BOREAS TE-18 Landsat TM Maximum Likelihood Classification Image of the NSA Summary The objective of this classification is to provide the BOREAS investigators with a data product that characterizes the land cover of the NSA. A Landsat-5 TM image from 20-Aug-1988 was used to derive this classification. A standard supervised maximum likelihood classification approach was used to produce this classification. The data are provided in a binary image format file. Table of Contents * 1. Data Set Overview * 2. Investigator(s) * 3. Theory of Measurements * 4. Equipment * 5. Data Acquisition Methods * 6. Observations * 7. Data Description * 8. Data Organization * 9. Data Manipulations * 10. Errors * 11. Notes * 12 Application of the Data Set * 13. Future Modifications and Plans * 14. Software * 15. Data Access * 16. Output Products and Availability * 17. References * 18. Glossary of Terms * 19. List of Acronyms * 20. Document Information 1. Data Set Overview 1.1 Data Set Identification BOREAS TE-18 Landsat TM Maximum Likelihood Classification Image of the NSA 1.2 Data Set Introduction This data set classifies BOReal Ecosystem-Atmosphere Study (BOREAS) Northern Study Area (NSA) into 11 land cover classes: wet conifer, dry conifer, mixed, deciduous, fen, water, bare soil/disturbed, recent burn, regeneration (young), regeneration (medium), and regeneration (old). This classification was done using a maximum likelihood classifier with training sites. 1.3 Objective/Purpose This classification was produced for BOREAS investigators who needed a land cover data set with which to compare other measurements. For example, one group used it to compare how CO2 flux varied with land cover along their aircraft flight path. 1.4 Summary of Parameters Land cover type The following classes were identified: Image Value Class ------------------------------ 1 Conifer (Wet) 2 Conifer (Dry) 3 Mixed (Coniferous and Deciduous) 4 Deciduous 5 Fen 6 Water 7 Disturbed 8 Regeneration (Younger) 9 Regeneration (Medium) 10 Regeneration (Older) 11 Visible Burn 1.5 Discussion The objective of this classification is to provide the BOREAS investigators with a data product that characterizes the land cover of the NSA. These data can be used for modeling purposes. The technique that was used to produce this classification is a standard maximum likelihood supervised classification. Training fields were selected for various land cover types. Statistics were derived from these training sites and subsequently used to classify the image. Specifically, the PCI programs CSG (signature generator) and MLC (multispectral classifier) were used to classify the image. 1.6 Related Data Sets BOREAS TE-18 Landsat TM Maximum Likelihood Classification Image of the SSA BOREAS TE-18 Landsat TM Physical Classification Image of the SSA 2. Investigators 2.1 Investigator Name and Title Dr. Forrest Hall Biospheric Sciences Branch NASA Goddard Space Flight Center (GSFC) 2.2 Title of Investigation BOREAS TE-18 Regional Scale Carbon Flux from Modeling and Remote Sensing 2.3 Contact Information Contact 1 ---------------- Dr. Forrest G. Hall NASA GSFC Greenbelt, MD (301) 286-2974 (301) 286-0239 (fax) Forrest.G.Hall@gsfc.nasa.gov Contact 2 ---------------- David Knapp NASA GSFC Greenbelt, MD (301) 286-1424 (301) 286-0239 (fax) David.Knapp@gsfc.nasa.gov 3. Theory of Measurements This product was produced for general use by modelers. The objective of this classification is to provide the BOREAS investigators with a data product that characterizes the land cover of the NSA. These data can be used for modeling purposes. The technique that was used to produce this classification is a standard maximum likelihood supervised classification. Training fields were selected for various land cover types. Statistics were derived from these training sites and used to classify the image. 4. Equipment 4.1 Instrument Description. The Landsat-5 Thematic Mapper (TM) sensor system records radiation from the seven bands described in Section 4.2.1. It has a telescope that directs the incoming radiant flux obtained along a scan line through a scan line collector to the visible and near-infrared focal plane, or to the mid-infrared and thermal-infrared cooled focal plane. The detectors for the visible and near- infrared bands (1 - 4) are four staggered linear arrays, each containing 16 silicon detectors. The two mid-infrared detectors are 16 indium-antimonide cells in a staggered linear array, and the thermal-infrared detector is a four-element array of mercury-cadmium-telluride cells. 4.1.1 Collection Environment The data that were used to produce this classification were collected by the Landsat-5 TM on 20-Aug-1988. Landsat-5 orbits the Earth at an altitude of approximately 705 kilometers. 4.1.2 Source/Platform Landsat-5 satellite 4.1.3 Source/Platform Mission Objectives The mission of the Landsat-5 satellite is to measure reflected radiation from Earth’s surface at a spatial resolution of 30-meters and to measure the temperature of Earth’s surface at a spatial resolution of 120-meters. 4.1.4 Key Variables Reflected radiation. Emitted radiation. Temperature. 4.1.5 Principles of Operation The TM is a scanning optical sensor operating in the visible and infrared wavelengths. It contains a scan mirror assembly that directly projects the reflected Earth radiation onto detectors arrayed in two focal planes. The TM achieves better imagery resolution, sharper color separation, and greater inflight geometric and radiometric accuracy for seven spectral bands simultaneously than the previous Multispectral Scanner (MSS). Data collected by the sensor are transmitted to Earth-receiving stations for processing. 4.1.6 Sensor/Instrument Measurement Geometry The TM depends on the forward motion of the spacecraft for the along-track scan and uses moving mirror assembly to scan in the cross-track direction (perpendicular to the spacecraft). The Instantaneous Field of View (IFOV) for each detector from bands 1-5 and band 7 is equivalent to a 30-m square when projected to the ground; band 6 (the thermal-infrared band) has an IFOV equivalent to a 120-meter square. 4.1.7 Manufacturer of Sensor/Instrument NASA Goddard Space Flight Center Greenbelt, MD 20771 Hughes Aircraft Corporation Santa Barbara, CA. 4.2 Calibration. The internal calibrator, a flex-pivot-mounted shutter assembly, is synchronized with the scan mirror, oscillating at the same 7-Hz frequency. During the turnaround period of the scan mirror, the shutter introduces the calibration source energy and a black direct-current restoration surface into the 100 detector fields of view. The calibration signals for bands 1 - 5 and band 7 are derived from three regulated tungsten-filament lamps. The calibration source for band 6 is a blackbody with three temperature selections, commanded from the ground. The method for transmitting radiation to the moving calibration shutter allows the tungsten lamps to provide radiation independently and to contribute proportionately to the illumination of all detectors. 4.2.1 Specifications The TM sensor is sensitive to the following spectral bands: Channel Wavelength (µm) Primary Use ------- --------------- ------------------------------------------ 1 0.451 - 0.521 Coastal water mapping, soil vegetation differentiation, deciduous/coniferous differentiation. 2 0.526 - 0.615 Green reflectance by healthy vegetation. 3 0.622 - 0.699 Chlorophyll absorption for plant species differentiation. 4 0.771 - 0.905 Biomass surveys, water body delineation. 5 1.564 - 1.790 Vegetation moisture measurement, Snow cloud differentiation. 6 10.450 - 12.460 Plant heat stress measurement, Other thermal mapping. 7 2.083 - 2.351 Hydrothermal mapping. Band Radiometric Sensitivity [NE(dP)]* ---- -------------------- 1 0.8% 2 0.5% 3 0.5% 4 0.5% 5 1.0% 6 0.5 K [NE(dT)] 7 2.4% Ground IFOV 30 m (bands 1-5, 7) 120 m (band 6) Avg. altitude 699.6 km Data rate 85 Mbps Quantization levels 256 Orbit angle 8.15 degrees Orbital nodal period 98.88 minutes Scan width 185 km Scan angle 14.9 degrees Image overlap 7.6 % * N.B. The radiometric sensitivities are the noise-equivalent reflectance differences for the reflective channels expressed as percentages [NE(dP)] and temperature differences for the thermal infrared bands [NE(dT)]. 4.2.1.1 Tolerance The TM channels were designed for a noise equivalent differential represented by the radiometric sensitivity shown in Section 4.2.1. 4.2.2 Frequency of Calibration The absolute radiometric calibration between bands on both sensors is maintained by using internal calibrators that are physically located between the telescope and the detectors and are sampled at the end of a scan. 4.2.3 Other Calibration Information Relative within-band radiometric calibration, to reduce "striping", is provided by a scene-based procedure called histogram equalization. The absolute accuracy and relative precision of this calibration scheme assumes that any change in the optics of the primary telescope or the "effective radiance" from the internal calibrator lamps is insignificant in comparison to the changes in detector sensitivity and electronic gain and bias with time and that the scene- dependent sampling is sufficiently precise for the required within-scan destriping from histogram equalization. Each TM reflective band and the internal calibrator lamps were calibrated prior to launch using lamps in integrating spheres that were in turn calibrated against lamps traceable to calibrated National Bureau of Standards lamps. Sometimes the absolute radiometric calibration constants in the "short-term" and "long-term parameters" files used for ground processing have been modified after launch because of inconsistency within or between bands, changes in the inherent dynamic range of the sensors, or a desire to make quantized and calibrated values from one sensor match those from another. 5. Data Acquisition Methods These data were acquired from the Landsat-5 TM sensor and received from the Canadian Centre for Remote Sensing (CCRS), who purchased it from the Earth Observation Satellite Company (EOSAT). As received from CCRS, the image had been processed from raw telemetry to a systematically corrected product within the CCRS MOSAICS system. 6. Observations 6.1 Data Notes This imagery was collected on 20-Aug-1988. This scene is Path 33, Row 21 in the Landsat Worldwide Reference System (WRS). The solar elevation angle at the time of image acquisition was 40.1 degrees. The solar azimuth angle was 146 degrees. The radiometric quality of this imagery was acceptable. 6.2 Field Notes Not applicable. 7. Data Description 7.1 Spatial Characteristics 7.1.1. Spatial Coverage The image area that was classified covers an area that is approximately 129 km by 86 km and includes areas just west of Thompson, Manitoba. The corners of the data set are as follows. These coordinates are in the BOREAS Grid projection. Corner X Y Longitude Latitude -------------------------------------------------------- Northwest 733.920 654.660 99.067W 56.313N Northeast 862.920 654.660 97.022W 56.099N Southwest 733.920 568.260 99.301W 55.549N Southeast 862.920 568.260 97.294W 55.339N 7.1.2 Spatial Coverage Map Not available. 7.1.3 Spatial Resolution Each pixel represents a 30-meter by 30-meter area on the ground. 7.1.4 Projection The area mapped is projected in the ellipsoidal version of the Albers Equal Area Conic (AEAC) projection. The projection has the following parameters: Datum: North American Datum of 1983 (NAD83) Ellipsoid: Geodetic Reference System of 1980 (GRS80) or Worldwide Geodetic System of 1984 (WGS84) Origin: 111.000°W 51.000°N Standard Parallels: 52° 30' 00"N 58° 30' 00"N Units of Measure: kilometers 7.1.5 Grid Description The data are referenced to the projection described in section 7.1.4. 7.2 Temporal Characteristics 7.2.1 Temporal Coverage This imagery was collected on 20-Aug-1988. This scene is Path 33, Row 21 in the Landsat Worldwide Reference System (WRS). The solar elevation angle at the time of image acquisition was 43.0 degrees. The solar azimuth angle was 148.8 degrees. The radiometric quality of this imagery was acceptable. 7.2.2 Temporal Coverage Map Not applicable. 7.2.3 Temporal Resolution This image was collected at a single point in time and is not multitemporal. 7.3 Data Characteristics 7.3.1. Parameter/Variable Land cover type. 7.3.2 Variable Description/Definition In a joint meeting of the Terrestrial Ecosystem (TE) modelers and the Remote Sensing Science (RSS) algorithm developers in Columbia, MD, July 1992, the following land cover classes were identified as necessary inputs to the TE models. This classification was performed using bands 1 - 5 and band 7 of a Landsat TM scene acquired on 20-Aug-1988. Class Descriptions Pixel Class Value Name ------------------ 1 Wet Conifer Wet Conifer is an area that contains coniferous trees, dominated by black spruce (picea mariana) or jack pine (pinus banksiana) growing on peat or poorly drained mineral soils that have a moss (typically sphagnum moss, sphagnum spp.) in conjunction with a herbaceous background. These areas are "non-dry" and will contain only small amounts of lichen on dry moss hummocks. 2 Dry Conifer Dry Conifer is an area that contains coniferous trees (primarily jack pine) with a lichen (cladina) background. These areas have sandy soils that are well drained. Areas of permafrost supporting conifers with a lichen background are also included in this class. 3 Mixed Deciduous and Coniferous Mixed Deciduous and Coniferous contains coniferous and aspen/birch (populus tremuloides/betula papyrifera) trees. The composition of this class contains less than 80% of the dominant species. 4 Deciduous The Deciduous class contains primarily aspen/birch. The composition of this class is generally greater than 80% deciduous trees. 5 Fen The Fen/Bog class is characterized by areas with a water table very near or at the surface. Fens experience lateral water transport whereas bogs are enclosed landforms experiencing only vertical transport. Fens typically contain sedges, moss, and bog birch associated with sparse to medium dense tamarack (larix laricina) stands. Bogs are usually treeless. 6 Water Water bodies such as ponds, lakes, and streams. 7 Disturbed The Disturbed class consists of areas that are dominated by bare soil, recently logged areas, or rock outcrops. This class also includes roads, airports, and urban areas. 8 Regeneration (Young) Areas that have been logged or burned. Logging is much more frequent in the Southern Study Area (SSA). These areas have usually been regenerating for 3 or 4 years since disturbance. These areas can contain aspen or jack pine replanted following logging. 9 Regeneration (Medium) Areas that have been burned or in some cases logged. These areas have typically been regrowing for up to 7 or 8 years since the disturbance. These areas are usually immature aspen or jack pine. 10 Regeneration (Older) Areas that burned and have been regenerating for about 12 to 20 years. These areas are usually mixed with class 4, indicating that the area is becoming recognizable as deciduous vegetation. 11 Recent Burn Areas that have been burned in the last 5 or 6 years. Distinguishable for their charred sphagnum background, they are usually areas of very intense burn where little or no vegetation survived. 7.3.3 Unit of Measurement Land Cover type - coded but unitless value. 7.3.4 Data Source Landsat-5 TM scene on 20-Aug-1988 from the CCRS. Forest cover maps of the area were acquired to identify training fields. These forest cover maps were from Forestry Canada and included species composition for various stands. The scale of the maps is 1:12500. These maps were compiled from aerial photography collected in 1984. 7.3.5 Data Range Pixel values of 0 to 11, stored as 8-bit integers. 7.4 Sample Data Record Not applicable for image data. 8. Data Organization 8.1 Data Granularity The smallest amount of data that can be ordered from these data is the entire data set. 8.2 Data Format 8.2.1 Uncompressed Data Files NSA classification product contains two files as follows: File 1: (80-byte American Standard Code for information Interchange (ASCII) text records) Documentation file File 2: (2,880 records of 4,300 bytes each) (1 byte per pixel) Classified image with values from 0 to 11. 8.2.2 Compressed CD-ROM Files On the BOREAS CD-ROMs, file 1 listed above is stored as ASCII text; however, file 2 has been compressed with the Gzip compression program (file name *.gz). These data have been compressed using gzip version 1.2.4 and the high compression (-9) option (Copyright (C) 1992-1993 Jean-loup Gailly). Gzip (GNU zip) uses the Lempel-Ziv algorithm (Welch, 1994) used in the zip and PKZIP programs. The compressed files may be uncompressed using gzip (-d option) or gunzip. Gzip is available from many websites (for example, ftp site prep.ai.mit.edu/pub/gnu/gzip-*.*) for a variety of operating systems in both executable and source code form. Versions of the decompression software for various systems are included on the CD-ROMs. 9. Data Manipulations 9.1 Formulae Not applicable. 9.1.1 Derivation Techniques and Algorithms The PCI programs CSG and MLC were used to create the needed statistics and the classified image. These programs should be referenced for specific formulae. 9.2 Data Processing Sequence 9.2.1 Data Processing Steps 1) No preprocessing of the image data was done before the maximum likelihood classification was performed. The raw digital numbers of the imagery were used as input for the classification. 2) Copy the ASCII and compress the binary files for release on CD-ROM. 9.2.2 Processing Changes None. 9.3 Calculations 9.3.1 Special Corrections/Adjustments None. 9.3.2 Calculated Variables Refer to PCI CSG and MLC programs. 9.4 Graphs and Plots None. 10. Errors 10.1 Sources of Error The sources of error in this classification can be attributed to several factors. In many cases, the spectral signature of one feature could be similar to the spectral signature of another feature, resulting in confusion. The similarity in spectral signatures could be the result of similar background components and variations in tree density. Error could also be a result of spectral mixing of various features that fall within a 30-meter pixel. 10.2 Quality Assessment 10.2.1 Data Validation by Source The imagery was spot checked at various locations, and the image class was compared to the forest cover map. An error assessment was performed on the classification. The BOREAS auxiliary sites were used as ground truth. The location of each auxiliary site was identified on the georeferenced image as a 3- by 3-pixel area. Each of the 9 pixels in these areas represents a test point. Many classes such as disturbed or water, were not represented by the auxiliary sites, and are not represented. 10.2.2 Confidence Level/Accuracy Judgment Although efforts have been made to make this classification as accurate as possible, there is bound to be some confusion between classes. In some areas, low productivity wet conifer or fen can be confused with the dry conifer, especially in areas of very young jack pine. 10.2.3 Measurement Error for Parameters and Variables The following tables and statistics were derived in assessing the classification accuracy: Confusion Matrix Classification Class 1 2 3 4 5 6 7 8 10 Truth ------------------------------------------------------------ Wet Conifer(1) 112 12 7 0 8 0 0 0 5 Dry Conifer(2) 4 45 20 3 0 0 9 0 0 Mixed (3) 0 0 12 6 0 0 0 0 0 Deciduous (4) 4 0 5 65 5 0 0 0 2 Fen (5) 0 0 0 0 9 0 0 0 0 Disturbed (7)* Regen.(Old)(10)* * no auxiliary sites available in this class. Class % Correct Wet Conifer 77.8% Dry Conifer 55.6% Mixed 66.7% Deciduous 80.2% Fen 100.0% Overall 73.0% Kappa = 0.633 or 63.3% better than chance agreement. (Campbell, 1987) 10.2.4 Additional Quality Assessments None. 10.2.5 Data Verification by Data Center This image was viewed to make sure that it matched the product description and appeared to be a classification image of the BOREAS NSA. 11. Notes 11.1 Limitations of the Data The user should keep in mind that these data are not 100% accurate. There may be errors or differences in class interpretation that might limit the use of these data for certain applications. 11.2 Known Problems With the Data Clouds in this classification show up in the disturbed class, and cloud shadows show up in the water class. These problems are readily apparent when looking at the imagery that was used to create this classification. 11.3 Usage Guidance Users should be aware of accuracy limitations as well as problems listed in Section 11.2. Before uncompressing the Gzip files on CD-ROM, be sure that you have enough disk space to hold the uncompressed data files. Then use the appropriate decompression program provided on the CD-ROM for your specific system. 11.4 Other Relevant Information None. 12. Application of the Data Set This data set can be used by anyone who wants to get an indication of the distribution of land cover types in the BOREAS NSA. See Section 1.4. 13. Future Modifications and Plans None. 14. Software 14.1 Software Description Various proprietary programs in the EASI/PACE image processing software from PCI, Inc. were used to classify the image. Questions related to the specific details of the software written to process this data set should be addressed to David Knapp (see Section 2.3). Gzip (GNU zip) uses the Lempel-Ziv algorithm (Welch, 1994) used in the zip and PKZIP commands. 14.2 Software Access EASI/PACE is a proprietary software package developed by PCI, Inc. Contact PCI for details. PCI, Inc. 50 West Wilmot St. Richmond Hill Ontario, Canada L4B 1M5 (905) 764-0614 (905) 764-9604 (fax) Gzip is available from many websites across the net (for example) ftp site prep.ai.mit.edu/pub/gnu/gzip-*.*) for a variety of operating systems in both executable and source code form. Versions of the decompression software for various systems are included on the CD-ROMs. 15. Data Access 15.1 Contact Information Ms. Beth Nelson BOREAS Data Manager NASA Goddard Space Flight Center Greenbelt, MD (301) 286-4005 (301) 286-0239 (fax) Elizabeth.Nelson@gsfc.nasa.gov 15.2 Data Center Identification See section 15.1. 15.3 Procedures for Obtaining Data Users may place data requests by telephone, electronic mail, or fax. 15.4 Data Center Status/Plans The TE-18 NSA maximum likelihood classification data are available from the EOSDIS ORNL DAAC (Earth Observing System Data and Information System) (Oak Ridge National Laboratory) (Distributed Active Archive Center). The BOREAS contact at ORNL is: ORNL DAAC User Services Oak Ridge National Laboratory Oak Ridge, TN (423) 241-3952 ornldaac@ornl.gov ornl@eos.nasa.gov 16. Output Products and Availability 16.1 Tape Products These data can be made available on 8 mm, Digital Archive Tape (DAT), or 9-track tapes at 6250 or 1600 BPI. 16.2 Film Products None. 16.3 Other Products These data are available on the BOREAS CD-ROM series. 17. References 17.1 Platform/Sensor/Instrument/Data Processing Documentation PACE Image Analysis Kernal Version 5.2. 1993. PCI Inc. Richmond Hill, Ontario. Richards, J. A. 1986. Remote Sensing Digital Image Analysis: An Introduction. Springer Verlag. Welch, T.A. 1984, A Technique for High Performance Data Compression, IEEE Computer, Vol. 17, No. 6, pp. 8 - 19. 17.2 Journal Articles and Study Reports Campbell, J.B. 1987. Introduction to Remote Sensing. Guilford Press. p.349. Sellers, P.and F. Hall. 1994. Boreal Ecosystem-Atmosphere Study: Experiment Plan. Version 1994-3.0, NASA BOREAS Report (EXPLAN 94). Sellers, P.and F. Hall. 1996. Boreal Ecosystem-Atmosphere Study: Experiment Plan. Version 1996-2.0, NASA BOREAS Report (EXPLAN 96). Sellers, P.and F. Hall. 1997. BOREAS Overview Paper. JGR Special Issue (in press). Sellers, P., F. Hall and K.F. Huemmrich. 1996. Boreal Ecosystem-Atmosphere Study: 1994 Operations. NASA BOREAS Report (OPS DOC 94). Sellers, P., F. Hall and K.F. Huemmrich. 1997. Boreal Ecosystem-Atmosphere Study: 1996 Operations. NASA BOREAS Report (OPS DOC 96). Sellers, P., F. Hall, H. Margolis, B. Kelly, D. Baldocchi, G. den Hartog, J. Cihlar, M.G. Ryan, B. Goodison, P. Crill, K.J. Ranson, D. Lettenmaier, and D.E. Wickland. 1995. The boreal ecosystem-atmosphere study (BOREAS): an overview and early results from the 1994 field year. Bulletin of the American Meteorological Society. 76(9):1549-1577. 17.3 Archive/DBMS Usage Documentation None. 18. Glossary of Terms None. 19. List of Acronyms AEAC - Albers Equal-Area Conic ASCII - American Standard Code for Information Interchange BOREAS - BOReal Ecosystem-Atmosphere Study BORIS - BOREAS Information System BPI - Bytes Per Inch CCRS - Canadian Centre for Remote Sensing CD-ROM - Compact Disk-Read-Only Memory DAAC - Distributed Active Archive Center DAT - Digital Archive Tape DEM - Digital Elevation Model EOS - Earth Observing System EOSAT - Earth Observing Satellite Company EOSDIS - EOS Data and Information System GMT - Greenwich Mean Time GRS80 - Geodetic Reference System of 1980 GSFC - Goddard Space Flight Center IFOV - Instantaneous Field of View MSA - Modeling Sub-Area MSS - Multispectral Scanner NAD27 - North American Datum of 1927 NAD83 - North American Datum of 1983 NASA - National Aeronautics and Space Administration NSA - Northern Study Area ORNL - Oak Ridge National Laboratory PANP - Prince Albert National Park RSS - Remote Sensing Science SSA - Southern Study Area TE - Terrestrial Ecology TM - Thematic Mapper URL - Uniform Resource Locator UTM - Universal Transverse Mercator WGS84 - World Geodetic System of 1984 WRS - Worldwide Reference System WWW - World Wide Web 20. Document Information 20.1 Document Revision Dates Written: 06-Apr-1995 Last Updated: 30-Jul-1998 20.2 Document Review Dates BORIS Review: 23-Dec-1997 Science Review: 20.3 Document ID 20.4 Citation These data were classified as a part of investigation TE-18, Principal Investigator F.G. Hall, using Landsat TM data from 20-Aug-1988 at 17:03:41 Greenwich Mean Time. The BOREAS science staff produced this data product. Any publication of these data should acknowledge the source of the data as the Canadian Centre for Remote Sensing (CCRS). 20.5 Document Curator 20.6 Document URL Keywords LAND COVER LANDSAT TM CLASSIFICATION TE18_NSA_Class_Max.doc 08/18/98