The influence of vegetative changes due to livestock grazing on grassland birds is well-recognized because these birds are heavily influenced by vegetative structure. Traditionally, species distribution models (SDMs) use direct variables, resources that the animal consumes or requires to persist in an area (e.g., water) to define and project a species’ niche and distribution. Indirect variables, which are features the animal does not consume or require for persistence but with which it may still interact, are often excluded. Our objective was to improve the traditional SDMs projecting the distribution of three summer resident South Texas grassland birds (Northern Bobwhite Colinus virginianus, Eastern Meadowlark Sturnella magna, and Cassin’s Sparrow Peucaea cassinii) by incorporating livestock grazing pressure, an indirect variable, into five SDM algorithms: BioClim, generalized linear model, MaxEnt, boosted regression tree, and random forest. We collected data from the Coloraditas Grazing Research and Demonstration Area (CGRDA), a 7684-ha area located on the San Antonio Viejo Ranch (SAV) in South Texas. We used several relevant environmental characteristics to build SDMs and compared model performance (AUC and TSS) with and without grazing pressure as an indirect variable. Machine learning models (MaxEnt and random forest) had the highest predictive performance for all species, with random forest being the most consistent for each analysis. BioClim and generalized linear model remained constant or only marginally improved with the addition of the grazing pressure. Our findings suggest that model selection for SDM should include consideration of species prevalence, and machine-learning algorithms should be preferred when the target species is of low or unknown prevalence. Further, livestock grazing has measurable influence on grassland bird species’ distributions and should be included in SDMs as an indirect variable in addition to associated vegetative changes.

中文翻译：

---

模拟牲畜放牧压力对草地鸟类分布的影响

众所周知，由于牲畜放牧而引起的营养变化对草原鸟类的影响，因为这些鸟类受营养结构的影响很大。传统上，物种分布模型（SDM）使用直接变量，动物在某个区域（例如水）中消耗或需要维持的资源来定义和预测物种的生态位和分布。间接变量是动物不消耗或不需要持久性但仍可与之交互的特征，通常会被排除。我们的目标是通过结合牲畜放牧压力（间接变量）来改善传统的SDM，以预测三只夏季居住在南德克萨斯州的草地鸟类（北鲍勃·怀特·科林努斯·维吉尼亚努斯，东梅多拉克·斯特恩拉·玛格纳和卡辛的麻雀Peucaea cassinii）的分布。分为五种SDM算法：BioClim，广义线性模型，MaxEnt，增强回归树和随机森林。我们收集了科罗拉多州放牧研究和示范区（CGRDA）的数据，该区位于德克萨斯州南部的圣安东尼奥维耶霍牧场（SAV），占地7684公顷。我们使用了几个相关的环境特征来构建SDM，并比较了是否有放牧压力作为间接变量的模型性能（AUC和TSS）。机器学习模型（MaxEnt和随机森林）对于所有物种都具有最高的预测性能，而随机森林对于每种分析而言都是最一致的。随着掠食压力的增加，BioClim和广义线性模型保持恒定或仅略有改善。我们的发现表明，SDM的模型选择应考虑物种流行程度，当目标物种的流行率较低或未知时，应首选机器学习算法。此外，放牧对草地鸟类的分布具有可衡量的影响，除相关的营养变化外，还应作为间接变量包括在SDM中。