Semi‐Automated Detection of Right Whales ( Eubalaena spp. ) in Very High‐Resolution Satellite Imagery

Abstract

Space‐based detection of whales is proliferating because it shows promise as a monitoring tool, yet tests of its application across taxa and environments are rare. The objective of this study was to develop an end‐to‐end, semi‐automated procedure for detecting two right whale species ( Eubalaena glacialis and E. australis ) in satellite imagery. We collected 35 new and archived images covering 5200 km 2 of ocean in three habitat types from various areas around the globe, then constructed and tested an EfficientNet model classifier and a Faster Region‐Based Convolutional Neural Network detection model to process the imagery. The model was trained using 428 large whales manually detected in 18 images. The test set included 119 large whales found in 5 images collected within right whale habitats that were not included in the training set. The model produced strong recall at detecting whales (> 0.73). Precision was lower in northern temperate E. glacialis habitat (0.11–0.20) than in coastal tropical E. australis habitat (0.84–0.95). Satellite tasking, detection, and species‐level identification efforts were more successful in coastal than in open‐ocean habitats. The end‐to‐end procedure was effective at detecting large whales in satellite imagery. Moving forward, responsive tasking options and the high price of imagery are challenges to scaling up this procedure for operational monitoring.

Bibliographic Information