International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
ground area a field-of-view (FOV) of +30 degrees and more is 
needed. Atmospheric absorption and scattering is highly angle 
dependent and so the atmospheric impact will be different 
within the image. This is true for framing cameras (classical 
aerial imaging cameras) or line scanner systems (like the 
ADS40). The appearance of the effect depends on the imaging 
principle: In framing cameras the atmospheric effects have a 
circular shape, while those in line scanner images have a linear 
shape. Usually linear shapes appear more disturbing, but this is 
a fact of human perception. 
The atmospheric effects originate in gaseous absorption of the 
directly transmitted light and the gaseous (Rayleigh-) and 
aerosol (Mie-) scattering of indirect components reaching the 
sensor (Figure 2). 
  
Figure 2. Radiation components reaching the sensor in the 
reflective wavelength range (400 nm — 2500 nm). 
In classical aerial imaging cameras atmospheric effects have 
been thwarted by coloured gradient filters at the time of image 
capture. The atmospheric conditions had to be anticipated and 
no afterwards correction was possible. The path scattered 
upwelled radiance (path radiance, cf. Figure 2, component C) 
can be equal to the direct radiance for dark targets (1 to 3 96 
reflectance) (Schott, 1997, p. 118). This causes a general 
brightening of the image and reduces the contrast. 
Nowadays with digitized aerial images and especially with 
digital cameras like the ADS40, atmospheric correction 
becomes an issue of post-processing. The difficult and time 
consuming task of removing the path radiance can be shifted to 
offline work. 
Exploiting the higher pixel dynamics of digital cameras will 
reduce the need for higher resolution as suggested by the 
General Image Quality Equation GIQE (Leachtenauer et al., 
1997): A doubling of the normalized relative edge response 
RER which is related to the contrast, can compensate for double 
the ground sampling distance GSD. 
Satellite imagery, which is inherently digital, has initiated the: 
development of a variety of procedures to convert sensor digital 
numbers (DN) to reflectance, a target property (the ratio of 
reflected to incoming light). 
The process of generating reflectance images involves several 
steps: 
|. Radiometric calibration: Convert DN to at-sensor 
radiances, a radiometric quantity measured in 
W/m°/sr/nm. 
Atmospheric correction: removal of the path radiance, 
ie. the stray light from the atmosphere. Correction of 
the adjacency effect, the outshining of the target by 
nearby bright objects (Dave, 1980), cf. Figure 2, 
component E. 
3. Correction of the anisotropic reflection properties of 
the targets (cf. sec. 3). 
t2 
+ 
N 
4.  Reflectance calibration: removal of the spectral effect 
of solar illumination by dividing through the 
incoming irradiation. 
Methods for reflectance image generation range from purely 
empirical methods to complex radiative transfer models. A more 
detailed discussion can be found elsewhere (Roberts et al., 
1986, Moran et al., 1992). Simple methods try to cover the 
above sequence in one step, more accurate methods will do the 
steps individually. 
The empirical methods can be divided into interactive methods 
where certain test areas have to be identified, like the "flat field" 
and the "empirical line fit" method (the latter requiring 
spectra) (Kruse, 1988). 
Non-interactive empirical methods use statistical methods like 
the ‘‘dark-object subtraction” method (Chavez, 1975, and an 
improved version, Chavez, 1988). 
eround 
= 
Physically based methods, like ATCOR (Richter, 1996, Richter, 
2002), ATREM/TAFKAA (Gao and Davis, 1997, Gao et al., 
2000), ENVI/FLAASH (Research Systems, Inc.), or ACORN 
(Analytical Imaging and Geophysics, LLC), will use radiative 
transfer (RT) models, like the 6S model (Vermote et al., 1997) 
or MODTRAN (Berk et al., 1998). 
Radiative transfer models require the knowledge of a set of 
parameters (atmosphere type/concentration profiles of gases, 
aerosol type and concentration, flight and ground elevation, 
illumination and view angles). Whereas the geometric 
parameters can be determined from flight management data, the 
atmospheric parameters are not immediately accessible to the 
user. In the case of imaging spectrometer data with adequate 
spectral channels, aerosol and water vapour concentration may 
be estimated from the data itself. However, this is not possible 
for the broadband channels of the ADS40. 
Furthermore RT models tend to be time consuming and need 
optimization steps to be runtime efficient. 
So according to the needs of different users, different strategies 
have to be taken for providing ADS40 reflectance images 
(Table 1). 
  
  
  
  
  
  
User Photogrammetry | Remote Sensing 
Amount of data Large Usually limited 
Processing speed High Moderate 
jeometric accuracy High Moderate to low 
Radiometric accuracy | Low High 
In-scene radiometric | Moderate High 
homogeneity 
  
  
  
  
  
Atmospheric Empirical Physical 
correction 
BRDF correction Empirical Empirical or 
physical 
  
  
Table 1. Requirements for radiometric data processing for 
photogrammetry and remote sensing. 
3. BIDIRECTIONAL EFFECTS 
Bidirectional effects have an impact on image quality of the 
same order of magnitude as atmospheric effects. 
The most prominent effect is the so-called hot-spot in aerial 
images. It is placed at the projected solar position, opposite the 
Ini 
sp: 
fre 
mi 
go 
lig 
in 
da 
th