urban areas are located in plains, whereas glaciers are located
in steep mountains and deserts have rolling surfaces. These
differences in topographic relief among the four landscapes
meant that changes in the standard deviation of the elevation
error were also primarily affected by the slope.
Conclusions
The distribution of
SRTM
error and its associations with vari-
ous influencing factors within China has been analyzed in
this study. The main conclusions are summarized as follows:
1. The spatial and statistical distributions of the main
topographic attributes derived from
SRTM
data accu-
rately represent the actual topographic characteristics
of China. The elevation distribution reflects the three
basic terrain terraces in China. The peak in the fre-
quency curve of slope was located near to 0° and an
inflection point occurred near 5°, which separated the
landscape into flat plains and rolling mountains.
SRTM
data was, however, found to have a significant problem
in the estimation of aspect in flat areas. On the other
hand, the aspect frequency distribution curves of the
areas with slopes greater than 5° were consistent with
independently derived and accurate
DEM
data and
indicated correctly that the mountains in China mostly
trend from east to west.
2. Most of the elevation errors were close to 0, and the
90 percent error for the entire sample was only 7.4
m, which met the original 16 m specification for the
SRTM
mission within the sub-samples of the different
factors. Only the steep slope samples and the glacier
samples failed to fully meet this specification. While
the magnitudes of both the maximum positive and
negative errors of the entire sample exceeded 1,000 m,
there were only a small number of sample points with
an absolute error greater than 50 m. In areas with large
errors,
SRTM
data had some obvious issues with “eleva-
tion anomalies” and “vague topography” where it does
not properly represent the structure of the land surface.
Although these areas accounted for only a small pro-
portion of the land in China, some special attention
should be paid to them.
3. The elevation error of the
SRTM
data was closely related
to the topography and land-cover. Different factors had
distinct impacts on the
SRTM
errors and caused distinct
error variation patterns. Slope had the strongest impact
on the errors, and it also affected the variation patterns
of error in the other factors. The spatial distribution of
the errors indicated that flat areas generally exhibited
small errors whereas mountains generally displayed
larger errors. The land cover type with greatest concen-
tration of positive errors was the urban built-up type,
with a mean error of 1.05 m, whereas the negative er-
rors were concentrated in deserts and wetlands (freez-
ing winter) with mean errors below −2 m. Because the
glacier-covered areas were dominated by mountains, no
concentration of positive or negative errors occurred,
but the mean error of the entire glacier area was −1.05
m. The statistical results for the various factors indi-
cated that the elevation errors changed from positive
to negative between plains and high mountains and
that the magnitude of the errors gradually increased
as the slope increased, with mean errors in the range
from 0.05 m to −0.86 m. For the different aspects, the
positive errors were concentrated in northern direc-
tions, while the negative errors occurred more often in
southwestern directions. In flat areas (slopes within the
range of 0° to 2°) with vegetation, the elevation errors
increased with increasing vegetation coverage and had
a mean error that varied within the range of 0.15 m to
1.67 m. In terms of standard deviation, the variation of
elevation errors increased with increasing slope and
vegetation coverage, but did not diverge greatly among
different aspects.
Contrary to other studies that concluded the
SRTM
error
usually showed an overall positive bias, mainly caused by
vegetation and artifacts on the bare ground, the overall mean
error of the
SRTM
data in China is negative. One reason for
this could be the presence of widespread deserts, freezing
wetlands, and glaciers where the radar signal can penetrate
to depths below the surface, and also to the abundant rugged
mountains with high slopes where the errors trend to negative.
For the users of
SRTM
data in China, the information
summarized above can lead to a better understanding of the
characteristics of elevation errors and how they may affect
their research. Although
SRTM
data have relatively good
quality taken over the full extent of China, users should take
notice of the errors and issues that exist locally in the data
when employing it in their applications. For example, the
stripes in flat areas; the obvious abnormalities in mountains,
glacier-covered areas and deserts; overestimation of elevation
in built-up areas and areas covered by vegetation; underes-
timation of elevation in desert, wetland and glacier areas;
and large negative errors in steeply sloping areas and their
predominance in southwest facing aspect areas. To use the
SRTM
data appropriately, users should first avoid employing
the data located around the obvious abnormalities and (if
possible) refer to an alternative (e.g., topographic map-derived
DEM
) source of data. Second, users are recommended to adopt
already widely-used methods of regression algorithms to
correct the remaining errors in
SRTM
data. There are already
methods for addressing the problems of striping, built-up
areas, and vegetation covered areas (Gallant and Read, 2009;
Baugh
et al.
, 2013), so the
SRTM
data should be updated to
meet the accuracy requirements of the applications.
It is reasonable to measure the quality of
SRTM
data by el-
evation error, but the relationship between
SRTM
morphology
and the actual surface morphology is another important indi-
cator of the quality of the data and has a significant impact on
numerous topographic analyses. There are already some pub-
lications which have addressed these issues, including river
networks (Da Paz
et al.
, 2007; Li and Wong, 2009; Hancock
et al.
, 2006; Liu 2008), and how the slope shape affects the
prediction of soil erosion (Verstraeten, 2006), but most have
only been studied at the local scale. In one important case,
however, the Hydro SHEDS (Lehner
et al.
, 2008) Project used
SRTM
data to provide hydrographic information for global-
scale applications, although there was no systematic evalua-
tion of the final data quality. Thus, there is still no complete
assessment of effect of
SRTM
quality on surface morphology at
the large scale, which, of course, includes the extent of China
and therefore needs further studies in the future.
Acknowledgments
This study was funded by the National Natural Science Foun-
dation of China (41371274, 41301284), the Ministry of Water
Resources industry-specific appropriation (201201081) and
the Natural Science Basic Research Plan in Shaanxi Province
of China (Program No.2013JQ5002). We thank the Institute
of Soil and Water Conservation of the Chinese Academy of
Sciences and Ministry of Water Resources (CAS /
MWR
ISWC
),
the Spatial Information Department of the Consultative Group
of International Agricultural Research, the Global Land Use
Database of the University of Maryland, and the Cold and
146
February 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
