PE&RS July 2019 - page 482

Archinal et al. (2015) used USGS ISIS to generate controlled po-
lar mosaics from LROC NAC images. The residuals were within
1 pixel with a maximal error of ~7 pixels, and the horizontal
accuracy of the mosaics was within 100 m with respect to the
Lunar Orbiter Laser Altimeter (LOLA) data (Lee et al. 2012).
Wu et al. (2014) developed a combined block adjustment
method to process Chang’e-2 stereo imagery and LOLA data of
the Chang’e-3 landing area. The resultant DTM has an aver-
age elevation difference of 15.19 m and a standard deviation
of 17.08 m compared to that of the LOLA DEM. Di et al. (2014)
developed a self-calibration bundle adjustment method for
Chang’e-2 stereo images to reduce the geopositioning inconsis-
tencies among the images of adjacent orbits from up to 20 pix-
els to a subpixel level. Kokhanov et al. (2017) used a self-devel-
oped photogrammetric method to obtain cartographic products
of potential landing sites for the “Luna-25” mission from 11
LROC NAC stereo pairs. The bundle adjustment results had pre-
cisions of 0.3 m and 1.0 m in the planar direction and 3.5 m in
the vertical direction. Haase et al. (2012, 2014) produced a DTM
and ortho-mosaic for the Apollo-17 landing site using LROC NAC
stereo images. The LROC -adapted German Aerospace Center
photogrammetric processing chain was used for stereo image
processing, and the resultant DTM had a standard deviation of
40 cm in elevational difference with the LOLA profiles.
As previously described, existing lunar mapping research
is generally based on stereo images. However, the LROC NACs
can only acquire stereo images from adjacent orbits using
off-nadir slew in limited interested locations, and most of
the time, the captured NAC images have nadir orientation.
There are only a few NAC stereo pairs in the Chang’e-5 plan-
ning landing area because the nadir NAC images cannot be
used to generate a high-resolution DOM mosaic using stereo
photogrammetric methods. Furthermore, most of the existing
studies have been focused on photogrammetric processing in-
volving several or tens of images. For large-area mapping, the
geometric and radiometric inconsistencies are more severe
and complicated. Therefore, it is necessary to develop a more
effective method to produce high-precision DOM mosaics for
large areas using nonstereo NAC images and a lunar control
source of limited resolution.
This research aimed to develop a systematic and effec-
tive method for generating a seamless DOM forthe planned
Chang’e-5 landing area. In this work, we used a two-step strat-
egy. First, the study area was divided into several overlapped
subareas, and a planar block adjustment with control points
was applied to each subarea to lower the geometric deviations
among the NAC images to the subpixel level at the same time
of tying the NAC images to the reference SLDEM2015. Then a
seamless DOM mosaic of each subarea was generated. Second,
the thin plate spline (TPS) model (Wahba 1990) was applied
to the subarea DOMs to remove the positional inconsistencies
between the adjacent subarea DOMs and guarantee seamless
DOM mosaicking throughout the whole research area. Using
the proposed two-step method, a controlled seamless mosaic
of the Chang’e-5 landing area was created with high geometric
precision and with a resolution of 1.5 m.
Data
The planned landing area of Chang’e-5 mission was chosen
to verify the proposed large-area DOM generation method. It
locates near Mons Rümker within Oceanus Procellarum and
covers an area of approximately 20° longitude × 4° latitude, or
approximately 413.8 km × 121.4 km (Di et al. 2018). LROC NAC
images were the data source used in this research. SLDEM2015,
a combined product of LOLA laser altimetry and DEMs generat-
ed from the Japanese Selenological and Engineering Explorer
(SELENE) terrain camera images (Barker et al. 2016), was used
as reference DEM for ortho-rectification. Figure 1 shows the
planned landing area of Chang’e-5 on the LROC WAC mosaic.
Figure 1. Planned landing area of Chang’e-5 marked as a
rectangle on the LROC WAC mosaic.
LROC NAC Images
The LROC is a system of three cameras onboard the Lunar
Reconnaissance Orbiter (LRO) that captures high-resolution
monochromatic images and moderate-resolution multispectral
images of the lunar surface. It consists of two NACs that are
designed to provide 0.5–2.0 m/pixel monochromatic narrow-
angle line scan images and a WAC that provides images at a
scale of 100 m/pixel in seven color bands over a 60-km swath
(Robinson et al. 2010).
NAC Experimental Data Record images were downloaded
from the NASA Planetary Data System (PDS) and preprocessed
using the USGS ISIS software. SPICE kernels (NAIF 2014) were
attached to each image using the “spiceinit” command, and
radiometric corrections and removal of echo effects were
realized by the “lronaccal” command and the “lronacecho”
command, respectively (PDS 2014; Henriksen et al. 2016).
Until December 2017, the planned Chang’e-5 landing area is
covered by about 2299 LROC NAC images. Considering the illu-
mination conditions, most of the chosen images have similar
solar azimuth angles that are higher than 180° (afternoon im-
ages) and incidence angles between 40° and 80°. The planned
landing area could not be completely covered with afternoon
images, so small gaps were filled with one or two morning im-
ages. A total of 765 NAC images were involved in this research
with a ground sample distance of mainly 1.5 m.
Control Source
SLDEM2015 was used as the control source for providing three-
dimensional control points in the block adjustment stage as
well as providing topographic correction during DOM gen-
eration. This product is a lunar shape model generated by a
combination of LOLA and SELENE data. This includes 43 200
stereo-derived DEMs from SELENE Terrain Camera images and
4.5 billion surface heights from LOLA (Barker et al. 2016). The
resultant near-global lunar DEM has an effective resolution
of approximately 60 m at the equator and a typical vertical
accuracy of approximately 3–4 m. In addition, the LROC WAC
mosaic (NASA 2011; Wagner et al. 2015) was used as a refer-
ence for grayscale balancing in the image mosaicking process.
Method for Large-Area Controlled DOM Generation
In this research, a two-stage method was used to generate a
large-area controlled seamless DOM. Figure 2 is the flowchart
showing the generation process of large-area seamless DOM. To
guarantee both the processing efficiency and mapping preci-
sion, the large landing area was partitioned into 10 subareas
and processed in parallel. Dividing the whole mapping area
into some subareas is a common strategy when dealing with
482 July 2019 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING