points affected by different sources of inaccuracy to varying
degrees of overall importance, in the context of improving
the reduction of the root mean square error over the
AOI
.
Such complementarity is also exploited in the method
developed by Huang
et al.
(2013), where different indicators
are used to filter distinct sources of inaccuracy. Our gain and
slope indicators showed the best performance with respect
to the fraction of points exclusively rejected, reflecting the
uniqueness of their filtering capabilities. However, in the
latter case, very few points were rejected and the overall
statistics of root mean square error reduction were not
significantly improved across the
AOI
. Nonetheless, we believe
that the slope indicator showed the potential for useful point
reduction in less flat terrain.
Impact of Snow
When the elevations acquired in the presence of snow were
removed (meaning the exclusion of all points acquired from
October to April), only 20 percent of the original points re-
mained. This fraction was further reduced to 5 percent when
the filtering process was applied to the subset. As a conse-
quence, more than 50 percent of the orbit lines were removed
completely. This resulted in a very sparse and irregular
spatial distribution for the remaining reference points over
the
AOI
. The
RMSE95
associated with this result was 5.25 m, a
decrease of 0.56 m compared with the value obtained for the
entire data set (last row of Table 1). In spite of this improve-
ment in vertical accuracy, the effect on the spatial distribu-
tion is too extreme for this filtering strategy to be considered.
The laser had very little operational time within this period
(
NSIDC
, 20114f) providing, in turn, few orbit lines to exploit
data from. This explained why this strategy yielded to those
results.
Conclusions
In this paper we presented a method for determining the
vertical accuracy of a
DEM
retrieved using satellite-based lidar
data. Due to several limitations, the data sources tradition-
ally used (
ASBD
) are not suitable for this purpose throughout
Northern Canada. The Global Land Surface Altimetry Data
(
GLA14
) from the
ICES
at
platform was identified as a viable
alternative (after outlier removal of inaccurate altitude values)
but still contained inaccuracies. Several sources of signal con-
tamination were identified (laser attitude, saturation, equip-
ment noise, atmosphere, and variable elevation within the
footprint) and indicators were selected to remove potentially
erroneous elevation values from this alternative source. These
indicators consisted of elevation differences with respect to
the
CDED
to enable outlier filtering, attitude quality, the gain
and the saturation level of the signal, the apparent reflectivity
of the target, the signal/noise ratio, the number of returned
echoes, and the slope. After the outliers were removed from
the
GLA14
, the overall filtering resulting from the combina-
tion of all the indicators removed 69 percent of the remaining
points and provided a 19 percent gain (1.32 m) relative to
the
RMSE95
calculated from the elevation difference between
GLA14
and the
CDED
. Furthermore, the high density of
GLA14
data allowed an acceptable spatial distribution of the eleva-
tion points over the
AOI
to be preserved. Snow presence filter-
ing, based simply on the exclusion of all elevations between
October and April, seriously deteriorated the distribution and
was not pursued further, even though a slight increase in the
vertical accuracy was achieved. Except for the signal/noise
indicator, which was deemed to be irrelevant, and possibly
the saturation indicator, the other indicators were found to be
useful and complementary, each exclusively filtering specific
points. The best performance was nominally achieved by
the slope indicator (90 percent of points exclusively rejected),
yielding a large value of
RMSE95
for the rejected points (and
thus, a significant impact in terms of that small number of
points) but with only a small improvement in
RMSE95
reduc-
tion across the
AOI
(−0.02 m when that indicator filter was
applied individually). The most significant overall impact
was achieved by the gain indicator whose exclusive filtering
rate was 42 percent accompanied by an individual
RMSE95
reduction of ‑1.08 m across the
AOI
. It was closely followed
by the indicator based on gain, which individually enabled a
reduction of 0.96 m on the
RMSE95
with a 23 percent exclusive
filtering rate.
Finally, this filtering method is applicable anywhere where
GLA14
data are available and aims to deal with the main
sources of inaccuracy in
GLAS
data. For operational use, it can
be automated from information directly available in the
GLA14
product, except for the outliers that can be removed using
the elevation model for which the vertical accuracy must be
calculated. The use of the pulse width (Gonzales
et al.
, 2010)
could be looked into as a potential additional indicator and
weighted against the selected indicators in this paper. To
our knowledge, the potential high performance of the slope
indicator has not yet been identified in the literature. Since
the
AOI
in this study was characterized by flat topography, the
attitude and the slope indicators should be tested in a rugged
environment to better evaluate their potential. In addition, the
number of peaks indicator should be used with caution over
forested or urban areas, as most points in the dataset could be
rejected. Comparing filtering techniques over different study
sites with variable topography could be the subject of further
studies. Furthermore, using ground reference data could
further help in determining the true potential of these filtering
strategies.
Acknowledgments
We would like to thank the Canadian Center for Mapping and
Earth Observation (
CCMEO
) for making this research project
possible. Also, special thanks to the
NSIDC
for their support
and for the free access to
ICES
at
data. Finally, we would like
to thank our colleagues from
CCMEO
, from l’Université de
Sherbrooke and the anonymous reviewers for their valuable
comments and suggestions.
References
Beaulieu, A., and D. Clavet, 2009. Accuracy assessment of
Canadian digital elevation data using ICESat,
Photogrammetric
Engineering & Remote Sensing
, 75(1):81–86.
Bae, S., and B.E. Schutz, 2002.
Precision Attitude Determination
(PAD), Algorithm Theoretical Basis Document
, version 2.2, Center
for Space Research, The University of Texas at Austin, 94 p.
Brenner, A.C., C.R. Bentley B.M. Csatho, H.J. Harding, M.A. Hofton,
J. Minster, L. Roberts, J.L. Saba, R. Schutz, R.H. Thomas, D.
Yi, and H.J. Zwally, 2011.
Derivation of Range and Range
Distributions from Laser Pulse Waveform Analysis for Surface
Elevations, Roughness, Slope, and Vegetation Heights, Algorithm
Theoretical Basis Document
, version 5.0, Goddard Space Flight
Center, Greenbelt, Maryland, 125 p.
Brenner, A.C., J.P. Dimarzio, and H.J. Zwally, 2007. Precision and
accuracy of satellite radar and laser altimeter data over the
continental ice sheets,
IEEE Transactions on Geoscience and
Remote Sensing
, 45(2):321–331.
Carabajal, C.C., and D.J. Harding, 2006. SRTM C-band and ICESat
laser altimetry elevation comparisons as a function of tree cover
and relief,
Photogrammetric Engineering & Remote Sensing
,
72(3):287–298.
Duda, D.P., J.D. Spinhirne, and E.W. Eloranta, 2001. Atmospheric
multiples scattering effects on GLAS altimetry - Part I:
Calculations of single pulse biais,
IEEE Transactions on
Geoscience and Remote Sensing
, 39(1):92–101.
706
September 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
