Change Detection

Wetlands Change

From the early 1990's, MDA has been developing techniques to identify wetland changes from raster imagery. The techniques use the U.S. Fish and Wildlife Service's National Wetland Inventory vector data as the historic base and more recent raster imagery (i.e., satellite imagery) as the data source for the most current information. Most of the techniques developed failed to identify adequately the wetland changes that had occurred. Only one technique, Cross-Correlation Analysis, was successful in identifying wetland changes. This technique was tested by the U.S. Fish and Wildlife Service on a study site in eastern Maryland and by the Environmental Protection Agency in conjunction with the U.S. Fish and Wildlife Service and the Natural Resource Conservation Service on a study site in eastern North Carolina. MDA's Cross-Correlation Analysis technique is the only known technique employing a wetland vector data base and raster images that has been able to successfully detect wetland changes.

Land Cover Updating

GeoCover Land Cover Updating Using Cross-Correlation Analysis

MDA has been tasked with mapping the world. The result of this task will be the ultimate land cover data set to date. It will be unprecedented due to its spatial resolution and accuracy. Unfortunately, it will also become obsolete as time changes the Earth's landscape. The nature of mapping is that only a generalization of a brief period of the landscape's history can be described. Time is as much a factor as space to geographers. This is why change detection was developed. When change detection is incorporated into the land cover mapping process, it prevents data from becoming obsolete. This is because change detection can be used to update the land cover data base. Also, when the change itself is analyzed, it adds more information to the land cover information. For example, change can be predicted, or trends described.

This project seeks to update the change in land cover derived in the GeoCover program. The methodology consists of two main steps: the detection of change in land cover, and the updating of that change. The change detection step is performed by using CCA while the automated updating uses Inverse CCA.

Cross Correlation Analysis (CCA) is a change detection technique developed by MDA specifically for the updating of land cover information. This project is a platform to demonstrate its effectiveness and efficiency in this role. It is not only a superior technique for the determination of change, but also for the automated interpretation of the change areas. CCA does not have the limitations of more typical change detection processes. The many resolution constraints of most change detections are not an issue with CCA. Seasonal, radiometric, and spectral discrepancies in data tested are not confounding factors in this analysis. The affects of seasonal differences are negligible.

The inverse CCA, or automated updating of the change area, theoretically reduces the human subjectivity in labeling the change areas. It labels by using the early date interpretation as training to label the later date. Theoretically, it labels the updates the same way the original interpreter would have labeled the features. In practice, however, the classes must be checked for discrepancies because there are some situations whereby the classes could be incorrectly labeled.

The bivariate analysis shows all of the necessary information for the analysis of change between two images. It is the third step in the CCA updating process after the CCA and inverse CCA. The CCA determines areas of change. The inverse CCA interprets that change. The bivariate analysis updates the image and aids in the analysis of the change. It is basically a contigency table. Also, it allows the analyzer to toggle between old and new date classifications to visually see the differences. One of the outputs of this procedure is the fully updated classified image.

The scope of the current land cover updating project is to update the GeoCover land cover product for the areas of Southwest Asia and Korea. As more and more of the world is mapped by the GeoCover program, the CCA update process will be applied behind it using current data. The updates can be applied indefinitely over the same area, continuously updating and thus producing time series analyses anywhere there is land cover data.

This process has been used for this purpose in two other projects, one for USFWS, and two for government mapping agencies. It has been tested and streamlined into a production environment to update large areas of land cover data. It is in the vision of MDA to use this process for continuous worldwide mapping of Landsat TM data indefinitely. As the process becomes more automated and accurate, and computer capabilities inevitably progress, this process will be faster, better and cheaper in the future.