International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan 2010
LAND SURFACE EMISSIVITY ESTIMATION FROM VNIR/SWIR REFLECTANCE
Masao MORIYAMA
Faculty of engineering, Nagasaki University
1–33 Bunkyo, Nagasaki 852–8521, JAPAN
matsu@rsirc.cis.nagasaki-u.ac.jp
Commission VIII/8
KEY WORDS: MODIS, Land surface temperature/emissivity, NDVI/NDWI
ABSTRACT:
For the Land surface estimation from the satellite on the basis of the split-window method, the land surface emissivity have to be
the known value. To estimate these value from the VNIR/SWIR land surface reflectance, the attempts are made that the emissivity
category are defined from the definition of the average emissivity and the spectral dependency of the emissivity, the number of pixels of
each category are counted and the statistical analysis between the emissivity of 10.8 and 12.0 micrometer and VNIR/SWIR reflectance
are made. Using 10 years of MODIS global mapped reflectance product (MOD09CFG) and Land surface temperature and emissivity
product (MOD11C1), these analysis are made and verified.
1
INTRODUCTION
The operational satellite based land surface temperature (LST)
estimation started around 2000 which are made by MODIS, ASTER
and AATSR. For the LST estimation, the split window algorithm
are mainly used for AATSR (Prata, 2002) and MODIS (Wan,
1999), which the land surface emissivity must be the known value.
The upcoming JAXA’s sensor SGLI (Second generation GLobal
Imager) (Honda, 2006) on board GCOM–C1 will have the similar spectral channel as AVHRR, MODIS and AATSR (10.8 and
12.0 [μm]) and the split windows algorithm are planned to use
for the LST estimation (Moriyama, 2009). For the land surface
emissivity definition, so far the land cover classification based
method are used (Wan, 1999, Prata, 2002), which provide the
emissivity set for the conventional land cover categories such as
forest, bare soil area and so on. This concept has assumptions
that the emissivity of each category are always the same and the
each pixel can be classified into the single land cover category.
Actually, these assumptions can be easily violated by the category mixture and water content fluctuation. This paper proposes
the reflectance based land surface emissivity estimation method
which are on the basis of the statistical analysis between the emissivity from MODIS Day/Night land surface temperature product
and the index derived from the MODIS land surface reflectance
product.
2
SPATIAL AND TEMPORAL DISTRIBUTION OF THE
LAND SURFACE EMISSIVITY
verify the classification based land surface emissivity by the radiometric consistency analysis, the emissivity represents the actual condition of the surface. Also as the surface reflectance, the
daily global mapped surface reflectance product (MOD09CMG)
is used. This product is generate from the atmospheric correction
process using the satellite detected radiance and the MODIS derived atmospheric data. Both of the dataset has the same spatial
resolution, 0.05[deg.]. These 2 data set from 5 Mar. 2000 to 31
Dec. 2009 are used for the analysis. From the LST product, the
pixel of the daytime LST Quality flag is used for the mask data
for the both product. As the example of the data, 1 Jan. 2001
mask data, ch. 31 emissivity and ch. 32 emissivity are shown in
Figure 1–3.
Figure 1: MODIS ch. 31 emissivity on 1 Jan. 2001
2.1 Dataset
Terra/MODIS continues to observe the earth for 10 years and
many kinds of the product are generated. Among the products,
this research uses the daily global mapped Day/Night land surface temperature product (MOD11C1) is used for the spatial and
temporal emissivity distribution analysis. This product is based
on the Day/Night land surface temperature estimation algorithm
which the simultaneous radiative transfer equation of the day and
night observation radiance of 6 emission spectral channel data
at the same location are used for the day and night land surface temperature estimation and the classification based land surface emissivity validation (Wan, 1999). Since this algorithm can
Figure 2: MODIS ch. 32 emissivity on 1 Jan. 2001
889
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan 2010
1
# of pixels
Relative accumlated frequency
1e+09
1e+08
0.8
# of pixels
1e+07
1e+06
0.6
100000
10000
0.4
1000
100
Figure 3: The mask image on 1 Jan. 2001
0.2
Relative accumlated frequency
1e+10
10
1
0
20
40
60
80
100
120
0
140
Emissivity category
2.2 Emissivity category
Figure 5: Total pixel of the emissivity category (All category)
1e+10
1
# of pixels
Relative accumlated frequency
1e+09
1e+08
0.8
ε̄ =
ε 1 + ε2
2
Δε =
# of pixels
1e+07
(1)
ε1 − ε2
2
1e+06
0.6
100000
10000
0.4
1000
(2)
100
0.2
Relative accumlated frequency
The land surface emissivity widely varies with the land cover and
the wavelength. For the rough estimation of the emissivity variation, the emissivity category is defined. To separate the spectral
dependency and the magnitude of the emissivity, the following
two parameters, the average emissivity ε̄ and the spectral dependency parameter a are defined as Eq. (1 – 3),
10
a=
Δε
1 − ε̄
1
(3)
0.95
66
55
45
36
28
21
15
10
6
3
1
0
79
67
56
46
37
29
22
16
11
7
4
2
122 107 93
80
68
57
47
38
30
23
17
12
8
123 108 94
81
69
58
48
39
31
24
18
13
9
19
14
32
25
82
70
59
49
40
83
71
60
50
41
33
26
20
61
51
42
34
27
84
72
127 112 98
85
73
62
52
43
35
128 113 99
86
74
63
53
44
129 114 100 87
75
64
54
130 115 101 88
76
65
131 116 102 89
77
126 111 97
0.85
0.8
20
1e+07
1e+06
100000
132 117 103 90
0.75
133 118 104
134 119
0.7
0.65
0.65
15
The variation of the pixel number of each category are shown in
Figure 7 and 8. It is clarified that the 0th. and 1st. category
varied in the inverse phase, the 2nd. and 4th. category varied in
the similar phase and the 7th. category kept constant. Also it is
clarified that this 5 emissivity categories are significant.
5
125 110 96
124 109 95
0.9
10
Figure 6: Total pixel of the emissivity category (0–20th. category)
Number of pixels
MODIS ch. 32 emissivity
78
121 106 92
120 105 91
5
Emissivity category
where ε1 and ε2 are emissivity at MODIS ch. 31 (11[μm]) and
ch. 32 (12[μm]) respectively. To make the 0.02 interval grid on
ε1 – ε2 plane, ε̄ and a are divided and the emissivity category is
defined as Figure 4.
1
0
0
135
0.7
0.75
0.8
0.85
0.9
0.95
1
10000
1000
100
MODIS ch. 31 emissivity
Figure 4: The emissivity category
10
1
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
2.3 Temporal and spatial distribution of the emissivity category
Year
0
From 10 years daily dataset, the total numbers of pixel of each
emissivity category are counted. The result of all category is
shown in Figure 5 and the result of 0 – 20th. category is shown in
Figure 6. The green dashed line in the both figure show the relative cumulative frequency, up to 7th. emissivity category, almost
all land cover are included.
1
2
3
4
5
6
Figure 7: Daily variation of the pixel number of the emissivity
category (0–6th. category)
890
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan 2010
1e+06
Number of pixels
100000
10000
1000
100
10
1
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Year
7
8
9
10
11
12
Figure 8: Daily variation of the pixel number of the emissivity
category (7–12th. category)
To identify the emissivity categories, the average and standard
deviation of LST of each emissivity category are computed, The
results are shown in Figure 9 and 10. Al so the 0 - 2nd. categories
and 3rd, 4th. and 7th. categories of 1 Jan. 2001 and 1 July 2001
are shown in Figure 11 and 12.
Figure 11: Emissivity category map of 1 Jan. 2001 (Upper: Red:
2, Green: 1, Blue: 0, Lower: Red: 7, Green: 4, Blue: 3)
330
320
Average LST [K]’
310
300
290
280
270
260
250
240
0
20
40
60
80
100
120
140
Emissivity category
Figure 9: The average and standard deviation of LST at each
category (All category)
330
320
Average LST [K]’
310
Figure 12: Emissivity category map of 1 July 2001 (Upper: Red:
2, Green: 1, Blue: 0, Lower: Red: 7, Green: 4, Blue: 3)
300
290
280
From the above analysis, it is clarified that the 0th. emissivity
category corresponds to the snow/ice area and 1st. emissivity
category corresponds to the vegetation. Also it is clarified the
most significant emissivity change is the alternation of the 0th.
and 1st emissivity category in the annual land cover change.
270
260
250
240
0
5
10
15
20
2.4 Land surface estimation from the VNIR/SWIR reflectance
Emissivity category
The surface emissivity is affected by not only the land cover also
the water content (J. W. Salisbery, 1992). To identify these condition, many kinds of indexes derived from the satellite data are
proposed. Among these, this research use NDVI (Normalized
Differential Vegetation Index) (de Griend and Owe, 1993) and
Figure 10: The average and standard deviation of LST at each
category (0 – 20th. category)
891
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan 2010
NDWI (Normalized Differential Water Index) (Gao, 1996) are
used.
This study estimates the average emissivity ε̄ (cf. Eq. (1)) and the
spectral dependency parameter a (cf. Eq. (3)) using NDVI and
NDWI derived from MOD09CMG product as follows,
N DV I =
ρ 2 − ρ1
ρ2 + ρ1
(4)
N DW I =
ρ2 − ρ 6
ρ2 + ρ6
(5)
where ρ means the surface reflectance and subscript shows the
MODIS spectral channel.
At first the degree of variation of ε̄ and a are computed from the
10 years dataset. The standard deviation of these variables are
shown in Figure 13.
Figure 14: Correlation coefficient between a and NDVI (Upper),
NDWI (Lower)
3
CONCLUSIONS
The previous discussions lead the following results. The land surface emissivity largely varies in the high latitude area because of
the alternation of ice/snow and vegetation. To estimated the emissivity, NDWI is effective, especially the wavelength dependency
of the emissivity estimation.
ACKNOWLEDGMENTS
This research was supported by JAXA/GCOM–C project.
REFERENCES
de Griend, A. A. V. and Owe, M., 1993. On the relationship between thermal emissivity and the normalized difference vegatation index for the natural surface. International Journal of Remote
Sensing 44(8), pp. 1119–1131.
Figure 13: Standard deviation of ε̄ (Upper) and a (Lower)
As the previous the emissivity category research, ε̄ does not so
much varied, but a varied especially in the high latitude area.
This is because the category 0 (ice/snow) and 1 (vegetation) is
the dominant category, the alternation of these 2 categories is the
significant emissivity change and these 2 categories have similar
emissivity. So that in this research, a is defined as the estimated
variable.
Gao, B. C., 1996. NDWI – A normalized difference water index
for remote sensing of vegetation liquid water from space. Remote
Sensing of Environment 58(2), pp. 257–266.
Honda, Y., 2006. The possibility of SGLI/GCOM-C for global
environment change monitoring. In: Proc. SPIE.
J. W. Salisbery, D. M. D., 1992. Emissivity of terrestrial materials in the 8 – 14 μm atmospheric window. Remote Sensing of
Environment 42(2), pp. 83–106.
To estimate a, the pixel–wise estimation is used, this means the a
estimation formula is made at each pixel. For the definition of the
independent variable of the estimation, the correlation coefficient
between a and NDVI, NDWI are computed, the results are shown
in Figure 14. The correlation between a and NDWI is high than
that of NDVI, especially the high latitude area. This shows that
the most significant emissivity change in the high latitude area
can estimate the NDWI.
Moriyama, M., 2009. Investigation of the split window algorithm
for the land surface temperature estimation. In: Proc. Fall Conf.
of Japanese Photogrammeetry and Remote Sensing, pp. 65–68.
Prata, F., 2002. Land surface temperature measurement from
space: AATSR Algorithm Theoretical Basis Document. Technical report, CSIRO.
Wan, Z., 1999. MODIS Land–Surface Temperature Algorithm
Theoretical Basis Document. NAS5–31370, NASA.
892