ARCSS Project: Land Atmospheric Ice Interactions (LAII),Arctic Transitions in the Land-Atmosphere System (ATLAS) Title: Land-Cover Map of the North Slope of Alaska Authors: S.V. Muller1 and D.A. Walker2 Affiliation: Tundra Ecosystem Analysis and Mapping Laboratory, Institute of Arctic and Alpine Research, CB 450 University of Colorado, Boulder, CO 80303-0450, USA. Corresponding Author: Steven V. Muller e-mail: steven.muller@colorado.edu phone: 303.492.6631 fax: 303.492.6388 Purpose of Map Creation: Give a spatially explicit representaion of the nature and distribution of land-cover and tundra types on the North Slope of Alaska. This was done in support of extrapolating ecosystem models developed in the Kuparuk to the North Slope. Final date of map: August 31, 1998 Final data of accuracy assessment: February 25, 2000 Funding: Arctic System Science (ARCSS), Arctic Transitions in the Land-Atmosphere System (ATLAS), National Science Foundation Grant No. OPP-9415554. Metadata file date: 25 February 2000 Key References Recommended for Citation: Muller, S.V., A.E. Racoviteanu, and D.A. Walker, 1999. Landsat-MSS-dervied land-cover map of northern Alaska: extrapolation methods and comparison with photo-interpreted and AVHRR-derived maps. International Journal of Remote Sensing, 20, 2921-2946. Muller, S.V. and D.A. Walker, in review. Accuracy Assessment of a MSS-derived land-cover map of northern Alaska. International Journal of Remote Sensing. ------------------------------------------------------------------------------- MAP PROJECTION AND OTHER BASIC MAP INFORMATION 1) Data model: RASTER 2) NAD27 datum (Clarke 1866 ellipsoid) 3) Albers Equal Area projection with following parameters 1st standard parallel 68 31 0.000 2nd standard parallel 70 54 0.000 central meridian -154 0 0.00 latitude of projection's origin 50 0 0.000 4) 100x100 m pixels 5) 4508 rows by 11040 columns 6) lower left corner: x = -551950.0 m E y = 1968450.0 m N Geometric Rectification: 1) Cubic Convolution resampling 2) 2nd order polynomial registration. 3) RMSE speculated to be between 100-150m range by USGS (Digital Data Code) Map Unit Names*: (1) Dry Prostrate-shrub Tundra and Barrens (2) Moist Graminoid, Prostrate-shrub Tundra (nonacidic) (9) Moist Tussock-graminoid, Dwarf-shrub Tundra (cold, acidic) (3) Moist Dwarf-shrub, Tussock-graminoid Tundra (typcial tussock tundra) (4) Moist Low-shrub Tundra and Other Shrublands (5) Wet Graminoid Tundra (6) Water (7) Clouds and ice (8) Shadows * A more detailed description of the legend is described later MAP EXPLANATION AND METHODS Note: for greater discussion and detail of the methods and references, see Muller et al., in press. This map was prepared for studies of the Arctic ecosystems of the North Slope of Alaska which includes the National Science Foundation supported Arctic System Science (ARCSS), Arctic Transitions in the Land-Atmosphere System (ATLAS) project. Satellite Mosaic: A 26 image mosaic of Multi-Spectral Scanner (MSS) satellite data covering the tundra regions of the North Slope of Alaska was obtained for the purpose of creating the land-cover map of the North Slope. The satellite mosaic was prepared by the National Mapping Division, U.S. Geological Survey, EROS Data Center, Sioux Falls, SD. The entire mosaic covered from approximately 68 degrees North latitude in Alaska to the Arctic Coast. Images for the mosaic were acquired during snow-free growing seasons between 14 August 1976 through 13 September 1986 (see Table 1 for a complete list). Due to prevalent cloud cover over the North Slope during most growing seasons, single time period (e.g., one week) mosaics of imagery from sun- synchronous satellites are generally not feasible. Initially two 'sub-mosaics' were created, one for the western half of the North Slope (NPR-A and west) and a mosaic covering eastern half of the North Slope (CAMA and ANWR). For these the MSS images were resampled from 80-m nominal spatial resolution to 50-m pixels. Each of these sub-mosaics were geometrically corrected using cubic-convolution interpolation by means of a second-order polynomial registration. The two sub- mosaics were then geometrically joined (mosaicked) and resampled to a 100-m spatial resolution. The NPR-A sub-mosaic was radiometrically referenced to the CAMA/ANWR (eastern) sub-mosaic. Table1. List of MSS scenes that comprised the MSS mosaic used to create the North Slope land-cover map. The starred (*) scene was the radiometric reference scene. ================================================================== Scene Date ================================================================== CAMA/ANWR sub-mosaic NPR-A sub-mosaic 82157020462 14 Aug 1976 82903212425 13 Jul 1977 82163320534 13 Jul 1979 82905213525 15 Jul 1977 82163320534 14 Jul 1979 82906214015 16 Jul 1977 82163420592 14 Jul 1979 82906214105 16 Jul 1977 82238620391 4 Aug 1981 82906214135 16 Jul 1977 82238720445 5 Aug 1981 82922212755 1 Sep 1977 82237221013 21 Aug 1981 82922212825 1 Sep 1977 85018121305 29 Aug 1984 82129721153 11 Sep 1978 85050321184 17 Jul 1985 83049621573 14 Jul 1979 85050821363 22 Jul 1985 82164021333 20 Jul 1979 85051921175 2 Aug 1985 82164021335 20 Jul 1979 *85051921181 2 Aug 1985 85085821422 7 Jul 1986 85092621154 13 Sep 1986 85085821424 7 Jul 1986 ================================================================== Classification techniques: First, in order improve the performance of image processing algorithms to be applied, the area south of the Brooks Range and east of approximately 150 degrees west longitude was cropped from the image since this area is south of treeline. In this same vein, the areas of ocean (which included sea-ice and clouds) were cropped from the image. This resulted in some loss of islands and spits off of the coasts and a mosaic of the North Slope that was best suited for our classification efforts. The land-cover mapping effort of the North Slope was an extrapolation of the land-cover classification developed and used to map the land-cover map of the Kuparuk River Region [KRR; Muller et al., 1998]. Mimicking as closely as possible the methods used to classify the KRR, the North Slope MSS mosaic was classified using the following steps. First, using a K-means unsupervised algorithm in the ENVI (version 3.0) image-processing software, 43 cluster classes were created. Using the KRR land-cover map, first-hand experience with the area, and other local areas maps from the North Slope [Walker, 1985; Walker and Acevedo, 1987; Walker and Walker, 1991; Walker and Walker, 1996; Walker et al., 1982; Walker et al., 1989; Walker et al., 1994; Walker et al., 1996], each cluster was interpreted and grouped into the most appropriate of the eight land-cover categories defined in the KRR classification. During the process of analyzing these 43 cluster classes, we identified seven clusters (29, 32, 34, 42, 43, 44, and 45) that spectrally overlapped two or more land- cover classes. These clusters were tagged and pooled into two groups of clusters depending on the land-cover categories they overlapped. The K-means algorithm was reapplied to each group of pixels in the MSS image and resulted in a total of thirty-one additional clusters. The spatial distribution of these new clusters was analyzed, and each one was grouped into the land-cover categories with the best match on the reference maps. One of the new clusters within the second group still overlapped land-cover types. This was further split with the K-means algorithm into three finer clusters which were grouped into appropriate land-cover categories. At this point, we concluded that further differentiation of any of the 69 cluster classes would not result in significant improvement of the classification. Corrections Using Ancillary Data: The image classification resulting from the above cluster analysis and grouping had two noticeable problems that were correctable using ancillary data. First, pixels shadowed by clouds and mountains were classified as Water due to their low DNs in both bands. Similarly, mountainous areas on the edge of shadowed areas, pixels were classified as Wet Graminoid Tundra, a result of spectral mixing of the shadows and adjacent barren tundra areas. These shadow problems were corrected by using a set of boundaries that demarcated the mountains and cloud-shadow areas. These boundaries were digitized directly on the screen using the false color infrared (CIR) version of the MSS image as a base map. Using the CIR image, it was possible to differentiate between uplands, clouded areas, and lowlands. Using the digitized data, all pixels in mountainous and cloudy areas that were originally classified as Water and Wet Graminoid Tundra were reclassified as Shadows and Barrens respectively. The second problem occurred with the Moist Graminoid, Prostrate-shrub Tundra (nonacidic) category. On the KRR classification, this nonacidic tundra type covers most moist tundra areas on the coastal plain. A similar, but acidic, tundra type is common west of the Colville River on stabilized sand dunes of a large late-Pleistocene sand sea [Carter 1981]. To correct this problem, we identified these areas as a separate land-cover class called Moist Tussock- graminoid, Dwarf-shrub Tundra (cold, acidic). We used a surficial geology map of the NPR-A [Williams et al. 1985, modified in Gryc 1985] to identify the sand- sea region. The boundary was digitized and then overlain on the classification results. Within this area, any pixel previously defined as Moist Graminoid, Prostrate-shrub Tundra was reclassified to the new class, Moist Tussock- graminoid, Dwarf-shrub Tundra (cold, acidic). To eliminate the pixelated, "salt and pepper", appearance, we smoothed the data by applying a 5-pixel moving window majority algorithm. This was only for display purposes. The digital data made available with this metadata file have NOT been filtered. Detailed Legend: Vegetation units are groupings of finer-level units mapped at numerous sites within the basin [Walker et al., 1994; 1996]. Common habitats associated with each category are listed in Table 2. Dominant plant communities found in each category are listed in Table 3. Table 2. Common habitats of individual land-cover categories. ================================================================================ (Code) Land-cover class: list of common habitats -------------------------------------------------------------------------------- (1) Dry Prostrate-Shrub Tundra and Barrens (DPTB): 1. Lichen-covered and partially vegetated siliceous rocks in foothills and mountains 2. Dry partially-vegetated alpine tundra 3. Limestone bedrock 4. Barren and partially vegetated river alluvium 5. Barren coastal mud flats 6. Dunes 7. Roads and gravel pads (2) Moist Graminoid, Prostrate-shrub Tundra (MGPT; nonacidic): 1. Moist nonacidic hillslopes and moderately well-drained surfaces with pH = 5.5 2. Dry nonacidic river terraces and gravely well-drained slopes 3. Dry acidic tundra on hill crests, moraines and kames 4. Nonsorted-circle and -stripe complexes on the coastal plain and in the foothills 5. Moist/wet patterned-ground complexes [e.g. low-centered polygon complexes], especially on the coastal plain, with more than 50% moist nonacidic tundra 6. Moist coastal tundra (9) Moist Tussock-graminoid, Dwarf-shrub Tundra (MTDT; cold acidic): 1. Moist tussock tundra in the sand region with pH<5.5 2. Moist/wet patterned-ground complexes in sand region [e.g. low-centered polygon complexes], especially on the coastal plain, with more than 50% moist nonacidic tundra (3) Moist Dwarf-shrub, Tussock-graminoid Tundra (MDTT; typical tussock tundra): 1. Moist acidic hillslopes and moderately drained terrain with pH < 5.5 (4) Moist Low-Shrub Tundra and other Shrublands (MLTS): 1. Riparian shrublands along rivers 2. Watertracks and shrublands in basins in foothills 3. Tussock tundra dominated by low shrubs 4. Shrublands on south-facing slopes 5. True shrub tundra on flat or gently rolling surfaces (5) Wet Graminoid Tundra (WGT: 1. Rich fens on coastal plain, along rivers, and foothill basins 2. Poor fens in foothills 3. Wet/moist patterned-ground complexes (e.g. ice-wedge polygon complexes) with >50% wet tundra (6) Water: 1. Water 2. Marshes and aquatic vegetation with more than 50% standing water (7) Clouds and ice: 1. Aufeis along braided rivers 2. Clouds mainly at high elevations (8) Shadows: 1. Mostly steep terrain in the mountains 2. Some cloud shadows ================================================================================ Table 3. Dominant plant communities of individual land-cover categories. See Muller et al.[in press] for references on studies of specific communities. ================================================================================ (Code) land-cover class: list of dominant plant communities -------------------------------------------------------------------------------- (1) Dry Prostrate-Shrub Tundra and Barrens: 1. Cetraria nigricans-Rhizocarpon geographicum 2. Selaginello sibiricae- Dryadetum octopetalae 3. Saxifraga oppositifolia-Saxifraga eschscholtzii 4. Epilobium latifolium-Castilleja caudata 5. Carex subspathacea-Puccinellia phryganodes 6. Elymus arenarius-Artemisia borealis 7. Unvegetated (2) Moist Graminoid, Prostrate-shrub Tundra (nonacidic): 1. Dryado integrifolia-Caricetum bigelowii, Astragalus umbellatus-Dryas integrifolia 2. Oxytropis bryophila-Dryas integrifolia 3. Selaginello sibiricae-Dryadetum octopetalae, Salici phlybophyllae Arctoetum alpinae 4. Juncus biglumis-Saxifraga oppositifolia, Astragalus umbellatus-Dryas integrifolia 5. Dryado integrifolia-Caricetum bigelowii, Carex aquatilis-Eriophorum angustifolium, Carex aquatilis-C. chordorrhiza 6. Saxifraga cernua-Carex aquatilis, Sphaerophorus globosus-Luzula confusa, Dryas integrifolia-Carex aquatilis (9) Moist Tussock-graminoid, Dwarf-shrub Tundra (cold acidic): 1. Eriophorum vaginatum-Ledum decumbens 2. Eriophorum vaginatum-Ledum decumbens, Carex aquatilis-Eriophorum angustifolium, Carex aquatilis-C. chordorrhiza (3) Moist Dwarf-shrub, Tussock-graminoid Tundra (typical tussock tundra): 1. Sphagno-Eriophoretum vaginati (4) Moist Low-Shrub Tundra and other Shrublands: 1. Salix alaxensis-S. lanata, Sphagno-Eriophoretum vaginati betuletosum nanae, Salix pulchra-Calamagrostis canadensis 2. Eriophorum angustifolium-Salix pulchra1 3. Sphagno-Eriophoretum vaginati2 4. Salix glauca-Alnus crispa3 5. ? Willow dominated uplands (5) Wet Graminoid Tundra: 1. Carex aquatilis-Eriophorum angustifolium, C. aquatilis-C. chordorrhiza 2. Sphagnum orientale-Eriophorum scheuchzeri, Carex aquatilis-Sphagnum lenense, Sphagnum lenense-Salix fuscescens 3. Carex aquatilis-Eriophorum angustifolium, C. aquatilis-C. chordorrhiza, Dryado integrifolia-Caricetum bigelowii (6) Water: 1. Unvegetated 2. Carex aquatilis, Hippuris vulgaris-Arctophila fulva, unvegetated (7) Clouds and ice: 1. Unvegetated 2. Mostly alpine vegetation types, barrens (8) Shadows: 1. Primarily barrens, also snowbeds Carici microchaetae-Cassiopetum tetragonae and Dryas integrifolia-Cassiopetum tetragona ================================================================================ MAP RESULTS Percent Area Cover by Land-Cover Categories: A summary of map unit areas for the entire area mapped is shown in Table 4. Table 4. Percent-cover of land-cover for the North Slope area mapped. ========================================================================= Category Percent-cover ------------------------------------------------------------------------- Dry Prostrate-shrub Tundra and Barrens 10.6% Moist Graminoid, Prostrate-shrub Tundra 22.9% Moist Tussock-graminoid, Dwarf-shrub Tundra 3.1% Moist Dwarf-shrub, Tussock-graminoid Tundra 26.5% Moist Low-shrub Tundra and Other Shrublands 18.3% Wet Graminoid Tundra 7.9% Water 5.6% Clouds and ice 0.7% Shadows 4.3% ========================================================================= Map Accuracy: During the summer of 1999 an accuracy assessment of this map was undertaken. For a full discussion regarding the success for extrapolation the land-cover classification from the KRR to the entire North Slope can be found assessment Muller et al. [in review]. The assessment of accuracy used two ground data sets: 1) data collected in 1996 in the Kuparuk Rive Region for the accuracy assessment of land-cover map covering that region (Muller et. al. 1998) and 2) data collected in 1998 in the west-central region of the map. Logistically, it was not feasible to collect ground data sample points from all areas of the map so we concentrated in the west-central region during our 1999 field work. Using the two data sets, the map and categories were estimated to have the accuracies shown in Table 5. Table 5. Comparison of map data to combined Kuparuk River Region and west-central region ground datasets. ================================================================================ Ground Data -------------------------------------------------------------------- Land-cover User's cover map DPTB MGPT MTDT MDTT MLTS WGT Water Totals Accuracy -------------------------------------------------------------------------------- DPTB 22 . 1 . . . 1 25 88.0% MGPT 4 75 4 8 11 16 . 118 63.6% MTDT . 3 15 . . 4 . 22 68.2% MDTT . 18 . 64 12 1 . 95 67.4% MLTS . 1 . 12 50 1 . 64 78.1% WGT 2 6 7 . . 31 2 48 64.6% Water . . 1 . . 1 40 42 95.2% -------------------------------------------------------------------------------- Totals 28 103 28 84 73 55 43 414 Porducer's Accuracy 78.6% 72.8% 53.6% 76.2% 68.5% 56.4% 93.0% -------------------------------------------------------------------------------- Classification Accuracy P = 71.7% Te = 67.0% * ================================================================================ * Te is a measure of accuracy that removes chance agreement from the estimate (Ma and Redmond 1995). The following conclusions were draw from the analysis of the accuracy assessment data. Some of the data discussed in the conclusions were from error matrices or methods of analysis that were not preseneted here, but discussed in detail in Muller et. al. (in review): 1. An accuracy estimate of 80.6% for the KRR portion of the NA-MSS map (versus 87% accuracy for the original KRR-MSS map; Muller et al. 1998) indicates that the extrapolation of the classification to northern Alaska did a good job of duplicating the KRR-MSS map results. 2. The estimate of 64% accuracy for the west-central area of the NA-MSS map indicates that the extrapolation of the classification to this region was acceptable but not as successful as we had expected. However, the fuzzy analysis methods (Gopal and Woodcock 1994) revealed that 77% of the samples in this region were correctly or acceptably classified. This indicates a fairly good classification, as well as a considerable amount of heterogeneity within pixels. 3. Estimates of overall map accuracy (71% from error matrix and 79% incorporating the fuzzy analysis results) indicate a fairly successful mapping effort for such a large area. However, since ground data was sampled for only two regions, map users' should take these results only as a rough indicator of overall map accuracy. 4. Detailed analysis of the error patterns within categories revealed that errors were generally caused by spectral overlap between classes and/or mixed land-cover within pixels. Given the spectral and spatial resolution of the source imagery, these errors were not deemed correctible. 5. Verification the unexpected patterns of the MLTS on the NA-MSS map in the western and southern portion of the map was a priority of this assessment. The results of this analysis reveal that although there were some cases of true shrub tundra in this area, most of the MLTS in the west-central region are shrubby versions of MDTT or MDTT interspersed with bands of riparian shrublands in areas of horsetail drainages. According to the classification legend it is acceptable for shrubby versions of MDTT to be mapped as MLTS. However, it is important for user's to know that these areas are not true shrublands. 6. We believe the results of this accuracy assessment indicate that the NA-MSS map is of sufficient quality to serve the purposes of the ATLAS project. However, potential users should evaluate this on a case-by-case basis. REFERENCES Carter, L.D., 1981. A Pleistocene sand sea on the Alaska Arctic Coastal Plain. Science, 211, 381-383. Gopal, S., and Woodcock, C., 1994, Theory and Methods for accuracy assessment of the matic maps using fuzzy sets. Photogrammetric Engineering and Remote Sensing, 60, 181-188. Gryc, G., 1985. The National Petroleum Reserve in Alaska: earth science consideration. U.S. Geologic Survey Professional Paper 1240-C, 94. Ma, Z., and Redmond, R.L., 1995, Tau coefficients for accuracy assessment of classification of remote sensing data. Photogrammetric Engineering and Remote Sensing, 61, 435-439. Muller, S.V. and D.A. Walker, in review. Accuracy Assessment of a MSS-derived land-cover map of northern Alaska. International Journal of Remote Sensing. Muller, S.V., A.E. Racoviteanu, and D.A. Walker, 1999. Landsat-MSS-dervied land-cover map of northern Alaska: extrapolation methods and comparison with photo-interpreted and AVHRR-derived maps. International Journal of Remote Sensing, 20, 2921-2946. Muller, S.V., D.A. Walker, F.E. Nelson, N.A. Auerbach, J.G. Bockheim, S. Guyer, and D. Sherba, 1998. Accuracy assessment of a land-cover map of the Kuparuk River basin, Alaska: considerations for remote regions. Photogrammetric Engineering and Remote Sensing 64:619-628. Walker, D.A., 1985. Terrain and vegetation types of the Sagavanirktok Quadrangle, Alaska, NASA-Ames consortium agreement no. NCA2-OR170-303 final report, 65 pp. Walker, D.A., and W. Acevedo, 1987. Vegetation and a Landsat-derived land cover map of the Beechey Point Quadrangle, Arctic Coastal Plain, Alaska, CRREL Report 87-5, 63 pp., US Army Cold Regions Engineering and Research Laboratory, Hanover, NH. Walker, D.A., and M.D. Walker, 1991. History and pattern of disturbance in Alaskan arctic ecosystems: A hierarchical approach to analyzing landscape change, J. Appl. Ecol., 28, 244-276. Walker, D.A., and M.D. Walker, 1996. Terrain and vegetation of the Imnavait Creek watershed, in Landscape Function and Disturbance in Arctic Tundra, Ecological Studies, Vol. 120, edited by J.F. Reynolds, and J.D. Tenhunen, pp. 73-108, Springer-Verlag, Berlin. Walker, D.A., N.A. Auerbach, L.R. Lestak, S.V. Muller and M.D. Walker, 1996. A hierarchic GIS for studies of process, pattern, and scale in arctic ecosystems: The Arctic System Flux Study, Kuparuk River Basin, Alaska, poster presented at the Arctic System Science All-Hands Workshop, Snowbird, Utah, 30 April-3 May. Walker, D.A., K.R. Everett, W. Acevedo, L. Gaydos, J. Brown, and P.J. Webber, 1982. Landsat-assisted environmental mapping in the Arctic National Wildlife Refuge, Alaska, CRREL Rep. 82-37, 59 pp., US Army Cold Regions Res. and Eng. Lab., Hanover, NH. Walker, M.D., D.A. Walker, and N.A. Auerbach, 1994. Plant communities of a tussock tundra landscape, Brooks Range foothills, Alaska, J. Veg. Sci., 5, 843-866. Walker, M.D., D.A. Walker, and K.R. Everett, 1989. Wetland soils and vegetation, Arctic Foothills, Alaska, Biol. Rep. 89-7, 89 pp., US Fish Wildl. Serv. Res. Devel., Washington, DC. Williams, J.R., W.F. Yeend, L.D. Carter, and T.D. Hamilton, 1985. Preliminary surficial deposits map of National Petroleum Reserve - Alaska. Open-File Report 77-868, scale 1:500,000, 2 sheets, U.S. Geological Survey. ******************************************************************************* The map was created using ENVI 3.0 image processing software (RSI, Inc.) and Arc/Info 7.1.2. Both software packages were run on a Sun Microsystems Ultra 60 running Solaris 2.6. File available in two formats: 1) ASCII GRID File name: ns_landcover.txt.gz File format: ASCII File size: 7.2 Megabytes, (NOTE!!!! 202.4 Megabytes when uncompressed) Description: The ASCII file consists of header information containing a set of keywords, followed by cell values in row-major order. The file format is: ncols 11040 nrows 4508 xllcorner -551950 yllcorner 1968450 cellsize 100 NODATA_value -9999 row 1 cell values row 2 cell values . . . row 4508 cell values 2) ArcInfo GRID in Export File format File name ns_landcov.e00.gz* File size: 13.7 megabytes (NOTE!!!! 706.7 Megabytes when uncompressed) Description of Grid ns_landcov Cell Size = 100.000 Data Type: Integer Number of Rows = 4508 Number of Values = 9 Number of Columns = 11040 Attribute Data (bytes) = 8 BOUNDARY STATISTICS Xmin = -551950.000 Minimum Value = 1.000 Xmax = 552050.000 Maximum Value = 9.000 Ymin = 1968450.000 Mean = 3.501 Ymax = 2419250.000 Standard Deviation = 1.929 COORDINATE SYSTEM DESCRIPTION Projection ALBERS Units METERS Spheroid CLARKE1866 Parameters: 1st standard parallel 68 31 0.000 2nd standard parallel 70 54 0.000 central meridian -154 0 0.00 latitude of projection's origin 50 0 0.000 false easting (meters) 0.00000 false northing (meters) 0.00000 * Note for TEAML data users only: on the TEAML file system this grid file (prior to export) is called final_landcov