Functional grouping and establishment oF distribution patterns oF invasive plants in china using selF-organizing maps and indicator species analysis

In the present study, we introduce two techniques – self-organizing maps (SOM) and indicator species analysis (INDVAL) – for understanding the richness patterns of invasive species. We first employed SOM to identify functional groups and then used INDVAL to identify the representative areas characterizing these functional groups. Quantitative traits and distributional information on 127 invasive plants in 28 provinces of China were collected to form the matrices for our study. The results indicate Jiangsu to be the top province with the highest number of invasive species, while Ningxia was the lowest. Six functional groups were identified by the SOM method, and five of them were found to have significantly representative provinces by the INDVAL method. Our study represents the first attempt to combine selforganizing maps and indicator species analysis to assess the macro-scale distribution of exotic species.


INTRODUCTION
Invasive species have become a serious socioeconomic problem worldwide.The costs of controlling invasive species have been increasing in recent years (Pimentel et al., 2005).A clear understanding of mechanisms the occurrence of invasive species in the environment (Usseglio-Polatera et al., 2000) would therefore be advantageous in creating efficient measures for the better management of invasive species.
In general, the more diverse a community is, the higher the chance that a particular taxon lives in a given habitat with various combinations of species traits (Usseglio-Polatera et al., 2000).Species traits therefore will provide useful information about species invasiveness, although there are some technical issues, e.g., local idiosyncrasies and phylogenetic constraints (Lloret et al., 2005).A growing body of research suggests that the success of invasive plants is controlled by a series of key processes or traits (Theoharides and Dukes, 2007).Consequently, some researchers recommend identifying groups of organisms with similar relationships among their species traits.
Functional groupings (Fox and Brown, 1993;Gitay and Noble, 1997;Wilson, 1999) denote species sharing common attributes.They have been used in different fields of ecological research, including vegetation studies, conservation, and so on (Salmaso and Padisak, 2007 and the references therein).Irrespective of the organisms studied, a common assumption within such a grouping is that the characteristics of a community can be better understood if species are grouped into classes that possess similar characteristics or behave similarly (Salmaso and Padisak, 2007).
Functional grouping analysis has some advantages in examining traits on an individual level because many traits manifest advantages only when acting synergistically (Lambdon et al., 2007).There are many works applying the concept of functional groups, for example, ones dealing with conservation Functional grouping and establishment oF distribution patterns oF invasive plants in china using selF-organizing maps and indicator species analysis target setting (Zhu et al., 2004) and invasive species management (Lososova et al., 2007;Statzner et al., 2007).
China is a vast country with rich biodiversity and regarded as a hotspot around the world (http://www.biodiversityhotspots.org/).however, it is also a very vulnerable country suffering from the ill effects of invasive species due to economic development (Ding et al., 2008).There is a long history of introduction of exotic species to China, especially those having economical benefits (Yan et al., 2001).however, negative feedbacks brought by exotic species have in recent decades been increasingly reported owing to global human activities and climate change.China is also a highly concerned nation (Yan et al., 2001;Ding et al., 2006;huang et al., 2008).
Understanding the functional grouping and distributional patterns of invasive species will be beneficial in establishing relevant controlling strategies and rapid assessment methods.In the present study, we apply two methods -self-organizing maps (SOM) and indicator species analysis (INDVAL) -to identify the functional groups of invasive plant species of China and select representative distributional provinces.

Distribution data
The provincial distributional information of 127 invasive plants of China was compiled from an online source (CSIS; China Species Information Service; http://www.chinabiodiversity.com/)and previously published papers (e.g., Yan et al., 2001;huang et al., 2008).Table 1 lists all the invasive plants and their corresponding numbers for subsequent analysis.

Physiological traits
The following physiological traits have been mentioned in previous works (e.g., Lososova et al., 2006Lososova et al., , 2007;;Lambdon et al., 2007;huang et al., 2008) as revealing the functional groups of invasive plants.For each attribute, its different states were quantitatively encoded in a series of binary variables.

Functional grouping analysis
We used self-organizing maps to group all species based on their functional similarity.After obtaining the resulting map, a cluster analysis using euclidean distance was performed to identify the final functional groupings.

SOM model
Self-organizing maps represent an artificial neural network model (Kohonen, 1982) aiming to classify high dimensional data, performing a non-linear projection of multidimensional data space onto two-dimensional space (Lek et al., 1996;Park et al., 2003aPark et al., , 2003b)).The SOM neural network consists of two layers of neurons: the input layer and the output layer.The output layer is represented by a map or a rectangular gird with l x m neurons (or cells), laid out in a hexagonal lattice (Worner and Gevrey, 2006).
We used a batch algorithm for SOM analysis (Worner and Gevrey, 2006).Details of the algorithm and its theoretical basis are given by Kohonen (2001).The software we used for implementing the SOM method was the Matlab programming language (Mathworks, 2001) and the SOM toolbox (version 2.0 beta), which was developed by the Laboratory of Information and Computer Science, helsinki University of Technology (http://www.cis.hut.fi/projects/somtoolbox/documentation/somalg.shtml).The geographic maps were generated by the software ArcView 3.2 (eSRI; http://www.esri.com/).

Cluster analysis
Sites that are neighbors on the grid are expected to be more similar to each other, whereas sites remote from each are expected to be distant in the feature space (Worner and Gevrey, 2006).To detect cluster boundaries on the map, cluster analysis was applied to the SOM model output (Park et al., 2003a(Park et al., , 2003b)).hierarchical cluster analysis can give cluster boundaries that are crisper than in the unified-matrix approach.A simple bootstrapping method was used to justify the choice of the number of clusters (hernandez et al., 2005).

Indicator species analysis
The indicator species analysis (INDVAL) method aims to identify representative species which can characterize groups of samples (Dufrene and Legendre, 1997).This analysis was performed using the INDVAL for PC package (http://biodiversite. wallonie.be/outils/indval/home.html).herein we use INDVAL to select representative areas for each functional invasive group.Significance was calculated using the method of computing the weighted distance between randomized values and the observed value (t-test).The Monte Carlo test was run using 50000 random iterations and five seeds per random number generator.The significance level was set at P<0.05 (Casazza et al., 2008).

ReSULTS AND DISCUSSION
We ascertained that Jiangsu had the highest richness out of 28 provinces, followed by Yunnan, Anhui, and Zhejiang.In contrast, Ningxia had the lowest richness.'hotspot' provinces containing more than 50 species include Jiangsu, Yunnan, Anhui, Zhejiang, Jiangxi, Fujian, and Liaoning.Most of these provinces are circumlittoral areas (Fig. 1).
The initial SOM model grouped species in a grid cell system (8×7) according to their trait similarities (Fig. 2).After applying hierarchical cluster analysis on the basis of the initial SOM map (Fig. 3a), our study revealed six functional groups based on their biological traits (Fig. 3b).Indicator species analysis identified representative invaded provinces for each functional group.Given below are detailed discussions of each group and relevant representative areas.
Group I included 12 species, which mostly had the flowering season in winter and the seed and vegetative reproduction mode.The representative species were Abutilon crispum, Lantana camara, and Wedelia trilobata.The conducted INDVAL analysis demonstrated that this group is distributed in the southern part of China and does not occur in Northern China.Representative provinces having significant indicator values (InV) were Fujian (InV=17.9,P<0.05), Guangdong (InV=29.8,P<0.05),  Guangxi (InV=19.8,P<0.05), hainan (InV=24.8,P<0.05), and Taiwan (InV=20.8,P<0.05).
Group II was composed of 27 species.The life mode of most species in this group was the annual grass mode.This group has the flowering season in spring.Representative species were Axonopus compressus, Cassia tora, and Hordeum jubatum.The conducted INDVAL analysis indicated that this group has no biased geographic distribution and is homogeneously invasive throughout the whole nation.No representative provinces were found.
Group III consisted of 18 species, which are generally distributed throughout the whole nation and have a range that is broader than for other groups.The typical species in this group were Mirabilis jalapa, Talinum paniculatum, and Ipomoea purpurea.The representative areas selected by INDVAL analysis were Southwestern China, principally including Yunnan Province (InV=18.1,P<0.05).
Group IV was composed of 20 species, which  have a dispersal mode related to abiotic factors, for example, hydrochory or anemochory.Representative species were Erigeron annuus, Senecio vulgaris, and Amaranthus albus.Typical distribution provinces were mainly located in the western part of China, including two provinces: Inner Mongolia (InV=12.5, P<0.05) and Xizang (InV=12.9,P<0.05).
Group V consisted of 24 species, most of which have been documented by published reports to have strong invasiveness and to be dangerous for eco- From the above analysis, we find that the SOM technique is a simple way to illustrate the associations of studied objectives through reducing the data dimensions.Its advantage is its ability to group objectives at a high speed compared to conventional cluster analysis.Thus, the SOM method is suitable for dealing with large data sets.
In our study, we not only ascertained functional groups of invasive species on the basis of their physiological traits, but also tried to understand the geographic patterns of these groups.By implementing the INDVAL method, we can better understand the distributional biases of functional groups of China's invasive plants.As far as we know, this represents the first attempt to combine SOM and INDVAL in application to distribution patterns of invasive species.In further studies including native species, the distributional overlapping and associated resource partition (Lambdon et al., 2007) between functional groups of exotic and native species could be compared using the methods introduced here.
a) Life span: annual grass, perennial grass, liane, shrub, and arbor.b) Invasive seriousness: common, middle, and strong.c) Regions of origin: Americas, europe, Australasia, Asia, or Africa.d) Introduction mode: accidental, human, fruit, ornamental, officinal, or pasture.e) Propagation habitat: human-related areas, wild natural areas, or both.f) Flowering season: spring, summer, autumn, winter, or all year.g) Reproduction mode: seed or seed and vegetative.h) Suitable range size: provincial or national.

Fig. 1 .
Fig. 1.Highlighted provinces of China are ones that contained more than 50 invasive plants.

Fig. 2 .
Fig. 2. Basal grouping of invasive plants based on similarities of trait characteristics.The species numbers correspond to the Latin names in Table1.

Fig. 3 .
Fig.3.The final six functional groups determined by hierarchical cluster analysis.a) The dendrogram generated by cluster analysis with the optimal cluster line using a bootstrapping technique.b) The corresponding SOM map (species occupying each cell can be identified in Fig.2).

Table 1 .
Latin names of 127 invasive species of China and associated numbers assigned to them for purposes of analysis.
systems.examples include Ambrosia artemisiifolia, Lepidium virginicum, and Agrostemma githago.The conducted INDVAL analysis demonstrated that this group has biased distribution in the northern part of China, principally in Jilin Province (InV=10.1,P<0.05).