579 
SPECTRAL VARIABILITY EFFECTS ON THE ATMOSPHERIC CORRECTION OF 
IMAGING SPECTROMETER DATA FOR SURFACE REFLECTANCE RETRIEVAL 
P.M. Teillet 
Canada Centre for Remote Sensing 
1547 Merivale Road, Nepean, Ontario, Canada K2G 4V3 
and 
J.R. Irons 
Biophysical Sciences Branch 
NASA Goddard Space Flight Center, Greenbelt, Maryland, USA 20771 
ABSTRACT 
Based on apparent radiances (nadir-looking) from 
the airborne Advanced Solid-State Array Spectro- 
radiometer (ASAS) and measured optical depth 
values, a surface reflectance spectrum was 
computed by an atmospheric code run (i) monochro- 
matically at band center wavelengths, (ii) for 
rectangular bandpasses, and (iii) for spectral 
response profiles. A comparison of the results 
from these different cases indicates that 
anomalous results for surface reflectance can be 
obtained in the vicinity of absorption features in 
the monochromatic case and that bandpass computa 
tions yield better results. Even with bandpass 
calculations, if there is no information available 
on gaseous absorption at the time of image 
acquisition and standard values are used in the 
atmospheric correction of high spectral resolution 
data, it is shown that the retrieved surface 
reflectances can depart significantly from normal 
values in spectral regions affected by absorption. 
INTRODUCTION 
In remote sensing, an atmospheric radiative 
transfer code is used to determine surface 
reflectances given image data acquired by 
satellite or aircraft sensors. The more accurate 
models of radiative transfer through the 
atmosphere are complex and can require a lot of 
computer processing time. Thus, in remote sensing 
studies, such models are often run monochromati- 
cally to represent a spectral band in order to 
save time. With the relatively narrow bands of 
imaging spectrometers, such an approach should 
work well throughout most of the relevant spectral 
domain. However, the narrowness of the imaging 
spectrometer bands also implies greater 
sensitivity to atmospheric absorption features, 
which can be very spectrally selective. 
The effects of these factors on atmospheric 
correction of imaging spectrometer data were 
examined using data from the airborne Advanced 
Solid-state Array Spectroradiometer (ASAS), used 
by the NASA Goddard Space Flight Center for 
multiple direction reflectance observations (Irons 
et al., 1989). The sensor primarily consists of 
a charge injection device with a 512 by 29 silicon 
photodetector array. A diffraction grating is 
used to distribute the spectral information onto 
the 29-array dimension, from which 29 channels are 
obtained in the 465-871 nm range. For a flight 
altitude of 5000 feet, the nadir pixel size is 4.5 
m across track. The data acquisition strategy is 
to change the pointing direction of the sensor 
from 45 degrees fore to 45 degrees aft while 
flying over a specific target location. In this 
way, seven images are acquired at 15 degree 
increments in view angle for a given target area. 
Thus, ASAS is a particularly useful instrument for 
studying bidirectional reflectance distribution 
functions of natural surfaces. 
ASAS data were acquired by NASA over Konza prairie 
grass in 1987 as part of the First ISLSCP Field 
Experiment (FIFE), where ISLSCP stands for the 
International Satellite Land Surface Climatology 
Project. Radiometric calibration coefficients 
were used to convert the data recorded during the 
flight to physical radiance units. An individual 
radiance spectrum was extracted from a nadir image 
acquired over vegetation. In the present 
atmospheric correction study, attention was 
focused on 9 channels encompassing the spectral 
range from about 0.670 to 0.805 micrometers, with 
each individual channel having a bandwidth of 
0.015 micrometers. This portion of the 
reflectance spectrum of vegetation includes the 
chlorophyll minimum in the visible and the rise up 
to the infrared plateau region. The feasibility 
of retrieving surface reflectances from the 
measured radiances has been investigated with the 
help of the 5S atmospheric code (Tanre et al., 
1986; Teillet, 1989). 
ANALYSIS AND RESULTS 
Monochromatic Case 
Based on the input conditions listed in Table 1 
and the ASAS apparent radiances, surface 
reflectances were computed by 5S monochromatically 
at the band center wavelengths listed in Table 2. 
However, computations in the 5S code are tied to 
a wavelength grid with 0.005 micrometer spacing. 
Thus, if the specified wavelength is 0.678 micro 
meters, for example, the code will actually 
generate results for a wavelength of 0.680 micro 
meters, the nearest grid wavelength. Results from 
this approach are listed in Table 2 under the 
heading "5S LAMBDA GRID". 
Rectangular Bandpass Case 
Each individual ASAS band is not monochromatic but 
rather has a bandwidth of 0.015 micrometers. A 
rectangular bandpass was adopted and used in a 
second series of 5S runs to approximate the 
spectral response profile of each ASAS channel. 
The results are presented in Table 2 under the 
heading "5S BAND (RECT)". 
Profile Bandpass Case 
It can be argued that a rectangular bandpass of 
0.015 micrometers is not very representative of