522
July 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
History of the Global Land Survey
The Global Land Survey (GLS) dataset is the product of a
partnership between the US Geological Survey (USGS) and
National Aeronautics and Space Administration (NASA). It
was designed to inform the U.S. Climate Change Science
and the NASA Land-cover and Land-use Change (LCLUC)
Programs. Building on the existing GeoCover dataset
representing nominal 1970’s, circa-1990, and circa-2000
“epochs” (Tucker et al., 2004), the GLS was selected to
provide wall-to-wall, orthorectified, cloud-free Landsat
coverage of Earth’s land area at 30-meter resolution in epochs
representing 1975, 1990, 2000, 2005, and 2010 (Franks et al.,
2009; Gutman et al., 2008). The GLS collection of Landsat
images is intended to provide one cloud- and error-free
image representing the peak growing season of each epoch
for each World Reference System (WRS) tile. This selection
significantly reduces end-user effort in accessing Landsat
data appropriate for land-cover classification and change
detection.
The GLS and its predecessor were the first open-access
satellite datasets to depict Earth’s global terrestrial surface
at high (30-m) resolution over the long term (35+ years). The
first three GeoCover epochs were geometrically corrected
at USGS and orthorectified by Earthsat Inc. (now MDA
Federal) under contract to the US government. Selection
of images was manual, by visualizing potential images for
each scene and checking for clouds, missing bands or scan
lines, and calibration metadata. Later collections used a
more automated approach, known as the Large Area Scene
Selection Interface (LASSI) (Franks et al., 2009). By querying
image metadata, images were selected based on low cloud
cover, proximity to the season of maximum photosynthesis
as measured by NDVI (Normalized Difference Vegetation
Index), and absence of sensor errors. In some areas, scenes
were selected manually where no cloud-free image was
available.
Since the 2008 opening of the USGS archive, terrain-
corrected Landsat images can now be downloaded from the
USGS EROS Data Center (EDC). However, the process
of selection and download using this approach is time-
consuming for acquiring large datasets. Though automated
tools, such as the Bulk Download Tool
, have been developed to reduce the number of
clicks required to download data, acquiring large volumes of
data still remains a tedious process. Opening of the archive
has also allowed for the improvement of characterization
Earth’s land surface when there is lack of data due to Scan
Line Corrector (SLC) -Off gaps, cloud and cloud shadow, via
maximum value compositing techniques (Holben, 1986).
The USGS Web-Enabled Landsat Data (WELD) project
is creating global composites of 3-year epochs, as well as
weekly, monthly, seasonal, and annual composites of the
conterminous United States and Alaska (Roy et al., 2010).
With multi-date composites an option, users continue to
rely on the GLS products to represent the surface at one point
in time. As of November 2014, more than six (6) terabytes
of GLS imagery have been downloaded from EDC, and over
177 terabytes have been downloaded from the Global Land
Cover Facility (GLCF)
). At the GLCF,
the GLS has served as the basis for mapping the first global
tree, water, and forest cover (Sexton et al., 2013, 2015; Feng
et al, in press) and for monitoring global forest cover change
(Townshend et al., 2012; D.-H. Kim et al., 2014; Sexton et al.,
2015). These applications have revealed issues in the original
GLS dataset. In this paper we characterize these issues
and describe their correction with newly available data. We
term this enhancement of the Global Land Survey or simply
“GLS+”.
Challenges and Enhancements
The principal challenges to analyses based on the original
GLS collection were images with: (1) excessive cloud cover, (2)
“off-season” phenology, (3) sensor artefacts, and (4) incorrect
or absent calibration metadata.
Cloud Cover
.
Landsat 7 uses the Automated Cloud-Cover
Assessment (ACCA) algorithm to map and estimate the
percentage of clouds in each image (Irish et al., 2006).
However, ACCA underestimates cloud cover (Gutman
et al., 2013), and so metadata queries will likely include
excessively cloudy images (Figure 1). Replacing ACCA with
improved algorithms is expected to improve cloud-cover
estimates (Lindquist et al., 2008; Huang et al. , 2010) and
thus automated metadata searches as well. For the GLS+,
we estimated coverage of clouds detected by the algorithm of
Huang et al. (2010), queried the USGS archive for suitable
replacements, and upon visual examination, selected suitable
images from the candidates.
The GLS collection of Landsat images is intended
to provide one cloud- and error-free image
representing the peak growing season of each
epoch for each World Reference System (WRS) tile.
“Global, long-term monitoring of changes in Earth’s
land surface requires quantitative comparisons
of satellite images acquired under widely varying
atmospheric conditions...”
