Data from: Species distribution models of an endangered rodent offer conflicting measures of habitat quality at multiple scales

1. The high cost of directly measuring habitat quality has led ecologists to test alternate methods for estimating and predicting this critically important ecological variable. In particular, it is frequently assumed but rarely tested that models of habitat suitability (“species distribution models”, SDMs) may provide useful indices of habitat quality, either from an individual animal or manager’s perspective. Critically, SDMs are increasingly used to estimate species’ ranges, with an implicit assumption that areas of high suitability will result in higher probability of persistence. This assumption underlies efforts to use SDMs to design protected areas, assess the status of cryptic species, or manage responses to climate change. Recent tests of this relationship have provided mixed results, suggesting SDMs may predict abundance but not other measures of high quality habitat (e.g., survival, persistence). 2. In this study, we created a suite of SDMs for the endangered giant kangaroo rat Dipodomys ingens at three distinct scales using the machine-learning method Maxent. We compared these models with three measures of habitat quality: survival, abundance, and body condition. 3. SDMs were not correlated with survival, while models at all scales were positively correlated with abundance. Finer-scale models were more closely correlated with abundance than the largest scale. Body condition was not correlated with habitat suitability at any scale. The inability of models to predict survival may be due to a lack of information in environmental covariates; unmeasured community processes or stochastic events; or the inadequacy of using models that predict species presence to also predict demography. Synthesis and applications: SDMs, especially fine scale ones, may be useful for longer-term management goals, such as identifying high quality habitat for protection. However, short-term management decisions should be based only on models that use covariates appropriate for the necessary temporal and spatial scales. Assumptions about the relationship between habitat suitability and habitat quality must be made explicit. Even then, care should be taken in inferring multiple types of habitat quality from SDMs.

1. 直接测定生境质量（habitat quality）的高昂成本，促使生态学家探索用于估算与预测这一极重要生态变量的替代方法。具体而言，生境适宜性模型（habitat suitability models）即物种分布模型（species distribution models, SDMs）常被认为可作为生境质量的有效指标，但该假设极少得到验证，其应用视角既涵盖个体动物层面，也包含资源管理者层面。尤为关键的是，SDMs正愈发频繁地被用于估算物种分布范围，其隐含假设为：适宜性较高的区域将拥有更高的种群存续概率。这一假设支撑了诸多研究实践：利用SDMs设计保护区、评估隐存物种的种群现状，或是制定应对气候变化的管理策略。近期针对该关联的验证结果存在分歧，提示SDMs或可预测种群丰度，但无法反映高质量生境的其他衡量指标（如存活率、存续率）。
2. 本研究以濒危物种巨更格卢鼠（Dipodomys ingens）为研究对象，采用机器学习方法最大熵模型（Maxent），基于3种不同空间尺度构建了多套SDMs。随后将所构建的模型与3项生境质量衡量指标（存活率、种群丰度与体况）进行对比分析。
3. 结果显示，SDMs与存活率无显著相关性，但所有尺度下的模型均与种群丰度呈正相关关系。相较于大尺度模型，细尺度模型与种群丰度的相关性更强。而任意尺度下，体况均与生境适宜性无显著相关性。模型无法预测存活率的原因可能有三：一是环境协变量（environmental covariates）的信息不足，二是存在未被监测的群落过程或随机事件，三是仅以物种出现为预测目标的模型难以同时反映种群动态。综合分析与应用启示：SDMs（尤其是细尺度模型）或可服务于长期管理目标，例如筛选可供保护的高质量生境。但短期管理决策仅能依托匹配对应时空尺度的协变量所构建的模型制定。生境适宜性与生境质量之间的关联假设必须予以明确说明。即便如此，依托SDMs推断多种类型的生境质量时仍需保持谨慎。