g. Click on “Color Scheme Options.”
i. Select “Approximate True Color” and leave the
band combinations as default.
h. In the Number of Classes field, type “25”.
i. This will create 25 classes in this spectral space.
The choice of 25 classes is somewhat arbitrary,
since these classes will be collapsed later,
but experience shows that it is a large enough
number to separate real classes on the ground
and yet small enough to interpret and integrate
i. In the Maximum Iterations field, type in “20”.
j. Leave the convergence threshold at 0.950.
k. Click on the AOI button.
i. In the window that appears, select the button for
(1) The file-field dialog will become active.
(2) Navigate to the NW aspects AOI that was
created in SOP 2, section X.
l. Click on OK button.
i. The progress meter will run as Imagine processes the
image. It will take several minutes.
2. Repeat this process with the baseline image, but use the
SE aspects AOI instead of the NW aspects AOI, and name
the output file accordingly. All other options are the
3. The result of this process will be two images, each with
25 spectral classes.
III. Collapsing Spectral Classes Into
Simple Physiognomic Classes
The 25 spectral classes in section I must be compressed
down to 9 simple physiognomic types (table 2).
These types occupy distinct regions of tasseled-cap
spectral space. To collapse the 25 classes into the simplified
classes, the tasseled-cap spectral space must be visualized,
nearby classes noted, and then classes recoded into
Conduct the following steps for both aspect classes. The
examples below show the NW aspects case.
Figure 2. Create Feature Space Images window.
Table 2. Physiognomic classes for the North Coast and
Cascades Network Parks.
Snow and ice
1. Viewing spectral space of NW aspects using feature space
a. From the main icon panel in Imagine, click the
Classification button, and then select “Feature space
image” from the list of options that appears.
i. The “Create Feature Space Images” window will
pop up (fig. 2).
b. Input raster layer field:
i. Use the open-folder icon to navigate to the
baseline tasseled-cap image that has been
clipped to the study area and to NW aspects.
(1) Recall that the baseline image is the one at
the beginning of the desired change interval.
(2) This image was created in section X of
0 Protocol for Landsat-Based Monitoring of Landscape Dynamics at North Coast and Cascades Network Parks
c. Output root name:
i. Imagine will produce a root name in the default output directory, which likely is not where it belongs. Thus, it is
necessary to specifically indicate the directory to which the output feature space image should be placed.
(1) Navigate to a desired output directory.
Follow the conventions of SOP 5 Data
Management for placement of the output
(2) Click on OK button.
d. Click on the AOI button
i. Select “AOI File.”
(1) The “Select the AOI file” field will become
ii. Navigate to the NW aspects AOI created in
SOP 2 section X.
e. Left-click and drag down through the three rows in
the table at the bottom of the Create Feature Space
Images window. All three rows (label 1,2,3) should
be highlighted in yellow.
f. Click on the OK button.
g. Examine the image outputs. For example, the
feature space image for bands 1 and 2 – tasseled-cap
brightness and greenness – are shown in figure 3.
h. In figure 3, brightness (band 1) values are arranged
Figure . An example of a greenness/brightness feature space
from left to right, and greenness values are arranged
from bottom to top. Color intensity corresponds
to pixel count—red indicates the greatest number
of pixels, magenta the fewest. Snow and ice are
relatively rare and are high in brightness but low
in greenness, and thus occupy a sparse population
spread out in a tail from left to right in figure 3.
Most of the pixels in this image (taken from the
area around Mt. Rainier National Park) are forested
and fall along a diagonal running from lower left to
upper center in figure 3.
2. View the 25 spectral classes in this spectral space.
a. From the main icon panel in Imagine, click the
Figure . Create thematic feature space images window.
“Classification” button, then select “Feature Space Thematic” from the list that appears.
i. The “Create Thematic Feature Space Images” window will popup (fig. 4).
b. Thematic Input File:
i. Use the folder icon to navigate to the unsupervised classification image for the NW aspects that was created in
section II above.
c. Feature Space Files:
i. Use the folder icon to navigate to the brightness-greenness feature space image created in step 1 above in this
(1) The brightness-greenness feature space image is noted by the phrase “_1_2.fsp.img” in the file name.
SOP . Physiognomic Change Detection 1
Figure . Feature-space thematic image corresponding to the
feature space image in
Figure 6. Approximate locations of physiognomic cover
classes in the brightness/greenness space.
d. Output Root Name:
i. Use the folder icon to open the file navigation dialog.
ii. As in step 1c above, it is useful to copy the default “output root name” onto the clipboard before navigating to the
desired output folder.
iii. Navigate to the same folder in which the feature
space images were placed.
iv. Follow naming and file location conventions in
SOP 5 Data Management.
e. In the Feature Space Files field, navigate and select
the greenness-wetness feature space image
i. This is the feature space image with “_2_3.fsp.
img” in the filename.
ii. When this is selected, a second row of filenames
should appear in the table at the bottom of the
(1) This second set of files will be placed in
the correct directory, assuming it has been
changed in step 2.d. above prior to this step.
f. Click on the AOI button and navigate to the AOI for
NW aspects. Click OK.
g. In the “Create Thematic Feature Space images”
window, select both rows by clicking and dragging the
numbers on the left-hand side of the table.
h. Click OK.
i. View the results.
i. In a viewer, open the feature space
thematic image. Figure 5 shows an
example of the output for the same image
shown in the feature-space image in
j. The colors in figure 5 correspond to the colors
of the classes in the 25-class unsupervised
classification image from section I.
k. To see which class numbers correspond to a
given color in the feature space thematic, use
the cursor utility.
i. Click the plus sign icon in the viewer (just
to the left of the hammer symbol; fig. 5).
The cursor utility window will pop up.
ii. A crosshair will appear in the viewer.
Click and drag at the center of that
crosshair to move it through the feature
space. The class number will be reported
in the “Pixel Value” column of the cursor
2 Protocol for Landsat-Based Monitoring of Landscape Dynamics at North Coast and Cascades Network Parks
Figure . Approximate locations of physiognomic cover classes in
the greenness/wetness space (greenness on X axis)
3. Identifying classes to combine.
a. Each of the nine physiognomic classes listed
in table 2 will be composed of several spectral
classes. A table of the nine classes and the spectral
classes that make them up must be determined by
the interpreter. Two tools can be brought to bear on
i. The shape of the brightness-greenness feature
space in figures 3 and 5 is typical. The
physiognomic classes can be identified based
on their approximate locations in this feature
space. Indeed, the relative stability of these
broad physiognomic classes across different
systems is one of the key reasons that the
tasseled-cap transformation was chosen as the
base for change detection. Figure 6 shows the
approximate regions of these physiognomic
classes in brightness/greenness space, while
figure 7 shows the same in greenness/wetness
space. These general locations serve as the starting point for determining how to aggregate spectral classes.
ii. In addition to noting the position in spectral space of the physiognomic classes, it also is useful to use direct
interpretation of the imagery to aid in labeling the spectral classes.
(1) Load the image of the 25 spectral classes into a new Imagine viewer.
(2) Select Raster/Attributes.
(a) A Raster Attribute window with a table of the 25 spectral classes will appear, with a column for the color
of each class in the image.
(b) To see where on the landscape a given class falls, use the left-mouse button to select the row of the given
class, and then right-click on the color patch of that row to change the class colors. Pick a new color that
is unlike most of those already on the image–often white or black are useful.
(c) Examine where on the landscape the class falls to aid in interpreting the class. Take into consideration
known elevation patterns, as well as a general knowledge of the cover types in the system. See the section
in SOP 4 Validation on S2S interpretation (section VI) for tips on interpreting directly from tasseled-cap
(d) When done with viewing the class, select “Edit/Undo last edit” to return the class to its original colors.
(i) When exiting the raster attribute editor, do not save changes.
(3) If desired, link the 25-class image with the original tasseled-cap image from which it was formed to aid in
(a) In a separate viewer, load the tasseled-cap image for the NW aspects.
(b) Right-click anywhere in the classified-image viewer, and select “Geo. Link/Unlink.”
(i) Left-click in the viewer with the tasseled-cap image.
(c) Open a cursor window in the classified image, if one is not open already.
(i) Select the “+” symbol in the viewer with the 25-class spectral image.
(ii) Click the center of the crosshairs in the viewer to move the crosshairs around, and note spectral
patterns in the tasseled-cap image to aid in interpretation of the imagery.
SOP . Physiognomic Change Detection
Figure . Recode window, where the 25-spectral class image is
recoded into 9 physiognomic classes.
Figure . An example of record
keeping for linking the 25-class spectral
image with 9 physiognomic classes.
(d) Use the interpretation aids in the supplied “spectral library”
that the authors have provided in an appendix to see examples
of different cover classes on airphotos, in the field, and on
iii. In practice, class calls will not always be clear cut, especially for the
physiognomic classes that involve mixtures of types. While good class
calls will improve the quality of later change detection, the method
does not assume that the physiognomic classes will be perfect. The
classes will be converted later into a statistical representation based on
these classes, adjusting the boundaries of the classes, and that statistical
representation will further be described by ground and/or airphoto
interpretation data to characterize within-class variability. Finally, the
method works on directional variation in spectral change and assumes
overlap of classes. Therefore, the precise definitions of the classes are
less critical than in a project where the classified image was the single
outcome. Nevertheless, the definitions of the class groupings may affect
outcomes, and it is assumed that lessons learned about class grouping
for each park eventually will need to be incorporated in a later revision
of this Protocol. See the Narrative section for a broader discussion of
b. Keep track of the link between the original spectral class number (1 through 25) and the new physiognomic classes (1
i. This can be done in a simple word processing document, or in a spreadsheet. An example in Excel is shown in
4. Merging classes.
a. Once a physiognomic class is associated with each of the 25 spectral classes, the 25-class image can be recoded into the
9 physiognomic classes (table 2).
i. Open the 25-class image in a Viewer, if not open already.
(1) NOTE: This is the geographic image of the 25 classes, not the feature space image of the type shown in
Protocol for Landsat-Based Monitoring of Landscape Dynamics at North Coast and Cascades Network Parks
Documents you may be interested
Documents you may be interested