Skip to main content

DAAC Home > Get Data > Field Campaigns > BOREAS Follow-On > Data Set Documentation
BOREAS FOLLOW-ON DSP-01 LANDSAT TM LAND COVER MOSAIC OF THE BOREAS TRANSECT
Get Data

Summary

The objective of this land cover mosaic is to provide a data product that characterises the detailed land cover of a significant portion of the BOREAS Region. Seven Landsat-5 TM images have been assembled to completely cover the BOREAS Transect. Entire TM scenes were used to create this land cover map. A detailed classification scheme was employed, permitting the extraction of virtually all land cover information that can be discerned from digitally enhanced TM images. The data are provided in a binary image format file.
      Note that some of the data files have been compressed using Zip compression. See Section 8.2 for details.

Data Citation

Cite this data set as follows (citation revised on October 29, 2002):

Beaubien, J., R. Latifovic, J. Cihlar, and G. Simard. 2001. BOREAS Follow-On DSP-01 Landsat TM Land Cover Mosaic of the BOREAS Transect. Data set. Available on-line [http://www.daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.

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 Follow-On DSP-01 Landsat TM Land Cover Mosaic of the BOREAS Transect

1.2 Data Set Introduction
      This data set provides land cover description for a significant portion of the BOREAS Region, by employing 7 Thematic Mapper (TM) scenes that collectively span the BOREAS Transect between the Southern and the Northern Study Areas (Sellers et al., 1995). The land cover is characterised as belonging to one of 29 land cover classes (section 7.3.2).
      A detailed classification scheme was employed, permitting the extraction of virtually all land cover information that can be discerned in digitally enhanced TM images. However, it should be noted that this classification does not include a separate category for wetlands. This is because wetlands in this region are generally composed of grass and/or shrub cover that can have very similar spectral reflectance characteristics to non-wetland shrub and grassland classes and it is extremely difficult to discern the two with single-date imagery.

1.3 Objective/Purpose
      This classification was produced for BOREAS Follow-on investigators who needed a land cover data set with which to compare other measurements. It is an improved version of a 6-scene mosaic described by Beaubien et al (1999). The objective was to map land cover at the highest level of detail possible with TM data, especially for the forest. This was achieved by a series of pre-processing and classification steps, so that all the land cover detail discernible in digitally enhanced images could be retained for preparing the land cover map.

1.4 Summary of Parameters
      Land cover type (refer to section 7.3.2 for a list of classes)

1.5 Discussion
      The objective of this land cover map is to provide BOREAS regional scale modellers with a product that characterises the land cover for a significant portion of the BOREAS Region at the TM scale. An unsupervised classification technique was used, preceded by several pre-processing steps to facilitate the extraction of all relevant land cover information.

1.6 Related Data Sets
BOREAS Level-3b Landsat TM Imagery: At-sensor Radiance in BSQ Format
BOREAS TE-06 Biomass and Foliage Area Data
BOREAS TE-06 Allometry Data
BOREAS TE-13 Biometry Data
BOREAS TE-20 SSA Site Characteristics Data

Return to top of document


2. Investigators

2.1 Investigator Name and Title
      Dr. Josef Cihlar

2.2 Title of Investigation
      BOREAS Follow-on DSP-1 TM land cover and primary productivity in the boreal forest

2.3 Contact Information

Contact 1:
Jean Beaubien
Laurentian Forestry Centre
Sainte-Foy, Quebec
(418) 648-5822
beaubien@cfl.forestry.ca

Contact 2:
Rasim Latifovic
Canada Centre for Remote Sensing
Ottawa, Ontario
(613) 947-1816 (tel)
(613) 947-1385 (fax)
rasim.latifovic@ccrs.nrcan.gc.ca

Contact 3:
Josef Cihlar
Canada Centre for Remote Sensing
Ottawa, Ontario
(613) 947-1265
(613) 947-1406 (fax)
Josef.Cihlar@geocan.emr.ca

Contact 4:
Guy Simard
Laurentian Forestry Centre
Sainte Foy, Quebec
(418) 648-5822 (tel)
simard@cfl.forestry.ca

Return to top of document


3. Theory of Measurements

Satellite sensors measure radiation reflected or emitted by land surface cover and its components. The assumption behind this approach to land cover mapping is that the reflected solar radiation is highly correlated with differences in land cover that are of interest to investigators studying land cover processes.

Return to top of document


4. Equipment

4.1 Instrument Description
      The Landsat-5 Thematic Mapper (TM) sensor system records radiation in the visible, near-IR, and thermal wavelengths. See BOREAS TM documentation for additional details.
4.1.1 Collection Environment
      Landsat-5 orbits the Earth at an altitude of approximately 705 kilometres.

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.

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 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.

4.2.1 Specifications
      The TM sensor is sensitive to the following spectral bands:
Channel  Wavelength (um)                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.
 
4.2.1.1 Tolerance
      None given.


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 equalisation. 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 equalisation. 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.

Return to top of document

5. Data Acquisition Methods

These data were acquired from the Landsat-5 TM sensor and received from the Canada Centre for Remote Sensing (CCRS). As received, the image had been processed from raw telemetry to a systematically corrected product within the CCRS MOSAICS system (Friedel and Fisher, 1987). The calibration coefficients equations incorporating 1994 calibration updates not included in the IGARSS '94 are as follows, where D = days since launch (01-March-1984), L* = (DSL - Offset)/G, and DSL = digital signal level:
TM         Calibration gain         Characteristic   Solar irradiance
Spectral   (counts/(W/m**2/         wavelength       (W/m**2/
Band       sr/micrometre))          (micrometre)     micrometre)
--------   ---------------------    ------------     -----------
1          G=(-3.58E-05)*D+1.376    0.4863           1959.2
2          G=(-2.10E-05)*D+0.737    0.5706           1827.4
3          G=(-1.04E-05)*D+0.932    0.6607           1550.0
4          G=(-3.20E-06)*D+1.075    0.8382           1040.8
5          G=(-2.64E-05)*D+7.329    1.677             220.75
7          G=(-3.81E-04)*D+16.02    2.223              74.960
Return to top of document

6. Observations

6.1 Data Notes
      In general, the data quality was acceptable from the atmospheric transparency viewpoint, except for scene Path 33/Row 31 which appeared to be hazy over its SW part. Significant radiometric differences were also apparent as a result of seasonal (day of acquisition) and interannual differences, most likely due to varying precipitation regimes. These differences were dealt with at the classification and labelling stages.

6.2 Field Notes
      Field observations were collected as part of the accuracy assessment procedure (refer to section 10.2.1).

Return to top of document


7. Data Description

7.1 Spatial Characteristics
7.1.1. Spatial Coverage
      The seven images cover an area of 142854.15 km2, over 159 million pixels, 30 metres by 30 metres in size, and include both the BOREAS Southern and Northern Study Areas. Because of the orbital geometry, a small area between images Path/Row 37/21 and 35/21 was omitted when the images were selected.
Corner                  Longitude            Latitude
------------------      ----------------    ----------------
Upper Left  Corner      106°46'55.71" W     56°51'25.89" N
Upper Right Corner       96°06'30.92" W     56°51'25.89" N
Image Centre            101°26'43.32" W     54°50'25.87" N
Lower Left  Corner      106°46'55.71" W     52°49'25.86" N
Lower Right Corner       96°06'30.92" W     52°49'25.86" N
7.1.2 Spatial Coverage Map

7.1.3 Spatial Resolution
      Each pixel represents a 30-meter by 30-meter area on the ground.

7.1.4 Projection
      The projection is defined by the following parameters:

 Georeference Units             :   LONG/LAT
 Projection                     :   Geographic (geodetic)
 Datum - Ellipsoid              :   WGS 1984 (Global Definition) - WGS 84
7.1.5 Grid Description
      The data are referenced to the projection described in section 7.1.4.
Pixel Size (in Degrees)        :   0.00046  Lon   0.00029  Lat
   (Equivalent Deg,Min,Sec)    :   0° 00' 01.65"     0° 00' 01.03"


7.2 Temporal Characteristics

7.2.1 Temporal Coverage
      See section 7.2.2.

7.2.2 Temporal Coverage Map
      The data that were used to produce this classification were collected by the Landsat-5 TM on the following dates:

Path   Row    Date and time of acquisition               Scene centre
----   ---    ----------------------------      ------------------------------
 33     21        June 12, 1992 16:50:20        55° 54' 46" N /  97° 59' 49" W
 34     21      August 06, 1992 16:56:30        55° 54' 42" N /  99° 32' 24" W
 35     21      August 11, 1991 17:06:50        55° 54' 48" N / 100° 59' 16" W
 37     21      August 28, 1998 17:36:24        55° 24' 44" N / 104° 18' 19" W
 35     22      August 11, 1991 17:06:50        54° 29' 37" N / 101° 43' 48" W
 36     22        July 30, 1996 17:00:42        53° 46' 34" N / 103° 26' 37" W
 37     22      August 09, 1991 17:14:50        53° 59' 30" N / 105° 06' 43" W
7.2.3 Temporal Resolution
      Each TM scene was collected at a single point in time as indicated in section 7.2.2.


7.3 Data Characteristics

7.3.1. Parameter/Variable
      Land cover type.

7.3.2 Variable Description/Definition
     The classification scheme was designed to meet three objectives: 1) retain as much information about land cover in the Region as possible, especially regarding forest types and conditions; 2) be compatible with other classifications employed by boreal ecosystem investigators; and 3) be applicable to Landsat TM (i.e. all the necessary information to differentiate between the classes should be available in a single-scene TM image).
 

Class Descriptions
Two sets of pixel values are shown for each class. The number preceding the class name refers to the final classification, each class having a unique identifier. The second set of numbers (after the name) refers to spectral clusters that have been agglomerated to produce the final class; they thus provide additional (spectral) information. Note that the file "tm_mosaic_cls1" (section 8.2.1) contains the first set of pixel identifiers, and the file "tm_mosaic_cls2" contains the second set.

Pixel    Class
Value    Name
------------------
  7 - Coniferous high crown density black spruce (includes spectral 
      clusters 6, 7)
 11 - Coniferous high crown density black spruce and Jack pine (11, 12, 15, 18)
 21 - Coniferous high crown density black spruce younger (21, 19, 17, 31)
 22 - Coniferous medium crown density jack pine (22)
 32 - Coniferous medium crown density black spruce, jack pine (33, 32, 34, 29, 
      10)
 25 - Coniferous medium crown density black spruce (20, 25, 28)
 43 - Coniferous low crown density black spruce, jack pine (43, 41)
 59 - Coniferous low crown density jack pine (59, 37, 49)
 55 - Coniferous very low density (55, 52, 50, 65)
 79 - Deciduous high crown density (83, 79, 78, 88, 86, 93)
 80 - Deciduous medium crown density (80, 75, 68, 66)
 99 - Deciduous low broadleaf cover (99, 111)
 39 - Mixed coniferous high density (39)
 36 - Mixed coniferous medium density (36)
 69 - Mixed deciduous forest (69, 73, 62)
 53 - Mixed forest (53, 48)
 85 - Shrubs and grassland (85, 72, 71, 126)
 13 - Burn recent bare area (13)
 35 - Burn recent sparse vegetation cover (35, 30, 54, 63, 47)
113 - Burn rock outcrops (51, 113, 130, 131)



 81 - Older burns shrub-grass cover (81, 106, 84, 109)
 64 - Old burns mixed regeneration cover (64, 77)
134 - Bare disturbed area (134, 143)
112 - Bare disturbed areas sparse vegetation cover (133, 112, 120, 123, 118, 
      149)
160 - Cropland high biomass (160)
161 - Cropland medium biomass (161)
162 - Cropland low biomass (162)
  1 - Water (1, 2, 3, 4, 5, 9)
150 - Clouds (150, 128, 127)
The class descriptions follow below. Note that the sole purpose of the numbering scheme used here is to indicate the hierarchy among the categories.
Class 1.0 Coniferous (>80% coniferous trees)
Class 1.1 High crown density (>60%)
Class 1.1.1 Black spruce (DN = 7)
This class, especially prevalent in the Northeast portion of the mapped area, contains older black spruce (BS) forest (older than ~40 years). 
Class 1.1.2 Black spruce, Jack pine (DN = 11)
This class is also dominated by old black spruce forest, like 1.1.1, but is more widely spread in the region and commonly includes jack pine (JP) trees; if desirable, 1.1.1 and 1.1.2 may be merged. 
Class 1.1.3 Black spruce, younger (DN = 21)
As most of this region was repetitively burned in the past, this class is particularly dispersed in the area. It may contain a small proportion of broadleaf species (<20%) and some jack pine stems; ëyoungerí refers to approximately 20-40 years. 
Class 1.2 Medium crown density (40-60%)
In this region, medium crown density forests are generally young (~ 20-40 years). It should be noted that the species (BS or JP) may influence the apparent density. For example, a dense (>60%) jack pine stand may be part of a medium density class because of the foliage pattern and the influence of the ground cover (particularly lichens) more easily apparent then under a dense black spruce stand.
Class 1.2.1 Jack pine (DN = 22)
Typically, a jack pine forest on a drier site with lichen commonly present in the understory. 
Class 1.2.2 Black spruce, Jack pine (DN = 32)
Lichens may also be present, but generally has a green understory; this class is particularly widespread in the region. 
Class 1.2.3 Black spruce (DN = 25)
Generally located on more humid sites, this class can include a small proportion of broadleaf species. 
Class 1.3 Low crown density (25-40%)
Class 1.3.1 Black spruce, Jack pine (DN = 43)
Commonly located on wet sites with a proportion of broadleaf vegetation (mostly shrubs); occasionally young tree canopy after older burns (> 10 years). 
Class 1.3.2 Jack pine (DN = 59)
Lichens commonly present in the understory; occasionally this class is located in burned areas with mixed cover (dead stems, bare soil, rock outcrops); an understory with abundant lichens seems to lower the apparent density. There is confusion in this class with some dried-up wetlands. 
Class 1.4 Very low crown density (10-25%) (DN = 55)
Commonly located on wet sites with a proportion of broadleaf vegetation (mostly shrubs) in the southern portion of the region; in the northern areas, some regenerating cover after burns.
Class 2.0 Deciduous (>80% deciduous trees)
Class 2.1 High crown density (> 60%) (DN = 79)
Almost exclusively in the southern part of the area; elsewhere some rare younger, dense broadleaf cover (or high shrubs) can be in this class.
Class 2.2 Medium crown density (40-60%) (DN = 80)
Mainly in the southern part of the area in the vicinity of high-density deciduous forest; transitions between high and medium densities may frequently occur locally; the density values refer mainly to the dominant broadleaf trees; some conifers may be present in lower strata. After perturbations (fires), this class may combine young and short broadleaf cover with possible conifer regeneration.
Class 2.3 Low broadleaf cover (DN = 99)
Mostly in the southern part of the area, after perturbations such as previous cut-overs, burns, and cropland areas with shrub cover.
Class 3.0 Mixed forest (20-80% coniferous or deciduous trees)
Class 3.1 Mixed coniferous (> 60% coniferous trees)
Class 3.1.1 High crown density (>60%) (DN = 39)
Mainly among southern mixed forest, some patches exist in northern old burns (>10 years). 
Class 3.1.2 Medium crown density (40-60%) (DN = 36)
Mainly young forest cover in northern old burns. 
Class 3.2 Mixed deciduous forest (> 60% deciduous, density > 60%) (DN = 69)
Mainly located in the southern part of the area within high crown density deciduous forest class of high crown density.
Class 3.3 Mixed forest (40-60% coniferous/deciduous, density > 60%) (DN = 53)
Among southern broadleaf and mixed forest cover, but also numerous patches of young forest cover after old burns in northern areas.
Class 4.0 Open land (tree crown density <10%)
Class 4.1 Shrubs and grassland (DN = 85)
Typically contains wetlands in southern areas; very few patches after more or less recent burns; some trees can be present.
Class 4.2 Burns
Note that some TM scenes predate fieldwork by up to seven years; during this period, vegetation on burns has regenerated.
Class 4.2.1 Bare areas, recent burns (DN = 13)
Dead standing stems are commonly present. Confusion exists with some patches of medium density jack pine cover class with lichen understory, class 22. 
Class 4.2.2 Rock outcrops (DN = 113)
Rock outcrops with very sparse vegetation after burns. 
Class 4.2.3 Recent burns, sparse vegetation cover (~ < 10 years) (DN = 35)
Conifer regeneration can be present; also some patches of open land with partial green vegetation, like some wetlands. 
Class 4.2.4 Older burns, shrub-grass cover (DN = 81)
Conifer regeneration can be present; also some patches of similar cover within wetlands. 
Class 4.2.5 Old burns, mixed regeneration (DN = 64)
Also mixed low cover in wetland areas. 
Class 4.3 Bare disturbed areas (DN = 134)
Very sparse vegetation cover, mainly located in areas recently disturbed by cut-overs in the southern portion of the area.
Class 4.4 Disturbed areas, sparse vegetation cover (DN = 112)
This class with sparse vegetation cover is located mainly in areas recently disturbed by cut-overs in the southern part of the area; a few exceptions are some patches in burned areas; conifer regeneration can be present (recall the acquisition dates of the imagery).
Class 5.0 Cropland
No attempt was made to differentiate crop types, because the primary interest is forest cover, and the varying acquisition dates make crop identification difficult. The categories below refer broadly to different crop types, but the phenological differences cause large confusion.
Class 5.1 High biomass (DN = 160)
Class 5.2 Medium biomass (DN = 161)
Class 5.3 Low biomass (DN = 162)
Class 6.0 Non-vegetated land
Class 6.1 Water bodies (DN = 1)
This class contains some cloud shadows
Class 6.2 Clouds (DN = 150)

7.3.3 Unit of Measurement
     Land Cover type - coded but unitless value.

7.3.4 Data Source
     Landsat-5 TM (refer to section 4.).Forest cover maps of the area were acquired to identify training fields. The following basic source maps were used:

  • Manitoba Department of Natural Resources, Provincial Forest Inventory Maps. These maps are compiled from aerial photographs at a scale of 1:15,840. Each map covers an area of about 93 km2. Maps identify stands by number, assigned within each map; a separate legend sheet for each map provides details about each stand.
  • Saskatchewan Department of Tourism and Renewable Resources, Inventory Maintenance Maps. These maps are compiled from aerial photographs at a scale of 1:12,500. Each map covers a 10x10 km area. Stand information code on the map include stand number, species age, height, and canopy information.
7.3.5 Data Range
     Pixel values of 0 to 162, stored as 8-bit integers.


7.4 Sample Data Record
     Not applicable for image data.

Return to top of document


8. Data Organisation

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
      The classification product contains two image files, each are unsigned 8 bit binary files, with a size of 23341 pixels by 14139 lines.
  • tm_mosaic_cls1.img: Classified image with values from 1 to 150 with zero values (0) used as fillers for non-image (no data) areas of the image: contains a unique identifier for each final class (refer to section 7.3.2).
  • tm_mosaic_cls2.img: Classified image with values from 1 to 150 with zero values (0) used as fillers for non-image (no data) areas of the image: contains a unique identifier for each spectral cluster; one or more clusters were merged to form the final classes (refer to section 7.3.2).
The image files have been compressed with the MS Windows-standard Zip compression scheme. These files were compressed using Aladdin's DropZip on a Macintosh. DropZip uses the Lempel-Ziv algorithm (Welch, 1994), also used in Zip and PKZIP programs. The compressed files may be uncompressed using PKZIP (with the -expand option) on MS Windows and UNIX, or with StuffIt Expander on the Mac OS. You can get newer versions from the PKZIP Web site at http://www.pkware.com/download-software/ [Internet Link].
Return to top of document

9. Data Manipulations

9.1 Formulae
      Not applicable.
9.1.1 Derivation Techniques and Algorithms
     None given.
9.2 Data Processing Sequence
9.2.1 Data Processing Steps
  1. Image selection and pre-processing

  2.       The intent was to obtain land cover classification compatible with the BOREAS experimental period (1994-1996). Given the limitations of data availability (primarily due to cloud cover), images from several years were used. For the most part, the resulting classification is representative of land cover during BOREAS because of the low rate of land cover change during the 1990's. The selected images are listed in section 7.2.2.
          A new radiometric normalisation procedure was used to establish radiometric consistency among the seven TM scenes. The principle of the method is to employ overlapping sections among the scenes, assuming that there are at least some areas within these whose reflectance properties have not changed between the imaging dates (Du et al., 2000). The procedure avoids any loss of radiometric detail during the normalisation, and ensures consistency across the entire mosaic, regardless of the sequence of scenes in the mosaicking.
  3. Image classification

  4.       A modified version of the Enhancement-Classification Method (ECM) (Beaubien et al., 1999) was used to classify the image mosaic. The ECM procedure is intended to capture most of the information visible in an enhanced image by converting the image into a classification through various generalization steps. It was designed to be as reproducible as possible for an interactive classification. The main modification to the original methodology is the use of an unsupervised classifier in step #2 below instead of image quantization. Image filtering, employed to help in the selection of the more important image mean values (signatures), was also modified. In summary, the modified ECM consists of the following phases:
    1. Image contrast enhancement

    2.       The purpose of the enhancement is to bring out the distinctions among various classes of interest so that they can be more easily viewed in the subsequent steps. It is carried out in three steps: viewing of the data, sampling of cover types with low and high values in each spectral band, and digital stretch.
    3. Unsupervised classification

    4.       The purpose of this part of the process is to produce a first classification, which captures most of the information visible in the initial enhanced image. A K-Means method (Tou and Gonzales, 1974) was used to classify the stretched image into 150 clusters. After computing a pseudo colour table (PCT) for the enhanced image, the classified image closely resembles the enhanced input image.
    5. Minimum distance re-classification

    6.       A second level of generalization is reached by a re-classification from a selection of the more significant signatures generated by the unsupervised classifier. This is carried out in three steps, first, by image filtering (mode and sieve filters) to "flatten" the image by making classes more uniform over groups of spatially adjacent pixels. The next step includes selection of all visible (thus more significant) cover types. Lastly, a minimum distance classification on the initial enhanced image according to the signatures of the selected cover types. To reduce the number of clusters before manual agglomeration, those three steps are optionally repeated. After each iteration, the classification results are carefully compared to the initial unsupervised classification, using the same pseudo-colour table. Other signatures may be added if suitable.
    7. Cluster agglomeration and labeling

    8.       The purpose of this step is to group spectral clusters representing the same ground cover types. Since the last classified image closely resembles the original image, where image colours associated with the basic cover types are known, a substantial degree of agglomeration can be completed on the basis of colour patterns. Additional tools may include examination of the cluster mean digital value through a distance table, and spatial proximity values between clusters. For the remainder, independent knowledge of ground conditions in specific areas is required, as in other classification methods.
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
      The area occupied by individual classes is as follows:

Pixel
Number     Area [km2]       Percent
-----------------------------------
7           4551.1461       3.1859%
11         11184.5817       7.8294%
21         12431.7315       8.7024%
22          3377.6496       2.3644%
32         16748.0874      11.7239%
25         10621.206        7.4350%
43          7990.4637       5.5934%
59          5389.1595       3.7725%
55         13722.4071       9.6059%
79          3908.4408       2.7360%
80          5158.6686       3.6111%
99          1160.6319       0.8125%
39          2958.075        2.0707%
36          2026.5147       1.4186%
69          3308.2668       2.3158%
53          4043.7774       2.8307%
85          4177.5588       2.9244%
13           861.1092       0.6028%
35         10667.6442       7.4675%
113         1683.2673       1.1783%
81          5954.6016       4.1683%
64          4617.3114       3.2322%
134          363.4947       0.2545%
112         1016.1828       0.7113%
160         2315.1312       1.6206%
161          799.0407       0.5593%
162         1818.0036       1.2726%
-----------------------------------
Total     142854.15          100%
Return to top of document

10. Errors

10.1 Sources of Error
      This classification does not include a separate category for wetlands. This is because wetlands in this region are generally composed of grass and/or shrub cover with different heights, densities, and the ratios of live-to-dead leaf biomass. Consequently, they can have very similar spectral reflectance characteristics to shrub, grassland, and some burn classes. It might be possible to use late spring images as an additional data source because surface water should have a stronger influence on the spectral reflectance. Alternatively, radar images might be helpful. However, such differentiation does not appear to be feasible with a single-date TM image taken during the growing season. Thus, most wetland cover types are in the shrub and grassland class (85), and some are contained within the burn classes (81, 64, 35). A similar situation is present in the southern part of the mosaic where grass fields in agriculture area are classified as class 81.

10.2 Quality Assessment

10.2.1 Data Validation by Source
      The classification was checked qualitatively using aerial photographs, and quantitatively using field data. Three types of field sites were used:
  1. BOREAS auxiliary sites 'B': these sites were carefully documented using aerial photographs, and described in the field using forest mensuration and site description methods (Halliwell and Apps, 1997). A total of 55 sites were documented in this manner.
  2. Additional sites: in late September and early October of 1998, a field survey was undertaken to obtain additional data for accuracy assessment. Due to access limitations and the large area to be covered, the criteria for site locations were the representativeness of the various forest classes and conditions, sufficiently large patch size (to compensate for potential geolocation uncertainties), and proximity to the existing road network. This was accomplished by pre-processing the mosaicked TM image to locate spectrally homogenous patches, overlaying the road network, and printing the result at a large scale for use in the field. The above procedure resulted in a somewhat uneven distribution of test sites, as indicated in the table below, but further improvements were not feasible with the available resources.

  3.       Two types of sites were used: 'S', sites with photographs (overstory, understory, horizontal view) and physiognomic descriptions (overstory, understory); and 'M', sites with a photograph (horizontal view) and vegetation class identification. The location of each 'S' site was established with a GPS, and each 'M' site by landmarks on the TM enhancements. A total of 94 'M' sites and 89 'S' sites were located across the region. In addition, high-resolution GPS point measurements were taken to facilitate accurate registration of the TM mosaic.
For the accuracy assessment, each site from the BOREAS auxiliary site set and the additional set was assigned to one of the 29 classes, based on a consensus of two interpreters. The corresponding class on the map was defined to be the majority class occurring in the 3x3 window centred on the site.

10.2.2 Confidence Level
      Based on the qualitative and quantitative accuracy assessments (section 10.2.1, 10.2.3), we have high confidence in the accuracy of the classification. As noted above, the main limitation is absence of a wetland class. In addition, accuracies could be lower in areas from which no field sites were located, e.g. areas farther from the existing road network.

10.2.3 Measurement Error for Parameters and Variables
      The following tables and statistics were derived in assessing the classification accuracy:

A. Number of sites by category and agreement

      Class      'M' Sites | 'S' Sites | 'B' Sites | All Sites    Pct
No.   Name       No. Match | No. Match | No. Match | No. Match    Match
------------------------------------------------------------------------
7    CHD BS       6    5      3    3      3    3     12   11      0.917
11   CHD BS/JP    1    1      1    1      4    4      6    6      1.000
21   CHD BSY      7    5     14   14      6    6     27   25      0.926
22   CMD JP       5    5      4    4      2    2     11   11      1.000
32   CMD BS/JP    4    4      8    7     10    8     22   19      0.864
25   CMD BS       3    3      8    7      3    1     14   11      0.786
43   CLD BS/JP    1    1      1    1      0    0      2    2      1.000
59   CLD JP       6    6      7    7      3    3     16   16      1.000
55   CVLD         4    4      3    2      5    2     12    8      0.667
79   BHD          8    8      7    7      6    6     21   21      1.000
80   BMD          5    5      3    2      0    0      8    7      0.875
99   BLD          3    3      0    0      0    0      3    3      1.000
39   MCHD         0    0      3    3      0    0      3    3      1.000
36   MCMD         2    2      2    2      2    2      6    6      1.000
69   MB           5    3      4    2      5    4     14    9      0.643
53   MIX          2    1      6    6      4    3     12   10      0.833
85   SHR/GRASS    3    3      2    2      0    0      5    5      1.000
13   B-recent     1    1      0    0      0    0      1    1      1.000
35   B-recent     4    3      1    1      0    0      5    4      0.800
113  B-rock-out-  2    2      1    1      0    0      3    3      1.000
      crops
81   B-old       11   11      6    6      1    1     18   18      1.000
      shrub/grass
64   B-old mixed  7    6      3    3      1    1     11   10      0.909
      regeneration
134  Bare         2    2      1    1      0    0      3    3      1.000
      disturbed
112  Bare         2    2      1    1      0    0      3    3      1.000
      disturbed sparse
160  CROP/HB*
161  CROP/MB*
162  CROP/LB*
1    WATER*
150  CLOUD*
------------------------------------------------------------------------
TOTAL            94   86     89   83     55   46    238  215      0.903

* no auxiliary sites available for this class
B. Confusion matrix
      The following confusion matrix shows results of the accuracy assessment. It is evident that the overall accuracy is quite high. The confusion that did occur was mostly between adjacent density classes or the proportions of coniferous and deciduous classes.

Error Matrix of the Classification Map BOREAS MOSAIC V3

Summarising the classes into major groups, the classification accuracy was as follows:

Class           % Correct
-------------------------
Coniferous        90.0%
Deciduous         96.9%
Mixed             80.0%
Disturbed/other   95.9%
------------------------
Overall Correct   90.7
The overall kappa value (Congalton, 1991) was 0.8977 or 89.8% better than chance agreement.
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 a part of the BOREAS region.

Return to top of document

11. Notes

11.1 Limitations of the Data
      The user should keep in mind that the original images were taken in different years. The class descriptions and the map accuracy should also be considered in using this product.

11.2 Known Problems with the Data
      None

11.3 Usage Guidance
      Users should be aware of accuracy limitations as well as problems listed in Section 10.

11.4 Other Relevant Information
      None.

Return to top of document


12. Application of the Data Set

This land cover map was created to provide BOREAS regional scale modelers with a driver data set that characterises the land cover for a significant portion of the BOREAS region at a relatively fine scale.

Return to top of document


13. Future Modifications and Plans

None given .

Return to top of document


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. 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)
Zip software is available from many Web sites across the Internet. You can get newer versions from the PKZIP Web site at http://www.pkware.com/download-software/ [Internet Link]. Versions of the decompression software for MS Windows, Mac OS, and several varieties of UNIX systems are included on the CD-ROMs.

Return to top of document


15. Data Access

15.1 Contact for Data Center/Data Access Information
      These BOREAS data are available from the Earth Observing System Data and Information System (EOS-DIS) Oak Ridge National Laboratory (ORNL) Distributed Active Archive Center (DAAC). The BOREAS contact at ORNL is:

ORNL DAAC User Services
Oak Ridge National Laboratory
(865) 241-3952
ornldaac@ornl.gov
ornl@eos.nasa.gov

15.2 Procedures for Obtaining Data
      BOREAS data may be obtained through the ORNL DAAC World Wide Web site at http://www.daac.ornl.gov/ [Internet Link] or users may place requests for data by telephone or electronic mail.

15.3 Output Products and Availability
      Requested data can be provided electronically on the ORNL DAAC's anonymous FTP site or on various media including, CD-ROMs, 8-MM tapes, or diskettes.

Return to top of document


16. Output Products and Availability

16.1 CD-ROM Products
      These data can be made available on CD-ROM.

16.2 Film Products
      None.

16.3 Other Products
      None.

Return to top of document


17. References

17.1 Platform/Sensor/Instrument/Data Processing Documentation
Friedel J. P., Fisher T. A. 1987. Mosaics - a system to produce state-of-the-art satellite imagery for resource managers. Geocarto International 2(3): 5-12.

PACE Image Analysis Kernel 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.

Du, Y., Cihlar, J., Beaubien, J., and Latifovic, R. 2000. Radiometric normalization, compositing and quality control for satellite high resolution image mosaics over large areas. IEEE Transactions on Geoscience and Remote Sensing (submitted).
 

17.2 Journal Articles and Study Reports
Beaubien, J., J. Cihlar, G. Simard, and R. Latifovic. 1999. Land cover from multiple Thematic Mapper scenes using a new enhancement-classification methodology. Journal of Geophysical Research 104(D22): 27909-27920.

Congalton, R.G. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment 37: 35-41.

Halliwell, D.H., and M. J. Apps. 1997. BOReal Ecosystem-Atmosphere Study (BOREAS biometry and auxiliary sites: locations and descriptions. Canadian Forest Service, Natural Resources Canada, Edmonton, Alberta. 120p.

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.

Tou, J.T., and R.C. Gonzales. 1974. Pattern recognition principles. Addison-Wesley Publishing Co.
 

17.3 Archive/DBMS Usage Documentation
      None.

Return to top of document


18. Glossary of Terms

None.

Return to top of document


19. List of Acronyms

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
EOS       - Earth Observing System
EOSAT     - Earth Observing Satellite Company
EOSDIS    - EOS Data and Information System
GMT       - Greenwich Mean Time
GPS       - Global Positioning System
IFOV      - Instantaneous Field of View
LCC       - Lambert Conformal Conic projection
MSA       - Modeling Sub-Area
MSS       - Multispectral Scanner
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
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
Return to top of document

20. Document Information

20.1 Document Revision Dates

Written: 06-Jul-2000
Last Updated: 10-Jan-2001 (citation revised on 29-Oct-2002)

20.2 Document Review Dates

Science Review: 12-Dec-2000

20.3 Document ID

dsp01_tm

20.4 Citation

Cite this data set as follows (citation revised on October 29, 2002):

Beaubien, J., R. Latifovic, J. Cihlar, and G. Simard. 2001. BOREAS Follow-On DSP-01 Landsat TM Land Cover Mosaic of the BOREAS Transect. Data set. Available on-line [http://www.daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.

20.5 Document Curator:

webmaster@daac.ornl.gov

20.6 Document URL:

http://daac.ornl.gov/BOREAS/FollowOn/guides/dsp01_tm_landcover_doc.html

Keywords
LAND COVER
LANDSAT TM
CLASSIFICATION

Return to top of document