=Paper=
{{Paper
|id=Vol-2030/HAICTA_2017_paper79
|storemode=property
|title=About Some Peculiarities of SRTM Digital Elevation Model Usage for Agricultural Land Use Planning
|pdfUrl=https://ceur-ws.org/Vol-2030/HAICTA_2017_paper79.pdf
|volume=Vol-2030
|authors=Arseniy Zhogolev,Igor Savin
|dblpUrl=https://dblp.org/rec/conf/haicta/ZhogolevS17
}}
==About Some Peculiarities of SRTM Digital Elevation Model Usage for Agricultural Land Use Planning==
https://ceur-ws.org/Vol-2030/HAICTA_2017_paper79.pdf
About some peculiarities of SRTM Digital Elevation
Model usage for agricultural land use planning
Zhogolev A.V. 1, Savin I. Yu.2
1
V.V. Dokuchaev Soil Science Institute, Moscow, Russia, e-mail: zhogolev_av@esoil.ru
2
V.V. Dokuchaev Soil Science Institute, Moscow, Russia, e-mail: savigory@gmail.com
Abstract. Digital elevation models (DEM) are widely used in agricultural
land use planning as a source of information about slopes, aspects, slope
forms and watersheds. Among different DEM products, one of the most
convenient is freely available SRTM. As well as other DEMs based on remote
sensing data, SRTM, actually, represents first reflective surface of the radar
signal such as top of the forest trees and bare-earth only if it is not obscured.
Using of such digital terrain model (DTM) in forested areas can lead to
artifacts in calculation of slopes, aspects, slope forms and watersheds. In the
present study, we provide the results of the quality assessment of SRTM and
different maps calculated from SRTM. We, also, proposed an easy-to-use
approach for adjustment of forest influence on SRTM and tested the approach
on key sites in Moscow, Russia.
Keywords: DEM, Geo-Information Systems, SRTM quality.
1 Introduction
Digital elevation models (DEM) are widely used in agricultural land use planning as
a source of information about slopes, aspects, slope forms and watersheds (Shukla,
2011; Zhogolev & Savin, 2016a). As a rule, remote sensing products representing the
first reflective surface of radar or laser signal are used. Such products represent the
tops of buildings, trees, other objects and the bare-earth if it is not obscured (Hirt,
2016). The influence of forest vegetation and other objects obstructing the bare-earth
can be adjusted.
Of laser illuminated detection and ranging (LIDAR), radar and stereo pair data
(United States National LIDAR Data-set, SRTM, ALOS PALSAR, ASTER GDEM,
SPOT DEM, etc.) one of the most widely used is SRTM product (Farr et al., 2007;
Nelson et al., 2009; Mulder et al., 2011; Du et al., 2015). The SRTM sensor was
launched on February 11, 2000 (Nelson et al., 2009). The mission lasted 11 days.
Since then a few updated versions of SRTM came out, the latest version – 4.1. The
spatial resolution of the SRTM is 1 arc second for the United States and 3 arc
seconds for global product. In 2015, SRTM with spatial resolution 1 arc second
became available globally. However, for many studies, at global and regional scales,
658
there is still a lot of interest in 3 arc seconds SRTM because of suitable
generalization and long practice of application. The three arc second product has
improvements that is not available for one arc second data.
The evolution of the SRTM has led to the improvement of the spatial reference,
filling of gaps and other improvements (CGIAR CSI, 2016). So far, there have been
many studies on the impact of noise, anchor errors, gaps in the data, the effect of
sensing at an angle. At the same time, there have been few studies on the problem of
the influence of vegetation, which use widely available spatial data (Hofton et al.,
2006; Shortridge & Messina, 2011; Gallant et al., 2012; Zhogolev & Savin, 2016b).
To assess accuracy of SRTM different data are used: GPS point data of field surveys;
topographic maps and accurate radar or LIDAR (Rodriguez et al., 2006; Ozah &
Kufoniyi, 2008; Karwel & Ewiak, 2012; Amans et al., 2013). In terms of practical
application, it is reasonable to estimate influence of vegetation on quality of SRTM
by a comparison with DSM based on traditional paper topographic maps of
comparable generalization scale. The generalization of topographic maps has been
perfected, taking into account long-term experience in application, so the comparison
can reveal the most significant errors of SRTM. Moreover, DEM based on
topographic maps can be replaced by SRTM for global scale application as better-
harmonized data or for local scale application in places where vectorized or up-to-
date topographic maps are not available (Mulder et al., 2011).
Adjustment of the forest vegetation influence on SRTM can be made using
approaches: replacing SRTM in forested areas with other DEM (Hengl et al. 2009),
reducing altitudes of SRTM in forested areas and smoothing the result with filters
(Hengl et al. 2009), replacing SRTM in forested areas with interpolated values
(Gallant & Read 2009; Gallant et al. 2012; Amans et al. 2013), adjustment of the
SRTM altitudes with the help of regressions model fitted using another DEM (Su &
Guo, 2014).
In this research, we provided the results of a comparison between slope, aspect,
slope form (convex-concave) and watershed maps calculated from original SRTM,
SRTM corrected using the proposed method and cartographic DEM created by
interpolation of isohypses of traditional topographic maps at 1:100 000 scale.
2 Study Area
As key sites, three lowland areas in the Moscow region (Russia) situated near
settlements Chashnikovo, Schebanovo and Serebryaniye prudy were selected (Tab.
1). The sites have different relief conditions and the share of forested area (Tab. 2).
The site "Schebanovo" has the greatest share of forests and the relief is flat. The
share of forested area in the site "Serebryaniye prudy" is the smallest, erosional
highly undulated relief is more pronounced than in other sites. The "Chashnikovo"
site has an average share of forested area and an average pronouncement of the relief.
A part of the Klyazma river floodplain is inside the boundaries of this site. The
dominant tree species on the site "Schebanovo" are pine (Pinus sylvestris L.), spruce
(Picea abies L.) and birch (Betula pendula Roth), with an average height of 24 m; on
the site "Chashnikovo": spruce (Picea abies L.) and birch (Betula pendula Roth),
659
average height 21 m; on the site "Serebryaniye prudy": linden (Tilia cordata Mill.)
and oak (Quércus róbur L.), average height 22 m.
Table 1. Coordinates of the key sites.
The name of key site Coordinates Lat/Long WGS84
Chashnikovo 1: 56005’15.04”N 37008’09.71”E
2: 56000’55.92”N 37008’18.50”E
3: 56005’23.08”N 37017’44.19”E
4: 56001’04.61”N 37017’55.85”E
Schebanovo 1: 55040’00.09”N 38029’56.73”E
2: 55036’57.64”N 38029’55.64”E
3: 55039’57.24”N 38046’34.15”E
4: 55036’57.70”N 38046’36.87”E
Serebryaniye prudy 1: 54035’29.90”N 38037’38.04”E
2: 54030’06.60”N 38037’43.59”E
3: 54035’29.74”N 38046’56.10”E
4: 54030’07.95”N 38046’59.12”E
Table 2. Description of the key sites.
The name of key site Minimum Maximum Percentage of Total site area,
altitude, m altitude, m forested area km2
Chashnikovo 180 240 55 80
Schebanovo 130 160 72 99
Serebryaniye prudy 130 210 24 100
3 Methods
For preparation and interpretation of satellite images Integrated Land and Water
Information System (ILWIS version 3.3.1) was used. Statistical analysis was carried
out in Microsoft Excel and R (https://www.r-project.org/).
Analysis of influence of vegetation on the quality of altitudes of the SRTM was
performed by comparing with the reference DEM based on traditional paper
topographic maps at 1:100 000 scale (further, “cartographic DEM”). The scale
1:100 000 was chosen because the mean error in the planned position of contours and
various objects lies in the range from 0.5 to 1 mm, i.e. from 50 to 100 m on the
ground (GKINP-05-029-84, 1984) that is close to the spatial resolution of the SRTM
which is 3 arc seconds or 90 m (Hengl, 2006). The analysis described in this article
was performed for three arc seconds void free SRTM v4.1 but it also can be applied
with little changes to one arc second SRTM as the main limitation of SRTM is its
vertical error, not spatial resolution.
Before the analysis, preparation of data was carried out. SRTM v4.1 data were
reprojected into UTM projection zone 37N on an ellipsoid WGS 84 with resampling
from 3 arc second (about 90 m resolution) to 30 m resolution using bilinear
660
interpolation. Higher resolution was chosen to maintain the accuracy during the
georeferencing to other data. Georeferencing error was assessed by comparison of
biases between borders of the forested areas on SRTM, Landsat and topographic
maps. The biases between the borders were less than one pixel (30 m resolution) for
all forested areas.
Preparation of cartographic DEM consisted of paper topographic maps scanning,
georeferencing, digitizing of isohypses and the construction of DEM by interpolation
between them. Georeferencing of scanned topographic maps was carried out to the
UTM projection zone 37N for the WGS84 ellipsoid at points of angles and the center
point by the affine transformation method. RMSE value for all maps proved to be
within one pixel, indicating a high quality of georeferencing. Next, digitization of
isohypses and other data on the altitudes for areas of key sites and their surroundings
was carried out. Additional points were placed on hilltops and in the bottoms of
depressions to simplify work of interpolation algorithm. Then, cartographic DEM
was built using linear interpolation algorithm built into ILWIS. The spatial resolution
of obtained cartographic DEM was 30 m. Additionally the georeferencing of
traditional paper topographic maps at 1:50 000 scale was also made for study areas.
The forested areas were mapped by decoding of satellite images LANDSAT
7TM+ acquired in May 2000 and 2001. These images were selected as the closest
date to the SRTM mission. For the recognizing of the forested areas, a training
sample set was built with the following objects: forests, croplands, grasslands,
settlements and water bodies. These samples were used for the automated
classification of the forested areas by maximum likelihood method. The accuracy of
forests classification was assessed by error matrix technique only for “Chashnikovo”
key site because of close spectral characteristics of forests on all key sites (Zhogolev
& Savin, 2016b). Random validation sample set of 642 pixels was used for
classification accuracy assessment. This was based on visual interpretation of
Landsat images with the help of high resolution images Worldview 2 (Zhu & Liu,
2014). The overall, producer’s and user’s accuracies were higher than 97%.
For the analysis of the spatial influence of forests on SRTM, according to DEM,
aspect, slope, slope form (convex-concave) and watershed areas maps were built. For
this purpose, we used algorithms described in the manual of ILWIS 3.31 (52North,
2016). They are based on using a sliding window of 5 × 5 pixels for the analysis of
surface curvature. Calculation of aspect and slope maps was done in the original
resolution for the SRTM of 90 m (UTM projection) to avoid the use of interpolated
altitude values. 8 points of the compass were used for aspect map. Slope maps were
built in increments of 1 degree.
Estimation of the forest influence was done according to the following method. At
first, SRTM was deducted from the cartographic DEM. Then statistical analysis was
performed separately for the altitude differences across forested areas and areas
without forests and settlements. Hypothesis of a normal distribution of DEM
differences was tested using the Kolmogorov-Smirnov test; histograms, arithmetic
means and the medians, extreme values, and the standard deviations of altitude
differences were analyzed. In addition, linear regressions were fitted, where the
independent variable was the altitude of cartographic DEM and dependent - altitude
of SRTM. The slope, aspect, and slope form maps were compared by calculating the
proportion of pixels with the same value to the total number of pixels. For a
661
comparison of the watershed maps, the numbers of recognized watersheds were
calculated. The difference in number of watersheds was calculated as the number of
watersheds for SRTM minus the number of watersheds for cartographic DEM.
We offer an easy approach for SRTM adjustment that can be implemented in most
GIS. Improvement of SRTM in forested areas can be made using a method of
altitude reduction and smoothing with the help of bilinear interpolation. The idea is
to resample the altitudes of the forested areas to a lower spatial resolution for
catching the main profile curvature and for smoothing influence of the tree heights
heterogeneity. Under the forest mask, we reduce median overestimation of altitudes
due to tree heights. Then, the forested parts of SRTM are returned to the original
spatial resolution also by using bilinear interpolation.
At first step, SRTM with spatial resolution 90 m was resampled to 30 m resolution
(as it is of Landsat satellite images). From altitudes in the forested areas, median
value of overestimation of altitudes was subtracted. Then smoothing was carried out
by resampling of the SRTM to a lower resolution (180 m) using bilinear
interpolation. After that, the SRTM was resampled to 30 m resolution and non-
forested areas of the SRTM were replaced with the original data (not resampled to
low resolution). At last step, the SRTM was resampled to the original 90 m
resolution (UTM). To assess the quality of the corrected DEM SRTM an analysis
similar to describe above was carried out.
4 Results
Median altitude differences between SRTM and cartographic DEM within the
forested areas were 3-4 times less than the mean height of the forest according to
topographic maps (Tab. 3). This is due to a systematic error of the SRTM vertical
positioning, the overgrowing of clearings with shorter trees, different distances
between the trees, different influence of various species of trees and other factors.
The greatest influence, presumably, is caused by the systematic error of the SRTM
vertical positioning.
662
Table 3. Descriptive statistics of the difference between DEMs.
Mean, Median, St.Dev.,
Key site Min., m Max., m
m m m
“Chashnikovo”, whole site 0.8 1.3 7.5 -17.2 29.1
(forest height– forest part 5.4 6.0 6.8 -16.0 25.5
21m) without
forests and
settlements -6.5 -7.0 4.8 -16.9 17.1
“Schebanovo”, whole site 4.9 5.9 7.0 -14.5 22.3
(forest height– forest part 7.8 8.0 5.0 -11.5 22.1
24m) without
forests and
settlements -2.5 -3.5 5.8 -14.5 18.8
“Serebryaniye whole site -2.1 -4.0 6.3 -16.0 25.5
prudy”, forest part 5.4 6.0 6.8 -16.0 25.5
(forest height– without
22m) forests and
settlements -4.5 -4.9 3.5 -16.0 21.3
General whole site 1.2 0.3 7.5 -17.2 29.1
information for forest part 6.8 7.1 5.4 -16.0 25.5
all key sites without
forests and
settlements -4.5 -5.0 4.4 -16.9 21.3
As the systematic error of the SRTM vertical positioning the median altitude
differences between DEMs in territories without forests and settlements can be
considered (Table 3). Consequently, for all key sites median overestimation of
altitudes due to the influence of forests is about 0.5 of the mean height of the forests
from topographical maps, which is consistent with other studies (Hengl et al., 2009).
So, this value of mean height can be used for the correction of SRTM and as an
estimation of maximum variation of tree heights in forested areas.
In addition to portions of images with positive difference between the altitudes on
SRTM and cartographic DEM, caused by the forests influence, there are vast areas
with high negative values of the differences. The biggest negative differences
between DEMs (up to -17 m) are observed in the floodplain of the Klyazma River in
the key site "Chashnikovo" (Tab. 3). Comparison with the topographic maps at
1:50 000 scale showed that large in magnitude negative values are usually caused by
the generalization of topographic maps at 1:100 000 scale. For example, in the
floodplain on the site "Chashnikovo" on maps at 1:50 000 scale there is an additional
contour, which lies close to the main contour, in comparison with maps at 1:100 000
scale, which leads to a bigger difference between SRTM and cartographic DEM
based on maps at 1:100 000 scale. In accordance with the values in table 4, the
SRTM DEM without correction has the same quality as DEM based on topographic
maps at 200 000 scale.
The proportion of coincidences of aspects and slopes constructed from SRTM and
cartographic DEM, were very low (Tab. 4).
663
Table 4. Comparison of maps calculated from SRTM and cartographic DEM.
Slope Difference
Aspect Slope form in the
Key site
matches,% matches,% matches, number of
% watersheds
“Chashnikovo” whole site 31 31 37 -15
forest part 30 29 37 -
without
forests and
settlements 32 35 38 -
“Schebanovo” whole site 16 32 40 -48
forest part 15 28 38 -
without
forests and
settlements 15 35 43 -
“Serebryaniye whole site 37 37 42 11
prudy” forest part 29 27 30 -
without
forests and
settlements 39 40 45 -
For the forest part, the share of coincidences is always lower than for open areas.
Biases of 1 degree on cartographic DEM often correspond to slopes of 2 degrees on
SRTM, which is typical not only for the forests, but also for open areas, although to a
much lesser extent. The proportion of such slopes for the site "Schebanovo" is more
than the proportion of matched, while there are more coincided slopes on the site
“Serebryaniye prudy”, and on the site "Chashnikovo" an intermediate situation is
observed. Decrease in the severity of the effect described above and increase of
altitude differences show that the SRTM does not convey the slope relief very well.
Similarly, differences in aspects decrease with the increasing of altitude differences.
Smaller proportion of matches in the forested areas is probably due to the influence
of SRTM by varying density and height of the trees. This effect can be adjusted by
isolating the total curvature of the relief and interpolating the intermediate values.
The adjustment of the effect of forest vegetation on the SRTM using bilinear
interpolation algorithm has led to an increase of the correlation with the data based
on the cartographic DEM. All parameters improved, the increase was quite moderate:
Spearman's correlation coefficient between the altitudes of DEMs increased by 0.05 -
0.14, the percent of coincided aspects by 1 - 4%, and the percent of coincided slopes
by 2 - 8% (Tab. 5). However, on the altitude map of adjusted SRTM, the forests
became hardly visible.
664
Table 5. Comparison of original SRTM and corrected SRTM.
Key site Statistics Original Corrected
2
“Chashnikovo” R (linear regression between 0.83 0.90
altitudes)
p - the level of significance of the < 2.2 × 10-16 < 2.2 × 10-16
linear regression
aspect matches, % 31 35
slope matches, % 30 38
slope form matches, % 37 46
difference in the number of -15 -12
watersheds
“Schebanovo” R2 (linear regression between 0.47 0.61
altitudes)
p - the level of significance of the < 2.2 × 10-16 < 2.2 × 10-16
linear regression
aspect matches, % 16 17
slope matches, % 64 72
slope form matches, % 40 60
difference in the number of -48 -46
watersheds
“Serebryaniy R2 (linear regression between 0.88 0.93
prudy” altitudes)
p - the level of significance of the < 2.2 × 10-16 < 2.2 × 10-16
linear regression
aspect matches, % 37 39
slope matches, % 37 39
slope form matches, % 42 45
difference in the number of 11 11
watersheds
The correction led to severe improvement in slope values on the edges of forests,
the sharp drop on the slope maps became hardly noticeable (Fig. 1). For the aspect
and slope form (convex-concave) maps, the difference between SRTM and adjusted
SRTM is not evident, but these maps visually became slightly smoother that is more
consistent with cartographic maps. The number of watersheds slightly increased,
which is more consistent with cartographic DEM. The correction was performed by
resampling the SRTM to the spatial resolution of 180 m and 360 m, however, for
most cases, filtering using a spatial resolution of 180 m was more effective, so the
table 5 shows data only for filtering with the resolution of 180 m.
665
Figure 1. Maps calculated from cartographic DEM, SRTM and adjusted SRTM (maps are
arranged in the following order from the top to the bottom: cartographic DEM, SRTM,
adjusted SRTM).
After adjustment of the SRTM, areas with the highest values of altitude
differences between the adjusted SRTM and cartographic DEM were analyzed. The
visual analysis of borders of the highest difference between DEMs using Quickbird
high-resolution images and topographic maps at 1:50 000 scale showed that the
greatest differences are confined to areas of maps at 1:100 000 scale with a
significant generalization. It could and did lead to the difference in altitude between
SRTM and cartographic DEM of more than 12 meters in the key site “Chashnikovo”.
Against the background of such large errors the influence of forest vegetation of
varying density and species composition on altitude of the model delineated from
Quickbird images proved to be insignificant. Thus, the quality of the DEM SRTM in
some forested areas is probably better than the quality of cartographic DEM based on
maps at 1:100 000 scale.
666
5 Conclusions
The influence of boreal forests on the SRTM in the studied region is clearly seen
visually and by statistical analysis when comparing with the DEM based on
traditional topographic maps at 1:100 000 scale. According to our data, the quality of
SRTM is comparable to DEM based on traditional topographic maps at 1:200 000
scale. If forest influence is completely removed, the quality will be close to the DEM
based on topographic maps at 1:100 000 or even 1:50 000 scale.
Maps of slopes, aspects, slope forms and watersheds calculated from SRTM
differ significantly from the same maps built from cartographic DEM at 1:100 000
scale. After the correction of SRTM using the proposed method, based on smoothing
DEM by the bilinear interpolation, the quality of aspect and watershed maps slightly
improved. The quality of the slope and slope form maps improved significantly. On
the slope maps, after adjustment, there was only little increase in the slope value on
the border of the forests, when before correction there had been the large jump in
slope value. After adjustment, the number of recognized watersheds increased which
is more consistent to the DEM built from topographic maps. Therefore, the changes
in SRTM after the correction would lead to the changes in land use planning, e.g. in
distinguishing lands with a high risk of soil erosion or located in different watershed
areas.
Proposed method based on bilinear smoothing of altitudes under the Landsat
forest mask allowed moderately improving the quality of SRTM. The method is
better to be applied to flat areas with forests which height varies little (less than a half
of their average height). The advantage of the proposed method is the simplicity of
its application and the opportunity for using in almost all GIS supporting bilinear
resampling. The further improvement of the method is required to consider trees
heights more accurately and to reduce noise on the forest edges associated with big
difference in tree heights. Such noise can be ignored in case when the forest edge
length is one pixel or less as the calculation of slopes, aspects, slope forms and
watersheds usually use five pixels and will not be affected significantly.
Acknowledgements. This research is supported by Russian Scientific Fund GRANT
NUMBER 15-16-30007.
References
1. 52North: ILWIS 3.0 User’s Guide [Internet]. (2016). Muenster: 52°North
Initiative for Geospatial Open Source Software GmbH; [cited 2017 May 10].
Available from: http://52north.org/communities/ilwis/ilwis-open/user-guide.
2. Amans, O.C., Beiping, Wu, Ziggah, Yao Y. (2013). Assessing vertical accuracy
of SRTM ver 4.1 and ASTER GDEM ver 2 using differential GPS measurements
- case study in Ondo State Nigeria. International Journal of Scientific and
Engineering Research, 4(12), p.523-531. ISSN 2229-551.
3. CGIAR CSI: SRTM 90m Digital Elevation Database v4.1 [Internet]. (2016).
Washington: The Consortium for Spatial Information of the Consultative Group
667
for International Agricultural Research: [cited 2017 May 10]. Available from:
http://www.cgiar-csi.org/data/srtm-90m-digital-elevation-database-v4-
1#version_history.
4. Du X., Guo H., Fan X., Zhu J., Yan Z., Zhan Q. (2015). Vertical accuracy
assessment of freely available digital elevation models over low-lying coastal
plains. International Journal of Digital Earth, 9, p.252–271. doi:
http://dx.doi.org/10.1080/17538947.2015.1026853
5. Farr T.G., Caro E., Crippen R., Duren R., Hensley S., Kobrick M., Paller M.,
Rodriguez E., Rosen P., Roth L., et al. (2007). The Shuttle Radar Topography
Mission. Reviews of geophysics, 45, p.1–43. doi:
http://dx.doi.org/10.1029/2005RG000183.
6. Gallant J.C., Read A. (2009). Enhancing the SRTM data for Australia.
Proceedings of Geomorphometry, 3, 149-154.
7. Gallant J.C., Read A.M., and Dowling T.I. (2012). Removal of tree offsets from
SRTM and other digital surface models. Int. Arch. Photogramm. Remote Sens.
Spatial Inf. Sci., XXXIX-B4, 275-280. doi: 10.5194/isprsarchives-XXXIX-B4-
275-2012, 2012.
8. GKINP-05-029-84. (1984). The main propositions to create and update
topographic maps at scales 1:10 000, 1:25 000, 1:50 000, 1:100 000, 1:200 000,
1:500 000, 1:1000 000. Moscow: Public and Editorial Department of military and
technical cooperation. (In Russian).
9. Hengl, T. (2006). Finding the right pixel size. Computers & Geosciences, 32(9),
p.1283-1298.
10. Hengl, T, Reuter, H.I., editors. (2009). Geomorphometry: concepts, software,
applications. Amsterdam: Elsevier. ISBN 978-0-12-374345-9.
11. Hirt, C. (2016) Digital Terrain Models. Encyclopedia of Geodesy: p.1–6.
doi:10.1007/978-3-319-02370-0_31-1.
12. Hofton, M., Dubayah, R., Blair, J.B., Rabine, D. (2006). Validation of SRTM
elevations over vegetated and non-vegetated terrain using medium footprint lidar.
Photogrammetric engineering and remote sensing, 72(3), p.279–285. doi:
http://dx.doi.org/10.14358/pers.72.3.279.
13. Karwel, A.K., Ewiak, I. (2012). Estimation of the accuracy of the SRTM terrain
model on the area of Poland. Int. Arch. Photogramm. Remote Sens. Spatial Inf.
Sci. (XXXIX-B7), 51–55. doi:10.5194/isprsarchives-xxxix-b7-51-2012.
14. Mulder, V.L., de Bruin, S., Schaepman, M.E., Mayr, T.R. (2011). The use of
remote sensing in soil and terrain mapping – A review. Geoderma, 162, p.1–19.
doi: http://dx.doi.org/10.1016/j.geoderma.2010.12.018.
15. Nelson, A., Reuter, H.I., Gessler, P. (2009). Chapter 3 DEM production methods
and sources. Developments in Soil Science, 33, p.65–85. doi:
http://dx.doi.org/10.1016/s0166-2481(08)00003-2.
16. Ozah, A.P., Kufoniyi, O. (2008). Accuracy assessment of contour interpolation
from 1:50 000 topographical maps and SRTM data for 1:25 000 topographical
mapping. XXI ISPRS Congress XXXVII-B7, p.1347-1353.
17. Rodríguez, E., Morris, C.S., Belz, J.E. (2006). A Global Assessment of the
SRTM Performance. Photogrammetric Engineering and Remote Sensing, 72(3),
p.249–260. doi:10.14358/pers.72.3.249.
668
18. Shortridge, A., Messina, J. (2011). Spatial structure and landscape associations of
SRTM error. Remote Sensing of Environment, 115, p.1576–1587. doi:
http://dx.doi.org/10.1016/j.rse.2011.02.017.
19. Shukla, M.K. (Ed.). (2011). Soil hydrology, land use and agriculture:
measurement and modelling. doi:10.1079/9781845937973.0000
20. Zhogolev, A.V., Savin, I.Y. (2016a). Automated updating of medium-scale soil
maps. Eurasian Soil Science, 49(11), p.1241-1249.
21. Zhogolev, A.V., Savin, I.Y. (2016b). The influence correction of boreal forest
vegetation on SRTM data. Geocarto International, (just-accepted), p.1-34.
22. Zhu, X., Liu, D. (2014). Accurate mapping of forest types using dense seasonal
Landsat time-series. ISPRS Journal of Photogrammetry and Remote Sensing, 96,
p.1–11. doi: http://dx.doi.org/10.1016/j.isprsjprs.2014.06.012.
669