MEPS 365:35-44 (2008)  -  DOI: https://doi.org/10.3354/meps07497

ABSTRACT:  Pop-up satellite archival tags (PSATs) and data loggers (archival tags) have become key tools for tracking movements of marine animals, but uncertainties in location estimates can range from tens to hundreds of kilometers. Sea surface temperature (SST) may be used in models to improve light-based geolocation by comparing SSTs measured in tags to those measured by satellites (e.g. with a Kalman filter). Daily SST measurements are retrieved from the data recorded in the tag by averaging the near surface temperatures. Raw satellite SST data are represented as points in a grid, but measurement noise and areas missing due to cloud cover can produce an uneven SST field that may not correspond well with the local average of SST measured by the tags. A smoothed satellite SST field is used to compensate for these problems. We used 2 crossvalidation schemes to analyze what degree of smoothing produces the optimal match with the SST from the tag. Simulations based on data returned by PSATs deployed on Atlantic bluefin tuna Thunnus thynnus are used as a test case. We demonstrate that the optimal scale of smoothing, which affects overall variance in any type of geolocation estimation, is influenced by the scale of diffusive animal movement and that treatment of satellite SST in a geolocation framework should be carefully considered. The developed crossvalidation scheme provides an objective method for choosing the optimal smoothing scale and allows for better control of the overall geolocation process.

KEY WORDS: Satellite SST · Kalman filter · Local polynomial regression · Animal movement · Bluefin tuna · Thunnus thynnus