North Pacific Wildlife Consulting, LLC, Seattle, WA, United States of America.
Laboratory of Behavior and Behavioral Ecology, Severtsov Institute of Ecology and Evolution Russian Academy of Sciences, Moscow, Russia.
PLoS One. 2024 Jul 16;19(7):e0307416. doi: 10.1371/journal.pone.0307416. eCollection 2024.
This study presents a semi-automated approach utilizing unoccupied aerial vehicle (UAV) surveys to accurately estimate the abundance of Pacific walruses at large coastal haulouts in Chukotka, Russia. Seven major haulout sites were surveyed during the summers and falls of 2017-2019. Walrus counts were performed using three distinct methods: traditional visual land-based counts, complete head counts utilizing georeferenced UAV imagery, and counting walruses within model polygons within the haulout outline and employing various extrapolation techniques to predict walrus abundance across the haulout area. The results indicated that traditional visual counts neither yielded consistent results nor allowed for uncertainty estimation, unlike the site- and date-specific direct extrapolation method and the non-specific linear regression model. These latter methods consistently provided estimates, on average, within 5% of the "true" abundance determined through complete photo-based head counts. Beside yielding accurate estimates, these semi-automated methods significantly reduced counting time by at least 63%, in contrast to complete head counts. The non-specific model, which allowed the estimation of walrus abundance based on the type of the terrain and the haulout area was less accurate compared with site and date specific estimates, but provided a tool to estimate abundance when no field visits are conducted, e.g., by using high-resolution satellite imagery to measure haulout area. This model revealed that the haulouts located on flat sandy beaches exhibited mean walrus densities approximately 30.5% times higher than those on rocky shores surrounded by cliffs: 0.879 (SD = 0.1302) and 0.648 (SD = 0.1753) walrus per m2 correspondingly. The estimated daily walrus abundance at major Chukotkan haulouts in 2017-2019 ranged between 15 and 94,660 (mean = 10,397, SD = 14,477) walruses with the maximum seasonal abundances reported at Cape Serdtse-Kamen as 94,960 on 10-Oct-2017, 26,850 on 10-Oct-2018, and 87,595 on 10-Oct-2019.
本研究提出了一种利用无人机(UAV)调查来准确估算俄罗斯楚科奇地区大型沿海聚居区太平洋海象数量的半自动化方法。在 2017-2019 年的夏季和秋季,对七个主要聚居地进行了调查。使用三种不同的方法进行海象计数:传统的基于陆地的视觉计数、利用地理参考 UAV 图像进行的完整头部计数、以及在聚居区轮廓内的模型多边形内计数海象,并采用各种外推技术来预测聚居区的海象数量。结果表明,传统的视觉计数既不能产生一致的结果,也不能进行不确定性估计,而基于地点和日期的直接外推方法和非特定线性回归模型则可以。后两种方法平均而言,在 5%的范围内提供了与通过完整基于照片的头部计数确定的“真实”丰度相符的估计值。除了提供准确的估计值外,这些半自动化方法还将计数时间至少减少了 63%,而与完整的头部计数相比。非特定模型允许根据地形类型和聚居区面积来估算海象数量,与特定地点和日期的估计相比,其准确性较低,但提供了一种工具,可以在没有实地考察时估算数量,例如,使用高分辨率卫星图像来测量聚居区面积。该模型表明,位于平坦沙滩上的聚居区的海象密度平均比周围有悬崖的多岩石海岸高出约 30.5%:分别为 0.879(SD=0.1302)和 0.648(SD=0.1753)头/平方米。2017-2019 年在楚科奇主要聚居区的估计每日海象数量在 15 至 94,660 头之间(平均值=10,397,SD=14,477),报告的最大季节性数量是 2017 年 10 月 10 日在 Cape Serdtse-Kamen 地区的 94,960 头、2018 年 10 月 10 日的 26,850 头和 2019 年 10 月 10 日的 87,595 头。