Geoscience Reference
In-Depth Information
ground measurements, are used as the reference data set. As classification schemes get more
complex, more variation in human interpretation is introduced. Other factors beyond just variation
in interpretation are important also.
In order to deal with ambiguity/variation in remotely sensed maps, Gopal and Woodcock (1994)
proposed the use of fuzzy sets to “allow for explicit recognition of the possibility that ambiguity
might exist regarding the appropriate map label.” In such an approach, it is recognized that instead
of a simple system of correct (agreement) and incorrect (disagreement) there can be a variety of
responses, such as absolutely right, good answer, acceptable, understandable but wrong, and
absolutely wrong. This approach deals well with the ambiguity issue. However, the results are not
presented in a standard error matrix format. Therefore, Congalton and Green (1999) and Green
and Congalton (2003) presented a fuzzy assessment methodology that not only deals with varia-
tion/ambiguity, but also allows for the results of the assessment to be presented in an error matrix.
1.3.5
Error Budget Analysis
Over the last 25 years many papers have been written about the quantification of error associated
with remotely sensed and other spatial data (Congalton and Green, 1999). As documented in this
chapter, our ability to quantify the total error in a spatial data set has developed substantially.
However, little has been done to partition this error into its component parts and construct an error
budget. Without this division into parts, it is not possible to evaluate or analyze the impact a specific
error has on the entire mapping project. Therefore, it is not possible to determine which components
contribute the most errors or which are most easily corrected. Some early work in this area was
demonstrated in a paper by Lunetta et al. (1991) and resulted in an often-cited diagram that lists
the sources of error accumulating throughout a remote sensing project.
It should be noted that each of the major error sources adds to the total error budget separately,
and/or through a mixing process. It is no longer sufficient to always just evaluate the total error.
For many applications there is a definite need to identify and understand (1) error sources and (2)
the appropriate mechanisms for controlling, reducing, and reporting errors. Perhaps the simplest
way to begin to look at an error budget is to create a special error budget analysis table (Congalton
and Green, 1999). This table is generated, column-by-column, beginning with a listing of the
possible sources of error for the project. Once the various components that comprise the total error
are listed, then each component can be assessed to determine its contribution to the overall error.
Next, our ability to deal with this error is evaluated. It should be noted that some errors may be
very large but are easy to correct while others may be rather small. Finally, an error index can be
created directly by multiplying the error contribution potential by the error control difficulty.
Combining these two factors allows one to establish priorities for best dealing with individual errors
within a mapping project. A template to be used to conduct just such an error budget analysis is
presented in Table 1.2.
1.3.6
Change Detection Accuracy Assessment
Much has recently been written in the literature about change detection (Lunetta and Elvidge,
1998; Khorram et al., 1999). This technique is an extremely popular and powerful use of remotely
sensed data. Assessing the accuracy of a change detection analysis has all the issues, complications,
and difficulties of a single date assessment plus many additional, unique problems. For example,
how does one obtain information on reference data for images/maps from the past? Likewise, how
can one sample enough areas that will change in the future to generate a statistically valid assess-
ment? Most of the studies on change detection do not present any quantitative results. Without the
desired accuracy assessment, it is difficult to determine which change detection methods are best
and should be applied to future projects.
