The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
The SRTM Digital Elevation Model was processed and
maintained by the Consultative Group for International
Agriculture Research Consortium for Spatial Information
(CGIAR-CSI). In the form compiled and maintained by
CGIAR-CSI, the SRTM elevation data have a spatial resolution
of 90m. This data set is seamless with all voids filled using a
methodology based on spatial filtering (Gorokhovich and
Voustianiouk, 2006). The CGIAR-CSI SRTM 90m digital
elevation data sets are provided to the general public in 5° X 5°
tiles in computer-compatible raster formats (GeoTiff and
ARC/INFO ASCII Grid). The data set is in LatLon coordinate
system projected on the WGS 84 Ellipsoid. For the purpose of
our study, one of the tiles covering the chosen study site was
downloaded from the CGIAR-CSI Web site at http://srtm.csi-
cgiar.org.
One topographic map sheet at the scale of 1/50,000 covering the
chosen study site was selected for use as a reference for
analyzing the SRTM elevation data. The map sheet was
digitized into layers (hypsometry, hydrography, transportation
and built-up) as part of the input data sets within the
framework of an on-going state-wide topographical mapping
project undertaken by the Ondo State Government of Nigeria.
In this accuracy assessment study, only the hypsometric and
hydrographic layers were obtained from the project consultants.
The selected layers were digitized from an existing paper map
sheet compiled in 1965. The map was based on the UTM
projection (Zone 31) on Clarke 1880 Ellipsoid and had a
contour interval of 50 feet.
GPS elevation data used in this study were acquired during a
field truthing mission organized by the Ondo State
Topographical Mapping project consultants as part of activities
within the framework of the state-wide 1/25,000 topographical
mapping project. During the GPS survey exercise, GPS
measurements were made with a vertical accuracy of ± lm at
randomly visited points. Since the GPS observations were
meant to validate contours interpolated from the existing
topographical maps, they were originally transformed into the
coordinate system of the maps (UTM projection on Clarke 1880
Ellipsoid).
3.2 Materials
Three major software packages were employed for the
processing of the data and the visualization and analysis of the
results. These included the royalty-free, open-source Integrated
Land and Water Information System (ILWIS 3.4), ArcGIS 9.2
(proprietary) and Microsoft Excel. In addition, we developed a
number of in-house programs in Visual Basic 6.0 for
performing some specialized functions such as coordinate
transformation, terrain profiling and elevation data extraction
from ASCII raster data sets.
3.3 Methodology
The methodology adopted in this study was in keeping with the
main objectives of the study as stated in Section 1 of this paper.
3.3.1 Data preparation: The data sets employed in this study
emanated from disparate sources based on different formats,
coordinate systems and projections. The first step in the
exploitation of the data sets was therefore the transformation of
all the data sets into a common system. Since the CGIAR-CSI
SRTM 90m digital elevation data sets were in LatLon WGS 84
system, the topographic map layers (contours and rivers) and
the GPS elevation data in UTM Clarke 1880 system were
transformed into the LatLon WGS84 system using tools
available in ILWIS 3.4 software. To restrict the test to the
chosen study site, it was expedient to extract only the GPS
points that fell within the extents of the study area. To perform
this operation, we implemented a small program in Visual Basic
6.0 to clip the GPS point set using the extents of the study site.
Using our program, the hypsometric layer with contour values
in feet was metricated by transforming it into a new layer with
all the contour values multiplied with a Z-factor of 0.3048
Since our study also involved the accuracy tests of contour
interpolation from the 1:50,000 topographical map, it was
necessary to process the source 1:50,000 topographic map into
a form appropriate for the test. To satisfy this requirement, we
created a grid-based digital elevation model with a resolution of
90m (corresponding to the resolution of the CGIAR-CSI SRTM
DEM) from the metricated 1:50,000 topographic map using the
contour interpolation function available in ILWIS 3.4. This
involved first rasterizing the contour map layer and then
interpolating between the isolines using the method described in
Gorte, B.G.H. and Koolhoven W. (1990).
3.3.2 Determination of the vertical accuracy of Topo DEM
and SRTM DEM: Several publications on the accuracy of
CGIAR-CSI SRTM 90m elevation data report that its absolute
vertical accuracy is in the order of ± 16m (Koch, A. and
Lohmann, P., 2000; Miliaresis, G. and Paraschou, C. V. E.,
2005; Muller, J. P., 2005). This accuracy value has been
extensively tested in different regions under different terrain
characteristics by many researchers (Giacomo F. et al, 2005;
Brown, C. G. et al, 2005). Results of such tests showed that the
absolute vertical accuracy of the SRTM elevation data depends
on terrain characteristics. In Gorokhovich and Voustianiouk
(2006) for example, it was shown that two topographic
derivatives, slope and aspect, significantly influence the
absolute vertical accuracy value. The study showed that steeper
slopes recorded higher vertical errors than gentler slopes, while
SRTM data underestimated elevations with North West aspect
and overestimated elevations with South East aspect. In our
study, emphasis was placed only on the absolute vertical
accuracy of Topo DEM and SRTM data covering our study site.
Determining the absolute vertical accuracy of SRTM data
basically involves computing the standard deviation statistic of
the elevation differences between the SRTM data and a
reference data set such as GPS point measurements. This
requires overlaying the GPS points on the SRTM and extracting
the heights from the two data sets at their position of
coincidence and using these values to compute the accuracy
statistic. Gorokhovich and Voustianiouk (2006) described a
method for performing the overlay. Their approach involved
first converting the SRTM raster data set into a vector-based
GIS layer containing as many polygons as there were grid cells
in the SRTM data and then performing a spatial join of the
point data and the new polygonal layer to extract the height data
for the statistical analysis. This method may prove to be highly
demanding in computer memory and may turn out to be
computationally-intensive to handle especially where the point
data set is large. In our study, we adopted a simpler method of
performing the spatial join. Our approach involved projecting
the X, Y Cartesian coordinates of the point data into their
equivalent grid image space (rows and columns). These rows
and columns were then used to access the value stored at the
corresponding grid cell location. We implemented a small
Visual Basic program (using the MapWindow programmable