TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5613 ~ 56
2
0
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.573
9
5613
Re
cei
v
ed
Jan
uary 4, 2014;
Re
vised Ma
rch 16, 2014; A
c
cepted Ap
ril 6, 2014
Detecting Rice Growth Using ALOS Multispectral and
Synthetic Aperture Radar
Bamban
g H. Trisason
g
ko
*, D
y
ah R. Panuju, La Od
e S. Iman
Dep
a
rtment of Soil Sci
ence a
nd La
nd R
e
so
u
r
ce, Bogor Agri
cultura
l
Univ
er
sit
y
,
Jala
n Meranti,
Dramag
a,
Bog
o
r 166
80, Indo
nesi
a
,
Phon
e/F
a
x: +
6
2 251
84
223
25
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: trisasongk
o
@
live.it
A
b
st
r
a
ct
Rice
mo
nitori
n
g
is a su
bstan
t
ial to Asia
n c
ount
ri
es, incl
u
d
in
g Indo
nes
ia
, since
most in
hab
itants
consu
m
e rice
o
n
a
dai
ly b
a
sis.
As fiel
d surve
y
requ
ires s
ubs
tantial ti
me a
n
d
bud
get, o
ne r
e
lies
on r
e
motel
y
sensed data,
espec
ially taken through spacebor
ne
pla
tform
. This r
e
search di
sc
usses m
u
ltispectral and
line
a
rly p
o
l
a
ri
zed Synth
e
tic
Aperture R
a
d
a
r
(SAR)
from
Advanc
ed L
a
n
d
Observin
g S
a
tellit
e an
d th
eir
app
licati
ons to
observe v
a
rio
u
s rice gr
ow
th infor
m
ati
on. It app
ears that b
o
th sens
ors pr
ovid
ed us
eful r
i
ce
grow
th data l
e
adi
ng to the
p
o
ssibi
lity o
n
i
m
provi
ng ric
e
fie
l
d infor
m
atio
n
extraction. C
l
a
ssificatio
n
sch
eme
by me
ans of R
and
o
m
F
o
rest sugg
es
ted that
both data w
e
r
e
fairly acce
pta
b
le for timely
monitor
i
ng.
Ke
y
w
ords
: de
cision tre
e
, mu
l
t
ispectral, rad
a
r
, rando
m fores
t, rice
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Rice i
s
a
sta
p
le foo
d
for Indon
esi
ans.
Although th
e
deman
d ten
d
s
to i
n
crea
se,
rice h
a
s
been pl
anted
in very few locatio
n
s in In
done
sia. Java has
been
serving a
s
p
r
imary produ
ct
ion
cente
r
for th
e co
untry
si
nce
ea
rly 19
00s [1]. Thi
s
has bee
n
sup
porte
d by
suitabl
e lan
d
resou
r
ces, water availabili
ty
and thoro
ugh irrigation
networks [2].
Monitori
ng
of rice can
be
establi
s
h
ed
by field su
rv
eys o
r
thro
u
gh remote
sensi
ng. In m
any ca
se
s,
combi
n
ing
b
o
th
approa
che
s
has b
een the
most relia
bl
e. Opti
cal re
motely-sen
se
d data have
been
servin
g
as
prima
r
y data
s
ets. Non
e
thel
ess, cl
ou
ds a
r
e pe
rsi
s
tent i
n
many
are
a
s, where com
p
lementa
r
y data
are then
req
u
i
red. An alternative to this sit
uation i
s
Synthetic Apert
u
re Rada
r (S
AR) data.
SAR data h
a
v
e been
studi
ed for
rice m
onitorin
g
con
s
ide
r
ably. Sci
entific literatu
r
e tend
s
to exploit m
o
re on
C-b
and
SARs, p
r
o
b
a
b
ly due
to
th
eir
co
ntinuou
s d
a
ta
sup
p
ly sin
c
e
the
era of
ERS-1/SAR
Advance
d
Mi
cro
w
ave Im
a
ger in
198
0s
-1990
s. Halda
r
an
d Patnai
k [3], for insta
n
ce,
studie
d
C-ba
nd Ra
darsat
-
1 cou
p
led
with AWiFS dat
a to monitor Indian
rabi
ri
ce. Another st
udy
in China [4]
reported that
C-
band Envis
a
t ASAR
data s
u
cc
es
s
f
ul
ly es
timated
leaf area index
(LAI).
L-ba
nd SA
R
wa
s expl
oite
d a
s
well, alt
houg
h in
a le
sser
extent. Japan
was the
only L
-
band SAR p
r
ovider op
erat
ing JERS
-1/S
AR at fi
rst an
d later ALOS
Phased
-Arra
y
L-Band SAR
(PALSAR).
I
s
hitsu
k
a
[5] prese
n
ted a rice
field
ob
se
rvation in
Japa
n u
s
ing
PAL
SAR si
ngle,
dual
and fully
pola
r
imetry. Th
e
rese
arch
indi
cated that
ob
servation
of
Ja
pane
se
small
-
scale
ri
ce
fie
l
d
wa
s quite u
n
su
cce
ssful,
althoug
h the
use of fu
lly polarim
etry
was
pro
m
ising. Usi
ng d
ual
polari
z
e
d
PALSAR data, Ling
et al
. [6] found that
mappin
g
rice
fields in Chi
na wa
s po
ssi
ble.
Those resea
r
ch
su
gge
st
that fully polarimetri
c SA
R may overcome limitatio
n in ri
ce pa
ddy
monitori
ng, d
e
spite
very li
mited sce
ne
coverage.
In
addition,
m
u
ltitemporal
an
a
l
ysis usi
ng si
ngle
or dual lin
ea
rly polari
z
e
d
data has b
een favora
bl
e. Wang
et al
. [7] for example, reli
ed
on
PALSAR mul
t
itemporal
ob
servatio
n to study cha
r
a
c
te
ristics of So
u
t
heast
Chin
a
rice fiel
ds. T
h
e
repo
rt of Ishitsuka [5] pionee
red PA
LSAR fully
polari
m
etri
c impleme
n
tatio
n
in rice fiel
ds.
Non
e
thele
s
s, literature
re
views
sug
g
e
s
ted that the
r
e have
bee
n a
few, if a
n
y, repo
rts o
n
the
comp
ari
s
o
n
o
f
multispe
ctra
l and fully po
larimetri
c
SA
R data for
rice field monit
o
ring in tropi
cal
regio
n
s.
Anticipating
n
ear future
lau
n
ch
of PALS
A
R-2,
thi
s
pa
per presents
a stu
d
y of S
A
R d
a
ta
analysi
s
to o
b
tain them
atic ma
p of ri
ce growth, in
comp
ari
s
o
n
with multi
s
pe
ctral
data
s
et. To
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5613 – 56
20
5614
extend Ishitsuka’
s work [5], we empl
oyed PALSAR
fully polari
m
etric d
a
ta to observe th
ei
r
perfo
rman
ce i
n
an Indon
esi
an test site.
2. Rese
arch
Metho
d
The
re
sea
r
ch
wa
s l
o
cated
on
rice field
s
m
anag
ed
b
y
PT Sang
Hyang
Seri
(SHS), a
state ente
r
pri
s
e re
sp
on
sibl
e sup
p
lying rice s
eedli
ng.
Planted rice
varieties in
clude Sintanu
r,
Benga
wan a
nd Cihe
ran
g
.
The latter has bee
n pop
ular
in Indon
esia an
d therefore sele
cte
d
as
the only vari
ety in our a
nalysi
s
. The
area
cove
rin
g
about 3,1
0
0 ha was l
o
cated in S
u
b
ang
R
e
ge
nc
y, W
e
s
t
J
a
va
(
F
ig
ure
1
)
.
Figure 1. Re
search Lo
catio
n
Rice is pla
n
te
d in blocks. A block occu
pi
es
ab
out 5-1
8
Ha with
sim
ilar width of 2
00m. A
grou
p of blo
c
ks i
s
m
anag
e
d
by a secto
r
manag
er
. T
r
a
n
spl
anting
by mean
s of
co
nventional
(n
on-
mechani
cal
)
way is usuall
y
accom
p
lished within 4-
7
days for each block. Th
is l
eads to variability
of ri
ce
gro
w
i
ng p
h
a
s
e
wit
h
in
site lo
cati
on, an
d all
o
ws continu
a
tion of
ri
ce
su
pply to the
seed
plant. A typical distan
ce b
e
twee
n plant
s is 20
cm.
Duri
ng the
rese
arch, several d
a
tasets were
colle
cted. ALOS AVNIR-2
dat
a we
re
obtained from KKP3T projec
t, ac
quired in
2008
and 2009. ALOS PALSAR
fully polarimetric
(PLR) data
s
e
t
s con
s
i
s
t of
2007, 200
9 a
nd 2010 d
a
ta
. The 2010 PLR data are a
subje
c
t of future
investigatio
n and are not b
e
ing re
po
rted
in this pape
r.
In orde
r to e
x
tend the an
alysis, a
dditi
onal d
a
tasets were a
c
qui
red. A scene
of ALOS
PRISM wa
s
obtaine
d to construct
baseline them
atic data
s
et in
GIS softwa
r
e
based o
n
d
a
ta
fusion
with A
V
NIR-2. Fou
r
algo
rithms were
ev
aluate
d
,
howeve
r
, In
tensity-Hue
-
Saturation
(IHS)
[8] prod
uct
was
sel
e
cte
d
(Figure 2
)
to
map
rice fiel
ds. Additio
n
a
l
data
s
ets in
cludi
ng b
a
sel
i
ne
maps
were obtained from BAKOSURT
ANAL.
Most of d
a
ta
sets to
a
ssi
st the analy
s
i
s
were colle
cted
in situ
.
Field
su
rve
y
s we
re
con
d
u
c
ted du
ring
the 200
9 and 201
0
d
r
y sea
s
o
n
s,
and
additio
nally i
n
20
11
(wet
season).
Base
d
on the preli
m
inary information provi
ded by the
company, sa
mple blo
c
ks
were sel
e
cte
d
. On
those bl
ocks, several m
easurement
s were ta
ken, including averaged hei
ght and ground
moistu
re con
d
ition (d
ry, wet or flooded
).
All ALOS im
ages were provided by Japan
A
e
rospace Expl
orati
on Agency
(JAXA) in
stand
ard
CE
OS format. T
hese data
were th
en p
r
e
-
pro
c
e
s
sed
using ASF Ma
p
R
ea
dy 2.3.17
to
retrieve the
digital numb
e
r (DN). Thi
s
format i
s
n
o
t recomme
n
ded for bi
op
hysical analy
s
is,
therefo
r
e co
mputation of radian
ce
wa
s
then pe
rform
e
d usin
g following equ
ation:
)
(
)
(
1
'
'
C
B
B
C
B
B
O
R
L
a
NL
a
(1)
W
h
er
e
L
: radian
ce (W/m2.sr.
μ
m)
R :
sensitivity
O
: signal outp
u
t
digital value
C
: dummy pixel output mean
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
cting Ri
ce Gro
w
th Usi
ng ALOS Mul
t
ispe
ctral an
d
Synthetic… (Bam
bang H. Trisasong
ko
)
5615
B
a
: offs
et of exp
o
s
u
re
B
NL
: non-line
a
r of
fset
B
: output digital value – dark
C’
: dummy pixel output – dark
B
a
’
: offset of exp
o
su
re – da
rk
Figure 2. A Subset of HSV Data to Con
s
truct Ri
ce Fiel
d Map, AVNIR data © JAX
A
Based
on
the
equ
ation, ra
dian
ce d
a
ta
were o
b
taine
d
an
d sub
s
e
quently fed i
n
to NDVI
analysi
s
. Pre
c
isely 75 pix
e
ls
were sam
p
led for ea
ch
date to ob
se
rve varia
b
ility of NDVI a
c
ross
r
e
sear
ch location.
For
co
mpa
r
ison, two
PAL
SAR PL
R
scene
s
we
re
st
udied
in thi
s
re
se
arch. T
he PL
R
mode preserves origi
nal radar
me
asurement and repre
s
e
n
ted i
n
compl
e
x numbe
r (kno
wn as
Single Loo
k
Compl
e
x, SLC). The follo
wing eq
uatio
n wa
s empl
oyed to deri
v
e backscatt
er
c
oeffic
i
ent of respec
tive polarization [9]
A
CF
Q
I
2
2
10
0
log
10
(2)
Whe
r
e I a
nd
Q re
spe
c
tivel
y
represents
real an
d im
agi
nary pa
rt of S
L
C d
a
ta. Co
n
v
ersio
n
facto
r
A
wa
s de
rived
usin
g a mo
de
l [9] which i
s
32.0. Mean
while, calib
ratio
n
factor
CF v
a
lue was
-83
dB,
as
s
u
gges
ted by previous c
a
libration
study
c
o
nduc
ted by J
AXA
[9]. PALSAR datasets
were
conve
r
ted
by
the same
soft
ware. It provi
des a
c
onve
n
i
ent way to
ex
tract b
a
cksca
tter co
efficien
ts
(s
igma nought, in dec
i
bel) from J
AXA CEOS format
. Similar
to t
hose of NDV
I, 75 pixels
were
sele
cted in fu
rther a
s
se
ssment and mo
del buildin
g.
Table 1. Ra
n
dom Fo
rest P
a
ram
e
ters
Parameters
AVNIR-2
PALSAR
Random F
o
rest
Options
Number of P
r
edi
ctors
4
3
Number
of
Tre
e
s
100
100
Random T
e
st Da
ta Proportion
0.3
0.3
Subsample
Prop
ortion
0.5
0.5
Seed for Ra
ndo
m Number
Gen
e
r
ator
1
1
Stopping Param
e
ters
Minimum Numbe
r
of Cases
75
75
Minimum Numbe
r
of Levels
10
10
Minimum Numbe
r
in Child Nodes
5
5
Maximum Numb
er of Nod
e
s
100
100
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5613 – 56
20
5616
The resea
r
ch employe
d
sup
e
rvi
s
ed
cla
ssifi
catio
n
app
r
oa
ch
to obtain
rice
map.
Sampling
strategy wa
s gu
ided by pla
n
n
i
ng data at
bl
ock level. Thi
s
was
also
e
n
han
ce
d by two
field su
rveys in July-Aug
u
s
t 20
09 a
nd
July 20
10. A
nother data
colle
ction
wa
s ta
ken i
n
Ju
ne
2011 to imp
r
ove field dataset
s. Based
on the coll
ection, ad
ditional sample
sets were
also
establi
s
h
ed. These sampl
e
s were
em
p
l
oyed
for
te
st
ing cl
as
sif
i
c
a
t
i
on rul
e
s.
De
cisi
on t
r
e
e
w
a
s
cho
s
e
n
as
cl
assificatio
n
a
l
gorithm, mai
n
ly due to
its ea
sy-to
-
co
mpre
hen
d
, flexible rul
e
[10].
Ran
dom
Fo
re
sts [11]
wa
s
particula
rly selecte
d
to
aid
the
co
nstruct
i
on of
cl
assifi
cation
rule
s.
The
algorith
m
ha
s bee
n teste
d
for variou
s application
s
with sup
e
ri
o
r
re
sults [12,
13]. Followi
ng
Ran
dom Fo
re
st spe
c
ificatio
n wa
s tested.
3. Results a
nd Analy
s
is
3.1. Fluctua
t
ion of NDVI Values
NDVI ha
s b
e
en exploited
on many bi
op
hysical
ob
se
rvations fro
m
remotely se
nsed data.
Usi
ng
radi
an
ce
data, NDV
I
values
we
re
retri
e
ved a
n
d
sub
s
equ
ent
ly analyze
d
a
c
cordi
ng to
the
field data (Fig
ure 3
)
.
On early ages
(early tilleri
ng)
,
observed NDVI tends to be ar
ound 0. Thi
s
i
s
due to the
fact that field
s
a
r
e
floode
d
by wate
r
and
the ri
ce
is very small,
whi
c
h lea
d
s to
strong
co
ntributi
on
of non-ve
get
ative comp
o
nents
(wat
er and
soil)
o
n
the re
sp
ective
pixel. When ri
ce
gro
w
s,
c
h
lor
o
ph
yll in
c
r
ea
se
s
,
w
h
ic
h
is
th
er
e
f
or
e
r
e
flec
ted by
rising NDVI
value.
Howev
e
r, in
cer
t
ain
perio
ds, NDVI
values are
d
r
opp
ed sig
n
ificantly.
From
f
i
eld ob
se
rvati
ons, it
wa
s fo
und that
man
y
blocks we
re
severely infe
sted. Two m
a
jor infe
stat
io
ns in resea
r
ch are
a
we
re
due to rat
s
a
n
d
golde
n apple
snail
s
(
Pom
a
cea canali
c
ul
ata
). All infested fields (Fig
ure 4)
were repla
c
ed by n
e
w
see
d
ling
s
. Th
is lead
s to un
even app
eara
n
ce o
n
parti
cular field
s
.
16-20
21-
25
26-30
31-
35
36-
40
41-45
46-
50
51-55
56-
60
61-
65
66-
70
71-
75
76-80
81-
85
91-95
96-
100
101-105
106
-110
Ag
e
-0
.2
-0
.1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
ND
V
I
Figure 3. Observed NDVI o
v
er Cihe
ra
ng
Cultivar
Figure 4. Infestation by Rat
s
NDVI in
crea
ses
duri
ng ve
getative ph
ase until
ear
ly gene
rative (h
eadin
g
) stag
e.
At
this
point, NDVI start
s
to decrease. As sho
w
n in Fi
gu
re
3, NDVI of
Cihe
ran
g
vari
ety reach
e
s t
he
pea
k
at
ab
ou
t
90 days. Th
is re
sult
a
p
p
r
oves a pr
evio
us study
by mean
s
of tim
e
seri
es MO
DIS
observation i
n
the same
si
te [14].
B
4
<
=
56.
36
B
2
<
=
54.
72
B
3
<
=
31.
88
Y
N
T
i
lle
r
i
n
g
B
3
<
=
51.
96
Y
N
B
2
<
=
64.
46
Ea
r
l
y
T
i
lle
r
i
n
g
Ea
r
l
y
T
i
l
l
e
r
i
n
g
Mat
u
r
e
N
Y
N
Y
B
4
<
=
90.
60
Mat
u
r
e
N
Y
B
1
<
=
63.
21
H
eadi
ng
Y
N
B
1
<
=
64.
97
B3
<=
3
0
.
8
7
N
Y
Mat
u
r
e
H
eadi
ng
N
Y
Mat
u
r
e
T
ill
e
r
i
n
g
N
N
Y
Figure 5. De
cision T
r
ee for
AVNIR-2
Dat
a
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
cting Ri
ce Gro
w
th Usi
ng ALOS Mul
t
ispe
ctral an
d
Synthetic… (Bam
bang H. Trisasong
ko
)
5617
NDVI curve i
ndicates that
at least four gr
o
w
th sta
ges a
r
e visib
l
e. Those are early
tillering
(le
ss than
50
da
ys), tilleri
ng
(50-7
0
),
hea
di
ng (71
-
90
), and mature (more
than
9
0
).
Acco
rdi
ng to
those
classes, all
pixels in the
i
m
ag
ery were the
n
cl
assified
e
x
ploiting Ran
dom
Fore
st. Follo
wing fig
u
re
shows
a resul
t
of Ran
dom
Forest
deci
s
ion tre
e
, whil
e its a
s
so
ciat
ed
accuracy is p
r
esented in T
able 2.
Table 2. Accu
racy in Per
Cent, Based o
n
30% Sampl
e
Pixels
Classified
O
b
served
Earl
y
Tillering
Tillering
Heading
Mature
Ear
l
y
T
iller
i
ng
99
0
0
1
T
iller
i
ng 3
96
0
1
Heading
0
18
65
17
Mature
3
1
1
96
As sho
w
n in
Figure 5, ba
n
d
4 of AVNIR-2 pl
aye
d
in
cru
c
ial
rol
e
in
the develo
p
m
ent of
rule
s, a
nd
sel
e
cted
by th
e
algorith
m
a
s
t
he m
a
in
stem
. This is due t
o
the sensitivity of the band
(820.6
nm) to
vegetation,
whi
c
h h
a
s
be
en wi
dely kno
w
n o
n
a va
rie
t
y of optical sensors. Band
1,
whi
c
h is oft
en used to
identify wate
r bodie
s
,
wa
s su
bsta
ntial
as well, e
s
peci
a
lly on th
e
discrimi
nation
of headi
ng a
nd matu
re
stage
s. Ri
ce fi
e
l
ds a
r
e d
r
ai
ne
d duri
ng the
maturity of ri
ce.
Mean
while, the Re
d ban
d (Band 3
)
is e
m
ployed
to separate tilleri
ng and matu
re paddi
es. Lo
we
r
value of Ban
d
3 in
dicates a con
s
ide
r
a
b
le invo
lvem
ent of soil/wa
ter ba
ckgrou
nd, and
there
f
ore
these values
fit into tillering class.
Overall a
c
curacy usi
ng AVNIR-2 data o
b
tained in thi
s
expe
riment
is rea
s
o
nabl
y good,
as i
ndi
cated
i
n
Ta
ble
2. Howeve
r, it a
p
pears that
h
e
ading
is com
paratively
co
nfuse
d
to tille
ring
and matu
re stage
s. This
is fairly unde
rstan
dabl
e si
nce the cl
ass ha
s no
sig
n
ificant featu
r
e
s
whi
c
h influe
n
c
e di
stin
ctive spe
c
tral
sig
n
a
ture. NDVI
pattern p
r
e
s
e
n
ted in Fig
u
re 3 also supp
orts
the argu
ment
, where h
ead
ing is indi
cat
ed by gent
le
slop
e, sho
r
tly before the p
eak (th
e
end
of
vegetative ph
ase
)
. Thi
s
al
so sugg
est
s
th
at even
Band
1
ca
n b
e
e
m
ployed fo
r
discrimi
nation
of
mature a
nd h
eadin
g
, the bias re
main
s consi
derably h
i
gh.
3.2. Back
sca
tter
Coe
fficie
n
t
a.
86-
90
b.
96-
1
0
0
c
.
101-
105
d.
106-
110
e.
111-
115
f
.
116-
12
0
g.
121-
1
2
5
h
.
126-
1
3
0
i
.
131-
135
Ag
e
-
26.
0
-
24.
0
-
22.
0
-
20.
0
-
18.
0
-
16.
0
-
14.
0
-
12.
0
-
10.
0
-8
.
0
-6
.
0
-4
.
0
-2
.
0
0.
0
2.
0
B
a
cksca
t
t
e
r
co
e
f
f
i
ci
e
n
t
(
s
i
g
m
a
nought
)
V
V
H
V
H
H
;
O
u
t
l
i
e
r
s
E
x
t
r
e
m
e
s
Figure 6. Backscatte
r Patterns of
Rice P
henol
ogy
. Prolong
ed ri
ce
planting
wa
s due to extensive
repla
n
ting in
2007.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5613 – 56
20
5618
Backscatter i
n
tensitie
s h
a
v
e been
em
ployed to o
b
se
rve a
nd
to model i
n
tera
ction
betwe
en ra
d
a
r si
gnal
s a
nd su
rfa
c
e
material
s. Th
is re
sea
r
ch focu
se
s on t
he ob
servati
on of
sigma
-
n
ought
among va
rio
u
s a
g
e
s
(afte
r tran
spl
antin
g). Du
e to m
ono
static n
a
ture of PALS
A
R
sen
s
o
r
, VH d
a
ta can b
e
o
m
itted as a consequ
en
ce
of the reci
pro
c
ity theorem.
Followin
g
fig
u
re
sho
w
s variabi
lity of backscatter data extracte
d from
combine
d
200
7 and 20
09 PLR data.
As sho
w
n, HH has the hi
ghe
st backscatter
coeffici
ents. On HH polari
z
ation,
higher
amplificatio
n
of backsca
tter is due
to stro
n
g
contributio
n o
f
double b
o
unce scattering
mech
ani
sm.
L-ba
nd
HH
si
gnal
s are ca
pable to
pen
etrate cano
p
y
layer and t
herefo
r
e i
n
teract
with
soil
ba
ckgroun
d. Sin
c
e th
e
soil
is usually
floo
ded
du
ring
p
eak vegetativ
e sta
ge,
sig
n
als
con
s
e
c
utively
interact
with soil an
d the ri
ce.
High
er ba
ckscatter
contrib
u
tion
of HH wa
s
al
s
o
re
po
r
t
e
d
b
y
nu
me
r
o
us r
e
s
e
a
r
c
h
er
s [6
-
7]. At the early tillering, Wang
et al
. [7] reported low back
sc
a
tter
returns
due to
s
p
arse
vegetation, a
nd the
r
efore, overall
scattering
is
domi
nated by
sp
e
c
ula
r
p
r
op
ag
ation. Thi
s
st
age
wa
s mi
ssed i
n
the
re
sea
r
ch due
to la
ck of PL
R
data
and
mo
st bl
ocks were
in
headi
ng o
r
p
r
e-
harve
st stage
s. Ho
wever, i
t
appears tha
t
our re
su
lts
are in line a
n
d com
p
leme
n
t
ary to previo
us
finding
s [7], with si
milar
a
b
sol
u
te value
of sigm
a-n
o
ught. De
clini
n
g SAR return
s at L
-
ba
nd
were
also
re
po
rted
[15]. Structu
r
al pr
ope
rty, i.e. drying
leav
es,
wa
s repo
rted a
s
the
ma
in cau
s
e
of th
e
decli
ne [7].
Amplification
of HH was
reporte
d in
m
e
ch
ani
cal-ba
sed plantin
g sy
stem i
n
Ja
pan [16]
and i
n
Chin
a
[6]. Although
the San
g
Hya
ng Se
ri di
d n
o
t implem
ent
fully mechani
cal
syste
m
, field
observation
discovered t
hat spa
c
e
b
e
twee
n ri
ce
wa
s re
gula
r
(sq
u
a
r
ed, wit
h
20 cm
sp
ace
betwe
en pla
n
t
s). Thi
s
re
gu
lar spa
c
e can
cre
a
te the B
r
agg
scatterin
g
, which ob
servable i
n
hig
h
percenta
ge of
HH si
gnal
s.
Figure 7. Reg
u
lar Pattern o
f
Rice Plantin
g
Specifically for VV pola
r
i
z
ation, Fig
u
re
4 sh
ows h
i
gh variatio
n of SAR backscatters.
Previou
s
ly, VV was d
e
mo
nstrate
d
lea
s
t useful for
ri
ce monitori
ng
[15]. This is
also reflecte
d
by
our re
sult
s. It app
ears that
disag
r
eem
e
n
t rem
a
in
s
in
fluctuate
d
b
a
ckscatte
r
co
efficients du
ri
ng
rice
g
r
o
w
th, e
s
pe
cially
at C-ban
d. Durde
n
et al
. [1
7], for in
stan
ce,
found
that mi
crowave
returns
tended to de
crea
se du
rin
g
gro
w
ing p
e
rio
d
at C-ba
nd
AIRSAR. Ho
wever, T
s
an
g
et al
. [18] mo
del
indicated that
there is an in
cre
a
se of SAR
bac
ksc
atters
of ER
S-1 V
V
polari
z
ation
.
Low retu
rn
s o
f
SAR si
gnal
s at HV
are
p
r
imarily du
e to
compl
e
x p
r
op
agation
of si
g
nals in
vegetative co
ver. Thi
s
wa
s extensively
repor
te
d by
n
u
merou
s
p
a
p
e
rs,
incl
udin
g
Santoro
et al
.
[19]. We
foun
d that
HV p
a
ttern
of Ci
her
ang
cultivar i
s
q
u
ite
simila
r to
Ling
et al
. [6] obs
e
rvation
in East Chi
n
a
using
dual
-p
olari
z
ed PAL
SAR. HV pattern i
s
also compa
r
abl
e to Wang
et al. [7]
finding
s.
PALSAR data spa
n
wa
s
quite differen
t
than AVNIR-2. This
creat
ed a co
nditio
n
whe
r
e
PALSAR dat
a co
ntain o
n
l
y headin
g
a
nd matu
re p
hases.
Usi
n
g
Ran
dom F
o
rest, a
simpl
e
sep
a
ratio
n
rul
e
bet
wee
n
h
e
ading
an
d ma
ture
wa
s e
s
ta
blish
ed
(Fig
ure 8).
It is i
n
teresting
that V
V
polari
z
atio
n wa
s sele
cte
d
by Ran
d
o
m
Fo
re
st
al
gorithm,
eve
n
a
s
sessm
e
nt of ba
ckscatter
sho
w
e
d
high
er prefere
n
ce
of HH. This
is pro
bably d
ue to the effect of gro
upi
ng rice age i
n
to
gro
w
th
cla
s
s
(hea
ding
an
d
mature). An
other
so
ur
ce of
the
bi
as would be
p
a
ra
meter sel
e
cti
on,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
cting Ri
ce Gro
w
th Usi
ng ALOS Mul
t
ispe
ctral an
d
Synthetic… (Bam
bang H. Trisasong
ko
)
5619
whi
c
h i
s
su
bject fo
r furth
e
r asse
ssment.
Du
e to
si
mpl
e
sepa
ration,
it is
obviou
s
to se
e that th
e
overall
accu
racy ten
d
s to
be hig
h
e
r
(Ta
b
le 3
)
. Simila
r to tho
s
e
in
AVNIR-2,
he
ading
wa
s
ha
rde
r
to detect in
al
l PALSAR PL
R data. M
o
st
of pixels
b
e
lo
ng to he
adin
g
wa
s repo
rted
ly miscla
s
sified
into mature
c
l
ass
.
Figure 8. De
cision T
r
ee for
PALSAR Line
ar Polari
zatio
n
Table 3. Co
nfusio
n Matrix Based o
n
30
% Samples
Classified
Observed
Heading
Mature
Heading
65 17
Mature
1
96
4. Conclusio
n
This re
se
arch de
mon
s
trat
es th
e
cap
a
b
ilities
of AL
OS sen
s
ors
to retri
e
ve in
valuable
informatio
n o
n
ri
ce
field
s
. It wa
s
sho
w
n
that grow
th
di
sturb
a
n
c
e,
especi
a
lly in tra
n
spl
anting
an
d
early vegetati
v
e stage
s, was
so
me
wh
at visible from
NDVI plot. V
egetative sta
ge of Cih
e
ra
ng
cultivar was
pea
ked at ab
out 90 days,
simila
r to
pre
v
ious ob
se
rvation usi
ng time se
rie
s
MODIS
data. Cla
ssifi
cation by me
ans of Rand
o
m
Fore
st
on AVNIR-2 dat
a
pro
duced substa
ntially
high
accuracy. Ho
wever, imp
r
o
v
ements a
r
e
still r
equi
red
since hea
di
ng stag
e wa
s co
nfused
with
tillering
and mature stages.
The
research was unabl
e
to
collect
PLR data
covering full phase of
rice
p
r
odu
cti
on. In thi
s
case,
early till
ering
an
d till
ering
ph
ases we
re mi
ssin
g. The
r
efore, full
con
c
lu
sio
n
cannot b
e
d
r
a
w
n u
s
in
g
sol
e
ly avail
able
PLR
data.
Ho
wever, it i
s
inte
re
sting
to
observe
that difficulties
to discrimi
nate headi
ng
sta
g
e
are q
u
ite similar on b
o
th AVNIR-2 a
n
d
PLR data.
These all
su
gge
st that AVNIR-2 an
d
PALSAR
fully pola
r
imetri
c
datasets a
r
e
useful t
o
monitor vari
o
u
s ri
ce g
r
o
w
ths. No
nethel
ess, seve
ral issue
s
nee
d to be addressed. Availabil
i
ty
has b
een a
primary ob
stacle for full
y polarimet
ri
c SAR. It is sugg
este
d that regul
ar f
u
lly
polari
m
etri
c
acq
u
isitio
n cycle n
eed
s t
o
be
revi
sed
to a
c
comm
odate
a tho
r
ough
ri
ce fie
l
d
observation. Acqui
sition
s
i
n
a co
uple
of
we
eks
after mi
d-sea
s
o
n
of
rainy m
onsoon
(a
bo
ut
De
cemb
er-Ja
nuary at
the
test site) are
sugg
es
te
d
to examin
e
appli
c
ability
of SAR d
a
ta
for
estimating rice
prod
uctivity.
Ackn
o
w
l
e
dg
ements
This research was
supported by JAXA and REST
EC, J
apan. Additional funding
was
provide
d
by Indone
sian
Ministry of Ag
ricultu
r
e
through K
KP3T proje
c
t. The authors
ackno
w
le
dge
sub
s
tantial a
s
sista
n
ce of Mr. Masato
shi
Kamei of RESTEC and Mrs. Ita Carolita
of
LAPAN who
managed i
m
age
acqui
sition. We are
grat
eful to our
Japanese
fellows for t
heir
enthu
sia
s
m a
nd supp
ort d
u
ring
the research.
We
th
ank
ou
r stu
d
ents at th
e Departm
ent of
Soil
Scien
c
e a
n
d
Land
Re
so
urces, Bog
o
r Agricultu
r
al
University for their
assi
sta
n
ce d
u
ri
ng the
r
e
sear
ch.
Referen
ces
[1]
Van V
a
lke
n
b
e
rg S. Java: T
h
e
Econom
ic Ge
ogra
p
h
y
of A T
r
opic
a
l Isla
nd.
Geogra
phic
a
l Review
. 19
25;
15(4): 56
3-5
8
3
.
[2]
Panu
ju
DR, Mi
zuno
K, T
r
isasongk
o BH. T
he D
y
n
a
mics
of
Rice
Pro
ducti
on i
n
In
do
nesi
a
1
961-
20
09
.
Journ
a
l of the
Saud
i Soci
et
y of Agricultur
al
Scienc
es
. 201
3; 12(1): 27-3
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5613 – 56
20
5620
[3]
Hald
ar D, Patn
aik C. S
y
n
e
rgis
tic Use of Mu
lti
-
te
mp
o
r
al
R
ada
rsa
t
SAR
an
d AWi
F
S D
a
ta
fo
r R
a
b
i
Ri
ce
Identificati
on.
J
ourn
a
l of Indi
a
n
Society of Re
mote S
ensi
n
g
.
201
0; 38(1): 15
3-16
0.
[4]
Che
n
J, Lin H,
Hua
ng C, Fan
g
C.
T
he Rel
a
tio
n
sh
i
p
bet
w
e
e
n
the Le
af Area I
nde
x (
L
AI) of Rice an
d th
e
C-ba
nd SA
R V
e
rtical/H
orizo
n
tal (VV/H
H
) Po
l
a
rizati
on
Rati
o.
Internati
o
n
a
l J
ourn
a
l
of R
e
mote Se
nsi
n
g
.
200
9; 30(8): 21
49-2
154.
[5] Ishitsuka
N.
Observatio
n o
f
Japan
ese P
addy F
i
elds
Using
PALSA
R Data
. IEEE Internationa
l
Geoscie
n
ce a
n
d
Remote Se
n
s
ing S
y
mp
osiu
m. Vancouver,
Can
ada. 2
011:
1938-
19
41.
[6]
Lin
g
F
,
Li Z
,
C
hen E, T
i
an X,
Bai L, W
ang
F
.
Rice Areas
Mapp
ing Us
in
g
ALOS PALSAR F
B
D Data
Consi
der
ing t
he Bra
gg Sc
attering
in
L-ban
d SAR I
m
ages
of Rice
F
i
elds
. IEEE Internationa
l
Geoscie
n
ce a
n
d
Remote Se
n
s
ing S
y
mp
osiu
m. Honol
ulu, U
SA. 2010: 14
6
1
-14
64.
[7]
W
ang C, W
u
J, Z
hang Y, Pan G, Qi
J, Salas W
A
.
Charact
e
rizin
g
L-b
and
Scattering of P
add
y Ric
e in
Southe
ast Ch
i
na
w
i
t
h
R
adi
ative T
r
ansfer M
ode
l an
d Mu
ltitempor
al AL
OS/PALSAR Imager
y.
IEEE
T
r
ansactio
n
s o
n
Geoscie
n
ce
and R
e
mote S
ensi
n
g
. 20
09; 47(4): 98
8-9
9
8
.
[8]
Lillesand T
M
,
Kiefer
RW, Chipma
n
JW
. Re
mote Se
nsi
ng
and
Imag
e In
terpretation. Fifth
Editi
on.
N
e
w
York: John W
i
l
e
y
& So
ns. 200
5: 546-5
50.
[9]
Shima
da M, Isoguc
hi O, T
adono T
,
Isono K. PAL
SAR Rad
i
ometric a
nd G
eometric C
a
li
br
ation.
IEEE
T
r
ansactio
n
s o
n
Geoscie
n
ce
and R
e
mote S
ensi
n
g
. 20
09; 47(1
2
): 391
5-3
932.
[10]
F
r
iedl MA, Bro
d
le
y CE. D
e
ci
sion T
r
ee Cla
ssificatio
n
of L
and C
o
ver fro
m
Remotel
y
S
ense
d
Data
.
Re
mote Se
nsi
ng of Enviro
n
m
ent
. 199
7; 61(3
)
: 399-40
9.
[11]
Breima
n L. Ra
ndom F
o
rests.
Machi
ne Le
arn
i
ng
. 20
01; 45(
1
)
: 5-32.
[12]
Z
hou L, W
a
n
g
H. Loa
n D
e
fault Pre
d
icti
on
on Lar
ge Imbala
n
ced D
a
ta
Usin
g
Ra
ndom
F
o
rests.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(6): 1
519-
152
5.
[13]
M
y
i
n
t SH. Overlo
ad P
a
ttern
Class
ificatio
n
for Server O
v
erlo
ad D
e
tect
ion.
Inter
natio
nal J
our
nal
of
Information &
Network Security
. 2013; 2(3): 260-
266.
[14]
Panu
ju
DR, H
e
idi
n
a
F
,
T
r
isason
gko B
H
,
T
j
ahjono
B, K
a
sno A, S
y
afril AH
A. Var
i
asi
Nil
ai In
dek
s
Vegetas
i MOD
I
S pada S
i
klu
s
Pertumbu
ha
n Pad
i
.
Jurna
l
Ilmi
ah Ge
omatika
. 20
09; 1
5
: 9-16. (i
n
Indon
esi
an).
[15]
Inoue
Y, Kur
o
su T
,
Maeno
H, Ur
atsuk
a
S,
Kozu
T
,
Dabro
w
ska-Z
ie
li
nsk
a
K, Qi J. S
e
a
s
on-l
ong
Da
il
y
Measur
ements
of Multifrequ
e
n
c
y
(K
a, Ku, X, C, and L) an
d Full-Po
lariz
a
tion Backsc
atte
r Signatur
es
over
Pa
dd
y Rice
F
i
eld an
d
their Rel
a
ti
onsh
i
p
w
i
t
h
Biol
ogic
a
l
V
a
ri
abl
es.
Re
mot
e
Sens
in
g o
f
En
vi
ronm
e
n
t
. 2
002; 81(
2-3): 1
94-2
04.
[16]
Ouchi K, W
a
n
g
H, Ishitsuk
a
N, Saito G,
Mohr
i K. On t
he Bra
gg Sc
at
tering Obs
e
rve
d
in
L-b
a
n
d
S
y
nt
hetic A
per
ture R
adar
Images
of F
l
o
o
d
ed R
i
ce
F
i
el
ds
.
IEICE T
r
ansactions
on
Co
mmu
n
icati
o
n
.
200
6; E89-B(8)
: 2218-2
2
2
5
.
[17]
Durd
en SL, M
o
rrisse
y
LA, Li
vingsto
n GP. Micro
w
av
e Ba
ckscatter and
Attenuatio
n De
pen
de
nce o
n
Leaf Ar
ea In
d
e
x for F
l
o
ode
d
Rice F
i
elds.
I
EEE Transactions on Geosci
ence
and Remote Sensing
.
199
5; 33(3): 80
7-81
0.
[18]
T
s
ang L, Ko
ng
JA, Di
ng K,
Ao CO. Scatte
ring
of
El
ectro
m
agn
etic W
a
v
e
s 2:
Numer
i
c
a
l Sim
u
l
a
tion
,
Ne
w
York: W
i
l
e
y
Intersc
i
enc
e
.
2001: 68
5-69
2.
[19]
Santoro
M, Fransso
n JES,
Eriksson
LEB,
Mag
nuss
on
M, Ula
nder
L
M
H, Olsson
H
.
Sign
atures
o
f
ALOS PALSA
R L-
band B
a
c
kscatter in S
w
edis
h For
e
st.
IEEE Trans
actions
on Geoscienc
e and
Rem
o
te Sensing
. 200
9; 47(1
2
): 4001-
40
19
Evaluation Warning : The document was created with Spire.PDF for Python.