U.S. Geological Survey/EROS, Sioux Falls, South Dakota 57198, USA.
Ecol Appl. 2011 Oct;21(7):2530-47. doi: 10.1890/10-1261.1.
The ability to predict amphibian breeding across landscapes is important for informing land management decisions and helping biologists better understand and remediate factors contributing to declines in amphibian populations. We built geospatial models of likely breeding habitats for each of four amphibian species that breed in Yellowstone National Park (YNP). We used field data collected in 2000-2002 from 497 sites among 16 basins and predictor variables from geospatial models produced from remotely sensed data (e.g., digital elevation model, complex topographic index, landform data, wetland probability, and vegetative cover). Except for 31 sites in one basin that were surveyed in both 2000 and 2002, all sites were surveyed once. We used polytomous regression to build statistical models for each species of amphibian from (1) field survey site data only, (2) field data combined with data from geospatial models, and (3) data from geospatial models only. Based on measures of receiver operating characteristic (ROC) scores, models of the second type best explained likely breeding habitat because they contained the most information (ROC values ranged from 0.70 to 0.88). However, models of the third type could be applied to the entire YNP landscape and produced maps that could be verified with reserve field data. Accuracy rates for models built for single years were highly variable, ranging from 0.30 to 0.78. Accuracy rates for models built with data combined from multiple years were higher and less variable, ranging from 0.60 to 0.80. Combining results from the geospatial multiyear models yielded maps of "core" breeding areas (areas with high probability values for all three years) surrounded by areas that scored high for only one or two years, providing an estimate of variability among years. Such information can highlight landscape options for amphibian conservation. For example, our models identify alternative areas that could be protected for each species, including 6828-10 764 ha for tiger salamanders, 971-3017 ha for western toads, 4732-16 696 ha for boreal chorus frogs, and 4940-19 690 ha for Columbia spotted frogs.
预测景观中两栖动物繁殖的能力对于告知土地管理决策和帮助生物学家更好地了解和修复导致两栖动物种群减少的因素非常重要。我们为在黄石国家公园(YNP)繁殖的四种两栖动物中的每一种构建了可能的繁殖栖息地的地理空间模型。我们使用了 2000-2002 年从 16 个流域的 497 个地点收集的实地数据和从遥感数据生成的地理空间模型的预测变量(例如,数字高程模型、复杂地形指数、地形数据、湿地概率和植被覆盖)。除了一个流域的 31 个在 2000 年和 2002 年都进行了调查的地点外,所有地点都只进行了一次调查。我们使用多类回归来为每种两栖动物从以下三种类型的统计模型中建立模型:(1)仅使用实地调查地点数据,(2)实地数据与地理空间模型数据相结合,(3)仅使用地理空间模型数据。基于接收者操作特征(ROC)分数的衡量标准,第二种类型的模型最好地解释了可能的繁殖栖息地,因为它们包含最多的信息(ROC 值范围为 0.70 到 0.88)。然而,第三种类型的模型可以应用于整个 YNP 景观,并生成可以与保护区实地数据进行验证的地图。针对单一年份构建的模型的准确率变化很大,范围从 0.30 到 0.78。使用来自多年的数据构建的模型的准确率更高且变化更小,范围从 0.60 到 0.80。结合地理空间多年模型的结果,得到了“核心”繁殖区域的地图(所有三年的概率值都很高),周围是仅在一到两年得分较高的区域,这提供了各年之间变异性的估计。这种信息可以突出显示两栖动物保护的景观选择。例如,我们的模型为每种物种确定了替代的保护区,包括老虎蝾螈的 6828-10764 公顷、西部蟾蜍的 971-3017 公顷、北方合唱蛙的 4732-16696 公顷和哥伦比亚斑点蛙的 4940-19690 公顷。