State–space model for light-based tracking of marine animals

Abstract

A coherent model is presented to estimate the most probable track of geographic positions directly from a series of light measurements. The model estimates two geographic positions per day, without reducing the daily light data to two threshold crossing times, its covariance structure is designed to handle high correlations due to for instance local weather conditions, and it can estimate the yearly pattern in latitudinal precision by propagating the data uncertainties through the geolocation process. The model is applied to one mooring study, one GPS drifter buoy study, and numerous simulated cases. The simulations are performed with realistic assumptions about the relationship between solar altitude and light and with realistic uncertainty parameters (all taken from real data). The simulations showed that all model parameters were identifiable, and that all tracks could be reconstructed within 1° or 2° latitude and 0.5° or 1° longitude. The mooring and drifter buoy data showed that the tracks could be reliably estimated, even in cases where the other methods had completely failed.