USGS 2005 GAP Landcover

A May, July, and September Landsat-5 Thematic Mapper image was acquired for each of 16 orbit path and row combinations required for complete coverage of North Dakota. The dates of all images for rows 27 and 28 in path 31, rows 26, 27, and 28 in path 32, and rows 27 and 28 in path 34 were identical within the path and the data were combined for the analysis. This reduced the number of path and row defined data sets from 16 to ten (path 30 row 27, path 30 row 28, path 31 row 26, path 31 rows 27 and 28, path 32 rows 26,27 and 28, path 33 row 26, path 33 row 27, path 33 row 28, path 34 row 26, and path 34 rows 27 and 28). The TM data for a path and row was clipped to the common spatial extent among the three dates of imagery. A vegetation and land cover classification was derived independently for each of the ten subsections of ND. The vegetation and land cover database for the state was produced by mosaicking the ten independently derived land cover maps. In the areas of overlap between the subsections, priority was given the path and row combination that had the most recent May TM image. The analysis of the TM imagery for each of the ten path and row subsections in ND was conducted using a sequential series of analyses in a hierarchal fashion. The analysis began with the discrimination of general land cover categories with similar physiognomy and progressed, for the natural and semi-natural land cover categories, to the discrimination of more detailed vegetation categories within the general land cover categories. A water mask was constructed for each image using a threshold on TM channel 5. A water mask for the path and row was created by combining the individual image water masks using Boolean logic and identifying pixels that were identified as water on two or more dates. Requiring a pixel to be identified as water on a minimum of two image dates before it was included in the water mask for the path and row reduced water commission errors. Cloud shadows and other commission errors in the water mask were manually excluded. Water remaining after incorporating National Wetlands Inventory data into the land cover classification was assigned to a generic wetland category. Cloud masks were created to exclude cloud pixels from training data. Cloud masks were created using a threshold on a TM channel 3 -TM channel 1 difference image, a threshold on the TM channel 1 image, and manual editing. Cloud masks were also used to determine if there were locations within ND which had cloud cover on all of the TM images and thus would require additional imagery to map the land cover or the inclusion of a cloud category in the land cover classification. Inspection of cloud masks in conjunction with extents of the TM images, including areas of overlap between paths and rows, revealed that it would be possible to map land cover for all areas of the state without acquiring additional images or the need for a cloud category. Wetlands are an important land cover component in ND, particularly in the Prairie Pothole Region. Many of the small temporary and seasonal wetlands in ND are small relative to the pixel size of the TM imagery and mapping these wetlands using only TM imagery would result in high omission errors for wetlands particularly if the imagery was acquired at a time when the wetlands were dry or have minimal surface water. Because of the difficulty mapping small wetlands using TM imagery, wetlands for the land cover database were extracted from digital National Wetlands Inventory (NWI) data despite the fact that NWI data for North Dakota were developed from aerial photography acquired during the period 1979-1984. A loss of wetlands was observed along the Missouri River and current land cover for these areas was identified from analysis of TM imagery. Wetland losses may have occurred in other areas but were not detected. There also appeared to be some gains in wetlands related to a wet period in the mid 1990's. Areas of water identified from analysis of TM imagery but not included in the NWI data were mapped as a generic wetland category. NWI for the Prairie Pothole Region of ND (areas to the east and north of the Missouri river) was obtained from the Region 6 U.S.Fish and Wildlife Service Habitat and Population Evaluation Team and NWI for areas west and south of the Missouri river was obtained from the U.S. Fish and Wildlife Service National Wetlands Inventory. Data were quality checked for valid wetland attribute codes as defined in the Classification of Wetlands and Deepwater habitats of the United States (Cowardin et al. 1979) and an exhaustive list of the attribute codes was created. For our purpose, we aggregated wetlands to seven wetland categories on the basis of the attributes at the system and class levels, and the water regime modifier. The seven categories are Lacustrine (System code L), Riverine (System code R), Palustrine Forested (System code P and subsystem code FO), Palustrine Scrub-shrub(System code P and subsystem code SS), Palustrine Semi-permanent (System code P, subsystem code not FO or SS, water regime code F), Palustrine Seasonal (System code P, subsystem code not FO or SS, water regime code C,D, or E), and Palustrine Temporary (System code P, subsystem code not FO or SS, water regime code A, B). There were a small number of Palustrine wetlands where the subsystem code is not FO or SS, and the water regime code is not A,B,C, D,E or F. Palustrine wetlands where subsystem code is not FO or SS with water regime G (intermittently exposed), H (permanently flooded),or K (artificially flooded) were classified as lacustrine wetlands. Wetlands which were complexes of 2 or more types are grouped for our purposes on the basis of the first wetland code. For example, a PSS/EMA wetland is identified as Palustrine Scrub-shrub. The NWI vectors were converted to rasters with 5 m pixels and these rasters were resampled to rasters with 30 m pixels and spatial extents corresponding to the appropriate TM image. The conversion from vector to raster overestimated the area of wetlands particularly for small wetlands. The two step vector to land cover data base raster conversion process reduced the overestimation of the area of wetlands to approximately 4 percent. Developed land cover categories were extracted from the ND National Land Cover data base. The four developed land cover categories included in the ND GAP land cover data base were Low Intensity Residential, High Intensity Residential, Commercial/Industrial/Transportation and Urban/Recreational Grasslands. Recent urban development in Fargo, Bismarck, Grand Forks and Minot and an omission of urban land cover in Williston, ND was mapped to create an additional developed land cover category. Multiple binary cropland not-cropland classification tree analyses were performed using different combinations of image dates and spectral channel subsets. Different combinations of image dates were used because of the wide range in the dates of the TM images for some paths and rows. For example, the TM images available for path 30 row 28 were included 4 May 1997, a 26 July 1998, and a 17 September 1994. The multiple binary classifications for cropland were combined to create a single image depicting all possible combinations of the cropland classifications. The digital number for a pixel in this image indexed the co-occurrence of cropland pixels among the binary classifications, e.g., pixels identified as cropland on all classifications, pixels identified as cropland on all classifications except one, croplands on different pairs of classifications, croplands on only one classification, etc. This cropland image was inspected by comparison with color composites of the TM imagery, digital orthophotographs, and the training data. Digital numbers in the image that contained a mixture of crop and not-cropland pixels were identified, and a kmeans clustering and maximum likelihood classification performed in an attempt to reduce the confusion for cropland in those spectral classes. A cropland mask was created by assigning the digital numbers in the cropland image to either cropland or not-cropland. Next, the cropland mask created in the previous step was combined with a cropland mask extracted from the ND National Land Cover dataset, and a cropland mask from either a Ducks Unlimited/U.S. Fish and Wildlife Service land cover classification (east and north of the Missouri River) or a U.S. Forest Service land cover classification (west and south of the Missouri River) to depict all possible combinations of cropland as defined by the three sources. This image was inspected for cropland omission and commission errors in the cropland mask created in the previous step and the cropland mask was modified to minimize the errors. Cropland commission errors for sparsely-vegetated badlands and sand dunes for path 30 row 28, path 32 rows 26,27,28, and path 34 rows 27 and 28 were reduced using polygons from the Ecoregions of North Dakota and Geologic Map of North Dakota vectors. Next binary woodland not-woodland classification tree analyses were performed in a manner similar to that described for croplands. Extensive use of digital orthophotographs was used to inspect the woodland classifications for omission and commission errors. If sufficient training data was available, a woodland community type analysis was performed using classification tree analysis. Otherwise, an unsupervised Kmeans cluster analysis and maximum likelihood classification was performed for the woodland mask and correspondence of spectral classes to woodland types was determined from landscape position and association information as determined from manual interpretation of color composites from the TM imagery, digital orthophotographs, and statistical summaries of the association of spectral classes with training samples. Ponderosa pine was mapped using conifer forest spectral classes with township, range, and section information for the four largest stands described in Potter, L.D. and D.L. Green, 1964, Ecology of Ponderosa Pine in Western North Dakota, Ecology 45:10-22. The single stand of limber pine was mapped using conifer forest spectral classes with township, range, and section information for the stand described in Potter, L.D. and D.L. Green, 1964, Ecology of a northeastern outlying stand of Pinus flexilis, Ecology 45:866-868. Next binary planted grassland not-planted grassland classification tree analyses were performed in a manner similar to that described for croplands and woodlands. In an effort to further reduce classification errors between planted grassland and natural grassland (prairie) due to their similarity in physiognomic structure and land uses, a Kmeans cluster analysis and Maximum likelihood classification were performed for pixels assigned to the planted grass mask. Statistical summaries of the association of spectral classes with planted grassland and natural grassland training samples and manual inspection of the spatial patterns of the spectral classes were to identify spectral classes as planted grassland, mixtures of planted grassland and natural grassland, and natural grassland. Spectral classes identified as planted grassland were used to create a planted grassland mask and spectral classes identified as mixtures of planted grassland and natural grassland were combined to create a mask for analysis of prairie land cover categories. Pixels remaining after constructing the cropland, woodland, planted grassland, wetlands, water, and developed land masks were assigned to the prairie land cover category. An analysis of prairie community types was performed using classification tree analyses with training data from range inventories of North Dakota School Lands managed by the North Dakota State Land Board. The assignment of classification tree nodes to prairie plant communities was evaluated by reference to landscape position and association as determined from manual inspection of TM color composites, digital orthophotographs, and statistical summaries of the association of nodes with prairie plant community and planted grassland training samples. Changes in assignment of prairie community type labels to classification tree nodes were made when evaluation of the classification suggested a different prairie plant community type or planted grassland would be more appropriate label for the node. There was insufficient School Land training data for a classification tree analysis of prairie plant communities for path 30 row 27, path 30 row 28, and path 31 row 26. A Kmeans cluster analysis and maximum likelihood classification was performed for path 30 row 28 and prairie plant community labels were assigned to the spectral classes by reference to prairie plant community maps created by the North Dakota Natural Heritage Program for Richland, Sargent, Barnes and Ransom counties in North Dakota. A prairie plant community analysis was not attempted for path 30 row 27 and path 31 row 26 because of the low abundance of prairies in this portion of the state and insufficient data training data. All prairie land cover in path 30 row 27 and path 31 row 26 was assigned to a single prairie land cover category. Development of a shrubland land cover category was challenging in part due to the difficulty defining training areas for small, sparsely distributed, and irregular shaped shrub patches and hence a small sample of training polygons and in part due to the spectral similarity of shrublands with other land cover categories. Because of these difficulties, we were not able to create a shrubland category for all areas of the state. The shrubland category was developed, when possible, during the analysis of the woodland and prairie land cover categories. Spectral classes for the shrubland category were identified when they were detected during the process of inspecting the woodland classifications using the digital orthophotos and using a threshold on a near infrared/red ratio for the July TM image for areas identified as prairie. The logic behind the use of the July near infrared/red ratio procedure was that the grass and forb component of the vegetation would be experiencing a water deficit at this date, while the shrub component with its deeper roots would have access to water and would have a canopy with more actively growing green leaves than grasses and forbs. These phenologic conditions did not appear to be met for many of the July images as attempts to define a shrubland category resulted in excessive shrubland commission error and hence a shrubland land cover category was not created. A shrubland land cover category was developed for paths 34 rows 27 and 28, path 33 row 27, path 33 row 28, path 31 rows 27 and 28, and path 31 row 26. The shrubland category for path 31 rows 27 and 28 was extended into the overlap areas for path 30 row 28 and path 32 rows 27 and 28. Refinement of a shrubland land cover category should be a high priorty in future land cover maps. The shrubland land cover category was partioned into an upland deciduous shrubland category, a sagebrush shrubland category (only for areas in path 34 rows 27 and 28), and a lowland shrubland category. The vegetation and land cover database for the state was produced by mosaicking the ten independently derived land cover maps. In the areas of overlap between the subsections, priority was given the path and row combination that had the most recent May TM image. Inspection of the state land cover mosaic revealed a 7.4 sq km area at the intersection of path 31 row 26, path 30 row 27 and the state boundary with Minnesota that did not have land cover data. Land cover data from the ND 1992 National Land Cover data set corresponding to the missing data location was crosswalked to the categories of the ND Gap land cover database and inserted into the state land cover mosaic to complete the coverage for ND. Five post-classification stratifications using ancillary data were performed to increase the information content of the land cover map and to reduce land cover classification errors. A Floodplain Woodland category was created by intersecting a binary mask created from the Geologic Map of ND with the Woodland land cover category. The initial binary geologic mask was created from two geologic categories, Holocene River Sediment, Qor, and Holocene to Pre-Wisconsin uncollapsed river sediment, Qcrf. The initial intersection of this mask with a binary image of the woodland land cover category revealed numerous commission errors for floodplain woodland. The binary geologic map was manually edited to exclude these commission errors. A final floodplain woodland land cover category was constructed after several iterations of manually editing the binary geologic map and inspection of its intersection with the woodland land cover category. An area in the northern portion of the Glacial Lake Agassiz physiographic region near Grand Forks, ND, was mapped to saline prairie land cover category by intersecting a binary mask of soil map unit ND073 from the ND State Soil Geographic (STATSGO) data base with planted grassland and prairie grassland land cover categories. The area of the sand prairie land cover category was increased by intersecting a binary mask of the prairie land cover category for path 31 rows 27 and 28 and path 31 row 26 with a binary mask of the Holocene Windblown Sand, Qod, category from the Geologic Map of ND. A fescue prairie land cover category was created in north-western ND by intersecting a binary mask of the Bluestem - Needlegrass-Wheatwass prairie land cover category from path 34 row 26 with a binary mask of a Fescue Prairie land cover category developed from regression tree analyses predicting the relative abundance of Heterostipa curtiseta and Agropyron dasystachyum from environmental variables (unpublished data presented at a poster session at the 55th Annual Meeting of the Society for Range Management, L. L. Strong, Integration of GIS and remote sensing for mapping rangeland plant communities of the Northern Great Plains). The prairie land cover category was reduced by changing pixels classified as prairie to planted grasslands if the pixels were classified as cropland on five or more years in a six year land cover data base constructed by the ND Agricultural Statistical Service.

The three developed land cover categories, low intensity residential, high intensity residential, and commercial/industrial/transporation and the urban/recreational grasses land cover categories of the North Dakota National Land Cover Data Set were used in the mapping of these land cover types in the North Dakota Gap Analysis Land Cover.

Used in the mapping of wetland land cover types.

Used in the production of the vegetatation and land cover map.

NatureServe. 2001. International Classification of Ecological Communities: Terrestrial Vegetation. Natural Heritage Central Databases. NatureServe, Arlington, VA.

U.S Department of Agriculture, Soil Conservation Service. 1976. National Range Handbook. Washington, D.C.

Barnes, P. W., L. L. Tieszen, and D. J. Ode. 1983. Distribution, production, and diversity of C3 and C4 dominated communities in a mixed prairie. Canadian Journal of Botany 61:741-751. 

Clements, F E. 1920. Plant indicators: The relation of plant communities to process and practice. Press of Gibson Brothers, Inc. Carnegie Institution of Washington. 73-139.

Cosby, H. E. 1965. Fescue grassland in North Dakota. J. Range Manage. 18:284-285.

Coupland, R T. 1992. Fescue prairie. Pages 291-295 in Coupland, R. T., ed., Ecosystems of the World 8A: Natural Grasslands: Introduction and Western Hemisphere. Elsevier, Amsterdam. 

Dix, R. L. and F. E. Smeins. 1967. The prairie, meadow, and marsh vegetation of Nelson County, North Dakota. Canadian Journal of Botany 45:21-58.

Hansen, P L, Hoffman, G R, and Bjugstad, A J. 1984. The vegetation of Theodore Roosevelt National Park, North Dakota: A habitat type classification. U.S. Dept. of Agriculture Forest Service General Technical Report. 35 pp.

Hegstad, G.D.  1973.  Vascular flora of Burke, Divide, Mountrail, and Williams counties in northwest North Dakota.  Ph.D. Diss., ND State Univ., Fargo, ND.

Johnson, W. C., R. L. Burgess, and W. R. Keammerer. 1976. Forest overstory vegetation and environment on the Missouri River floodplain in North Dakota. Ecological Monographs 46:59-84

Küchler, A W. 1964. Manual to accompany the map potential natural vegetation of the conterminous United States. American Geographical Society American Geographical Society Special Publication. 19 pp. 

Redmann, R. E. 1975. Production ecology of grassland plant communities in western North Dakota. Ecological Monographs 45:83-106.

Shantz, H. L. 1923. The natural vegetation of the Great Plains region. Annals of the Association of American Geographers 13:81-107.

Stewart, R. E. 1975.  Breeding birds of North Dakota.  Tri-College Center for Environmental Studies. Fargo, ND.

Weaver, J.E. and F.E. Clements.  1938.  Plant Ecology.  McGraw-Hill Book Company, New York.

Weaver, J.E. and T. J. Fitzpatrick.  1934.  The prairie.  Ecological Monographs. 4:113-295.

Whitman, W.C. 1979.  Analysis of grassland vegetation on selected key areas in southwestern ND.  North Dakota Regional Environmental Assessment Program. 199 pp.

Whitman, W.C. and Barker, W.T.  1994. Rangeland cover types of the Northern Great Plains region .  Pages 69-84 in T.N. Shiflet, ed. Rangeland Cover Types of the United States.  Society for Range Management. Denver, CO.

Value  Area(sq km) land cover category
 0     36178.1     background - outside ND boundary
 1     88165.9     cropland
 2     30542.7     planted herbaceous perennials
10      1402.3     prairie - wet-mesic tall grass
11       156.2     prairie - mesic tall grass
12      2054.3     prairie - mesic tall and mixed grass
13      5851.7     prairie - bluestem-needlegrass-wheatgrass
14      8291.1	   prairie - wheatgrass prairie
15      7392.0     prairie - needlegrass prairie
16      4298.5     prairie - little bluestem
17       444.4     prairie - fescue
18      4267.7     prairie - sand
19      1523.2     prairie - saline
20      4299.2     shrubland - upland deciduous
21        79.1     shrubland - lowland deciduous
22       902.8     shrubland - sagebrush
30         9.6     woodland - ponderosa pine
31         0.3     woodland - limber pine
32       193.6     woodland - rocky mountain juniper
33       450.2     woodland - mixed conifer and deciduous woodland
34       688.2     woodland - floodplain
35      1601.9     woodland - deciduous
36       498.7     woodland - green ash
37       290.1     woodland - aspen
38       234.4     woodland - bur oak
39       317.2     woodland - aspen and bur oak
40      3482.5     wetland - lacustrine
41       442.0     wetland - riverine
42      3086.6     wetland - palustrine temporary
43      5162.5     wetland - palustrine seasonal
44      2794.7     wetland - palustrine semipermanent
45      1329.2     wetland - water
50       310.9     barren land 
51      1479.0     sparse vegetation - badlands
52       106.8     sparse vegetation - riparian
60        89.3     developed - high intensity residential
61       232.8     developed - low intensity residential
62       475.7     developed - commerical/industrial/transporation
63       108.6     developed - urban grasslands
64        46.9     developed - recently developed or omissions in 1992 NLC