TELKOM
NIKA
, Vol. 11, No. 12, Decem
ber 20
13, pp.
7290
~72
9
4
e-ISSN: 2087
-278X
7290
Re
cei
v
ed
Jun
e
18, 2013; Revi
sed
Jul
y
1
2
, 2013; Acce
pted Augu
st 12, 2013
Multidimensio
n
al Data Mining using a K-mean
Algorithm based on the Forest Management Inventory
of Fujian Province, China
Yanrong Gu
o
1
,
Baog
uo Wu*
1
, Yang Liu
2
1
Schoo
l of Information Sci
enc
e and T
e
chno
l
o
g
y
of Beij
in
g F
o
restr
y
Un
ive
r
sit
y
, Beij
in
g 10
008
3, Chi
n
a
2
Ke
y
L
abor
ator
y for Silvic
ultur
e
and C
ons
erv
a
tion of Mi
n
i
stry of Educ
atio
n, Beiji
ng F
o
restr
y
Un
iversit
y
,
Beiji
ng 1
0
0
083
, China
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
w
b
ao
gu
o@
yeah.n
e
t
A
b
st
r
a
ct
T
o
deter
mi
ne r
e
lati
onsh
i
ps b
e
t
w
een stand vo
lu
me a
nd
s
i
te factors in th
e a
b
senc
e of infor
m
ati
o
n
abo
ut stand a
ge an
d de
nsit
y, a classifica
tion patte
r
n
w
a
s establ
ish
e
d
using
a clus
tering a
n
a
l
ysis
alg
o
rith
m an
d app
lie
d to Chi
na fir in F
u
ji
an
Province
. T
h
e
results show
e
d
that slope
p
o
sitio
n
, elevati
on,
elev
ation a
nd
hu
mus d
epth
w
e
re imp
o
rtan
t factors
affecting the stand
volu
mes of youn
g/i
m
mature
forests, near-
m
atur
e for
e
sts, and
matur
e
/o
vermature
f
o
re
sts, respective
l
y
. T
he K-
mea
n
alg
o
rith
m c
oul
d
be use
d
to eva
l
uate the i
n
flue
nces of site fact
ors on stand volu
me und
er different stand
age gr
oups a
n
d
dens
ity cond
itions.
Ke
y
w
ords
:
da
ta min
i
n
g
, K-means a
l
g
o
rith
m, site
factor, forest ma
nag
e
m
e
n
t inventory
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Fore
st re
so
urce d
a
ta play i
m
porta
nt role
s in forest m
anag
ement a
nd de
ci
sion
makin
g
.
Gene
rally, the fore
st resource
data
mainly con
s
i
s
t of nation
a
l fore
st inventory, fore
st
manag
eme
n
t inventory
and inve
stig
ation of fixe
d sa
mplin
g. The
s
e d
a
ta co
ntrib
u
te
to
su
staina
ble fore
st manag
ement, but ru
les fo
r hu
ge datasets hav
e not been d
e
fined. Existing
data ca
nnot
be mined fo
r rule
s, preve
n
t
ing the pre
d
i
c
tion of future trend
s. Ra
pid and effici
ent
data mining h
a
s be
com
e
n
e
ce
ssary to e
nable forest h
a
rvestin
g
.
Knowle
dge i
s
mined an
d
then analyzed from mul
t
iple angle
s
, aiding in deci
s
ion
sup
port, process co
ntrol,
and info
rmati
on ma
n
agem
ent [1]. Beca
use of the
s
e
benefits, dat
a
mining is u
s
e
d
in many ind
u
strie
s
. It has been ap
plied
to urban resi
dential load
s
[2], ecologi
ca
l
environ
ment
comp
en
satio
n
[3], intelligent desig
n sy
stems [4], and
forest
ry [5]. In fore
stry, da
ta
mining techni
que
s ben
efit long-te
rm fore
st manag
eme
n
t.
Chin
a fir i
s
a
n
impo
rtant
conifero
us pla
n
ta
tion tre
e
speci
e
s i
n
Fuji
an Provin
ce,
whe
r
e
the cli
m
ate i
s
ari
d
a
nd
su
b
-
tropi
cal.
Chi
na fir play
s a
n
imp
o
rtant
role b
e
cau
s
e i
t
is th
e mai
n
t
r
ee
spe
c
ie
s for a
fforestation,
providin
g wo
od that ec
on
omically be
n
e
fits the regi
on. We expe
ct to
improve
Chi
n
a fir growth
and effici
en
cy. The
relati
onship
s
bet
ween
stand v
o
lume a
nd
si
te
factors mu
st be define
d
cl
early to ena
bl
e pro
per
man
ageme
n
t. Pre
v
ious pa
pe
rs
have re
porte
d
relation
shi
p
s
betwe
en t
r
ee
growth
an
d
site
con
d
ition
s
[6
-8], but
these a
nalyse
s h
a
ve b
een
unidime
nsi
o
n
a
l.
Ho
wever, in the real p
r
o
c
e
ss of China fir
gro
w
th, the stand volum
e
is affected b
y
the
age, den
sity and site
con
d
ition. Determining the
p
r
odu
ctivity level of China
fir is import
ant
becau
se it p
r
ovide
s
a
ba
sis for thi
nni
ng ma
nag
em
ent. To u
nde
rstan
d
tree g
r
owth,
re
alize
multidimen
sio
nal data a
nal
ysis, an
d find
out t
he rul
e
s, we int
r
od
u
c
ed
data min
i
ng to fore
stry
and fo
re
st m
anag
ement. I
n
this
pap
er,
we
so
ught
to identify re
lationship
s
b
e
twee
n sta
n
d
volume a
nd
site facto
r
s un
der
differe
nt
stand
ag
e a
n
d
de
nsity
con
d
itions.
The
result
s p
r
ovid
e
more a
c
curate deci
s
io
n su
pport for tree
gro
w
th evalu
a
tion in fore
st reso
urce ma
nagem
ent.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Multidim
ensio
nal Data Mini
ng usi
ng a K-m
ean Algorithm
based on
the Fore
st… (Yanro
ng Gu
o
)
7291
2. Materials
and Method
s
2.1. Data
Col
l
ection
In this study, plots we
re i
dentified u
s
in
g
the forest
manag
eme
n
t inventory of Fujian
Province. Plo
t
s of China
fir through
out
the provin
ce
were
sele
cte
d
u
s
ing th
e f
o
llowin
g
su
rvey
data requirements: availability of
data
on the different site condit
ions and afforestation times,
con
s
i
s
tent st
and man
age
ment measu
r
es, an
d re
l
a
tively little
destructio
n
o
f
the stand by
human
s. Threefold stan
da
rd deviation
wa
s use
d
to eliminate ab
n
o
rmal data, a
nd 52,920
Ch
ina
fir sampl
e
plo
t
s we
re chosen for ra
ndo
m sampli
ng a
nalysi
s
. Annu
al data for th
ese pl
ots were
distrib
u
ted a
s
uniformly as
possibl
e.
The m
a
in
su
rvey facto
r
s
were the
co
mpartme
n
t, subplot,
stand
age,
domin
ant tre
e
spe
c
ie
s, tre
e
sp
ecie
s
co
mpositio
n, a
nd
stand
averag
e h
e
ight
. The
comp
onent
s of th
e
environ
menta
l
variabl
es were
co
ntaine
d. The
s
e
va
riable
s
were l
andform; ele
v
ation; slo
p
e
,
slop
e dire
ctio
n, slope po
sit
i
on; soil type, textur
e, and structu
r
e; hu
mus
thickn
ess; stand ag
e;
manag
eme
n
t measure; hea
lth level; site
type; and afforestation time
(195
6–2
006
).
2.2. Data Min
i
ng Frame
w
ork
Figure 1 d
e
p
i
cts the
de
si
gn of the a
s
se
ssm
ent sy
stem, incl
udi
ng data mi
ni
ng. The
asse
ssm
ent pro
c
e
ss wa
s divided
into
t
he
followi
ng st
ep
s: (i) d
a
ta pre
paration
,
(ii) clu
s
teri
n
g
analysi
s
fo
r
data mini
ng,
and
(iii) cate
gori
z
ation
of
volume a
nd
site
con
d
ition
s
fo
r differen
t
stand a
g
e
s
a
nd den
sitie
s
.
Figure 1.
Flow Ch
art of Ch
ina Fir Data Mining
2.3. Data Pre
p
aratio
n
2.3.1. Data
Cleaning
The data
cont
ained
some o
u
tliers, noi
se,
and mi
ssing
or incon
s
iste
nt values. In su
ch
ca
se
s, we re
placed data p
o
ints with me
an
value
s
of the co
rrespon
ding varia
b
le
s.
2.3.2. Data T
r
ansformatio
n
In this stud
y, data tran
sform
a
tion
consi
s
ted of
gene
rali
zati
on and
normalize
d
pro
c
e
ssi
ng.
Gene
rali
zatio
n
processin
g
repl
aced
th
e lo
wer level
s
of
data
obj
ects with
mo
re
abstract
con
c
epts. Stan
d a
ge
wa
s d
e
fin
ed a
s
yo
ung
gro
w
th, mid
d
l
e
-ag
e
d
fore
st
, nea
r-m
ature
forest a
nd m
a
ture. In no
rmalize
d
processing,
attrib
ute data were proj
ecte
d p
r
opo
rtion
a
lly onto
a sp
ecifi
c
sm
all scale. T
h
is pro
c
e
s
s was use
d
in
data
mining to
eli
m
inate d
e
viations
amon
g t
he
different
attribute d
a
ta. T
he di
men
s
ion
s
of
the
attributes were n
o
t co
nsi
s
tent
or compa
r
a
b
le
.
The stan
da
rd
ization meth
o
d
wa
s applie
d to solve
the
proble
m
of non-u
n
ified di
mensi
o
n
s
in all
indicators an
d then com
p
a
r
ed with the a
s
sessme
nt index:
H
ij
′
=
H
ij
-
H
ijm
i
n
/
H
ij
m
a
x
-
H
ij
m
i
n
’
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 729
0 – 7294
7292
Whe
r
e
H
ij
′
is the stan
da
rdization value,
H
ij
is the ob
served value,
H
ijm
ax
is the maximu
m
of all obse
r
ve
d values, an
d
H
ijm
in
is the minimum of all observed val
ues.
2.3.3. Data
Reduc
tion
Much tim
e
is wa
sted o
n
the an
alysi
s
of
large
and
compl
e
x data
s
ets. T
o
avoi
d this
probl
em, d
a
ta redu
ction
method
s m
u
st be
re
searche
d
. Compl
e
x data
s
ets
contai
ning
so
me
correl
ation
ca
n be redu
ce
d
to a few indi
cators that
ful
l
y reflect the
origin
al information an
d a
r
e
indep
ende
nt of one anoth
e
r. In other
words, thi
s
techn
o
logy can
maintain the
integrity of the
dataset while
allowin
g
effici
ent data mi
ni
ng and imp
r
o
v
ing the quali
t
y of results.
Extensive re
search ha
s be
en perfo
rme
d
on e
fficient algorith
m
s th
at can man
a
ge high
dimen
s
ion
a
lity [9-12]. High-dime
nsi
onal
data are o
fte
n transfo
rme
d
into lower-d
i
mensi
onal d
a
ta
by princi
pal compon
ent de
comp
ositio
n [13]. Principal
compo
nent analysi
s
wa
s used for dat
a
redu
ction in t
h
is stu
d
y.
2.4. Cluste
r Algorithms
The dista
n
ce
measu
r
e
was u
s
ed to
comp
ute clu
s
ter
similarity
for most cl
usteri
ng
algorith
m
s. In
data mining, clu
s
terin
g
is
a disc
overy p
r
ocess that g
r
oup
s o
r
com
partme
n
talize
s
a dataset to maximize intra-cl
uste
r and
minimize
int
e
r-clu
s
ter
sim
ilarity. In cluster analysi
s
, the
K-mean al
go
rithm is one of
the most efficient an
d wid
e
ly used met
hod
s in pra
c
ti
ce [13, 14].
The K
-
mea
n
algo
rithm
is initialized f
r
om
some
ra
ndom
or a
p
p
r
oximate
sol
u
tion, a
s
follows [15, 16]: (i) K obj
ects
are
sel
e
cted rand
oml
y
as initial cl
uster
ce
nters from n data
obje
c
ts, (ii) th
e distan
ce of each obje
c
t from the
mean
of each clu
s
tering o
b
ject (clu
ster center)
was calculat
ed and a
new
partition i
s
created
usi
n
g the mini
m
u
m distance, (iii) new
cluster
cente
r
s
are
computed, a
n
d
(iv) ste
p
s
(ii) and
(
iii) a
r
e iterated u
n
til no ch
ang
e occurs in a
n
y
clu
s
t
e
r.
The
spe
c
ifics of the K-m
e
ans
algo
rithm
ar
e
de
scrib
e
d
belo
w
. Ea
ch re
petition a
ssi
gn
s
each point to
its nea
re
st cl
uster, a
nd p
o
i
n
ts bel
ongin
g
to the sam
e
clu
s
ter a
r
e th
en average
d to
derive
ne
w cl
uster cente
r
s.
Each repetiti
on su
cces
siv
e
ly improves
the cl
us
te
r
ce
nters u
n
til the
y
become sta
b
l
e
[10, 13]. The algorith
m
u
s
e
s
the equ
ation:
2
1
i
K
i
iP
C
Ep
m
(
2
)
Whe
r
e
E
i
s
the sum of squared e
r
rors for all
obje
c
ts in th
e dat
aba
se,
p
i
s
the dat
a
matrix, and
m
i
is the cent
roi
d
of cluste
r
C
i
. In the K-mean method, th
e k clu
s
te
r must be kept a
s
c
o
mpa
c
t as
po
ss
ib
le
in
the in
te
r
i
or
o
f
the
c
l
us
te
r, a
n
d
clu
s
ters mu
st be ke
pt as d
i
stant from
on
e
anothe
r as p
o
ssi
ble.
3. Results of Expression
and Visualization
3.1. Dete
rmination of Site Factors
The p
r
in
cipal
com
pon
ent
decompo
sitio
n
metho
d
wa
s u
s
ed
for
d
a
ta re
du
ction
in this
study.
Th
e eigenvalue
s of
the eight main co
m
pone
nts excee
d
e
d
1, and the accu
mulativ
e
contri
bution
rate rea
c
he
d
86.17% (T
abl
e 1). The
ma
in comp
one
nts we
re la
ndf
orm, elevatio
n,
slop
e, slop
e positio
n, exposu
r
e, soil
type, humu
s
de
pth, and soil t
h
ickne
s
s.
Table 1. Statistics of the Main Com
pon
e
n
ts
Main component
Eigenvalue
Contribution r
a
te
Accumulative co
ntribution rate
1 2.98275402
18.64
18.64
2 2.35229036
14.70
33.34
3 2.11518018
13.22
46.56
4 1.88209686
11.76
58.33
5 1.63179546
10.20
68.53
6 1.23453778
7.72
76.24
7
8
1.02892308
1.00982300
6.43
3.50
82.67
86.17
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Multidim
ensio
nal Data Mini
ng usi
ng a K-m
ean Algorithm
based on
the Fore
st… (Yanro
ng Gu
o
)
7293
3.2. Classific
a
tion Dete
rmination
Acco
rdi
ng to importa
nt level, the orders in
which the
eight main compon
ents a
ffected
stand
volume
we
re:
slop
e
positio
n, sl
op
e, expo
sure,
soil thi
c
kne
ss, elevation, h
u
mus de
pth,
soil type, an
d landfo
r
m fo
r young fo
re
sts; elevati
on,
slop
e po
sitio
n
, soil thi
c
kn
ess, expo
sure,
humu
s
d
epth
,
slop
e p
o
siti
on, lan
d
form,
and
soil typ
e
for i
mmatu
re timb
er; el
evation, hum
us
depth,
soil
thickne
ss,
soil
t
y
pe, landfo
r
m
,
slop
e
po
sition, expo
su
re,
and
slope
fo
r n
ear-matu
r
e
forest
s; an
d
humu
s
d
epth
,
elevation, soil thickn
e
s
s, slop
e p
o
sitio
n
, soil type, l
andform, slo
pe,
and expo
su
re
for mature a
nd overm
a
ture forest
s (T
a
b
le 2).
Table 2. Cate
gori
z
ed
Re
sul
t
s of Relation
ship
s bet
wee
n
Site Factors and Volu
me
for Different
Ages an
d De
nsitie
s
Age
groups
Densit
y
(tree
·
ha
-
1
)
Landform
Elevation
Slope
Slope
position
Exposure
Soil
ty
p
e
Humus
depth
Soil
thickness
Y
oun
g
forest
1000-
4500
— -0.025
-0.073
-0.076
-0.049
-0.005
0.013
0.037
Immature
timber
600-
4200
-0.068
-0.257
-0.069
-0.241
-0.158
-0.029
0.076
0.237
Near-
mature
forest
450-
3600
-0.169
-0.345
-0.010
-0.100
-0.097
-0.222
0.317
0.249
Mature/
overmature
forest
450-
3300
-0.185
-0.419
-0.144
-0.230
-0.123
-0.189
0.505
0.390
4. Discussio
n
and concl
u
sion
Stand den
sit
y
was
neg
atively correlat
ed with
fo
re
st illumination
and temp
erature in
forest
s of different de
nsitie
s, but tree growth
was p
o
sitively asso
ci
ated with the
s
e compo
nen
ts.
Becau
s
e th
e
den
sities
of young fo
re
sts
and imm
a
ture
timber
were
high, growth
of Chin
a fir was
limited mainl
y
by illuminat
ion and temperature. Co
nsistent with t
h
is
situation,
exposure and
slop
e po
sitio
n
impa
cted f
o
re
st illumin
a
t
ion and
phot
osynthe
s
i
s
, there
b
y affecti
ng tre
e
growt
h
.
Although
gen
tle slop
es (<25°) were
co
ndu
cive to
China fir
growth [17], sam
p
le plot
s
with
slop
es
>25
°
accou
n
ted fo
r 83.23% of
young fore
st
s. Thus, slo
p
e
may be a key factor in the
stand volum
e
of young fore
st.
The density of
immature timber was
suffici
ently high and then
result
in
illumi
nation
affected the t
r
ee g
r
o
w
th. T
he illumin
a
tio
n
of uppe
r
sl
ope
s was a
d
equate,
whi
c
h pro
m
ote the
tree g
r
o
w
th
(Tabl
e 2). Sl
ope p
o
sition
wa
s a
key
factor influ
e
n
cin
g
the
stand volum
e
of
immature timber.
The
den
sitie
s
of
ne
ar-ma
t
ure, matu
re,
and
ove
r
ma
ture fo
re
sts
were
relativel
y
low.
Such sta
n
d
s
sho
u
ld have
adeq
uate illu
mination an
d
abunda
nt sh
rub
s
, herb
s
, and fore
st litter.
The
soil
co
nd
ition is t
he m
o
st imp
o
rta
n
t
dire
ct fa
cto
r
b
e
ca
use
soil t
h
ickne
s
s influ
ences the
roo
t
sy
st
em
ca
pa
cit
y
and
f
e
rt
ili
zer
ab
so
rpt
i
o
n
.
Thu
s
,
s
o
il t
h
ic
kne
s
s
wa
s
a
key
f
a
ct
or
inf
l
uen
cing t
h
e
stand volum
e
s of nea
r-mat
ure, matu
re, and overmatu
re fore
sts.
The elevatio
ns of youn
g forest
s, i
mmature tim
ber, nea
r-m
ature fore
st
s, and
mature/ove
rmature
fore
st
s
were 1
40–
990 m,
108
–
1225
m, 15
0
–111
5 m, a
n
d
15
0–12
25
m,
respe
c
tively. There a
r
e
sig
n
ificant
differences in
th
e
stand
volume
amon
g the
fore
st types f
o
r
four age g
r
o
u
p
s (T
able 2
)
. Thus, a
n
elevation had a la
rge imp
a
ct on
the growth of
China fir.
In this study,
because all
plots
were in
low an
d mid
d
le mou
n
tain
area
s, lan
d
form did
not signifi
can
t
ly influence tree g
r
o
w
th. If the study
plots had b
een
in different a
r
ea
s, su
ch
a
s
hills
and l
o
w,
middle,
and
high
mou
n
ta
in area
s,
the
re
sults woul
d have
differed. In lo
w
an
d
middle m
ount
ain a
r
ea
s, pl
ant re
sid
u
e
s
decompo
se
d rapidly und
er
mild conditio
n
s,
p
r
od
uci
n
g
loose an
d fert
ile soil. Th
ese are
a
s
had
greate
r
a
nnu
al rainfall
s, which
enh
ance
d
tree g
r
o
w
th
. In
high m
ountai
n area
s, tre
e
gro
w
th
wa
s i
nhibited
by lo
wer temp
erat
ure
s
a
nd
hig
her
evapo
rati
on
rates.
In con
c
lu
sion
, under
different stan
d ag
e gro
u
p
s
an
d
den
sity cond
itions, sta
nd
volume
wa
s influen
ce
d by the different com
pone
nts of si
te factor for the yo
ung, immatu
re, near-matu
r
e
and matu
re/o
vermature forests of China
fir. Giv
en the relation
ship
s betwee
n
sta
nd volume an
d
basi
c
tree fa
ctors, su
ch a
s
diamet
er, tree heig
h
t, crown
width, a
nd bra
n
ch he
ight, we sh
ou
ld
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 729
0 – 7294
7294
con
s
id
er th
at
site fa
ctors al
so
affected
th
ese
ba
si
c fact
ors.
To fu
rthe
r evalu
a
te the
influen
ce
s of
site facto
r
s
o
n
tree h
e
ight
and dia
m
et
er un
der th
e
same
age
grou
ps
but d
i
fferent den
si
ty
con
d
ition
s
, stand de
nsity should b
e
divided into mo
re
classe
s.
Ackn
o
w
l
e
dg
ement
This stu
d
y was supp
orted
by the national
natural scien
c
e foun
d
a
tion-fun
ded
proje
c
t
(No. 31170513) and
the
national high tec
h
nology res
e
arc
h
and development program (863
prog
ram
)
(2
0
12AA102
003
).
Referen
ces
[1]
PN T
an, Steinbach M, Kumar V. In
troduction
to data mini
ng.
W
P
Co. 2006.
[2]
Y He, Y W
ang,
T
Luo, A He, J W
ang. Urba
n
Resi
d
entia
l Lo
ad Com
b
in
ed
F
o
recast Mode
l Based
on
Data Mi
ni
ng T
e
chn
i
qu
es
and
Pan
e
l
Data T
heor
y.
J
ourn
a
l
of Co
mputati
ona
l Inform
ation System
s
.
201
0; 6(6): 180
1-18
08.
[3]
JQ Xi
an
g. Res
earch
on
min
i
n
g
d
e
vel
opm
ent
in
yu
n
nan
u
n
d
e
r ec
olo
g
ica
l
e
n
viro
nment c
o
mpens
ation.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(5): 2
853-
285
9.
[4]
T
Jing, Y Yuan. Intelli
ge
nt desig
n s
y
ste
m
of
mechan
i
c
al pro
ducts
base
d
on
dat
a mini
ng a
nd
kno
w
l
e
d
ge ba
sed eng
in
eeri
n
g.
Jour
nal
of
T
heoretic
al
an
d Ap
pli
ed I
n
fo
rmati
on T
e
chn
o
lo
gy
. 20
12;
46(1): 23
7-2
4
4
.
[5]
CP Ch
en, BG
W
u
, YG Jia , D
D
Lu.
A Stud
y
on A
ppl
yi
n
g
T
e
chni
ques
of D
a
ta Mini
ng
in
F
o
rest in F
r
om
Ati
o
n
Ma
na
geme
n
t.
He Bei J
ourn
a
l of F
o
res
t
ry and Orchar
d Rese
arch
. 20
04; 19(2): 1
49-
153.
[6]
JC Lu. Affection of terrain fac
t
ors on gro
w
th
of
Plantation
Chin
ese fir in
hig
h
altitud
e
mounta
i
ns i
n
eastern F
u
j
i
an
Provinc
e
.
Jour
nal of F
u
ji
an F
o
restry Scienc
e and T
e
ch
no
l
ogy
. 200
6; 33(
2): 120-1
28.
[7]
J Hua
ng, W
M
i
n, CC
Ca
i, S
H
Lu. Effects
of di
ffere
nt
de
nsities on
t
he cunn
ing
ham
ia lanc
eol
ata
i
n
middl
e ag
e.
Ap
plicati
on of Sta
t
istics and Man
age
ment
. 200
6
;
25(1): 111-1
1
6
.
[8]
JH Men
g
. Buil
din
g
forestr
y
d
a
ta
w
a
r
eho
use
fo
r forest man
agem
ent e
n
ter
p
rise
and th
eir
app
licati
o
n
PhD diss
ertatio
n
.Chi
na: Beij
in
g F
o
restr
y
U
n
iv
ersit
y
, Bei
j
i
ng. 201
1.
[9]
CC Ag
gar
w
a
l,
PS Yu. F
i
ndi
ng
gen
era
lize
d
proj
ected
cl
u
s
ters in
hi
gh
dime
nsio
nal
s
paces. A
C
M.
200
0.
[10]
Hinn
e
b
u
rg, K
e
i
m
DA. Optima
l
gri
d
-cluster
ing
:
T
o
w
a
rds
br
ea
king
the
curse
of dim
ensi
o
n
a
li
t
y
i
n
hi
gh-
dime
nsio
nal cl
usterin
g
. Cites
eer. 199
9.
[11]
C Ordonez. Cl
usterin
g
bin
a
r
y
data streams w
i
t
h
K-means.
In, Proceedi
ng
s of the 8th ACM SIGMOD
w
o
rksh
op on R
e
searc
h
issues
in data min
i
n
g
and kn
o
w
l
e
d
g
e
discover
y
. 20
0
3
; 12-19.
[12]
H W
e
i, XJ Li, Y Guan, et
al. On the model
checki
ng of th
e space
w
i
r
e l
i
n
k
interface.
TELKOMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(2): 7
40-7
46.
[13]
C Ding,
X He.
K-means cl
usterin
g
via pri
n
ci
pal
com
pon
ent
anal
ys
is. In, Procee
din
g
s of the t
w
ent
y-
first internatio
n
a
l confer
enc
e on Mach
ine l
e
a
r
nin
g
. ACM. 2004.
[14]
JA Hartiga
n
, MA W
ong. Algori
t
hm
AS 136: A k-means clust
e
rin
g
alg
o
rithm
.
Applie
d statistics
. 1979;
100-
108.
[15]
AK Jain, Dub
e
s
RC. Algorith
m
s for clus
terin
g
data. Prentic
e-Hal
l
, Inc. 198
8.
[16]
AK
Jain. Data clusteri
ng:
5
0
ye
ars
b
e
y
o
nd K-means.
Pattern Recogni
tion Letters
.
201
0
;
31(8): 6
51-
666.
[17]
HB Z
hang. F
u
ji
an F
o
rest. Be
iji
ng: Chi
na F
o
re
str
y
Press.19
9
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.