Data from: Potential breeding distributions of U.S. birds predicted with both short-term variability and long-term average climate data

Climate conditions, such as temperature or precipitation averaged over several decades strongly affect species distributions, as evidenced by experimental results and a plethora of models demonstrating statistical relations between species occurrences and long-term climate averages. However, long-term averages can conceal climate changes that have occurred in recent decades and may not capture actual species occurrence well because the distributions of species, especially at the edges of their range, are typically dynamic and may respond strongly to short-term climate variability. Our goal here was to test whether bird occurrence models can be predicted by either covariates based on short-term climate variability or on long-term climate averages. We parameterized species distribution models (SDMs) based on either short-term variability or long-term average climate covariates for 320 bird species in the conterminous U.S., and tested whether any life-history trait-based guilds were particularly sensitive to short-term conditions. Models including short-term climate variability performed well based on their cross-validated AUC score (0.85), as did models based on long-term climate averages (0.84). Similarly, both models performed well compared to independent presence/absence data from the North American Breeding Bird Survey (independent AUC of 0.89 and 0.90, respectively). However, models based on short-term variability covariates more accurately classified true absences for most species (73% of true absences classified within the lowest quarter of environmental suitability versus 68%). In addition, they have the advantage that they can reveal the dynamic relationship between species and their environment because they capture the spatial fluctuations of species potential breeding distributions. With this information we can identify which species and guilds are sensitive to climate variability, identify sites of high conservation value where climate variability is low, and assess how species' potential distributions may have already shifted due recent climate change. However, long-term climate averages require less data and processing time and may be more readily available for some areas of interest. Where data on short-term climate variability are not available, long-term climate information is a sufficient predictor of species distributions in many cases. However, short-term climate variability data may provide information not captured with long-term climate data for use in SDMs.数十年来平均的气候条件（如气温、降水）对物种分布具有显著影响，这一点已通过实验证据与大量模型得到证实：诸多模型均揭示了物种出现记录与长期气候平均值间的统计关联。然而，长期气候平均值会掩盖近数十年间发生的气候变化，且可能无法很好地反映物种的实际出现情况——这是因为物种分布（尤其是分布范围边缘的种群）通常处于动态变化中，且会对短期气候变异性产生强烈响应。本研究旨在验证：基于短期气候变异性的协变量，或是基于长期气候平均值的协变量，能否用于构建鸟类出现预测模型。我们针对美国本土的320种鸟类，分别基于短期气候变异性协变量与长期气候平均值协变量构建了物种分布模型（Species Distribution Models, SDMs），并检验是否存在基于生活史特征的功能群对短期气候条件尤为敏感。纳入短期气候变异性的模型，其交叉验证AUC（受试者工作特征曲线下面积，Area Under the Receiver Operating Characteristic Curve）得分为0.85，表现优异；基于长期气候平均值的模型得分亦达0.84，表现同样出色。类似地，相较于北美繁殖鸟类调查（North American Breeding Bird Survey, BBS）的独立出现/不出现数据，两类模型均表现优异：其独立验证AUC得分分别为0.89与0.90。不过，基于短期气候变异性协变量的模型，对多数物种的真实无记录点的分类准确率更高：73%的真实无记录点被归类至环境适宜性最低的四分位区间，而基于长期平均值的模型这一比例为68%。此外，短期气候变异性模型还具备一项优势：其能够捕捉物种潜在繁殖分布的空间波动，进而揭示物种与环境之间的动态关联。借助这类信息，我们可以识别出对气候变异性敏感的物种与功能群，定位气候变异性较低的高保护价值区域，并评估物种潜在分布因近期气候变化已发生的偏移情况。不过，长期气候平均值所需的数据与处理时长更少，且在部分目标区域中更易获取。在无法获取短期气候变异性数据的场景下，长期气候信息在多数情况下足以作为物种分布的预测因子。不过，针对物种分布模型（SDMs）的构建而言，短期气候变异性数据能够提供长期气候数据无法涵盖的额外信息。