Home - Table of Contents

An Alternative Approach to Land Cover Mapping in Complex Terrain

John McCombs, Scott Klopfer, Dave Morton, and Jeff Waldon
Virginia Tech, Blacksburg, Virginia

Researchers working for the Virginia Gap Analysis Project (VA-GAP) are developing a technique for forest-type mapping utilizing both remote sensing and abiotic factor modeling. This combined technique will assist in classification of forest types in Virginia where remote sensing alone is insufficient.

The technique uses a base land cover map combined with a landform-based moisture map and a fine-scale physiographic map of Virginia. The base land cover map (Morton in prep.) was synthesized from Landsat TM imagery and ancillary data layers such as DLG roads and NWI wetland maplets. Each image was classified to a modified Anderson Level I scheme, with forest tracts classified as either coniferous, deciduous, or mixed.

Once the Level I map (Morton in prep.) was created, further discrimination of the three forest type classes (deciduous, coniferous, and mixed) was needed. Due to the complex topographic nature of the state of Virginia and the relationships known to exist between forest type distribution and topographic characteristics, we are investigating abiotic factor modeling as an additional method of estimating forest-type distribution.

To produce the component data layers necessary for the abiotic factor modeling, digital elevation models (DEM) for the state corresponding to 7.5 minute topographic quadrangles were obtained from the USGS. The DEMs have a resolution of 30 m, and an elevation value (along 1 meter increments) associated with each raster cell.

The individual data layers used to create the moisture map were slope, aspect, and a landform index (McCombs 1997, McNab 1989). The slope and aspect maps were generated directly from the DEMs. The landform index value was a measure of the convexity or concavity of the land surrounding a location. The landform index was calculated based on the changes in elevation from the point of interest to points located on the edge of a 9 x 9 pixel window. Locations in a cove or a sheltered area would have a lower elevation than the points around it, and thus would have positive landform index value. The opposite would occur on ridge tops or knobs. The more extreme the landform, index value (positive or negative) would indicate the degree of convexity or concavity. The moisture map was created through the analysis of combinations of the slope, aspect, and landform index maps. The resulting map had three classes of moisture: xeromorphic, mesomorphic, and unclassed. For example, a southwest-facing, steep, convex site would be classified as xeromorphic. A north-facing, relatively flat, concave site would be mesomorphic. Many combinations of slope, aspect, and landform index could not be definitively classed as xeromorphic or mesomorphic and were left "unclassed."

There are portions of five physiographic provinces in the state of Virginia—Coastal Plain, Piedmont, Blue Ridge Mountains, Ridge and Valley, and the Appalachian Plateau. The delineation (and naming) of these regions vary by author and by map (Daniels et al. 1973, Fenneman 1938). Physiography is of interest because several forest types are restricted to, or predominantly occur in, only one or two provinces. To simplify the analysis, three major physiographic provinces—Coastal Plain, Piedmont, and Mountains—were mapped. This was accomplished through the use of a relative phenological index value that was computed based on Hopkins’ Law of Bioclimatics (1938).

Hopkins studied the effect geographical location (in the Northern Hemisphere) had upon the timing of biological events. He found that for every 1 latitude increase north, 5 longitude increase west, or 122 m increase in elevation, the onset of biological events was delayed 4 days. It was then possible to compare any two points and compute a phenological difference between them. This was done for Virginia using Great Dismal Swamp National Wildlife Refuge in the southeast corner of the state as a base point.

The forest type, landform moisture, and physiographic map were then combined to create a 27-class map. The 27 class values result from the combination of three forest types (coniferous, deciduous, mixed), three moisture classes (xeromorphic, mesomorphic, unclassed), and three physiographic regions (Coastal Plain, Piedmont, Mountain). Possible forest types were then assigned to each class based on known ecological associations with the variables. For instance, only Eastern Hemlock, White Pine, Hemlock/White Pine, Red Spruce/Fraser Fir, or Virginia Pine would be expected to be found at a pixel classified as mountain mesomorphic conifer. To determine the most probable forest type for that pixel, the specific landform data would be re-incorporated. For instance, elevation would provide information which could either include or exclude Red Spruce/Fraser Fir - a type found only at elevations above 1520 m in Virginia. Slope, aspect, and landform index are also used to narrow the classification possibilities.

This technique will not allow classification of all forest types in all regions of the state. Other techniques being investigated include further use of Landsat TM imagery and landscape-scale climate estimations (Klopfer 1997). The 27-class forest type map can be used as a mask for spectral classification of specific types of forest which cannot be delineated from landform alone. With these masks, variability in the Landsat reference data may be lessened. Research into estimating landscape-scale climate factors may also prove useful in future investigations of forest type distribution in Virginia.

Although the development of this technique is ongoing, preliminary indications are that the methods described will allow VA-GAP to produce an accurate forest type map at a fine resolution for Virginia. This technique should prove useful for other state GAP projects attempting to map forest types/alliances in areas with high forest type diversity and complex topography. In addition, the intermediate data layers developed during this research have their own value for land-planning, research, and teaching activities.

Literature Cited

Daniels, R. B., B. L. Allen, H. H. Bailey, and F. H. Beinroth. 1973. Chapter 2 in S. W. Buol, editor. Physiography. Soils of the southern states and Puerto Rico. USDA Southern Cooperative Series Bulletin No. 174. 105 pp.

Fenneman, N. M. 1938. Physiography of the Eastern United States. McGraw-Hill Book Co., Inc. 714 pp.

Hopkins, A. D. 1938. Bioclimatics, a science of life and climate relations. U. S. Dept. Agr., Misc. Publ. 280. 188 pp.

Klopfer, S. D. 1997. Insolation, precipitation and moisture maps for a Virginia geographic information system. M.S. Thesis. VPI&SU. Blacksburg, VA. 184 pp.

McCombs, J. W. 1997. Geographic information system topographic factor maps for wildlife management. M.S. Thesis. VPI&SU. Blacksburg, VA. 141 pp.

McNab, W. H. 1989. Terrain shape index: quantifying effect of minor landforms on tree height. For. Sci. 35(1):91-104.

Morton, D.D. in prep. Land cover of Virginia from Landsat Thematic Mapper imagery and ancillary data. MS Thesis, Virginia Polytechnic Institute and State University, Blacksburg, VA.

Home - Table of Contents