banner literature about projects tools meetings search links BBS GAP home GAP home


<- Previous | First | Next ->

Challenges in Land Cover Classification in Areas of Rapid Urban Expansion


Illinois Natural History Survey, Champaign

The Chicago region in northeastern Illinois is an area of rapid urban expansion. Areas that were
once agricultural fields and mining sites have been converted to residential and commercial
properties, many of which include man-made drainage ponds. New roads have been built to link
new development to older urban centers, providing a causeway for more urban expansion. As a
result, open grassland and woodland habitat has been lost to a highly fragmented landscape.
Landsat imagery covering the Chicago area also encompasses diverse rural areas, including
farmland, forest preserves, parks, beaches, and wetlands. This wide array of landscape types
required a way to separate significantly diverse spectral regions.

To understand what changes have occurred in this urban landscape, we studied the expansion of
the Chicago region. Our first challenge was found in creating an urban mask in a region of
explosive urban growth. An all-inclusive urban mask of high-density urban areas was essential
in separating urban spectral properties from vegetative spectral properties. The most recent
regionwide roads coverage was ten years old. Much development had since taken place in the
Chicago region, which made creating an urban mask directly from the roads coverage
impractical. A conventional classification of imagery to create the mask would have been
inconclusive because older urban residential areas appear as “woodland” and do not indicate
road density.

We solved this problem with a two-part approach. First, we followed methods outlined by
Morisette et al. (1996). The theory is to account for areas of high road density, which is a
measure of urban density.

Part I involved four steps performed in ARC/INFO GRID, v. 7.2:
1. the road vector coverage to grid format using LINEGRID.
2. Increase the road zone by five pixels (150 meters total) using EXPAND.
3. Reduce the encompassed area of high road density using SHRINK.
4. Convert Increase the encompassed area once again to produce an area of urban density using

Part II of the process involved on-screen digitization of urban expansion areas not encompassed
by the area expanded from the roads zone. On-screen digitizing was performed in ArcView 3.2
with the July 1999 Landsat TM image displayed as background. This process captured new
residential and commercial developments that did not have high-density road networks when the
roads coverage was created. In addition to adding in urban areas, on-screen work was used to
eliminate rural areas that had a high density road network but were not true urban areas, such as
farmsteads, state parks, and interstate cloverleaf exchanges. The benefits of using the road
density approach to creating an urban mask were that it captures roads in older urban residential
areas and the amount of time spent performing on-screen digitizing of urban expansion was
greatly reduced.

Once the urban mask was generated, we were able to run two separate unsupervised classifications on the Landsat imagery–one for "urban" areas and one for "non-urban" areas. We are confident that this extra effort led to better accuracy in our classification in both the urban and non-urban areas. As an example, we were able to capture small urban parks as well as older tree-lined neighborhoods within the Chicago metro area, which may have been lost otherwise. In this urban landscape, these "green pockets" provide critical habitat for wildlife that live and migrate through this area.

Literature Cited

Morisette, J.T, H. Chesire, C. Stallings, and S. Khorram. 1996. An urban mask raster image for vector street files. In S. Morain and S. Lopez Baros, editors. Raster Imagery in Geographic Information Systems. Onward Press, Santa Fe, New Mexico.

<- Previous | First | Next ->