Existing vegetation and land cover of the Little Missouri, Grand River, and Sheyenne National Grasslands

Publication Type:

Report

Source:

The University of Montana, Montana Cooperative Wildlife Research Unit, Wildlife Spatial Analysis Lab, Missoula, MT (1997)

Call Number:

U97RED01IDUS

Abstract:

We constructed a digital map of existing vegetation and land cover across nearly 10 million hectacres (~24 million acres) in the Dakotas and eastern Montana based on the classification of 3.25 Landsat Thematic Mapper (TM) scenes. Information was stored in a raster database comprised of ARC/INFO grids (one per TM scene). The database was built through a two-stage classification involving both unsupervised and supervised procedures. First, we carried out an unsupervised classification (based on Euclidean distance) for each TM scene. This pixel classification was designed to maintain patterns observed in a color composite of TM channels 4, 5, and 3. The resulting spectral classes (n ~ 130) then were merged to a 0.4 ha minimum map unit (MMU; >4 pixels) in areas where woody draw or riparian vegetation was likely to occur, and to a 2 ha MMU (> 22 pixels) in upland settings. Next, a raster database was constructed in ARC/INFO; base regions (or raster polygons) were delineated, and attributes for each region were collected (including majority aspect values and mean values for TM channels 1-7, elevation and slope). Adjacent TM scenes then were 'virtually' edge-matched such that the files could be seamlessly merged after classification. Ground-truth data (2,919 unique plots form the Forest Services), were compiled in a reference database and subjected to a series of logical and positional tests to verify their accuracy. Plots in each TM scene that passed all tests were used as training data in a supervised classification which assigned cover type labels using the NEAREST MEMBER OF GROUP classifier (based on Euclidean distance). Decision rules were also used for manual modifications specific to each TM scene. In addition, some cover types were labeled through visual interpretation of the imagery, such as urban areas, agricultural lands, and clouds. In all, 36 cover types were mapped for the project area; these included urban, agriculture (irrigated and dryland), six grassland types, seven shrub types, four coniferous and one broadleaf forest types, eight riparian types, three badland types, plus water, rock, and clouds. The predominant cover types in the project area were grasslands and agriculture (46% and 28%, respectively). Shrublands and forests covered another 13% of the area, and the remaining 13% was mostly woody draw/riparian types and badlands. Classification accuracy was assessed using a bootstrap method and fuzzy sets. accuracy values ranged from 45-88% and were relatively high for lower cover grassland types, ponderosa pine forests, badlands, and rock; shrub types, especially silver sage, tended to classify poorly.

Notes:

Reference Code: U97RED01IDUS

Full Citation: Redmond, R. L., J. C. Winne, T. P. Tady, M. Thornton, J. Troutwine, and Z. Ma. 1997. Existing vegetation and land cover of the Little Missouri, Grand River, and Sheyenne National Grasslands. Report prepared for the U.S. Department of Agriculture, Forest Service, Northern Regional Office. The University of Montana, Montana Cooperative Wildlife Research Unit, Wildlife Spatial Analysis Lab, Missoula, MT. 44 pp. plus appendices.

Location: