Counting breeding gulls with unmanned aerial vehicles: camera quality and flying height affects precision of a semi-automatic counting method

Abstract

The use of Unmanned Aerial Vehicles (UAVs) to monitor large colonies of seabirds avoids challenges associated with conventional methods, but manual image processing is expensive. Development of semi-automated analytical methods rely on high image spatial resolution, which requires a trade-off between securing low area coverage and high spatial resolution flying at low altitude vs high area coverage but low spatial resolution flying at higher altitudes. Increasing individual bird detection probabilities requires maximizing contrast between target and background, which can be enhanced using thermal sensors. We applied a semi-automatic analytical method to multispectral UAV derived imagery to count a mixed breeding colony of Herring Gulls (Larus argentatus) and Lesser Blackbacked Gulls (L. fuscus). We trained the computer to detect different image classes by their spectral signature in several orthomosaics obtained from UAV flights at different altitudes using different cameras. Highest agreement with the manual counts was achieved by low flying (20 m) using the highest camera resolution (97.7 ± 1.1% for the Herring Gulls, omission error 2.6%, commission error 0.5%; 94.8 ± 1.8% for Lesser Black-backed Gulls, omission error 6.5%, commission error 1.6%). Method precision varied between trials, confirming the importance of low altitude flying with high quality cameras, and a 40% reduction in detection noise from adding a thermal sensor.