See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/225220768

Application of self-organizing neural networks

to classification of plant communities in
Pangquangou Nature Res....

Article in Frontiers of Biology in China · December 2008

DOI: 10.1007/s11515-008-0061-7

CITATIONS READS

13 35

2 authors:

Jintun Zhang Hongxiao Yang

Beijing Normal University Qingdao Agricultural University
144 PUBLICATIONS 1,568 CITATIONS 20 PUBLICATIONS 144 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Biodiversity View project

Quantitative ecology and phytochemistry of Gentiana macrophylla communities View project

All content following this page was uploaded by Jintun Zhang on 01 April 2015.

The user has requested enhancement of the downloaded file.

Front. Biol. China 2008, 3(4): 512–517
DOI 10.1007/s11515-008-0061-7

RESEARCH ARTICLE

Application of self-organizing neural networks to classification

of plant communities in Pangquangou Nature Reserve, North
China
Jintun ZHANG (*), Hongxiao YANG
College of Life Sciences, Beijing Normal University, Beijing 100875, China

E Higher Education Press and Springer-Verlag 2008

Abstract Vegetation classification is an important topic 1 Introduction

in plant ecology and many quantitative techniques for
classification have been developed in the field. The arti- Classification of plant communities is the fundamental
ficial neural network is a comparatively new tool for data work in study of vegetation ecology which has attracted
analysis. The self-organizing feature map (SOFM) is much attention from ecologists (Yang and Lu, 1981; Yu,
powerful tool for clustering analysis. SOFM has been 1995; Zhang, 2004a, 2004b). The systems of vegetation
applied to many research fields and it was applied to the classification established in different periods reflect
classification of plant communities in the Pangquangou human knowledge of vegetation and the development
Nature Reserve in the present work. Pangquangou stage of ecologic science at that time. In the past decades,
Nature Reserve, located at 37u209–38u209 N, 110u189– numerical classification has been recognized to be essen-
111u189 E, is a part of the Luliang Mountain range. tial methodologies in this field (Zhang, 1994a, 2004a,
Eighty-nine samples (quadrats) of 10 m 6 10 m for for- 2004b). There have been a number of effective quantitat-
est, 4 m 6 4 m for shrubland and 1 m 6 1 m for grass- ive methods in plant ecology research (Zhang and Zhang,
land along an elevation gradient, were set up and species 2000). Neural network theory is a comparatively novel
data was recorded in each sample. After discussion of the branch of mathematics, developed based on the mech-
mathematical algorism, clustering technique and the pro- anism of the human brain (Li and Zheng, 2003).
cedure of SOFM, the classification was carried out by Compared with other quantitative methods, artificial
using NNTool box in MATLAB (6.5). As a result, the neural networks (ANN) show advantages in dealing with
89 samples were clustered into 13 groups representing 13 complicated problems (Yuan, 2000; Wu and Huang,
types of plant communities. The characteristics of each 2005). Theoretically, ANN can describe natural phenom-
community were described. The result of SOFM clas- ena and rules better, and have been successfully applied in
sification was identical to the result of fuzzy c-mean clus- artificial intelligence, computer, industry, geosciences and
tering and consistent with the distribution patterns of medical studies, etc. (Yuan, 2000; Feisi Center for
vegetation in the study area and shows significant eco- Scientific Products, 2004). An ecosystem is a natural com-
logical meanings. This suggests that SOFM may clearly plex system and the application of ANN to an ecosystem
describe the ecological relationships between plant com- may be a convenient alternative tool to traditional statist-
munities and it is a very effective quantitative technique in ical methods. The Kohonen Self-Organizing Feature Map
plant ecology research. (SOFM) is one of the most well-known neural networks
with unsupervised learning rules and is applied to clas-
Keywords Neural network, self-organizing feature map, sification of plant communities in the Pengquangou
vegetation, quantitative classification Reserve in the present study. The Pengquangou Reserve
was established for the conservation of the first-class
nationally protected bird, Crossoptilon mantchuricum,
and the cold-temperate coniferous forests. The diversity
of plant communities is the basis for conservation of
Translated from Acta Ecologica Sinica, 2007, 27(3): 1005–1010 [译 endangered animals and plants. Recently, the
自: 生态学报] Pengquangou Reserve has become an important eco-tour-
E-mail: zhangjt@bnu.edu.cn ism target (Cheng et al., 2006) and hence, the study of
Application of self-organizing neural networks to classification of plant communities in Pangquangou Nature Reserve 513

plant community types is significant for natural conser-

vation and also for development of local economy (Li and
Zhang, 2003).

2 Theory and methods of SOFM

2.1 SOFM theory

The Self-organizing feature map (SOFM) is a self-organ-

izing competition neural network established by Kohonen
in 1981. In the cerebral cortex, the input signal of neurons
is partially from sensory tissues or from outside input and
partially from feedback in the same area. The interaction
between neurons has the same characteristics, i.e. the two
adjacent neurons stimulate each other while distant neu- Fig. 1 Structure of self-organizing feature map (SOFM).
rons are restrained with each other. Each neuron strength-
ens itself and its adjacent neurons and affects other 2.2 SOFM clustering method and procedure
neurons simultaneously. If one neuron is excited, it will
restrain other neurons from getting excited through its Suppose the input data vector:
branches. This will lead to competition between neurons
and as a result, the most excited neuron will win over other Pk ~(Pk1 ,Pk2 , . . . ,PkN ), ðk~1,2, . . . ,qÞ
neurons (Yuan, 2000). SOFM is set up based on this the-
ory of the biological structure of the brain.
The associated weight vector,
SOFM network consists of two layers, an input layer
and an output layer (Fig. 1). There are N nodes (neuron)

Wij ~ Wj1 , Wj2 , . . . , Wji , . . . , WjN i~1, 2, . . . , N;
in the input layer and M nodes (neuron) in output layer.
The input node and the output node are connected by j~1, 2, . . . , M:
two-direction weights. The input layer contains a neuron
Then, the SOFM clustering steps are:
for each variable in the vegetation data set. The input
1) Initializing. Giving initial values of Wij within [0, 1]
units operate in a similar way to those in other neural
randomly (i 5 1, 2, …, N; j 5 1, 2, …, M), initial values of
networks, effectively presenting the data for each quadrat
learning rate g(0) and neighborhood Ng(0) and determin-
to the network in an appropriate format. The output layer
ing total learning times T.
is a two-dimensional array of units (M 5 m2) and each of
2) Inputting a random quadrat unit drawn from the
these units is linked to every unit in the input layer by a
input dataset Pk into the network and calculating Pk :
weighted connection. Lateral interaction between units in
the output layer also ensures that learning is a competitive

Pk Pk1 ,Pk2 , . . . ,PkN
process in which the network adapts to response in differ- Pk ~ ~ h
2
2
2 i1=2
ent locations for inputs that differ. Consequently, quad- kPk k
Pk1 z Pk2 z . . . z PkN
rats that are similar should be associated with units that
are close together in the output layer while a dissimilar 3) Calculating wj :
quadrat would be associated with a distant unit elsewhere

in the output layer. Wj Wj1 ,Wj2 , . . . ,WjN
W j~  
Wj  ~ h
2
2
2 i1=2
SOFM realizes network learning and training by use of
Wj1 z Wj2 z . . . z WjN
self-organizing and non-supervising training. The struc-
ture of network and connected weights are adjusted auto-
4) Defining Euclidean distance between wj and Pk :
matically according to clustering regulations and this
procedure will be ended when the distribution rule of sam- " #1=2
N 
X 2
ples is illustrated clearly. In practice, we only need to dj ~ Pki {Wji , ðj~1,2, . . . ,M Þ
adjust weights for each input to make the weight vector i~1
closer to or further away from the input vector. This is an
5) Determining the minimum distance dg, g is chosen as
integrated competitive learning process and the classifica-
the winning neuron, called the Best Matching Unit (BMU).
tion of quadrats will be carried out automatically during
this process (Tran et al., 2003; Feisi Center for Scientific  
Products, 2004). dg ~ min dj , j~1,2, . . . ,M
514 Jintun ZHANG, Hongxiao YANG

6) Adjusting the weights (Wij) elevation varies from 1700 m to 2831 m. The natural
h i vegetation is conserved well in the Pengquangou
Wji (tz1)~Wji (t)zg(t): Pki {Wjt (t) Reserve and most areas of this reserve are covered by
forests, with a small area close to the mountain top cov-
ðj~1,2, . . . ,M; 0vgð0Þv1Þ
ered by subalpine shrubland and meadow (Chen and
7) Selecting another random quadrat unit and input- Zhang, 2000). According to the system of national vegeta-
ting it into the network, and return to step 3) until all q tion regionalization (Zhang and Chen, 2004), the basal
quadrat units have been input into the network. vegetation zone of the Pengquangou Reserve is warm tem-
8) Defining learning rate g(t) and neighborhood Ng(t): perate deciduous broad-leaved forest. The vertical vegeta-
g(t) is the learning rate at t time, here we define it as tion belts are deciduous broad-leaved forest (800–
follows: 1600 m), coniferous and deciduous broad-leaved mixture
forest (1600–1750 m), cold-temperate coniferous forest
gðtÞ~gð0Þð1{t=T Þ ð0vgð0Þv1Þ (1750–2600 m) and subalpine shrubland and meadow
(2600– 2831 m).
g(0) is the initial learning rate, t the learning time, T the Along the elevation gradient of 1700–2700 m in
total learning times.Suppose a neuron in competition Shenweigou valley, 21 transects separated by 50 meters
layer g has coordinates (xg, yg) in two-dimensional array, in altitude were set up and species data of cover, height,
and its neighborhood should be a square within points basal area, and individual number from 4–6 quadrats in
xg + Ng(t), yg + Ng(t), (xg 2 Ng(t) and yg 2 Ng(t). The each transect were recorded. The quadrat size is
modified formula is: 10 m 6 10 m (based on the minimum community area),
 in which 4 m 6 4 m and 1 m 6 1 m small quadrats were
Ng ðtÞ~INT Ng ð0Þð1{t=T Þ, used to record shrubs and herbs, respectively, and to cal-
culate frequency. There were totally 198 species in 89
INT[x] represent taking positive integer, Ng(0) is initial quadrats recorded. Elevation, slope, aspect and the depth
value of Ng(t). of litters for each quadrat were also measured and
9) Increasing time t to t + 1. If t , T then go to step 2), recorded.
or else stop the training. We used the Importance Value of each species as data in
Through training, the winner neuron g and weight vec- community analysis. The importance value was calculated
tor must approach to input vector, and must make the according to the following formulas (Zhang, 2004a,
clustering a realization. By use of the neural network tool- 2004b):
box in MATLAB, the network will provide classification
results automatically after defining learning rate, learning IV Tree and shrub 5 (Relative cover + Relative frequency +
times, neighborhood radius, network dimensions, etc. Relative density)/300
(http://www.mathworks.com; Schreer et al., 1998). IV Herbs 5 (Relative cover + Relative height)/200

Therefore, the species data matrix is the importance

3 Plant community data in the Pengquangou values of 198 species in 89 quadrats.
Reserve

The Pengquangou Reserve is located in the midst of 4 SOFM classification of plant communities
Luliang mountain range, at 37u209–38u209 N, and in the Pengquangou Reserve
110u189–111u189 E. Its elevation varies from 1700 m to
2831 m. The climate of this area is warm temperate, Initializing, training, and stimulating for SOFM cluster-
semi-humid, with continental characteristics and con- ing were carried out in the neural network toolbox of
trolled by seasonal winds. The annual mean temperature MATLAB (6.5). The training process was divided into
varies from 3uC to 4uC. The monthly mean temperatures two phases with different learning rates and neighbor-
of January and July are 210uC and 16.1uC, respectively hood radius (Yuan, 2000; Feisi Center for Scientific
and the annual cumulative temperature more than 10uC is Products, 2004). The matrix data of 198 species in 89
2100uC. The annual frost-free period is about 100 d. The quadrats
were input intothe network and the input model
annual mean precipitation is 838.8 mm and over 60% of was Pk ~ Pk1 ,Pk2 , . . . ,PkN , k 5 1, 2, …, q (q 5 198, N 5 89).
rains fall in July, August and September. The main soil There were 89 groups of sample vectors (quadrats) and
parent includes granite, gneiss, quartzite, etc. The vertical each sample vector contained 198 elements (species). In
distribution of soils is apparent and several soil types, such other words, there were 89 neurons in input layer. The
as mountain cinnamon soil, brown forest soil and moun- number of neurons in the competitive layer depends on
tain meadow soil, can be found in this area. The study the number of clusters of classification of the 89 quadrats
area, Shenweigou, is a main valley in the east region. Its and is given by researchers based on ecological knowledge
Application of self-organizing neural networks to classification of plant communities in Pangquangou Nature Reserve 515

and experience. By reference to former research of vegeta- brown soil. The common species are Betula platyphylla,
tion in the Pengquangou Reserve, 11, 12 and 13 clusters of Evonymus alatus, Spiraea pubescens, Cotoneaster acutifo-
plant communities were selected and compared in SOFM lius, Carex lanceolata, Ploygonatum verticilatum,
clustering. Epilobium hirsutum and Agrimonia pilosa. Comm. Pinus
SOFM network was chosen in the neural network tool- tabulaeformis is grows in areas with an altitude of 1500–
box of MATLAB. The learning rate was 0.1 for the ordi- 1900 m, slope around 20u. The soils are mountain cin-
nating phase and 0.02 for adjusting phase; the learning namon soil and brown forest soil. The common species
phase was broken down into 5000 steps for the ordinating in the community are Quercus liaotungensis, Betula platy-
phase and 50 000 steps for the tuning phase. The neigh- phylla, Salix wallichiana, Ostryopsis davidiana, Rosa
borhood radius was 1.0 for adjusting phase. xanthina, Vitex negundo var. heterophylla, Spiraea pubes-
Compared with 11 and 12 clusters, 13 clusters of plant cens, Potygala tenuifolia, Atractylodes chinensis and
communities from SOFM classification was closest to Artemisia gmelini. Comm. Quercus liaotungensis is abund-
vegetation reality. The 13 clusters of SOFM result and ant in areas with an altitude of 1750–2050 m, slope
the corresponding types of plant communities are listed around 30u and the soils are mainly mountain brown soil.
in Table 1. This classification is identical to the result of The common species in the community are Betula platy-
the fuzzy C-mean clustering and can be interpreted well phylla, Picea wilsonii, Crataegus pinnatifida, Rosa
with ecological meanings (Zhang, 2004a, 2004b). xanthina, Corylus mandshurica, Phlomis umbrosa, Carex
This classification is reasonable according to ecological lanceolata, Convallaria keiskei, Chamaenerion angustifo-
and environmental characteristics of vegetation in the lium and Bupleurum chinense. Comm. Populus dividiana,
Pengquangou Reserve. It reflects the general picture of Betula platyphylla is distributed from 1950 to 2500 m in
plant communities and their environments and the key hills with a slope around 20u and its soil is brown forest
community types to be conserved. It also reveals the soil. The common species in the community are Quercus
relationships of plant community types, vegetation and liaotungensis, Spiraea pubescens, Hippophae rhamnoides,
environmental variables, such as elevation, water and Carex lanceolata, Lathyrus humilis, Phlomis umbrosa,
heating conditions and soils. Each community has its Galium bungei, Clematis macropetala and Maianthemum
own features. bifolium. Comm. Populus dividiana is distributed from
Comm. Valley bottom meadow is dominated by 1950 to 2300 m in the hills with slope around 25u and its
Ranunculus japonicus and Taraxacum dealbatum, distrib- soil is brown forest soil. The common species in the com-
uted in hills above 1700–1850 m with slope around 15u munity are Quercus liaotungensis, Lespedeza bicolor,
and its soil is subalpine meadow soil. The common species Crataegus pinnatifida, Evonymus alatus, Spiraea pubes-
are Sanguisorba officinalis, Thalictrum petaloideum, cens, Carex lanceolata, Convallaria keiskei, Ledebouriella
Taraxacum asiaticum and Carex spp. Comm. Hippophae divaricata and Bupleurum chinense. Comm. Larix princi-
rhamnoides and Ostryopsis davidiana is distributed in the pis-rupprechti is distributed from 1900 to 2500 m in hills
low hills of 1700–1800 m with slope around 25u and its with slope around 20u, and its soil is brown forest soil. The
soil is brown soil. The common species in the community common species in the community are Betula platyphylla,
are Ribes burejense, Rosa bella, Dasiphora glabra, Picea wilsonii, Quercus liaotungensis, Acer ginnala,
Lonicera chrysantha, Artemisia apiacea, Galium verum, Lonicera chrysantha, Lespedeza sp., Corylus mandshurica,
Thalictrum squarrosum and Ranunculus japonicus. Carex lanceolata, Lathyrus humilis, Bupleurum chinense
Comm. Populus cathayana is distributed in hills about and Thalictrum petaloideum. Comm. Picea wilsonii grows
1800 m with slope around 20u and its soil is mountain in areas at an altitude of 1950–2050 m and a slope of 15–

Table 1 Types and their sample composition resulting from SOFM clustering

SOFM clustering result types sample composition of each type name of plant communities
Type 1 9, 13, 39–42, 54–57 valley bottom meadow
Type 2 1–4, 12, 14, 24, 34, 38 Hippophae rhamnoides, Ostryopsis davidiana scrubland
Type 3 19, 23, 26, 32–33, 35–37 Populus cathayana forest
Type 4 18, 27–30 Pinus tabulaeformis forest
Type 5 5–7, 43–44, 58–59, 67, 71 Quercus liaotungensis forest
Type 6 8, 25, 68–69 Populus dividiana, Betula platyphylla forest
Type 7 20–21 Populus dividiana forest
Type 8 10–11, 15–17, 22, 31, 64 Picea wilsonii forest
Type 9 46–49, 52–53, 60–62, 65 Larix principis-rupprechtii forest
Type 10 50–51, 63, 79–80 Picea meyeri forest
Type 11 45, 88–89 Potentilla glabra, Spiraea alpina scrubland
Type 12 66, 78, 85–87 Evonymus hamiltonianus scrubland
Type 13 70, 72–77, 81–84 subalpine meadow
516 Jintun ZHANG, Hongxiao YANG

25u in mountain brown forest soil. The common species in meanings. The cold-temperate coniferous forests are dom-
the community are Populus davidiana, Betula platyphylla, inant. Larix principis-rupprechti forest and Picea forest
Corylus mandshurica, Rosa xanthina, Vicia cracca, develop well covering large areas and are the main habi-
Ledebouriella divaricata, Adenophora elata, Carex lanceo- tats of Crossoptilon mantchuricum and the important tar-
lata and Lathy rushumili. Comm. Picea meyeri is distrib- get forest of conservation in North China. Other forest
uted from 2000 to 2500 m in altitude with slope 20–25u communities, such as Quercus liaotungensis forest,
and brown forest soil. The main species in the community Populus davidiana, Betula platyphylla forest and Pinus
are Picea wilsonii, Larix principis-rupprechtii, Betula pla- tabulaeformis forest are significant in maintaining species
typhylla, Lonicera hispida, Rosa bella, Lonicera chry- diversity in this area and should be protected. The sub-
santha, Carex sp., Lathyrus humilis, Phlomis umbrosa, alpine shubland and meadow related to high elevation and
Maianthemum bifolium, Cerastium arvense and Vicia uni- cold climate play an important role in enriching ecosystem
juga. Comm. Potentilla glabra, Spiraea alpina grows at an diversity and species diversity (Zhang and Chen, 2004).
altitude of 2150–2450 m and a slope of 20–25u in moun- Ecological analysis shows that SOFM clustering is a very
tain meadow soil. The common species in the community effective method for classification of plant communities
are Spiraea pubescens, Dasiphora fruticosa, Carex leor- and suitable for vegetation study.
hyncha, Artemisia japonica, Sanguisorba officinalis and The operation of SOFM clustering for plant community
Scutellaria scordifolia. Comm. Evonymus hamiltonianus classification is very simple especially when it is carried out
is distributed from 2600 to 2800 m, close to mountain in the neural network toolbox of MATLAB and the clas-
top, and its soil is mountain meadow soil. The common sification becomes simpler. As long as the data matrix and
species are Dasiphora fruticosa, Trollius chinensis, parameters are provided in the network, the clustering
Saussurea mongolica, Sanguisorba officinalis, Thalictrum result will come out. Notably, when using SOFM to cluster,
petaloideum, and Taraxacum dealbatum. Comm. the number of groups of classification is determined sub-
Subalpine meadow is dominated by Kobresia bellardii jectively. However, this is not a serious disadvantage
and Potentilla nivea, distributed from 2100 to 2800 m with because many numerical classification techniques, such as
a gentle slope and its soil is subalpine meadow soil. The fuzzy C-mean clustering demand similar action (Zhang,
common species are Geum aleppicum, Papaver nudicaul, 1994b). The final classification results need to be evaluated
Rhodiola dumulosa, Trollius chinensis, and Carex spp. by researchers based on ecological knowledge and research
experiences which can not be replaced by mathematical
methods (Zhang, 2004a, 2004b, 2005).
5 Discussion Acknowledgements This study was supported by the National Natural
Science Foundation (Grant No. 30070140) and the Teachers’
SOFM neural network can deal with imprecise and incom- Foundation of the Education Ministry of China.
plete fuzzy information and has advantages in solving non-
linear problems and in studying complicated system.
Theoretically, SOFM can describe natural phenomena References
and rules better. The network can distribute work in par-
allel and hence can do calculation very quickly. It can dis- Chen T G, Zhang J T (2000). A study of plant community diversity
in Shenweigou of the Pengquangou Reserves, Shanxi. Acta
tribute information within the whole network with
Botanica Boreali-Occidentalia Sinica, 20(4): 638–646 (in Chinese)
variation of weights and problems for some units cannot Cheng Z H, Zhang J T, Wu B H, Niu L Q (2006). Relationships
affect the network function. Therefore, it is suitable for between tourism development and vegetation environment in
analysis of complex systems (Li and Zhang, 2003). Luya mountain nature reserve: vegetation landscape types and
Vegetation ecosystem is a complex system with various ordination. Acta Ecologica Sinica, 26(6): 1940–1946 (in Chinese)
Feisi Center for Scientific Products (2004). MATLAB 6.5 Designe
non-linear and fuzzy relations between species, communit-
and analysis of artificial neural network. Beijing: Electronic
ies and environmental factors. Therefore, SOFM clustering Industry Press (in Chinese)
should be a perfect methodology for the study of classifica- Li B, Zhang J T (2003). Analysis of relationships between vegetation
tion of plant community. Additionally, SOFM is self- and climate variables in Loess Plateau. Acta Botanica Sinica,
organizing learning without supervision and should be 23(1): 82–89 (in Chinese)
Li S C and Zheng D (2003). Progress of the application of artificial
objective. Namely, after inputting data matrix and defining
neural network to earth science. Progress of Earth Science, 18(1):
network parameters, the network will provide clustering 69–70 (in Chinese)
results automatically without human interference (Feisi Schreer J F, Hines R J O, Kovacs K M (1998). Classification of dive
Center for Scientific Products, 2004). profiles: A comparison of statistical clustering techniques and
SOFM clustering classified vegetation into 13 com- unsupervised artificial neural networks. Journal of Agricultural
Biological and Environmental Statistics, 3(4): 383–404
munities in the Pengquangou Reserve, which is consistent Tran L T, Knight C G, O’Neill R V (2003). Self-organizing maps for
with the result of fuzzy C-mean clustering. This classi- integrated environmental assessment of the Mid-Atlantic region.
fication system is reasonable with apparent ecological Environmental Management, 31(6): 822–835
 Application of self-organizing neural networks to classification of plant communities in Pangquangou Nature Reserve 517

Wu P Q and Huang M S (2005). The application of SOM on the Zhang J T (1994b). Canonical indicator species analysis, a new
classification of Fujian cities’ functions. Economic Geography, extrinsic classification technique. Acta Phytoecologica Sinica,
25(1): 68–70 (in Chinese) 18(4): 331–337 (in Chinese)
Yang H X, Lu Z Y (1981). Methods of quantitative classification in Zhang J T (2004a). Fuzzy mathematic classification and ordination
plant ecology. Beijing: Science Press (in Chinese) of plant communities in Pangquangou Natural Reserve. Journal
Yu S X (1995). Introduction to mathematical ecology. Beijing: of Beijing Normal University (Natural Science), 40(2): 249–254
Scientific Literature Press (in Chinese) (in Chinese)
Yuan Z R (2000). The artificial neural network and its application. Zhang J T (2004b). Quantitative ecology. Beijing: Science Press (in
Beijing: Qinghua University Press (in Chinese) Chinese)
Zhang F, Zhang J T (2000). Progress of quantitative classification Zhang J T (2005). Canonical principal component analysis of forest
and ordination research in China. Journal of Shanxi University communities in Lishan Natural Reserve, Shanxi Province, North
(Nat Sci Ed), 23(3): 278–282 (in Chinese) China. Journal of Biomathematics, 20(2): 213–218 (in Chinese)
Zhang J T (1994a). A comparison of fuzzy c-mean clustering and Zhang J T, Chen T G (2004). Variation of plant communities along
TWINSPAN. Journal of Wuhan Botanical Research, 12(1): 11– an elevation gradient in the Pengquangou Reserves, North China.
17 (in Chinese) Community Ecology, 5(2): 227–233

View publication stats