location of animal-movement that uses recorded twilight times to estimate location (Sumner et al. 2009); inclusion of other contributing factors such as weather, bathymetry and ocean currents; geographic layers that use knowledge of land masses or other potential
obstructions to guide the system away from unfeasible locations e.g. a seal does not travel across land; non-animal related work such as modelling vehicular traffic flows for the purpose of reducing traffic jams (Karaaslan et al. 2009) or accident prevention (Tiwari 2000).