GLS Scene Selection: North America

Scenes for the GLS are being selected through an automated tool (LASSI – Landsat Automated Scene Selection Interface) that uses non-linear optimization to identify the “best” selection of scenes from the archive given a set of weighting conditions. These weights balance factors such as image cloud cover, image greenness (NDVI), choice of sensor, and the ability of a Landsat-7 SLC-off pair to completely fill image gaps. LASSI was developed by NASA Ames Research Center with support from NASA GSFC and USGS EROS. For more information on the algorithmic basis for the tool, see L. Khatib, J. Gasch, R. Morris, S. Covington, "Local Search for Optimal Global Map Generation Using Mid-Decadal Landsat Images", In Proceedings of AAAI Workshop on Preference Handling for Artificial Intelligence, Vancouver, BC, 2007.

We invite you to examine the characteristics of a “Base Case” for the North American portion of the GLS, as well as particular variations given below. Please comment on the results presented below, and how you would like to see the GLS scene selection optimized for North America. Results for additional regions will be posted in the future.

1. Base Case (view the weighting factors used)

NDVI (Percent of Maximum)
White: ~1.0, Green: ~0.8, Yellow: ~0.5, Orange: ~0.3, Red: ~0.1, Black: ~0.0



Click on one of the options below to see the particular output parameter mapped in the window above:

Sensor selected (Landsat-5 vs. Landsat-7)
Percent image cloud cover from ACCA (Automated Cloud Cover Algorithm)
Day of Year of the Image Acquisition
Spatial coverage of SLC-off gaps using L7 image pairs (=100% for L5 images)
Time of Year coincidence compared to Geocover 2000
Acquisition Year proximity to 2005
NDVI of image (or NDVI of base image for Landsat-7)
NDVI difference between L7 base and fill images (L5 images set to 0)
Time of Year difference between L7 base and fill images (L5 images set to <1 cycle)

2. Variations on Base Case. Below we provide several examples of how varying an individual weighting factor changes the data set characteristics. Clicking on one of the links below will open a Powerpoint file with maps of the various data set characteristics for that set of weightings. (Be sure to scroll down in the Powerpoint file to view all maps.) Compare these with the maps shown above for the Base Case. Note that varying any factor is a trade-off - improving one aspect of the data set will tend to degrade some other aspect.

1. No preference for Landsat-7 versus Landsat-5
2. Preference for less time difference between L7 base and fill scenes
3. No preference for matching seasonality of Geocover 2000
4. Decreased priority on choosing peak-NDVI imagery
5. Increased priority on choosing peak-NDVI imagery