Data from: Establishing macroecological trait datasets: digitalization, extrapolation, and validation of diet preferences in terrestrial mammals worldwide

Ecological trait data are essential for understanding the broad-scale distribution of biodiversity and its response to global change. For animals, diet represents a fundamental aspect of species' evolutionary adaptations, ecological and functional roles, and trophic interactions. However, the importance of diet for macroevolutionary and macroecological dynamics remains little explored, partly because of the lack of comprehensive trait datasets. We compiled and evaluated a comprehensive global dataset of diet preferences of mammals ("MammalDIET"). Diet information was digitized from two global and cladewide data sources and errors of data entry by multiple data recorders were assessed. We then developed a hierarchical extrapolation procedure to fill-in diet information for species with missing information. Missing data were extrapolated with information from other taxonomic levels (genus, other species within the same genus, or family) and this extrapolation was subsequently validated both internally (with a jack-knife approach applied to the compiled species-level diet data) and externally (using independent species-level diet information from a comprehensive continentwide data source). Finally, we grouped mammal species into trophic levels and dietary guilds, and their species richness as well as their proportion of total richness were mapped at a global scale for those diet categories with good validation results. The success rate of correctly digitizing data was 94%, indicating that the consistency in data entry among multiple recorders was high. Data sources provided species-level diet information for a total of 2033 species (38% of all 5364 terrestrial mammal species, based on the IUCN taxonomy). For the remaining 3331 species, diet information was mostly extrapolated from genus-level diet information (48% of all terrestrial mammal species), and only rarely from other species within the same genus (6%) or from family level (8%). Internal and external validation showed that: (1) extrapolations were most reliable for primary food items; (2) several diet categories ("Animal," "Mammal," "Invertebrate," "Plant," "Seed," "Fruit," and "Leaf") had high proportions of correctly predicted diet ranks; and (3) the potential of correctly extrapolating specific diet categories varied both within and among clades. Global maps of species richness and proportion showed congruence among trophic levels, but also substantial discrepancies between dietary guilds. MammalDIET provides a comprehensive, unique and freely available dataset on diet preferences for all terrestrial mammals worldwide. It enables broad-scale analyses for specific trophic levels and dietary guilds, and a first assessment of trait conservatism in mammalian diet preferences at a global scale. The digitalization, extrapolation and validation procedures could be transferable to other trait data and taxa.生态性状数据是解析生物多样性大范围分布格局及其对全球变化响应机制的核心基础。对于动物类群而言，饮食是物种进化适应、生态与功能角色以及营养相互作用的核心方面。然而，饮食在宏观进化与宏观生态动态中的关键作用仍鲜有探索，其核心阻碍之一便是缺乏综合化的性状数据集。我们汇编并评估了一套覆盖全球陆生哺乳动物的综合性饮食偏好数据集——MammalDIET。该数据集的饮食信息源自两个全球范围且涵盖所有类群的数据源，并对多名数据录入人员的操作误差开展了系统性评估。随后，我们构建了层级化外推流程，用于补全缺失饮食信息的物种数据：基于属、同属内其他物种乃至科等更高分类层级的信息，对缺失数据进行外推，并分别通过内部验证（对已汇编的物种级饮食数据应用刀切法（jack-knife））与外部验证（采用来自综合性大陆级数据源的独立物种级饮食信息）对该外推流程的可靠性进行校验。最终，我们将哺乳动物物种划分为营养级与饮食功能群，并针对验证表现良好的饮食类别，以全球尺度绘制了其物种丰富度及总丰富度占比的空间分布图。本次数据录入的正确率达94%，表明多名记录者间的数据一致性较高。原始数据源共为2033个物种提供了物种级饮食信息（基于国际自然保护联盟（International Union for Conservation of Nature, IUCN）的分类体系，占5364种陆生哺乳动物的38%）。剩余3331个物种的饮食信息大多通过属级饮食数据外推得到（占所有陆生哺乳动物的48%），仅少量通过同属内其他物种（6%）或科级分类单元（8%）的信息完成外推。内部与外部验证结果显示：（1）针对核心食物类别的外推结果可靠性最高；（2）多个饮食类别（"Animal"、"Mammal"、"Invertebrate"、"Plant"、"Seed"、"Fruit"与"Leaf"）的饮食等级预测正确率均处于较高水平；（3）特定饮食类别的正确外推潜力在类群内部及类群之间均存在显著异质性。物种丰富度与占比的全球空间分布图显示，不同营养级间的分布格局具有一致性，但不同饮食功能群间则存在显著差异。MammalDIET是一套覆盖全球所有陆生哺乳动物的综合、独特且可免费获取的饮食偏好数据集，可为特定营养级与饮食功能群的大范围宏观分析提供支撑，并首次实现了全球尺度下哺乳动物饮食偏好性状保守性的系统性评估。本研究采用的数字化录入、数据外推与验证流程，可推广至其他性状数据与类群的相关研究中。