Geoscience Reference
In-Depth Information
All this information should be represented in a well-defined, mutually exclusive, and totally
exhaustive classification scheme. However, in many instances, it would be useful to know more
about the boundaries or edges of different LC types. For example, when performing change detection
(i.e., looking for changes over time), it is important to know whether real change exists and that
change is going to occur along the boundaries between cover types. Therefore, it is important that
more research and study be undertaken to better understand this boundary and edge issue.
1.3.3
Reference Data Collection
Reference data are typically assumed to be correct and are used to evaluate the results of the
LC mapping. If the reference data are wrong, then the LC map will be unfairly judged. If the
reference data are inefficiently collected, then the project may suffer from unnecessarily high costs
or an insufficient number of samples to properly evaluate the results. Reference data are a critical,
very expensive, and yet often overlooked component of any spatial analysis. For example, aerial
photographic interpretation is often used as reference data for assessing a LC map generated from
digital satellite imagery. The photographic interpretation is assumed correct because it often has
greater spatial resolution than the satellite imagery and because photogrammetry has become a
time-honored skill that is accepted as accurate. However, photographic interpretation is subjective
and can be significantly wrong. If the interpretation is wrong, then the results of the accuracy
assessment could indicate that the satellite-based map is of poor accuracy when actually it is the
reference data that are inappropriate.
There are numerous examples in the literature documenting problems with collecting improper
or inadequate reference data. One especially insidious problem with reference data collection is
the size of the sample area in which the reference data are collected. Clearly, it is important to
collect reference information that is representative of the mapped area. In other words, if the map
is generated with remotely sensed data that have 30-
¥
30-m pixels, it does not make sense to
collect reference data for a 5-m
area. A current example of this situation is the use of the Forest
Inventory and Analysis (FIA) plots collected by the U.S. Forest Service across the country. It is
important that these inventory plots be large enough to provide valid reference data.
The opposite situation must also be carefully monitored. For example, it is not appropriate to
assess the accuracy of a 1-ha mapping unit with 5-ha reference data. The reference data must be
collected with the pixel size and/or the minimum mapping unit of the map in mind. Additionally,
the same exact classification scheme using the same exact rules must be used to label the reference
data and to generate the map. Otherwise, errors will be introduced by classification scheme
(definitional) differences and the error matrix created will not be indicative of the true accuracy of
the map. Using well-designed field forms that step the collector through the process can be very
helpful in ensuring that the reference data are collected at the proper scale and with the same or
appropriate classification scheme to accurately assess the map.
2
1.3.4
Beyond the Error Matrix: Fuzzy Assessment
As remote sensing projects have grown in complexity, so have the associated classification
schemes. The classification scheme then becomes a very important factor influencing the accuracy
of the entire project. Recently, papers have appeared in the literature that point out some of the
limitations of using only an error matrix with a complex classification scheme. A paper by Congalton
and Green (1993) recommends the error matrix as a jumping-off point for identifying sources of
confusion and not just error in the remotely sensed classification. For example, the variation in
human interpretation can have a significant impact on what is considered correct and what is not.
As previously mentioned, if photographic interpretation is used as the reference data in an accuracy
assessment and the interpretation is not completely correct, then the results of the accuracy assess-
ment will be very misleading. The same statements are true if ground observations, as opposed to
