Improving light and temperature based geolocation by unscented Kalman filtering

Tracking marine animals with electronic tags has become an indispensable tool in understanding biology in relation to movement. Combining light based geolocation estimates with an underlying movement model has proved helpful in reconstructing the most probable track of tagged animals. These tracks can be further improved by including the tag measured sea-surface temperature and matching it to external sea-surface temperature (SST) data. The current methodology for doing this in a state-space model requires that external sea-surface temperature be smoothed before it is used in the model, and further that its gradient field is pre-calculated. This two-step approach has a number of technical drawbacks, and the final statistical inference about the most probable track is consequently less convincing. This paper presents a new methodology (refer to as UKFSST) where all steps, including the SST smoothing, are handled within one coherent model. An additional benefit is that even the degree of smoothing, which was previously pre-determined and fixed, can now be optimally selected. UKFSST offers better handling of non-linearities in Kalman filter, and provides a statistically sound model for geolocation applications, as opposed to ad hoc SST matching approaches.