CCT products, will be available to the user in geometrically corrected 
format. The implementation of capabilities to perform a series of stan- 
dard geometric and radiometric corrections at the EROS Data Center is 
planned for the late 1970's. A standard geometric correction function 
will be applied to LANDSAT data prior to dissemination of data from the 
EROS Data Center. Realizing that many investigators prefer to work with 
raw data and perform their own specific corrections, uncorrected data 
will continue to be made available. 
Many geometric correction algorithms employ resampling techniques 
to calculate the brightness value of a given pixel in the corrected 
image from the brightness values of pixels in the distorted image. Sev- 
eral resampling techniques sample the data and determine new values by 
interpolation or averaging with weighted values for a series of pixels 
around the pixel in question. The effects of such resampling techniques 
on the radiometric properties of LANDSAT data and subsequent classifi- 
cation with machine oriented decision rules are not certain due to a 
lack of thorough documentation. Detailed studies are needed to document 
these effects with respect to the type of geometric correction algorithm 
and accuracy of image classification at different levels of detail.  Rec- 
ommendations with supporting evidence are needed with regard to the type 
of geometric correction algorithm to use to achieve specific accuracy 
levels and whether such geometric correction functions should be applied 
prior to or after image classification. 
Research has shown that when more than one date of imagery is used 
in an analysis, improved classification results can be achieved. Also, 
multidate image overlays have proven useful in monitoring change in spec- 
ific resources. The capability to register two different scenes or to 
register one scene to a specific map base for location of training data 
is available. Currently, registration of two scenes can be accomplished 
with estimated registration errors ranging from less than one picture 
element to as much as two or three picture elements. High order poly- 
nomial geometric correction functions and an increased number of ac- 
curately located ground control points are needed to achieve registration 
errors of less than one picture element. The cost of implementing such 
procedures is quite high. Efforts should be focused to continue imple- 
mentation of such techniques on high speed parallel processors to reduce 
the processing time and subsequent cost to the user. 
Although there is reasonable justification for image classification 
with multidate data, one must question the validity of such classifi- 
cation when registration errors of plus or minus one to three picture 
elements may be encountered. Thorough documentation of the effect of 
registration error on classification accuracy is needed. Another aspect 
of multidate analyses which must be investigated is the value of the 
incremental increase in classification accuracy with respect to the 
increased cost associated with classification of multidate data. For 
°xample, if classification accuracy is improved from 89% to 92%, the 
increased accuracy would not appear to justify the substantial increased 
cost of processing the multidate data. Research is needed to invest- 
igate the significance of multidate data on accuracy of multispectral 
classification. Such research must be designed to assess incremental 
gains in accuracy in terms of classification accuracy, placement of 
boundary lines, geometric accuracy of the map overlay products, and 
reduction of sampling error and confidence limits of estimates derived