DataSheet_1_Predicting Seabird Foraging Habitat for Conservation Planning in Atlantic Canada: Integrating Telemetry and Survey Data Across Thousands of Colonies.docx

Conservation of mobile organisms is difficult in the absence of detailed information about movement and habitat use. While the miniaturization of tracking devices has eased the collection of such information, it remains logistically and financially difficult to track a wide range of species across a large geographic scale. Predictive distribution models can be used to fill this gap by integrating both telemetry and census data to construct distribution maps and inform conservation goals and planning. We used tracking data from 520 individuals of 14 seabird species in Atlantic Canada to first compare foraging range and distance to shorelines among species across colonies, and then developed tree-based machine-learning models to predict foraging distributions for more than 5000 breeding sites distributed along more than 5000 km of shoreline. Despite large variability in foraging ranges among species, tracking data revealed clusters of species using similar foraging habitats (e.g., nearshore vs. offshore foragers), and within species, foraging range was highly colony-specific. Even with this variability, distance from the nesting colony was an important predictor of distribution for nearly all species, while distance from coastlines and bathymetry (slope and ruggedness) were additional important predictors for some species. Overall, we demonstrated the utility of tree-based machine-learning approach when modeling tracking data to predict distributions at un-sampled colonies. Although tracking and colony data have some shortcomings (e.g., fewer data for some species), where results need to be interpreted with care in some cases, applying methods for modeling breeding season distributions of seabirds allows for broader-scale conservation assessment. The modeled distributions can be used in decisions about planning for offshore recreation and commercial activities and to inform conservation planning at regional scales.

在缺乏移动生物的活动与栖息地利用详细信息的情况下，对其开展保护工作难度极大。尽管追踪设备的小型化降低了此类信息的收集难度，但在大范围地理尺度上对众多物种开展追踪，仍面临后勤与财务层面的双重难题。预测性分布模型可通过整合遥测（telemetry）与普查数据，构建物种分布图谱并为保护目标制定与规划提供依据，以此填补这一研究空白。我们利用加拿大大西洋海域14种海鸟共520只个体的追踪数据，首先对比了不同繁殖集群下各物种的觅食范围与距海岸线的距离；随后构建基于树的机器学习（tree-based machine-learning）模型，对沿5000余公里海岸线分布的5000余个繁殖位点的觅食分布进行预测。尽管不同物种的觅食范围存在较大差异，但追踪数据显示，部分物种会利用相似的觅食栖息地（如近岸觅食者与远岸觅食者）；且在同一物种内部，觅食范围高度依赖于具体繁殖集群。即便存在上述差异，对于几乎所有物种而言，距繁殖集群的距离仍是其分布的重要预测因子；而对于部分物种，距海岸线的距离以及水深地形（bathymetry，坡度与崎岖度）则是额外的重要预测因子。总体而言，我们验证了基于树的机器学习方法在利用追踪数据建模、预测未采样繁殖集群分布方面的实用性。尽管追踪与繁殖集群数据存在一定局限（例如部分物种的数据量较少），部分场景下需谨慎解读研究结果，但通过应用海鸟繁殖季分布建模方法，可开展更大尺度的保护评估工作。所构建的预测分布可用于近海娱乐与商业活动规划的决策制定，同时可为区域尺度的保护规划提供参考依据。