Skip to content

Colour drone images gave better harbour seal length estimates than thermal images

Research: Ecology, Other research

Drone images can capture many seals at a haulout, but turning them into body-length estimates is not straightforward. In a preprint that has not yet undergone peer review, researchers tested a system using surveys of harbour seals (Phoca vitulina) in British Columbia in 2023 and 2024. The system traced seals in aerial mosaics—maps assembled from overlapping photographs—and estimated each animal’s length.

Using manual length annotations of 2,483 seals on the same aerial mosaics as the reference—not physical measurements of the animals—the system’s estimates from standard colour (RGB) images were shorter by an average of 0.6 cm. That near-zero average bias did not mean every estimate was close: the root mean square error (RMSE) was 12.5 cm, a measure that gives extra weight to larger individual errors. Thermal images performed worse, underestimating length by an average of 19.9 cm, with an RMSE of 24.1 cm. Under the conditions tested, the researchers concluded that thermal images were unsuitable for this automated image-based length-measurement approach.

With RGB images, the system assigned seals to the correct size class—pup, juvenile, or subadult/adult—78.7% of the time. Accuracy rose to 90.9% when seals near class boundaries were excluded. These are size-based categories, not confirmed ages; small measurement errors can shift a seal across a boundary. The accuracy of the aerial mosaics also varied between survey conditions, with geometric errors ranging from 1 to 18 cm. That matters especially for seals close to a size-class boundary, where even a small length error may change their classification.

The results suggest that colour drone surveys could help researchers monitor the size composition of seals at haulouts, provided image type and survey conditions are taken into account. The study tested a measurement tool, rather than tracking population change.

Sources and image credit

  1. Automated detection, morphometric measurement, and size-class determination of harbor seals (Phoca vitulina) from UAV orthomosaics: a cascaded deep-learning pipeline with stratified validation across platforms, sensors, altitudes, and substratesPrimærkilden er et preprint. Publiseringsdato og artikkelens sammendrag ble kontrollert; fullteksten var ikke tilgjengelig under verifiseringen.
  2. Preprint Match — authors' bioRxiv abstract reprinted with posting dateThe original bioRxiv server currently blocks this reader; independent preprint index reprints the authors' original abstract word for word and identifies posting date 8 Oct 2026, with RGB bias −0.6cm RMSE 12.5cm, thermal bias −19.9cm RMSE 24.1cm and 2,483 annotated seals. Preprint, not peer reviewed.

Image: Brocken Inaglory / Wikimedia Commons, CC BY-SA 3.0. Representative photograph. · CC BY-SA 3.0 · Source

About this content: This story was produced with AI assistance within an editorial workflow developed by Wildlife Vagabond. Editorial responsibility remains with Wildlife Vagabond.How AI is used

Ecology

SPECIES IN THIS STORY

Species in this story

Harbour Seal Phoca vitulina Explore species

Independent research and conservation news archive

Did you find this information useful?

Wildlife Vagabond is independently built and maintained. Voluntary support helps cover source verification, hosting and continued work on research and conservation news.

Support Wildlife Vagabond

The news archive will remain freely available.