Using remote sensing to detect whale strandings in remote areas: The case of sei whales mass mortality in Chilean Patagonia.
Peter T FretwellJennifer A JacksonMauricio J Ulloa EncinaVreni HäussermannMaria J Perez AlvarezCarlos OlavarríaCarolina S GutsteinPublished in: PloS one (2019)
We test the ability of Very High Resolution satellite (VHR) imagery to detect stranded whales using both manual and automated methods. We use the 2015 mass mortality event in the Gulf of Penas locality, central Patagonia, Chile, as an initial case study. This event was the largest known mass mortality of baleen whales, with at least 343 whales, mainly sei whales (Balaenoptera borealis), documented as stranding. However, even with such a large number of whales, due to the remote location of the gulf the strandings went unrecorded for several weeks. Aerial and boat surveys of the area were conducted two to four months after the mortality event. In this study we use 50cm resolution WorldView2 imagery to identify and count strandings from two archival images acquired just after the stranding event and two months before the aerial and ground surveys, and to test manual and automated methods of detecting stranded whales. Our findings show that whales are easily detected manually in the images but due to the heterogeneous colouration of decomposing whales, spectral indices are unsuitable for automatic detection. Our satellite counts suggest that, at the time the satellite images were taken, more whales were stranded than recorded in the aerial survey, possibly due to the non-comprehensive coverage of the aerial survey or movement of the carcases between survey acquisition. With even higher resolution imagery now available, satellite imagery may be a cost effective alternative to aerial surveys for future assessment of the extent of mass whale stranding events, especially in remote and inaccessible areas.
Keyphrases
- deep learning
- cross sectional
- cardiovascular events
- high resolution
- machine learning
- risk factors
- optical coherence tomography
- convolutional neural network
- coronary artery disease
- high throughput
- magnetic resonance imaging
- healthcare
- magnetic resonance
- mass spectrometry
- current status
- single molecule
- nucleic acid
- dual energy