I. GENERAL OVERVIEW OF DEVELOPMENTS AND PROPOSED SATELLITE PROGRAMS 
One of the emerging areas of technology development in the United 
States and elsewhere is the development of algorithms for the treatment 
of Landsat digital data. : 
In a simplistic manner, the processing of the data may be viewed 
as a four-step process: 
1) The removal of radiometric irregularities among the detectors 
(basically, a calibration process) and the application of systematic 
geometric corrections required by a system observing a round, rotating 
Earth and converting them to a plane surface, such as a photograph or 
a map; 
2) combining ephemeris data with the sensor data to remove, to 
the degree possible, variables that are nonsystematic, such as space- 
craft attitude and altitude. (It should be noted that to this point 
the radiometric fidelity of the data, depending on adequacy of sensor 
calibration, is preserved); 
3) the transformation of the geometry of the data, from that 
recorded and preliminarily corrected, to match a selected map projec- 
tion. (In this process, the area recorded by a given pixel of 
information is shifted with respect to the surface of the Earth, and 
the radiometric quantities associated with this pixel must be modified 
to accommodate this shift). This process of radiometric adjustment, 
caused by geometric shift, is commonly referred to as resampling. A 
number of algorithms have been developed to accomplish this step; the 
optimum is the sine x over x approach but this is extremely costly in 
terms of computer time hence, most algorithms in current use, are 
approximations; the most commonly used is the so-called cubic 
convolution; but whatever algorithm is used, the net effect is a 
diminution of contrast among adjacent pixels (smoothing of the data); 
4) the extraction of categories of information (themes) based 
largely on radiometric properties. These algorithms can be largely 
grouped into two categories: a) Those that train on a given scene or 
scenes (This is commonly referred to as a supervised approach); and 
b) those that mathematically cluster and separate increments of the 
scene on the basis of discrete characteristics that may or may not 
relate to basic discipline breakdowns (This is commonly referred to as 
the unsupervised approach). Recently, hybrid algorithms have been de- 
veloped that combine the two approaches by providing a cluster analysis 
from within a given training sample. 
use 
fou 
ste 
Ber 
exa 
of 
exa 
(US 
gen 
Riv 
rai 
acc 
1:2 
rec 
sta 
rep 
bee: 
Mor: 
Wer: 
of 
the 
Tec] 
Roy: 
Oct 
(19 
Lan 
COV 
ine: 
expe 
gro 
ERO: 
loc: