The International Archives of the Photogrammetrv. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1034
4.4 Modeling Algorithm
The integrated approach, as shown in Fig. 2, enhances Facets
Stereo Vision (chapter 3.1) with photoclinometry (4.1) and in
tegrates the derivation of surface reflectance (2 and 4.2) and
atmospheric properties (4.3). Depending on the actual investi
gation area, desired quality, and resolution, it might become
useful or even necessary to constrain or omit individual parame
ters. In general, surface reflectance and atmospheric models are
interchangeable and can be replaced. E.g., Lambert’s Model
may be used for very simple surface parameterization.
In practice, the complexity of the algorithm should be increased
along with the increase of resolution. Photoclinometry is not ne
cessary to obtain a coarse DTM, and meaningful photometric
parameters can only be obtained if geometric details of the sur
face are modeled. In this context, it may still be useful to regu
larize the least squares adjustment as described in 3.2.
The integrated algorithm can be regarded as a complete surface
modeling approach for Mars that makes use of all radiometric
and geometric information contained in HRSC data. Because it
is based on the entire image content - literally every pixel that
is obtained in the investigation area it can be applied to small
regions only, typically below 100x100 DTM facets.
5. APPLICATION
The integrated surface modeling algorithm has been implemen
ted in MATLAB. It has been applied to several regions of Mars
using HRSC imagery of different resolution obtained under va
rious illumination and viewing geometries. In the following,
two representative examples are discussed: a 25x25 km area of
hills and mesas, which is located in southern Gusev, as well as a
2.8x2.8 km area containing two small impact craters.
5.1 Investigation Areas and HRSC Data
The described surface modeling approach can make use of pan
chromatic HRSC bands: stereo (S1/S2), photometry (P1/P2),
and nadir (ND). (See chapter 6 for remarks on color.)
The Gusev area has been imaged multiple times (cp. Jehl et al.,
2008), e.g., during orbit 648 from very high altitudes (Fig. 3;
Table 1). Although this inevitably leads to rather low ground re
solutions, the major advantage is large stereo angles due to orbit
curvature - up to 31.2° in comparison to nominal HRSC angles
of 18.9°. Local viewing angles range from 0° to 60°. The illu
mination is generally low; it varies locally between 51° and 90°,
casting some shadows. Such geometry potentially allows for the
derivation of reliable atmospheric optical depth and surface re
flectance. Therefore, radiometric results are discussed in-depth.
Table 1: Overview of HRSC bands and viewing geometries
Area:
Gusev Southern Hills
Two Craters
Orbit:
648
894
Band
Ground
Viewing
Ground
Viewing
Resolution
Angle
Resolution
Angle
SI
281 m/pixel
31.2°
36 m/pixel
21.6°
PI
289 m/pixel
21.4°
-
-
ND
165 m/pixel
2.1°
16 m/pixel
o
o
P2
448 m/pixel
29.5°
-
-
S2
n/a
n/a
32 m/pixel
21.7°
Fig. 3: Investigation area (red rectangle) within Gusev crater.
The second investigation area - two small craters in the Nanedi
Valles region, imaged during orbit 894 (Fig. 6) - has been cho
sen to demonstrate the approach’s strength in modeling fine
surface details. With special emphasis on high resolution geo
metry, HRSC photometry bands with ground resolutions around
65 m/pixel (due to pixel binning) have not been used (Table 1).
5.2 Gusev Southern Highlands
For the Gusev area, surface radiometry and geometry as well as
atmospheric properties have been successfully derived. The re
sulting DTM features a post spacing of 500 m (Fig. 4).
17S.J
Fig. 4: Gusev hills DTM as color-coded perspective view.
The optical depth has been determined with x = 1.55 ± 0.06,
meaning that only 21% of the reflected light directly reached
the nadir-looking detector. This number appears low but, how
ever, it is well within the range that Hoekzema et al. (2006)
derived for regions in and around Gusev crater.
Based on the integrated adjustment results, one Hapke Model
has been calculated for the entire investigation area, excluding
shadows. The initial approach involved all four parameters of
equation (1); it lead to a large c value that was also highly cor
related with the other phase function parameter, b (Table 2).