The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
The ASD field spectrometer is measuring in the range of 325 -
1075 nm. This range is covering the spectral range of Proba
CHRIS hyperspectral image data. The field spectrometer
measurements are collected randomly over the study area (Fig.
1). Before going to the field, the measurement sites are defined
from a Terra ASTER image (27.04.2001) and ancillary data
including different surface covers and accessible sites. As many
as possible sites have been chosen. At the field study 106 sites
are measured. Samples of the geological materials covering the
surface of the site are collected at 30 sites to be measured with a
laboratory spectrometer. Throughout the field study, at some
days there have been no stable weather conditions. This has to
some extent affected the field study. Due to this, before each
spectral measurement the spectrometer has been calibrated and
hand specimens collected.
M-ai* W3?ts intm .«-«w
»3JS*t M*WB M-361 J.r«B
Figure 1. Proba CHRIS image of the study area with field
measurement sites (yellow asterisks)
Not all the field measurement sites had laboratory samples. The
laboratory measurements of the hand specimens are done to
verify spectral details and provide information for removing
artifacts from the field and image data. After receiving the
image data it has been observed that some of the image is cloud
covered and cloud shadowed. These no data areas have even
included some field measurement sites.
METHODOLOGY
The Proba CHRIS image mode 1 with the acquisition angle -55
has been covering the study area. In the image data the missing
pixels have been filled and stripes have been removed using the
HDFclean program (Cutter, M., A., 2006). The clouds and
cloud shadows have been masked.
The Proba CHRIS image data is atmospherically corrected to
remove the effects of scattering and absorption of the
atmosphere and to convert from radiance values received at the
sensor to reflectance values of the imaged surface materials.
Three different atmospheric correction methods are used
(Bertels, et al., 2006 and Van der Meer, et al., 2001):
1) Scene derived correction: IARR
The internal average relative reflectance normalizes the image
to a scene average spectrum. This shifts all spectral radiances to
the same relative brightness. The resulting spectral values
represent reflectance relative to the average spectrum.
2) Ground-calibrating method: Empirical Line
Calibration
The technique is used to force image data to match selected
field reflectance spectra. It requires field or laboratory
reflectance spectra of at least two uniform ground targets. For
each spectral band a linear regression is calculated between the
reference spectra and the image spectra.
3) Radiative transfer model: 6S
The 6S (Second Simulation of the Satellite Signal in the Solar
Spectrum) model is used to predict the atmospheric radiative
properties and to model the radiance at the sensor (Vermote, et
al., 1997).
After calculating reflectance values of the Proba CHRIS image
the backscatter response curves are derived for different surface
materials to integrate the image, field and laboratory spectral
data. These data are used to create a spectral library of the study
area for the end member selection.
Following these, the dimensionality, the inherent redundancy
and noise in Proba CHRIS data is reduced by the principal
components transform, minimum noise fraction. Then the
reference spectra and the image spectra are compared and a
spectrum match, spectral angle mapper, is applied. The spectral
angle mapper is a spectral classification that matches pixels to
the reference spectra. The algorithm determines the spectral
similarity between two spectra by calculating the angle between
the spectra. Following the classification process the lithological
map of the (^ankiri-Eldivan area is produced and verified with
the ancillary data.
RESULTS AND DISCUSSION
At first, the atmospherically corrected Proba CHRIS
hyperspectral image data is compared with field and laboratory
spectra. Three different methods have been used for the
calibration of the hyperspectral image data: IARR, empirical
line calibration and 6S. The results of the internal average
relative reflectance correction have not matched the field and
laboratory spectra. There are a few green agricultural areas and
a small forest area in the study area. The vegetation and the
small dam lakes caused strong absorption and affected the result.
The empirical line calibration requires at least two reference
targets: white and dark. In the image area there are good white
targets but none dark target with field spectroscopy data is
supplied for the empirical line calibration. This caused a minor
mismatch with the field and laboratory spectra.
The best atmospheric correction has been the 6S radiative
transfer model. 6S is quite effective but needs a lot of input
parameters to execute the program. Some of these parameters
need additional field measurements. Nevertheless, the program
has let choose between default parameters.
Wavelenghts
range (nm)
Xa
Xb
Xc
Total
irradiance
405.0-486.9
0.0031
0.2242
0.19069
1372.602
486.9-586.9
0.0028
0.1078
0.13097
1439.093
586.9-697.8
0.003
0.0581
0.09385
1263.639
697.8-749.0
0.0036
0.041
0.07557
1044.318
749.0-1008.8
0.0052
0.0265
0.05653
712.782
Table 1. Coefficients for atmospheric correction simulated
using 6S code
440