TELKOM
NIKA
, Vol. 13, No. 4, Dece
mb
er 201
5, pp. 1376
~1
383
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i4.2272
1376
Re
cei
v
ed
Jul
y
3, 2015; Re
vised Septem
ber
18, 20
15;
Accept
ed O
c
tober 2, 201
5
Poisson Clustering Process o
n
Hotspot in Peatland
Area in Sumatera
Annisa Pus
p
a Kirana
*1
, Imas Sukaesi
h Sitanggan
g
2
, Lailan Sy
aufina
3
1,2
Department of Computer S
c
ienc
e,
F
a
cult
y of Natural Sci
ence a
nd Math
ematics, Bogor
Agricultur
al
Univers
i
t
y
, Ind
ones
ia
3
Departme
n
t of Silvicult
u
re, F
a
cult
y of F
o
res
t
r
y
, Bogor Agr
i
cultura
l
Univ
er
sit
y
, Indo
nesi
a
*Corres
p
o
ndi
n
g
author, em
ail
:
puspakir
an
a
@
ap
ps.ipb.ac.i
d
1
, imas.sitang
gan
g@i
pb.ac.i
d
2
,
lail
ans@
i
p
b
.ac.id
3
A
b
st
r
a
ct
T
he incre
a
se i
n
pe
atlan
d
fire
’s intensity h
a
s
enco
u
rag
ed
pe
opl
e to dev
elo
p
meth
ods of pr
eventi
n
g
wi
l
d
fi
re
. On
e of th
e
p
r
e
v
en
ti
on
m
e
tho
d
s
i
s
re
co
gn
i
z
ing
the
distri
butio
n
pa
ttern of
hotspo
t
as o
n
e
of for
e
s
t
and l
a
n
d
fire i
ndic
a
tors. W
e
coul
d deter
min
e
the ar
ea tha
t
has hig
h
fire
s density b
a
se
d on d
i
strib
u
ti
o
n
patterns so
a
n
y early
prev
e
n
tion ste
p
s co
uld
be p
e
rf
or
me
d in th
at a
r
ea. T
h
is rese
arch pr
opos
ed
to
recog
n
i
z
e
th
e
distrib
u
tion
pat
tern of hotsp
ot
clusters
in t
h
e
peatl
and
are
a
s
in Su
matera
in the y
ear 2
0
14
usin
g Kul
l
d
o
rff’
s Sc
an Statis
tics (KSS) me
thod w
i
th
Poi
sson
mo
del.
This ap
proac
h
w
a
s specific
all
y
desi
gne
d to
d
e
t
ect clusters
an
d ass
e
ss th
eir
signific
anc
e v
i
a Mo
nte
Carl
o
replic
at
io
n. R
e
sults sh
ow
ed t
hat
the
meth
od
is r
e
lia
bl
e to
dete
c
t the clusters
of hot
sp
ots w
h
i
c
h hav
e th
e ac
curacy of
95%.
Ria
u a
nd S
out
h
Sumatera
prov
ince
hav
e the
hig
hest d
ensity
of cluste
r
distr
i
buti
ons
of the
hotspot. Bas
e
d
on th
e
maturit
y
level
of
pe
at, cl
uster d
i
stributi
o
ns of
hotsp
ot w
e
re
most
l
y
foun
d
i
n
‘
hem
i
c
’
ma
tu
ri
ty le
ve
l
.
Ba
se
d on
pe
a
t
lan
d
thickness, clust
e
r distrib
u
tion
of hotspot
w
a
s mostly fo
und i
n
‘
v
ery d
e
e
p
’
thi
ckness.
Ke
y
w
ords
: clu
s
tering, hotsp
o
t, peatlan
d
, Po
i
sson proc
ess, scan statistics
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Indone
sia
faces
a
se
riou
s
probl
em
of fo
rest
an
d lan
d
fire
every ye
ar. Th
at ki
nd
of fire
prod
uces
ma
ssive
smo
g
a
nd carb
on p
o
llutions
whi
c
h
lead to the
decrea
s
e
of health a
nd th
e
environ
menta
l
damage [1]. There a
r
e two kin
d
s of fa
ctors which trigge
r the occurre
n
ce of that
fire: huma
n
f
a
ctors
and
n
a
tural fa
ctors. Based
on
t
he data
[2], in mo
st of th
e fire
s cases are
cau
s
e
d
by human facto
r
s, where
a
s a
small fracti
o
n
of the cases is cau
s
ed
by the natural
factors. Co
nsi
derin
g the b
a
d
impa
ct on f
o
re
st and l
a
n
d
fire an
d its
variou
s trig
ge
ring fa
ctors, it is
very import
a
n
t
to develop
an ea
rly wa
rning
system
t
o
prevent forest an
d lan
d
fire. Ho
weve
r, in
orde
r to
deve
l
op the
syste
m
, we first n
eed to
re
cog
n
ize th
e di
stri
bution p
a
ttern of hot
spot
s as
one of fo
re
st
and la
nd fire
indicators. By re
cog
n
izi
ng t
hat pattern,
we a
b
le
dete
c
t the a
r
ea
th
a
t
has hi
gh fire
s den
sity so
any ear
ly pre
v
ention step
s can be p
e
rf
orme
d in tha
t
area. Severa
l
studie
s
have
been
cond
u
c
ted
on d
e
velopin
g
an
e
a
rly warni
n
g
system
for
fire p
r
eventio
n.
Sitanggan
g e
t
al (201
4) in
[3] ap
plieda
spatial
de
ci
si
on tree
algo
ri
thm on
spati
a
l data
of fo
rest
fires for predi
cting
hotspot
s o
c
cu
rre
nce
s
. Hotsp
o
t o
c
curre
n
ce a
s
an in
dicator o
f
fore
st an
d l
and
fires is
essent
ial in developi
ng an ea
rly
warnin
g syste
m
for fire pre
v
ention.
This work used clu
s
teri
ng
method to reco
gni
ze the
distributio
n pattern of the hotspot
occurre
n
ce. Clu
s
terin
g
i
s
a process of
grou
ping
dat
a int
o
cla
s
s
e
s o
r
cl
ust
e
r
s
,
su
ch t
h
at
ob
ject
s
within
a
clu
s
ter
ha
s hi
gh
simila
rity in
compa
r
ison to
one
an
othe
r, but a
r
e ve
ry dissimila
r t
o
obje
c
ts in
ot
her
clu
s
te
rs [
4
]. The
r
e a
r
e several
po
pular metho
d
s
p
r
op
osed
b
y
resea
r
chers to
perfo
rm
such
hotsp
ot pattern
re
cog
n
ition, incl
udi
n
g
hiera
r
chy, p
a
rtition, mod
e
l ba
sed, d
e
n
sity,
and
grid
ba
se
d cl
uste
ring
[5]. In [6], Sitangga
ng
et al
develop
ed a
web
ba
se
d O
L
AP appli
c
ati
o
n
for hot
spot
clusteri
ng in
Indon
esi
a
u
s
i
ng the K-me
ans
algo
rith
m. The K-m
ean
s alg
o
rith
m is
partition
ba
sed
clu
s
terin
g
method.
Usman [7]
appl
ied h
o
tsp
o
t clusteri
ng i
n
Sumatera
in
the
perio
d of 2
0
0
2
and
201
3 b
a
se
d on
the d
ensity ap
proa
chu
s
in
g Den
s
ity-Based S
p
atial Clu
s
te
rin
g
.
Sitanggan
g [8] developed
hotspot pred
iction mod
e
ls
using de
ci
si
on tree alg
o
ri
thms.All of these
previou
s
ly d
e
scrib
ed
wo
rks only p
r
o
c
ess the
sp
atial dom
ain i
n
formatio
n o
f
hotsp
ot da
ta,
whe
r
ea
s tho
s
e kind of data
may also co
ntain the ti
me domain information. Therefore, it could
be
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 137
6 – 1383
1377
benefi
c
ial to
develop
a me
thod that ta
kes into
both
spatial an
d te
mporal do
mai
n
inform
ation
on
hotsp
ot
cl
ust
e
ring.In
thi
s
rese
arch, we apply st
atisti
cal
app
roa
c
h
to re
co
gni
ze
the di
strib
u
tion
pattern of hots
pot in both s
patial and tempor
al domain us
ing Kulldorff’s
Scan
Statis
tic
(KSS)
method with
Poisson mo
d
e
l [9]. The basi
c
idea i
s
that there is
scan
ning
wind
ows that moves
across spa
c
e
and/or time. For each location and si
ze of the wind
ow, the num
ber of ob
serv
ed
and
expe
cte
d
cases is n
o
ted. Every
circula
r
scan
ning
a
windo
w
will find t
h
e likelihoo
d
ratio
usin
g Poi
s
so
n mod
e
l. Th
en evalu
a
tion
of the
st
atistical
significa
nce
of the
cl
uster calculat
ed
usin
g Monte
Carl
o sim
u
lat
i
on. The regi
on with t
he
h
i
ghe
st value
of likeliho
od ratio is the a
r
ea
that has the
most potent
ial clu
s
ters. There are
several relate
d re
sea
r
che
s
about dete
c
ting
clu
s
ters of
spatial a
nd te
mporal dim
e
nsio
n d
a
ta,
e
s
pe
cially in
the forest
ry field. Fo
r exa
m
ple
Devis et
al (2
008)
in
[10] a
pplied scan
statisti
c meth
o
d
to ide
n
tify and a
nalysed
spatiote
mpo
r
al
clu
s
ters in 19
97–2
003 fire
seq
uen
ce
s of
Tuscany re
gi
on.
The pu
rpo
s
e
of this re
sea
r
ch is to
re
co
gnize
the di
stribution
patte
rn of hot
spot
clu
s
ters
in peatland
areas in Su
matera in the year 2
014 using KSS method with
Poisson models.
We
analysed th
e re
cog
n
ize
d
distri
butio
n pattern
o
f
hotspot
cl
usters b
a
se
d on p
h
ysi
c
al
cha
r
a
c
teri
stics of peatla
nd
. The physi
cal cha
r
a
c
teri
stic of peatlan
d
that we u
s
e is the matu
rity
level of peat
, thickne
s
s o
f
peat, and land u
s
e of
peat. Peatla
nd in Sumat
e
ra
categ
o
ri
sed
according
to t
he level
of m
a
turity consi
s
t
of thr
ee
kin
d
s
the
r
e
are fi
bric,
hemi
c
, a
nd
sap
r
ic.
Fib
r
ic
is the early
stage of pea
t decomp
o
siti
on wh
ere
re
cog
n
izable pl
ant fires do
minate. Hemi
c is
interme
d
iate
stage
of p
e
a
t de
comp
osition, betwee
n
fibri
c
a
n
d
sa
pri
c
. Th
e
n
, sa
pri
c
i
s
the
advan
ced
sta
ges of p
eat
d
e
com
p
o
s
ition
into o
r
g
ani
c-matter rich
’e
arth’ without
visible
fires
[
11].
The matu
rity level of peat and thickne
s
s i
ndi
ca
ted in the categ
o
ry of peat. Suppo
se,
’Hemi
c
/Sapri
c
(60/4
0
), very deep’ mea
n
s that Hemi
c/Sapri
c
is th
e maturity level of peat. Va
lue
(60/40
)
sh
ows the
a
r
ea
covering
60
p
e
rcent
of
he
mic
and
40
percent
of
sa
pric.
’Very
d
eep’
mean
s that in the catego
ry of
the thickn
ess of peat with a
depth 400-8
00 cm. T
he thickne
s
s of
peat can b
e
grou
ped
to: 1) Ve
ry sh
a
llow thi
c
kne
s
s (D0
)
(havi
ng a thi
c
kne
ss
<5
0 cm)
with
maturity level
Hemi
c/Sap
r
i
c
, Hemic/Mi
n
e
ral
s
a
nd Sa
pric/
H
emi
c
, 2
)
Shallo
w thi
c
kne
s
s (D1
)
(50-
100
cm
with
maturity lev
e
l Fib
r
ic/Sap
ric,
Hemi
c/S
apri
c
, Hemi
c/mineral, Sa
pric/
H
emi
c
a
nd
Sapri
c
/Mine
r
al, 3) Mode
rate thickne
s
s (D2)
(10
0
-200 cm) wit
h
maturity level of peat are
Hemi
c/Sapri
c
, Hemic/Mi
ne
ral, Sapri
c
, S
apri
c
/He
m
ic
and Sap
r
ic/M
ineral, 4
)
Dee
p
thickne
s
s (D3)
(200
-4
00
cm
with matu
rity level of peat are
He
mi
c/Sapric, Sap
r
ic an
d Sapri
c
/Hemic, 5
)
V
e
ry
Dee
p
thickn
ess (D4
)
(4
00-8
00
cm) with matu
ri
ty level of peat are Hemic/Sap
r
ic
an
d
Sapri
c
/Hemi
c
[11]. The ki
nd ofland
uses d
e
tecte
d
in peatlan
d
a
r
ea in Sumat
e
ra
con
s
i
s
t o
f
25
types [11].
The be
nefits
of this re
se
arch a
r
e a
s
e
a
rly
warni
ng
system an
d ea
rly detection
o
f
forest
and l
and fi
re,
espe
cially in
the p
eatland
are
a
in
Sum
a
teraby
providing illu
strations of
clu
s
tering
pattern
of h
o
tspot
ba
sed
on
the
sp
atial an
d tem
p
oral
a
s
pe
ct.
We vi
suali
s
e
d
the
re
sult
of
clu
s
terin
g
pat
tern by devel
oping p
r
o
ne
area
map.
Th
e rem
a
inde
r
of this pap
er
is organi
ze
d as
follows. In section 2, we
describ
e the methodol
o
g
y and algo
rithm used in
this re
sea
r
ch. In
se
ction 3, we
discuss the
analysi
s
of th
e re
su
lts o
b
ta
ined from
ap
plying these
methodol
ogie
s
in
the given dat
aset
s. Finally, we co
ncl
ude
this pap
er in
se
ction 4.
2. Rese
arch
Metho
d
2.1. Stud
y
Area and For
e
s
t Fires
Data
This
work cl
u
s
ters hot
spot
dataset on
p
eatland a
r
ea
in Sumatera in the year 2
014 by
con
s
id
erin
g b
o
th sp
atial an
d tempo
r
al d
o
main. Total
peatlan
d
in In
done
sia i
s
ab
out 20.6 milli
on
hecta
re
s
whi
c
h 3
5
% of p
eatland
are l
o
cate
d in Su
matera
[11]. Data
sets that
we
use in t
h
is
resea
r
ch i
s
h
o
tspot
data,
esp
e
ci
ally in
the pe
at
land
area
in
Sum
a
tera
Isla
nd i
n
the ye
ar 20
14.
Hotspot data are obtained from
FIRMS MODIS Fire/Hotspot, N
ASA/Univers
i
ty of Maryland.
Peatland
dist
ribution
data
obtaine
d fro
m
Wetland
In
ternation
a
l. S
patial d
o
main
of hot
spot
d
a
ta
refers to
the l
o
catio
n
of
hot
spot fire from
longi
tu
de a
n
d
latitude
fiel
ds.
While, te
mporal refers to
the occurre
n
ce time of fire from date field
s
.
Once the
dat
a are gath
e
re
d, we first pe
rform
pre
-
p
r
o
c
e
ssi
ng d
a
ta
whi
c
h
con
s
i
s
t of two
kind
data the
r
e are
hotspot
and pe
atland
data.
Hot
s
po
t data pre
p
ro
ce
ssi
ng
can
be de
com
p
o
s
ed
into four
step
s. The
first
st
ep of p
r
ep
ro
cessing
hotsp
ot data i
s
sel
e
cting i
m
po
rtant attribute
s
for
the cl
uste
rin
g
process; the
r
e a
r
e
l
a
titude,
longitu
de
co
ordin
a
te
of th
e hot
spot
an
d
date
of
hotspot
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Poisson Cl
ust
e
ring Process on Hots
pot in Peatland Area in Sum
a
tera
(Anni
sa Pu
spa Kirana
)
1378
occurre
n
ce. The
se
cond
step of prepro
c
e
ssi
ng d
a
ta
is sele
cting of
hotsp
ots in t
he no
n-p
eatl
and
area
and h
o
t
spot in the
peatland
area. The thir
d step of p
r
epro
c
e
s
sing
is selectin
g
the
distrib
u
tion l
o
cation
of the
hotsp
ot in
ea
ch
distri
ct o
r
city. The final
step
is loadi
ng info
rmatio
n
into the
data
base. Pre
p
ro
ce
ssi
ng
of p
eatland
dat
ai
s
cal
c
ulate
the a
r
eal
pe
r
kilomet
r
e
sq
u
a
re
based
on th
e
physi
cal
cha
r
acte
ri
stic i
n
formatio
n of
p
eat, incl
udin
g
the m
a
turity
level of p
eat,the
thickne
ss of p
eat, and land
use of pe
at.
2.2. Kulldorf’
s Scan Sta
t
istic Me
thod
S
c
an
st
at
ist
i
c ca
n det
e
c
t
incr
ea
ses
o
f
case
s
in th
e local tem
p
oral an
d (o
r) spatial
dimen
s
ion
s
,
examine
whe
t
her this i
n
cre
a
se i
s
d
ue to
rand
om vari
a
t
ion and
dete
c
t the existe
n
c
e
of clustering in a
cert
ain region
as well
as the preci
s
e posi
tion the cluster [12].
KSS method
is
the extens
ion of
s
c
an
s
t
atis
tic
.
KSS
us
es
larg
e collec
t
ions
of overlapping s
c
an windows
to
detect cl
uste
rs, both the l
o
catio
n
and t
he si
ze, an
d
evaluate the
i
r sig
n
ifican
ce [9]. Scanni
ng
wind
ows are colle
ction
s
in the ar
ea that potentially co
ntains g
r
ou
ps
of hotspots
on it. Basicall
y,
scan statisti
cs works
by compa
r
ing a set
of
event
s
occurrin
g in
si
de a
scan
nin
g
wi
ndo
w
with
those that lie outside. KSS employs a large
coll
ec
ti
on of ci
rcular scanning wi
ndows to detect
clu
s
ters, as ill
ustrate
d
in Fi
gure
1.
G
is t
he wh
ole a
r
e
a
of study,
Z
is a ci
rcula
r
scan
ning
wind
ow,
µ(
Z
)
is
the
to
tal population
which i
s
located in a circular sca
nning
windo
w,
µ(
G
)
is all the total
popul
ation which a
r
e in t
he area of st
udy,
n
z
is th
e numb
e
r of
ca
se
s in a
circula
r
sca
n
n
ing
wind
ow
Z
,
n
G
is th
e n
u
mb
e
r
of
ca
se
s i
n
t
he
study a
r
ea
of
G
,
p
i
s
the
avera
ge
rate
of o
c
curren
ce
event in a
ci
rcula
r
scanni
n
g
a
wind
ow a
nd
q
i
s
th
e ra
te of the eve
n
t outsi
de a
circul
ar scan
ni
ng
wind
ow [13].
Figure 1. Study area and circul
ar
window illustration i
n
KSS method [13]
In this res
e
arc
h
,
we
use the KSS method wi
th a Poiss
o
n model to determine the value
of the poss
ibility s
c
a
nning
window ratio
(lik
eli
hood
rat
i
o) [9]. KSS
with a Pois
son model
used to
comp
are the
numbe
r of
ca
se o
c
curre
n
ces in
and
o
u
t
sca
nnin
g
wi
ndo
w. It use
d
to se
arch f
o
r
clu
s
ters of hotspot in this rese
arch a
n
d
fo
llowing th
e pro
c
e
ss in
homog
ene
ou
s Poisson. Unde
r
the null hypot
hesi
s
of spati
a
l and tem
p
o
r
al rando
mne
ss, the
s
e
events are di
strib
u
ted a
c
cordi
n
g
to a kno
w
n
discrete
-state
ran
dom
pro
c
e
ss
(Poi
sso
n
or B
e
rn
oull
i
), whi
c
h
pa
rameters
can
be
estimated. Gi
ven this assu
mption,
it is possible to test whether
o
r
not the null hypothesi
s
hol
ds
in a s
p
ec
ific
area [9].
Equation
1 i
s
used to
cal
c
ulate
the fu
nction
of the
ratio
of the
possibility for ea
ch
a
circula
r
scan
ning win
d
o
w
Z
with Poisson model. Th
e
n
z
is
the numbe
r of cases in the circula
r
scanni
ng win
dow
Z
,
e
z
is
the expe
cte
d
ca
se
s in
the ci
rcula
r
scan
ning
win
d
o
w
Z
,
n
G
i
s
the
numbe
r of ca
se
s in the stu
d
y area of
G
[15].
λ
Z
n
Z
e
Z
n
Z
.
n
G
-n
Z
n
G
-e
Z
n
G
-n
Z
, if
n
Z
>e
Z
(1)
1, otherwise
The
e
Z
ca
n
be
calculated
usin
g Equ
a
t
i
on 2 i
n
which the
µ (
Z
)
i
s
total p
opul
ation in
scanni
ng win
dow
Z
,
n
G
is the total
num
b
e
r
of cases in
study
area
G
,
µ (
G
)
is the
total num
ber
of
popul
ations i
n
study are
a
G
[15].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 137
6 – 1383
1379
e
Z
μ
Z
n
G
μ
G
(2)
The m
o
st li
ke
ly clu
s
ter i
s
calcul
ated fo
r
each
simulat
ed d
a
taset in
exactly the
same
way
as
the real d
a
ta.Statistical
signifi
can
c
e is
ev
alu
a
ted
usin
g Monte
Carl
o hypoth
e
si
s testin
g [9].
Monte Carlo
simulatio
n
wa
s pe
rform
ed
with at leas
t 9
99 re
plicatio
n
s
und
er the n
u
ll hypothe
sis. If
the cl
uste
r of
intere
st
cont
ains mo
re th
an 9
5
% of t
he repli
c
atio
ns, the
cl
ust
e
r i
s
said
to
be
signifi
cant at
the 95 % l
e
vel [9]. In the
impleme
n
tation of thi
s
research,
we uti
lized
clu
s
te
ri
ng
packa
ge
s
of data
in R sta
t
istical
to
ols, and
Po
stg
r
e
SQL for data
base ma
nag
ement. We u
s
e
‘spatial
epi’ R
packa
ge dev
elope
d by [14] to cluster d
a
t
aset.
3. Results a
nd Discu
ssi
on
3.1. Distribu
tion of Ho
tsp
o
t in the Pea
t
land Ar
ea in
Sumatra in the Year 201
4
The n
u
mb
er
of hotspots i
n
the ye
ar
2
014
am
ou
nte
d
to 3
9
4
07
hotsp
ots.
Wh
ile in th
e
peatlan
d
are
a
amou
nted to 26 19
3 hot
spot
s. The n
u
mbe
r
of hot
spot
s is
scattered i
n
an a
r
ea of
peatlan
d
in the year
201
4
in Sumatra
can
be
seen
in Figu
re 2.T
he hig
hest d
ensity of hot
spot
based on th
e
maturity level of peat in Sumatera in
th
e year 20
14 l
o
cate
d in the
maturity level of
’Hemi
c
/Sapri
c
(60/4
0
), ve
ry deep’
with
average d
e
n
sity value 7
.
913/km
2
. Th
e highe
st de
nsity
based on the
thickn
ess of
peatland in
Sumatera
fro
m
2001 to 2
014 domi
nat
ed by very deep
thickne
ss
whi
c
h rea
c
he
d 4
.
016/km
2
. La
nd u
s
e that d
e
tected i
n
th
e peatla
nd a
r
ea in Sum
a
te
ra
there a
r
e 2
5
types. The hi
ghe
st den
sity base
d
on th
e land u
s
e of
peatland in
Sumatera i
n
the
perio
d of 201
4 dominate
d
by swam
p forest whi
c
h rea
c
he
d 1.425/
km
2
.
Figure 2. Nu
mber of hot
sp
ot non-p
eatla
nd are
a
s in S
u
matera in the year 201
4
3.2. Cluste
r Distribu
tion
s of Ho
tspo
t
in Sumatera
in the Year 2
014
Hotspot sp
re
ad in 39 districts in Sum
a
tera
with to
tal area
s of peatlan
d
dist
ribution
rea
c
he
d 77
8
32 km
2
. Lo
cat
i
ons
whi
c
h h
a
v
e the most d
i
stribut
io
n of
hotsp
ot duri
n
g the year
20
14
are Ri
au province (Ro
k
a
n
Hilir: 5 793 h
o
tspot
s,
Beng
kali
s: 5 343 h
o
tspot
s, Siak: 3 010 hotsp
o
t
s,
Indragi
ri Hili
r:
1 910 h
o
tsp
o
ts, Dum
a
i: 1 295
h
o
tsp
o
ts an
d Pela
lawa
n: 671 h
o
tspot
s), Sou
t
h
Sumatra province
(Ogan
Komeri
ng Ilir:
4 463
hotspots). Cluste
ring hotspots with
KSS method in
peatlan
d
are
a
in Sumatera can dete
c
t whe
r
e the
o
c
curre
n
ce of cluster h
o
tsp
o
ts, whe
n
clu
s
te
r
hotsp
ot hel
d, and
the
geo
grap
hical
size of
clu
s
ter
hotsp
ots. KS
S clu
s
ters
an
d mea
s
u
r
e
their
signifi
can
c
e v
i
a Monte Carl
o repli
c
ation.
Re
sults
sho
w
ed that the method is
relia
ble to detect t
he
clu
s
ters of ho
tspots
whi
c
h
have the accura
cy of 95%.
Table 1
sh
ows the
clu
s
ter
distrib
u
tion
s i
n
the
pe
atlan
d
are
a
in Su
matera th
at h
e
ld inthe
year 2
014.In
the year 20
1
4
con
s
ist
s
of
4 cl
us
te
rs. A
s
sh
ows in
Ta
ble 1
mo
st likely clu
s
ter (
P
) in
Dumai,
B
eng
kali
s,
an
d
Ro
kan
Hili
r di
st
r
i
ct
in
Ria
u
p
r
ovince.
The
radiu
s
of
mo
st likely
clu
s
te
r is
defined
94.1
3
km. Th
e cl
uster is
re
sul
t
ed from
sc
a
nning
a wi
nd
ow that it is
centred o
n
t
he
latitude a
nd l
ongitud
e
coo
r
dinate
s
(1
01
.58391
1, 1.5
7505
1). T
he
se
con
dary
cl
uster I (S
1) i
s
i
n
Ogan Kom
e
ri
ng Ilir in Sou
t
h Sumatera
provin
ce. Cl
u
s
ter
cent
re of
se
cond
ary cl
uster II (S2
)
i
s
Padang di
stri
ct in We
st Sumatera p
r
ovin
ce. A loca
tio
n
which ha
s the most
di
strib
u
tion of hotsp
ot
is Siak di
stri
ct in Riau p
r
ov
ince. As
sho
w
s in ta
ble 1
the radi
us of
se
con
dary
cl
uster II is d
e
fined
432.16
km. T
he secondary
cluster III
(S3) is i
n
Siak,
Bengkalis,
I
ndragiri Hilir, Pelalawan
and
Karimun
in
Riau p
r
ovin
ce.
The
radiu
s
of mo
st likely clu
s
te
r i
s
d
e
f
ined 9
4
.13
km. The
cl
ust
e
r
409
7222
9066
228
247
1192
1464
406
2353
2541
1036
29
0
2000
4000
6000
8000
10000
Hotspot
Month
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Poisson Cl
ust
e
ring Process on Hots
pot in Peatland Area in Sum
a
tera
(Anni
sa Pu
spa Kirana
)
1380
resulting fro
m
scanni
ng
a win
dow centr
ed o
n
the latitud
e
and l
ongi
tude coordin
a
tes
(103.4
717
61,
0.7250
88). T
he p
-
value fo
r this
clu
s
ter is 0.00
000
001
that wa
s obt
ained
usi
ng t
h
e
999 Monte
Carlo si
mulatio
n
s.
Table 1. The
clu
s
ter di
strib
u
tions in p
eat
land area in
Sumatera in t
he years 201
4
Cluster District
Province
Longitude
Latitude
Radius (km)
Likelihood Ratio
P
Dumai Riau
101.583911
1.575051
94.13
5
139.63
Bengkalis Riau
102.295207
1.178453
Rokan Hilir
Riau
100.736996
1.54702
S1
Ogan Kome
ring I
lir
South Sumater
a
105.591389
-3.115678
0
730.02
S2
Padang
West
Sumatera
100.383694
-0.940575
279.21
432.16
Padang Pariama
n
West Sumatera
100.297522
-0.739932
Pariaman
West
Sumatera
100.135707
-0.625686
Agam West
Sumatera
99.921643
-0.268635
Pasaman West
Sumatera
99.723302
0.160544
Kerinci Jambi
101.349276
-1.793403
Pesisir Selatan
West Sumatera
100.998983
-2.373818
Kampar
Riau
101.219395
0.48254
North Bengkulu
Bengkulu
101.128132
-2.527019
Pekan Baru
Riau
101.396441
0.532964
Indragiri Hulu
Riau
102.338812
-0.597247
Rokan Hulu
Riau
100.789814
1.021444
Mandailing Natal
North Sumat
e
ra
99.067844
0.904823
Merangin
Jambi
102.473472
-1.930053
Pelalaw
a
n
Riau
102.469822
0.096434
Siak Riau
101.971053
0.766098
S3
Karimun Riau
103.471761
0.725088
166.81
129.08
Indragiri Hilir
Riau
103.0559
-0.369494
Pelalaw
a
n
Riau
102.469822
0.096434
Bengkalis Riau
102.295207
1.178453
Siak Riau
101.971053
0.766098
*Note: P = Most likely
clusters (pri
mari
l
y
cluster) an
d S = Secondar
y clusters
Figure 3
sho
w
s cl
uste
r
di
stributio
ns of
hots
pots in
peatlan
d
a
r
e
a
s i
n
Sum
a
te
ra i
n
the
year
2
014 which are con
s
ist of
4 clu
s
ters
d
e
tecte
d
(1
cl
uste
r as
m
o
st
li
kel
y
clu
s
ter and
3
clu
s
ters a
s
seco
nda
ry clu
s
ter). The lo
cation th
at
has mo
st likel
y cluste
r is i
n
Riau
province
,
esp
e
ci
ally
in Ro
kan
Hili
r, Bengkalis, an
d
Dumai.
Og
an Kom
e
rin
g
Ilir di
stri
ct in
South Su
ma
tra
provin
ce is th
e location tha
t
has the hig
hest den
si
ty of seconda
ry clu
s
ter. Figu
re 4 sho
w
s m
o
st
likely clu
s
ter
and second
ary cluste
r distri
bution
s
of hotspot in the pe
atland area in
Sumatera.
Figure 5 sh
o
w
s
clu
s
ter di
stribution
s
of hots
pot
s ba
se
d on the mat
u
rity level of peatlan
d
in Sumatera i
n
the year 20
14. The dist
ri
bution cl
uste
rs of hotsp
ot base
d
on the
maturity level of
peatwere d
o
m
inated on th
e maturity level of He
mi
c/Sapri
c
(6
0/40)
and very dee
p peatlan
d
. The
locatio
n
that has the hig
h
e
st den
sity of cluste
r h
o
tspot on the maturity level of Hemic/Sa
pric
(60/40) and
very deep
is
in Riau provi
n
ce, especi
al
ly in t
he di
strict of Rokan
Hilir, Bengkal
i
s,
Dumai a
nd S
i
ak. The lo
ca
tion that has the highe
st
den
sity of cluster h
o
tsp
o
t on the maturi
ty
level of He
mi
c/Sapri
c
(60/
40)
and m
o
d
e
rate
peatlan
d
is
in
Ria
u
p
r
ovince e
s
pe
cially in the
di
stri
ct
of Bengkali
s
, Siak and
Du
mai.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 137
6 – 1383
1381
Figure 3. Clu
s
ter di
strib
u
tions of hot
spo
t
s
in peatlan
d
area in Sum
a
tra in the year 201
4
Figure 4. Clu
s
ter di
strib
u
tions of hot
spo
t
s
in peatlan
d
area in Sum
a
tra in the year 201
4
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Poisson Cl
ust
e
ring Process on Hots
pot in Peatland Area in Sum
a
tera
(Anni
sa Pu
spa Kirana
)
1382
Figure 5. Clu
s
ter di
strib
u
tions of hot
spo
t
s based
on t
he maturity le
vel of peat in the year 201
4
Based o
n
Fig
u
re 5, many hotsp
ots foun
d in peat
land
with the maturity level of ’hemic’
and the
thickness level
of
’mode
rate’ th
at gene
rally
l
o
cate
d in th
e
dome
of pe
a
t
and
clo
s
e from
resi
dential.
Whe
n
‘hemi
c
’ maturity levels of
peat d
o
minated o
n
the outskirts of peat do
mes
indicated the
damag
e of the pe
at dom
e very hi
gh.
Combi
ned
wi
th the drai
na
ge to overco
me
dryne
ss of pe
at it makes e
x
cretion of water that
is in
the dome pe
a. If
the wate
r at the dome
of
peat missing,
hence peat e
x
perien
c
e
d
le
aka
ge so fun
c
tion hydrolo
g
y peat being
broken [2].
Figure 6 sh
o
w
s
clu
s
ter di
stributio
ns of
hots
pot
s ba
sed o
n
the thickne
ss of P
o
rtland in
Sumatera
in
the yea
r
2
014.
Cluste
r distri
butio
n
s
of hot
spot
are
domin
ate
d
by ’very d
eep
thickne
ss (D4
)
(40
0
-8
00 cm)’ and ’mo
d
e
rate thick
n
e
ss (D2
)
(1
00
-200 cm
)’. The
location that
has
the hig
h
e
s
t d
ensity of
clu
s
ter h
o
tspot
with ’very
de
e
p
thickn
ess
(D4) (400
-80
0
cm
)’ b
e
in
Riau
provin
ce, e
s
p
e
cially in
the
distri
ct of
Ro
kan
H
ili
r, Sia
k
, Indragiri
HIlir and B
eng
kali
s. Whil
e t
he
locatio
n
s that
have the hig
hest de
nsity of cluste
r hot
spot with ’m
o
derate thi
c
kn
ess (D2)
(10
0
-
200 cm)’ be i
n
South Sum
a
tera p
r
ovin
ce, espe
cially
i
n
the distri
ct
of Ogan Kom
e
ring Ilir; in
Riau
provin
ce
esp
e
cially i
n
the
dist
rict
of Beng
kali
s, Sia
k
, an
d Ind
r
a
g
iri
HIlir. A
s
deman
d by
an
Indone
sia
n
g
o
vernm
ent re
gulation
num
ber
32 ye
ars 1990, peatla
nd
a
r
ea
that have
’de
ep’ and
’very deep’ t
h
ickne
s
s (>3
00 cm
)
sho
u
l
d
not be
ope
ned for
ag
riculture d
e
velo
pment. Base
d on
RTRWN that
in line with con
s
titutio
n
numbe
r 2
1
years 1
9
9
2
rega
rdi
n
g
area
s plan
ning
(UUT
R),the p
r
otectio
n
agai
nst the peat sho
u
ld be do
ne to control hydrolo
g
ical area
s, se
rve
s
as
a fastenin
g
water, flood prevention, and
protectin
g
the eco
s
ystem
that is typical of the area [1
6].
Figure 6. Clu
s
ter di
strib
u
tions of hot
spo
t
s bas
ed on t
he thickne
s
s of peat in the year 201
4
Figure 7 sh
o
w
s
clu
s
ter di
stributio
ns
of hots
pot
s ba
sed o
n
the la
nd use of pe
atland in
Sumatera
in
the year
20
14. The
dist
ribution
cl
u
s
te
rs
of hotspot
based o
n
t
he lan
d
u
s
e
o
f
peatlan
d
are
dominate
d
by swam
p fore
st. The loca
ti
on that has the highe
st d
ensity of cluster
0
1000
2000
3000
4000
5000
6000
7000
Fibrists/Saprists (60/40),moderate
Hemists (100),moderate
Hemists/min (30/70),shallow
Hemists/min (30/70),moderate
Hemists/min (70/30),moderate
Hemists/min (90/10),moderate
Hemists/Saprists (60/40),deep
Hemists/Saprists (60/40),very
deep
Hemists/Saprists (60/40),moderate
Saprists (100),moderate
Saprists/Hemists (60/40),deep
Saprists/Hemists (60/40),very
deep
Saprists/Hemists (60/40),moderate
Saprists/min (50/50),shallow
Saprists/min (50/50),moderate
Saprists/min (90/10),moderate
Hotspot
Maturity
level of Peat
Most Cluster
Secondary
Cluster
I
Secondary
Cluster
II
Secondary
Cluster
III
0
2000
4000
6000
Peaty
soil
(peat
thickness <50
cm)
(D0)
Shallow
peat
(peat
thickness 50–100
c
m
) (D1)
Moderate deep
peat
(peat thickness
101–
200 cm) (D2)
Deep peat
(peat
thickness 201–400
c
m
) (D3)
Very
deep peat (peat
thickness >400
cm)
(D4)
Thickness of
peat
Hotspot
Most Cluster
Secondary
Cluster
I
Secondary
Cluster
II
Secondary
Cluster
III
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 137
6 – 1383
1383
hotsp
ot is in
Riau
p
r
ovince, espe
cially
in the
di
st
rict of
Ro
ka
n Hili
r,
Beng
kali
s, and
Si
ak;
S
o
uth
Sumatera p
r
o
v
ince e
s
pe
cia
lly in the district of
Ogan Komeri
ng Ilir. The swam
p fore
sts a
r
e to be
an are
a
of forest fires on th
e drou
ght an
d in the
dry season. The d
a
mage
cau
s
e
d
by forest fire in
the swamp fo
rest aim
s
seri
ous n
egative
impact.
Figure 7. Clu
s
ter di
strib
u
tions of hot
spo
t
s bas
ed on t
he land
cover of peat in the year 2014
4. Conclusio
n
This
work
applied Kulldorff’s
Sc
an Statis
tic
(
KSS) on the s
patial
temporal fores
t
fires
dataset. The
dataset con
s
i
s
ts of the
spa
t
ial aspe
ct
(l
o
ngitude, l
a
titude) of the
hot
spot, tem
poral
asp
e
ct (date
of occurre
n
ce
) of
hotspot, and p
eatland
physi
cal
cha
r
acteri
stics th
at may influe
nce
the distrib
u
tio
n
pattern of h
o
tspot cl
uste
rs in
the study
area Sum
a
te
ra province, Indon
esi
a
. KSS
method
uses circul
ars scannin
g
wi
nd
ows to dete
c
t the clu
s
ters of h
o
tsp
o
ts. Every ci
rcula
r
scanni
ng win
dow
will find the ratio of the po
ssi
bilit
y using Poi
s
son model. Th
e test statisti
c is
determi
ned a
s
the maximu
m likeliho
od ratio over a
ll circul
ar
wind
o
w
s. Evaluatin
g the statistical
signifi
can
c
e i
s
calculated
by gene
ratin
g
a
large n
u
m
ber
of ran
d
o
m data
s
ets unde
r the n
u
ll
hypothe
sis of
no clu
s
terin
g
. Next step is cal
c
ul
atin
g
the value of the test statistic for e
a
ch of
those
data
s
et
s. The
regi
on
with the hi
g
hest valu
e of
likelih
ood
ra
tios is th
e a
r
ea that ha
s t
h
e
most potential clusters.
Cluste
ring
hotspots with
KSS method ca
n detect where the
occurrence
of cluste
r hot
spot
s, whe
n
cluster h
o
tsp
o
t hel
d, and the
geographi
cal
size of cl
uste
r hotsp
ots.
Clustering hotspot in the
peatland areas in
Sumat
e
ra i
n
the year 2014
with the KSS
method
disco
v
er patterns
of hotsp
ot di
stribution
wh
i
c
h co
nsi
s
t of 4
clu
s
ters. Th
e
radi
us
of mo
st
likely clu
s
ter
is defined 9
4
.
13 km. The provin
ce
s
wit
h
the occurre
n
ce
s of the highe
st hotsp
ot
clu
s
ter a
r
e locate
d in Ri
au provin
ce
and South
Sumatera p
r
ovince, espe
cially in Du
mai,
Bengkalis,
Roka
n HIlir
an
d Oga
n
Kom
e
ring Ili
r
di
st
rict. Based
o
n
the matu
rity level of peat,
clu
s
ter di
strib
u
tions of hot
spot were m
o
stly f
ound in
‘hemic’ matu
rity level. Ba
sed on p
eatl
and
thickne
ss,
clu
s
ter
distri
buti
on of hot
spot
wa
s
mo
stly found in ‘ve
r
y deep’ thi
c
kness. Based
on
land u
s
e of p
eatland, cl
ust
e
r dist
ributio
n
s
of
hotsp
ot were mo
stly found in ’
s
wa
mp fore
sts’.
Referen
ces
[1]
Glover D, Jessup T
.
Mahalny
a Harg
a Seb
u
a
h
Benca
n
a
. Ba
ndu
ng (ID): IT
B. 2002.
[2]
Sy
aufina
L.
Ke
bakar
an Hut
a
n
dan La
ha
n di Indo
nesi
a
. Mala
ng: Ba
yume
di
a
Publis
hin
g
. 20
08.
[3]
Sitang
ga
ng IS
, Yaako
b
R,
Mustaph
a N
,
Ainud
di
n A
N
. A Decis
i
o
n
T
r
ee Base
d on
Spati
a
l
Relati
ons
hips
f
o
r Pre
d
ictin
g
Hotspots
in
Pe
atlan
d
s.
T
E
LK
OMNIKA T
e
lec
o
mmunic
a
tio
n
, Co
mputin
g,
Electron
ics an
d Contro
l
. 201
4; 12 (2): 511-
518.
[4]
Han J, Kam
ber
M, Pei J. Data
Minin
g
C
once
p
t
and T
e
chni
q
ues T
h
ird Editi
on.
Un
ited of S
t
ates (US):
Elsevi
er Inc. 2012.
[5]
Madh
ulata
T
.
An Overvi
e
w
o
n
Cl
usteri
ng M
e
thods.
IOSR Journ
a
l of
En
g
i
ne
erin
g
.
Apr
. 201
2;
2(4)
:
719-
725.
[6]
Sitang
ga
ng IS, F
uad T
,
Annisa. K-Means
Cluster
i
ng Vis
ualiz
atio
n of W
eb-Bas
ed OLA
P
Operatio
ns
for Hotspot Dat
a
.
IEEE
. 2010; 978-
1-42
44-
67
16-7/1
0
/$2
6
.00
.
0
2000
4000
6000
8000
10000
Bog shrub
sw
amp forest
Rubbe
r
Coconut on a former
sw
amp
forests
> 5 y
e
a
r
Palm
oil
on
a former
sw
amp
forests > 5 y
e
a
r
Logging concessions
land
Planting land
an
industrial plant
Open land
for farm
Paddy
fields and coconut
Intensive rice
fields
(rice
crops, fallow
)
, orange
Rainfed rice (rice crops, fallow
)
Shrubs, marsh grasses and
former
fire
Bush,
Law
n on
former rice
fields
Hotspot
Land cover of peat
Most Cluster
Secondary
Cluster
I
Secondary
Cluster
II
Secondary
Cluster
III
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Poisson Cl
ust
e
ring Process on Hots
pot in Peatland Area in Sum
a
tera
(Anni
sa Pu
spa Kirana
)
1384
[7]
Usman M, Sitangg
an
g IS, Syaufin
a L.
Hotspot Distrib
ution
Analyses b
a
s
ed on Pe
at Ch
aracteristic
s
usin
g D
ensity-
Based
Sp
atial
Clusteri
n
g
. T
h
e 1st International S
y
m
pos
ium on LAPAN-IPB
Satellite
for
F
ood Sec
u
rit
y
and
Envir
o
n
m
ental M
onit
o
ring. Proc
ed
ia
Enviro
nmenta
l
Scienc
es. 20
15; 24:
13
2
–
140.
[8]
Sitang
ga
ng IS,
Yaak
ob
R, Mu
stapha
N, Ai
nu
ddi
n
AN. B
u
rn
Area Pr
ocessi
n
g
to Ge
ner
ate
F
a
lse A
l
arm
Data for Hotspot Prediction Models.
T
E
LKOMNIKA T
e
leco
mmu
n
icati
on,
Co
mp
uting, El
ectronics a
n
d
Contro
l
. 201
5; 13(3): 10
37-
10
46.
[9]
Kull
dorff M. A
Spatia
l Sca
n
S
t
atistic. Comm
uni
c
a
tions
in
S
t
atistics: T
heory
an
d Meth
ods
. 199
7; 26:
148
1-14
96.
[10]
Devis T
,
F
r
éderic R, R
o
sa
L, Luci
a
n
o
T
,
Mikhai
l K. Sca
n
statistics a
n
a
l
y
sis
of forest
fire clust
e
rs.
ELSEVIER Comm
unications
in Nonlinea
r Scienc
e and Num
e
r
i
cal Simulation
. 20
08; 13(
80): 168
9–
1
6
94
[11]
W
a
h
y
unto, Rit
ung S, Sup
a
rto
,
Subag
jo H. S
ebar
an Gamb
u
t
dan Kan
d
u
n
g
an Karb
on
di S
u
matera d
a
n
Kalima
n
tan. Bo
gor (ID): W
e
tlands Internati
o
n
a
l IP. 2005.
[12]
Cao C
X
,
Xu M
,
Chen JQ, W
u
YS, Xie
X. Sp
ace-
T
i
me Scan
Statistic Base
d Earl
y W
a
rn
in
g of H1N
1
Influenz
a A
in
Shenz
he
n, Ch
i
na.
Intern
atio
n
a
l Arc
h
ives
of t
he P
hoto
g
ra
mmetry, R
e
mote
Sens
ing
a
n
d
Spatia
l Informa
tion Scie
nce
. K
y
ot
o: Japa
n. 2010; 28(
8).
[13]
Wen S, Kedem
B. A semipa
ra
metric Cl
uster
Detectio
n Meth
od a
Com
p
re
h
ensiv
e Po
w
e
r
Comp
ariso
n
w
i
t
h
Kul
l
dorffs Method.
Intern
ation
a
l Jo
urna
l of Health Ge
og
raph
ics. BioMe
d
Centra
l
. USA. 2009
[14]
Kim AY, W
a
kef
i
eld
J. R D
a
ta
and M
e
tho
d
for
S
pati
a
l E
p
id
e
m
iolo
g
y
: T
he Spatia
l Ep
i Pack
age. D
ept
of
Statistics, Univ
ersit
y
of W
a
sh
i
ngton. 2
010.
[15]
Kull
dorff M.
201
4. SaT
S
can User Gui
de
[do
w
n
l
oad
ed at 20
14
Jul
y
2]. Availa
bl
e a
t
http://
w
w
w
.
satscan.org/.
[16]
Agus F
,
Hairia
h K, Mul
y
an
i
A. Measurin
g C
a
rbo
n
Stock in
Peat
Soil. World Agrofor
e
stry
Centre and
Indon
esi
an
Ce
ntre for A
g
ricu
l
t
ural
Lan
d R
e
s
ources
Res
ear
ch a
nd
Deve
lo
pment. Bo
gor
(ID): W
o
rld
Agroforestr
y
Centre (ICRAF). 2011.
Evaluation Warning : The document was created with Spire.PDF for Python.