Spectral Characteristic Comparison of Rice Plants under Healthy and Water-Deficient Condition Using Landsat ETM+ Data.

論文

S pectr al char acter istic compar ison of r ice plants under healthy and
water - def icient conditions using L andsat E T M ＋ data

Landsat 衛星による水稲耕作地への水供給状態の違いによる
水稲の分光特性比較研究
I Wayan NU A R SA , F umihiko NISH IO and Chiharu H ONGO
I Wayan NU A R SA ・西尾 文彦 ・本郷 千春
A bstr act : T he spectral characteristics of rice plants subject to water deficiency were analysed and compared
with healthy rice plants. U nder water deficiency conditions, the reflectance of visible bands of the rice plants
increased significantly, while the reflectance of near - and middle- infrared bands decreased. F or the early detection
of water deficiency, the visible band of L andsat E T M ＋ is more sensitive than the infrared bands. B and- 3 showed
the highest reflectance differences between rice under water - deficient and healthy conditions. T he use of the
vegetation index can distinguish rice under water deficiency more clearly than when only using a single band. T he
R atio V egetation Index (R V I) is the best vegetation index for both early detection of water deficiency and
distinguishes the two rice conditions.
和文概要：稲作地での水供給状態の違いによる水稲の分光特性を比較した。水供給不足の状況では，水稲からの可視バ
ンドの反射率は著しく増大したが，近赤外及び中間赤外バンドは低下した。水供給が不充分な初期には，L andsat E T M ＋
の可視バンドの反射率は，赤外バンドよりも敏感であった。なかでもバンド３は，水不足と充分な状況では大きな反射率
の違いを示した。植生指標（N DV I ）は，単バンドの反射率を用いるよりも水供給不足と十分な状態における生育状況を区

別することが出来る。稲作指標（R V I ）は，初期の水稲作付け状況での水供給状況を検知するために導入し有効であること
が明らかになった。

worldwide for large- scale field monitoring of plant

１. Introduction

populations, and, through its use, timely and precise
site- specific actions may be tak en (B ouman, 1995 ;

R ice is the primary food source for more than

P ierce et al., 1999). If rice crops can be attended to

three billion people and is one of the world s major

using accurate k nowledge of their status in the

staple foods. P addy rice fields account for approx i-

growth stages, it is possible to respond positively to

mately 15% of the world s arable land (K hush, 2005 ;

the supply and demand of varietal growth and

IR R I, 1993). T he growth and yield of rice crops are

production rates.

not only determined by their genetic compositions

V egetation reflectance is a function of optical

but are also affected by the environmental factors

properties, canopy biophysical attributes, viewing

of their habitats. T he progress of a crop is conven-

geometry, illumination conditions, and back ground

tionally evaluated by periodical plant samplings,

effects (A sner, 1998 ; B arrett and Curtis, 1992 ; J ac-

and management practices are guided by the perfor-

quemoud et al., 1992 ; M yneni et al., 1989). T he

mance in the field (Y ang et al., 2008). N owadays,

biophysical attributes of a vegetation crop drive

remote sensing is the preferred technology adopted

variation in the canopy reflectance characteristics
due to the three- dimensional orientation, which

F aculty of A griculture, U dayana U niversity

Center for E nvironmental R emote S ensing, Chiba
U niversity
写真測量とリモートセンシング」V OL . 50，N O. 2，2011
66

provides a better structure and opportunities for
photons to interact with multiple surfaces of different plant parts (e.g., leaves, stems), thus favouring

radiometric reception (Y ang and Cheng, 2001 ;

2007 ; T ong et al., 2001). T he use of actual remote

B aret et al., 1994 ; J acquemoud, 1993 ; K upiec and

sensing data of rice plants under abnormal condi-

Curran, 1995 ; Y ang and S u, 2000).

tions has been limited by difficulties in finding

T he reflectance of agricultural crops in the visible

sufficient areas under such conditions (Currey et al.,

and infrared regions is currently being used to

1987). T his study used the actual remote sensing

assess different crop parameters and plant growth

data (L andsat E T M ＋ images) to monitor the spec-

status (Carter and K napp, 2001 ; L ee et al., 2008 ;

tral properties of rice plants under water stress. T he

Y ang and Chen, 2004). M ost green plants ex hibit a

ground data were obtained from the harvest failure

spectral reflectance profile of a higher reflectance

of farms caused by drought (low rainfall) in a broad

in the near - infrared and a lower reflectance in

area of about 421 ha. T he objectives of the studies

visible light when grown under normal conditions

was to analyse differences in the spectral character-

(H all et al., 2002). In water stress environments, a

istics of rice plants under water stress compared to

reverse fashion is observed due to a decrease in

those under normal conditions using L andsat

plant vigour and canopy coverage and an increase

E T M ＋ data.

in the reflectance of chlorophyll absorption (K nipling, 1970 ; Y ang et al., 2007). Crop foliage density

２. S tudy Area, Data, and Methods

and changes in geometry play a k ey role in the
reception of incident radiation and the following

2. 1 Backg round and S tudy Area

biomass formation and accumulation. V ariations in

T he study area was located in T abanan R egency,

crop vegetation, in regular conditions or under stres-

B ali P rovince, Indonesia, centred at a latitude of

ses, differ greatly in reflectance behaviour, and this

8°31 50 S and a longitude of 115°02 30 E (F ig. 1).

may be assessed from the canopy spectral charac-

T he T abanan R egency was selected for the study

teristics (Carter and K napp, 2001 ; Y ang and Chen,

area due to T abanan being the central production

2004).

area of rice in B ali. In the study area, the rice

T he monitoring of rice plants using remote sensing data has been widely done. H owever, the studies
have been focused on estimations of rice field areas
and production estimations using either optical

remote sensing (B ailey et al., 2001 ; Ok amoto and
K awashima, 1999 ; F ang et al., 1998 ; N uarsa et al.,
2005 ; N uarsa and N ishio, 2007) or radar systems
(S hao et al., 2001 ; K urosu et al., 1997 ; Chak raborty et al., 2000 ; R ibbes and L e T oan, 1999). Investigations of rice plants under abnormal conditions are
currently still very limited. M ost of them concern
ａ

ｂ

the spectral characteristics of rice plants infested by
pests and plant diseases (Qin and Zhang, 2005 ;
Y ang and Cheng, 2001 ; Y ang et al., 2007), and only
a few studies relate to water deficiency (Y ang and
S u, 2000 ; K ok sal et al., 2008). S tudies of the rice
plants under abnormal conditions have been done
mostly using ground sensing spectroradiometers to
acquire the reflectance of rice plants (Y ang et al.,

Fig . 1 Location map of the study area. T he black colour
indicates rice field areas, and the white colour

indicates other land use areas. Left below ⒜ is
the study area for the rice under water deficiency.
T he g reen point next to the study area is the
nearest rainfall station used to obtain a rainfall
data. Rig ht below ⒝ is the study area for the rice
under healthy conditions. T he three orang e points
in that imag e are the g round sites where the field
measurement was done.
67

planting is organised by a subak . T he subak is the

T able 1

Monthly rainfall of the nearest rainfall station
from the study area (BMG, 2006)

farmer s social organisation that manages the irriR ainfall
(mm)

gation water. E ach subak controls around 150−300
M onth

ha of rice fields (F ood Crops A griculture DepartJ anuary
F ebruary
M arch
A pril
M ay
J une
J uly
A ugust
S eptember
October
N ovember
December

ment, 2006). T hey usually plant rice at the same
times. A n advantage for the remote sensing study
with this farming system is that the monitoring is
made easier due to the availability of a wide rice
area.
B ecause irrigation water is a constraining factor
in the study area, rice planting is done in turns
between the Subaks. A lternative plants to rice plants
include session agriculture crops such as corn, soy-

322
183
295
138
0
27
106
14
76
214
53
391

beans, etc. T he length of the growth cycle of rice
can vary from three to six months for different
varieties (Casanova, 1998). In the study area, the
average length of the growth cycle of rice is 115
days, with production reaching 5.0 tons/ ha (F ood

T he zero value of rainfall in M ay indicated that the
rainfall in that month was less than 1 mm. P lanting was
done on 15 A pril 2005. B eginning from M ay, the rice
plants did not get water, and in J une the rainfall was
very low. T his condition caused the rice to ex perience
drought.

Crops A griculture Department, 2006). In one year,
the farmer usually plants rice twice in two planting

T able 2

S ite characteristics of rice under healthy and
water- deficient conditions

sessions.
T his study was carried out on two different sites
and during two planting sessions. One ex perimental
session was conducted under healthy rice conditions, and the other session was conducted under
water deficiency. F or the healthy rice conditions,
field surveys were done in three different site stations every 16 days when the L andsat S atellite
passed the study area. S ome rice growth parameters, such as the rice height, the total number of
stem, the percentage of rice coverage, and the rice
age, were measured. F or the water deficiency conditions, field surveys were not performed, as in this
case the condition was not anticipated. T his event
became k nown after the A griculture Department of

Parameters

H ealthy rice

W ater- deficient rice

Soil type
L and slope (% )
E levation (m above
sea level)
(F ertiliser and pesticide used)

L atosol / V ertisol
0- 15
38- 52

L atosol / V ertisol
0- 8
5- 15

A griculture Department recommendation
W ater source
Permanent irrigation
R ice variety
Ciherang
Peak of T ransplant- 12 J une 2005
ing
A verage temperature 25.7
in planting session
(°C)
A verage duration of 75
solar radiation in
planting session (% )

A griculture Department recommendation
R ainfall
Ciherang
15 A pril 2005

26.1

79

the local government released relevant reports.
H owever, the locations and planting dates of these
areas were precisely recognised. H ence, the investi-

outside the normal planting season. T he farmers

gation of rice under healthy and water deficiency

had speculated on rain coming soon after planting.

conditions could evidently be done successfully.

U nfortunately, the rainfall had not begun from the

A ccording to the A griculture Department of the

rice age of 15 days after planting until one month

local government, the cause of the occurrence of

after (T able 1). T his situation caused around 421 ha

drought was that the planting of rice was done

of rice plants in the study area to ex perience a

68

drought. T he rice production under these circum-

L andsat E T M ＋ images is 16 days, the total of the

stances was approx imately 0.35 ton/ ha of grain dry

time series images that can be collected in one rice

weight (less than 6% of normal condition) (F ood

growth cycle is around six images for different

Crops A griculture Department, 2006).

acquisition dates. S ome of the images cannot be

T he site characteristics of the rice under healthy

used, due to cloud conditions or the appearance of

and water - deficient conditions during the rice grow-

gap values in the S L C - off mode of the station

ing circle were the same for the soil type and rice

points. T his caused a reduced availability of images.

variety and were similar for the land slope, eleva-

F ortunately, this study area was covered by two

tion, average temperature, average duration of

scenes of L andsat images in different paths. T he

solar radiation as well as fertiliser and pesticide

total of the L andsat images used in this study is

treatment. H owever, the water resources for the

shown in T ables 3 and 4 for rice in sites with healthy

rice field irrigation were different (T able 2). In the

and water - deficient conditions, respectively.

healthy rice area, water was mainly derived from
the permanent irrigation water. On the other hand,

2. 3 Data analysis

for rice in water - deficient sites, water comes solely

2. 3. 1 Radiometric corrections

from rainfall. T he permanent irrigation water guar-

In the temporal analysis of remote sensing data,

antees the availability of water and the sus-

radiometric corrections are an important part of the

tainability of healthy rice crops more than rainfall

image analysis. T he digital number (DN ) of the

can. T he peak of transplanting was done on 12 J une

L andsat E T M ＋ at different acquisition dates

2005 at healthy rice and 15 A pril 2005 at W ater -

should be converted to the corrected digital number

deficient rice.

(cDN ) to eliminate the radiometric and atmospheric
effects, so that they have comparable values. In this

2. 2 Landsat Imag e Data

study, we used the simple radiometric correction

L andsat satellite images had eight bands includ-

model introduced by P ons and S ole- S ugranes (1994).

ing thermal and panchromatic bands. In the visible,

T he formula of this model is shown in the following

near - infrared, and middle- infrared regions, L andsat

equation :

E T M ＋ had a 30- m spatial resolution. H owever, in
cDN ＝1000a(DN −K d /［μ S e

e

］
,

(1)

the thermal and panchromatic regions, the spatial
(i)

(if 250＜cDN

study used visible and reflectance infrared bands

(ii)

(if cDN ＞318.3 ; cDN ＝255),

(B ands 1- 5 and B and 7) of L andsat E T M ＋.

(iii) (if μ

resolutions were 60 m and 15 m, respectively. T his

318.3 ; cDN ＝254),

0 ; cDN ＝255),

A lthough the L andsat E T M ＋ used in this study has
where cDN ＝corrected digital number
a scan- line corrector (S L C) off, the better spatial,
a ＝gain value of each L andsat band
spectral, and temporal resolution of its images
DN ＝digital number
caused it to still be relevant for use. W ith the 16
K ＝radiances value of zone totally in shade
days of temporal resolution, L andsat E T M ＋ is the
d＝actual S un- E arth distance
ideal satellite image for rice monitoring seeing as
μ ＝cosine of the incident angle
rice has only a 115- day cycle from planting to
S ＝ex oatmospheric solar irradiance
harvest. T hese images can be downloaded at U S GS
τ ＝optical depth of the atmosphere
Global V isualization V iewer (http:/ / glovis.usgs.
μ ＝cosine of the solar zenith angle
gov).
μ ＝cosine of sensor view angle
B ecause the length of the rice growth cycle is
around 115 days and the temporal resolution of

cDN is the corrected digital number and the conver69

T able 3

Values of d ⑴, μ ⑵, and K ⑶ for every acquisition date of Landsat E T M＋ in healthy rice
K

A cquisition date

P ath

R ow

DOY

d

μ

26 A pril 2002
21 M ay 2002
7 J uly 2005
1 A ugust 2005
17 A ugust 2005
24 A ugust 2005
4 October 2005
12 N ovember 2005

116
116
117
116
116
117
116
117

66
66
66
66
66
66
66
66

116
141
188
213
229
236
277
316

1.00626
1.01210
1.01669
1.01497
1.01244
1.01103
1.00033
0.98983

0.79414
0.75100
0.71514
0.74624
0.78085
0.80312
0.88006
0.90081

ET M1 ET M2 ET M3 ET M4 ET M5 ET M7
51
53
52
50
53
55
54
52

31
32
34
31
33
35
34
31

23
22
22
23
23
24
24
22

20
23
25
19
22
23
21
24

17
19
21
16
18
15
17
21

11
14
11
12
13
10
12
11

S ources : ⑴ Chander et al. (2009) ; ⑵ our image with calculations ; ⑶ the DN of our image. DOY is the day of year.

T able 4

Values of d ⑴, μ ⑵, and K ⑶ for every acquisition date of Landsat E T M＋ in water- deficient rice
K

A cquisition date

P ath

R ow

DOY

d

μ

117
116
116
117
116
117
116
117

66
66
66
66
66
66
66
66

108
117
133
140
149
172
181
204

1.00409
1.00653
1.01043
1.01191
1.01355
1.01625
1.01665
1.01592

0.80647
0.79287
0.76522
0.75302
0.73857
0.71480
0.71309
0.72621

ET M1 ET M2 ET M3 ET M4 ET M5 ET M7
18
27
13
20
29
21
30
23

A pril 2005
A pril 2005
M ay 2005
M ay 2005
M ay 2005
J une 2005
J une 2005
J uly 2005

52
52
56
48
54
62
57
51

33
31
34
30
34
41
36
33

23
23
23
22
23
25
27
23

18
23
22
18
22
23
21
24

15
18
17
15
18
13
18
22

11
15
12
12
12
9
12
12

S ources : ⑴ Chander et al. (2009) ; ⑵ our image with calculations ; ⑶ the DN of our image. DOY is the day of year.
T able 5

Values of τ ⑴, S ⑵, and a ⑶ for every spectral
Landsat Band for both healthy and waterdeficient rice

aspect of each pix el ; d depends on the date of the
satellite pass, and μ depends on the sensor viewing
angle. T he parameter μ

B and

τ

S (W m μ m )

a

ET M1
ET M2
ET M3
ET M4
ET M5
ET M7

0.5
0.3
0.25
0.20
0.125
0.075

1997
1812
1533
1039
230.8
84.9

0.7757
0.7957
0.6192
0.9655
0.1257
0.0437

is 1 in most L andsat

images because v is 0 at the nadir and has small
values on the rest of the images (P ons and S oleS ugranes, 1994). T he parameters needed in the study
to calculate cDN are shown in T ables 3- 5.
P ractically, to apply the algorithm above we only
need a DE M of enough quality (altimetrically and
S ources : ⑴ Dozier (1989) ; ⑵ Chander et al. (2009) ; ⑶
our image with calculations.

planimetrically) because all of the other parameters
are k nown (e.g., S ) or can be inferred from images

sion of the effective reflectance to the common 8- bit

(e.g., K ). T o avoid overcorrections and undercorrec-

format of most image processors. T he range of the

tions on the ridges and channels and to account for

output values has been limited to between 0 and 255.

local phenomena, it is important to use a DE M with

N ote that a, K , S , and τ depend on the wavelength

a planimetric resolution comparable to the geomet-

and have different values for each spectral band ;

ric resolution of the image. N augle and L ashlee

K varies for each image because it is related to and

(1992) showed that a DE M of 95 m can be insuffi-

depends on atmospheric conditions ; μ

cient for a L andsat T M image over rugged terrain.

and μ

depend on the latitude, date, and time of the satellite

In this study, we derived a DE M from the topo-

pass ; besides that, μ also depends on the slope and

graphical map with a spatial resolution of 30 m.

70

T able 6

S everal veg etation indexes used in the study
for identifying the spectral characteristics of
rice in healthy and water- deficient conditions

analysis. In each image of an acquisition date, a 200
- pix el sample was tak en randomly in the field
observation site at both healthy and water - deficient

No

V egetation Index

F ormula

N ormalised D i f f erence
V egetation Index (N DV I)
R atio V egetation Index
(R V I)
Infrared P ercentage V egetation Index (IP V I)
D i f f erence V egetati on
Index (DV I)

nir −r
NDV I ＝
nir ＋r

sites (F ig. 1). T he average value of the sample was
1

2

3

4

used as a representative of the spectral value of that
acquisition date. B ased on the spectral value, the

RV I ＝

nir
r

nir
IPV I ＝
nir ＋r

vegetation index was calculated. S everal vegetation
index es that were evaluated in the study are shown
in T able 6.

DV I ＝nir −r

B ased on the spectral and vegetation index val100
nir −r ＋
0.5
nir ＋r

ues, spectral analysis was performed by means of

S oil - A djusted V egetation
(1＋L )(nir −r )
SA V I ＝
6
nir ＋r ＋L
Index (S A V I)
N ormalised D i f f erence
nir −swir 1
ND WI 1＝
7
nir ＋swir 1
W ater Index (N DW I - 1)
N ormalised D i f f erence
nir −swir 2
ND WI 2＝
8
nir ＋swir 2
W ater Index (N DW I - 2)

water - deficient conditions, using statistical analysis

H ere, nir, r, swr1, swr2, and L are the near - infrared
band, red band, middle- infrared (B and- 5 of L andsat
E T M ＋), middle- infrared (B and- 7 of L andsat E T M ＋),
and canopy back ground brightness correction factor,
respectively.

3. 1 Rice g rowth parameters under healthy condi-

5

T ransformed V egetation T V I ＝
Index (T V I)

comparing the values for rice under healthy and

(paired- samples T test).

３. R esults and Discussion

tion
B ased on field survey data and statistical analysis
of rice under healthy condition, there was a quadratic relationship between some of rice growth

2. 3. 2 S pectral Characteristic Analysis of Rice

parameters (rice height and stem total) and rice age.
H owever, percentage of rice coverage was better

P lants
Gathering reflectance values from the L andsat

ex pressed by a linear relationship with rice age.

pix els for both rice conditions over the entire time

R ice coverage height showed the best relationship

period of rice growth was the nex t step of the data

with the rice age, followed by rice height and stem

ｂ

ａ

ｃ
Fig . 2 Relationship between rice heig ht ⒜, stem total ⒝ and rice coverag e ⒞ with rice ag e.
71

Fig . 3 S pectral characteristic of rice under healthy and water- deficient conditions for all reflective bands of Landsat
E T M＋. T he white areas in the chart indicate that both reflectances of the rice have similar forms, while the dark
areas show different forms of rice reflectance.

total with the R

value were 0.9617, 0.9544, and

0.9294 respectively (F ig. 2a, F ig. 2b, and F ig. 2c).

3. 2 S pectral characteristics of rice under healthy
and water- deficient conditions
T he trends of rice spectra under both conditions
for rice aged 30 days after planting are similar. T he
tendency towards differences in reflectance values
is only seen when the rice enters its second month
(F ig. 3). In the first month, when rice from both
conditions are in a healthy state, the reflectance of
rice under water deficiency appears higher than that
in healthy condition sites for most L andsat E T M ＋
bands (F ig. 4). T his result is caused by the effect of
water coverage. In healthy rice sites, the availability of water for irrigation is sufficient. H owever, in
water - deficient rice sites, the availability of water

Fig . 4 Differences of the spectral characteristics of rice
at similar ag es at times when the rice plants
under water deficiency did not yet show sig ns of
water stress.

is limited. T herefore, the reflectance of the rice area
for healthy rice sites is controlled by the water,

influenced by the moist soil surface. M oist soil has

whereas in water - deficient sites, the reflectance is

a higher reflectance than water in the visible, near

72

Fig . 5 T he total monthly rainfall in water- deficient sites
beg inning from the plantation time throug h the
next three months. Rice was transplanted in the
middle of April. S tarting from a rice ag e of two
weeks, there was no rain at all until the next
month. Rainfall in small amounts occurred at a
rice ag e of seven weeks. T he chart also illustrates
the relationship between monthly rainfall and
periods of water stress (see the dashed line). T he
connected line just shows a comparison of the
rice under normal conditions in the same rice ag e
(it does not the actual of rice planting ).

- infrared and middle- infrared regions of the electromagnetic spectrum (H arrison and J upp, 1989).
In the second month, the trend of spectral changes
for the rice became inverted. U nder normal conditions, when the rice enters the second month, the
reflectance values of B and- 1, B and- 2, and B and- 3
should decrease, while in B and- 4, B and- 5, and B and
- 7 they should increase. H owever, at the water -

Fig . 6 Differences of the spectral characteristics of rice
at similar ag es at times when the rice plants in
water deficiency did show sig ns of water stress.
T he third comparisons were done for rice ag es
g reater than one month.

deficient site, the reverse occurs (see F ig. 3 in the
dark area). T his result shows that there has been

two months (67−70 days after plantation time) old

water stress. T he appearance of water stress in the

(F ig. 6). T his is caused by differences in the chloro-

second month in the water - deficient site is support-

phyll and water contents of the rice plants. On rice

ed by the rainfall data obtained from the nearest

aged two months, most of the rice area coverage is

rainfall station from the study area (T able 1).

attributed to rice plants. T he reflectance properties

A lthough the rain did not come until the third week

of the rice will be similar to vegetation in general

after rice planting, signs of water deficiency are

(M artin and H eilman, 1986). A n increase in leaf

only seen in the second month. T he N DV I of rice

reflectivity in the visible light portion as a response

under water deficiency decreased significantly in

to water stress is caused by the metabolic sensitivity

the beginning of the second month compared with

of chlorophyll, resulting in a less efficient absorp-

the rice under normal conditions (F ig. 5).

tion of light (increased reflection) (K nipling, 1970).

B eginning from the second month after transplan-

T he water absorption properties of the middle-

tation, when the rice in the water - deficient site

infrared (T henk abail et al., 1994b) cause the re-

shows signs of water stress, the spectral values at

flectance values of rice under water deficiency in

all L andsat bands are higher than those from the

B and- 5 and B and- 7 of L andsat E T M ＋ to be higher

healthy rice, ex cept for B and- 4 (F ig. 3). T he max i-

than those of healthy rice. H owever, in B and- 4, the

mum difference was seen for rice that was around

chlorophyll pigments that are present in leaves have
73

T able 7

Averag e differences of spectral values and t - tests between rice under water deficiency and healthy conditions
for similar rice ag es.

R ice Condition

R ice age
(days)

cDN V alue
B and- 1

B and- 2

B and- 3

B and- 4

B and- 5

B and- 7

H ealthy conditions (A vg / S td)
W ater deficiency (A vg / S td)
Difference of value
S ig. (2- tailed) t - test

42
44

63.38/ 1.91 42.28/ 1.67 32.39/ 1.63
77.11/ 2.15 56.03/ 1.90 52.38/ 1.93
21.65%
32.52%
61.74%
＜0.01
＜0.01
＜0.01

122.99/ 3.60 65.91/ 2.27 29.35/ 1.93
106.99/ 3.48 89.79/ 2.27 45.37/ 1.93
−13.01%
36.23%
54.56%
＜0.01
＜0.01
0.034

H ealthy conditions (A vg / S td)
W ater deficiency (A vg / S td)
Difference of value
S ig. (2- tailed) t - test

70
67

67.03/ 1.96 46.80/ 1.73 32.39/ 1.63
77.01/ 2.02 51.08/ 1.70 52.79/ 1.92
14.90%
9.14%
62.99%
＜0.01
＜0.01
＜0.01

140.28/ 3.71 68.04/ 2.27 27.57/ 1.93
104.51/ 3.55 81.36/ 2.31 38.50/ 1.95
−25.50%
19.58%
39.67%
＜0.01
＜0.01
＜0.01

H ealthy conditions (A vg / S td)
W ater deficiency (A vg / S td)
Difference of value
S ig. (2- tailed) t - test

77
76

70.67/ 1.96 50.03/ 1.73 42.95/ 1.63
83.01/ 2.19 60.38/ 1.92 59.29/ 1.92
17.46%
20.69%
38.04%
＜0.01
＜0.01
＜0.01

125.15/ 3.71 71.23/ 2.27 25.78/ 1.93
103.71/ 3.60 73.04/ 2.29 34.46/ 1.95
−17.13%
2.55%
33.67%
＜0.01
＜0.01
0.038

T - tests of paired samples used 200 pix el samples. A value less than 0.05 in the T - tests of paired samples indicates
that there is a significant different between the rice under water deficiency and healthy conditions for two- sample
paired tests (two- tailed).

higher levels in healthy rice. In the near - infrared

rice ages provided a significant distinction of spec-

portion, the radiation is scattered by the internal

tra between rice under water deficiency and healthy

spongy mesophyll leaf structure, which leads to

conditions (S ig.＜0.05) in a two- tailed t - test (T able

higher values on the N IR channels (B aret and

7). It means that rice field spectral in water defi-

Guyot, 1991). T herefore, the reflectance of B and- 4

ciency and healthy was different statistically.

in healthy rice is higher than that of water - deficient
rice (F ig. 3).
F or the early detection of water deficiency, the

3. 2 Veg etation index of healthy and water- deficient
rice

visible band (B ands 1- 3) is more sensitive than near -

U sing multiple bands of L andsat E T M ＋ in the

and middle- infrared bands of L andsat E T M ＋. In

form of a vegetation index delivers a clearer distinc-

the visible band, the changing trend of reflectance

tion between rice under healthy and water - deficient

values was detected at the beginning of the third

conditions than when using a single band. F ig. 7

observation (before rice is one month old) ; in near -

shows that in the first month after plant time, the

and middle- infrared, however, it was just detected

vegetation index has a similar pattern due to both

at the fourth observation (rice aged more than one

rice still being under normal conditions. H owever,

month) (F ig. 3). T his means that the visible band is

when the rice plants under water deficiency enter

more sensitive for early water deficiency detection

the second month of growth, when there is a period

than the near - and middle- infrared bands of L and-

of water stress, the difference in the vegetation

sat E T M ＋. T he highest percentage difference of

index for the two rice conditions is high for all

values of L andsat E T M ＋ bands between healthy

vegetation index es evaluated in this study. A ll vege-

rice and water - deficient rice at similar rice ages

tation index values for the water - deficient rice are

was shown by B and- 3 (T able 7).

significantly lower than those for the healthy rice,

T he comparison by means of statistical tests (t -

ex cept for the T V I. T he distinction of T V I trends

test) showed that the entire L andsat E T M ＋ bands

compared with other vegetation index es is because

evaluated in this study (B 1- B 5 and B 7) for similar

the T V I equation is in a reverse form from other

74

Fig . 7 Veg etation index comparison of rice under healthy and water- deficient conditions. E ig ht veg etation indexes
were evaluated, including NDVI, T VI, RVI, DVI, S AVI, IP VI, NDWI- 1, and NDWI- 2. T he white areas in the chart
indicate that both rice are still under normal conditions, while the dark areas show rice plants experiencing
water deficiency.

vegetation index equations (T able 6).

rice under water stress after the third observation

F or three observations of the vegetation index for

(after rice age of 28 days). Other vegetation indices

similar rice ages, N DW I - 1 provides the highest

provided response after the fourth observation

distinction between two rice conditions at rice ages

(after rice age of 35 days). A high distinction of

of 42- 44 days and 67- 70 days, at −71.19% and

vegetation index between rice under healthy condi-

−63.98% , while for rice ages of 76- 77 days the

tions and in water stress was also shown by the

highest difference of a vegetation index value is

statistical test (t - test). T he entire vegetation index

given by DV I with the value of −45.96% (T able 8).

provided a significant difference between two rice

T he highest difference average is provided by

conditions, with significant values in two- tailed t -

N DW I - 1 with the value of −57.36% followed by

tests lower than 0.05 (T able 8). A lthough N DW I - 1

R V I, S A V I and N DV I. H owever, for the early

provided the highest difference average between

detection of water deficiency, the N DV I, R V I, and

two rice conditions compared with other vegetation

S A V I provided better responses. T hey could detect

index , the differences of N DW I - 1 value between
75

T able 8

Averag e differences of veg etation index values and t - tests between rice under healthy and water- deficient
conditions in rice of similar ag es. Neg ative values indicate that the veg etation value of water- deficient rice
is lower than that in healthy rice and vice versa.
N DV I

RV I

IP V I

DV I

TVI

SA V I

N DW I - 1

N DW I - 2

H ealthy conditions 42
0.583
W ater deficiency
44
0.343
−41.17%
Difference of value
＜0.01
S ig. (2- tailed) t - test

3.797
2.042
−46.22%
＜0.01

0.792
0.671
−15.28%
＜0.01

90.602
54.608
−39.73%
＜0.01

96.087
108.938
13.37%
＜0.01

0.872
0.512
−41.28%
＜0.01

0.302
0.087
−71.19%
＜0.01

0.615
0.404
−34.31%
＜0.01

H ealthy conditions 70
0.625
W ater deficiency
67
0.329
−47.36%
Difference of value
＜0.01
S ig. (2- tailed) t - test

4.331
1.98
−54.28%
＜0.01

0.812
0.664
−18.23%
＜0.01

107.896
51.724
−52.06%
＜0.01

94.287
109.842
16.50%
＜0.01

0.935
0.492
−47.38%
＜0.01

0.347
0.125
−63.98%
＜0.01

0.672
0.462
−31.25%
＜0.01

H ealthy conditions 77
0.489
W ater deficiency
76
0.273
−44.17%
Difference of value
＜0.01
S ig. (2 tailed) t - test

2.914
1.749
−39.98%
＜0.01

0.744
0.636
−14.52%
＜0.01

82.201
44.42
−45.96%
＜0.01

100.555
113.775
13.15%
＜0.01

0.731
0.408
−44.19%
＜0.01

0.275
0.174
−36.90%
＜0.01

0.658
0.501
−23.86%
＜0.01

A verage difference
values

−46.83%

−16.01%

−45.92%

14.34%

−44.28%

−56.78%

−29.81%

R ice Condition

A ge

−44.23%

T - tests of paired samples used 200- pix el samples. A value less than 0.05 in the t - test of paired samples indicates that
there is a significant different between the rice under water deficiency and healthy conditions for a two- sample
paired test (two- tailed).

healthy and water deficiency rice is actually still

of rice under water deficiency had significant differ-

smaller than the R V I. T his is caused by differences

ences compared to those of healthy rice. R ice plants

of N DW I - 1 values was already high when both rice

under water deficiency in the study area were

plants are still in good health (F ig. 7). T herefore, the

detected when the rice entered the second month

R V I is the best vegetation index for both early

after planting. W hen water deficiency happened, the

detection of water deficiency and distinguish the

reflectance of the visible bands increased signifi-

two rice conditions.

cantly, while the reflectance of near - and middle-

T he advantages of using a vegetation index

infrared bands decreased. F or the early detection of

compared with a single band is to reduce the spec-

water stress, the visible band was more sensitive

tral data to a single number that is related to the

than the near - and middle- infrared bands. T he red

physical characteristics of vegetation (e.g., leaf

band (B 3) of L andsat E T M ＋ was the best band to

area, biomass, productivity, photosynthetic activity,

distinguish rice plants under water stress and

or percent cover) (B aret and Guyot, 1991 ; H uete,

healthy conditions because it provided the highest

1988) while minimising the effects of internal (e.g.,

difference in the reflectance percentage between

canopy geometry and leaf and soil properties) and

rice under healthy and water - deficient conditions.

ex ternal (e.g., sun- target - sensor angles and atmo-

T he utilisation of the vegetation index to discrimi-

spheric conditions at the time of image acquisition)

nate between rice under healthy and water - deficient

factors on the spectral data (B aret and Guyot, 1991 ;

conditions provided better results than when using a

H uete and W arrick 1990 ; H uete and E scadafal,

single band. T he R V I is the best vegetation index

1991).

for both early detection of water deficiency and
distinguish the two rice conditions. A lthough L and-

４. Conclusions

sat E T M ＋ had S L C - off, it can still be used for rice
monitoring because it has good spatial, spectral, and

Corrected digital number and vegetation indices
76

temporal resolutions, especially for session plants.

Chander, G., M ark ham, B .L ., Dennis L . H elder, D.L .,

Acknowledg ments

2009. S ummary of current radiometric calibration

I would lik e to ex press my profound gratitude to

E O- 1 A L I sensors. R emote Sensing of E nviron-

coefficients for L andsat M S S , T M , E T M ＋, and

my advisor for his invaluable support, encourage-

ment , 113, pp.893- 903.

ment, supervision, and useful suggestions through-

Carter, G.A . and K napp, A .K ., 2001. L eaf optical

out this research work . I am also very thank ful to

properties in higher plants : L ink ing spectral

J S P S R onpak u for their funding support of this P hD

characteristics to stress and chlorophyll concen-

research program.

tration. A merican J ournal of Botany , 88, pp.677-

( 受付日2010. 8. 23，受理日2011. 3. 11)

684.
Casanova, D., E pema, G.F . and Goudriaan, J ., 1998.
M onitoring rice reflectance at field level for

References
A sner, G.P ., 1998. B iophysical

and biological

sources of variability in canopy reflectance.

estimating

biomass

and

L A I.

F ield

C rops

R esearch, 55, pp.83- 92.

R emote Sensing of E nvironment , 64, pp.234- 253.

Chak raborty, M . and P anigrahy, S ., 2000. A P rocess-

B ailey, J .S ., W ang, K ., J ordan, C. and H iggins, A .,

ing and software system for rice crop inventory

2001. U se of precision agriculture technology to

using multi - date R A DA R S A T S canS A R data.

investigate spatial variability in nitrogen yields in

ISPR S J ournal Photogrammetry and emote Sens-

cut grassland. Chemosphere 42, pp.131- 140.

ing , 55, pp.119- 128.

B aret, F . and Guyot, G., 1991. P otentials and limits

Currey, B ., F raser, A .S . and B ardsley, K .L ., 1987.

of vegetation indices for L A I and A P A R assess-

H ow useful is L andsat monitoring. Nature, 328,

ment. R emote Sensing of E nvironment , 35, pp.161

pp.587- 590.

- 173.
B aret, F ., V anderbilt, V .C., S teven, M .D. and J acquemoud, S ., 1994. U se of spectral analogy to

Dozier, J ., 1989. S pectral signature of A lpine snow
cover from the L andsat T hematic M apper,
R emote Sensing of E nvironment , 28, pp.9- 22.

evaluate canopy reflectance sensitivity to leaf

F ang, H .L ., W u, B .F ., L iu, H .Y . and H uang, X ., 1998.

optical properties. R emote Sensing of E nviron-

U sing N OA A A V H R R and L andsat T M to esti-

ment , 48, pp.253- 260.

mate rice area year - by - year. International J our-

B arrett, E .C. and Curtis, L .F ., 1992. Introduction to

nal of R emote Sensing , 19, pp.521- 525.

environmental remote sensing. In : Introduction

F ood Crops A griculture Department, 2006. A nnual

to E nvironmental R emote Sensing , 3, p.426 (B ar-

report of food crops, 3 pp.125- 130 (Department

rett, E .C. and Curtis, L .F ., E ds.), Chapman & H all,

A griculture of L ocal Government, B ali P rovince

L ondon, U .K .

Indonesia).

B ouman, B .A .M ., 1995. Crop modeling and remote
sensing for yield prediction. Netherlands J ournal
of A gricultural Science, 43, pp.143- 161.

Gausman, H .W ., 1974. L eaf reflectance of near infrared.

Photogrammetric

E ngineering

and

R emote Sensing , 40, pp.183- 191.

F ood Crops A griculture Department, 2006. A nnual

H all, A ., L amb D.W ., H olzapfel, B . and L ouis J .,

report of food crops. Department A griculture of

2002. Optical remote sensing applications for

L ocal Government, B ali P rovince Indonesia, 3,

viticulture−a review. A ustralian

125- 130.

Grape and Wine R essearch, 8, pp.36- 47.

J ournal

of

B M G, 2006. R eport of Monthly R ainfall . M eteoro-

H arrison, B .A . and J upp, D.L .B ., 1989. Introduction

logical and Geophysical A gency, 2, pp.45- 50 (B ali

to remotely sensed data, 1, p.65 (CS IR O P ublica-

P rovince, Indonesia).

tions).
77

H uete, A .R . and E scadafal, R ., 1991. A ssessment of
biophysical soil properties through spectral decomposition techniques. R emote Sensing of E nvi-

735 nm to assess nitrogen status of rice canopy.
A gronomy J ournal , 100, pp.205- 212.
L illesand, T .M . and K iefer, R .W ., 1994. R emote
Sensing and Image Interpretation , 3, pp.427- 517

ronment , 35, pp.149- 159.
H uete, A .R . and W arrick , A .W ., 1990. A ssessment of

(N ew Y ork : J ohn W iley and S ons).

vegetation and soil water regimes in partial can-

M artin, R .D.J . and H eilman, J .L ., 1986. S pectral

opies with optical remotely sensed data. R emote

reflectance patterns of flooded rice. Photogram-

Sensing of E nvironment , 32, pp.115- 167.

metric E ngineering and R emote Sensing , 52, pp.

H uete, A .R ., 1988. A soil - adjusted vegetation index
(S A V I). R emote Sensing of E nvironment , 25, pp.
295- 309.
IR R I,

1993. 1993−1995 IR R I

1885- 1890.
M yneni, R .B ., R oss, J . and A srar, G., 1989. A review
on the theory of photon transport in leaf canopies.

R ice

A lmanac .

M anila : International R ice R esearch Institute.

A gricultural and F orest Meteorology , 45, pp.1- 153.
N augle, B .I. and L ashlee, J .D., 1992. A lleviating

J acquemoud, S ., 1993. Inversion of the P R OS -

topographic influences on land- cover classifica-

P E CT ＋S A IL canopy reflectance model from

tions for mobility and combat modeling, Photo-

A V IR IS equivalent spectra : theorectical study.

grammetric E ngineering and R emote Sensing , 58,

R emote Sensing of E nvironment , 44, pp.281- 292.

pp.1217- 1221.

J acquemoud, S ., B aret, F . and H anocq, J .F ., 1992.

N iel, T .G.V . and M cV icar, T .R ., 2001. R emote S ensing

ce. R emote Sensing of E nvironment , 41, pp.123-

R eview. A vailable on http:/ / www.clw.csiro.au/

132.

publ i c a t i ons / c onsul t a nc y / 2001 / C R C - R i c e -

K hush, G.S . 2005. W hat it will tak e to feed 5 billion
rice consumers in 2030. Plant Molecular Biology ,
59, pp.1- 6.
K nipling, E .B ., 1970. P hysical and physiological
basis for the reflectance of visible and near - infrared radiation from vegetation. R emote Sensing of
E nvironment , 1, pp.155- 159.

of

R ice- B ased Irrigated A griculture : A

M odeling spectral and bidirectional soil reflectan-

T R P 11050101.pdf (A ccessed 15 December 2009.)
N uarsa, I.W . and N ishio, F ., 2007. R elationships
between rice growth parameters and remote sensing data. J ournal of R emote Sensing and E arth
Sciences, 4, pp.102- 112.
N uarsa, I.W ., K anno, S ., S ugimori, Y . and N ishio,
F ., 2005. S pectral characterization of rice field

K ok sal, E .S ., K ara, T ., A pan, M ., ̈ U stun, H . and

using multi - temporal L andsat E T M ＋ data. J our-

I lbeyi, A ., 2008. E stimation of green bean yield,

nal of R emote Sensing and E arth Science. 2, pp.65

water deficiency and productivity using spectral

- 71.

index es during the growing season. Irrigation and
Drainage Systems, 22, pp.209- 223.
K upiec, J .A . and Curran, P .J ., 1995. Decoupling
effects of the canopy and foliar biochemicals in
A V IR IS spectra. International J ournal of R emote
Sensing , 16, pp.1731- 1379.

Ok amoto, K . and K awashima, H ., 1999. E stimation
of rice - planted area in the tropical zone using a
combination of optical and microwave satellite
sensor data. International J ournal of

R emote

Sensing , 20, pp.1045- 1048.
P ierce, F .J ., N owak , P . and R oberts, P .C., 1999.

K uroso, T ., F ujita, M . and Chiba, K ., 1997. M onitor-

A spects of P recision A griculture, p.1- 84. In D.L .

ing of rice fields using multi - temporal E R S - 1 C -

Sparks (ed.) A dvances in A gronomy , 67, A cademic

band S A R data. International J ournal of R emote

P ress, S an Diego, CA , U S A .

Sensing , 18, pp.2953- 2965.

P ons, X . and S ole- S ugranes, L ., 1994. A simple

L ee, Y .- J ., Y ang, C.- M ., Chang, K .- W . and S hen Y .,

radiometric correction model to improve auto-

2008. A simple spectral index using reflectance of

matic mapping of vegetation from multispectral

78

satellite data. R emote Sensing of E nvironment ,

S pectroscopic determination of wheat water sta-

48, pp.191- 204.

tus using 1650- 1850 nm spectral absorption fea-

Qin, Z. and Zhang, M ., 2005. Detection of rice sheath
blight for inseason disease management using

tures, International J ournal of R emote Sensing ,
22, pp.2329- 2338.

multispectral remote sensing. International J our-

Y ang, C.- M . and S u, M .R ., 2000. A nalysis of spectr

nal of A pplied E arth Observation and Geoinfor-

al character istics of r ice canopy under water

mation, 7, pp.115- 128.

deficiency : M onitor ing changes of spectral char-

R ibbes, F . and T oan, L .T ., 1999. R ice field mapping
and monitoring using R A DA R S A T data. Interna-

acteristics of dehydrating rice canopy. A sian
Conference on R emote Sensing , 21, pp.13- 18.

tional J ournal of R emote Sensing , 20, pp.745- 765.

Y ang, C.- M ., and Chen, R .- K ., 2004. M odeling rice

S hao, Y ., F an, X ., L iu, H ., X iao, J ., R oss, S ., B risco,

growth using hyperspectral reflectance data. Crop

B ., B rown, R . and S taples, G., 2001. R ice monitor-

Science, 44, pp.1283- 1290.

ing and production estimation using multitempor-

Y ang, C.- M ., and Cheng, C.- H ., 2001. S pectral char-

al R A DA R S A T . R emote Sensing of E nvironment ,

acteristics of rice plants infested by brown planth-

76, pp.310- 325.

opper. Proceedings of the National Science Coun-

T henk abail, P .S ., S mith, R .B . and P auw, D.E ., 2000.

cil , 25, pp.180- 186.

H yperspectral vegetation indices and their rela-

Y ang, C.- M ., Cheng, C.- H . and Chen, R .- K ., 2007.

tionships with agricultural crop characteristics.

Changes in spectral characteristics of rice canopy

R emote Sensing of E nvironment , 71, pp.158- 182.

infested by leaf folder and brown planthopper.

T henk abail, P .S ., W ard, A .D., L yon, J .G. and M erry,

Crop Science, 47, pp.329- 335.

C.J ., 1994b. T hematic M apper vegetation indices

Y ang, C.- M ., L iu, C.C., and W ang, Y .W ., 2008. U sing

for determining soybean and corn growth parame-

F ormosat - 2 S atellite Data to E stimate L eaf A rea

ters. Photogrammetric E ngineering and R emote

Index of R ice Crop. J ournal of Photogrammetry

Sensing , 60, pp.437- 442.

and R emote Sensing , 13, pp.253- 260.

T ong, Q.T .Q., P u, R ., Guo, X ., and Zhao, C., 2001.

79