Demonstration of a Likelihood Framework for light-based geolocation

For many marine fish, light-based geolocation using twilight data from archival or pop-up tags is still the only way to estimate location. It is an inherently imprecise exercise because of short-term fluctuations in incident... [ view full abstract ]

For many marine fish, light-based geolocation using twilight data from archival or pop-up tags is still the only way to estimate location. It is an inherently imprecise exercise because of short-term fluctuations in incident light, and the resulting uncertainty in location is big enough to need accounting for when reconstructing tracks and making inferences about habitat use. The key ingredient for any reliable geolocation-based statistical model of movement or habitat selection, whether Bayesian or “classical”, is a valid likelihood function for each set of twilight data (i.e. a way to compute the relative probability of the observed data for any assumed location). However, the complex autocorrelations and non-Gaussian errors make it very difficult to devise and compute such a likelihood directly. Instead, we use data from moored tags to obtain an experimentally derived likelihood with correct confidence interval properties. This likelihood can then be computed directly from real tag data for use in state-space models which can construct a movement track with appropriate uncertainty. These methods are built in to a cohesive package as part of a wider program of electronic tagging related tools and to demonstrate the general nature of the system, examples of geolocation using light data obtained from a variety of species, some with corresponding GPS tracks are presented.