Many earth science disciplines require detailed analysis of changes to the earth's surface over a period of time. Examples include studies of deforestation, urban growth, desertification, glacier and ice pack change, agricultural land use and many others. Satellite data can be very useful to such studies. Often these changes are made on the basis of visual comparison of individual band files, or true color or false color images composited from such band file data. Another approach is to subtract one band image from another band image in order to obtain a quantitative measure for the degree of change.
PANCROMA™ offers a powerful multispectral tool for assessing such changes. Rather than using a single multispectral band, the Change Distance™ tool uses all available Landsat multispectral bands in order to determine the degree of change. This can often be more accurate than relying on a single band, because apriori knowledge of the most affected band is not necessary in order to detect the change. The utility considers each pixel in the scene as a vector in six-space, and computes the vector distance between the top of atmosphere (TOA) reflectance value of each pixel in the base and changed scene. Using the TOA reflectance provides a more accurate comparison between the two scenes than using raw digital numbers (DNs). Those pixels with near zero vector distances may be assumed to be unchanged, while those with larger vector distances are identified as changed.
Using the tool requires two multispectral Landsat data sets, one representing the base scene and one for the changed scene. Each set should include bands 1, 2, 3, 4, 5, and 7. Because a pixel-by-pixel comparison will be made between the two scenes, it is important that they are correctly co-registered. The registration requirements are exactly the same as those required for gap filling, and any and all of the PANCROMA™ tools that are used for preparing Landsat gap fill data can be used for change detection.
The first step in the process in to obtain two Landsat scenes with at least some overlapping area, i.e. two scenes with the same path and row number. Make sure that you download the metadata files with the Solar Elevation Angles and Acquisition Dates, as these will be needed in the process. We are going to use a very obvious feature to demonstrate change detection in this example: cloud cover. The base image subset area has very little cloud cover, while the changed target has a lot. If change detection works properly, we should expect the cloud cover to be highlighted as a very measurable change.
Base image and changed image with heavy cloud cover.
The next step is to co-register the bands. There are several ways to do this but the easiest is probably to use the six file subsetting method. This method will crop the images to the Maximum Common Extents coordinates and rescale the images if necessary so that the row and column counts will match exactly. (The Maximum Common Extents are the corner coordinates of the common area between the two scenes). The main proviso with this technique is band file labeling. Since the utility was designed for gap filling, you have to be careful to get your file naming conventions set so that your registered band files do not get confused.
PANCROMA™provides a file-naming tool in order to facilitate this if you choose to use it. To illustrate its use, suppose that we have two Landsat scenes as follows:
Start by opening the first three bands of Group 1 and the first three bands of Group 2, in order:
After the six band files are opened, select 'Subset Utilities' | 'Subset Images' | 'Subset Six Bands'. When the Subset Data Entry form becomes visible, check the 'Select Change Subset' check box and click on the 'First Group' radio button as shown in the image below.
Now select 'Enter'. The first six band files will be co-registered and displayed as shown in the following image.
First Data Group
Note that the captions in the image forms read: Group 1 Band 1, Group 1 Band2, etc. Now select 'File' | 'Save Subset Images' | 'Geotiff'. Select a base file name like 'changeTest' or the like. The first group of registered files will be saved. In this example, the file names will be:
Now select 'Close Graphics Window and Reset'. Repeat the process outlined above for the second group of six band files, entering the files in this order:
except this time when the Subset Data Entry form becomes visible, check the 'Select Change Subset' check box and click on the 'Second Group' radio button . Select 'Enter'. The next group of files will be displayed as shown below:
Second Data Group
Select 'File' | 'Save Subset Images' | 'Geotiff' again. Select the same base file name as before ('changeTest' in this example). The second group of registered files will be saved. In this example, the file names will be:
You now have as set of 12 co-registered band files for you scenes. Select 'Close Graphics Windows and Reset' and move on to the next step: change detection. To start the change detection analysis, select 'File' | 'Open' and open your 12 band files in this order:
Group 1 bands 1-7
Group 2 bands 1-7
In this example the order would be:
Now select 'Spectral Analysis' | 'Change Analysis' | 'Landsat' | '12 File Vector Distance'. The TOA Reflectance Data Form will appear. Enter the values for the Solar Elevation Angle and Acquisition Date for the first Landsat Scene as described in Section 51 into the form and select 'OK'. All 12 of your band files will be read into memory. The TOA form will again appear. Enter the Solar Elevation Angle and Acquisition Date for the second Landsat Scene and select 'OK'.
IMPORTANT NOTE: The satellite metadata will typically not be the same for two Landsat scenes even if the path and row numbers are the same. It is important to enter the correct values for each scene when prompted so that the TOA reflectances are computed accurately. Select 'OK'. The change detection plot will be computed and displayed. The plot along with the base image is shown below. (Note that you can switch views between the distance plot and the base image using the 'Switch Images' menu on the Color Distance Form.
Distance Change™ Plot and Base Image
Notice that the areas of the most dense cloud coverage are colored red, indicating the most pronounced change from the base image. Note the small blue hole in the cloud coverage denoted by the lower arrow. This coincides with one of the few clouds in the base image. This has been correctly identified as a region of lesser change.
This example, although depicting a trivial change, illustrates how the Change Distance™ utility can be used to identify much more subtle changes in Landsat scenes with great sensitivity