The objective of this study is to evaluate the effectiveness
of using more easily automated image composites for the
modeling of
PTCC
. The two types of composites investigated
were based on the maximum
NDVI
Landsat-5 values and the
univariate median Landsat-5 values. These composite images
and their respective
PTCC
models were compared to the model
II regression mosaic and its
PTCC
model to determine if the
products of the more automated compositing methods could
adequately replace the model II regression mosaic.
Study Area
The
PTCC
prototype study (Coulston
et al.
, 2012) used five
MRLC
mapping zones (16, 23, 48, 54, and 59; Figure 1) to
develop the
PTCC
methodology. These same zones were used
for this study.
Zones 16 and 23 occur in the western United States and
comprise the western edge of the Rocky Mountains (zone 16)
and the Colorado Plateau (zone 23). Zone 16 is 5.8 million ha.
The average elevation is 2,353 m and the maximum elevation is
4,083 m. March is the wettest month, receiving 64 cm of precipi-
tation while January is the driest month, receiving 5 cm of pre-
cipitation (Daly
et al.
, 2008). Pinyon-juniper woodland, cover-
ing over 1 million ha, is the predominant forest type. Aspen and
deciduous oak woodlands, covering 1.5 million ha collectively,
are the predominant deciduous forest types. Other common
coniferous forest types, covering 1.2 million ha collectively, in-
clude lodgepole pine, Douglas-fir, Englemann spruce, white fir,
subalpine fir, and ponderosa pine (Ruefenacht
et al.,
2008).
Zone 23 is 10.1 million ha and consists mostly of high
elevation desert with scattered mountain ranges. The average
elevation is 1,878 m and the maximum elevation is 3,788 m.
October is the wettest month, receiving 38 cm of precipitation,
and January is the driest month, receiving 3 cm of precipita-
tion (Daly
et al.,
2008). The predominant forest type is pin-
yon-juniper, covering nearly 3 million ha. The predominant
deciduous forest types are deciduous oak woodland and as-
pen, covering 900,000 ha. Ponderosa pine, juniper woodland,
and Douglas-fir are the other common coniferous forest types,
covering 350,000 ha collectively (Ruefenacht
et al.,
2008).
Zones 48, 54, and 59 are 7.9, 7.1, and 7.7 million ha,
respectively. They occupy parts of Tennessee, Alabama,
Georgia, North and South Carolina, and Virginia. The average
elevation for these three zones is 222 m and the maximum
elevation is 1,035 m. The wettest month for these zones is
March, receiving 139 cm of precipitation while January is the
driest month, receiving 13 cm of precipitation (Daly
et al.,
2008). Loblolly pine, covering 7.5 million ha, is the predomi-
nant forest type. Various hardwoods including white oak, red
oak, northern red oak, hickory, yellow poplar, and sweetgum,
covering 6.6 million ha collectively, are the predominant
deciduous forest types (Ruefenacht
et al.,
2008).
Methods
For this investigation, three datasets were developed and
used to model
PTCC
. A description of how these three datasets
were developed is provided below.
Image Selection and Preprocessing
For each of the 51
WRS-2
paths/rows that intersected the five
zones, 15 L5 scenes were obtained using the
USGS
Global
Visualization Viewer (
GloVis
) (
glovis.usgs.gov
). Selecting 15
scenes per path/row ensured there were anomaly-free pixels
for the compositing process. Since the response dataset used
to model
PTCC
was photo-interpreted using aerial photography
acquired in 2009 and 2010, these were the primary years from
which the L5 scenes were selected, but 2011 was also includ-
ed. Additionally, a few L5 scenes were from 2007 and 2008.
An assumption of this study was that the photo-interpreted
PTCC
values were related to the spectral values of the satellite
images. Landscape changes might have occurred between the
time the satellite image was acquired and the time the aerial
photography was flown thereby weakening the relationships.
The same data was used for all models making all models
subjected to the same data impurities. The objective of this
study was to examine differences in model performances.
Since all models were subjected to the same data impurities,
the comparisons conducted in this study are still valid.
Path/row
NDVI
curves provided by
GloVis
were used to
determine the dates of L5 scenes to download; see Figure 2
for an example. Only
NDVI
curves for deciduous forest, mixed
forest, evergreen forest, and woody wetlands land-cover types
were examined. For the predominant land-cover type from
the above list for each path/row, the peak
NDVI
values were
obtained. A date range was determined by finding the dates
Figure 1. Contiguous United States with Multi-Resolution Land Characteristic mapping zones 16, 23, 48, 54, and 59 highlighted.
200
March 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING