ASTER Polar Surface and Cloud Classification
Version 2.8, [May 2002]
This provisional release provides a pixel level polar surface and cloud
classification based upon a supervised neural network classifier. The
possible classifications of surface and cloud types are Water, Snow / Ice,
Ice cloud, Land, Thin water cloud, Water cloud over water, Sea ice, Water
cloud over land and Bare ground / tundra.
Table of Contents
1. Where to get detailed information on this product
2. How to report problems
3. Description of this version
* Changes
* Product Quality
* Problems and limitations
* Additional information
4. Version history
1. Where to get detailed information on this product
There are several sources of information that users will find useful:
* ASTER Polar Cloud Mask Algorithm Theoretical Basis Document. Describes the algorithm used to create the product. Available at
http://eospso.gsfc.nasa.gov/eos_homepage/for_scientists/atbd/viewInstrument.php?instrument=24
* US ASTER Website. Describes the ASTER instrument, how to obtain data, the various data products that are available, the activities of the US
ASTER Science Team, etc.
http://asterweb.jpl.nasa.gov/
* Level 1 product information. Because the Polar Surface and Cloud Classification product is derived from a Level 1 product, understanding
the L1 products is useful.
Level 1 ATBD available at
http://eospso.gsfc.nasa.gov/eos_homepage/for_scientists/atbd/viewInstrument.php?instrument=24
Level 1 User Guide available at
http://www.science.aster.ersdac.or.jp/en/index.html
2. How to report problems
To report a problem with the product, please send mail to edc@eos.nasa.gov or call EDC DAAC User Services at (605) 594-6116.
3. Description of this version
This provisional release provides a pixel level polar surface and cloud classification based upon a supervised neural network classifier. The
possible classifications of surface and cloud types are as follows:
Code Class
1 Water
2 Snow / Ice
3 Ice cloud
4 Land
5 Thin water cloud
6 Water cloud over water
8 Sea ice
12 Water cloud over land
30 Shadow
31 Forest
Changes:
Version 2.6 was released in October 2001 as a Beta Release
Version 2.8 Provisional Release, updated metadata to include code explanations
and incorporated improved classifier.
Product quality:
The neural network classifier was trained to 93% overall accuracy. The product has been visually evaluated over a variety of polar scenes with
good overall performance. A more objective independent validation is underway.
Problems and limitations:
Currently, only a daytime algorithm is available. The classifier sometimes misclassifies pixels as cloud at high solar zenith angles
(> 80). Some thin clouds over ice are not detected. Currently, Aerosols are poorly represented in the training set and may be
misclassified as land, ice, or cloud.
4. Version History
Version 2.6, October 2001
Version 2.8, May 2002
|
|
|
|