Accuracy Assessment of Digital Elevation Model Generated from Pleiades Tri stereo-pair Salman Nasir*1, Irfan Akhtar1 Iqbal, Zahir Ali1, Atif Shahzad1 1
Training, Research and Development Directorate, Pakistan’s Space and Upper Atmospheric Research Commission (SUPARCO), P.O.Box 8402, Karachi Pakistan. *Corresponding
Author:
[email protected]
KEYWORDS :( DTM, DEM, DSM, Pleiades, Tri Stereo-pair, RPC)
ABSTRACT: Digital Elevation Model (DEM) is crucial for several purposes like town planning, hydrological analysis, land sliding, flash floods, earthquake, road construction, surface analysis, ortho-rectification of satellite imagery, 3D visualization, precise farming and forestry, base mapping, flight simulation and disaster management. Pleiades is a French constellation of very high resolution satellites. It acquires both panchromatic as well as multispectral imagery in Visible Near Infra-red (VNIR) range. The added benefit of Pleiades is that it provides tri stereo-pair imagery at 0.5m spatial resolution, unlike its other contemporary systems like Quickbird and IKONOS. Tri stereo-pair is used for Digital Terrain Model (DTM) and Digital Surface Model (DSM) extraction because of its backward and forward look angles. Moreover, Pleiades provides Rational Polynomial Coefficient (RPC) and sensor model data which augment the accuracy of its 3D products. Being a newly launched system, it provides fertile field for researchers to analyze the strengths and weaknesses of this system. This study explores the potential of Pleiades Tri stereo-pair in generating high resolution DEM, and comparing its accuracy with Shuttle Radar Topography Mission (SRTM) and Advanced Space Thermal Emission Radiometer (ASTER). In this study, data from space borne LiDAR (ICESat/GLAS) was used as a reference due to its reported unprecedented accuracy. Comparison of Pleiades with LiDAR resulted in an R2 of 0.92 with an RMSE of 5.2m. Similarly, comparison of SRTM and ASTER resulted in an R2 of 0.74 (RMSE 7.5m) and R2 0.84 (RMSE 6.6m), respectively. Conventionally, DEM/DTM is generated from the satellite
1. Introduction
stereo-pair which does not inform on the surface heights (DSM).
Pleiades-1A was launched successfully on December 16, 2011 in
Tri-stereo pair, on the other hand, gives information about the
its orbit from a Russian Soyuz ST rocket out of French Guiana.
terrain (DTM) as well as the height of the surface above ground
Pleiades-1A is the first satellite of the Pleiades constellation. The
(DSM). Objectives of this study were:
Pleiades constellation will comprise of two very high-resolution
i) To explore the potential of Pleiades Tri stereo-pair in
optical Earth-imaging satellites on a Sun-synchronous orbit at
generating a high resolution DEM (10 m)
694 km (Poli et al., 2013). The most prominent benefit of the
ii) To assess the accuracy of Pleiades Tri stereo-pair DEM with
Pleiades system is that it offers stereoscopic coverage of quite
space
high resolution. It has the ability to produce highest accuracy
borne
LiDAR(ICESat/GLAS)
of
unprecedented
accuracy (15cm)
through both a forward and backward looking stereo pair.
iii) To compare the accuracy of Pleiades DEM with SRTM-90m
However, this combination has limitations over gentle terrain
and ASTER-30m DEMs
areas. In terrains with high topographic variations, a nadir, forward and backward looking tri-stereo pair can be used to
2. Study Area
overcome the inaccuracies due to topography (Philip Cheng, 2012).The sensor model & Rational Polynomial Coefficients
Melbourne is located at latitude 37°48′49″S and longitude
(RPCs) are provided with the data. All this information is
144°57′47″E in the south-eastern part of mainland Australia,
required to extract a high resolution DSM (also referred to as a
within the state of Victoria. Melbourne’s major bayside beaches
DEM) to represent the earth’s surface as well as the objects such
are located in the south-eastern suburbs along the shores of Port
as buildings and trees on it. However, many applications require a
Phillip Bay, in areas like Melbourne. In comparison with other
DTM, which is derived from DSM, to represent the bare ground
Australian cities, Melbourne’s buildings have unrestricted height
surface with no objects.
limits. Out of six, Melbourne contains five tallest building of Australia. Eureka Tower is the tallest of all and it is situated at Southbank. 553
2.3.1 Acquisition of Tri Stereo-pair imagery Pleiades 0.5m resolution Tri Stereo-pair of Melbourne was acquired. Tri Stereo-pair is used for Digital Terrain Model (DTM) and Digital Surface Model (DSM) extraction because of its backward and forward looking-angles. Angles were identified with the help of buildings’ shadow. 2.3.2 Block File Generation Block file (*.blk) is an extension file that stores all the steps of processing for DEM extraction. This file stored the information of assigned projection, interior and exterior orientation, GCPs information (Tie, Control, and Check) and Root Mean Square
Figure 2: Pleiades satellite Imagery of Melbourne
(RMS) accuracy of extracted DEM. This tower also has an observation deck, from where all of the Melbourne’s structures are visible. It has an average elevation is 31m (102) feet (Melbourne).The study area covers variation in
A
topography; it provides accuracy for DEM assessment at lower and higher elevations. 2.2 Data Sets Used Following datasets was used for DEM/DSM generation and for accuracy assessment:
B
Pleiades Tri Stereo-pair imagery having 0.5m spatial resolution SRTM (Shutter Radar Topographic Mission) 90m DEM ASTER (Advance Space Borne Thermal Emission Radiometer) 30m DEM LiDAR points data with 15 cm accuracy
2.3 Methodology
C
The DEM generation is a very complex process and requires high accuracy in inputs and data handling. The methodology adopted in this paper is shown in Figure 2:
Figure 3: Left (A), right (B) and third angle(C) Tri Stereo-pair imageries
2.3.3 Geometric Model Pleiades has a “push broom” scanner. It has HRG2 instrument (High Resolution Geometry) that captures panchromatic stereopair imagery. It provides automated geometric correction which is helpful in self-calibration of interior orientation. Figure 1: Methodological flow of the study
554
in metadata supplied by the vendors of satellite imagery will
2.3.4 Defining Projection
result in IO errors. This can be rectified through efficient
Universal Transverse Mercator (UTM) south zone-55, projection
modeling approach.
was defined as Metric Coordinate system and WGS-1984
This
method
is
straightforward to
implement and it can be widely applied to other satellite
projection was defined as Geographic Coordinate System for
imagery for precise geo-referencing (Zhang et al., 2012).
stereo-pair imagery. 2.3.5 Interior Orientation Precise geo referencing is required for exploiting the full geometric or metric quality of optical satellite imagery. For this purpose numerous senor orientation models have been designed that are capable of utilizing the maximum metric potential of these images which had been developed over the last few decades. Specifically speaking, “Rational Function” model is
Figure 3: Rotation Angles in Degrees (Gessler, Spring, 2009)
more widely employed because it accounts for an entire physical imaging process. This model also takes the information about the
2.3.6 Exterior Orientation
factors such as Interior Orientation (IO) and satellite orbit A method applied for improving the Exterior Orientation (EO)
models. This information is mostly provided by the satellite
by using accurate ground control points, tie points and check
imagery vendors. Each satellite provides unique quality of IO
points for DEM extraction. Parameters of the EO, at the time the
parameters which significantly affects the geo-referencing
image is captured, are stored in the metadata or RPC file which
performance. Various methods for self-calibration have been
include Perspective Center (in meters), Rotational Angles (in
developed albeit such methods have a high requirement for
degrees)(see figures 4 & 5), focal length of the satellite, ground
substantial amount of ground control along with good point
principal points, pixel size and incident angle along and across
distribution (Zhang et al., 2012).
track. In addition to these, information about the right, left and third angle imagery is also provided.
But, due to correlation between different model parameters the results are not always stable. The “Rational Function” sensor model has successfully been applied to satellite sensors like Quickbird, IKONOS, World View and Pleiades. These satellites produce Rational Polynomial Coefficient (RPC) file as metadata file, which is used to illustrate the modeling procedures. The IO parameters are firstly computed by spatial resection using the vendor-supplied instrument view angles provided in the metadata. This information is provided for each individual pixel used to model the line-of-sight with respect to pixel location in the focal plane. The computed IO parameters are then adopted in the rational function sensor model and the metric performance of
Figure 4: Perspective Center in Meters (3D-Axis Rotation, 2010)
the satellite imagery is examined. The precision of line-of-sight There are three types of points namely i) Control points, ii)
data depends on the quality of IO estimations. If errors occur in
Check points and iii) Tie points where each has three types of
these data then it will affect the results and IO parameters will
orientation which is Full, Horizontal and Vertical. “Full”
be of poor quality, this will lead to degradation of the geometric
orientation has X, Y and Z coordinates. “Horizontal” orientation
potential of the imagery.
has X and Y coordinates; whereas the “Vertical” orientation Detailed examination reveals that the IO errors in these satellites
contains the Z values which is unknown and can be estimated or
demonstrate a similar distribution pattern. The errors in IO
ignored during triangulation. “Control” points are used for
parameters are compensated through rational functions. This
triangulation purpose. “Check” points are used for verifying the
modeling approach greatly improves metric performance with
quality of triangulation and “Tie” points indicate the position of
geo-referencing accuracy at sub-pixel level. Any impreciseness
ground point that appeared in the overlapping areas of two or 555
more images (Jacobsen, 2001). The X, Y and Z coordinates are determined by triangulation. These points can be collected manually or automatically. In this study, 80 well-spread ground control points were generated. Further, 143 tie points were generated using these GCPs. It should be noted that a higher number of GCPs result in higher accuracy of the extracted DEM by reducing the shift in ortho-rectification of the imagery. Extreme care should be taken Figure 5: Pleiades-10m DSM to DEM conversion
in selecting the GCPs, because the accuracy of the resulting DEM is highly dependent on them. Using all the parameters of
DSM has information about surface elevation and DEM has only
interior and exterior orientation in Leica Photogrammetry Suite
terrain elevation information. So Pleiades-10m DSM now be
(LPS), a 10m resolution DSM was generated, which was later
converted into Pleiades-10m DEM.
converted into DEM. 3.1 Accuracy Assessment using LiDAR (ICESAT/ GLASS)
Geoscience Laser Altimeter System (GLAS) onboard the Ice, Cloud and land Elevation Satellite (ICESat) is the first space
The coincident elevation values were extracted from Pleiades,
borne LiDAR and is acknowledged for its unprecedented
SRTM and ASTER DEMs using LiDAR data for comparison
accuracy of 15cm.LiDAR points are available in line strips
(Figure 8).
spaced at about 172 m (Iqbal,2013). On the basis of data available for Melbourne city, Pleiades, SRTM and ASTER DEMs were subset to match the extent of LiDAR point data. Moreover, the SRTM-90m and ASTER-30m DEMs were resampled to 10m for the purpose of comparison and accuracy assessment. 3.0 Results and Discussion As DSM (Figure 6) describes the visible surface elevation and DEM describes the bare ground elevation, for accuracy assessment DEM is required. For this purpose PCI Geometica was used to convert DSM to DEM (Figure 7).
Figure 8: Accuracy assessment with ICESat/GLAS point data
The comparison of Pleiades-10m DEM with LiDAR point elevation resulted in anR2 0.92 and RMSE 5.2m (Figure 9).A systematic error could also be noticed in the figure showing a greater potential for correcting the Pleiades data. Similarly, the comparison of ASTER-30m & SRTM-90m DEMs with LiDAR point elevation data gave correlation 0.91, RMSE 6.65 as well as
Figure 6: DSM (10 m) generated using IO and EO parameters
correlation 0.86, RMSE 7.5 m respectively. Results show that Pleiades-10m DEM is closest to LiDAR data as compared to ASTER & SRTM DEM.
556
For this purpose, ASTER-30m & SRTM-90m DEMs were also resampling at 10m DEMs. The correlation graph between Pleiades-10m & LiDAR was very good which 0.963%.All the elevation points were very close but lower side of one-to-one line. It shows that Pleiades-10m & LiDAR was very close to each other in vertical and horizontal accuracy (Figure 9).
Figure 11: Correlation between SRTM-10m resampled DEM with LiDAR Table 1: Correlation between LiDAR, Pleiades-10m, ASTER10m and SRTM-10m
Pleiades Pleiades
Figure 9: Correlation between Pleiades-10m DEM with LiDAR. The one-one
1
LiDAR
ASTER
0.963
SRTM
0.912
0.844
Table 4 shows that Pleiades-10m DEM was in close in
line is shown in red dashed line.
agreement with elevations from LiDAR (ICESAT/GLAS). This was followed by ASTER-10m and SRTM-10m respectively. The correlation graph between ASTER-10m & LiDAR shows
Here also generated correlation between LiDAR, Pleiades-10m,
that at low elevation ASTER has good relation with LiDAR and
ASTER-10m and SRTM-10m DEMs.
gradually it becomes weak as the elevation points goes to higher 4.0 Conclusion
latitude (Figure 10).
Results showed that Pleiades 10 m DEM gave RMSE of 5.2m with LiDAR as a reference. Pleiades Tri Stereo-pair gave precise results due to its forward and backward look angles. Tri stereo-pair can also generate DSM with 1 m positional biased. Comparison
with
SRTM,
ASTER
and
LiDAR
(ICESAT/GLAS), showed that Pleiades Tri Stereo-pair is very useful for DEM/DSM generation due to its maximum information content and high resolution. Pleiades Tri Stereo-pair is very helpful for 3D model generation.
Figure 10: Correlation between ASTER-10m resampled DEM with LiDAR
The correlation graph between SRTM-10m & LiDAR shows that there was weaker correlation. The elevation points were scattered upper and lower side of the one-to-one line. At mid elevation ranges between 18m to 25m; all the points were near to one-to-one line. That is why correlation of these points was good (Figure 11).
557
5.0 References Poli,D.,Remondino, F., Angiuli,E.,Agugiaro, G., 2013. Evaluation of pleiades-1a triplet on Trento testfield. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-1/W1, pp. 287292,ISPRS Hannover Workshop, Hannover, German. Philip, Cheng, 2012.Geometric Correction, Pan sharpening and DTM extraction from Pleiades Satellite. Geoinformatics Magazine for Surveying, Mapping & GIS Professionals, Volume 15, pp.10-12. Melbourne, 2013.http://en.wikipedia.org/wiki/Melbourne (Last access November 2013). Zhang, C., Fraser, C. S., Liu, S., 2012. Interior Orientation Error Modeling and Correction for Precise Georeferencing of Satellite Imagery. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B1, 2012, pp.285-290. Iqbal, I.A. and Dash, J. and Ullah, S. and Ahmad, G. (2013) A novel approach to estimate canopy height using ICESAT/GLAS data: a case study in the New Forest National Park, UK. International journal of applied earth observation and geoinformation, 23. 109 - 118. ISSN 1569-8432 Jacobsen, 2001. Exterior Orientation Parameters, Institute for Photogrammetry and GeoInformation, University of Hannover, Germany, pp.1321 – 1332. 3D-Axis Rotation, 2010. http://File:Roll Pitch Yaw.JPG Wikimedia Commons (Last access November 2013). Gessler, Spring, 2009. Introduction to Geospatial Analysis for Natural Resource Management. Leica Photogrammetry Suite Project Manager: Users Guide, Leica Geosystems LLC. Leica Photogrammetry Suite 9.2, pp.10.
558
