regression mosaic are lower. The spectral values for the
model II regression mosaic came from dates in August
to October (Table 5). The imagery used for the model
II regression mosaic procedure is more constrained,
because of the necessity to use quality images that are
as cloud/shadow free as possible. The model II regres-
sion mosaic did experience a loss of information, but
the other composites did not.
2. Second, the
PTCC
models produced were similar in
terms of model fit. The random forest
OOB
percent
variance explained and the mean of squared residu-
als differed by a maximum of 5.3 percent and 87.6,
respectively, between the different
PTCC
models cre-
ated using the different composite images (Table 6).
Wilcoxon signed rank tests did find significant differ-
ences in three comparisons of
OOB
PTCC
predictions
(Table 7). Two of these significant differences occurred
when comparing the model II regression mosaic
OOB
PTCC
predictions with the median composite and the
maximum
NDVI
composite
OOB
PTCC
predictions. These
differences are likely attributed to the late growing sea-
son imagery chosen for the model II regression mosaic,
which caused the
OOB
PTCC
predictions to be on average
two percent lower than the
OOB
PTCC
predictions de-
rived using the other composites. The other significant
difference occurred when comparing the
OOB
PTCC
pre-
dictions for the median composite and the maximum
NDVI
composite for zone 54. The
OOB
PTCC
predictions
for the median composite was on average five percent
higher than the
OOB
PTCC
predictions for the maxi-
mum
NDVI
composite. Additionally, when comparing
the
PTCC
estimates created from the different models,
the majority of
PTCC
estimates (>89 percent, Table 8)
fell within +2 standard errors of the other
PTCC
model
estimates. The
PTCC
estimates created using the median
composite had the most estimates that fell within +2
standard errors of the other
PTCC
model estimates (>97
percent, Table 8) and the maximum
NDVI
had the fewest
(>89 percent, Table 8). While there were differences
between the
PTCC
models, these differences are not
substantial enough to choose one model over another.
3. Third, as is shown in Plates 1 and 2, the quality of the
median composite image is greater than that of the
maximum
NDVI
composite. The maximum
NDVI
com-
posite has Landsat anomalies, clouds, shadows, and
abnormal pixelation. Because the median value is being
selected for the median composite, these image defects
were excluded from the median composite. The quality
of the model II regression mosaic surpasses the quality
of the median composite. However, producing a model
II regression mosaic is a time-intensive manual process
than can take several days whereas the median compos-
ite procedure can be fully automated. Based on these ob-
servations and analysis, the median composite method
is an efficient approach to statistically reduce data on a
per-band, per-pixel basis from a much larger ensemble of
images to create images suitable for vegetation modeling.
Acknowledgments
I would like to thank Craig Baker, Robert Benton, Tanushree
Biswas, Mark Finco, Vicky Johnson, Kevin Megown, and Mi-
chael Walterman for their assistance in this project. I would
like to also thank two anonymous reviewers for their assis-
tance in improving the manuscript.
References
Adelabu, S., O. Mutanga, and E. Adam, 2015. Testing the reliability
and stability of the internal accuracy assessment of random
forest for classifying tree defoliation levels using different
validation methods,
Geocarto International,
30(7):810−821.
Bailey, R.G., 1983. Delineation of ecosystem regions,
Environmental
Management
, 7(4):365−373.
Beaty, M., M. Finco, and K. Brewer, 2011.
Using Model II Regression
to Radiometrically Normalize Landsat Scenes for the Purpose of
Mosaicking
, USDA Forest Service Remote Sensing Applications
Center, RSAC-10012-RPT1, Salt Lake City, Utah, 6 p.
Breiman, L., 2001. Random forests,
Machine Learning
, 45(1):5−32.
Breiman, L., 1996. Out-of-bag estimation, URL:
.
berkeley.edu/~breiman/OOBestimation.pdf
(last date accessed:
11 January 2016).
Chander, G., B.L. Markham, and D.L. Helder, 2009. Summary
of current radiometric calibration coefficients for Landsat
MSS, TM, ETM+, and EO-1 ALI sensors,
Remote Sensing of
Environment
, 113(2009):893−903.
Chavez, P.S., 1988. An improved dark-object subtraction technique
for atmospheric scattering correction of multispectral data,
Remote Sensing of Environment
, 24(1988):459-479.
Conese, C., and F. Maselli, 1991. Use of multitemporal information
to improve classification performance of TM scenes in complex
terrain,
ISPRS Journal of Photogrammetry and Remote Sensing
,
46(4):187−197.
Coulston, J.W., D.M. Jacobs, C.R. King, and I.C. Elmore, 2013. The
influence of multi-season imagery on models of canopy cover:
a case study,
Photogrammetric Engineering & Remote Sensing
,
79(5):469−477.
Coulston, J.W., G.G. Moisen, B.T. Wilson, M.V. Finco, W.B. Cohen,
and C.K. Brewer, 2012.
Modeling percent tree canopy cover: a pilot study,
Photogrammetric
Engineering & Remote Sensing
, 78(7):715−727.
Crist, E.P., and R.C. Cicone, 1984. A physically-based transformation
of Thematic Mapper data - The TM tasseled cap,
IEEE Trans-
actions on Geoscience and Remote Sensing
, GE-22(3):256−263.
Daly, C., M. Halbleib, J.I. Smith, W.P. Gibson, M.K. Doggett, G.H.
Taylor, J. Curtis, and P.A. Pasteris, 2008. Physiographically-
sensitive mapping of temperature and precipitation across
the conterminous United States.
International Journal of
Climatology
, 28(15):2031−2064.
Flood, N., 2013. Seasonal composite Landsat TM/ETM+ images using
the medoid (a multi-dimensional median),
Remote Sensing
,
2013(5):6481−6500.
Greenfield, E.J., D.J. Nowak, J.T. Walton, 2009. Assessment of 2001
NLCD percent tree and impervious cover estimates,
Photo-
grammetric and Engineering & Remote Sensing,
75(11):
1279−1286.
Griffiths, P., S. van der Linden, T. Kuemmerle, and P. Hostert, 2013.
A pixel-based Landsat compositing algorithm for large area land
cover mapping,
IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing
, 6(5):2088−2101.
Hansen, M.C., D.P. Roy, E. Lindquist, B. Adusei, C.O. Justice,
and A. Altstatt, 2008. A method for integrating MODIS and
Landsat data for systematic monitoring of forest cover and
change in the Congo Basin,
Remote Sensing of Environment
,
112(2008):2495−2513.
Homer, C., J. Dewitz, J. Fry, M. Coan, N. Hossain, C. Larson, N.
Herold, A. McKerrow, J.N. VanDriel, and J. Wickham, 2007.
Completion of the 2001 National Land Cover Database for the
conterminous United States,
Photogrammetric Engineering &
Remote Sensing
, 73(4):337− 341.
Homer, C., C. Huang, L. Yang, B. Wylie, and M. Coan, 2004.
Development of a 2001 National Land-Cover Database for the
United States,
Photogrammetric Engineering & Remote Sensing
,
70(7):829–840.
Homer, C.G., and A. Gallant, 2001. Partitioning the conterminous
United States into mapping zones for Landsat TM land cover
mapping, US Geological Survey, URL:
/
pdf/homer.pdf
(last date accessed: 11 January 2016).
210
March 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
