a. In two new viewers, open the baseline and change
tasseled-cap images on which the POM images were
i. These images are the tasseled-cap images listed
in the “Apply to file:” lines of the PROBDIFF
control file (see section IV, step 2.c.v. above).
b. In two additional viewers, open the predifferencing
i. These are the nine-layer images corresponding to
the POM in each of the physiognomic classes.
c. In a fifth viewer, open the POM difference image
d. In any of the viewers, select “View/Tile Viewers” to
better distribute all five viewers on the screen.
e. Geolink all viewers to the POM difference image.
i. Right-click in the POM difference image, select
Geo Link/Unlink, and then select another viewer.
Repeat for all four other viewers.
f. Start a cursor using the “+” icon.
g. Move the cursor around and ensure that the patterns
in the difference image make sense.
h. How to interpret the images:
i. There are nine bands in these POM images,
only three of which are shown. Depending on
the configuration of Imagine, the three shown
initially may differ, but are typically bands
4, 3 and 2 for Red, Green, and Blue on the
screen, respectively. Each band corresponds to a
physiognomic class, with band 1 corresponding
to the first physiognomic class. The order of
the classes is the same as the order they were
entered in the covariance matrix file passed to
the PROBDIFF program.
(1) The choice of which bands are displayed can
be changed using Raster/Band Combinations
in the viewer. There, the bands associated
with each color gun (red, green, and blue) on
the screen can be chosen.
(2) No-change: If an area has not changed in
any of the three bands currently displayed
in the viewer, it will appear grey because
all three bands are zero. Grey is roughly
midway between the negative and positive
ranges of the changes seen across the image.
(3) Change: Colors other than grey involve
some change in the combined probabilities
of the three classes being displayed.
Because the color reflects potential positive
and negative changes in each of the three
displayed classes as well as no-change, the
exact interpretation requires some careful
(a) Consider the case where classes 1, 2,
and 3 are represented by the red, green,
and blue color guns on the screens.
(b) If a pixel shows up red, it could be
the result of two distinct processes.
The more intuitive process is where
class 1 has increased in POM, and
classes 2 and 3 are unchanged. The
less intuitive process is where classes 2
and 3 have decreased dramatically, and
class 1 has remained the same. Thus,
the color combination represents the
relative change among the three classes
(4) It is safest to use the colors as a preliminary
indicator of change, and then to use the
cursor (crosshairs) to query individual pixels
in a changed area to confirm the direction of
change (positive or negative) of each class.
If the difference image appears to have many geometric
artifacts, it may be that the map information was entered
incorrectly, or that the registration of the images was
(5) In the case of the latter problem, it is
necessary to confirm that the terrain
correction in the geometric registration was
done correctly, and that the tie points were
located correctly. See SOP 2, section VI on
ii. If there appear to be many false positives—areas
where change is claimed in the difference image
but where there is no apparent cause in the
source tasseled-cap images—then the patterns of
the false positives must be examined to develop
hypotheses about what might be going wrong.
(1) Note that low probability changes in class
membership are often due to normal noise in
illumination, processing, etc. False positives
should be considered as those changes
where a high probability (>50) of change
is indicated, but where there is no obvious
change in the parent images.
(2) If false positives appear to have no
2 Protocol for Landsat-Based Monitoring of Landscape Dynamics at North Coast and Cascades Network Parks
relationship to the patterns on the actual
landscape, it may be that the radiometric
normalization has not been achieved
appropriately. Go back to the SOP 2 section
VIII on radiometric normalization and check
the output images to make sure that the
normalization has not failed.
(3) If false positives are restricted to particular
features on the landscape, first check to
determine if those features are always
located in one of the physiognomic classes.
If so, first check to make sure that the
covariance matrices of the classes were
entered correctly into the covariance matrix
comma-delimited file (see section IV, step 1
The steps in this section need to be repeated in their
entirety for the other aspect class.
The difference POM images at this point can be used in
one of three ways. First, they can be used directly as visual
estimates of the continuous-variable probability of change
across all classes. This directly fulfills one of the objectives
of the protocol, and makes this image a final product (see
SOP 5). Secondly, these nine-layer probability-of-change
images can be masked to only include areas where a certain
threshold of change has occurred, to simplify viewing and
interpretation. Finally, these continuous-variable images can
be categorized according to rules deemed as useful for an
individual park’s needs. This will be described in section VIII
VI. Merging Aspects
Sections II through V must be done separately for each
of the two aspect classes (SE and NW). Once this has been
done, the POM difference images should be merged for further
In Imagine’s main icon panel, click on the Interpreter
button, and select “Utilities/Operators.” The operator window
will popup. Enter the SE aspect image as Input File #1,
the NW aspect image as Input File #2. Retain the “+” as
the operator. This will add the two POM difference images
together on a pixel-wise basis. Because the two images have
zeros where their aspect class is not present, this approach
is a simple approach to merge the two aspect classes. Name
the output according to the conventions in SOP 5 Data
VII. Using Pacific Meridian Resources
Data to Describe Physiognomic Class
The physiognomic classes created in section IV above
were based on a direct understanding of the physical basis
for the distribution of reflectances in spectral space. Broadly
speaking, regions of spectral space describe vegetation
physiognomic properties anywhere in the world. While
changes in these classes can often be used to describe where
disturbance is occurring and generally how that disturbance
affected the vegetation, the physiognomic classes contain no
information on park-specific vegetation-community types that
In the parks of the NCCN, field data were recorded
for the purposes of Landsat-based mapping by the private
consulting company Pacific Merdian Resources (PMR) in
the mid-1990s. While these field data do not cover enough
of the spectral space to be used as the basis for statistical
class building, they can be used as a first approximation of
the park-specific vegetation within each of the physiognomic
classes. This step only need be taken during the startup phase
of monitoring, when the image from the era near collection of
ground data from PMR is being processed.
Using the PMR field data to describe the physiognomic
classes consists of three steps. First, the database with the
field data must be summarized to the plot level with summary
statistics that are desirable to the user. An example is
described below, but any statistics could be used. The key
is simply that each plot with a unique geographic location
have a set of numeric descriptors. This is done in Microsoft©
Access, or could be done in any similar database-software
package. The second step is to extract the physiognomic class
associated with each plot. This is done in Imagine using the
geographic coordinates in the database. Finally, the plots are
grouped by physiognomic class and summarized to provide a
description of that class.
1. Summarizing PMR data in the database.
a. This can be done to any level of detail desired.
Below, we provide a sketch of the process.
b. The PMR data were provided to the authors
in the form of an MS Access database called
“PMRPlotData.” Each tree or each cover component
is stored as a separate entry. The goal is to collapse
all tree observations and cover observations into a
single entry per plot.
i. A query with all plots and measures for the
NCCN existed in that database. From it, a
simple query was used to develop a new table
with the plots and measures for a single park.
SOP . Physiognomic Change Detection
Figure 1. Pixel to Table window.
(1) Record cleaning involved filling in values in
the FIXED column for trees where no entry
was made. The FIXED value was inferred
from the DBH of the observed tree.
(a) if DBH is between 4 and 18, assign 40.
(b) if DBH is between 18 and 36, assign 10.
(c) if DBH is >36, assign 5.
(2) A select query was then used to extract only
the live-tree observations.
(3) Conifer and broadleaf components were
(a) The broadleaf and conifer species were
assigned manually in the species table
of the database, and linked to the live-
(b) This was then linked to the result of
the query of live tree observations (step
(c) A query could then be made on conifer
and hardwood designation. In this
query, the DBH was converted to area
per acre using the expansion factor (the
FIXED value) in the formula: [exp_fact
(i) A new column with area per acre
was created in the query.
(d) The results were two queries, one for
broadleaf and one for conifer, where
each tree’s DBH was scaled to a plot-
level basal area.
(e) Then a crosstab query was used to add
up the total basal area by broadleaf and
conifer per plot.
ii. For all components where the COMPCODE was
not LT (live tree), a crosstab query was used to
add the cover percentages by plot number.
iii. This was combined with the conifer and
broadleaf tree components to result in a single
table that had all plot coordinates from the
original plot file, along with the broadleaf and
conifer basal area and the percentages of the
2. Extracting these points in Imagine.
a. Copy the X,Y coordinates of the plots into a separate
i. In the database program, select all of the plot
X and Y coordinates. In Access, the column
headers X_coord and Y_coord are selected, and
the Edit/Copy function selected.
(1) NOTE: It is critical that these coordinates
be listed in the same coordinate system as
the image (i.e. NAD83, if standard NCCN
conventions are used). The images in the
PMR database are in NAD83 already, so this
should not be an issue.
ii. Paste these into an empty Excel spreadsheet.
iii. Remove any leading text rows, so that only two
columns have data.
iv. Save this as a comma-delimited file.
v. In the Windows Explorer, change the extension
of this file to .dat (this is so that Imagine will
b. In Imagine’s main console, select “Utilities/Pixels to
i. The Pixel to Table window will pop up (fig. 19).
ii. The Input Image is the nine-class physiognomic
image created by recoding the 25-class image in
section III above.
(1) Once the file is located and shows in the
Input File field, the “Add” button must be
iii. Under “Type of Criteria:” select “Point File,” and
then find the “.dat” file just created.
iv. The Output File (ending in .asc) should be placed
in this park’s folder.
v. Click on OK.
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