geographic positioning accuracy of + 55m should be obtainable through 
digital processing of LANDSAT MSS data with 25 to 30 ground control 
points per scene. 
Major sources of geometric distortions are skew, aspect ratio, 
mirror scan rate, scale distortions, and topographic variations within 
the scene. These geometric corrections are applied to bulk processed 
MSS film data (ERTS Data Users Handbook). Unlike the film products, 
LANDSAT CCT's are not geometrically corrected for skew and non-linear 
mirror scan rate nor are they corrected to a given map projection (Thomas, 
1973). Image rectification procedures have been developed by numerous 
investigators (Goetz et al, 1975; Bernstein, 1973a and 1973b; Bernstein 
and Ferneyhough, 1975; Taber, 1973; Rifman, 1973; Peet, Mack and Crosson, 
1973; and Peet, 1975). These techniques require location of ground con- 
trol points, determining geometric correction functions and applying 
these functions to the original distorted image. The intensity or bright- 
ness values of picture elements ín the corrected image are determined with 
one of several resampling techniques. 
Resampling techniques vary in the way the brightness value for a 
pixel in the corrected image is determined from the distorted image. 
Nearest neighbor, bilinear interpolation, and cubic convolution algorithms 
are common resampling techniques. Goetz et al (1975), Bernstein and Fer- 
neyhough (1975) and Tabor (1975) provide detailed descriptions of these 
resampling techniques. Although many variables will affect the accuracy 
of geometric correction, techniques currently available are capable of 
producing images geometrically corrected to approximately sixty meters 
(Bernstein, 1973 and Nichols, 1975). 
Sampling procedures 
One of the most promising advancements in remote sensing technology 
during the past decade involves the merging of statistical sampling 
theory with the collection and interpretation of remote sensing imagery 
and ancillary data. The specific use of sampling may vary, but its ap- 
plication is based on the general concept that remote sensing data ac- 
quisition is, in essence, a sampling of the physical environment in both 
time and space. Once this is accepted, it follows that statistical theory 
provides a tool for optimizing that sampling in terms of accuracy and 
cost. 
In the temporal sampling context, much work has been done in more 
carefully defining the optimum data (or dates) for image acquisition 
based on expected changes in the appearance of the target. For example, 
the LACIE (Large Area Crop Inventory Experiment) Project, a large NASA 
undertaking concerned with developing efficient methods for conducting 
worldwide crop inventories, has prompted considerable study of the use 
of imagery acquired at several times during the growing season and inter- 
preted in conjunction with crop calendars adjusted for growing conditions, 
region by region. For many vegetation applications it has been recognized 
that changes through the growing season, rather than acting as a confusion 
factor, can, if understood, be used to increase the amount of information 
that can be extracted from the imagery or to define that time at which 
imagery should be acquired to maximize interpretability. 
The concept of multistage sampling has recently gained acceptance 
as a means for increasing the efficiency of data collection, particularly