1998 vegetation/land cover map of Georgia

Back to Base Layers and Information

Metadata:

Identification Information
Data Quality Information
Spatial Data Organization Information
Spatial Reference Information
Entity and Attribute Information
Distribution Information
Metadata Reference Information

Identification Information:

Citation:

Citation Information:

Originator:: Natural Resource Spatial Analysis Laboratory, Institute of Ecology, University of Georgia
Publication Date: July 1, 2003
Title: 1998 vegetation/land cover map of Georgia
Geospatial Data Presentation Form: raster digital data

Description:

Abstract:: This land cover map was produced from Landsat TM imagery with a spatial resolution of 30x30m. The classification process used six of the original seven bands of the imagery; the 120x120m thermal infrared band was removed from the data sets before processing. Additional ancillary geospatial and non-geospatial statistical data were incorporated in the mapping process. Information regarding the types of data used will be found in the following text.
An accuracy assessment was performed using field verification, aerial videography, and digital ortho quarter quads (DOQQ). The overall statewide accuracy is 75.46%. Although the data are available at a 30 m pixel resolution, accuracy was not assessed on patches of less than 4 pixels. A further description of the accuracy information will be found in the data quality statement.
Purpose:: These data can be used in a Geographic Information System (GIS) for any number of purposes including conservation planning, wildlife habitat assessment, etc. The University of Georgia assumes no liability for the use of these data.
Accuracy of the data is reported at the scale of the state. For this reason, it is recommended that additional field reconnaissance be used before performing analysis with the data at higher resolutions. Images used in the production of the land cover range from 1996-1998 and therefore do not reflect change that has occurred beyond these dates.

Time Period of Content:

Time Period Information:

Single Date/Time:

Calendar Date: 1998

Currentness Reference: ground condition

Status:

Progress: Complete
Maintenance and Update Frequency: None planned

Spatial Domain:

Bounding Coordinates:

West Bounding Coordinate: -87.334182
East Bounding Coordinate: -79.122568
North Bounding Coordinate: 35.673024
South Bounding Coordinate: 29.989335

Keywords:

Theme:

Theme Keyword: vegetation
Theme Keyword: Georgia
Theme Keyword: GAP
Theme Keyword: map
Theme Keyword: 1998
Theme Keyword: land cover

Place:

Place Keyword: Georgia

Temporal:

Temporal Keyword: 1998

Access Constraints: None.

Use Constraints: None.

Point of Contact:

Contact Information:

Contact Person Primary:

Contact Person: Matthew Elliott
Contact Organization: University of Georgia

Contact Position: Program Coordinator, Georgia Gap Analysis Program

Contact Address:

Address Type: mailing and physical address
Address: Institute of Ecology, University of Georgia
City: Athens
State or Province: GA
Postal Code: 30601
Country: U.S.A.

Contact Voice Telephone: (706)542-3489

Contact Facsimile Telephone: (706)542-6040

Contact Electronic Mail Address: melliott@uga.edu

Data Set Credit:

Matt Elliott, J.P. Schmidt, Kevin Samples, Bill Bumback, Liz Kramer; University of Georgia, College of Environment and Design, Natural Resources Spatial Analysis Laboratory (NARSAL)

Native Data Set Environment:

Microsoft Windows NT Version 4.0 (Build 1381) Service Pack 6; ESRI ArcCatalog 8.1.0.642

Data Quality Information:

Attribute Accuracy:

Attribute Accuracy Report:

The Georgia GAP 1998 landcover was created in two stages. Initially, an 18-class Anderson-level map was produced. From this, we refined the initial map into a final version with 44 classes (the GAP landcover). Accuracy assessment was conducted both at the 18-class level and the 44-class level.

For accuracy assessment of the 18-class map, land cover clumps were stratified by ecoregion and land cover class. Ecoregions included the Coast, Coastal Plain, Fall Line, Piedmont, Mountains (including Blue Ridge, Ridge and Valley and Cumberland Plateau), and Atlanta regions. The number of random patches to be selected by eco-region was calculated from the formula n = p(1-p)/s-squared, where p is the presumed accuracy and s is the standard error (Cochran 1977). GAP requires a standard error of 8%, therefore with a presumed accuracy of 50%, 40 patches per cover class (as an average) were necessary in each ecoregion. The total number of points was calculated by multiplying 40 by the number of cover classes. Aerial extent per class was calculated, and the final number of patches assessed per class was based on the percent of aerial extent for that class in the region. A minimum of 5 points in a region was selected for any given class. Only patches of four or more pixels were randomly selected for interpretation.

In all of Georgia except the Atlanta area, an initial assessment to the Anderson-level was conducted by means of aerial videography. A flight plan was designed to maximize coverage of each ecoregion. Video acquisition was conducted over four days: October 31, November 1, 2, and 3, 2000. During fall color change, this time period allowed individual tree species to be more easily distinguished. The setup of aircraft, video equipment, GPS, and dual cameras followed that of Slaymaker (1996).

The equipment used for the assessment consisted of a Canon GL1 Digital Video Camcorder, a Horita FP-50/TR GPC GPS3 SMPTE Time Code Reader, a 13" television, and a computer using ArcView. The videocassettes were viewed on the television using the Horita Time Code Reader to display the time code on the television screen.

For each point, the interpreter identified a first and second choice of land cover classes, and entered a location confidence variable. This location information was used to compensate for changes in location due to the movement of the airplane during video filming. Interpreters assigned a one (1) for an interpretation where they could reasonably identify the location of the clump on the video and a zero (0) if the interpreter had very little confidence in the location.

In the Atlanta area, verification was accomplished using 1999 color-infrared DOQQ's, which cover the thirteen county metropolitan area. As with the video, a blind interpretation of the DOQQ point locations was conducted with only the land cover polygons overlaying the photographs and the interpreter having no knowledge of the corresponding land cover. The formula for selecting points was the same as that employed for the areas assessed with videography.

In assessing the 43-class GAP landcover, the state was stratified by three ecoregions: Coastal Plain (combining the Coast, earlier Coastal Plain, and Fall Line regions), Piedmont, and Mountains. Within ecoregions, the number of points selected per class followed the previously illustrated methodology.

Among classes that changed between the 18-class map and the GAP landcover, we selected new accuracy assessment clumps. In the Mountains and Piedmont, these were randomly chosen along road transects. In the Coastal Plain, the were interpreted from aerial videography. Among classes that did not change between the 18-class map and the GAP landcover, we retained clumps from the initial accuracy assessment.

A total of 4120 points were used in accuracy assessment of the GAP map. The combined overall accuracy is 75.46%. For additional information regarding the accuracy assessment refer to Elliott et al. (2003).

Positional Accuracy:

Horizontal Positional Accuracy:

Horizontal Positional Accuracy Report:: A total of 4120 points were used in accuracy assessment of the GAP map. The combined overall accuracy is 75.46%. For additional information regarding the accuracy assessment refer to Elliott et al. (2003).

Lineage:

Source Information:

Source Citation:

Citation Information:

Originator: EROS Data Center, Sioux Falls, SD
Title: Landsat 5 Thematic Mapper
Other Citation Details:: (Row,Path) [Date, Date, Date]; (20,36) [August 2, 1997, July 17, 1997, March 30, 1998*]; (19,36) [September 25, 1996, October 17, 1998, January 2, 1998*]; (19,37) [September 25, 1996, June 27, 1998, January 2, 1998*]; (19,38) [April 2, 1996, June 27, 1998, January 2, 1998*]; (19,39) [April 2, 1996, June 27, 1998, January 2, 1998*]; (18,36) [April 27, 1996, June 20, 1998, November 24, 1997*]; (18,37) [April 27, 1996, May 19, 1998,January 11, 1998*]; (18,38) [April 27, 1996, May 19, 1998, January 11, 1998*]; (18,39) [April 27, 1996, May 19, 1998, January 11, 1998*]; (17,37) [June 23, 1996, May 12, 1998, November 17, 1997*]; (17,38) [June 23, 1996, May 12, 1998, December 19, 1997*]; (17,39) [April 20, 1996, May 12, 1998, December 19, 1997*]; (16,38) [March 12, 1996, March 2, 1998, January 26, 1997*]; (16,39) [March 12, 1996, May 5, 1998, January 26, 1997*];

Source Time Period of Content:

Source Currentness Reference: ground condition

Process Step:

Process Description:

The 1998 Georgia GAP landcover was created in two stages. Initially, an 18-class Anderson-level map was produced. For the 18-class map the methodology is as follows. More information may be obtained in Kramer et al. 2003 and Payne et al. 2003.

The first step of the process involved creating subsets of the leaf-off (winter) Landsat TM images by county. Linear features such as transportation infrastructure, utility swaths, and water bodies were then identified and classified using datasets obtained by the GA GIS Clearinghouses. These included transportation infrastructure; roads, railroads, and airports; utility lines; mining sites; wetlands; and linear and polygonal hydrographic features. Feature data incorporated into the land cover map were visually assessed with 1993 Digital Ortho Quarter Quads (DOQQ) and corrected if necessary. This was especially true for utility line data. County DOQQ mosaics were created and used as reference sources for the mapping process.

Urban and agricultural areas selected from the National Land Cover Dataset (U.S. Geological Survey 1999) and used to subset the raw 6-band images into areas of potential urban and agriculture. The subsets of raw imagery were then clustered using the Iterative Self-Organizing Data Analysis Technique (ISODATA). This clustering method, referred to as an unsupervised classification, uses spectral patterns to assign groupings of pixels into classes. The number of classes in each ISODATA image varied depending on the size of the subset and the complexity of the category and county. ISODATA images usually fell within a range of 25 to 100 classes. Classes from the initial ISODATA image that were difficult to classify were run through a second round of ISODATA processing (cluster bust).

Each of the clustered images resulting from the unsupervised classifications were interpreted using visual inspection of multiple dates of the raw TM images guided by a combination of ancillary data including; the black and white 1993 DOQQs, National Wetlands Inventory, digital elevation models, and polygonal hydrology features.

While leaf-off images were the primary data source for most unsupervised classifications, the spring images (leaf-on) were useful for separating clear-cut areas from agriculture/pasture, for reducing topographic effects in mountainous areas, and for identifying clear-cut areas from deciduous forest along the fall line and in sandhills.

The identification of low intensity residential areas was enhanced using road data provided by the Georgia Department of Transportation in conjunction with visual assessment of the DOQQ's. The methods associated with mapping the extent of low intensity residential are reported in Epstein et al. 2002.

All of the data layers were combined into a single county map and then all county maps were combined into a single statewide land cover. Corrections to the statewide map were then performed on a regional or statewide basis.

From the 18-class map, a number of classes were split into more refined vegetation classes for the GAP landcover. Classes split included deciduous forest, evergreen forest, mixed forest, forested wetland, non-forested wetland, and, in the Coastal Plain, clearcut.

Refined vegetation and other classes were derived through a combination of supervised classifications (live oak, open pine), further unsupervised classifications (cypress/gum, bottomland hardwood, evergreen forested wetland, forested vs. non-forested residential classes), topographic modelling using the National Elevation Dataset (all deciduous and mixed forest classes, white pine, white pine/hemlock, xeric ridge pine), utilization of National Wetlands Inventory data (U.S. Fish and Wildlife Service 2002) (saltwater marsh, freshwater marsh, shrub wetland), and decision rules incorporating other ancillary datasets (longleaf pine (The Nature Conservancy 2002), loblolly/shortleaf, loblolly/slash, and Okefenokee (Florida Cooperative Fish and Wildlife Research Unit 1997) and Ft. Benning areas (Natureserve 2001).