8 
at different incidence angles is therefore highly dependent on knowledge of the shape of the 
surface. The precision location of measurements is essential if we are to be able to integrate and 
fuse data sets. Compensation for the relief distortion effect which arises in optical sensors viewing 
off-nadir, and the layover effect inherent in the geometry of synthetic aperture radar both require a 
priori knowledge of the shape of the underlying terrain. A digital elevation model is therefore a 
critical element in the construction of the measurement database. 
The acquisition of a DEM on a worldwide basis is therefore a high priority task. There are two 
technical approaches to such an undertaking: Stereo methods using optical sensors, and 
synthetic aperture radar interferometry. The first method has been demonstrated to yield the 
necessary accuracies (Kauffman and Wood, [1987], Simard, [1988] ), and the second method, 
though less mature, shows significant promise (Im, [1990] , Li and Goldstein [1990]. 
Interferometric radar has the significant advantage that it is capable of operating through cloud 
cover, and therefore presents the basis of a system which will be able to acquire a world-wide 
model in a predictable interval of time. 
QeQCPding 
The requirement for precise location of each pixel independent of the acquiring satellite implies 
that the data should be transformed into a standard coordinate system, as part of the geometric 
rectification process. The necessity to be able to relate each pixel to the topography in order to 
determine the reflectance model, and the general requirement that different data sets all relating 
to the same area of the earth's surface should be easily combinable, imply that the selected 
standard coordinate system should be a standard geographic projection. Satellite data 
geometrically corrected to a standard geographic grid is called geocoded data. 
One can see that if, for each data set, the position and attitude of the sensor is available together 
with the characteristics of the illumination source and the time of acquisition, and each pixel is 
calibrated and geocoded, it is possible to: 
(a) reconstruct the path travelled by the energy through the atmosphere, and hence 
derive the appropriate atmospheric corrections (assuming one has information 
which allows proper characterization of the atmosphere as it was at the time of 
data acquisition). 
(b) reconstruct the illumination conditions on each surface pixel (assuming the 
availability of a DEM). 
(c) combine the data set with other geographic data and with other geocoded 
remote sensing data sets. 
Geocoding the data is central to successfully accomplishing these objectives. 
Geocoded Data Sets 
Detailed treatments of the geocoding of remote sensing data are to be found elsewhere 
(Friedmann [1981], Guertin and Shaw [1981], MacDonald and Friedmann [1985]), hence the 
discussion presented here will cover only the important features of this type of data set. 
The basic idea behind the geocoding of a data set is to transform it from the coordinate system in 
which it is acquired, which is dependent on the acquisition system (satellite and sensor), into a 
geographic coordinate system which is linked into a standard mapping system, and therefore 
independent of the geometric characteristics of the spacecraft orbit etc. This concept is illustrated 
in Figure 5. r 
The figure shows the satellite scanning track running from top right to bottom left. Image data long 
this track is divided into segments known as scenes, and the pixels are arranged in rows 
perpendicular to the satellite track. This coordinate system, in the case of the LANDSAT satellite,