Geography Reference
In-Depth Information
LANDSAT -MSS-June-1975
LANDSAT -TM-August-2007
P173-r034
P186-r034
P171-r035
P171-r036
P184-r035
P172-r035/Master
scene
P173-r035
P186-r035
P185-r035
P184-r036
P185-r036
P172-r037
Image by image processing
Subset each image on ERB-borders
No atmospheric correction/no atmospheric
parameters
Only relative radiometric
normalization/iMAD
Atmospheric correction/ATCOR-2 +
Relative radiometric normalization/iMAD
Training samples/
Spectral characteristics
Mosaic
Five general supervised classification/MLC
Subset on ERB-borders
Five general supervised classification/MLC
Resempling to (30m)
One thematic map of
classification results/pixel size
(30m)
Mosaicing the six images after
classification/pixel size (60m)
Geometric registration/image to image
Change detection products
Change detection statistics/change matrix
Thematic map/s of 20 possible change-types
Fig. 5.61 Flow chart of the post-classification change detection mapping approach that was
performed for the two remotely sensed datasets (TM-May-1987 and TM-May-2007)
5.13 Accuracy Assessment
The execution estimation of an applied classification approach is a complex pro-
cess, involving various criteria. Cihlar et al. ( 1998 ) suggested six criteria. These
are: ''accuracy; reproducibility; robustness; ability to fully use the information
content of the data; uniform applicability; and objectiveness''. These requirements
are difficult to satisfy using only one classification approach. The reason for this
relates to the different environmental settings and datasets used. The acceptable
accuracy values are relative, determined generally by the users themselves
depending on the type of application. Accuracy values that are acceptable for
specific application may be unacceptable for others.
Generally, there are no dependable rules for determining the testing samples
that are required to evaluate the classification accuracy. However, there are some
useful suggestions, including those made by Fitzpatrick-Lins ( 1981 ). Another idea,
put forward by Congalton and Green ( 1999 ), is to use 50 testing samples as
minimum for each classified LULC-category. If the study area is larger than
1,000,000 ha, or if there are more than 12 classified categories, then there should
Search WWH ::




Custom Search