Illinois Natural History Survey - University of Illinois

Land-cover Classification for Forests

A land-cover map depicting major vegetational communities is one example of how satellite imagery is used in natural resource mapping. For many projects at the Illinois Natural History Survey (INHS), the creation of maps and databases that allow analyses of landscapes begins on a computer screen. While many scientists at INHS conduct research on ecology or some other area of biology, others classify satellite imagery, a difficult task.

Satellites actually measure the reflectance of visible light and other electromagnetic radiation off the land. As a satellite looks down upon the earth, it records data in pixels. Each pixel is about the size of a baseball diamond (30- m square) and must be classified based upon its reflectance of different wavelengths of light and other radiation. All these data are recorded and classified into similar categories. Additional processing is needed to determine what the various classifications mean in terms of ground vegetation.

The task of classifying the land cover of the state--a daunting task--is being shared. Don Luman and others at the Illinois State Geological Survey have classified the agricultural portion of the state (about 75% of the land area) as part of an agreement with the National Agricultural Statistical Service. INHS researchers are in the process of classifying the forests, wetlands, and grasslands of Illinois. The project is about half complete in generating a full land-cover map of Illinois based upon imagery acquired in 2000. Eventually this land-cover map will be used to model wildlife habitats and to evaluate land-cover conditions. The land-cover map will be used to examine land-use practices in Illinois, with hopes of assisting state agencies as they plan conservation measures including acquisitions, stewardship programs, and incentive programs.

The forests of Illinois are a particular challenge in classification. Because many wildlife species make their homes in forests, we have found that detailed mapping of the forests is very desirable. One of the difficulties in mapping the forests is identifying the dominant plant community in any given area. Because each pixel represents an area of 30 meters by 30 meters, some pixels may include more than one plant community. To improve interpretation of land-cover type from the imagery, we use "spatial patterning," that is, we look for patterns in the landscape that indicate dominance of a specific tree species in that area.

For example, we are currently examining the area around the Wabash River from the southern part of Edgar County to the northern part of White County to the western edge of Effingham County. A computer program statistically places pixels into different classes based on the reflectance characteristics, grouping similar pixels together. Next, we examine the classes created by the computer program and assign each a name. For example, many forests around rivers have different spectral and spatial patterns than forests in upland areas (see photo). Researchers may also use an elevation model to determine the trees in upland versus bottomland locations. Soil surveys and aerial photographs are also very helpful in defining tree species based on location and soil type, but the work may not end there. We also rely on experts to tell us whether our analysis is reasonable. With the challenges of this work, a land-cover map specialist at INHS becomes part scientist, detective, computer jockey, and naturalist. Who says computer work is boring?

Brooke Bahnsen and Patrick Brown, Center for Wildlife Ecology

Charlie Warwick, editor