PREFORMING AN IMAGE CLASSIFICATION For this exercise I downloaded an ortho image from the USGS’s National Map: Viewer & Download Platform (http://nationalmap.gov/viewer.html). The image that I download is of Clarks Reservation State Park and the surrounding area, it is the NAIP image. The reason I chose this type of area is because it has many classes of land use. It has a quarry, conifer forest, deciduous forest, urban areas, wet lands, agriculture land, open fields and open water. In the United States there are land use maps ready to download. The mentioned website also has the land use files for this area that I downloaded as well. However these land use maps can be out of date or be too broad of an area. What we are going to do is perform an image classification for land use for Clar ks Reservation and the surrounding area. Step 1: Determining what classes you want to group with. For this exercise I am going to use class groups that are normally used in Land Use maps, I will be using a broader group than the Land Use maps. Were the L and Use map I downloaded straight for the USGS site has classes like “Developed – open field”, “Developed –light”, “Developed –medium”, and “Developed –heavy” I will group all of these simply under “Developed”. The Classes I will be using for this exercise will be: “Developed Land”: Thisclass will include “Developed –open field”, “Developed –light”, “Developed –medium”, and “Developed –heavy” which are urban areas. This includes roads, houses, etc... “Barren Land”: This class will include areas of construction, bare earth, and quarries. “Deciduous Forest”: This class will include forest areas where a majority of the trees in that area shed their leaves in the fall. “Conifer Forest”: This class will includeforest areas where a majority of the trees in the area have needles all year long. “Open Water”: This class will include bodies of water, normally deep and year round. “Wetlands”: This class will include any wetland areas.“Fields”: This class will incl ude shrub and open fields. Some of these may be agricultural fields that have not been used for a few years. “Agriculture”: This class will include active farming and pasture lands. Step 2: Add your Ortho Image and optionally the Land Use raster. Before you add anything make sure your coordinate system is set up. For this exercise I will just add a US Counties Outline to my map. This will set the coordinate system for all succeeding raster images. Once a coordinate system is set add your raster image. I will also be adding a USGS supplied land use map. The Land Use raster will use codes for what it is showing. The codes for the area that I am using are 11, 21, 22, 23, 24, 31, 41, 42, 43, 52, 71, 81, 82, 90, and 95. Some of these codes I will group together to make things more simple. Below each class it will tell whe ther it will be grouped with others. For more information about these and other classes look at the raster image’s metadata.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
For this exercise I downloaded an ortho image from the USGS’s National Map: Viewer & Download
Platform (http://nationalmap.gov/viewer.html). The image that I download is of Clarks Reservation State
Park and the surrounding area, it is the NAIP image. The reason I chose this type of area is because it has
many classes of land use. It has a quarry, conifer forest, deciduous forest, urban areas, wetlands,agriculture land, open fields and open water. In the United States there are land use maps ready to
download. The mentioned website also has the land use files for this area that I downloaded as well.
However these land use maps can be out of date or be too broad of an area. What we are going to do is
perform an image classification for land use for Clarks Reservation and the surrounding area.
Step 1: Determining what classes you want to group with.
For this exercise I am going to use class groups that are normally used in Land Use maps, I will
be using a broader group than the Land Use maps. Were the Land Use map I downloaded straight for
the USGS site has classes like “Developed – open field”, “Developed – light”, “Developed – medium”,
and “Developed – heavy” I will group all of these simply under “Developed”.
The Classes I will be using for this exercise will be:
“Developed Land”: This class will include “Developed – open field”, “Developed – light”,
“Developed – medium”, and “Developed – heavy” which are urban areas. This includes roads,
houses, etc...
“Barren Land”: This class will include areas of construction, bare earth, and quarries.
“Deciduous Forest”: This class will include forest areas where a majority of the trees in that area
shed their leaves in the fall.
“Conifer Forest”: This class will include forest areas where a majority of the trees in the areahave needles all year long.
“Open Water”: This class will include bodies of water, normally deep and year round.
“Wetlands”: This class will include any wetland areas.
“Fields”: This class will include shrub and open fields. Some of these may be agricultural fields
that have not been used for a few years.
“Agriculture”: This class will include active farming and pasture lands.
Step 2: Add your Ortho Image and optionally the Land Use raster.
Before you add anything make sure your coordinate system is set up. For this exercise I will just
add a US Counties Outline to my map. This will set the coordinate system for all succeeding raster
images. Once a coordinate system is set add your raster image. I will also be adding a USGS supplied
land use map. The Land Use raster will use codes for what it is showing. The codes for the area that I am
using are 11, 21, 22, 23, 24, 31, 41, 42, 43, 52, 71, 81, 82, 90, and 95. Some of these codes I will group
together to make things more simple. Below each class it will tell whether it will be grouped with others.
For more information about these and other classes look at the raster image’s metadata.
You will also want to go to the Display tab and have the transparency set to 50-60%. This will
show us how accurate the Land Use map is according to the ortho image. Below Figure 3 shows that the
areas in red boxes no longer are for agricultural use. There can be some other small inaccuracies.
Figure 3: The red boxes show large areas that no longer match their land use code.
Step 3: Figuring out what is what.
We will start with the most difficult to identify first then move on to the easier to identify
classes. The biggest difficulty can be with separating deciduous and conifer forests. In the late spring,
summer and early fall both these forests will look green. By using the Principal Component Classification
we will be able to see the difference between conifer and deciduous trees. There are other ways of
doing this that will be shown as well.
You will want to select the Image Classification toolbar for this. You will go to Customize, go toToolbars and then go down the list until you see Image Classification (the toolbars are in alphabetic
order).
Figure 2: Land Use with grouped values and decent color scheme.
The other way of finding out the forest types is to have another ortho image from the early
spring when all the deciduous trees are without leaves. Figure 10 shows another ortho image from early
spring over the Clarks Ortho I want to classify, the conifer forest are in the red boxes and the deciduous
forest is in the yellow boxes. Now that we have a way of separating the different kinds of forest we can
move on to the easier identifiable classes.
Figure 10: An ortho image of the area during the early spring when the deciduous trees have no leaves. The red boxes are the
conifer forests and the yellow boxes are the deciduous forests.
For developed land you will want to look at the ortho image and find houses, roads, sporting
fields, and other urban areas. Some land use maps break these up further but we will keep it simple and
have all developed land in one class. Next you will want to identify which areas is open water, for these
you will look for lakes, ponds, and rivers. For barren land you will want to look for bare earth, this can be
bedrock outcrops and quarries. For wetlands you will want to look for areas of shallow water with lots of
vegetation mixed in with it, these areas are often found near lakes. For agriculture you will want to look
for farmland, these plots are normally rectangular in shape. Fields are found by looking for grasslands
and areas with small scrubs and a few small trees. We can create our samples now that we know what
we are looking for.
Step 4: Creating class samples.
Now that we know what will make up our classes we will make samples so that the Interactive
Supervised Classification tool can make a raster showing us land use for the entire raster image. We will
be using the Image Classification Toolbar for this step as well. If you closed it you will need to open it
again, instructions how to do so are in the beginning of step 3. At this point you may have multiple
raster images added to your map, you will want to make sure the raster image you want to classify isshown in the Image Classification Toolbar display window (red box in Figure 11).
Figure 11: Image Classification Toolbar with desired raster selected.
When you have samples for all the classes that you want you will want to click on the
Classification button on the Image Classification Toolbar and select Interactive Supervised Classification
from the drop down menu (Figure 22).
Figure 22: Selecting the Interactive Supervised Classification tool from the Image Classification Toolbar.
Once you click on the Interactive Supervised Classification tool it will begin the tool, no tool
window will open up. When the tool is finished you will get a raster similar to the one in Figure 23. There
are bound to be some errors in the classification, the main cause of these errors are shadows. Another
source of error is the lack of enough samples. If your classification does not meet your standards youcan look and see where those errors are and go back and make sure those are areas to sample. The
more you do this the more accurate your classification will be. As you can see this map in Figure 23 is
much easier to look at than the USGS provided land use map in Figure 24.