Separation of the Bioclimatic Spaces of Himalayan Tree Rhododendron Species Predicted by Ensemble Suitability Models

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Elsevier - Publisher Connector

Global Ecology and Conservation 1 (2014) 2–12

Contents lists available at ScienceDirect

Global Ecology and Conservation

journal homepage: www.elsevier.com/locate/gecco

Original research article Separation of the bioclimatic spaces of Himalayan tree rhododendron species predicted by ensemble suitability models Sailesh Ranjitkar a,b, Roeland Kindt c, Nani Maiya Sujakhu a,d, Robbie Hart e, Wen Guo a,d, Xuefei Yang a,∗, Krishna Kumar Shrestha f, Jianchu Xu a,b,∗∗, Eike Luedeling c

a Key Laboratory of Plant Diversity and Biogeography of East Asia, Kunming Institute of Botany, Kunming 650204, China b World Agroforestry Centre East and Central Asia, Kunming 650204, China c World Agroforestry Centre, United Nations Avenue, Gigiri, 30677, Nairobi, 00100, Kenya d University of Chinese Academy of Sciences, Kunming, China e Missouri Botanical Garden, 4344 Shaw Blvd., St. Louis, MO 63110, USA f Central Department of Botany, Tribhuvan University, Kirtipur, Kathmandu, Nepal

article info a b s t r a c t

Article history: The tree rhododendrons include the most widely distributed Himalayan Rhododendron Received 1 April 2014 species belonging to the subsection Arborea. Distributions of two members of this sub- Received in revised form 3 July 2014 species were modelled using bioclimatic data for current conditions (1950–2000). A sub- Accepted 4 July 2014 set of the least correlated bioclimatic variables was used for ecological niche modelling Available online 23 July 2014 (ENM). We used an ENM ensemble method in the BiodiversityR R-package to map the suitable climatic space for tree rhododendrons based on 217 point location records. Ensem- Keywords: ble bioclimatic models for tree rhododendrons had high predictive power with bioclimatic Rhododendron arboreum Distribution variables, which also separated the climatic spaces for the two species. Tree rhododen- BiodiversityR drons were found occurring in a wide range of climate and the distributional limits were Consensus method associated with isothermality, temperature ranges, temperature of the wettest quarter, and Ensemble model precipitation of the warmest quarter of the year. The most suitable climatic space for tree Hindukush–Himalaya–Hengduan Mountain rhododendrons was predicted to be in western Yunnan, China, with suitability declining towards the west and east. Its occurrence in a wide range of climatic settings with highly dissected habitats speaks to the adaptive capacity of the species, which might open up future options for their conservation planning in regions where they are listed as threatened. © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

1. Introduction

Since habitat suitability of plants and animals is strongly determined by climate (Marino et al., 2011; Pearson and Dawson, 2003), niche models, driven by climate variables, have emerged as primary tools for projecting climate change impacts on species ranges (Elith et al., 2006; Hijmans and Graham, 2006; Hill et al., 2012). Species are assumed to occur within a certain climatic space determined by the climatic needs of the species (Trivedi et al., 2008). Climate envelope models characterize

∗ Corresponding author. Tel.: +86 871 65223909. ∗∗ Corresponding author at: World Agroforestry Centre East and Central Asia, Kunming 650204, China. Tel.: +86 871 65223355. E-mail addresses: [email protected] (X. Yang), [email protected] (J. Xu).

http://dx.doi.org/10.1016/j.gecco.2014.07.001 2351-9894/© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/3.0/). S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12 3

Fig. 1. Distribution range of the subsection Arborea (section Ponticum and genus Rhododendron) in the Himalayas and the Hengduan Mountains, Southeast and South Asia. The inset shows the altitudinal distribution of Rhododendron arboreum (R. arb) and R. delavayi (R. del); Locations were obtained from herbarium collections of tree rhododendron specimens. R. delavayi was described in West Yunnan (YW), Central Yunnan (YC) and East Yunnan (YE); R. arboreum was described in the remaining locations [Central Himalayas—CH, Eastern Himalayas—EH, Meghalaya—MG, Northeast India and adjoining hills of Myanmar—MM, South India—SI, Sri Lanka—SL, and Western Himalayas—WH]. these climatic needs (Bakkenes et al., 2002; Trivedi et al., 2008), and classify locations on the map according to climatic suitability for the species. Climate envelope approaches are rooted in ecological niche theory (Grinnell, 1917; Guisan and Thuiller, 2005; Hutchinson, 1957), and they are increasingly being applied in ecological and evolutionary research (Austin, 2007; Elith and Leathwick, 2009; Sagarin et al., 2006), as well as for studies in the fields of biogeography and biological conservation (Araújo et al., 2004; makowsky et al., 2010). Understanding the ecological needs of keystone species, such as rhododendrons (Singh et al., 2009) in the Himalayas, is of critical importance for ecosystem conservation. Among Rhododendron, the subsection Arborea of the section Ponticum and the subgenus Hymenanthes (Chamberlain, 1982) contains only a few species. Two species in this subsection, Rhododendron lanigerum Tagg and R. niveum J.D. Hooker, are endangered/threatened (Fang et al., 2005) species with small distributional ranges, and they are not included in the present work due to insufficient data. Some literature (sensu Fang et al., 2005) divided the remainder of the subsection Arborea into two species, R. arboreum Smith and R. delavayi Franchet, while some includes delavayi as one of the sub-species of R. arboreum (sensu Chamberlain, 1982). Five lower taxa of R. arboreum have been listed, three from the Himalayas (including subsp delavayi) and two from tropical highlands (Giriraj et al., 2008; Long, 1991; Press et al., 2000; Sekar, 2010). Because of the wide distributional range and the complex infra-generic taxonomy, we summarily refer to this taxonomic group as ‘‘tree rhododendrons’’, where appropriate in this paper. Tree rhododendrons occur in Southern China, Myanmar, India, Bhutan, Nepal, and Northern Pakistan (R. arboreum subsp arboreum and R. arboreum subsp cinnamomeum) at elevations between approximately 1000 and 3800 masl (Fig. 1)(Chamberlain et al., 1996; Fang et al., 2005; Long, 1991; Press et al., 2000; Sekar, 2010). In addition, isolated populations, possibly representing remnant vegetation from glacial periods of the Pleistocene (Burkill, 1924; Giriraj et al., 2008), exist in mountainous regions of South India (R. arboreum subsp nilagiricum) and Sri Lanka (R. arboreum subsp zeylanicum). Tree rhododendrons are an important component of Himalayan ecosystems. They fulfil multiple functions within ecosystems throughout the mountain range. They grow in a wide range of habitats, including steep areas of high rainfall and acidic soils, and so help to stabilize slopes and maintain watershed functions (Cox, 1990; Gibbs et al., 2011). This widely- distributed species likely plays a major role in sustaining a wide-range of insect and bird pollinator populations due to its profuse flowering and nectar provision from early spring to early summer (Ranjitkar et al., 2013; Singh et al., 2009). Tree rhododendrons also serve as very important habitat components for several endangered species, including the red panda Ailurus fulgens (Dorji et al., 2011), the Himalayan musk deer Moschus chrysogaster, the Asian black bear Ursus thibetanus (Rai et al., 2012), and the cheer pheasant Catreus wallichii (Jolli et al., 2012). 4 S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12

Table 1 Bioclimatic variables used for calibrating ecological niche models. Abb. Variable Abb. Variable

Bio1 Mean annual temperature Bio12 Total (annual) precipitation Bio2 Mean diurnal range (mean of max temp–min temp) Bio13 Precipitation of wettest month Bio3 Isothermality (Bio2/Bio7) (* 100) Bio14 Precipitation of driest month Bio4 Temperature seasonality (standard deviation *100) Bio15 Precipitation seasonality (coefficient of variation) Bio5 Max temperature of warmest month Bio16 Precipitation of wettest quarter Bio6 Min temperature of coldest month Bio17 Precipitation of driest quarter Bio7 Temperature annual range (Bio5–Bio6) Bio18 Precipitation of warmest quarter Bio8 Mean temperature of the wettest quarter Bio19 Precipitation of coldest quarter Bio9 Mean temperature of driest quarter AI Annual aridity index Bio10 Mean temperature of warmest quarter PET Annual potential evapotranspiration Bio11 Mean temperature of coldest quarter Abb.—abbreviation.

In addition to their ecological value, tree rhododendrons are of cultural importance to local people. As an edible, medicinal and sacred species, they are an important part of traditional knowledge for many ethnic groups living in the Sino-Himalayan region (Bhattacharyya, 2011; Cox, 1990; Dolui et al., 2004), while trunks and branches of rhododendrons are harvested extensively for fuel, charcoal and timber (Chettri et al., 2002; Ranjitkar et al., 2014a; Schmidt-Vogt, 1990). Yet human- induced land transformation and excessive harvesting for firewood and timber has resulted in the loss of habitat and reduction in the population of tree rhododendrons—causing the species to be included on the red-list of threatened species maintained by the International Union of Conservation of Nature (IUCN) (Gibbs et al., 2011). They are listed as endangered (EN) in southern India (Gibbs et al., 2011), as vulnerable in Sri Lanka (MOE, 2012) and northeast India (Singh et al., 2009), and as threatened in the north-western part of its native range, including parts of India and Pakistan (Haq, 2011). In some regions; however, there is no particular concern about the conservation of tree rhododendrons (Gibbs et al., 2011). Species with superior capacity to disperse, ability to migrate fast enough to keep pace with climate change, high competitiveness with already-established vegetation and adaptability to a wide-range of habitat conditions have shown the ability to occupy large areas, often as invasive species (Feeley and Silman, 2010; Normand et al., 2011; Sexton et al., 2009). Tree rhododendrons (particularly R. arboreum) must share some of these characteristics, since it is the only species out of hundreds of Rhododendron species that occurs throughout the entire Himalayan mountain range, as well as in isolated temperate regions in South India and Sri Lanka. On the other hand, tree rhododendrons show mass foliar damage due to sudden change in temperature (Rai et al., 2012) and a tendency to flower earlier with rising temperature (Ranjitkar et al., 2013), which indicates their sensitivity to climatic influences. Therefore, the study of the climatic space occupied by keystone species, such as tree rhododendrons, is important for gaining insights into the response of temperate Himalayan species to climate change and related conservation issues. In this study, we examine the general biogeography of tree rhododendrons, regarding the climatic needs of two species. Our model aims to describe climatic suitability, which would have to account for various contributing factors for which no suitable data are available. Specific objectives of this investigation are (1) to identify key climatic constraints and determine the climatic space for tree rhododendrons, (2) to separate the climatic space for two widely-distributed tree rhododendrons, R. arboreum and R.delavayi, which are believed to have niche overlap, and (3) to describe dispersal of tree rhododendrons in relation to climate, including possible barriers, corridors, and conservation issues.

2. Material and methods

2.1. Bioclimatic data

Bioclimatic data layers of the WorldClim Dataset (www.worldclim.org/bioclim)(Hijmans et al., 2005), as well as the Global Aridity Index and Global Potential Evapo-Transpiration data from the CGIAR-CSI (www.csi.cgiar.org) were used as predictors of tree rhododendron distribution (Table 1). All datasets used had a spatial resolution of 30 arc-seconds.

2.2. Species location data

We compiled location data of individual tree rhododendrons from herbarium specimens housed in the National Herbarium and Plant Laboratory in Kathmandu, Nepal, and the Kunming Institute of Botany, China. In addition, online databases of the Chinese Virtual Herbarium, the Royal Botanical Garden at Edinburgh, United Kingdom, and the Herbarium at the University of Tokyo, Japan, were used as data sources (Appendix A). Field notes from our previous work were also included as sources of occurrence data. Specimen positions were verified using local flora, scientific publications (Appendix B), and field observations. We excluded records with obvious geocoding errors, as well as textual location references without coordinates. We also excluded repeated locations and locations that occurred in the same cell of a grid, with a spatial resolution of 10 min, to reduce spatial autocorrelation due to clustering of collection activity at certain places. Out of more S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12 5 than 1200 records throughout the Sino-Himalayan and Southern regions of distribution, 152 R. arboreum and 65 R. delavayi records that were collected after 1950, and were used for building climatic envelope models (see Fig. A.1 in Appendix A).

2.3. Selection of predictor variables

A subset of the bioclimatic indicators was chosen as explanatory variables. Variables in this subset should not be highly correlated, so we used the Variance Inflation Factor (VIF) to select a suitable set of predictors (Fox and Weisberg, 2011; Ranjitkar et al., 2014b). We used the VIF, in combination with a correlation matrix, to summarize climatic information associated with tree rhododendrons locations. VIF and correlation coefficients were calculated using BiodiversityR (Kindt, 2014), a package for biodiversity analysis for the R programming language (see Table C.1 in Appendix C for various packages used within BiodiversityR). The VIF is used as an indicator of multi-collinearity, and identifies sets of predictors that are highly correlated. We applied the frequently-used VIF threshold of 5 (Rogerson, 2001) for deciding which predictors to accept for further analysis. The ensemble test was run with location and pseudo-absence points, which were split into 75% calibration and 25% test data, and were performed until a set of predictors with VIF values <5 and Pearson correlation values <0.8 were obtained (see Tables D.1–D.4 in Appendix D).

2.4. Modelling process in BiodiversityR

A three-step model, run in the BiodiversityR package, was performed to prepare for consensus mapping of species distributions. Consensus mapping is a mapping technique that is based on an ensemble of several niche modelling algorithms, rather than on a single approach. This strategy has been shown to increase the accuracy of species distribution predictions over single-algorithm procedures, and it has great potential for applications in conservation biology and biogeography (Marmion et al., 2009). The first step was the calibration of ENM algorithms, which are listed in Table 2. In the BiodiversityR package, sets of functions were developed for a more-direct handling of these model formulae for ensemble output (Kindt, 2014). Function (ensemble.test.spilts) in the package was used for calibration that uses the tuning procedure. The tuning procedure uses an internal cross-validation procedure, in which the calibration dataset is split into k subsets. Subsequently, the Area Under the Receiver Operator Curve (AUC) values of the subsets are used to determine the appropriate weights (range between lowest 0 to highest 1) for the ensemble model (Kindt, 2014). For each focal species, this test run was repeated multiple times, resulting in conclusive average weights for modelling algorithms. A second calibration procedure (ensemble.test function) was applied where modelling algorithms with weights >0.05 (default value) were retained. In the second calibration, 10 internal test runs and 4-fold cross-validations resulted in final weights for the selected algorithms, which were used to produce the final ensemble output. In addition, the tuning procedure explored how differences in weights of the sub-models resulted in changes in the AUC of the ensemble model, and it selected the weights that result in the greatest accuracy of the ensemble model. The automated process in BiodiversityR uses weighted average (wmod) of individual algorithms (Pmod) to calculate the ensemble model (Pensemble) using the following formula:  (w P ) = mod mod Pensemble  . wmod The third step included preparation of consensus raster layers using function (ensemble.raster) that combined modelling algorithms with positive weights into ensemble outputs (Kindt, 2014). Only the most successful algorithms, according to the AUC criterion, were used for constructing the ensemble prediction. The first consensus layer summarized the ensemble output that indicated the presence of focal species for each grid cell, based on a threshold defined by maximizing the sum of the true presence and true absence rates. Other layers produced included a presence–absence (1–0) map and a consensus suitability map based on the number of individual suitability models that predict the presence of the focal organism.

2.5. Ensemble forecast of climatic space

The aforementioned process was slightly modified for the ensemble forecast of tree rhododendron’s climatic space in the present work. In the first calibration step, we split the presence and background dataset geographically, where we calibrated 18 algorithms (Tables 2 and3) for predicting tree rhododendron’s habitat. The Rhododendron arboreum distribution was split into five sub-regions, four in the Himalayas and one in the southern region of the distribution. Similarly, the R. delavayi distribution was split into two sub-regions, the western and eastern part of the species’ range. Following Ranjitkar et al. (2014b) we used 4-fold cross-validation, where the presence and background data were divided into 75% calibration and 25% evaluation subsets for each sub-region. Weight values obtained from each run for each sub-region were again averaged to obtain final weights for the region. Modelling algorithms with weights >0.04 were selected for calibration and modelling in the region (Table 3). Consensus mapping and statistics of replicate runs were used as final products to describe the climatic space for tree rhododendrons. 6 S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12

Table 2 Species distribution modelling algorithms available in current version of BiodiversityR to build ensemble model. Class of model Method Description of sub-model as used in BiodiversityR

Envelope model BIOCLIM It computes the similarity of a location by comparing the fi(xi) at any location to a percentile distribution of the values at known locations of occurrence. Multivariate distance DOMAIN It computes the Gower distance between environmental variables at any location and those at any of the known locations of occurrence. Regression: Multivariate adaptive MARS It is a non-parametric regression technique that automatically models regression splines non-linearities and interactions between variables. Flexible discriminant analysis FDA It is a supervise classification method. The method combines different models for multi-group non-linear classification. Additive models: Generalized additive GAM Generalized additive models are a semi-parametric approach to predicting models non-linear responses to a suite of predictor. The method uses the backfitting algorithm that provides a very general modular estimation method capable of

using a wide variety of smoothing methods to estimate the fi(xi). Stepwise GAM GAMSTEP Builds a GAM model in a step-wise fashion. Mixed GAM Computation Vehicle MGCV It provides functions for generalized additive and generalized additive mixed modelling, while the estimation of models is based on the smooth functions using penalized regression splines. MGCV with fixed d.f. regression splines MGCVFIX It uses a GAM technique with fixed degree of freedom regression splines. Boosted regression models: GBM GBM is based on prediction components, where each component consists of a Generalized boosted regression models different weighted sum of nonlinear transformations of the predictor variables. Each prediction component is fitted to the residuals of the previous component in the model. Stepwise boosted regression tree models GBMSTEP It is a technique that aims to improve the performance of a single model by fitting many models based on stepwise selection and combining them for prediction. Generalized linear models GLM GLM generalizes linear regression by allowing the linear model to be related to the response variable via a link function and by allowing the magnitude of the variance

of each measurement to be a fi(xi). Stepwise generalized linear models GLMSTEP It includes regression models in which the predictive variables are selected by an automated algorithm that involves backward elimination or forward selection. Maximum entropy MAXENT It is a machine-learning method that estimates the species distribution probability by assessing the maximum entropy distribution, so that the most spread-out, or closest to uniform. Artificial neural networks NNET It is a machine learning approach that employs an adaptive structure, which can be trained with application data to capture complex relationships between input and out variables. Random forests RF It is a collection of tree-structured weak learners that comprised identically distributed random vectors where each tree contributes to a prediction. Recursive partitioning and regression RPART It is a simple nonparametric regression approach, where the space spanned by all trees predictor variables, is recursively partitioned into a set of rectangular areas. The partition is created such that observations with similar response values are grouped and a constant value of the response variable is predicted within each area. Support vector machines SVM Machine-learning methods that are based on classification (C-svc, nu-svc), novelty detection (one-class-svc), and regression (eps-svr, nu-svr). SVME It is used to train an SVM and carry out general regression and classification (of nu and epsilon-type), as well as density-estimation.

fi(xi) = function of predictor variable ‘i’.

The ensemble model of species distribution assigned quantitative scores for habitat suitability for all pixels of the analysis grids. The model output suggests the cut-off point, above which suitable climatic space for the focal species can be found. We classified all pixels in the consensus map output according to this criterion, such that all pixels with suitability scores above the cut-off point were included in the species’ bioclimatic space. All steps involved in the modelling procedure were executed using the BiodiversityR (ver. 2.4-1) package within the R 3.0.2 environment, and the final distribution map was prepared using ArcGIS 10 (see Appendix A).

3. Results

3.1. Ensemble model

Stepwise elimination of the bioclimatic variables with VIF values >5 resulted in a set of least correlated predictor variables. Seven bioclimatic variables were used in predicting the climatic space for R. arboreum, while six were identified for R. delavayi. Mean weights obtained from split test runs at sub-regions, and the average weight used in the final model run, are tabulated in Table 3. Among sub-regions, differences were found for the contributions of these variables (Fig. 2), as well as for the number of algorithms used in the ensemble modelling (Table 3). Models based on the combination of these variables provided good fits to the data in regional-level prediction. Mean AUC values of the algorithms selected for the ensemble were high and fairly similar, with a mean AUC of the ensemble model of 0.98 for both species. The threshold, indicating the cut-off point, below which climatic space was found to be unsuitable, S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12 7

Table 3 Species distribution modelling algorithms calibrated, weight for each algorithm from each sub-region and values in bold represent weight reassigned for final model calibration. Algorithm/sub-models R. arboreum R. delavayi FW W C E S Mean Final WY EY Mean Final

BIOCLIM 0 0 0 0 0.032 0.006 0 0 0 DOMAIN 0 0 0 0 0 0 0 0 0 EARTH* 0.069 0.067 0.051 0.083 0.118 0.078 0.083 0 0.025 0.013 FDA 0.025 0 0 0 0 0.005 0 0.016 0.008 GAMĎ 0.134 0.139 0.139 0.134 0.138 0.136 0.145 0.146 0.134 0.14 0.147 GAMSTEP# 0 0 0.043 0.014 0.022 0.016 0.033 0.059 0.046 0.048 MGCV* 0.018 0.092 0.092 0 0.016 0.043 0.046 0 0.015 0.008 MGCVFIXĎ 0.094 0.117 0.111 0.092 0.121 0.107 0.114 0.132 0.098 0.115 0.121 GBMĎ 0.154 0.141 0.144 0.151 0.1 0.138 0.147 0.161 0.159 0.16 0.168 GBMSTEPĎ 0.151 0.137 0.141 0.149 0.131 0.142 0.151 0.157 0.156 0.156 0.164 GLMĎ 0.103 0.109 0.118 0.088 0.111 0.106 0.113 0.122 0.086 0.104 0.109 GLMSTEP 0 0.019 0 0 0 0.004 0 0 0 MAXENT 0 0 0 0 0 0 0 0 0 NNET 0.05 0 0 0.025 0.026 0.02 0 0.014 0.007 RFĎ 0.154 0.142 0.146 0.152 0.143 0.147 0.157 0.163 0.16 0.162 0.17 RPART* 0.015 0.037 0.015 0.096 0.044 0.041 0.044 0 0.025 0.012 SVM 0 0 0 0 0 0 0 0 0 SVME# 0.034 0 0 0.017 0 0.01 0.085 0.053 0.069 0.073 Null values (in ‘Mean’ columns) indicate algorithms not included for determination of weight in first step *Algorithms used in ensemble for predicting climate space of R. arboreum; Ď Algorithms used in ensemble for predicting climate space of both species; #Algorithms used in ensemble for predicting climate space of R. delavayi FW—Far western Himalayas, W—Western Himalayas, C—Central Himalayas, E—Eastern Himalayas, S—Southern Himalayas, WY—Western Yunnan, and EY—Eastern Yunnan.

Fig. 2. The climatic space for tree rhododendron throughout South and Southeast Asia, based on geographically-divided calibration. Grey lines indicate longitudinal division based on occurrence of the species in the Himalayas and the Hengduan Mountain region. was 0.068 for R. arboreum and 0.075 for R. delavayi. Fig. 3(a) indicated a presence–absence (0–1) map, based on these cut- off values. The species occurrence area in Fig. 3(a) was equivalent to the suitable range predicted with ensemble of seven algorithms, as shown in the consensus map (Fig. 3(b)). 8 S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12

Fig. 3. Consensus map layer produced as a result of the ensemble modelling, based on average weights obtained from geographically divided calibration: (a) Presence of two focal species within the area of prediction, (b) number of algorithms used in the ensemble modelling and (c) forecast of climatic space for the two focal species.

3.2. Bioclimatic space

Our predictions show suitable climate for tree rhododendrons between the Indus River Basin in the west and the Yangtze River Basin in the east (Fig. 3(c)). No suitable climatic space was predicted on the Tibetan Plateau to the north of the Himalayas, restricting the entire space to the southern part of the mountain range. The eastern regions (including the Hengduan Mountain region) were predicted to be most suitable for tree rhododendrons (Fig. 3). The temperature ranges (both annual and diurnal) were important in this region (Fig. 2), where climatic space of both species overlapped, although R. arboreum occurrence was not confirmed with herbarium data or literature towards southern Yunnan (south to 27° latitude). Suitable climatic conditions for R. delavayi were identified in the highlands of Yunnan (central and western Yunnan and in the Kachin state in Myanmar). The Yangtze River Basin, and associated lowlands, separate the climatic space for this species in the western Yunnan highlands to the eastern Yunnan–Guizhou Plateau (Fig. 3). Three parallel rivers – the Mekong, the S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12 9

Salween and the Irrawaddy – divide the bioclimate space of the two Rhododendron species. East of the Mekong river, only R. delavayi was reported while R. arboreum has been confirmed to west of the Salween (Fig. 3(c)). From western Yunnan and Kachin (i.e. west of the Salween), a narrow band of climatic space for R. arboreum extends along the mountains to the north of the Brahmaputra River Basin, westward along the Himalayan Mountains and into Northern Pakistan. The Brahmaputra Basin and associated lowlands were identified as barriers that divided the climatic space of tree rhododendrons. Directly to the south of the Brahmaputra Basin, suitable climatic space was predicted in Meghalaya (Northeast India). Towards the southeast, suitable areas extended into Chin State (Myanmar), the hilly regions of Nagaland, Manipur and Mizoram in Northeast India, and the highlands of Thailand. Temperature variables were also important in this region. Suitability continued throughout the Himalayas, with high mountains, deep gorges, and river basins constituting barriers to dispersal. Many river valleys, especially in the upper catchments, connected suitable habitats. This pattern of coalition and disjunction of climatic space for tree rhododendrons was noticed throughout the Himalayas. Major river basins and river systems that interrupted the continuous distribution of these species are shown in Fig. 3. Compared to the Eastern Himalayas, habitat suitability was lower in the Western Himalayas where precipitation variables were found to be more important than temperature. Small, isolated patches of climatic space were also predicted in South India and Sri Lanka, in tropical montane areas.

4. Discussion

4.1. Performance of the modelling procedure

On a regional scale, the climatic space for tree rhododendrons identified by our ecological niche model corresponded closely to the actual distribution of the species throughout the Himalayas, the Hengduan Mountains, and other parts of Southeast and South Asia. However, the actual habitat is likely smaller than what is indicated by our models, because climatic variables are not the only determinants of habitat suitability. Edaphic and bio-geographic factors limit the distribution of species, even in areas with suitable climate. Seed dispersal mechanisms, the ability of a species to colonize a new territory, presence of existing vegetation and human-induced land transformations also confine a species’ actual distribution (Nor- mand et al., 2011; Sexton et al., 2009). On the other hand, when modelling is carried out in a large geographical area, climate is typically considered the most important determinant of species occurrence (Pearson and Dawson, 2003), and it provides basic information on suitable habitat for the species (Marino et al., 2011). An ensemble approach can improve the accuracy of species distribution predictions (Marmion et al., 2009), which makes us confident that the approach taken in this study provided robust habitat predictions.

4.2. Climatic space for tree rhododendrons

Tree rhododendrons are responsive to temperature, and it is an important delimiting factor. Cold temperatures, in particular, limit the distribution of a number of species in the Himalayas (Ranjitkar et al., 2013; Vetaas, 2002). Our model diagnosed a strong response of tree rhododendrons to temperature, particularly to temperature ranges and isothermality throughout the distribution range, and ‘‘temperature during the wettest quarter’’ in the Himalayas for delimiting the climatic space of R. arboreum. Nonetheless, while the contribution of temperature was important for the bioclimatic space, our model also diagnosed a strong response to precipitation variables. Warm temperatures affect moisture availability, and sufficient rainfall is necessary to overcome the moisture stress. This could be essential for defining the suitable habitat of tree rhododendrons. The same might be true for many temperate trees, as several studies have shown that, near the warmer edge of their suitable range, such trees grow best where soil moisture is high (Bonan and Sirois, 1992; Fang and Lechowicz, 2006). Field surveys and published studies (Appendix A) have shown that tree rhododendrons occur in the temperate forest, with significant amounts of rainfall and high humidity. An important role of precipitation in explaining the distribution of tree rhododendrons could thus be expected. In the Himalayas, almost 80% of annual rainfall occurs during the monsoon season (Ranjitkar et al., 2013; Vetaas, 2000; Wangda and Ohsawa, 2006), which coincides with the warmest and the wettest quarter of the year, when thermal stress is higher. The monsoonal climate is reflected by high rainfall seasonality values in our model. For different sub-regions, precipitation at different times of year represents an important factor that determines the suitability of R. arboreum (i.e. ‘‘precipitation during the driest and coldest quarter’’ in the far-western sub-region, and ‘‘precipitation during the warmest quarter’’ in the western sub-region). Humid conditions during this season likely mitigate moisture stress caused by high temperatures in the comparatively-drier Western Himalayas. The high density of tree rhododendrons in the wetter parts of the Eastern Himalayas, compared to the drier Western Himalayas, also suggests that high rainfall increases the chance of encountering the species. Southern populations of tree rhododendrons receive rainfall throughout the year, which may be necessary for the species to survive at lower latitudes. Giriraj et al.(2008) also reported the importance of precipitation variables (which almost exactly concur with our model) for R. arboreum subsp nilagiricum in the Western Ghats, South India. 10 S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12

4.3. Dispersal

The monsoonal climate in the Sino-Himalayas might be a climatic factor for dispersal of tree rhododendrons (subsection Arborea). This area is characterized by a very high rate of plant endemism (Zhang and Sun, 2011), as well as rapid diversification of the entire subgenus Hymenanthes (Milne et al., 2010). Therefore, it seems plausible that R. delavayi, which occupies almost the entire eastern region (the Hengduan Mountain range and the eastern sub-region) of the distributional range, is representative of this ancient gene pool. This species must have a narrower range of climatic suitability, as shown in our predictions. Therefore, its range expansion towards the west and south from this region might have been promoted by climatic fluctuations, speciation under climatic stress, and the ability of rhododendrons to disperse over long distances. Rhododendron seeds are dispersed through air, water, and by animals (attached in animal fur) (Higgins, 2008). Seed can travel over 100 m (up to 1000 m) in windy conditions, as shown for Rhododendron ponticum (Higgins, 2008; Stephenson et al., 2007). In addition, animal movement can help in long-distance dispersal of the species, while streams and rivers can spread seeds even farther. A role of rivers in shaping species distributions and transferring invasive species to new areas is suggested by several studies conducted in Yunnan (Wang et al., 2011; Yue et al., 2012; Zhang and Sun, 2011). It seems plausible that the migration occurred primarily along the river valleys that currently connect bioclimatic spaces for tree rhododendrons. The different sub-species within species of this subsection – several of which, at the present time, share habitats – are likely to have arisen during periods in which habitats were more fragmented than today. Such fragmentation might have occurred during glacial periods of the Pleistocene (Irving and Hebda, 1993). Our model does not incorporate identification of the sub-species level in the initial stage of modelling because most of the specimens used as source of information do not provide nomenclature details at a sub-species level. Therefore, model outputs do not recognize climatic niche at a sub-species level, particularly in the Himalayas.

4.4. The southern populations of tree rhododendrons

The climate variables selected for modelling were capable of predicting the bioclimatic spaces of the two southern sub- species, but their connection to the Himalayan ecotypes could not be explained on the basis of the present model. The two sub-species R. arboreum subsp nilagiricum and R. arboreum subsp zeylanicum are present in geographically-separate tropical montane areas of South India and Sri Lanka, respectively, potentially representing remnants of plant communities that occupied the subcontinent during Pleistocene glaciations (Burkill, 1924). As in the case of Yunnan (Wang et al., 2011; Yue et al., 2012; Zhang and Sun, 2011), the role of South Asian rivers should not be underestimated in species dispersal. The Paleo- Ganges River may have contributed to the dispersal of Himalayan species to southern latitudes, from where some species settled into their current habitat during the glacial period. According to the so-called Satpura hypothesis, the Satpura range to the south of the Ganges Basin (Fig. 3(c)) might have served as a migration route of temperate species to southern latitudes. The Satpura hypothesis also elaborates on the role of rivers in this region for the dispersal of animal and plant species (Hora, 1949). Population genetic studies have not been used to confirm these theories in the current study, but based on some other studies (e.g. Jain et al., 2000; Kuttapetty et al., 2014) and modelled bioclimatic space, the aforementioned drivers are likely to be true for current patterns of distribution.

4.5. Tree rhododendrons in the context of climate change

Tree rhododendrons are relatively long-lived species and populations at low elevations may be remnants from larger populations present during a cooler period 150–500 years ago (Vetaas, 2002). Thus, it is possible that suitable habitats have already been shrinking for some time, putting additional pressure on the species which is also threatened by human-induced habitat loss (Gibbs et al., 2011; Giriraj et al., 2008; Haq, 2011; MOE, 2012; Singh et al., 2009). On the other hand, strategies that enable species to cope with climatic change can put some insights in tree rhododendron distribution. A species can genetically adapt through evolutionary processes, or the organisms’ physiology may be plastic enough to enable it to function under a wide range of climates (Austin and Moehring, 2013). Such wide distribution is suggestive of high genetic diversity (Geml et al., 2010), which is reported within the tree rhododendron’s gene pool (Jain et al., 2000; Kuttapetty et al., 2014), particularly R. arboreum. The southern species of tree rhododendrons are likely a manifestation of this genetic adaptability. A further advantage of the species may be its high phenotypic plasticity that allows tree rhododendrons to thrive even in regions that are outside its climatic ‘‘comfort zone’’. The most striking example of this phenotypic adaptability is the existence of a viable reproducing population of Rhododendron arboreum in the highlands of Jamaica, where a patch of naturalized forest of Rhododendron arboreum was reportedly established during the early 1900s (Goodland and Healey, 1996; Leach, 1957). This suggests that, if tree rhododendrons are left undisturbed, they may be able to cope with rapid climate change, and little conservation effort would be needed to help them persist. However, as evidenced by its listing as threatened in some areas of the Himalayas, land-use pressures on the species and its habitat are intense and unabated. S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12 11

5. Conclusions

Our ecological niche models were very successful in delineating the current climatic space of two tree rhododendrons species. The distributional limits of tree rhododendrons were associated with thermal factors (isothermality and temperature during the wettest quarter of the year), except in the drier Western Himalayas, where the moisture regime (precipitation during the coldest and warmest quarter of the year) is an important contributor. Its occurrence in a wide range of climatic settings with highly dissected habitats speaks of the adaptive capacity of the species, and the existence of isolated viable populations in South India, Sri Lanka and Jamaica indicates high phenotypic plasticity. More information about the age structure, climatic adaptability and detailed genetic studies from the region of occurrence, would provide evidence of previous migrations and genetic adaptability of this Himalayan species. Such information, along with the plastic nature of tree rhododendrons, might open up future options for their conservation planning in regions where they are listed as threatened.

Acknowledgements

Modelling, field visits, and data collection were supported by the National Science Foundation of China (Grant No. 31270524) and the Independent Research Program of the Chinese Academy of Sciences (Grant No. KSCX2-EW-J-24). We acknowledge the support from the CGIAR research programs on ‘‘Forests, Trees and Agroforestry’’ (CRP6.2) and ‘‘Climate Change, Agriculture and Food Security’’ (CRP7). We would like to thank the herbaria, mentioned in the manuscript and the appendix, for providing access to their collections. We also express our thanks to Dr. R. Edward Grumbine for reviewing our manuscript. The first author would like to express sincere thanks to Jonathan Teichroew for English editing.

Appendix A. Supplementary data

Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.gecco.2014.07.001.

References

Araújo, M.B., Cabeza, M., Thuiller, W., Hannah, L., Williams, P.H., 2004. Would climate change drive species out of reserves? An assessment of existing reserve-selection methods. Glob. Chang. Biol. 10, 1618–1626. Austin, M., 2007. Species distribution models and ecological theory: a critical assessment and some possible new approaches. Ecol. Modell. 200, 1–19. Austin, C.J., Moehring, A.J., 2013. Optimal temperature range of a plastic species, Drosophila simulans. J. Anim. Ecol. 82, 663–672. Bakkenes, M., Alkemade, J.R.M., Ihle, F., Leemans, R., Latour, J.B., 2002. Assessing effects of forecasted climate change on the diversity and distribution of European higher plants for 2050. Glob. Chang. Biol. 8, 390–407. Bhattacharyya, D., 2011. Rhododendron species and their uses with special reference to Himalayas—A Review. Assam Univ. J. Sci. Technol. 7, 161–167. Bonan, G.B., Sirois, L., 1992. Air temperature, tree growth, and the northern and southern range limits to Picea mariana. J. Veg. Sci. 3, 495–506. Burkill, J.H., 1924. The botany of the Abor expedition. Rec. Bot. Surv. India 10, 420. Chamberlain, D.F., 1982. A Revision of Rhododendron II. Subgenus Hymenanthes, Notes from the Royal Botanic Garden Edinburgh. Royal Botanic Garden Edinburgh, Edinburgh. Chamberlain, D., Hyam, R., Argent, G., Fairweather, G., Walter, K.S., 1996. The Genus Rhododendron: Its Classification And Synonymy. Royal Botanic Garden Edinburgh, Edinburgh. Chettri, N., Sharma, E., Deb, D.C., Sundriyal, R.C., 2002. Impact of firewood extraction on tree structure, regeneration and woody biomass productivity in a trekking corridor of the Sikkim Himalaya. Mt. Res. Dev. 22, 150–158. Cox, P.A., 1990. The Larger Rhododendron Species. OR:Timber Press, Portland. Dolui, A.K., Sharma, H.K., Marein, T.B., Lalhriatpuii, T., 2004. Folk herbal remedies from Meghalaya. Indian J. Tradit. Knowl. 3, 358–364. Dorji, S., Vernes, K., Rajaratnam, R., 2011. Habitat correlates of the red panda in the temperate forests of Bhutan. PLoS One 6, e26483. Elith, J., Graham, C.H., Anderson, R.P., Dudí k, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberón, J., Williams, S., Wisz, M.S., Zimmermann, N.E., 2006. Novel methods improve prediction of species’ distributions from occurrence data. Ecography (Cop.) 29, 129–151. Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697. Fang, M., Fang, R., He, M., Hu, L., Yang, H., Chamberlain, D.F., 2005. In: Wu, Z.Y., Raven, P.H., Hong, D.Y. (Eds.), Rhododendron Linnaeus. In: Flora of China, vol. 14. Science Press, Beijing, and Missouri Botanical Garden Press, St. Louis, pp. 260–455. Fang, J., Lechowicz, M.J., 2006. Climatic limits for the present distribution of beech (Fagus L.) species in the world. J. Biogeogr. 33, 1804–1819. Feeley, K.J., Silman, M.R., 2010. Land-use and climate change effects on population size and extinction risk of Andean plants. Glob. Chang. Biol. 16, 3215–3222. Fox, J., Weisberg, S., 2011. An R Companion to Applied Regression, second ed. Sage. Geml, J., Kauff, F., Brochmann, C., Taylor, D.L., 2010. Surviving climate changes: high genetic diversity and transoceanic gene flow in two arctic-alpine lichens, Flavocetraria cucullata and F. nivalis (Parmeliaceae, Ascomycota). J. Biogeogr. 1529–1542. Gibbs, D., Chamberlain, D., Argent, G., 2011. The Red List of Rhododendrons. Botanic Gardens Conservation International, Richmond, UK. Giriraj, A., Ramesh, B.R., Karunakaran, P.V, Jentsch, A., Murthy, M.S.R., 2008. Mapping the potential distribution of Rhododendron arboreum Sm. ssp. nilagiricum (Zenker) Tagg (Ericaceae), an endemic plant using ecological niche modelling. Curr. Sci. 94, 1605–1612. Goodland, T., Healey, J.R., 1996. The Invasion of Jamaican Montane Rainforests by the Australian Tree Pittosporum undulatum Bangor, UK. Grinnell, J., 1917. The niche-relationships of the California Thrasher. AUK 34, 427–433. Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more than simple habitat models. Ecol. Lett. 8, 993–1009. Haq, F.U., 2011. Conservation status of the critically endangered and endangered species in the Nandiar Khuwar catchment district Battagram, Pakistan. Int. J. Biodivers. Conserv. 3, 27–35. Higgins, G.T., 2008. Rhododendron ponticum: a guide to management on nature conservation sites. In: Irish Wildlife Manuals, No. 33. National Parks and Wildlife Service. Department of the Environment, Heritage and Local Government, p. 74. 12 S. Ranjitkar et al. / Global Ecology and Conservation 1 (2014) 2–12

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surface for global land areas. Int. J. Climatol. 25, 1965–1978. Hijmans, R.J., Graham, C.H., 2006. The ability of climate envelope models to predict the effect of climate change on species distributions. Glob. Chang. Biol. 12, 2272–2281. Hill, M.P., Hoffmann, A.A., Macfadyen, S., Umina, P.A., Elith, J., 2012. Understanding niche shifts: using current and historical data to model the invasive redlegged earth mite, Halotydeus destructor. Divers. Distrib. 18, 191–203. Hora, S.L., 1949. Satpura hypothesis of the distribution of the Malayan fauna and flora to Peninsular India. Proc. Nat. Inst. Sci. 15, 309–314. Hutchinson, G.E., 1957. Concluding remarks. Cold Spring Harb. Symp. Quant. Biol. 22, 415–427. Irving, E., Hebda, R., 1993. Concerning the origin and distribution of rhododendrons. J. Am. Rhododendron Soc. 47, 139. Jain, A., Pandit, M.K., Elahi, S., Jain, A., Bhaskar, A., Kumar, V., 2000. Reproductive behaviour and genetic variability in geographically isolated populations of Rhododendron arboreum (Ericaceae). Curr. Sci. 79, 1377–1381. Jolli, V., Srivastav, A., Thakur, S., 2012. Patch occupancy for cheer pheasant Catreus wallichii in the Great Himalayan national park conservation area. Int. J. Gall. Conserv. 2, 74–81. Kindt, R., 2014. BiodiversityR: GUI for biodiversity, suitability and community ecology analysis. Kuttapetty, M., Pillai, P.P., Varghese, R.J., Seeni, S., 2014. Genetic diversity analysis in disjunct populations of Rhododendron arboreum from the temperate and tropical forests of Indian subcontinent corroborate Satpura hypothesis of species migration. Biologia (Bratisl) 69, 311–322. Leach, D.G., 1957. The discovery of rhododendrons on Jamaica. J. Am. Rhododendron Soc. 11, 1–5. Long, D.G., 1991. Rhododendron L.. In: Grierson, A.J.C., Long, D.G. (Eds.), Flora of Bhutan Vol. 2 Part 1. The Royal Botanical Garden Edinburgh, pp. 358–387. makowsky, R., Marshall, J.C., McVay, J., Chippindale, P.T., Rissler, L.J., 2010. Phylogeographic analysis and environmental niche modeling of the plain-bellied watersnake (Nerodia erythrogaster) reveals low levels of genetic and ecological differentiation. Mol. Phylogenet. Evol. 55, 985–995. Marino, J., Bennett, M., Cossios, D., Iriarte, A., Lucherini, M., Pliscoff, P., Sillero-Zubiri, C., Villalba, L., Walker, S., 2011. Bioclimatic constraints to Andean cat distribution: a modelling application for rare species. Divers. Distrib. 17, 311–322. Marmion, M., Parviainen, M., Luoto, M., Heikkinen, R.K., Thuiller, W., 2009. Evaluation of consensus methods in predictive species distribution modelling. Divers. Distrib. 15, 59–69. Milne, R.I., Davies, C., Prickett, R., Inns, L.H., Chamberlain, D.F., 2010. Phylogeny of Rhododendron subgenus Hymenanthes based on chloroplast DNA markers: between-lineage hybridisation during adaptive radiation? Plant Syst. Evol. 285, 233–244. MOE,, 2012. The National Red List 2012 of Sri Lanka: Conservation Status of The Fauna And Flora. Ministry of Environment, Colombo, Sri Lanka. Normand, S., Ricklefs, R.E., Skov, F., Bladt, J., Tackenberg, O., Svenning, J.-C., 2011. Postglacial migration supplements climate in determining plant species ranges in Europe. Proc. Biol. Sci. 278, 3644–3653. Pearson, R.G., Dawson, T.P., 2003. Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Glob. Ecol. Biogeogr. 12, 361–371. Press, J.R., Shrestha, K.K., Sutton, D.A., 2000. Annotated Checklist of the Flowering Plants of Nepal. The Natural History Museum, London. Rai, I.D., Adhikari, B.S., Rawat, G.S., 2012. Mass foliar damage at subalpine-timberline ecotone in Western Himalaya due to extreme climatic events. Am. J. Clim. Chang. 01, 104–107. Ranjitkar, S., Luedeling, E., Shrestha, K.K., Guan, K., Xu, J., 2013. Flowering phenology of tree rhododendron along an elevation gradient in two sites in the Eastern Himalayas. Int. J. Biometeorol. 57, 225–240. Ranjitkar, S., Sujakhu, N.M., Jati, R., Xu, J., Schmidt-Vogt, D., 2014a. Yield and household consumption of Rhododendron arboreum as a fuelwood species in Eastern Nepal. Biomass Bioenerg 61, 245–253. Ranjitkar, S., Xu, J., Shrestha, K.K., Kindt, R., 2014b. Ensemble forecast of climate suitability for the trans-Himalayan Nyctaginaceae species. Ecol. Modell. 282, 18–24. Rogerson, P.A., 2001. Statistical Methods for Geography. Sage, London. Sagarin, R.D., Gaines, S.D., Gaylord, B., 2006. Moving beyond assumptions to understand abundance distributions across the ranges of species. Trends Ecol. Evol. 21, 524–530. Schmidt-Vogt, D., 1990. High Altitude Forests in the Jugal Himal (Eastern Central Nepal) - Forest Types And Human Impact. In: Geoecological Research, vol. 6. Franz Steiner Verlag Stuttgart, Germany. Sekar, K.C., 2010. Rhododendrons in Indian Himalayan region: diversity and conservation. Am. J. Plant Sci. 01, 131–137. Sexton, J.P., McIntyre, P.J., Angert, A.L., Rice, K.J., 2009. Evolution and ecology of species range limits. Annu. Rev. Ecol. Evol. Syst. 40, 415–436. Singh, K.K., Rai, L.K., Gurung, B., 2009. Conservation of rhododendrons in Sikkim Himalaya: an overview. World J. Agric. Sci. 5, 284–296. Stephenson, C.M., Kohn, D.D., Park, K.J., Atkinson, R., Edwards, C., Travis, J.M., 2007. Testing mechanistic models of seed dispersal for the invasive Rhododendron ponticum (L.). Perspect. Plant Ecol. Evol. Syst. 9, 15–28. Trivedi, M.R., Berry, P.M., Morecroft, M.D., Dawson, T.P., 2008. Spatial scale affects bioclimate model projections of climate change impacts on mountain plants. Glob. Chang. Biol. 14, 1089–1103. Vetaas, O.R., 2000. Separation of subspecies along a temperature gradient, in: Proceedings IAVS Symposium. pp. 28–31. Wang, R., Wang, J., Qiu, Z., Meng, B., Wan, F., Wang, Y., 2011. Multiple mechanisms underlie rapid expansion of an invasive alien plant. New Phytol. 191, 828–839. Vetaas, O.R., 2002. Realized and potential climate niches: a comparison of four Rhododendron tree species. J. Biogeogr. 29, 545–554. Wangda, P., Ohsawa, M., 2006. Structure and regeneration dynamics of dominant tree species along altitudinal gradient in a dry valley slopes of the Bhutan Himalaya. For. Ecol. Manage. 230, 136–150. Yue, L.-L., Chen, G., Sun, W.-B., Sun, H., 2012. Phylogeography of Buddleja crispa (Buddlejaceae) and its correlation with drainage system evolution in southwestern China. Am. J. Bot. 99, 1726–1735. Zhang, T., Sun, H., 2011. Phylogeographic structure of Terminalia franchetii (combretaceae) in southwest China and its implications for drainage geological history. J. Plant Res. 124, 63–73.