Distinct Ecological Adaptations and Habitat Responses to Future Climate Change in Three East and Southeast Asian Sapindus Species


 Background: Sapindus is an important biodiesel, biomedical, and multifunctional economic forest species in Asia, however its germplasms have been persistently damaged or lost. It is imperative to conserve the diversity of Sapindus. This study aimed to reveal the potential habitat distribution patterns of Sapindus mukorossi, Sapindus delavayi, and Sapindus rarak in response to current environment and future climate change, and identify hotspots of habitat degradation/expansion to facilitate climate change-adaptive biological conservation. Methods: Using current environmental data and future climate projections (2021–2100), we simulated the present and potential future habitats of Sapindus mukorossi, Sapindus delavayi, and Sapindus rarak in east and southeast Asia using a maximum entropy (MaxEnt) model that was developed based on 2041 occurrence records. Results: The model showed that precipitation may play an important role in framing the potential habitats of Sapindus; however, S. delavayi was more sensitive to minimum temperatures (-2 °C to 3 °C) and elevation (1200-2000 m), while S. rarak was more demanding in terms of solar radiation (annual mean Uvb of 4600 to 5000 J/m2/day). Under the current environment, S. mukorossi has the widest suitable habitat distribution (250.24 × 104 km2), followed by that of S. rarak (173.49 × 104 km2), and S. delavayi (78.85 × 104 km2). Under future climate change scenarios, the habitat distribution of S. mukorossi will expand and contract, that of S. delavayi exhibited significant expansion. In contrast, future S. rarak habitat distribution exhibited significant contraction. Conclusions: There were significantly distinct ecological adaptations among Sapindus mukorossi, Sapindus delavayi, and Sapindus rarak in east and southeast Asia. The contraction areas should be subject to germplasm collection and ex situ conservation preferentially. The modelled unchanged areas should be used for potential future Sapindus mukorossi, Sapindus delavayi, and Sapindus rarak conservation and utilization.


Background
Humans have a signi cant in uence on future climate change, primarily re ected in the generation of anthropogenic greenhouse gas emissions ( has experienced several climatic upheavals, and plants tend to alter their bioecological characteristics, phenology, or resilience response mechanisms through natural selection and genetic evolution to gradually spread or migrate to more suitable habitats to avoid being affected by adverse climatic conditions (R. T. Corlett & Westcott, 2013; Hewitt, 2000; Thomas et al., 2003). However, in the face of rapid climate change, it is unrealistic for plants to evolve adaptation strategies in such a short period, especially for arbour species. Therefore, exploring the potential habitat distributions of cherished species under future climate change is crucial to their diversity and conservation.
Sapindus is an evergreen, or deciduous, tree in the Sapindaceae family, which comprises 13 species with a global distribution in warm-temperate to tropical regions, found mainly in southeast Asia and North and South America (J. Liu et al., 2017). Among these, Sapindus mukorossi (S. mukorossi), Sapindus delavayi (S. delavayi), and Sapindus rarak (S. rarak) are concentrated in east and southeast Asia (J. Liu et al., 2017). Owing to the high yield of its seed oil (26.69-44.69%), as well as the high medium-chain monounsaturated fatty acid content (Sun, Jia, Xi, Wang, & Weng, 2017), Sapindus seed oils are suitable for biodiesel production according to American Society for Testing and Materials (ASTM) D6751 and the European EN 14214 standards ( (Singh & Singh, 2008) properties. Saponin serves as an e cient natural surfactant in commercial soaps, shampoos, and cosmetic cleansers (Muntaha & Khan, 2015), and the root and fruit of Sapindus trees are used in traditional Chinese medicine. Therefore, Sapindus is regarded as an important biodiesel, biomedical, and multifunctional economic forest species (J. Liu et al., 2017; Sun, Jia, Ye, Gao, & Weng, 2016). Sapindus germplasm resources are widely distributed; however, they are generally scattered in the form of single plants or extremely small populations. Further, with global deforestation and the rapid anthropogenic development, Sapindus germplasms have been persistently damaged or lost in recent decades, especially in China, India, and Nepal (Jia and Sun, 2012;Liu et al., 2017). We doubt that Sapindus will experience a greater loss of diversity under future climate change. It is imperative to conserve the diversity of Sapindus, and natural populations at risk of destruction should be protected through in or ex situ conservation.  (Pal, Vaishnav, Meena, Pandey, & Singh, 2020) also simulated the adaptability and limiting factors of Sapindus emarginatus Vahl in the face of future climate regimes through two ecological niche models. Therefore, ecological niche model appear one of the optimal solutions to establishing how species will adapt to future climate scenarios. Plant niches are habitats with a minimum threshold necessary for a plant's survival (Barry, Moore, & Cox, 1980). The forest niche is strongly affected by the environment, and its niche changes or moves with environmental changes. The principle of ecological niche models is to infer the ecological needs of species through mathematical models based on occurrence data and environmental factors to draw a statistical or mechanistic model of their potential distributions (Araújo & Peterson, 2012;Elith & Leathwick, 2009;Zhu, Liu, Bu, & Gao, 2013). Bene tting from the simplicity of niche models and data accessibility (Merow et al., 2014), researchers can often estimate the potential habitat distribution and range shifts of species under future climate change. At present, the most commonly used niche models are the GARP (Stockwell, 1999), MaxEnt (Elith et al., 2006) , 2020) using the MaxEnt model. However, S. mukorossi is the most widely distributed species of Sapindus in Asia, and thus far, there are no reports exploring its habitat distribution using ecological niche models. Adopting ecological niche models to explore the distribution of Sapindus habitat distribution and their responses to future climate change is likely an effective strategy to address the conservation of Sapindus germplasm diversity.
In this study, we combined principal component analysis, the MaxEnt model, and ArcGIS for the three major Sapindus species in east and southeast Asia, namely, S. mukorossi, S. delavayi, and S. rarak. Based on 2,041 occurrence records, we drove models using current climate normals (averages between 1970 and 2000) and future climate projections (averages between 2021 and 2100) to investigate the habitat distribution and centroid shifts of S. mukorossi, S. delavayi, and S. rarak. The three objectives for this study were as follows: (1) to determine the spatial heterogeneity of the distribution of S. mukorossi, S. delavayi, and S. rarak; (2) to project suitable habitats for these three Sapindus species under current environmental conditions; and (3) to reveal the potential habitat redistribution patterns of these three Sapindus in response to future climate change, and identify hotspots of habitat degradation/expansion to facilitate climate change-adaptive biological conservation.

Study area
Our study area was located in east and southeast Asia, the natural range of S. mukorossi, S. delavayi, and S. rarak, which comprises 15 countries, including China, Afghanistan, Myanmar, Vietnam, Brunei, Cambodia, Indonesia, Malaysia, the Philippines, Japan, North Korea, and South Korea. The study area covered an area of 14,770,700 square kilometres, lying between 40°N latitude and 10°N latitude and between 90°E and 150°E longitude, and could be divided into climate zones of subtropical monsoon, temperate monsoon, tropical rainforest, and tropical monsoon. The annual mean temperature uctuated from 3.82°C in the Qinghai-Tibet Plateau to 27.60°C in the tropical region, and annual precipitation varied from 608 mm in the Qinghai-Tibet Plateau to 3685 mm in Indonesia, with elevations ranging from 1 to 3817 m.
There were some spatial clusters of occurrence records, particularly in China, Vietnam, and Japan. When such spatial clusters of occurrence exist, models often over t environmental biases and in ate model performance values (Veloz 2009; Hijimans et al. 2012; Boria et al. 2014). Therefore, we employed the spatially rated occurrence data tool of SDMtoolbox 2.0, and set the spatial interval to 10 km to eliminate spatial occurrence clusters. The elimination resulted in 419 occurrence records, which were subsequently employed in the models, which included 342 S. mukorossi, 41 S. delavayi, and 36 S. rarak occurrence records.

Environmental factors
In total, 32 environmental factors were applied to the projection, including bioclimatic, topographical, UV-radiation, and soil parameters, to model potential suitable habitat for Sapindus ( Table 1). The bioclimatic factors for current climate conditions were extracted from the 2.5 min resolution historical climate (averages for 1970-2000) database (https://www.worldclim.org/data/worldclim21.html). Topographical factors were obtained from the WorldClim dataset (https://www.worldclim.org/data/worldclim21.html). Soil factors were extracted from the Centre for Sustainability and the Global Environment dataset (https://nelson.wisc.edu/sage/). UV radiation variables were obtained from the glUV dataset (https://www.ufz.de/gluv/) (Beckmann et al., 2014).  depending on the amount of greenhouse gases emitted in the near future. All environmental factors were statistically resampled to 2.5-minute resolution using ArcGIS.

Parameterization and model evaluation
The MaxEnt model utilises the maximum entropy principle, applying ve different feature constraints (linear, product, hinge, quadratic, and threshold) to environmental variables to calculate the potential geographic distribution probability of species. We used the SDMtoolbox to carry out the MaxEnt model on the ArcGIS platform, with 25% of the occurrence data as testing data and 75% of the occurrence data as training data. To explore the key factors that shaped the habitat distribution of each species, we utilised a jackknife test with all model replications to rank the relative importance of the environmental factors, and the response curves of each factor were visualised. To calibrate and validate the robustness of the MaxEnt model evaluation, receiver operating characteristic curve (ROC curve) analysis was conducted. An area under the receiver operating curve (AUC) was employed to estimate the accuracy of the model predictions

Quantifying the magnitude and direction of habitat shifts
To quantify the magnitude of habitat distribution change in the projected distributions of each species under future climate scenarios, we compared the future habitat distributions to the initial distributions and classi ed pixels as (a) expansion, (b) unchanged, and (c) contraction (Brown, 2014). Subsequently, we calculated the expanded, unchanged, and contracted habitat areas under all future climate scenarios using ArcMap 10.5. Moreover, we calculated centroids for both current and future species distributions and employed these centroids to project a vector arrow to indicate the magnitude and direction of habitat shifts of each species using the SDMToolbox in ArcMap 10.5 (Brown, 2014).

Testing for Niche Divergence among Sapindus species
We used ArcMap 10.5 to obtain environmental factor data based on 32 environmental factor raster les for 419 occurrence data points from the three Sapindus. The species-level divergence associated with each of the environmental factors was examined using a nonparametric Kruskal-Wallis multiplerange test (Conover, 1980). Without a priori designation of species distribution, principal component analysis (PCA) was applied to the scaled data for the 32 environmental factors corresponding to all available occurrence data. The relative contribution of each environmental parameter to the formation of niche spaces was then illustrated using a PCA distance biplot (Hu et al., 2017). A nonparametric Kruskal-Wallis multiple-range test and PCA were implemented in the vegan and mvstats packages in R 3.6.3, respectively, and the biplots were constructed using a combination of the dvtools and ggplot packages in R 3.6.3.

Niche Divergence of S. mukorossi, S. delavayi, and S. rarak
The niche divergence of S. mukorossi, S. delavayi, and S. rarak for each of the 32 environmental parameters were analysed by the nonparametric Kruskal-Wallis rank sum test, and revealed that three species of Sapindus showed a signi cantly distinct adaptability range to environmental variables when compared with each other (p = 9.981e − 14 ). The PCA of the 32 environmental factors identi ed four components that collectively explained 79.85% (PC1:39.44%, PC2:17.30%, PC3:13.58%, and PC4:9.55%) of the observed variation among the 419 occurrences (Fig. 2). Variables with the highest loading scores on PC1 were Bio17, Bio14, Bio6, and Bio9, which are closely associated with precipitation and temperature. PC2 correlated strongly with soil pH and elevation. PC3 was closely associated with soil nutrients, while PC4 was associated with the highest UVb and precipitation of the warmest quarter. The 419 occurrences were clustered into three distinct environmental spaces in the Cartesian coordinates formed by the rst two principal components, and the clustering correlated closely with the three Sapindus species.
S. mukorossi, S. delavayi, and S. rarak exhibited signi cantly different ecological adaptations. S. mukorossi was the most widely distributed tree species and showed broad suitability. Compared with that of S. mukorossi, the ecological adaptation of S. delavayi indicated that the species preferred higher altitudes, higher mean diurnal range, and higher UV radiation. S. rarak showed a wider range of altitude adaptations and exhibited a higher demand for UV radiation and minimum temperatures.

Habitat distribution and key environmental factors of Sapindus under the current environment
Suitable distribution modelling for the three Sapindus species performed better than random distribution modelling, with testing AUC values ranging from 0.944 to 0.983, indicating that the models performed excellently in predicting suitable habitats in the current environment. We found that the predicted current suitable habitat distributions matched well with the actual ranges of the S. mukorossi, S. delavayi, and S. rarak, with overlaps found in the margins of neighbouring species distribution zones, such as in

Habitat distribution and key environmental factors of S. mukorossi
The MaxEnt model's internal jackknife test of factor importance showed that the precipitation of the warmest quarter (Bio18, 51.8 % of variation), isothermality (Bio 3, 18.9 % of variation), and net primary productivity (Npp, 15.6 % of variation) were the major contributors to the distribution model of S. mukorossi, with a cumulative contribution of 86.3% (Table 2). According to the MaxEnt results and environmental factor response curves, the ecological thresholds for the key environmental factors were 950-350 mm for the precipitation of the warmest quarter, 26-33 isothermality, and 1.02-1.08 kg/m 2 net primary productivity. The total habitat of S. mukorossi was 250.24 × 10 4 km 2 , and mainly located in the Jiangxi Province, Guangdong Province, Guangxi Zhuang Autonomous Region, Hainan Province, Taiwan in southern China, and the Kinki, Shikoku and Kyushu regions in Japan and Vietnam (Fig. 3). The area of suitable habitat was 77.28 × 10 4 km 2 , and the area of low suitability was 171.58 × 10 4 km 2 , accounting for 30.88% and 68.57% of the total habitat area, respectively (Fig. 6).
Areas of high suitability covered an area of only 1.38 × 10 4 km 2 (0.56%), concentrated in Hainan Province, local areas of Taiwan in China, and local areas in Japan.
The habitat distribution of S. delavayi was quite dissimilar to that of S. mukorossi. The habitat distribution of S. delavayi was highly concentrated in southwest China, with a total habitat area of 78.85 × 10 4 km 2 (31.51% of the habitat area of S. mukorossi) (Fig. 6). The areas of high suitability for S. delavayi were concentrated in the Sichuan Basin of the Sichuan Province, Kunming, Lijiang, Chuxiong, Qujing, and Zhaotong of Yunnan Province, and local areas in Linzhi City, Tibet, while areas of low suitability radiated to southern Shanxi Province, western Hubei Province, and local areas of the Guizhou, Henan, and Shandong Provinces (Fig. 4).

Potential distribution of three Sapindus species under future climate scenarios
By comparing the current suitable habitats (Fig. 3-5) with the projected suitable habitats from 2020 to 2100, we predicted the potential redistributions of S. mukorossi, S. delavayi, and S. rarak in response to 21st century climate change under four future climate scenarios (Fig. 7). The dynamics of the potential habitat areas of these three species exhibited different trends during the 21st century. Generally, our projections indicated that all species signi cantly differed in their habitat changes among the SSP126, SSP245, SSP370, and SSP585 scenarios, with some species dramatically expanding or contracting their habitat distributions.
Under the SSP126, SSP245, SSP370, and SSP585 scenarios, by 2100, the suitable habitat distribution of S. mukorossi showed a signi cant expansion toward higher latitudes and contraction in lower latitudes, with expansion occurring predominantly in the Sichuan Basin, Hubei Province, Hunan Province in China, and localised in southern Korea, and contraction occurring in Chongqing, Guangxi Zhuang Autonomous Region, Fujian Province, and Hunan Province in China occurred to Tibet, and only slightly occurred in the periphery of the Sichuan Basin ( Fig. 7; Figure S2). In contrast, the projected suitable habitat distribution of S. rarak underwent signi cant contraction and minor expansion. During 21st century (present day to 2100), the areas of suitable habitat distribution for S. rarak contracted from 42.73 × 10 4 km 2 to 72.30 × 10 4 km 2 , mainly in the northern Yunnan Province in China, eastern Myanmar, Thailand, Cambodia, Malaysia, Indonesia, and the Philippines ( Fig. 7; Figure S3). These changes were predicted to be more intensive under the SSP370 and SSP585 scenarios than under the SSP126 and SSP245 scenarios ( Fig. 7; Figure S3). Furthermore, climate change was predicted cause an expansion from 18.73 × 10 4 km 2 to 51.23 × 10 4 km 2 of suitable S. rarak habitat distribution in Mizoram and Manipur in India, northern Myanmar, Indonesia, and the Philippines (Fig. 7).

Habitat and spatial centroid shifts in the 21st century
The vectors between the present and future centroids indicated that the magnitudes and directions of the range shifts of these S. mukorossi, S. delavayi, and S. rarak differed under different future climate scenarios (Fig. 8). The geographical centroid of the potential habitat of S. mukorossi in the current scenario was currently located at 113.74°E, 27.64°N, in Liling County, Hunan Province, China (Fig. 8A)

Discussion
This study projected the current and future potential suitable habitats for three Sapindus species in east and southeast Asia using the MaxEnt model based on occurrence data sets with mean AUCs of 0.964 and 0.959 for current and future models, respectively. Therefore, we believe that our model performance is robust and adequate for construing the overall suitable habitat distribution of Sapindus. To the best of our knowledge, this is the rst study to analyse the suitable habitat distribution of S. mukorossi, S. delavayi, and S. rarak for the present and future using MaxEnt model.

Habitat distributions of S. mukorossi, S. delavayi, and S. rarak under the current environment
Sapindus is an ancient and widely distributed species; Engler (A. Engler, Melchior, & Werdermann, 1989) and Taylor (Taylor, 1990) found leaf fossils of Sapindus in Tertiary strata of North America. Xia (Xia, 1995) stated that Sapindus was widely distributed and highly differentiated from the Eocene to the Tertiary. Long periods of natural selection and geographical isolation have allowed Sapindus to differentiate and adapt in their respective regions, eventually forming their own unique environmental suitability and species-speci c physiological tolerance. Liu  In this study, we demonstrated that there were signi cant variations in the ecological suitability among S. mukorossi, S. delavayi, and S. rarak through PCA and Kruskal-Wallis multiple-range tests. The MaxEnt model results indicated that precipitation and net primary productivity were common key environmental factors affecting the distribution of S. mukorossi, S. delavayi, and S. rarak (Table 2). Therefore, we suggest that precipitation and net primary productivity are essential requirements for the survival of Sapindus. However, when accounting for additional factors, apparent differences exist in the importance of elevation, minimum temperature, and solar radiation between S. mukorossi, S. delavayi, and S. rarak. Minimum temperature (-2°C to 3°C) and elevation (1200-2000 m) were the secondary critical environmental factors in determining the suitable habitat distribution of S. delavayi, and solar radiation (annual mean Uvb ranged between 4600 and 5000 J/m 2 /day) was the secondary critical environmental factor for S. rarak ( Table 2). The results of the biplot (Fig. 2) in the PCA supported these ndings. Based on the genetic structure and geographical variation of Sapindus germplasms, Sun (Sun et al., 2018) found that areas of high altitudes and low minimum temperatures are suitable for the growth of S. delavayi. Ashish Kumar Pal (Pal et al., 2020) found that maximum temperature and annual precipitation were the most critical environmental factors determining the distribution of suitable habitats for S. emarginatus through the MaxEnt model. These ndings also demonstrate that precipitation is a prerequisite for suitable habitat distribution of Sapindus, and that S. emarginatus similarly diverges into signi cantly different ecological suitability.
Our study indicated that S. mukorossi exhibited the widest distribution of suitable habitats, followed by that of S. rarak, and S. delavayi, concentrated in the high-altitude areas of southwest China. Sun (Sun et al., 2016) and Cai (Cai et al., 2018) found that S. delavayi has larger seed kernels than S. mukorossi and, thus, more suitable for cultivation as a biodiesel species. Therefore, it is recommended that S. delavayi biodiesel forests should be developed in the Sichuan and Yunnan Provinces in China. Suitable habitats of S. mukorossi were widely distributed throughout southern China, southern Korea, and southern Japan (Fig. 3), whereas the suitable habitat areas for S. rarak were widely distributed in tropical and subtropical regions. Sun  found that S. mukorossi shows strong breeding potential, and Liu (J.  found that S. rarak exhibits strong biochemical potential for saponins. Combined with the MaxEnt results from this study, we suggest that S. mukorossi could be developed and cultivated in the Fujian, Jiangxi, Guangdong, and Guangxi Provinces in China, and S. rarak could be cultivated in the Yunnan Province in China, Thailand, and Myanmar.

Response of suitable habitat distribution to future climate change
Climate is the most critical ecological factor for plants, and changes in plant distribution are the clearest and most direct responses to climate. Future climate change may substantially change the structure and function of terrestrial ecosystems, resulting in signi cant changes in the extent and distribution of biological habitats (Barry et al., 1980;Lawler, 2009). The results of this study indicate that S. mukorossi, S. delavayi, and S. rarak display signi cantly distinct ecological adaptations. Therefore, they are also predicted to exhibit dramatically divergent responses in the face of future climate change. The habitat distribution of S. mukorossi will expand and contract in future climate change, with a trend towards expansion at higher latitudes and contraction at lower latitudes, with the centroid moving mainly eastward and southward. We hypothesised that rising minimum temperatures and better water circulation at higher latitudes and mid-altitudes (Gao, Pan, & Zhang, 1991), caused by climate change, will allow S. mukorossi to expand into these areas. Clark (Clark & D., 2004) indicated that average temperature and rainfall were generally negatively correlated in the tropics, and de la Peña and Hughes stated (Pea & Hughes, 2007) that climate change will lead to more erratic rainfall patterns and unpredictable high temperature spells in the tropics. In the face of future climate change, extreme heat and erratic rainfall and drought at low latitudes will likely lead to localised areas that are no longer suitable for the distribution of S. mukorossi.
S. delavayi showed signi cantly different distribution trends in response to climate change compared to that of S. mukorossi. Bene ting from future climate change, a large number of currently unsuitable areas in Sichuan and Yunnan Provinces in China are predicted to potentially become suitable for S. delavayi, with the centroid shifting south-eastward ( Fig. 7; Fig. 8). We speculate that increased precipitation caused S. delavayi to expand outward from its original habitats; however, elevation remains a key ecological factor in determining the habitat distribution of S. delavayi (Sun et al., 2018), which prevents S. delavayi from expanding to lower elevations.
Interestingly, S. rarak exhibited a signi cant contraction trend in distribution, mainly in the northern Yunnan Province in China, eastern Myanmar, Thailand, Cambodia, Malaysia, Indonesia, and the Philippines, with the most signi cant contraction occurring in the SSP370 and SSP585 climate scenarios ( Fig. 7;  Fig. 8 Whether a species bene ts or suffers under anthropogenic climate change depends on geographical location, ecological niches, and the migration ability of the species, which provide essential insights for spatial conservation assessments of future biodiversity hotspots. In the face of rapid climate change, it is unrealistic for plants to evolve in correlation with physiological adaptation strategies in short periods; therefore, migration or dispersal ability is a prerequisite for their survival (Richard T Corlett, 2009;Liao et al., 2020). This study found that S. mukorossi, S. delavayi, and S. rarak exhibited varying degrees of expansion under future climate change scenarios, especially in S. delavayi (Fig. 7). However, we doubt that these species will be able to migrate and spread smoothly into suitable habitats within 100 years. Corlett  In contrast, Guangdong province, Guangxi Zhuang Autonomous Region, Fujian Province, Hainan Province, and Taiwan in China, which will be relatively less affected by climate change, could be used as a base for resource conservation, breeding, large-scale cultivation, and utilisation of S. mukorossi in the future. The Sichuan Basin, Kunming, and Qujing City in the Yunnan Province, China can serve as a base for S. delavayi, and the Yunnan Province, China, eastern Myanmar, northern Thailand, and northern Laos can be used as a base for S. rarak.

Uncertainties
In this study, we have likely developed the largest compilation of occurrence records (2041 in total) for S. mukorossi, S. delavayi, and S. rarak to date; however, the sample size is still relatively small compared to the large study area. Moreover, although climatic, soil, topographical, and solar radiation factors were considered in this model, species habitats are also constrained by other factors, including migration ability, adaptive capacity, inter-species interactions, human activities, and land use. Therefore, the actual habitat distributions of S. mukorossi, S. delavayi, and S. rarak may be smaller than our model-predicted potential habitat distributions. However, introducing all variables into the model may lead to more collinearity problems, and the effects of key variables may be weakened; thus, the model results may not be more accurate than the current model are. Nevertheless, the modelled current habitat distributions of the three Sapindus species investigated in this study matched well with current observations of Sapindus distribution patterns (J. Sun et al., 2018) and accurately re ected the ecological adaptation differences among these three species. We are con dent that the MaxEnt model used in this study correctly predicted the potential habitat distribution of these three Sapindus species within the context of climate change. We also believe that our results provide an important theoretical basis and recommendations for the conservation and sustainable exploitation of Sapindus genetic diversity.

Conclusions
We rst demonstrated that there were signi cantly different ecological adaptations among S. mukorossi, S. delavayi, and S. rarak in east and southeast Asia.
Climate niche models showed that precipitation may play an important role in framing the potential habitats of Sapindus; however, there were signi cant ecological adaptive divergences among them. S. mukorossi exhibited the widest range of adaptation. Compared to that of S. mukorossi, S. delavayi was more sensitive to minimum temperatures and elevation, and S. rarak was more demanding in terms of solar radiation.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Competing interests
The authors declare that they have no competing interests Availability of data and materials The datasets generated during and/or analysed during the current study are available in the Chinese National Plant Specimen Resource Center, Global Biodiversity Information Facility, Chinese National Specimen Information Infrastructure, WorldClim, Centre for Sustainability and the Global Environment, glUV repository.  Spatial distribution of occurrence records of S. mukorossi, S. delavayi, and S. rarak. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 3
Suitable habitat distribution of S. mukorossi under the current environment. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 4
Suitable habitat distribution of S. delavayi under the current environment. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 5
Suitable habitat distribution of S. rarak under the current environment. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 6
Distribution of suitable habitats areas for S. mukorossi, S. delavayi, and S. rarak under the current environment. Figure 7