Climate Adaptation Flagship and Ecosystem Sciences, Commonwealth Scientific and Industrial Research Organisation, Townsville, Queensland, Australia.
PLoS One. 2010 Oct 22;5(10):e13569. doi: 10.1371/journal.pone.0013569.
Accurate predictions of species distributions are essential for climate change impact assessments. However the standard practice of using long-term climate averages to train species distribution models might mute important temporal patterns of species distribution. The benefit of using temporally explicit weather and distribution data has not been assessed. We hypothesized that short-term weather associated with the time a species was recorded should be superior to long-term climate measures for predicting distributions of mobile species.
We tested our hypothesis by generating distribution models for 157 bird species found in Australian tropical savannas (ATS) using modelling algorithm Maxent. The variable weather of the ATS supports a bird assemblage with variable movement patterns and a high incidence of nomadism. We developed "weather" models by relating climatic variables (mean temperature, rainfall, rainfall seasonality and temperature seasonality) from the three month, six month and one year period preceding each bird record over a 58 year period (1950-2008). These weather models were compared against models built using long-term (30 year) averages of the same climatic variables.
Weather models consistently achieved higher model scores than climate models, particularly for wide-ranging, nomadic and desert species. Climate models predicted larger range areas for species, whereas weather models quantified fluctuations in habitat suitability across months, seasons and years. Models based on long-term climate averages over-estimate availability of suitable habitat and species' climatic tolerances, masking species potential vulnerability to climate change. Our results demonstrate that dynamic approaches to distribution modelling, such as incorporating organism-appropriate temporal scales, improves understanding of species distributions.
准确预测物种分布对于气候变化影响评估至关重要。然而，使用长期气候平均值来训练物种分布模型的标准做法可能会掩盖物种分布的重要时间模式。使用时间明确的天气和分布数据的好处尚未得到评估。我们假设与记录物种时间相关的短期天气对于预测移动物种的分布应该优于长期气候措施。
我们通过使用最大熵算法 Maxent 为澳大利亚热带草原 (ATS) 中发现的 157 种鸟类生成分布模型来检验我们的假设。ATS 的多变天气支持具有可变运动模式和高游牧发生率的鸟类组合。我们通过将气候变量（平均温度、降雨量、降雨量季节性和温度季节性）与每个鸟类记录前三个月、六个月和一年的时间段相关联，开发了“天气”模型，这是在 58 年期间（1950-2008 年）进行的。这些天气模型与使用相同气候变量的 30 年平均值构建的模型进行了比较。
天气模型始终比气候模型获得更高的模型得分，特别是对于广泛分布、游牧和沙漠物种。气候模型预测了物种更大的范围面积，而天气模型则量化了栖息地适宜性在月份、季节和年份之间的波动。基于长期气候平均值的模型高估了适宜栖息地的可用性和物种的气候耐受性，掩盖了物种对气候变化的潜在脆弱性。我们的结果表明，采用动态方法进行分布建模，例如纳入适合生物体的时间尺度，可以提高对物种分布的理解。
