Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

ATMOSPHERIC EFFECTS ON VEGETATION INDICES OF TM AND ETM+ IMAGES

FROM A TROPICAL REGION USING THE 6S MODEL


Mauro Antonio Homem Antunes1, Jos Marinaldo Gleriani2, Paula Debiasi1
1

Engineering Department, Federal Rural University of Rio de Janeiro, IT/DE, Br 465 km 7, Seropdica,
RJ, Brazil, <mauroantunes@ufrrj.br>, <paula@ufrrj.br>

Forestry Engineering Department, Federal University of Viosa, Avenida P. H. Rolfs s/n, Viosa, MG,
Brazil, <gleriani@gmail.com>
ABSTRACT

Vegetation indices have been widely used for monitoring


vegetation and to estimate vegetation biophysical
parameters. The atmosphere affects differently sensor bands
depending on the waveband, thus affecting the indices
obtained from top of the atmosphere reflectances. The
objective of this paper is to evaluate the atmospheric effects
on the NDVI and SAVI using the 6S model adapted for
atmospheric correction of images of eight bits digital
numbers. Images from the TM and ETM+ sensors of the
same area were corrected for the atmosphere and the
vegetation indices were compared with those from noncorrected images. For these images the results show that the
vegetation indices at the top of the atmosphere were lower
than those from the surface. This underestimation of the
vegetation indices due to the atmosphere is attributed to a
higher absorption of radiation in the near infrared than in the
red spectral region. This is exceptionally pronounced on
tropical atmospheres where absorption by water vapor tends
to be higher than in mid-latitudes. It is concluded that
atmospheric correction is necessary when calculating
vegetation indices over tropical regions and the adapted 6S
model is suitable for this task.
Index Terms Vegetation indices, atmospheric
correction, 6S model
1. INTRODUCTION
Vegetation indices are band transformations that have been
largely used for vegetation assessment and monitoring.
Indices were introduced to maximize the vegetation signal
while decreasing the background effects. The most used
vegetation indices are the Normalized Difference Vegetation
Index-NDVI [1] and the Soil Adjusted Vegetation IndexSAVI [2]. It has been demonstrated from model simulations
that these indices are affected by the atmosphere [3].
Vegetation indices with reduced atmospheric effects have
been developed for the MODIS sensor, e.g. SARVI [4] and

978-1-4673-1159-5/12/$31.00 2012 IEEE

EVI [5]. However, the other indices like NDVI and SAVI
continue to be used without atmospheric corrections, mainly
for sensors other than MODIS.
In order to retrieve reliable information from vegetation
indices it is necessary to use ground reflectances instead of
apparent reflectances to calculate the indices. One
methodology that has been largely used for atmospheric
correction is the dark pixel subtraction [6]. This
methodology has been found the least accurate for
atmospheric correction [7]. However, the dark pixel
subtraction is not appropriate for tropical atmospheres due
to high differences in atmospheric effects between red and
near infrared bands due to absorption by water vapor. The
6S model [8] was developed and implemented for
simulating the radiance reaching a sensor for a target with
known surface reflectance and as a function of illumination,
viewing and atmospheric conditions. The model has the
possibility of calculating the atmospheric correction based
on the radiance level at the sensor and the defined
illumination/viewing and atmospheric conditions. An
adaptation has been implemented in the model so that it can
be used for atmospheric corrections of images in a pixel by
pixel basis [9]. The adapted model calculates the
atmospheric correction, reads the image, calculates the
surface reflectance for each pixel and stores it in an image
scaled between 0 and 255 for an 8 bits image.
The objective of this paper is to evaluate the effects of
the atmosphere on the NDVI and SAVI using Landsat TM
and ETM+ images from a tropical region corrected with the
6S model versus the non-corrected images. For comparison
purposes the Simple Ratio Vegetation Index-SRVI was also
evaluated in this study.
2. DATA AND METHODOLOGY
Landsat images from the TM and ETM+ sensors from five
dates and path/row 217/76 were used in this research. The
images were from October 5, 2007 (TM), June 22, 2004
(TM), December 31, 2004 (TM), July 8, 2001 (ETM+) and
December 12, 2000 (ETM+).

6549

IGARSS 2012

The images were registered using control points


extracted from 1:10,000 scale maps with residual values
below one pixel and imported to the SPRING software [10].
Bands 3 (red) and 4 (near infrared) were then exported as
binary images in raw format and the atmospheric correction
was carried out using the 6S model. The used parameter
conditions for the atmospheric correction are listed on Table
1 below.

calculated from apparent reflectances and from ground


reflectances using the following equations:

Table 1. Input parameters for the atmospheric corrections


used for all TM and ETM+ images.
All images:
- Atmospheric model:
Tropical
- Aerosol model:
Continental
- Longitude:
-43.689800
- Latitude:
-22.764883
- Viewing angle:
0
- Target altitude:
16 meters above sea level
TM image from October 5, 2007:
- Sun-Earth distance:
0.99996 AU*
- Time:
12.74611 UT**
- Solar zenith angle:
50.85
- Horizontal visibility:
12 km
TM image from June 22, 2004:
- Sun-Earth distance:
1.03281 AU
- Time:
12.55889 UT
- Solar zenith angle:
35.45
- Horizontal visibility:
15 km
TM image from December 31, 2004:
- Sun-Earth distance:
0.96691 AU
- Time:
12.63028 UT
- Solar zenith angle:
30.47
- Horizontal visibility:
18 km
ETM+ image from December 12, 2000:
- Sun-Earth distance:
0.96942 AU
- Time:
12.7044 UT
- Solar zenith angle:
29.11
- Horizontal visibility:
20 km
ETM+ image from July 8, 2001
- Sun-Earth distance:
1.03371 AU
- Time:
12.6875 UT
- Solar zenith angle:
56.40
- Horizontal visibility:
17 km
* Astronomical Units
** Universal Time

Where R and NIR are respectively the red and near


infrared reflectances (bands 3 and 4) and L is the canopy
background adjustment factor, assumed to be 0.5 [2]. The
Simple Ratio Vegetation Index (SRVI) was also calculated
for comparison purposes. The SRVI is calculated as:

The visibility was estimated by visual analysis of the


RGB combinations of all bands for all images as aerosol
profiles or surface visual estimates for those dates were not
available. This is a rather simplification but it represents
well most of the real conditions in remote sensing where
these data have to be estimated on an image basis.
Atmospherically corrected images were uploaded in the
SPRING software and the vegetation indices were

(1)
(2)

(3)

The indices were clipped to the coordinates from 43


48 1.31W to 43 36 54.15W and from 22 41 40.75S
to 22 56 45.46S. The average altitude of the area was
obtained from SRTM DTM on the WGS84 reference
ellipsoid and transformed to altitude above the sea level.
Figure 1 shows a TM image of the area using the RGB 453
band combination to give an appreciation of the study area.
Green vegetation appears in red, pasture and grasses appear
in in green, bare soil and urban areas in blue and clear water
in black.
Scatterplots were made using the vegetation indices
from the top of the atmosphere (apparent) reflectance and
ground reflectance to compare the results.
3. RESULTS AND DISCUSSION
The area used in this research has a large variety of land
uses such as urban areas, water, industrial areas, tropical
forests, planted forests, bare soil, exposed rocks, pastures
and ponds filled with water with sediments from sand
extraction. This allowed a large variation in the values of the
indices, as can be seen in Figures 2 and 3. The TM image
from October 5, 2007 was chosen to present the results
graphically because it is from a date close to spring
Equinox, while the other images represent extreme
situations in terms of solar zenith angle and Sun-Earth
distance conditions.
Figures 2 and 3 show that the atmosphere causes a
decrease in both vegetation indices and the atmospheric
correction causes an increase in the indices. The SAVI
presented a more confined scatterplot alongside the 1:1 line
than the NDVI. The highest atmospheric effects were
present in the SRVI (Figure 4). It highlights the differences
in atmospheric effects on red and near infrared bands,
although for the date shown these differences were
decreased due to a low visibility compared to the other
images. In a tropical atmosphere the near infrared
absorption by water tends to be more important than the
scattering by aerosols and gases. As a consequence in the

6550

near infrared the corrected image tends to have the


reflectance increased from satellite level to ground level. In
the red waveband the absorption by water is compensated
by the scattering from aerosols and gases with an intensity
that depends on the aerosols concentration.

vegetation indices that use these two bands when comparing


vegetation indices from ground and top of the atmosphere
reflectances. This can be observed by comparing Figure 4
with Figure 5, where the higher horizontal visibility as
compared to the image from October 5 led to higher
differences in top of the atmosphere versus ground SRVI.
It can be seen from these results from image data that the
atmospheric correction enhances the vegetation indices and
the degree of enhancement will depend on the atmospheric
aerosol concentration given by the visibility, as the used
gaseous atmosphere as the same for all images.

Fig. 2. Scatterplot of NDVI from ground reflectance (y axis)


versus NDVI from top of the atmosphere reflectance (x
axis) for the TM image from October 5, 2007.
Fig. 1. Study are in a RGB 453 combination for the TM
image of October 5, 2007. The image is from ground
reflectance to which a linear stretch was applied.
The atmospheric effects on vegetation indices were
different between the images from different dates, although
the patter was similar in all images. Differences in
vegetation indices from ground reflectance were due to
differences in surface spectral characteristics (e.g., dry
season versus wet season) and differences in illumination
conditions (solar zenith angle). However, the differences in
vegetation indices between ground and top of the
atmosphere reflectances varied among the different dates,
because of differences in aerosol concentration as given by
the horizontal visibility.
The Mie scattering due to suspended aerosols in the
atmosphere tend to decrease the differences in atmospheric
effects between red and near infrared. On the other hand, for
clear atmospheres the effect of water absorption causes a
large difference in atmospheric effects between red and near
infrared, leading to enhanced differences between

Fig. 3. Scatterplot of SAVI from ground reflectance (y axis)


versus SAVI from top of the atmosphere reflectance (x axis)
for the TM image from October 5, 2007.

6551

decrease of the differences between top of the atmosphere


and ground SRVI when horizontal visibility increased.
It is concluded that ground reflectance from
atmospherically corrected images should be used instead of
apparent reflectance at satellite level in order to obtain more
realistic values of vegetation indices.
The influence of the atmospheric effects on the
relationships between vegetation indices and vegetation
status, e.g. LAI and biomass, remain to be investigated
using atmospherically corrected images, ground measured
reflectances and observed vegetation biophysical
parameters. It is also recommended to investigate the effects
of aerosols and gaseous atmosphere on other vegetation
indices like SARVI and EVI and the consequences for
surface parameters estimation.
11. REFERENCES

Fig. 4. Scatterplot of SRVI from ground reflectance (y axis)


versus SRVI from top of the atmosphere reflectance (x axis)
for the TM image from October 5, 2007.

[1] Rouse, J.W., R.H. Haas, J.A. Schell, D.W. Deering and J.C.
Harlan, Monitoring the vernal advancement of retrogradation
(greenwave effect) of natural vegetation, NASA/GSFC, Type III,
Final Report, Greenbelt, MD. 371 p. 1974.
[2] Huete, A.R., A soil adjusted vegetation index (SAVI), Remote
Sensing of Environment, Vol. 25, pp. 295-309, 1988.
[3] Myneni, R.B and G. Asrar, Atmospheric effects and spectral
vegetation indices, Remote Sensing of Environment, Vol. 47, pp.
390-402, 1994.
[4] Kaufman, Y. J., and D. Tanr, Atmospherically resistant
vegetation index (ARVI) for EOS-MODIS, IEEE Transactions on
Geoscience and Remote Sensing, Vol. 30, pp. 261270, 1992.
[5] Huete, A., K. Didan, T. Miura, E.P. Rodriguez, X. Gao and
L.G. Ferreira, Overview of the radiometric and biophysical
performance of the MODIS vegetation indices, Remote Sensing of
Environment, Vol. 83, pp. 195213, 2002.
[6] Chavez, P.S., An improved dark-object subtraction technique
for atmospheric scattering correction of multispectral data, Remote
Sensing of Environment, Vol. 24, pp. 459-479, 1988.

Fig. 5. Scatterplot of SRVI from ground reflectance (y axis)


versus SRVI from top of the atmosphere reflectance (x axis)
for the TM image from December 31, 2004.
4. CONCLUSIONS
The effects of the atmosphere on the NDVI and SAVI
vegetation indices were investigated using Landsat TM and
ETM+ images and the 6S model. Atmospheric corrections
increased both the NDVI and SAVI. The atmosphere affects
differently the red and near infrared bands. For a tropical
atmosphere the radiation absorption in the near infrared
decreases the signal reaching the sensor more than in the red
band. For the same gaseous atmosphere the aerosols
concentration plays an important role as can be seen in the

[7] Moran, M.S., Jackson, R.D., Slater, P.N., Teillet, P.M.,


Evaluation of Simplified Procedures for Retrieval of Land Surface
Reflectance Factors from Satellite Sensor Output, Remote Sensing
of Environment, Vol. 41:169-184, 1992.
[8] Vermote, E.F., D. Tanr, J.L. Deuze, M. Herman and J.J.
Morcrette, Second simulation of the satellite signal in the solar
spectrum, 6S: An overview., IEEE Transactions on Geoscience and
Remote Sensing, Vol. 35, pp. 675-686, 1997.
[9] Antunes, M.A.H., Atmospheric correction of remotely sensed
images using the 6S model: An evaluation with Landsat images,
IGARSS 2012, Munich, July, 2012.
[10] Cmara, G., R.C.M. Souza, U.M. Freitas and J. Garrido,
SPRING: Integrating remote sensing and GIS by object-oriented
data modeling, Computers & Graphics, Vol. 20, pp. 395-403,
1996.

6552

You might also like