Document Actions
CCA Update Data Products
In today's rapidly changing world the moment a feature is mapped it is potentially out of date. As time passes there is a greater probability the feature has changed. With traditional methodology, the only way to provide updated land cover information is to create entirely new data from more-recent imagery.
Such an effort is not only cost-prohibitive, it is also unlikely to keep pace with the rapid changes in some parts of the world. While some automated update methods are available, the results often fail to meet users' minimum mapping unit and accuracy requirements. MDA's patented Cross Correlation Analysis (CCA) technique addresses both cost and accuracy concerns: it uses the synoptic strengths of the satellite image in conjunction with the accuracy of GeoCover LC data to locate change.
Description:
Using CCA for land cover updating provides a more cost- and time-efficient alternative to complete re-classification. CCA allows first the finding of the areas of change, and then the interpretation and updating of just those areas of change so that no areas are being re-interpreted unnecessarily. The change detection approach minimizes the introduction of new error into the data. A complete re-mapping may correct some errors in the initial mapping but will also introduce new errors as some features will be mapped differently.
CCA is able to overcome many limitations of conventional change detection methods. It performs well regardless of seasonal differences because it uses former class boundaries summarized with new class signatures to determine the relationship between pixel values and a feature class. There is no reliance on direct pixel value comparison between the different scenes. This approach isolates change, or reduces change detection artifacts better than conventional change detection procedures. Spatial, spectral, radiometric, and temporal resolutions are compensated for by this procedure.
- CCA overcomes many of the problems associated with seasonal variation in imagery, and the need to select images on the same anniversary date is minimized (however, it is recommended that the scenes be selected during the growing season)
- The change file is extremely useful for a variety of environmental and security applications
- The change file can be easily manipulated to depict land cover from either time period (earlier-date or later-date) or for the change from one to the other
- Distributed on CD_ROM
CCA can detect change in any pair of earlier-date and later-date images. At present, CCA is performed to detect change between earlier-date CY 1990 Landsat 4/5 images and later-date CY 2000 Landsat 7 images. A brief description of MDA's CCA land cover updating process follows.
For more on GeoCover LC data products, select an individual product to the right. Or you may read more about the preparation of these products.
Change Analysis
CCA is used to identify changes between the earlier-date and later-date scenes. CCA correlates the earlier-date ungrouped thematic file (240 spectral classes) to the 6 MSI bands of the later-date scene, and correlates the earlier-date 6 MSI bands to an ungrouped thematic file (240 spectral classes) for the later-date scene. Each correlation generates a separate Z-statistics file. The maximum value of each Z-statistics file is used. This analysis results in a thematic image of Z-statistics with values ranging from 0 to 60,000. The higher the value, the more likely the land cover has changed. The Z-score is a measure of the distance exhibited by an individual member of a population from the central tendency of the population. Population is stratified however, usually spectrally by clusters.
Semi-Automated Land Cover Updating
Trained image interpreters evaluate CCA results to determine if the change that has been detected is legitimate and to verify the new land cover category for valid areas of change. Updates are applied to GeoCover LC 1990 data, creating GeoCover LC CCA Update 2000 data. The entire suite of GeoCover LC data products are then derived from the updated data.
Creation of Bivariate Land Cover Classification
The edited change file derived from the newer TM data and the land cover file from the older TM data are merged to create a land cover change product, identifying the earlier-date and later-date land cover for each pixel.
