U of M wordmark

navigation edge

 

website logo

 

 

Land Cover Classification

Creating a census of the landscape

One of the most common and useful applications of satellite remote sensing is using satellite imagery to create a land cover classification.  This means taking the signals that the satellite senses (the energy reflected in the visible and infrared wavelength) and translating them into a map of recognizable categories of land cover.

Historically, aerial photography provided an important source of information on land cover and land use. However, the cost of its acquisition and interpretation is prohibitively expensive for larger geographic areas. An alternative is to acquire the needed information from digital images of the Earth taken from satellites orbiting the globe. This approach has several advantages:

  1. The broad view from a satellite provides coverage of large, multi-county geographic areas;
  2. The digital form of the data lends itself to more efficient analysis using computers, and the classified data can be used in geographic information systems
  3. The land cover maps can be generated at considerably less cost than by other methods.

Picking the classes

Classification categories attempt to group land cover into classes based on structure, taxonomy, or function.  There can be many “levels” of complexity to a classification.    The land cover classifications displayed on this website are all considered “level one” or fairly simple categorically.  Our land cover classification system includes seven classes. 

Land Cover Class

Description

Agriculture

Agricultural cropland including row crops, forage crops and small grains.  Examples: corn, soybeans, alfalfa, oats, wheat, barley and sugarbeet.

Forest

Land covered with trees reaching a mature height of at least 6 feet tall with a definite crown. Examples: white pine, red pine, black spruce, fir, mixed conifer, aspen, maple, oak, and mixed deciduous.

Grassland

Golf courses, lawns, and sod fields in the TCMA maps. For the statewide maps, the class is expanded to include upland areas covered by cultivated or non-cultivated herbaceous vegetation predominated by grasses, grass-like plants and forbs.  Examples: pasture and dry prairie.

Shrubland

An upland or lowland area with vegetation that has woody stems, generally with several basal shoots, low growth of less than 20-feet height, and fairly uniformly distributed throughout and moderate to high density. Examples: alder, willow, buckthorn, hazel, sumac, and scrub oak.

Wetland

A lowland area with a cover of persistent and non-persistent herbaceous plants standing above the surface of wet soil or water.  Examples: cattails, march grass, sedges and peat.

Water

Permanent open water, lakes, reservoirs, streams, bays and estuaries.

Urban

Residential, commercial, industrial, transportation, industrial and commercial, mixed urban or build-up land, other urban or built-up land.

This classification scheme was modeled after the Upper Midwest Gap Analysis Program Image Processing Protocol.  It identifies the major land cover types of the Upper Midwest, is compatible with existing national systems, and provides a realistic classification hierarchy for Landsat TM and ETM+ data.  Further, these classes are consistent with earlier classifications allowing for year-to-year classification comparisons.

Image processing and classification

To reduce complicating atmospheric effects, only clear, cloud-free images were chosen for processing.  All images were rectified to UTM zone 15, GRS1980, NAD83 using at least 35 well distributed ground control points and nearest neighbor resampling. To reduce the volume of data, but maintain its information content, a principal components like transformation, known as the “tasseled cap” transformation was performed.  The components or features of the transformation are:  brightness, greenness and wetness. Representative samples of each cover type class were used to “train” the classifier which then assigned each pixel into one of the seven cover type classes based on its spectral-radiometric-temporal properties.

Map minutiae

Statewide 2000: Nineteen images for each of three dates, spring, summer, and fall were collected.   Fifty seven images, acquired by both Landsat-7 ETM+ and Landsat-5 TM between 1999 and 2001, were required for statewide coverage.  The images were jointed to form a statewide image mosaic for each time period.  The three statewide images were then registered to create a single image with nine features for classifications. The imagery was stratified based on acquisition date and ecoregion into 20 separate strata, “spectrally consistent classification units,” for classification.

Statewide 1990: This is the GAP classification by the Minnesota DNR recoded to match our Statewide 2000 classes.  For more information on how the 1990 Minnesota GAP data was created, see http://deli.dnr.state.mn.us/metadata.html?id=L390002710606.

Twin Cities Metro Area (TCMA) series: Four pairs of bitemporal clear, cloud-free Landsat images were selected to classify the study area: June 2 and August 23, 1986; June 16 and September 4, 1991; May 18 and September 7, 1998; and May 21 and July 16, 2002. The seven-county TCMA is entirely contained within Landsat path 27, rows 28–29. The images were Landsat-5 TM, except for a Landsat-7 ETM+ image for May 2002.

For more information on the accuracy of these maps, continue on to the Accuracy Assessment page.

 




A “level one” land cover map for the entire state of Minnesota for the year 2000. 

This is a “level one” land cover map for the entire state of Minnesota for the year 2000.  The cover types were derived from multitemporal, multispectral image classification of satellite imagery acquired by the Landsat satellites. 

Some terms to know

Multitemporal: simply means more than one time period. Numerous studies have demonstrated the synergistic value of using two or more dates of imagery for classification of land cover. For our land cover classifications we used a combination of spring, summer and fall dates of clear, cloud-free Landsat images for classification.

Multispectral:  refers to the several portions (e.g., blue, green, red in the visible, plus the reflective infrared) of the electromagnetic spectrum used to identify unique signatures for each land cover type.