Evaluation of Radar and Cameras as Tools for Automating the Monitoring of Waterbirds at Industrial Sites

Conflict occurs between people and birds at industrial sites around the world, where birds can endanger human lives (e.g. airports) and where bird populations are endangered by human activities (e.g. wind farms). Mitigating these conflicts requires accurate detection of birds and measures of their abundance and distribution. At industrial sites, detection of flying birds and the deployment of deterrents are often automated through detection by avian radar. Such sites include the various oil sands mining operations in northern Alberta, where operators are required to protect migrating waterfowl from landing on potentially toxic waste-water ponds. I tested two technologies for detecting birds in this context, one for flying birds (radar), and one for birds that have landed (cameras). I tested radar to establish its accuracy for detecting flying birds, based on birds detected by paired human observers. I used X-band marine radar and tested two types of radar antennas, one parabolic and one open-array, across a range of conditions at both process-affected water ponds and freshwater ponds. I found that the two antennas failed to detect about half of all detections confirmed by visual observers, both when they were each in operation separately (open-array antenna failed to detect 43% of targets that were confirmed as birds; parabolic antenna failed to detect 56.4% of targets that were confirmed as birds) and when they were in operation together (both antennas operating simultaneously on two radars failed to detect 43% of targets that were confirmed as birds by the visual observers). My results suggest that antenna type, height of radar station, substrate around the station, and site-specific knowledge of target birds should be more explicitly addressed when marine radar is used as part of bird protection programs. A combination of radar types, antennas, and other detection methods may be needed to achieve more comprehensive bird detection strategies at industrial sites. I also tested cameras to monitor birds in the context of industrial ponds. Birds that have landed on ponds are not detectable by radar, and standardised monitoring by human observers has documented tens of thousands of birds landing annually on oil sands process-affected water ponds. Such counts provide information on bird abundance, but there is considerable variation between observers and sites. To overcome these limitations, I evaluated the potential for cameras to monitor birds on industrial water bodies. I compared counts from high-resolution panoramic photos and photos taken by conventional remote cameras to counts conducted by field observers. I also tested the success of a computer algorithm to process photos automatically. High-resolution panoramas recorded two-thirds of bird counts recorded simultaneously by field observers, for distances of approximately 500 m from survey stations. Conventional remote cameras recorded two-thirds of birds in photos clearly, but only to a distance of 100 m. Both single-frame SLR panoramas and single-frame wildlife photos failed to capture birds that dove, birds that were behind other birds, and birds with oblique aspects to the camera. The presence of these birds could be revealed by capturing bird motion with multiple photo frames in short succession (time-interval). Automated processing of time-interval photos produced a very high true negative rate (95%), suggesting that it can substantially reduce the time spent by humans to process photos. The combined application of high resolution photos taken at frequent intervals and a specialized bird detection code makes cameras a viable alternative to human observers. Understanding the distributions and abundance of migratory waterfowl in the oil sands is in the interest of hunters, naturalists and citizens across North America. Radar and cameras can both contribute to this understanding, while simultaneously improving human safety, reducing cost and inter-observer variation, and increasing the duration and frequency of monitoring.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.