A Landsat Data Tiling and Compositing
Approach Optimized for Change Detection in the
Conterminous United States
Kurtis J. Nelson and Daniel Steinwand
Abstract
Annual disturbance maps are produced by the
LANDFIRE
program across the conterminous United States (
CONUS
).
Existing
LANDFIRE
disturbance data from 1999 to 2010 are
available and current efforts will produce disturbance data
through 2012. A tiling and compositing approach was de-
veloped to produce bi-annual images optimized for change
detection. A tiled grid of 10,000 × 10,000 30 m pixels was
defined for
CONUS
and adjusted to consolidate smaller tiles
along national borders, resulting in 98 non-overlapping
tiles. Data from Landsat-5,-7, and -8 were re-projected to
the tile extents, masked to remove clouds, shadows, water,
and snow/ice, then composited using a cosine similar-
ity approach. The resultant images were used in a change
detection algorithm to determine areas of vegetation
change. This approach enabled more efficient processing
compared to using single Landsat scenes, by taking advan-
tage of overlap between adjacent paths, and allowed an
automated system to be developed for the entire process.
Introduction
Satellite image-based change detection is used for a wide
variety of applications, in part because of the ability to map
change over large areas, long time periods, and at low cost
relative to ground or airborne sampling. Global archives of
imagery provide historical baselines, while an advent of
new sensors (e.g., Landsat-8 Operational Land Imager (
OLI
),
Sentinel) and relaxed data access policies (Woodcock
et al.
,
2008; Aschbacher and Milagro-Pérez, 2012) combine to offer
vast quantities of image data for analysis. As computing and
digital storage systems advance and associated costs dimin-
ish, change detection algorithms are able to take advantage of
more data and provide more detailed descriptions of how the
land surface of the Earth is constantly changing. The resultant
change products can be used by research and monitoring pro-
grams for forest tracking (Goward
et al.
, 2008), carbon account-
ing (Zhu
et al.,
2010), land cover change (Jin
et al.
, 2013), and
wildfire applications (Vogelmann
et al.
, 2011), among others.
One program that relies on image-based change detection
for monitoring and updating its products is the Landscape
Fire and Resource Management Planning Tools (
LANDFIRE
)
program (
.
LANDFIRE
.gov
).
LANDFIRE
is an interagen-
cy collaboration that provides consistent and comprehensive
vegetation and wildland fuels data for the entire United States
(Rollins, 2009; Ryan and Opperman, 2013).
LANDFIRE
products
are used for strategic planning and tactical decision-making
on wildfire incidents, resource management plans, fuel treat-
ment projects, and many other non-fire applications.
LANDFIRE
products are periodically updated to account for landscape
change due to natural and anthropogenic disturbances. One
component of
LANDFIRE
updating is the Remote Sensing of
Landscape Change (
RSLC
) effort that utilizes Landsat data to
provide analysts with bi-annual change detection products
for the conterminous United States (
CONUS
), which are then
manually processed to determine location, time, and types of
disturbance occurring on the landscape (Nelson
et al.
, 2013a),
resulting in annual disturbance products depicting these attri-
butes spatially on the landscape. These products are currently
available for the years 1999 to 2010.
The
LANDFIRE
2012 (
LF
2012) effort is under way at the time
of this writing and is expected to be available in early 2015.
The goal of
LF
2012 is to update available datasets to
circa
2012 conditions. Unique data constraints required modifica-
tions to the
RSLC
approach for
LF
2012 compared to previous
efforts. Bi-annual image data are required for each year that
disturbance products are being created and one year before
and after. Since
LF
2012 is mapping disturbances in 2011
and 2012, imagery was needed for the years 2010 to 2013. In
previous
LANDFIRE
updates, imagery was processed as single
scenes on a Worldwide Reference System (
WRS
) path/row
basis (Nelson
et al.
, 2013a; Nelson
et al.
, 2013b), requiring
445 individual image stacks to cover
CONUS
. However, the
only Landsat sensor that collected imagery for this entire time
period was the Landsat-7 Enhanced Thematic Mapper Plus
(E
TM
+). Due to the Scan Line Corrector (
SLC
) malfunction in
2003 (NASA, 2005), the exclusive use of single-scene E
TM
+
data was undesirable because it would require significant
manual effort to produce disturbance products, would result
in substantial amounts of missing data with no disturbance
information, and would introduce scan line artifacts into
the derived products. With Landsat-5 Thematic Mapper (
TM
)
available for 2010 to 2011, and
OLI
data becoming available in
2013, the decision was made to use all available data to miti-
gate the
SLC
-Off issue. Only in 2012, when no other Landsat
data were available, were
ETM
+ data used exclusively. In addi-
tion to the
SLC
-Off scan line gaps, missing data due to clouds,
shadows, and other undesirable features proved problematic
in previous
LANDFIRE
updates. Therefore, a tiling and compos-
iting approach was developed for
LF
2012 where multi-tempo-
ral Landsat data were combined on a pixel basis to produce
nominally complete and seam-free image tiles for change
detection processing. Existing data from the Web-Enabled
Landsat Data (
WELD
) project (Roy
et al.
, 2010) were evaluated
but not utilized for several reasons including unavailability
of data past 2012, exclusive use of Landsat-7 for all years, and
the use of maximum greenness compositing.
United States Geological Survey (USGS) Earth Resources
Observation and Science (EROS) Center, Sioux Falls, SD
57198 (
).
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 7, July 2015, pp. 573–586.
0099-1112/15/573–586
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.7.573
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2015
573
