Analytical techniques for the interpretation of satellite derived marine animal locations

• Main Author: Frydman, S
• Format: Thesis
• Language: English
• Published: 2011
• Subjects: geolocation; Argos; kernal smoothing; interpolation; visualisation; utilisation distribution; filter; Antarctic; Austral; Heard Island; Antarc*; Antarctic Fur Seals; King Penguins
• Online Access: https://eprints.utas.edu.au/12442/; https://eprints.utas.edu.au/12442/1/Thesis_Sascha_Frydman.pdf

Over the past few decades there has been a dramatic increase in the scale and complexity of data collected from free ranging animals. Advances in data collection have outstripped the development of sufficiently powerful analytical tools. As such, researchers require the ability to store, analyse, interpret and visualise large quantities of animal derived telemetry data. This is achieved through a combination of statistical analysis algorithms and accompanying data analysis software. During the austral summer of 2003/04, a multi-species animal tracking project was conducted at Heard Island. Over a period of 40 days, the offshore movements of Antarctic fur seals, Macaroni penguins and King penguins were tracked using animal-borne transmitters and the Argos satellite system. This project required real-time analysis capabilities to integrate data from multiple predator movements with other contemporaneous data sources. This thesis describes the development of these tools, from the initial package used to direct field efforts to the creation of sophisticated mathematical approaches to the analysis and interpretation of these data. Though developed in response to the Heard Island project, the work presented here is generally applicable to all Argos derived animal movement data. The outcomes of this work are: 1) A purpose built software system that archived, analysed and visualised satellite derived location data. This system provided researchers with near real-time estimates of areas of intensive foraging activity. It was designed to support the quantitative requirements of the predator-prey study at Heard Island. 2) A concise set of equations for the generation of weighted kernel smoothed utilisation distributions (UD). These equations can directly take geographic coordinates (latitude and longitude) without the need for preliminary conversion to a Cartesian based system such as universal transverse Mercator. UDs of animal location tracks were created using weighted and non-weighted kernels and the effect of using different methods for determining smoothing parameter were explored. This study found that the predictive ability of the UD was most effective when using a weighted kernel in conjunction with likelihood crossvalidation.3) A new type of speed filter that uses an evolutionary optimisation algorithm to incorporate an elliptical location error into the positional estimates provided by Argos. When tested against the output of the commonly used filter defined in McConnell et al. 1992, the optimising filter was found to remove half as many locations. 4) A time weighted probability density function (PDF) of speeds travelled by multiple individuals was used to create a population level estimate of movement behaviour. This PDF was incorporated into an algorithm for interpolating movement between locations. The optimisation framework was used to guide the interpolated system so that its PDF matched that of the population level PDF. This system utilised knowledge of location error in conjunction with the population distribution. The interpolated data was then used to generate a weighted kernel density estimate or UD of the animal’s movements. This method is called model interpolated kernel smoothing (MIKS). A series of validation tests were conducted to assess the performance of this MIKS to simpler forms of UD such as that created with linear interpolation. MIKS was found to out perform the other types of UD. 5) A flexible software platform for the development of location filtering and location interpolation algorithms that incorporates a form of evolutionary optimisation known as extremal optimisation. This system, written in C++, provides all the functionality required to implement the algorithms described in this thesis. It includes a database for the archival and rapid retrieval of Argos satellite data. It also provides a framework for building a spatial location analysis system. Finally, it provides links to various data analysis and presentation packages including ParaView and R. This series of studies provides a set of analytical tools and supporting software for the interpretation of animal movement data. It utilises a novel approach to exploring probability distributions which will create new opportunities for the development of analytical techniques in the future.