Identifying optimal wavelengths to maximise the detection rates of marine fauna from aerial surveys

• Published in: Biological conservation Vol. 257; p. 109102
• Main Authors: Colefax, Andrew P., Kelaher, Brendan P., Walsh, Andrew J., Purcell, Cormac R., Pagendam, Dan E., Cagnazzi, Daniele, Butcher, Paul A.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2021
• Subjects: Availability perception bias; Drone hyperspectral sensing; GAMM generalized additive model; Marine aerial survey; Marine megafauna detection; UAV fauna survey; GAMM generalized additive model; Marine megafauna detection; Drone hyperspectral sensing; Availability perception bias; Marine aerial survey; UAV fauna survey
• ISSN: 0006-3207; 1873-2917
• DOI: 10.1016/j.biocon.2021.109102

Aerial surveys are increasingly used for assessing the presence and abundance of many large faunal species. However, detection rates often suffer significant sightability errors. Using sensors that capture thermal infrared can sometimes improve the detectability of fauna by increasing the contrast of target animals against their surrounds. However, there are several situations where this does not work well. Submerged fauna in marine environments are one example, and where the detection error can be substantial. To identify what wavelength selection might enhance detectability rates beyond standard cameras, we used a drone-based hyperspectral sensor to measure the reflectance of 64 animals and their corresponding surrounds. Fauna included white sharks (Carcharodon carcharias), bottlenose dolphins (Tursiops aduncus), and various other species of ray (Rhinoptera neglecta, Aetobatus narinari, Rhynchobatus australiae) and fish (Seriola lalandi). Data were analysed using Generalized Additive Mixed-Models (GAMM) to compare the difference in spectral reflectance between fauna and their surrounds. The GAMMs showed little difference between submerged fauna groups, but indicated that a band of wavelengths between 514 and 554 nm provided the greatest contrast between fauna and their surrounding backgrounds. This was also consistent across the range of environmental conditions. We contend that restricting the spectral input to a standard high-resolution camera, or focusing on the channels that accentuate the green colours in post-processing, will likely result in increased accuracy in the detection of submerged fauna. This approach may also be applied to fauna surveys in other landscapes, particularly where the use of thermal sensors is not appropriate.