EM 1110-2-2907
1 October 2003
(1) Resolution and RMS (Root Mean Squared). Some errors are simple to
quantify. For instance, the image pixel in a TM image represents the average radi-
ance from a 30- 30-m area on the ground. So, measurements within a TM scene
will only be accurate to within 30 m. Positional accuracy may be established by a
comparison with a standard datum giving an absolute uncertainty value. The RMS
(root mean squared) error is automatically calculated during image rectification.
This error can be improved while designating GCPs (Ground Control Points; see
Paragraph 5-17).
(2) Overall Accuracy. Overall accuracy can be established with "Ground
truth." Ground truth is site-specific and measures the accuracy by sampling a
number of areas throughout a scene. Overall accuracy of an image is then calcu-
lated by modeling the difference between the observed pixel DN signature and
known object on the ground.
(3) Error Matrices. Assessing classification error is more involved. Solv-
ing for this type of error requires a numerical statistical analysis. Some software
incorporates accuracy assessment within the classification function. For instance,
classification error assessment compares an image classification matrix with a
reference matrix. See Paragraph 5-20d(4) for information on classification accu-
racy. In this type of assessment, the reference data are assumed correct. Pixels are
assessed in terms of their mistaken inclusion or exclusion from an object class;
this is known as commission and omission (see Congalton and Green, 1999). All
known error should be noted and included in any assessment. Review Congalton
and Green (1999) for further information on the practice of error assessment.
k. Presenting the Data. Once a visual and statistical evaluation has been per-
formed, the analysis must be presented in a manner that best communicates the
information needed. The information may be presented as a hardcopy printout of
the image or presented as a map (Figure 5-40). The information may also be dis-
played as a statistical database, which includes data tables and graphs. Knowledge
of GIS, cartography, and spatial analysis is helpful in choosing and executing the
manner in which the data will be presented. For instance, a number of GIS soft-
ware programs are capable of displaying the image in a map format with a linked
data set. Be sure to keep in mind the final product needed by the client.
5-58