Mapping the ecological dimensions and potential distributions of endangered relic shrubs in western Ordos biodiversity center

Potential distributions of endemic relic shrubs in western Ordos were poorly mapped, which hindered our implementation of proper conservation. Here we investigated the applicability of ecological niche modeling for endangered relic shrubs to detect areas of priority for biodiversity conservation and analyze differences in ecological niche spaces used by relic shrubs. We applied ordination and niche modeling techniques to assess main environmental drivers of five endemic relic shrubs in western Ordos, namely, Ammopiptanthus mongolicus, Amygdalus mongolica, Helianthemum songaricum, Potaninia mongolica, and Tetraena mongolica. We calculated niche overlap metrics in gridded environmental spaces and compared geographical projections of ecological niches to determine similarities and differences of niches occupied by relic shrubs. All studied taxa presented different responses to environmental factors, which resulted in a unique combination of niche conditions. Precipitation availability and soil quality characteristics play important roles in the distributions of most shrubs. Each relic shrub is constrained by a unique set of environmental conditions, the distribution of one species cannot be implied by the distribution of another, highlighting the inadequacy of one-fits-all type of conservation measure. Our stacked habitat suitability maps revealed regions around Yellow River, which are highly suitable for most species, thereby providing high conservation value.

The investigation of ecological niches conservatism has become a significant research topic, together with the identification of main environmental constraints on species' distribution, given the expected impacts of climate change on biodiversity [1][2][3][4] . This outcome has prompted new tools development to assess how ecological niche of a species can shrink, expand or persist, in environmental and geographic spaces while anticipating climate change effects [5][6] . Currently, ecological niche differences among species can be visualized and analyzed to evaluate the possible ecological and evolutionary forces that shape geographical distributions and habitat preferences [7][8] . Species groups that are highly diverse may present a varied set of ecological adaptations along an environmental gradient which are of importance for understanding ecological niche differences and to prepare mitigation actions against climate change impacts. Information about the potential changes in ecological niches can be used to implement or guide conservation actions, especially in biodiversity-rich areas 9 .
Various methods have been proposed to study niche conservatism, the study of how species vary in their requirements for or tolerance of these factors has advanced, in part due to the conceptual development of ecological niche 10 . Methods for quantification and estimation of niche differences typically rely on either ordination techniques 11 or ecological niche modeling (ENM) 12 . The former approach allows direct comparisons of speciesenvironment relationships in environmental spaces 13 , whereas the latter is widely adopted in conservation biogeography. ENM assumes that each species occupies its own particular Grinnellian niche on a macroscale 7 . This modeling technique uses occurrence-associated environmental variables to assess potential distribution 12 . Niche overlap is then estimated through model predictions across a landscape. The development of ENM has facilitated the extraction of ecological niche characteristics, which can assist in biodiversity conservation 14 .
The western Ordos plateau is a relatively independent physiographic unit in the semi-arid zone in central-northern China (Fig. 1). Climate in this area is extremely cold during winter and spring and very dry during summer and fall. Shrubs are the dominant plant life forms in this region 15,16 . This geographically restricted area has developed not only rich shrub diversity but also high percentage of relic species and endemic taxa 17 , being as one of the eight floristic endemic centers in China 18 . These relic shrubs belong to the Mediterranean flora, they are playing an important role in prevention of soil erosion and desertification in arid areas (Fig. 1). However, the ecological requirements and geographical distributions of these relic shrubs were poorly mapped, many scattered populations were discovered beyond historical ranges. Additionally, the western Ordos plateau is rich in mineral resources (e.g. coal, kaolin, quartzite, and ironstone). Biodiversity in this area is suffering from heavy human disturbances 17 , which predominantly include mining activities and associated refining processes, infrastructure construction (e.g. highway) associated with urbanization. Grazing has also posed a serious threat to endangered relic shrubs in this area.
In present study, we selected five endangered endemic relic shrub taxa, namely, Ammopiptanthus mongolicus, Amygdalus mongolica, Helianthemum songaricum, Potaninia mongolica, and Tetraena mongolica based on their importance in western Ordos. For example, Tetraenoideae is a monotypic subfamily from Zygophyllaceae with only one genus, which contains only one species (i.e. Tetraena mongolica). This single species is endemic to western Ordos plateau ( Fig. 1) and defined as a rare and endangered species because of its extremely restricted distribution and very scarce population 19 . Here we analyzed how species-specific responses to environmental factors and differences in ecological niche space can aid future conservation. To this end, we used ENM and ordination techniques to characterize ecological niches of five relic shrubs and quantify the similarities between them. First, we identified main environmental variables that constrain their distributions. Second, we used information on their environmental constraints to generate consensus habitat suitability map to highlight hotspots of habitat suitability to inform conservation planning. Third, we assessed the similarities and differences of ecological spaces occupied by relic shrubs to investigate how differences in the distributions of ecological niche spaces and species-specific responses to environmental factors may inform conservation plans. Finally, we mapped the suitability of each shrub in dimensions of human disturbance to assess human threats. Similarities existed in the morphological and physiological characteristics of these shrubs. Thus, we assumed their ecological niches to be similar. However, due to the different adaptations to diverse environments, we expected ecological niches to be non-equivalent.

Results
Importance of environmental variables. Ten environmental variables (i.e. temperature mean diurnal range (BIO2), maximum temperature of warmest month (BIO6), aridity index (AI), aspect, compound topographic index (CTI), growing degree days (GDD), normalized difference vegetation index (NDVI), slope, soil nutrient availability (SQ1) and soil rooting conditions (SQ3) were finally used by considering biological relevance and spatial correlation (see below method for detail). Distributions of relic shrubs were underpinned by their different responses to the environment (Fig. 2). In western Ordos, shrub distributions were mainly constrained by a combination of AI, slope, and rooting conditions (Table 1). For A. mongolicus and A. mongolica, the AI and slope degree were important variables. In A. mongolicus, the suitability increased along BIO2, GDD, slope degree, and SQ1; however, such suitability decreased along AI and NDVI. The suitability of A. mongolica exhibited a unimodal response to AI, aspect, GDD, and slope degree; nonetheless, a linear response to BIO2, CTI, NDVI, SQ1, and SQ3 was observed (Fig. 2). For H. songaricum, soil qualities were important variables, including both SQ1 and SQ3, in which a remarkably negative response to SQ1 was observed (Fig. 2). NDVI and slope degree showed significant contribution to P. mongolica's distribution, whereas NDVI and AI contributed significantly to T. mongolica's distribution (Table 1). A unimodal response to AI, BIO2, and NDVI was observed in T. mongolica (Fig. 2). Different from expected, none of the shrubs were strongly constrained by BIO6, aspect, and CTI. Most shrubs demonstrated responses to variables that usually overpassed logistic suitability estimates of 0.5.  Table 1), which strongly supported their predictive power (except for P. mongolica). The null model protocol suggests that our results are significantly better than those expected by a random model. In fact, all ENMs performed significantly better than expected by chance alone (P < 0.01; Supplementary Table 1), except P. mongolica. These models successfully identified three wide-range species (A. mongolicus, A. mongolica, and P. mongolica) and two narrow-range species (T. mongolica and H. songaricum) (Fig. 3). Stacking of five distribution models resulted in a map with centers of high suitability for the shrubs (Fig. 3a). Centers of high suitability were mainly located near Yellow River around Wuhai City. Additional hotspots were found in southwestern Zhongwei. A narrow area in northwestern Yellow River was identified as highly suitable for most taxa (Fig. 3a). In an attempt to visualize suitable climate space within human disturbance gradient, the model that identified a suitable space was mapped in dimensions of human footprint and population density ( Supplementary Fig. 1). The suitable space for A. mongolicus and A. mongolica fell into areas of both high and low human disturbances, whereas P. mongolica tended to occupy areas of low human disturbance. For the two narrow-range species, T. mongolica can survive in areas with low to medium population density, whereas H. songaricum tended to occupy in sparsely populated region, both the two taxa tended to occupy areas of low human disturbance.

Biodiversity hotspots.
Ecological niche properties. Analysis of ecological niche properties rendered a PCA with the first axis mainly loaded by GDD, AI, slope, and NDVI, which explain 30.6% of total variation (Fig. 4). The second axis explained about 14.6% of the variation and was loaded by SQ1, SQ3, and BIO6.
Results from niche breadth assessment showed a high variation in environmental suitability for relic shrubs (Supplementary Table 1). The highest niche breadth that we found was A. mongolica (0.1109), which presented the broadest distribution of suitable habitat (Fig. 3). Two other wide-range species (i.e. A. mongolicus and P. mongolica) also exhibited broad niche breadth. T. mongolica (0.0228) and H. songaricum (0.0075) exhibited narrow niche breadth, corresponding to their limited geographic distributions (Supplementary Table 1, Fig. 3). Niche overlap results suggest high variations in environmental space inhabited by different shrubs (Table 2, Fig. 4). Great overlaps were observed between A. mongolicus and A. mongolica (0.585) and between H. songaricum and T. mongolica (0.484). However, the niche overlaps among other pairs were extremely low, ranging from 0.078 to 0.266, suggesting they occupy considerably different environment niches. Even in comparisons between wide-range species, niche overlaps (e.g. between A. mongolicus and P. mongolica, and between A. mongolica and P. mongolica) were extremely low (Table 2), these species differed in their occupied niche spaces (Fig. 4). All niche overlap values are presented in Table 2.
Null hypothesis of niche equivalency test was rejected for all comparisons between the five shrubs, except between A. mongolicus and A. mongolica (Table 2). By contrast, in our analysis of niche similarity, the null hypothesis held for all pairs of shrubs (i.e. niche similarity in Table 2

Discussion
Identification of main environmental constraints on species distribution is important for conservation actions of investigate future climate change effects 1 . In this study, we identified biodiversity hot spots and key environmental constraints on distributions of five relic shrubs by mapping ecological dimensions and potential distributions using state-of-the-art ENM and ordination techniques. The wide distribution of relic shrubs in western Ordos  plateau biodiversity center underlines the variety of environmental conditions in which they are adapted, these conditions reflected physiological differences inherent in these shrubs.

Environment shaping distribution. Environmental factors that shape shrub distribution varied consider-
ably. Arid index is found to be the most important factor for A. mongolicus (Fig. 2). Here aridity index was used to quantify precipitation availability over atmospheric water demand 20 . In western Ordos, climate is extremely dry, the annual precipitation is about 272.3 mm. Our results suggest that precipitation is an important limiting factor for A. mongolicus, A. mongolica, and T. mongolica. These plants have developed high capability in water absorption and decrease in transpiration to expand their geographical ranges (e.g. A. mongolicus) 21 . A linear response to AI was observed in A. mongolicus, whereas unimodal and semi-unimodal responses were observed in T. mongolica and A. mongolica, respectively (Fig. 2). Slope steepness affects plant growth through differential incidence of solar radiation, wind velocity, or soil type. Xerophytic plants are known to inhabit the south-facing slope 22 . Here, slope was found to be an important factor in three wide-range species. Less exposure to insulation and moisture abundance in soil probably be important to these species. For H. songaricum, soil quality (i.e. SQ1 and SQ3) was found to be the most important factors. H. songaricum is a deciduous shrub, this plant is found primarily on lithoid hillsides (about 1000 m to 1300 m) in western Ordos. Su et al. suggested that the species dispersed from Central Asia into western Ordos plateau through Hexi Corridor during Tertiary period 23 . Our results suggest that soil properties, including soil texture, bulk density, coarse fragments, and soil organic carbon (SQ1 and SQ3), mainly restricted present distribution of H. songaricum in western Ordos. As founding above, variables reflecting precipitation availability and soil quality characteristics usually play an important role in these relic shrubs. These findings will not only help us in understanding the present distribution, but also pave the way for investigations of species responses to climate change and future conservation actions 1 .
Potential distribution. Given wide variation in environmental conditions in which shrubs grow, low niche overlap between these shrubs was expected. The low values of niche overlap were also reflected in their different environmental constraints. Geographical distributions of shrubs were generally consistent with their niche breadth because a broad niche allows species to persist in a wide range of habitat (Supplementary Table 1, Fig. 3). By contrast, a narrow niche restricts a species into the few places 24,25 . The limited distribution of T. mongolica and H. songaricum might be primarily due to their narrow niche breadth. Nonetheless, our results of ENM predictions showed that much suitable environmental space existed beyond known distributions of five shrubs, these areas might be useful in future investigation or transplantation actions. Biodiversity hotspots identified for five shrubs were mainly found around Yellow River in western Ordos, specifically in Wuhai area, southwestern Zhongwei, and a narrow area along northwestern Yellow River (Fig. 3). However, these areas fell into intensive human residence region bearing high human disturbance ( Supplementary Fig. 1). In western Ordos, the human activities caused by urbanization or mining have seriously impacted relic shrubs 17 .
Implications for conservation. Importance of relic shrubs in western Ordos were because of their floristics in plant evolution and biogeography, and their ecological function against desertification. Shrubs in Ordos plateau can resist wind, stabilize sand, preserve biodiversity, and protect habitats in certain degrees 17 . Notably, biodiversity conservation, ecological function and economic development are correlated and equally important in western Ordos, nonetheless this reality was not fully recognized by regional government, although a nature reserve has been set up for these species and other rare and endangered plants in western Ordos. Our results suggest that each relic shrub in western Ordos plateau is constrained by a unique set of environmental conditions, their non-equivalence of ecological niches implied that the distribution of one species cannot be implied by the distribution of another, highlighting the inadequacy of one-fits-all type of conservation measure. Conservation for each shrub should designed carefully to reflect its unique environmental requirements. Both T. mongolica and H. songaricum have declined drastically over the past decades and are classified as endangered species in China Species Red List 26 . Much suitable ecological space existed beyond known distributions of the five shrubs. However, these areas fell into areas of human activities and were severely affected by urbanization or mining. Thus, efforts to balance habitat protection and economic development should be prioritized in western Ordos. Mapping distribution and ecological dimension. We used ENMs to analyze spatial distribution of relic shrubs and identify key environmental variables constraining their distributions. Maximum entropy modeling implemented in Maxent was adopted 38 , Maxent follows the principle of maximum entropy and distributes probability as uniformly as possible. Recent studies have shown great advancements on geographic background selection, and reduction of space correlation or model complexity in ENM [39][40][41] . We used SDM tool 30 to prepare background bias file, as well as to fine tune model feature and regularization parameters for each species to best model their ecological niche and potential distribution. Model strength was estimated using area under the curve (AUC) of receiver operator characteristic (ROC) generated by Maxent. Null model approach 42 was used to test significance of our model predictions. Added advantage of testing against a null model is that all collection localities can be used for model calibration 42 . The relic shrub models were projected on study area to identify suitable habitats for their distribution and conservation. Model predictions were thresholded to produce binary maps using 10th training presence threshold, this approach is conservative in ecological niches estimation because this technique eliminates extreme values that may result from erroneous identification or georeference (i.e. partial ROC approach) 43 . Variables in models capable of predicting shrubs' presence were identified through permutation importance test, a high percentage of permutation importance indicates a large relative decrease in AUC value after random permutation, thereby signifying high reliance on the variable 44 .
Niche breadth, overlap, equivalency, and similarity. Niche breadth was estimated by applying inverse concentration measure of Levins as implemented in ENMTools 5,45 , we obtained niche breadth of each species to assess degree of shared niche space between shrubs. Niche breadth ranges from 0 (when all but one grid cell exhibits non-zero suitability) to 1 (when all the grid cells in the study area are equally suitable) 46 , species with a wider environmental distribution render higher niche breadth values. Assessment of niche overlap allows quantification of niche shared by the shrubs, niche overlap between pairs of shrubs was computed using Schoener's D statistics 47,48 , ranging from 0 (when two species present no overlap in environmental space) to 1 (when two species share same environmental space). Niche equivalency and similarity were determined between pairs of shrubs using a kernel smoothing script 6 . Multivariate niche overlaps in gridded environmental space between pairs of taxa were compared using PCA_env function (i.e. principal component analysis on entire environmental space of the two ranges) 6 . Niche overlap was measured along the gradients of multivariate analysis; furthermore, statistical tests of niche equivalency and similarity 48 were computed from the density in environmental space as described by Broennimann et al. 6 .
Human disturbance. In an attempt to visualize human disturbances under which models identified suitability across western Ordos Plateau, human disturbance variables associated with model prediction were extracted and visually displayed in scatter plots. In this step, human disturbances were represented by human footprint and population density. The former is a composite summary of human influence on land surfaces 49 , whereas the latter is a gridded population distribution, which was obtained from Oak Ridge National Laboratory 49 .