993 
VERY HIGH RESOLUTION STEREO DTM EXTRACTION AND ITS APPLICATION TO 
SURACE ROUGHNESS ESTIMATION OVER MARTIAN SURFACE 
J. R. Kim*, J-P. Muller 
Milliard Space Science Laboratory, Department of Space and Climate Physics, University College London, Holmbury 
St. Mary, Dorking, Surrey, RH5 6NT, UK jk2@mssl.ucl.ac.uk, jpm@mssl.ucl.ac.uk 
VVG VI/7 
KEY WORDS: Planetary Mapping, Geology, Automation, Imagery, GIS, DEM/DTM, Fusion, Algorithms 
ABSTRACT: 
We have developed a processing workflow to extract very high resolution DTMs up to 0.5-4m grid-spacing from HiRise (Kim and 
Muller, 2007) and 12- 18m gridded DTMs from CTX stereo pairs. This workflow relies on combining these data with the outputs of 
the UCL HRSC stereo processor, which is capable of producing 30-100m DTMs. This provides a unique capability to observe 
Martian topography at multiple resolutions. One of the most interesting applications for these data sets is local surface roughness 
extraction. An iterative method is employed to try to reduce noise in DTM extraction and applied to reconstruct surface roughness 
data, fusing stereo DTM and MOLA beam broadening effects. Accurate local surface roughness over a wide area will give important 
clues about the composition and origin of the Martian surface which has not been revealed before with methods developed to date. 
1. INTRODUCTION 
The local surface roughness is a very interesting and important 
property of planetary surfaces. It defines how momentum is 
transported from the atmosphere to the surface through the 
aerodynamic surface roughness length, Zo (Muller et al., 2001); 
how dust is raised into the atmosphere and for rock and boulder- 
strewn landscapes how their distributions came about due to 
events such as impact cratering and other explosive events. The 
vertical and horizontal resolution of conventional topographic 
data extraction methods such as stereo analysis have been 
severely limited in the past to extract this valuable information. 
An alternative for planetary surfaces, albeit only for along-track 
footprints, is through the use of laser altimeters. The 
relationship between laser beam pulse spreading and local 
surface roughness was first defined by Gardner (1982). Laser 
beam broadening was first demonstrated for Shuttle Laser 
Altimeter (SLA-01/2, Harding et al., 1994) and the Mars 
Orbiter Laser Altimeter (Abshire et al., 2000). On the other 
hand, local surface roughness extraction directly from height 
measurements has also been tested. The power spectrum 
method (Smith et al., 2001) and the median differential slope 
method (Kreslavsky and Head., 2000) to extract km scale 
surface roughness are good examples. However, in both cases, 
the vertical resolution is largely dependent on the horizontal 
density of MOLA spots which are limited to km scale. This is of 
less interest to surfaces such as Mars. Garvin et al., (1999) first 
demonstrated metre resolution Martian surface vertical 
roughness from the MOLA beam broadening effect. Then Smith 
et al., (2001) analysed the local roughness using MOLA beam 
broadening at the global scale and finally Neumann et al., 
(2003) extracted the local roughness at !4 of a degree resolution 
from improved pulse characteristics. However, all of their 
studies have problems to apply surface slbpe correction. For 
example, Garvin et al., (1999) used the along track slope 
extracted from the footprint-to-footprint height differences and 
Neumann et al. (2003) employed a 1/64° gridded MOLA DTM. 
In both cases, these very crude slopes cannot effectively remove 
slope effects within the laser beam footprint as the footprint-to- 
footprint distance (320m) and the projected footprint (~150m) is 
too coarse to characterise the with in-footprint slope distribution. 
A new solution to this problem is proposed here to employ 
multi-scale stereo DTMs. In this research, we show how 
effectively stereo DTMs extracted from different data sources 
can be combined with MOLA laser beam properties to calculate 
a more reliable and precise local surface roughness. 
2. ALGORITHMS 
2.1 High resolution stereo DTM extraction 
Considering the MOLA footprint size (150m) and across track 
distance (» 1km near the equator), it is clear that the horizontal 
resolution of MOLA data is not enough to provide details on 
key geomophological features. Therefore, stereo image analysis 
is still important even though it has relatively poor vertical 
accuracy (=TFoV). Stereo image coverage of the Martian 
surface has dramatically increased since the successful orbital 
insertion of HRSC onboard ESA’s Mars Express. Recently, the 
stereo image coverage of HRSC (<20m) is up to 45.7% of 
Martian surface (Jaumann, 2008). One of the biggest attractions 
of HRSC stereo imagery is that it’s positional accuracy is well 
co-registered with MOLA. As a result, a HRSC pixel has an 
inherent planimetric accuracy of 25-40m and a 3D space 
intersection of up to 6-8m vertical registration with MOLA, if 
the improved exterior and interior orientation from bundle 
adjustment is employed (Speigel, 2007). This means that the 
HRSC image and derived DTM can be employed as the base 
data for other optical image’s geometric calibration. A barrier 
to this idea is the inherent difficulty in co-registration due to the 
resolution differences between HRSC and other high resolution 
(>few metre pixel resolution) imagery such as CTX or MOC- 
NA and very high resolution («lm resolution) imagery such as 
HiRlSE. Our solution, to address this problem, has been to 
employ a hierarchical co-registration between different 
resolution imagery or re-sampled images. For example, in one 
area, which includes coverage by HRSC, CTX and HiRISE 
stereo images, the HRSC intersection points and orthorectified 
images (ORIs) provide the first geodetic control information for 
the CTX imagery because the registration between CTX and 
HRSC is not so difficult (a factor of ~2-3 difference in 
* Corresponding author