A simple new algorithm to filter marine mammal Argos locations

TLDR

Satellite telemetry using the Argos system has tracked many marine mammals, but their aquatic behavior generates many low‑accuracy locations that are often filtered by unrealistic swimming speeds, inadvertently discarding high‑speed, good‑quality points taken close together. The authors propose an alternative filter that incorporates swimming speed, inter‑location distance, and turning angle to better discriminate location quality. They evaluated this filter on 67 tracks from nine marine mammal species, including seals, walruses, belugas, and narwhals. Compared to a speed‑only filter, the new method removed similar amounts of low‑quality LC B and A points while retaining 4.1 % vs.

Abstract

Abstract During recent decades satellite telemetry using the Argos system has been used extensively to track many species of marine mammals. However, the aquatic behavior of most of these species results in a high number of locations with low or unknown accuracy. Argos data are often filtered to reduce the noise produced by these locations, typically by removing data points requiring unrealistic swimming speeds. Unfortunately, this method excludes a considerable number of good‐quality locations that have high traveling speeds that are the result of two locations being taken very close in time. We present an alternative algorithm, based on swimming speed, distance between successive locations, and turning angles. This new filter was tested on 67 tracks from nine different marine mammal species: ringed, bearded, gray, harbor, southern elephant, and Antarctic fur seals, walruses, belugas, and narwhals. The algorithm removed similar percentages of low‐quality locations (Argos location classes [LC] B and A) compared to a filter based solely on swimming speed, but preserved significantly higher percentages of good‐quality positions (mean ± SE% of locations removed was 4.1 ± 0.8% vs. 12.6 ± 1.2% for LC 3; 6.8 ± 0.6% vs. 15.7 ± 0.9% for LC 2; and 11.4 ± 0.7% vs. 21.0 ± 0.9% for LC 1). The new filter was also more effective at removing unlikely, conspicuous deviations from the track's path, resulting in fewer locations being registered on land and a significant reduction in home range size, when using the Minimum Convex Polygon method, which is sensitive to outliers.