Landsat Missions
Segmentation is defined as the process of grouping individual pixels into homogeneous landscape units. The smallest landscape units, or segments, are derived first. The larger landscape units are derived by grouping the smaller segments into increasingly larger segments.
This segment-based approach utilizes a multi-scale segment model derived from Landsat ETM+ SLC-on imagery to guide the interpolation of spectral data across the gaps in SLC-off imagery. Application of this model allows the gap pixels to be filled with concurrent spectral data which is essential for applications that require same-day spectral information such as crop mapping, crop monitoring, and disaster assessment.
A segment model is first derived for each Landsat ETM+ (SLC-on) circa 2000 image in the GeoCover data set.
The model is generated by applying a segmentation algorithm to the Landsat image at three scales (10, 15, and 20), as demonstrated in the images in Figure 1, for a region in northeastern Nebraska.
SLC-off gaps are filled using a combination of the segment model and a specially designed nearest neighbor algorithm, as demonstrated in Figure 2. The segment map for scale 10 is first superimposed on the SLC-off image and the gap pixels intersecting each individual segment boundary are interpolated. Edge pixels (28.5m for reflective bands, 57m for thermal bands) and anomalous pixels are excluded from the gap-fill process. Normally, about 80% of the gap pixels are filled using the scale 10 segments.
The segment map for scale 15 is then used to fill additional gap pixels. Scale 20 segments are then applied. Previously interpolated gap pixels are not used in the gap-fill algorithm at these two scales. A final ‘clean-up’ pass is applied to fill any remaining gap pixels that were not filled using scales 10, 15, or 20. A thematic map identifying the scale at which each pixel was filled is provided with each order.
Applications that have been evaluated using Segment-Based data include general land cover and crop specific mapping, evapotranspiration modeling, tree plantation monitoring, burn severity mapping, and disaster response support.
Note that the segment-based gap-filled product is based on a patch-level approach and is therefore more appropriate for landscape level applications, such as land cover mapping and visual assessment.
Applications requiring a high degree of accuracy at the pixel level, such as impervious surface mapping, small or narrow feature identification (e.g., roads), or applications using texture features should be cautious of using the gap-filled areas.
The final product will include the gap-filled SLC-off image and a gap mask. The original digital elevation models (DEMs), segmentation maps, and source imagery used to generate the product will not be distributed with the product.
To download samples of the SLC-off products, see the Sample Products page.
All SLC-off data products may be searched and ordered through EarthExplorer or the USGS Global Visualization Viewer (GloVis)
The distribution of Landsat 7 ETM+ Segment-based Terrain correction data is restricted to U. S. Government and its affiliated users. Please contact USGS EROS Customer Services if you have questions regarding this policy.
U.S. Department of the Interior |
U.S. Geological Survey
URL: http://landsat.usgs.gov
Sitemap
Page Contact Information: landsat@usgs.gov
Page Last Modified: May 12, 2008
