The habitat resources are structured across different spatial scales in the environment, and thus animals perceive and select habitat resources at different spatial scales. Failure to adopt the scale-dependent framework in species habitat relationships may lead to biased inferences. Multi-scale species distribution models (SDMs) can thus improve the predictive ability as compared to single-scale approaches. This study outlines the importance of multi-scale modeling in assessing the species habitat relationships and may provide a methodological framework using a robust algorithm to model and predict habitat suitability maps (HSMs) for similar multi-species and multi-scale studies. We used a supervised machine learning algorithm, random forest (RF), to assess the habitat relationships of Asiatic wildcat (Felis lybica ornata), jungle cat (Felis chaus), Indian fox (Vulpes bengalensis), and golden-jackal (Canis aureus) at ten spatial scales (500–5000 m) in human-dominated landscapes. We calculated out-of-bag (OOB) error rates of each predictor variable across ten scales to select the most influential spatial scale variables. The scale optimization (OOB rates) indicated that model performance was associated with variables at multiple spatial scales. The species occurrence tended to be related strongest to predictor variables at broader scales (5000 m). Multivariate RF models indicated landscape composition to be strong predictors of the Asiatic wildcat, jungle cat, and Indian fox occurrences. At the same time, topographic and climatic variables were the most important predictors determining the golden jackal distribution. Our models predicted range expansion in all four species under future climatic scenarios. Our results highlight the importance of using multiscale distribution models when predicting the distribution and species habitat relationships. The wide adaptability of meso-carnivores allows them to persist in human-dominated regions and may even thrive in disturbed habitats. These meso-carnivores are among the few species that may benefit from climate change.

中文翻译：

---

使用随机森林算法的多尺度生境选择和气候变化对四个同伴中生食肉动物分布的影响

栖息地资源在环境中跨不同空间尺度构成，因此动物在不同空间尺度上感知和选择栖息地资源。未能在物种栖息地关系中采用与规模有关的框架可能会导致推论有偏差。因此，与单尺度方法相比，多尺度物种分布模型（SDM）可以提高预测能力。这项研究概述了多尺度建模在评估物种栖息地关系中的重要性，并且可以提供一种使用健壮算法为相似的多物种和多尺度研究建模和预测栖息地适宜性图（HSM）的方法框架。我们使用有监督的机器学习算法随机森林（RF）来评估亚洲野猫（Felis lybica ornata）的栖息地关系，在人为主导的景观中，在十个空间尺度（500–5000 m）上，有丛林猫（Felis chaus），印度狐狸（Vulpes bengalensis）和金狐（犬金黄色）。我们计算了十个尺度上每个预测变量的袋外（OOB）错误率，以选择最具影响力的空间尺度变量。尺度优化（OOB速率）表明模型性能与多个空间尺度上的变量相关。在更广泛的尺度（5000 m）下，物种的出现往往与预测变量最相关。多元RF模型表明，景观成分是亚洲野猫，丛林猫和印度狐狸发生的有力预测指标。同时，地形和气候变量是决定黄金jack狼分布的最重要预测因子。我们的模型预测了未来气候情景下所有四个物种的范围扩展。我们的结果突出了在预测分布和物种栖息地关系时使用多尺度分布模型的重要性。中度食肉动物的广泛适应性使它们能够在人类占主导的区域中生存，甚至在受干扰的栖息地中都能繁衍生息。这些中食肉食动物是少数可能受益于气候变化的物种。