A probabilistic algorithm to process geolocation data.

Børge Moe,Benjamin Merkel,Richard A. Phillips,Nigel G. Yoccoz,Sébastien Descamps,Hallvard Strøm

doi:10.1186/s40462-016-0091-8

Abstract

BackgroundThe use of light level loggers (geolocators) to understand movements and distributions in terrestrial and marine vertebrates, particularly during the non-breeding period, has increased dramatically in recent years. However, inferring positions from light data is not straightforward, often relies on assumptions that are difficult to test, or includes an element of subjectivity.ResultsWe present an intuitive framework to compute locations from twilight events collected by geolocators from different manufacturers. The procedure uses an iterative forward step selection, weighting each possible position using a set of parameters that can be specifically selected for each analysis.The approach was tested on data from two wide-ranging seabird species - black-browed albatross Thalassarche melanophris and wandering albatross Diomedea exulans – tracked at Bird Island, South Georgia, during the two most contrasting periods of the year in terms of light regimes (solstice and equinox). Using additional information on travel speed, sea surface temperature and land avoidance, our approach was considerably more accurate than the traditional threshold method (errors reduced to medians of 185 km and 145 km for solstice and equinox periods, respectively).ConclusionsThe algorithm computes stable results with uncertainty estimates, including around the equinoxes, and does not require calibration of solar angles. Accuracy can be increased by assimilating information on travel speed and behaviour, as well as environmental data. This framework is available through the open source R package probGLS, and can be applied in a wide range of biologging studies.Electronic supplementary materialThe online version of this article (doi:10.1186/s40462-016-0091-8) contains supplementary material, which is available to authorized users.

Highlights

The use of light level loggers to understand movements and distributions in terrestrial and marine vertebrates, during the non-breeding period, has increased dramatically in recent years
A number of methods have been developed to estimate locations from light data (Table 1), and to filter the resulting outputs in various ways [14,15,16,17]. These are mainly based on either a threshold [7, 18] or template-fit approach [19]. In the former, longitude is computed from the timing of local noon, and latitude from day length, based on the timing of twilight events which are determined using a pre-defined light intensity threshold
The aim of this paper is to propose an intuitive, probabilistic algorithm, implemented in R [22] through the new package probGLS, that can be used on data from all existing geolocator brands

Summary

Introduction

The use of light level loggers (geolocators) to understand movements and distributions in terrestrial and marine vertebrates, during the non-breeding period, has increased dramatically in recent years. Geolocators record ambient light intensities and elapsed time, from which longitude and latitude can be estimated [3, 7]. They can record data for up to a year or longer, and cover one or several annual migration cycles [8, 9]. A number of methods have been developed to estimate locations from light data (Table 1), and to filter the resulting outputs in various ways [14,15,16,17] These are mainly based on either a threshold [7, 18] or template-fit approach [19]. The templatefit method involves fitting a simplified geophysical model for various latitudes (i.e. the template) to recorded light intensities for each day at a longitude estimated in the same way as in the threshold method [21]

Objectives

Methods

Results

Discussion

Conclusion