Urbanization as a Component of Global Change, Boston University
Home
Local monitoring
Change detection
Modeling

Part 1: Measuring growth in urban areas

For this work, identification of changes in urban areas will be accomplished in three phases: (1) image preprocessing, (2) change detection using unsupervised methods, and (3) accuracy assessment. The first phase of work includes collection of Landsat TM imagery and ancillary map data for each city, coincident with two time periods: 1990 and 2000. Geometric correction will be used during the first phase to ensure that pixels of each image align in a common map coordinate system.

The second phase of work includes estimating the quantity and type of land use/land cover changes apparent in the images. To do this, a simple unsupervised multi-date k-means clustering of stable and changed land cover classes will be employed. The basic premise of this technique is that new change classes should have distinct combinations of spectral signatures from non-change classes, and hence be separable. Spatial information will be extracted from the images and combined with the clustered output to produce per-region map of change. A segmentation algorithm (Harward and Woodcock 1990) will be used that exploits the correlation between neighboring pixels to aggregate neighboring pixels into polygons. The resulting polygons are combined with the class map and labeled on the basis of the classes inside each polygon using a majority rule.

To validate the maps of land use change, the third phase involves ground-based accuracy assessment. Validating the maps is critical to improve methodology, improve area estimates, and understand biases in analysis (Congalton 1991). Validation will consist of collection of GPS guided ground truth measurements in a random stratified sampling design, which will be compared to results in the change map.


Preliminary results

The following cities provide a sample of the types of growth seen in this study. Stay tuned for new results in the following weeks.
1989 Landsat TM image of Brasilia, Brazil, where urban areas appear purple, vegetation green. 1999 Landsat TM image of Brasilia, Brazil, where urban areas appear purple, vegetation green. Brasilia, Brazil

1989 and 1999 images

Minimal growth in pockets.
1992 Landsat TM image of Chengdu, China, where urban areas appear purple, vegetation green. 2002 Landsat TM image of Chengdu, China, where urban areas appear purple, vegetation green. Chengdu, China

1991 and 2002 images

Growth along radial axes.
1987 Landsat TM image of Toronto, Canada, where urban areas appear purple, vegetation green. 1999 Landsat TM image of Toronto, Canada, where urban areas appear purple, vegetation green. Toronto, Canada

1987 and 1999 images

Outward densification.
1992 Landsat TM image of Beijing, China, where urban areas appear purple, vegetation green. 2001 Landsat TM image of Beijing, China, where urban areas appear purple, vegetation green. Beijing, China

1992 and 2001 images

Growth along urban fringe.




<- Back   |   Part 1: Change detection   |   Part 2: Modeling   |   Next ->



Boston University      |     Department of Geography      |     Center for Remote Sensing      |     MODIS Land Cover Project

Urban Areas Research
 Department of Geography | Boston University
675 Commonwealth Avenue | Boston Massachusetts 02215
(617) 353-2525 | fax (617) 353-3200

Acknowledgment and Disclaimer