A greener Loess Plateau in the future: moderate warming will expand the potential distribution areas of woody species

Understanding the effect of future global warming on the distribution and diversity of woody species in the Loess Plateau is critical to the vegetation restoration and rebuilding of this area and yet is highly challenging. In the absence of enough experimental data, projection based on species distribution models is the best option for assessing the future shift in species distribution areas. Here, via a comprehensive habitat suitability model, we present an assessment of potential distribution area change for two plant life forms with a total of 42 key woody species, including 21 tree species and 21 shrub species, on the Loess Plateau under multiple scenarios, and based on this information, we explore the responses of regional potential woody species diversity to future global warming. The results showed that moderate warming will promote the expansion of potential distribution areas for most woody species and generally increase regional species diversity, which will result in a greener Loess Plateau. Our results also show that shrub species are more drought-tolerant and less adversely affected by climate change and thus should be considered a priority in vegetation restoration, especially in the arid area of the northern Loess Plateau. These results are helpful for identifying priority restoration areas, selecting appropriate species for artificial planting, and providing useful information for vegetation restoration and management in the future.


Introduction
The Loess Plateau is a crucial region for soil erosion control and vegetation restoration in China and an indispensable region in China's 'two barriers and three belts' ecological security strategic configuration, which impacts national ecological security , Sun and Wang 2022. The Loess Plateau is a traditional agricultural area in China, in where agriculture started approximately 7000 years ago (Roberts et al 2001, Huang et al 2004. With climate change and increased agricultural activity since the modern era, soil erosion is becoming increasingly severe in this area, leading to a fragile and insecure ecological environment. Some researchers argue that this area is the most severe soil erosion region globally (Cao et al 2018, Kong et al 2020).
Vegetation change is one of the most important indicators to evaluate the interaction between climate and terrestrial ecosystems (Walker and Steffen 1997). Especially in arid and semiarid ecosystems, vegetation dynamics are highly sensitive to climate change. Understanding vegetation change's spatial pattern and driving factors can provide an effective theoretical basis to prevent regional land degradation (Xie et al 2016, Kong et al 2020. More importantly, in the Loess Plateau, the decline in vegetative cover has been suggested to produce more land degradation and increase the frequency of meteorological disasters (Zhao et al 2013). Moreover, global warming in the future will bring great uncertainty to the vegetation change in this region, which has attracted wide attention from researchers (Liu et al 2014, Feng et al 2016.
China adopted the 'Conversion of Cropland to Forest and Grassland Programme' (CCFGP) in 1999, which aims to improve the quality of the ecological environment by increasing forestland and grassland (Wang et al 2007). The Loess Plateau is the key demonstration area of the CCFGP (Naeem et al 2021), and between 1999 and 2010, this programme has turned approximately 16 000 km 2 of non-irrigated arable land and sloping farmland into grassland and forests; as a result, the vegetation cover on the Loess Plateau increased by 25% (Bai et al 2019). Over the past decade, numerous studies have verified an increasing trend in vegetation greenness cover on the Loess Plateau (Naeem et al 2021), with the tree cover increasing by 41% (Xiao 2014). From 2001 to 2015, the overall annual normalized difference vegetation index (NDVI) showed an increasing trend with a growth rate of 0.0057 yr −1 in this area (Li et al 2021a). Most changes occurred in the central region, which encompasses approximately 54.99% of the plateau, and were mainly distributed in gullies and hills (Cao et al 2018, Yu et al 2020.
However, some researchers question the longterm impact of afforestation activities on the ecosystem in the Loess Plateau, and they believe that more artificial forests will consume more deep soil moisture, which may exacerbate soil desiccation and dry soil layer formation (Jiao et al 2012, Jia et al 2017. Conversely, afforestation on the Loess Plateau requires considerable human and material resources; hence, understanding the vegetation planting potential of the area is necessary to scientifically plan the vegetation restoration objectives and process.
Shrubs are usually selected as pioneer tree species for afforestation because of their ability to grow on barren soil , Li et al 2020. Compared with tree species, shrub species have a wider niche and are more suitable for various site conditions; they can grow in harsh ecological environments, such as broken terrain and loess cliffs formed by long-term soil erosion, and form a dense pure community (Wei et al 2013, Hao et al 2016. Hence, shrubs are important for soil moisture conservation and soil erosion reduction and can promote forest restoration and vegetation succession. Therefore, planting shrubs is an important part of greening and improving the ecological environment of the Loess Plateau. However, there are few studies on the suitability of shrub planting on the Loess Plateau, a subject that deserves further attention.
In the practice of vegetation restoration, we need to answer the following questions. From the perspective of vegetation restoration cost and effect, which places are more suitable for vegetation growth in the Loess Plateau and can be maintained without high human and material resources after vegetation restoration? Moreover, which places can be divided into priority restoration areas under the uncertain future climate change scenario? The potential distribution of plant species that can be supported by natural conditions without human influence is key information to fill this gap (Han andPeng 2021, Zhao et al 2021a). Species distribution models (SDMs) are a useful tool to perform this analysis, as they simulate the species niche according to the statistical information provided by sampling points in the multidimensional mathematical space composed of environmental variables to determine the preference of species for habitats with probability degrees, and then predict the distribution of species across space and time (Anderson 2013, Guo et al 2020. Here, we used SDMs to explore the baseline potential distribution of 42 main woody plants, including 21 tree species and 21 shrub species, which are also the key species used in afforestation on the Loess Plateau. To prepare adaptive strategies for the potential consequences of future climate warming, we also explored the potential distribution of these woody plants under four shared socioeconomic pathways (SSPs: SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) and four periods (2021-2040, 2041-2060, 2061-2080, and 2081-2100). Specifically, our study aimed to (a) determine the potential afforestation scope in the Loess Plateau under baseline conditions and identify priority restoration areas and (b) simulate climate change impacts on afforestation species distribution to explore possible vegetation change trends in the future. , we chose the typical planting trees and shrubs, common species, pioneer species, constructive plants, and dominant plants in the Loess Plateau, and ultimately, 21 tree species and 21 shrub species were selected for analysis. The occurrence data of these 42 woody plants were obtained from field surveys, published work, and online species databases such as the Chinese Virtual Herbarium (www.cvh. accn/) and Global Biodiversity Information Facility (www.gbif.org/). To obtain a more complete species niche estimation, the model analysis was performed across all of China, and finally, after removing duplicate coordinates and incomplete information, 3521 occurrence data points for 21 tree species and 2669 occurrence data points for 21 shrub species were selected nationwide (figure 1, table S1).

Environmental variables
Here, we collected 21 climate variables, including 19 bioclimatic variables (Fick and Hijmans 2017), wind speed, and surface soil moisture (Feng et al 2022) as climatic data. The 19 bioclimatic variables were downloaded from the WorldClim database (www. worldclim.org/) with a resolution of 2.5 arc-minutes (approximately 4.5 km at the equator) under baseline conditions (from 1970 to 2000) and four SSPs (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) in four periods (2021-2040, 2041-2060, 2061-2080, and 2081-2100) in the future (O'Neill et al 2017). The wind speed variables under baseline conditions were also downloaded from the WorldClim database and the future projected wind speed variables originated from the Coupled Model Intercomparison Project Phase 6 (CMIP6) climate projections with a resolution of 1.125 arc-degrees, which were downloaded from Climate Data Store (https://cds.climate.copernicus.eu). We used the statistical downscaling approach to improve the spatial resolution to 2.5 arc-minutes. More specifically, we chose the bioclimatic variables and wind data originating from the mediumresolution Beijing Climate Center Climate System Model version 2. The soil moisture data were from Feng et al (2022), which used a deep learning method to merge and downscale data from CMIP6 projections. These data were downloaded from the National Tibetan Plateau/Third Pole Environment Data Center (TPDC, http://data.tpdc.accn/en/). We used correlation analysis to screen all 21 climate variables to reduce the influence of multicollinearity (Feng et al 2019), with a threshold of 0.75. We screened variables with small correlations with each other (Robertson et al 2001, Guo et al 2022, and finally, nine climate variables were retained for the model analysis (table 1).
We used ten topsoil property variables and two topographical variables as non-climatic data (table 1), and the soil variables were the result of format conversion of the Harmonized World Soil Database (Nachtergaele et al 2009). Among the two topographical variables, slope data was downloaded from the TPDC (Yang 2019), and the geomorphological type data was downloaded from the Resource and Environmental Science and Data Center (www.resdc.cn/ Default.aspx). There is weak multicollinearity among all non-climatic variables, with correlation coefficients of <0.75.

Model construction 2.3.1. Distribution model strategy
We used the comprehensive habitat suitability (CHS) model (Zhao et al 2021b, Guo et al 2022 to predict the habitat suitability for all 42 woody species. The CHS model is an integrated model strategy that can simultaneously simulate the species' habitat shifts caused by climate change and the maximum suitable habitat range defined by the nonclimatic variables. In this process, we separately performed two model processes. One was the climate suitability model, which used nine bioclimatic variables as input data and an ensemble model combining ten popular model algorithms (i.e. GLM, GAM, GBM, CTA, ANN, SRE, FDA, MARS, RF, and MaxEnt) to explore the species' habitat distribution under every climate condition (Grenouillet et al 2011, Guisan et al 2017, Guo et al 2022. In practice, we used the biomod2 package (Thuiller et al 2016) based on the R-language to perform model analysis. For every species, all ten algorithms used the same pseudo-absences points (background points), and 2000 points were randomly created across China as  (1)) to reduce the uncertainty caused by different modelling algorithms and model inputs where EM i refers to the result of the climate niche model of the evaluation unit (grid) i; w j is the weight of the results of model j, determined by dividing the TSS max value of each model result by the sum of all models' TSS max ; and x ij is the value of evaluation unit i in the results of model j. Finally, via the maximum TSS value as thresholds, we binarized the results of the ensemble model. The other model is the distribution limitation model, which used the 12 non-climatic variables as input data, via MaxEnt to assess soil suitability requirements (Phillips et al 2006(Phillips et al , 2017 the model was evaluated by five-fold cross-validation to reduce the uncertainty and used the AUC as the model evaluation index. Finally, based on a threshold value, we binarized the results of MaxEnt, and the threshold value was defined as the minimum value that ensured that the true positive rate of the MaxEnt model was 0.9 (Bean et al 2012, Guo et al 2022). Lastly, the result of the CHS model was defined by the intersection of these two models in equation (2): For evaluation unit i, CHS i refers to the results of the CHS model, TM i is the result of the climate suitability model in evaluation unit i, and L i is the result of the distribution limitation model. Technical details of the CHS model followed the study of Zhao et al (2021b) and Guo et al (2022).

Species richness suitability maps
To identify priority restoration areas that are suited for most woody species to grow and to explore possible vegetation change trends on the Loess Plateau in the future, we superimposed suitability maps to create richness maps using equation (3) For evaluation unit i, where N i refers to the number of species, EM i represents the result of the climate niche model, L i is the result of the distribution limitation model, and n is the number of species.

Model evaluations
The climate suitability model performed well for the 42 woody species, as shown by the accuracy metrics (figure S1). The mean AUC was [0.84, 0.99], the mean value was 0.94, the mean TSS was [0.71, 0.93], and the mean value was 0.79, all of these indicated that our model result has gratifying precision. Concerning the plant life forms, shrub species had better model performance, with an AUC value of 0.95 ± 0.03 and TSS value of 0.81 ± 0.06, whereas the AUC value of tree species was 0.92 ± 0.04, and the TSS value was 0.77 ± 0.07. The distribution limitation models for the 42 woody species were successful and had precise results, with the mean AUCs being excellent (average 0.95, range 0.87-0.99).

Potential distribution area and species richness at baseline
Here, for ease of comparison and analysis, we used the proportion of the distribution area in the Loess Plateau (PALP) to define the size of the species' suitable habitat ( figure S2). The result shows that the PALP for all 42 woody species was 7.63%-65.09%, with an average value of 40.17%, and concerning the life forms, there was not much difference in PALP between trees and shrubs. The species richness suitability maps of all 42 woody species (figure 2(c)) showed that in nearly 90% of the Loess Plateau, the number of species was greater than 8. The areas with a number of species less than 8 were mainly located in the desert in the north of the Loess Plateau, and in such areas, the number of tree species suitable for growth was less than 4 ( figure 2(b)), but the number of shrub species was greater than 4 ( figure 2(a)). Therefore, shrub species should be given preference in vegetation restoration for desertification control in such areas. Nearly half of the Loess Plateau has numbers of species greater than 26; in such areas, climatic conditions are suitable for most plants, and thus they can be used as the key areas for regional vegetation restoration and construction.

Changes in the potential distribution area and number of species for woody species under climate change
According to the CHS model, we calculated the PALP for each species under every projected climate scenario (figure 3, table S1), and the results showed that 21 species will certainly expand their potential distribution area under all future scenarios, and only four species will certainly shrink their potential distribution area in the future. Under the mildest warming scenario (i.e. 2021-2040 under SSP1-2.6), the potential distribution area of 67% (28) of the species will expand, but under the most severe temperature increase scenario (i.e. 2081-2100 under SSP5-8.5), only half of species (21%) will expand their potential distribution area. In addition, the split violin density plot of PALP also showed an increasing trend under nearly all future climate warming scenarios, except for the most severe temperature increase scenario. To summarize, moderate warming will promote the expansion of potential distribution areas for most woody species, which will contribute to regional vegetation establishment and protection, but severe warming will have no benefit for regional greening.
To further clarify the spatial characteristics of species richness change, we mapped the change in the number of species in the Loess Plateau between baseline conditions and all projected climate change scenarios (figures 4, S3 and S4). The results showed that the woody species diversity in the desert area in the northern Loess Plateau tended to increase, and the species diversity of existing woody species in the southern Loess Plateau tended to decrease.
Concerning the different warming degrees ( figure 4), under the mildest warming scenario (i.e. 2021-2040 under SSP1-2.6, with a temperature rise of 1.28 • C), the number of species remained unchanged in most areas, and a slight increase occurred in the west and northeast of the Loess Plateau. Under the moderate warming scenario (i.e. 2081-2100 under SSP2-4.5, with a temperature rise of 3.29 • C), the largest greening trend will occur, as species diversity will increase in almost half of the regions, especially in northeast parts of the Loess Plateau, and the number of species will increase by more than 10. Under the most severe temperature increase scenario (i.e. 2081-2100 under SSP5-8.5, with a temperature rise of 5.01 • C), in the west and north of the Loess Plateau, the increasing trend in the number of species waned, and in the southern and even central parts of the Loess Plateau, the decreasing trend in the number of native species intensified.
Concerning the plant life forms (figures S3 and S4), shrub species and tree species have the same increasing trend in the western and northern Loess Plateau, but the decreasing trend in the number of shrub species is not as considerable as that of tree species in the southern Loess Plateau, which means that shrub planting on the Loess Plateau may benefit more from climate warming.

Importance of shrubs for vegetation rebuilding on the Loess Plateau
Recent work suggests that on the Loess Plateau, the potential distribution area changes of species with different life-form have the same trend under future climate change, but the details differ. Shrub species can grow in more arid areas and are less affected by climate change than tree species. This phenomenon can be supported by trait information of species with different life-form. Previous research has demonstrated that in this area, the resistance of radial growth to extreme drought events is stronger in shrubs (e.g. Sophora viciifolia and Rosa xanthina) than in trees (e.g. Pinus tabulaeformis) (Li et al 2020). Yang et al (2019) confirmed that shrub species (e.g. Salix psammophila and Caragana korshinskii) were morphologically more effective than tree species (e.g. Pinus tabuliformis and Armeniaca vulgaris) in funnelling rainwater to the soil of the basal area, which resulted in shrub species being more suitable for arid and semiarid environments. In addition, the observation of leaf economic and hydraulic traits of 31 wood species from different successional statuses on the Loess Plateau showed that compared with tree species, shrub species have a wider range of photosynthetic and hydraulic capacity, highlighting the functional diversity and wider adaptability range of shrubs (Yin et al 2019).
To further explore the niche differences between shrub and tree species, we analysed the response characteristics of different life forms to environmental variables. First, we calculated the average variable importance of all climate suitability models for the two life forms of species. The results showed that the importance of the climate variable for tree and shrub species was similar, and bio1 (mean annual air temperature) and bio12 (annual precipitation) were the most important climatic variables, with contribution rates of 31.40% and 25.15%, respectively (figure S5). Shrub species were more sensitive to climate variability than the average trend compared with tree species. Additionally, we used the MaxEnt model to create response curves for all two dominant climatic variables for these two life forms of species (figure 5) (Phillips et al 2006(Phillips et al , 2017. The results show that for mean annual air temperature, the response curves of the two life forms are almost the same, but for annual precipitation, shrub species have a wider suitable range, especially in the range of 150-250 mm, and shrub species should be the preference for regional vegetation restoration.

A greener Loess Plateau in the future
Our analysis indicates that future moderate warming will cause a generally increasing trend in the distribution area of woody species on the Loess Plateau. Several reports have shown that future moderate warming could result in a general northward expansion of the dominant woody plants on the Loess Plateau, e.g. Xanthoceras sorbifolium, Robinia pseudoacacia, Populus davidiana, P. tabuliformis, and C. korshinskii. Overall, with the projected increase in temperature, the range shift of these species to higher latitudes and elevations will become gradually more significant (Hu et  Here, we explore the shift in the distribution centre of all 42 woody species (figure 6) and PALP for different numbers of woody species under all four projected warming scenarios in 2081-2100 ( figure 7). The result shows that the projected moderate warming (SSP1-2.6 and SSP2-4.5) will cause the distribution centre to move to the interior of the plateau, which will increase the species diversity in the central and northern parts of the study area and lead to an increase in the vegetation distribution potential in these parts. Figure 7 also indicates that moderate warming will significantly increase species diversity in the whole region. However, under severe temperature increase scenarios, the area with high species diversity (more than 36 woody species) will decrease significantly. Notably, under SSP5-8.5, most species shift to the west of the Loess Plateau to higher elevations, which will reduce the increasing trend of species diversity in the north of the study area compared with other climate scenarios.
Recent work suggests that afforestation has resulted in soil desiccation, which negatively impacts the ecosystems on the Loess Plateau (Ding and Wang 2022). However, Wei et al (2022) have noted a positive correlation between NDVI and surface soil moisture across the whole Loess Plateau, and in the west and central-north (including Central Shanxi, northern Shaanxi, and Inner Mongolia) of the Loess Plateau, of the soil moisture showed an increasing trend in all layers. In addition, some reports suggested that there was still a 25%-50% potential for vegetation cover increase in the areas of hilly gullies and windy sandy areas in the same area (Zhao et al 2021a). Our results showed that in this area, the potential woody species diversity will increase under future moderate warming scenarios, which will support the further greening of the region.
Feng et al (2016) found large-scale afforestation has resulted in a significant increase in net primary productivity and evapotranspiration, but limited by rainfall and water, some areas of the Loess Plateau are close to the upper limit of sustainable use of water resources. This is the practical ceiling that cannot be ignored for vegetation restoration in the region at present and even in the future. Hence, for a greener Loess Plateau, a possible solution is selecting local plant species for afforestation and optimizing vegetation structure. Thus, future afforestation should consider species that use less soil water (Zhou et al 2016). In this process, shrub species should receive more attention, especially in the relatively arid northern part of the Loess Plateau. Introducing shrubs to semiarid grasslands could significantly increase species diversity, above-and belowground biomass, and surface soil water content (Hao et al 2016), and shrub cover can significantly improve the soil conservation capacity (Zhou et al 2016). Compared with tree species, shrub species show higher drought tolerance and lower soil water consumption, which will enable shrub species to play an important role in future vegetation restoration.
To summarize, future moderate warming will expand the habitat range and increase the diversity of the most dominant wood plants on the Loess Plateau. Most importantly, the northern boundary of species distribution will shift northward to hilly gullies and windy sandy areas, which will conveniently enhance future greening in this area. In addition, with further optimization of vegetation structure and the development and utilization of water-saving shrubs, there will be a potential to support a greener Loess Plateau in the future.

Limitations and suggestions
Some limitations of this study need to be clarified. One is the influence of planting activities on sampling points. For tree species with high economic value, planting samples may lead to the expansion of their natural niche in the model simulation, which should be identified and eliminated. On the other hand, the vegetation restoration of the Loess Plateau will still be dominated by human activities in the future; hence, the bias due to the planting samples is acceptable to some extent. The second is limited environmental variables. Due to the data availability and the limitation of the resolution, our model cannot include all possible important environmental variables, such as soil nutrients and future land use and cover data. The last is limited by simulation capability. Furthermore, in data collection, we did not distinguish between adult and regeneration samples; hence, we cannot identify the difference in niches of the same species at different growth stages.
Our results provide an optimistic projection that the future shift and habitat increase in woody species under moderate warming will produce a greener Loess Plateau, which will be beneficial to the restoration and rebuilding of regional vegetation and the implementation of the CCFGP. With the rising temperature, the regional potential species diversity will increase, and more woody species will be selected for artificial afforestation, which will increase the stability of the regional ecosystem. However, with the range shift of wood species, in the south of the Loess Plateau, the native tree species diversity will decrease, and it will be necessary to plant non-native species. Thus, in this process, the species' ecological adaptability and the risk of biological invasion should receive close attention from managers (Lenzner et al 2020, Guo et al 2022.

Conclusions
Understanding how the potential distribution of woody species might change in the future in the Loess Plateau is a necessary task that has thus far defied accurate projection. We suggest that the CHS model is a valuable tool to support quantitative assessments of possible impacts of future climate change on regional woody plant species distribution and diversity. We demonstrated that climate warming will promote the expansion of potential distribution areas for most woody species and generally increase regional species diversity, which will result in a greener Loess Plateau. Our projection can be used to identify priority areas for vegetation restoration and preferred species for regional vegetation restoration. In particular, our findings provide a scientific basis for regional vegetation restoration and rebuilding, which will provide useful information for the CCFGP.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.