Considering that most protected areas and mammal fauna data were organized according to administrative units, species richness, functional diversity, and phylogenetic diversity, as well as functional and phylogenetic structure, were calculated using the prefecture as the basic unit rather than dividing the study area into grid cells (Huang et al. 2020). The Global Aridity Index dataset was used to identify dryland regions in China (Zomer et al. 2022). The aridity index was less than 0.03 for extremely arid regions, 0.03–0.2 for arid regions, 0.2–0.5 for semi-arid regions, 0.5–0.65 for semi-humid regions, and greater than 0.65 for humid regions (Zomer et al. 2022). The study specifically focused on regions with an aridity index of less than 0.5. These regions were then overlaid with the administrative boundaries of Chinese prefecture-level cities. Finally, 96 prefecture-level cities that overlapped with the arid and semi-arid regions were selected as the study area.

Terrestrial mammal distribution data for the 96 prefecture-level cities were collated from various faunal works and a PhD thesis (Suppl. material 1: table SS1), including Heilongjiang Fauna (Ma 1986), Liaoning Fauna (Xiao 1988), and a PhD thesis on the species diversity and distribution patterns of veterinary species in Northeast China (Zhang 2000). A total of 249 mammal species were recorded across the 96 prefecture-level cities, comprising 3,098 prefecture-level presence records. Species richness per prefecture ranged from two to 70 species (Suppl. material 1: fig. S1; table S2). Angiosperm plant species richness data for the corresponding prefectures were obtained from the Chinese Vascular Plant Distribution Database, which was compiled from several floras, published literature, and herbarium records (Lu et al. 2018). Mammal endangerment status was based primarily on information in the Red List of Chinese Vertebrates (Jiang et al. 2016). Missing information was supplemented using the IUCN website (https://www.iucn.org/) and published literature (Jiang et al. 2016; IUCN 2021; Ding et al. 2022).

The climatic variable parameters were obtained from http://www.worldclim.com at a spatial resolution of 30 arc-seconds, including mean annual temperature (MAT), annual precipitation (MAP), and MAT and MAP during the Last Glacial Maximum (LGM) (Hijmans et al. 2005). The average values from two models, i.e., the Community Climate System Model version 3 (CCSM3) and the Model for Interdisciplinary Research on Climate version 3.2 (MIROC3.2), were used to represent MAT and MAP during the LGM. Climate change from the LGM to the present was calculated as contemporary MAT or MAP minus MAT or MAP during the LGM and was expressed as anomalies in MAT or MAP. All climatic variables were extracted as average values for each prefecture-level city. Elevation data were downloaded from http://earthenv.org/DEM.html, and elevation range (Range ELE) was calculated as the maximum elevation value minus the minimum elevation value for each prefecture (Robinson et al. 2014).

The HYDE v3.2 database provides gridded cropland and pasture coverage at a spatial resolution of 0.083° (Klein Goldewijk et al. 2017). The dataset was divided into two periods (1900–1950 and 1950–2000), and land-use change in cropland and pasture was calculated as the absolute difference between the end year and the start year. For each period, the absolute changes in cropland and pasture were then summed to represent land-use change (Suppl. material 1: table S3).

The phylogenies of mammals in the arid and semi-arid regions (249 species) and all mammals in China (600 species) were pruned from a global mammal phylogeny obtained from http://vertlife.org/phylosubsets using the option “Mammal birth–death node-dated completed trees (all 5,911 species, set of 10,000 trees)” (Upham et al. 2019). A posterior sample of 10,000 mammal phylogenetic trees was used to construct the maximum clade credibility tree using the mean node heights algorithm in TreeAnnotator software (version 1.8.2) of the BEAST 2 package (Bouckaert et al. 2014).

A functional dendrogram was constructed based on five key traits: body mass, diet, litter size (number of pups produced in a litter), habitat breadth, and activity time (Suppl. material 1: table S4). Data on these traits were obtained from published literature (Cooke et al. 2019; Ding et al. 2022). These five traits collectively represent the morphology, functionality, and ecological strategies of mammal species (Chillo and Ojeda 2012). Because body mass (log10-transformed), litter size (log10-transformed), and habitat breadth (square-root-transformed) were continuous variables, whereas diet and activity time were categorical variables, Gower’s distance and the ‘complete’ clustering method in the R package vegan were used to construct the functional dendrogram (Oksanen et al. 2015).

The mammal occurrence records for the 96 prefecture-level cities were first converted into a binary matrix of species presence and absence, with each city represented by a row and each mammal species represented by a column. A value of 1 indicated the presence of a species in a prefecture-level city, and 0 indicated its absence. Taxonomic diversity was calculated as the total number of mammal species recorded in each prefecture-level city. Before phylogenetic and functional metrics were calculated, species names in the community matrix were matched with the tip labels of the phylogenetic tree and functional dendrogram.

Phylogenetic and functional diversity were calculated using the ‘pd’ function in the ‘picante’ package based on the binary species presence matrix and the corresponding phylogenetic tree or functional dendrogram. Phylogenetic diversity was calculated using the pruned mammal phylogenetic tree, whereas functional diversity was calculated by replacing the phylogenetic tree with the functional dendrogram. Thus, both phylogenetic and functional diversity were measured as the total branch length connecting all mammal species present in each prefecture-level city. Mean pairwise distance (MPD), which was used as a metric of phylogenetic and functional structure, was calculated using the ‘mpd’ function (Kembel et al. 2010). Phylogenetic MPD was calculated from the pairwise phylogenetic distance matrix, whereas functional MPD was calculated from the pairwise functional distance matrix derived from the functional dendrogram.

To control for the effects of species richness on phylogenetic and functional diversity and structure, standardized effect sizes (SES) were used to assess phylogenetic and functional metrics (Swenson 2014). Standardized effect sizes were calculated by comparing the observed phylogenetic or functional diversity or structure with values generated from 999 randomized trees or functional dendrograms. A tip-label randomization null model was used. In this null model, the species composition and species richness of each prefecture assemblage were kept unchanged, whereas species labels on the phylogenetic tree or functional dendrogram were randomly shuffled 999 times. The formula for calculating SES was as follows:

SES= obs − mean rndsdrnd

where obs is the observed phylogenetic or functional diversity or structure of each mammal assemblage, mean_rnd is the mean of the 999 null models (distance matrix labels were shuffled 999 times), and sd_rnd is the standard deviation of these 999 null models. For phylogenetic and functional diversity metrics, a positive SES indicates higher-than-expected phylogenetic or functional diversity, whereas a negative SES indicates lower-than-expected phylogenetic or functional diversity. For phylogenetic and functional structure metrics, a positive SES indicates a phylogenetically or functionally overdispersed structure relative to random expectations, whereas a negative SES indicates a clustered structure. The ‘pd’ function in the R package picante was used to calculate phylogenetic and functional diversity separately for the study region and all of China. This allowed determination of the proportion of mammal phylogenetic and functional diversity in the study region relative to the total mammal phylogenetic and functional diversity in China.

Spatial A spatial autoregressive (SAR) model was used to account for spatial autocorrelation in model residuals and to assess the relationships between environmental, biotic, and anthropogenic variables and each diversity or assemblage structure metric. The response variables included species richness, standardized phylogenetic diversity, standardized functional diversity, standardized phylogenetic structure, and standardized functional structure. The explanatory variables included mean annual temperature, annual precipitation, glacial–interglacial mean annual temperature change, glacial–interglacial annual precipitation change, elevation range, plant species richness, land-use change 1900–1950, and land-use change 1950–2000.

A spatial weights matrix was constructed using the k-nearest-neighbors method based on the geographic coordinates of the prefecture-level cities. For each prefecture-level city, the four nearest prefecture-level cities were defined as its spatial neighbors (k = 4). Because the coordinates were longitude–latitude data, great-circle distances were used to identify the nearest neighbors. The resulting neighbor object was then converted into a spatial weights matrix and used in the SAR models.

Species richness was square-root-transformed to obtain normally distributed residuals. To compare regression coefficients, all variables were standardized (mean = 0, standard deviation = 1). Multicollinearity among explanatory variables was assessed using the variance inflation factor (VIF). All explanatory variables had VIF values lower than 3, suggesting that multicollinearity did not substantially affect the models (Suppl. material 1: fig. S2). All possible subsets of candidate variables were fitted. To control for the effect of the area of each prefecture-level city on the results, this variable was included as a fixed effect in all candidate models. Candidate models were ranked according to Akaike’s information criterion corrected for small sample size (AICc), and ΔAICc and Akaike weights (w) were used to assess the relative support for each model (Suppl. material 1: table S5). ΔAICc was calculated relative to the model with the lowest AICc. Models with ΔAICc < 2 were considered to have equivalent support and were retained as the confidence set for model averaging (Burnham and Anderson 2002). This confidence set was averaged using the ‘model.avg’ function in the ‘MuMIn’ package. Model-averaged regression coefficients, unconditional standard errors (SE), and 95% confidence intervals (CIs) were extracted (Barton and Barton 2015). Explanatory variables were considered significant if their 95% CIs did not overlap zero. In addition, the relative importance of each explanatory variable was evaluated by summing its AICc weights across all candidate models. All analyses were conducted in R version 4.1.1 (R Core Team 2016).

Plant species richness was positively associated with mammal taxonomic diversity. This relationship may be explained by the food resource diversity hypothesis, which suggests that areas with richer plant assemblages can provide a wider range of potential food resources and thereby support more consumer species occupying different trophic niches (Novotny et al. 2006; Kissling et al. 2007). For mammals, this mechanism may operate both directly and indirectly. Plant resources can directly support herbivorous mammals, while more diverse plant communities may also support a greater diversity of insects, rodents, and other prey resources, thereby indirectly benefiting insectivorous and carnivorous mammals (Feng et al. 2019). This mechanism is supported by evidence from Inner Mongolia, where mammal richness increased with food resource diversity across herbivorous, insectivorous, and carnivorous groups (Feng et al. 2019). Therefore, the positive relationship between plant species richness and mammal taxonomic diversity in this study may reflect the role of plant resources in maintaining trophic resource diversity and supporting the coexistence of mammals with different feeding niches.

In addition to plant species richness, taxonomic diversity may also be shaped by contemporary climate (Romdal et al. 2012; Boucher-Lalonde et al. 2015). The tropical niche conservatism hypothesis suggests that most species originated in the tropics and partially retain their ancestral physiological tolerances, such that warmer and wetter regions have higher species richness (Romdal et al. 2012; Boucher-Lalonde et al. 2015). In addition, previous studies have demonstrated that precipitation is the dominant limiting factor determining species richness in arid and semi-arid regions (Noy-Meir 1973; Hawkins et al. 2003; Fei et al. 2022). However, unexpectedly, the results showed a negative correlation between annual precipitation and species richness. This unexpected pattern may partly reflect the use of annual precipitation summarized at the prefecture level across large administrative units. This spatial aggregation may obscure local precipitation gradients within prefectures, especially in large administrative units with substantial environmental heterogeneity (Suppl. material 1: fig. S3). Therefore, this negative association should be interpreted cautiously because it may be influenced by spatial grain and environmental variation within prefectures (Nogués-Bravo and Araújo 2006; Hortal et al. 2008; Stein et al. 2014). In addition, low-precipitation areas may harbor a relatively large number of species adapted to arid conditions, such as Camelus ferus, Lepus yarkandensis, Gazella subgutturosa, and Euchoreutes naso. This may partly offset the expected decline in species richness under low precipitation. Therefore, this relationship should not be interpreted simply as a direct negative effect of precipitation on mammal richness (Fuller et al. 2021).

Beyond modern climate, historical climate dynamics may also have contributed to current patterns of taxonomic diversity. Contrary to the initial hypothesis, taxonomic diversity was positively correlated with glacial–interglacial precipitation change. This suggests that greater increases in precipitation since the LGM may have shaped mammal range dynamics, allowing some regions to support higher current mammal richness (Svenning et al. 2015; Morales-Barbero et al. 2018; Williams and Blois 2018). Because water is a key limiting factor in arid and semi-arid systems, historical increases in moisture may have enhanced primary productivity, vegetation cover, and the extent of suitable habitats (Noy-Meir 1973; Hawkins et al. 2003). These changes could have allowed some mammal species to persist locally, expand their ranges, or recolonize newly suitable areas during postglacial community assembly (Svenning et al. 2015; Morales-Barbero et al. 2018; Williams and Blois 2018). This explanation is consistent with studies showing that past climate change has shaped mammal diversity through historical range shifts, refugial persistence, and postglacial recolonization (Davies et al. 2011; Svenning et al. 2015; Morales-Barbero et al. 2018; Santos et al. 2020).

Anthropogenic land-use change is a major driver of biodiversity loss through habitat loss and fragmentation (Durán et al. 2020; Cabernard et al. 2024). Unexpectedly, mammal phylogenetic diversity was positively associated with land-use change between 1900 and 1950. This may reflect the fact that, during 1900–1950, large-scale mechanized agriculture and modern infrastructure had not yet been widely implemented in these regions, and land-use change was largely driven by small-scale cultivation, seasonal grazing, and traditional agropastoral mosaics (Perkins 2017). Such changes in cropland and pasture coverage under relatively low-intensity traditional land-use systems may have increased the diversity of habitat types (e.g., croplands, pastures, and semi-natural wastelands). This may have provided habitat opportunities for species occupying different ecological niches, which is consistent with the intermediate disturbance hypothesis (Tripathi et al. 2021; Viljur et al. 2022).

In contrast, the relationship between mammal phylogenetic diversity and land-use change between 1950 and 2000 was not significant. This pattern may reflect a mismatch between the land-use metric used here and the dominant forms of anthropogenic disturbance after 1950. During this period, land-use changes included not only changes in cropland and pasture coverage but also agricultural mechanization, irrigation, fertilizer and pesticide inputs, infrastructure expansion, and changes in grazing pressure (Foley et al. 2005; Erb et al. 2013; Kuemmerle et al. 2013). These processes can affect mammal habitat quality, landscape connectivity, and community composition, thereby influencing phylogenetic diversity (Newbold et al. 2015; Powers and Jetz 2019). Therefore, the lack of a significant relationship does not necessarily indicate that land-use change during this period had no ecological effect.

Modern climate may likewise play an important role in shaping mammal phylogenetic diversity (Barreto et al. 2019; Sun et al. 2020). In this study, mammal phylogenetic diversity was positively correlated with modern temperature. This pattern is consistent with the tropical niche conservatism hypothesis, which proposes that many mammal lineages originated in warm environments and that relatively few lineages have adapted to cold conditions (Barreto et al. 2019). However, this relationship may vary among regions and taxonomic groups. A previous study in the Wuling Mountains found a negative correlation between the phylogenetic diversity of small mammals and temperature, suggesting that higher temperatures may constrain the distribution of certain lineages under specific regional conditions (Sun et al. 2020). Therefore, the positive temperature effect observed in this study may indicate that, in arid and semi-arid China, warmer conditions reduce cold limitation and allow more evolutionarily distinct mammal lineages to persist.

Land-use change may also have influenced mammal phylogenetic structure (Graham et al. 2019). Mammal phylogenetic structure was positively associated with land-use change 1900–1950, indicating that greater land-use change was linked to a more overdispersed phylogenetic structure. This pattern may be due to niche differentiation facilitated by heterogeneous landscapes resulting from moderate land-use change (Cavender-Bares et al. 2004; Escobedo et al. 2017). In heterogeneous landscapes, distantly related species may use different resources or microhabitats, thereby reducing direct competition and promoting an overdispersed phylogenetic structure (Cavender-Bares et al. 2004; Escobedo et al. 2017). Thus, this positive association may reflect niche differentiation in historically heterogeneous landscapes rather than a direct benefit of land conversion.

A previous study found that glacial–interglacial climate change had negative impacts on the functional richness of terrestrial vertebrates, indicating that strong historical climate change can leave a legacy on present-day functional diversity patterns (Huang et al. 2023). However, functional diversity was positively correlated with glacial–interglacial precipitation change (Fig. 2). This contrasting result may reflect the role of historical precipitation variability in dryland ecosystems, where past shifts in precipitation conditions could have altered the balance between dry and more productive habitats (Noy-Meir 1973; Gherardi and Sala 2015). Larger historical changes in precipitation may have promoted the coexistence of mammals with different drought tolerances, dietary strategies, and dispersal abilities (Stein et al. 2014; Gherardi and Sala 2015).

The findings also indicated a significant positive correlation between mammal functional diversity and elevation range. A larger elevation range could provide more niches, thereby promoting functional diversity (Stein et al. 2014). In addition, the greater complexity of vegetation structure and the larger number of habitat types associated with a larger elevation range could provide more resources than a more uniform environment, supporting the coexistence of more species (Tews et al. 2004; López-González et al. 2015). Water availability under current climatic conditions may represent another important mechanism underlying variation in mammal functional diversity. A positive correlation was also found between mammal functional diversity and annual precipitation. Higher precipitation could increase the availability of food and habitat types, thereby facilitating functional diversity (Qian et al. 2009; Wang et al. 2017; Gorczynski et al. 2021). This interpretation is also consistent with evidence from cricetid rodents in southern Mexico, where precipitation was positively correlated with functional diversity, and low functional diversity occurred mainly in areas with low precipitation and strong climatic seasonality (Martín-Regalado et al. 2019).

Mammal community structure can be shaped by both modern climate and historical legacies, but their relative importance may vary among regions and mammal groups (Rowan et al. 2016, 2020). The results did not show significant correlations between phylogenetic structure and modern climate, but functional structure was significantly positively correlated with annual precipitation. Under drier conditions, water limitation may filter mammal species according to drought tolerance and resource-use strategies, leading to more functionally clustered assemblages (Le Bagousse-Pinguet et al. 2017; Liang et al. 2020). For example, a study of the phylogenetic and functional structure of birds on university campuses in northern China found that functional structure tended to be more clustered on relatively dry university campuses, consistent with the findings of this study (Liang et al. 2020). In addition, functional structure was positively correlated with elevation range, i.e., a greater elevation range promoted a more overdispersed functional structure. Habitats with greater heterogeneity typically support greater differences in functional traits among species, leading to a more dispersed functional structure (Barr and Biernat 2020).

This study indicated that the conservation value of mammals in arid and semi-arid China cannot be evaluated solely based on species richness. Although these regions supported only approximately one-third of China’s mammal species, they accounted for 52% of the total phylogenetic diversity and 85% of the total functional diversity of Chinese mammals. Moreover, the spatial hotspots of taxonomic, phylogenetic, and functional diversity showed limited overlap. This suggests that the identification of conservation priority areas based mainly on species richness may overlook regions with substantial evolutionary history or ecological functional importance (Devictor et al. 2010; Brum et al. 2017; Cadotte and Tucker 2018). Therefore, conservation strategies should incorporate multidimensional assessments.

In addition, the current national nature reserve network provides insufficient coverage for some areas important for multidimensional mammal diversity (Suppl. material 1: fig. S4; tables S7, S8). Among the 96 prefecture-level cities, 74 had less than 10% of their area covered by national nature reserves (Suppl. material 1: fig. S4; table S8). Among the prefecture-level cities in the top 20% for taxonomic diversity, phylogenetic diversity, and functional diversity, 52.4%, 100%, and 70.0%, respectively, had national nature reserve coverage below 10%. This finding indicates that prefecture-level cities with high phylogenetic diversity may be substantially underrepresented in the current national nature reserve network. These results suggest that the current national nature reserve network may not only exclude some areas with high species richness but also fail to adequately cover areas with high phylogenetic and functional diversity. The effectiveness of national nature reserves depends not only on the total area covered but also on whether they represent key components of biodiversity (Rodrigues et al. 2004; Maxwell et al. 2020). The potential conservation gaps identified here can serve as candidate areas for more targeted conservation gap assessments in the future (Suppl. material 1: fig. S4).