During 13-18 May, 2019, we censused Red-footed Boobies (Sula sula) at Kīlauea Point National Wildlife Refuge (KPNWR), Kauaʻi. To evaluate ground-based visual counts and assess variability among methods, we employed photographic techniques to count nesting Red-footed Boobies at KPNWR in three distinct sub-colony areas: Crater Hill East, Crater Hill Interior, and Crater Hill West (see associated data series report for delineation of sub-colony areas). For ground-based photos, we used a Sony a7rii camera (42 megapixel full-frame) equipped with a Sony 100-400 mm F4.5-5.6G lens to allow sufficient resolution to count nests and birds. We achieved pixel resolutions (as a function of camera resolution, focal length, and distance to birds) of 0.28-0.91 cm pixel-1 for photos used for refuge-wide population estimates. In addition, photos of the one sub-colony (Crater Hill West) were taken at two focal lengths (100 mm, 0.84-2.03 cm pixel-1; 300 mm, 0.28-0.68 cm pixel-1) to test the effect of pixel resolution on nest counts. These photo sets were taken within 5 minutes of each other to minimize variability in numbers of roosting birds at the colony. Additionally, we collected oblique aerial photographs from a helicopter flying along the length, and just offshore, of Red-footed Booby nesting areas at KPNWR during the late morning on 13 May 2019. Aerial photographs were taken with a Canon 5DSr camera (51 megapixel full frame) equipped with a Canon EF USM 135-mm telephoto lens, resulting in pixel resolutions of 0.61-1.53 cm pixel-1. Both ground-based and aerial photos were taken with overlap between individual photos in each set so that observers could later ensure locational reference and complete counting without double-counting.
For each sub-colony, we identified all nesting, roosting, and flying Red-footed Boobies in photos using the program DotDotGoose (version 1.5.1; Ersts 2020). For each photo, we evaluated image quality, geographical location of the image, the degree of overlapping areas between images, and other general observations. We used the DotDotGoose interface to individually count and label each Red-footed Booby with one of four classifications based on its appearance and behavior at the time the image was captured: Flying, Nesting, Roosting, Unknown. If a nest was not clearly visible, the body alignment of a sitting bird with horizontal posture was also used to indicate the presence of a nest. We classified Red-footed Booby adults sitting with tails down, chest out, and an extended body with no visible nest as Roosting. If a Red-footed Booby was present, but showed no definitive roosting or nesting cues, or if it was displaying cues of both roosting and nesting behaviors, we classified it as Unknown. Once we counted and categorized each individual, we summed the total number of birds in each class based on their classifications. When other bird species were observed in a photo, they were marked but we did not classify behavior. We avoided double-counting individual birds seen in more than one image by identifying areas of overlap among sequential photos. As we counted through the photo sets, we took note for reference of which of the preceding photos overlapped with the current photo being counted. To test the counter’s consistency at identifying overlaps between images and classifying behavior of birds, we randomly re-counted 10 percent of the photos from two photo sets: the 100 mm Crater Hill West photo set (1 photo recounted) and the 300 mm Crater Hill West photo set (7 photos recounted). Each photo was cross-referenced with prior photos to determine overlap, and the original classes were used to categorize birds.
The DotDotGoose software interface requires that photos being counted are organized in a single folder and creates a pnt output file referencing all photos in the folder and the locations (in pixel space) and classification of each object identified by the user. This data release maintains that structure for each unique sub-colony photo set and photographic platform (ground or aerial). The pnt file in each folder can be opened in DotDotGoose to see classification results and locations of birds identified in each photo. For example, the CHW_Ground300 folder contains all 300-mm ground-based photos used to count the Crater Hill West sub-colony, as well as the file CHW_Ground300.pnt which can be loaded in DotDotGoose. The CHW_Ground300 and CHW_Ground100 folders also contain a second pnt file (CHW_Ground300_SampleRecount.pnt and CHW_Ground100_SampleRecount.pnt, respectively), for results of random recounts of a subset of each photo set, as described above.
References:
Ersts, P.J., 2020, DotDotGoose (version 1.5.1): American Museum of Natural History, Center for Biodiversity and Conservation, accessed May 30, 2020, at https://biodiversityinformatics.amnh.org/open_source/dotdotgoose/