Three‐dimensional Bayesian passive acoustic tracking of walruses in the Chukchi Sea.

Abstract

A Bayesian travel‐time inversion method was developed for 3‐D localization and tracking using multipath arrival times of knock‐type sounds produced by walruses. The method accounts for data errors and uncertainties in environmental and geometric parameters. Data were collected on three seafloor sound recorders, arranged in an equilateral triangle with approximately 500‐m sides, deployed in 30 m of water in the Chukchi Sea near the Hanna Shoal, west of Barrow, Alaska. Pacific Walrus (Odobenus rosmarus divergens) calls were recorded by these recorders for 2.5 months in the summer of 2009. A regularized, linearized inversion algorithm was used to estimate 3‐D tracks and track uncertainties for these calling walruses. Regularization incorporates prior information (expected values and uncertainty estimates) for the recorder locations and acoustic environment (water depth and sound speed) as well as preferred track characteristics. Inverting for the smoothest (simplest) track consistent with the acoustic data and prior information mitigates the risk of over‐interpreting track structure due to data errors and environmental and geometric uncertainties. Hence, this approach provides more plausible sequences of source positions than a non‐regularized solution. Tracks for walrus dives from the 2009 dataset will be presented.