Вы находитесь на странице: 1из 6

2008 International Workshop on Earth Observation and Remote Sensing Applications

Evaluation of Atmospheric Correction Using


FLAASH
Jinguo Yuan a,b,c Zheng Niu b
a. b
College of Resource and Environmental Sciences, Hebei State Key Laboratory of Remote Sensing Science,
Normal University, Shijiazhuang, China Institute of Remote Sensing Applications, Chinese
b
State Key Laboratory of Remote Sensing Science, Academy of Sciences, Beijing, China
Institute of Remote Sensing Applications, Chinese E-mail: zheng_niu@263.net
Academy of Sciences, Beijing, China
c.
Hebei Key Laboratory of Environmental Change and
ecological Construction. Shijiazhuang, China
E-mail: yuanjinguo8@163.com

invariant-object methods, histogram matching methods and


Abstract—This paper evaluated the capability of FLAASH in empirically linear correction [4]. Empirically statistical method
ENVI software to make atmospheric correction for Hyperion is simple and easy, but it can not obtain true object radiation
hyperspectral image and ALI image. Hyperion and ALI sensors information. Radiation transfer model has higher accuracy, but
are two of the three instruments onboard NASA EO-1 satellite, it is restricted to real-time atmospheric parameters acquisition,
New Millennium Program (NMP) launched on November 21, having characteristic of many parameters, large amount of
2000, and Hyperion is the first spaceborne hyperspectral imaging computation and difficult application. Many researchers have
spectrometer. The study area is Zhangye city(37˚28′N-39˚57′N, made studies on atmospheric correction methods. FLAASH
97˚20′E-102˚12′E ) in Heihe River valley of Gansu province, was used to make atmospheric correction of AVIRIS and
China. Using TM data with UTM projection, Hyperion ASTER images and the effects were evaluated [5.6]. 6S model
hyperspectral data acquired on September 10, 2007 and ALI and empirically linear method were used to make atmospheric
data on September 20, 2007 were geometrically and correction of Hyperion image [7, 8]. Based on the application
radiometrically corrected, and then atmospherically corrected of atmospheric radiation transfer model to make atmospheric
using FLAASH. Surface reflectance spectra of corn, water body, simulation of PHI hyperspectal data, Li et al. (2006) developed
desert and buildings were extracted from these two images and effective atmospheric correction and reflectance inversion
made comparison with apparent reflectance. Canopy reflectance
algorithm, and analyzed the effects of cross-radiation [9]. Long
spectra of corn were recorded using ASD Fieldspec
et al. (2002) used Rahman bidirectional reflectance model and
spectroradiometer in near-real time to coincide with EO-1
satellite sensor overpass. According to filter function of ALI and BRDF based atmospheric correction method to iteratively
central wavelength and Gaussian filter function based on full extract multi-angle atmospheric corrected NOAA images [10].
width at half maximum (FWHM) of Hyperion, the ASD The Hyperion and ALI sensors are two of three instruments
reflectance spectra were resampled to corresponding Hyperion onboard National Aeronautics and Space Administration
and ALI bands. Results showed that resampled ASD spectra of (NASA) Earth Observing (EO-1) satellite, New Millennium
corn were consistent with spectra on Hyperion and ALI images Program (NMP) launched on November 21, 2000, with sun-
after FLAASH. This demonstrated the effectiveness of
synchronous orbit at 705km altitude. Hyperion is the first
atmospheric correction using FLAASH.
spaceborne hyperspectral instrument to acquire a 10nm
sampling interval over the contiguous visible/near-infrared and
I. INTRODUCTION shortwave infrared spectral data of 356-2577nm [11]. It
provides 242 potential bands with 30m spatial resolution,
Atmospheric correction is the premise of land surface
which bears the potential to provide data for accurate
parameters quantitative analysis using remote sensing data, its
vegetation identification. This paper used FLAASH to make
purpose is to remove or decrease the influence of scattering and
atmospheric correction of Hyperion and ALI images and
absorption of atmospheric molecule and aerosol on object
analyzed the effect of atmospheric correction.
reflectance. The methods of atmospheric correction are mainly
empirically statistical method and physically-based radiation
transfer model. 6S (Second simulation of the satellite signal in II. STUDY AREA AND DATA SETS
the solar spectrum), FLAASH (Fast Line-of-sight Atmospheric
Analysis of Spectral Hypercubes) and ACORN are based on A. Study Area
radiation transfer theory [1, 2, 3]. Empirically statistical The study area is Zhangye city, Gansu Province, China in
methods are mainly based on field measured spectra and to Heihe River valley, located at 37˚28′N-39˚57′N, 97˚20′E-
build empirically linear equation to finish atmospheric 102˚12′E (Fig. 1). It lies at the plain in front of Qilian
correction, including ratios methods, dark-object methods, Mountain. Gansu province belongs to temperate monsoon

1-4244-2394-1/08/$20.00 ©2008 IEEE.


2008 International Workshop on Earth Observation and Remote Sensing Applications

climate, having characteristics of transitional climate to ALI provides Landsat type panchromatic and multispectral
continental. The study area has mean annual air temperature of bands (Table Ⅰ) [13]. ALI data were acquired on September
4-9˚C and mean annual precipitation of 40-200mm. It is oasis 20, 2007 at 03:36 UTC. Image center location was 38.7753N,
agricultural area, and the main crops are corn, wheat, sugar 100.396E. Its data format was L1R HDF. ALI sensor includes
beet, potato, etc. four separate sensor chip assembly (SCAs), so ALI L1R
product includes four SCA image file (.M#R) stored in BSQ
format, one metadata file (.MET) and auxiliary file (.hdf).

TABLE I. CHARACTERISTICS OF ALI

Band Wavelength (µm) Resolution (m)


Pan 0.48 - 0.69 10
MS-1́ (Blue) 0.433 - 0.453 30
MS-1 (Blue) 0.45 - 0.515 30
MS-2 (Green) 0.525 - 0.605 30
MS-3 (Red) 0.63 - 0.69 30
MS-4 (VNIR) 0.775 - 0.805 30
MS-4́ (VNIR) 0.845 - 0.89 30
MS-5́ (SWIR) 1.2 - 1.3 30
Figure 1. The study area at Zhangye city, Gansu Province, China MS-5 (SWIR) 1.55 - 1.75 30
MS-7 (SWIR) 2.08 - 2.35 30

B. Field Plot Data


Field data of 35 corn plots were gathered in the study area. III. REMOTE SENSING I MAGE PROCESSING
Each plot size covered a minimum of 90m×90m, three times
greater than 30m. At each plot, three sampling points’ canopy A. Geometric Correction
reflectance spectra were recorded on September 2-12, 2007, in The Landsat ETM+ image was subset to form the image of
near-real time to coincide with EO-1 satellite overpass. The the study area, shown as Fig. 2a. Using ETM+ image as
instrument used was Analytical Spectral Devices (ASD) reference image, more than forty ground control points (GCP)
Fieldspec spectroradiometer, spectral range is 350-2500nm were selected, and Hyperion image was geometrically
with 1nm sampling interval. Geographical locations and corrected to UTM projection using second order polynomial
elevation of sampling points were recorded by Global interpolation, the total root mean square error was within one
Positioning System (GPS). pixel size (30m). According to the field sampling plots,
Hyperion image of the study area was shown in Fig. 2b, on
C. Remote Sensing Data which the vector GPS point coverage was overlaid. For ALI
Remote sensing data used in this study are Lansat-7 ETM+ data, four SCA image data were mosaicked to obtain a
data, Hyperion hyperspectral data and ALI data. Hyperion and continuous ALI image. Using the corrected Hyperion image to
ALI are two of the three sensors of EO-1 satellite. Hyperion correct ALI image, 35 GCPs were selected, the total root mean
swath width is 7.7km. ALI cross-track ground swath width is square error was also less than 30m. ALI image was shown in
37km and along-track length of an ALI scene generally is Fig. 2c.
42km or 185km.
According to World Reference System (WRS) path-row
(133-33) of the study area, ETM+ data were downloaded at
Earth Science Data Interface (ESDI) at the Global Land Cover
Facility [12], acquired on July 7, 1999. The projection of the
ETM+ image was UTM, reference ellipsoid was WGS84, and
zone number was 47N. In ARC/INFO software, the GPS data
were converted into vector point coverage and projection was
transformed into UTM.
Hyperion hyperspectral data were acquired on September
10, 2007 at 03:46 UTC. It was located at 38°28'33.96"N-
39°23'35.80"N, 100°17'34.79"E-100°39'35.05"E. Data Format
was HDF radiometrically corrected (Level 1R) data. In 242
channels, bands 1-7 (356-417nm), bands 58-70(936-1058nm),
bands 71-76(852-902nm) and bands 225-242 (2406-2578nm) a. Landsat ETM+ image on July 7, 1999
are not calibrated and set to zero in L1R product [13]. Thus, the
final L1R data provides a total of 198 bands representing 427-
2395nm continuous spectra with 10nm spectral resolution. The
data files are organized in band-interleaved-by-line (BIL)
format.

1-4244-2394-1/08/$20.00 ©2008 IEEE.


2008 International Workshop on Earth Observation and Remote Sensing Applications

Absolute radiance and apparent reflectance of four typical


objects in this study area on Hyperion image were shown in
Fig. 3.

b. Hyperion image on September c. ALI image on September


10, 2007 20, 2007
a. Absolute radiance
Figure 2. Remote sensing data of the study area (cross symbols are GPS
field sampling points)

B. Radiometric Correction
The DN values of Hyperion and ALI L1R product are
scaled at-sensor radiance and stored as 16-bit signed integer.
VNIR bands (8-57) and SWIR bands (77-224) of Hyperion
data have a scaling factor of 40 and 80, respectively, so the
absolute radiance is calculated by (1) and (2) [13]. The absolute
radiance of ALI image is calculated by (3).

LVNIR = DN / 40
(1)

LSWIR = DN / 80 (2)
b. Apparent reflectance

Figure 3. Absolute radiance and apparent reflectance of four typical objects


on Hyperion image(427-2396nm)
L = DN ⋅ scalingfactor + offset (3)
IV. ATMOSPHERIC CORRECTION AND RESULT ANALYSIS
Apparent reflectance is also called planetary reflectance, or
reflectance at top of atmosphere (TOA). This combined surface A. FLAASH Introduction
and atmospheric reflectance of the Earth is determined by (4).
The purpose of atmospheric correction is to remove or
decrease the influence of atmosphere on object reflectance, and
π ⋅ Lλ ⋅ D 2 separate the contribution of land object from atmosphere-land
ρ TOA = mixed signal. Radiometric calibration and atmospheric
ESUN λ ⋅ cosθ s (4) correction is the premise of quantitative remote sensing.
Atmospheric correction of the Hyperion and ALI images was
Where, ρTOA is planetary reflectance (unitless); Lλ is performed using FLAASH (Fast Line-of-sight Atmospheric
Analysis of Spectral Hypercubes) in ENVI's atmospheric
spectral radiance at sensor's aperture; D is Earth-Sun distance correction module [3]. FLAASH supports both hyperspectral
in astronomic units at the acquisition date, and it can be sensors (such as HyMAP, AVIRIS, HYDICE, Hyperion,
interpolated; ESUN λ is mean solar exoatmospheric irradiance; Probe-1 and CASI) as well as multispectral sensors (such as
Landsat, SPOT, IRS and ASTER). FLAASH corrects
θ s is solar zenith angle. wavelengths in the visible through near-infrared and short-
wave infrared range. Unlike many other atmospheric correction

1-4244-2394-1/08/$20.00 ©2008 IEEE.


2008 International Workshop on Earth Observation and Remote Sensing Applications

programs that interpolate radiation transfer properties from a shallow and not pure, its reflectance spectra were not typical.
pre-calculated database of modeling results, FLAASH Most of the reflectance of water body after FLAASH
incorporates MODTRAN4 radiation transfer code. You may atmospheric correction was lower than those before
choose any of the standard MODTRAN model atmospheres atmospheric correction, while in the 1114-1154nm, it is higher
and aerosol types to represent the scene, and a unique than those before correction. The reflectance of desert after
MODTRAN solution is computed for each image. FLAASH FLAASH atmospheric correction at wavelengths greater than
also includes correction for adjacency effect, an option to 508nm was higher than those before correction. The
compute a scene-average visibility, uses the most advanced reflectance of buildings after FLAASH atmospheric correction
techniques for handling particularly stressing atmospheric at wavelengths less than 915nm and 1003-1094nm were lower
conditions (such as the presence of clouds), and cirrus and than those before correction.
opaque cloud classification map [3].

B. Atmospheric Correction of Hyperion and ALI Images


Input parameters of FLAASH for the Hyperion and ALI
images were shown in Table Ⅱ. To accurately calculate
surface reflectance, reflectance image after FLAASH
atmospheric correction removed automatically non-calibrated
bands, repeated bands and atmospheric vapor absorption bands
(1346-1467nm and 1800-1971nm) , and the resulted image had
reflectance of 158 bands (427-2355nm) .

TABLE II. INPUT PARAMETERS OF FLAASH FOR HYPERION AND ALI


IMAGES
Hyperion ALI image
Parameters
image
38˚50΄21.85˝N, 38˚46΄31.08˝N, a. Reflectance on Hyperion image
Image center location
100˚26΄36.69˝E 100˚23΄45.61˝E
Sensor altitude 705km
Ground elevation 1.5km
Pixel size 30m
Flight date Sept. 10, 2007 Sept. 20, 2007
Flight time GMT 03:46:32
Atmospheric model Mid-latitude summer
Aerosol model Rural
Initial visibility 20km
Aerosol scale height 2km
CO2 mixing ratio 390ppm
Modtran resolution 5cm-1
Modtran multiscatter model Scaled DISCORT
DISCORTstreams number 8
Non-nadir looking instrument
175.8408˚, 171.695˚
Zenith angle
15˚ 0˚
Azimuth angle

b. Reflectance on ALI image


C. Result Analysis Figure 4. Reflectance of four typical objects on Hyperion and ALI images
Reflectance spectra of four typical objects (corn of plot 15, after FLAASH atmospheric correction
water body, desert and buildings) on Hyperion and ALI images
after FLAASH atmospheric correction were shown in Fig. 4a Reflectance at atmospheric vapor absorption bands (near
and Fig. 4b. Spectral resolution selected was 5cm-1, and 1400nm and 1900nm) on Hyperion image was removed.
spectral polishing was made using the average of neighboring Reflectance of corn at near infrared bands on ALI image was
nine-band spectra. The results showed that reflectance on lower than that on Hyperion image, because the acquisition
Hyperion image showed the rich spectra information of objects. date of ALI image was September 20, 2007, ten days later than
After FLAASH atmospheric correction, corn showed typical Hyperion image.
reflectance spectra curve with high reflectance at green band To remove the noise during atmospheric correction of
and low reflectance at red band, but the depth of red band was images, spectral polishing was needed. It was made using the
not obvious, maybe due to some processing errors or the poor average of neighboring multiple bands. For Hyperion
growth because of insect pest. In the 8-36 band (427-712nm), hyperspectral data with 10nm resolution, the number of 9 was
the reflectance of corn were lower than those before used. Fig. 5 showed the effect of spectral polishing processing
atmospheric correction, while at other bands (> 712 nm), the on the reflectance of four objects. After spectral polishing, the
results were opposite. As the water body in the study area was

1-4244-2394-1/08/$20.00 ©2008 IEEE.


2008 International Workshop on Earth Observation and Remote Sensing Applications

obvious noise of reflectance spectra was removed. The effects To test the effect of atmospheric correction of FLAASH for
of spectral polishing for corn, desert and buildings were better Hyperion image, according to the sampling points determined
than that for water body. by GPS, corn ASD reflectance spectra was processed. Using
spectral response function of Hyperion, that is, the wavelength
4000 5000 center of full width at half maximum (FWHM) of each band
and Gaussian filter function, ASD spectra were resampled to
4000 corresponding Hyperion bands. The formula of resampling was
Reflectance(*10000

3000 Before After


3000
(5).
2000
2000
N N
1000
1000 ρ = ∑ ( β (λ ) * ρ (λ )) ∑ β (λ ) (5)
λ =1 λ =1
0 0
447 651 854 1054 1255 1457 1659 1861 2063 2264
Wavelength (nm)
In which, ρ is resampled ASD reflectance corresponding
to Hyperion bands; ρ ( λ ) is ASD reflectance, β (λ ) is the
a. Corn weight of spectral response function corresponding to each
band.
1200 2000
Take corn of plot 15 for example, resampled ASD
Before After
1600 reflectance spectra of 1nm interval to Hyperion reflectance
Reflectance(*10000

800 spectra and corn spectra on Hyperion image after atmospheric


1200
correction were shown in Fig. 6a. It was clear that these two
800
reflectance were consistent with each other, the correlation
400 coefficient was 0.987, the reflectance difference was -0037-
400 0.034.
0 0 According to ALI spectral response function, ASD spectra
447 651 854 1054 1255 1457 1659 1861 2063 2264 were resampled to ALI image. Corn spectra on ALI image after
Wavelength (nm) atmospheric correction and resampled ASD reflectance spectra
were shown in Fig. 6b. Because ASD reflectance measurement
b. Water body was on September 2-12, 2007, while the acquisition date of
ALI image was September 20, 2007, about ten days later than
6000 8000
ASD measurement, and during that period, corn was developed
7000
5000 to mature and to senesce, the chlorophyll of leaf began to
Reflectance(*10000

6000
4000
5000
decrease, and absorption of red light decreased. Therefore,
3000 4000
compared to ASD spectra, reflectance on ALI image after
3000 FLAASH atmospheric correction at red band (660nm)
2000
2000 increased and reflectance at near infrared (790nm and 866nm)
Before After
1000
1000 decreased.
0 0
447 651 854 1054 1255 1457 1659 1861 2063 2264
Wavelength (nm)

c. Desert

1600 2000

1600
Reflectance(*10000

1200
1200
800
800

400
Before After 400

0 0
447 651 854 1054 1255 1457 1659 1861 2063 2264
Wavelength (nm) a. Resampled ASD spectra and corresponding Hyperion spectra

d. Buildings

Figure 5. The effects of spectral polishing on reflectance spectra

1-4244-2394-1/08/$20.00 ©2008 IEEE.


2008 International Workshop on Earth Observation and Remote Sensing Applications

The authors would like to greatly thank the China’s Special


Funds for Major State Basic Research Project
(2007CB714406), the Knowledge Innovation Program of the
Chinese Academy of Sciences (KZCX2-YW-313) and the
foundation of the State Key Laboratory of Remote Sensing
Science (KQ060006), which subsidized this research.

REFERENCES
[1] Vetmote E F, Tanre D, Deuze J L, et, al., “Second simulation of the
satellite signal in the solar spectrum, 6S: an overview,” IEEE
Transactions on Geoscience and Remote Sensing, 1997, 35, pp. 675-
686.
[2] Zheng Wei , Zeng Zhiyuan, “A review on methods of atmospheric
correction for remote sensing images,” Chinese, Remote Sensing
Information, 2004, (4), pp. 66-70.
[3] FLAASH Module User’s Guide, ENVI FLAASH Version 4.2, Research
b. Resampled ASD spectra and corresponding ALI spectra Systems, Inc. August, 2005 Edition
[4] LIANG Shunlin. Quantitative Remote Sensing of Land Surface. A JOHN
Figure 6. Comparison of resampled ASD spectra of corn and corresponding WILEY & SONS, INC. PUBLICATION. Jin Au Kong, editor. New
spectra on Hyperion and ALI images after FLAASH atmospheric correction Jersey, 2004, pp. 196-230.
[5] Song Xiaoyu, Wang Jihua, Liu Liangyun, Huang Wenjiang, Zhao
Chunjiang, “Atmospheric correction of hyperspectral imagery:
V. CONCLUSIONS evaluation of the FLAASH algorithm with AVRIS data,” Chinese,
Remote Sensing Technology and Applications, 2005, 20(4), 393-398.
Atmospheric correction is the premise of quantitative study [6] Lv Fengjun, Hao Yuesheng, Li Chuanping, Han Shuling, Wang Juan,
using remote sensing data. This paper used FLAASH module “Application of atmospheric correction of remote sensing data based on
in ENVI to make atmospheric correction of Hyperion and ALI FLAASH module,” Chinese, Hebei Geology, 2007, 2, pp.23-26.
images. FLAASH incorporates MODTRAN4 radiation transfer [7] Yuan Jinguo, Niu Zheng, “Nitrogen and chlrophyll mapping based on
code, and atmospheric correction of all bands is made Hyperion hyperspectral image,” Chinese, Nongye Gongcheng Xuebao,
simultaneously, and also it is fast, especially it has obvious 2007, 23(4), pp.172-177.
advantage for atmospheric correction of hyperspectral image. [8] Shen Yan, Niu Zheng, Chen Fang, Wand Changyao, “Surface
reflectance retrieval of Hyperion hyperspectral imagery based on the
Reflectance spectra curve after spectral polishing can remove empirical line calibration method,” Chinese, Geography and Geo-
the obvious noise during atmospheric correction of images. information Science, 2007, 23(1), pp. 27-30.
Using spectral response function of Hyperion and ALI images, [9] LI Qingli, Xue Yongqi, Wang Jianyu, Bai Zhiquan, “Atmospheric
ASD spectra of corn were resampled to corresponding correction of PHI hyperspectral image,” Chinese, Journal of Infrared and
Hyperion and ALI bands. The results showed that reflectance Millimeter Waves, 2006, 25(4), pp. 316-320.
spectra on Hyperion and ALI images after atmospheric [10] Long Fei, Zhao Yingshi, “Atmosphere correction of multi-angle NOAA
correction were consistent with resampled ASD spectra, which data based on BRDF reflectance,” Chinese, Journal of Remote Sensing,
demonstrates that atmospheric correction using FLAASH is 2002(6), pp. 173-179.
effective. [11] Richard Beck. EO-1 User Guide, V.2.3. 2003.
[12] http://glcfapp.umiacs.umd.edu:8080/esdi/index.jsp
[13] http://eo1.usgs.gov/faq.php

ACKNOWLEDGMENT

1-4244-2394-1/08/$20.00 ©2008 IEEE.

Вам также может понравиться