The Campaign
- Soil Moisture
- Soil Properties
- Soil Temperature
- Vegetation and Land Cover
- Aircraft Remote Sensing
- Satellite Remote Sensing
|
Land Cover Classification of the SGP99 Region
- Collected all available "cloud-free" Landsat-5 (March 9,
May 12, and July 15) and Landsat-7 scenes (July 7 and July
23) from March 9 to July 23, 1999. No Landsat images for the
months of April or June were available from Space Imaging or
EOSAT.
- Imported the 3 Landsat-5 (March 9, May 12, and July 15)
and the two Landsat-7 scenes (July 7 and 23) into GIS
software (PCIworks)
- Mosaic the two Landsat scenes (Path 28, Rows # 35-36)
together for each date
- Screen digitized training sites from ground survey data
collected during July 7-July 22, 1999. The ground-survey data
identified 44 different land-cover categories
- Re-grouped the 44 different land-use categories into the
following 15 land-cover categories
- Alfalfa
- Bare soil
- Corn
- Pasture grazed
- Legume
- Pasture ungrazed
- Trees
- Urban
- Water
- Wheat stubble
- Bare ground with wheat stubble
- Bare ground with green vegetation
- Shrubs
- Sand and quarries
- Outcrops
- The July 7th scenes had too many clouds and were
discarded. Of the other scenes, March 9 had the most clouds,
July 23 had the least clouds, and May 12 and July 15 have
intermediate cloud cover relative to March 9 and July
23.
- Created cloud and shadow masks for each image by screen
digitizing cloud and shadow training sites and determining
the mean digital numbers of these training sites from their
signature statistics. The following values define the
threshold digital numbers used to remove clouds and shadows
from each image (where # designates the channel number):
- March 9:
Clouds: removed pixels where #1 > 90 and #2 > 65 and
#3 > 80.
Shadows: removed pixels where #1 < 55 and #2 < 20
and #3 < 20 and #4 < 25.
- May12:
Clouds: removed pixels where #1 > 110 and #2 > 56
and #3 > 67 and #4 > 100.
Shadows: removed pixels where #1 < 56 and #2 < 30
and #3 < 30 and #4 < 70.
- July 15:
Clouds: removed pixels where #1 > 120 and #2 > 150
and #3 > 150.
Shadows: removed pixels where #1 < 65 and #2 < 25
and #3 < 25 and #4 < 32.
- July 23:
Clouds: removed pixels where #1 > 110 and #2 > 110
and #3 > 160 and #4 > 190.
Shadows: removed pixels where #1 < 73 and #2 < 50 and #3 < 50 and #4 < 110.
- Created two final cloud/shadow masks by combining masks
from the March9/May12/July15/July23 images and combining
masks from the July15/July23 images.
- Performed two maximum likelihood classifications by
grouping the images into two time series -
March9/May12/July15/July23 and July15/July23.
- Defined the following parameters during the
classification:
- Working channels as bands 3 (red), 4
(reflective near infrared), 5 (reflective mid-infrared),
and 7 (reflective mid-infrared) for each mosaic image.
- Training sites by converting the digitized
training site vectors into bitmaps and importing these
bitmaps into the training site classification channel.
- Threshold and bias values for the training site
categories by using the default values for all
categories.
- Masks by converting the combined cloud/shadow
masks described in step #7 into bitmaps
- Created classification images for the two time series
images with the 15 land-cover categories described in step
5.
- Overall accuracy for the March/May/July classification
was 72.60%, while overall accuracy for the July15/July23
classification was 70.15%.
- The March/May/July classification was chosen as the
primary classification because the accuracy was better than
the July15/July23 classification and visual inspection
revealed better classification results in the study
areas.
- The July15/July23 classification image was merged into
the May/March/July classification image by substituting the
unclassified pixels from the May/March/July cloud/shadow mask
with pixels from the July15/July23 image.
- The confusion and separability matrices indicated most
classes had good separability (above 1.9 on a scale from 0.0
to 2.0). The classes with less than good separability
included the following:
Separability table
Category name |
Category name |
Separability |
Pasture ungrazed |
Pasture grazed |
0.639471 |
Bare soil with wheat stubble |
Bare soil |
1.211651 |
Shrubs |
Trees |
1.333874 |
Bare soil with green vegetation |
Bare soil |
1.373694 |
Bare soil with wheat stubble |
Wheat stubble |
1.507923 |
Bare soil with green vegetation |
Bare soil with wheat stubble |
1.594767 |
Bare soil with green vegetation |
Pasture ungrazed |
1.623797 |
Legume |
Corn |
1.667574 |
Bare soil with green vegetation |
Pasture grazed |
1.746619 |
Pasture grazed |
Alfalfa |
1.747247 |
Shrubs |
Pasture grazed |
1.793274 |
Pasture ungrazed |
Corn |
1.817496 |
Bare soil with green vegetation |
Corn |
1.825002 |
Shrubs |
Pasture ungrazed |
1.836492 |
Bare soil with green vegetation |
Alfalfa |
1.837461 |
Pasture ungrazed |
Alfalfa |
1.846080 |
Pasture grazed |
Corn |
1.876988 |
Wheat stubble |
Bare with wheat stubble |
1.888443 |
Bare soil with wheat stubble |
Pastu4re ungrazed |
1.897055 |
Bare soil with green vegetation |
Wheat stubble |
1.898382 |
Data File Specifications
Data File Specifications
Data type |
8 bit binary |
Projection |
UTM |
Datum |
NAD27 |
Ellipsoid |
Clark 1866 |
Zone |
14 S |
Units |
meters |
No of pixels |
8559 |
No of lines |
12359 |
X min |
456000.000 E |
X max |
712770.000 E |
Y min |
3724830.000 N |
Y max |
4095600.000 N |
Pixel Size |
30.000 E 30.000 N |
Land cover categories: |
1 |
Alfalfa |
2 |
Bare soil |
3 |
Corn |
4 |
Pasture grazed |
5 |
Legume |
6 |
Pasture ungrazed |
7 |
Trees |
8 |
Urban |
9 |
Water |
10 |
Wheat stubble |
11 |
Bare ground with wheat stubble |
12 |
Bare ground with green vegetation |
13 |
Shrubs |
14 |
Sand bars and quarries |
15 |
Outcrops |
File Names and Formats
File Names and Format Information
File name |
File type |
Format |
Description |
File size |
Storage required |
LC99.bil |
Binary |
Layout BIL
NROWS 12359
NCOLS 8559
NBANDS 1
NBITS 8 |
Land-cover classification for the SGP99 study
area |
~105.8 MB |
~106 MB
|
LC99.gif |
Image |
Gif |
Gif file displaying the binary file |
~ 43 KB |
FTP Site
The Land Cover Classification data set from SGP99 resides
on DAAC anonymous FTP. You may access it from this
document,
Land
Cover Classification Data
- or directly via FTP at
- ftp disc.gsfc.nasa.gov
- login: anonymous
- password: < your internet address >
- cd /data/sgp99/LandCover/
Points of
Contact
The Principal Investigator for the SGP99 Land Cover
Classification data is
Thomas J. Jackson
USDA ARS Hydrology Lab
Bldg. 007, Rm. 104, BARC-West
Beltsville, MD 20705
tjackson@hydrolab.arsusda.gov
301-504-8511 (voice)
For information about or assistance in using SGP99 DAAC
data, contact
Hydrology Data Support Team
EOS Distributed Active Archive Center (DAAC)
Code 610.2
NASA Goddard Space Flight Center
Greenbelt, Maryland 20771
hydrology-disc@listserv.gsfc.nasa.gov
301-614-5165 (voice)
301-614-5268 (fax)
Last updated: February 28, 2008 12:36:11 GMT
Page Author: Hydrology Data Support Team -- hydrology-disc@listserv.gsfc.nasa.gov
Web Curator: -- Website Curator:
NASA official: Steve Kempler, GES DISC Manager -- Steven.J.Kempler@nasa.gov
|