IEEE Geoscience and Remote Sensing Magazine - June 2022 - 133

DISCUSSION
This workflow is presented as a cookbook for environments
that have steep slopes. The climatic environments used here
to apply our method share contrasted reliefs but with different
distributions of slopes because of various geological,
morphodynamical, and climatic contexts. In addition,
urbanization and land use have evolved at different paces
during the considered period. However, in both cases, our
workflow enables detecting and quantifying morphodynamical
and landscape evolution through a decennial time
period. However, a few issues need to be discussed. Some of
them, related to the nature of the raw data, are unsolvable,
while others can be partially resolved.
Aerial image sensors operate in visible spectra and are sensitive
to climatic conditions. Therefore, aerial photos often
show partial to significant cloud coverage. This point is critical
for tropical areas as well as high-relief regions, causing two
permanent limitations: 1) the clouds are not optically transparent,
and 2) in cases of partial coverage, cloud shadows on
the ground limit DSM reconstruction. This is even more critical
under windy conditions, when clouds are moving from
one image to another during acquisitions. At this stage, the
best practice would be to get rid of such images.
Another issue that can be encountered is related to the
topography of investigated scenes since steep landscapes will
lead to shadowed hillslopes. In the shadow, the radiometry
range drops by a factor two or three compared to other areas
and makes dense matching difficult. Attempts to increase
the brightness in such areas make better pixel matching possible.
Nonetheless, topographic reconstruction of these areas
remains less accurate than for illuminated ones. We show, on
the other side of the spectra, that very bright areas, such as
glaciers, are well reconstructed because associated pixels are
not saturated on the image data sets [e.g., Figure 7(a)].
Finally, since our workflow enables the production of
DSMs with metric resolution, and because biomass growth
can be fast in different environments, the method can help
assess canopy height evolution and hence biomass dynamics
through decades in various settings. This is a critical
aspect of land use science and ecology research, and we
suggest that a workflow using archived aerial images can
be used for other purposes, not only for quantifying geomorphological
processes.
CONCLUSION
A complete workflow for deriving time series DSMs from
scanned archived image has been presented. The global, automatic
method accounts for various kinds of fiducial mark patterns
in various campaigns. It requires minimal use of GCPs,
which dramatically reduces manual editing and its associated
limitations. The resulting block association and compensation
enables us to obtain well-aligned canvases. Algorithms
are provided, and users can implement them using their
software of choice. Consequently, the workflow can obtain
time series of DSMs through decennial time periods
with a submetric sampling. An additional outcome
JUNE 2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
is the provision of accurate orthoimage time series. This
is especially of interest for time evolution assessment for
landscapes experiencing climate change [27].
When applying the workflow to steep terrains, surface
processes with annual rates of change on the order of 1 m
are quantifiable. This facilitates studying landscape dynamics
with sediment flux and mass balance assessment on a decennial
timescale, which is critical in geomorphology. The DSMs
resulting from the method presented here are therefore useful
for mass movement assessment, such as landslides, glaciers,
and the like. Finally, as we derive DSMs, canopy height
can be also calculated, and hence so can biomass evolution.
Consequently, the resulting products of such a pipeline offer
substantial new insights into critical zones.
ACKNOWLEDGMENTS
We thank our undergrad students that contribute to the hard
work of collecting data, identifying GCPs, and contributing
to initial results: David Talec, Adrien Sy, Marie Lirzin, Meropi
Kostopoulou, and Coline Hopquin. We also thank Luc Girot,
Matthieu Gougeon, and Laurent Michon for their constructive
inputs. We thank the anonymous reviewers and associate
editor for their helpful comments. This research benefits from
IGN aerial surveys available from https://www.geoportail.
gouv.fr. Numerical computations were partly performed on
the S-CAPAD/DANTE platform, IPGP, France. The authors
acknowledge financial support by LabEx UnivEarthS (grants
ANR-10-LABX-0023 and ANR-18-IDEX-0001).
2
A
1.5
1
0.5
A′
(a)
2
1
-1
-2
-3
1,000
2,000
3,000
Profile A-A′ Distance (m)
(b)
FIGURE 9. The detection of scanner band defaults when comparing
stable areas between 1978 and 1984. (a) Difference between the
DSM of 1978 and 1984 of the north part of Langevin River next to the
Dolomieux rim, showing unexpected parallel bands. Such variation of
elevation on a stable area actually corresponds to the scanner band
defaults. (b) Cross-section of the difference along profile A-A'
133
A
A′
-0.5
-1
-1.5
-2
Difference (m)
Elevation Change (m)
Between 1978 and 1984
https://www.geoportailouv.fr https://www.geoportailouv.fr
IEEE Geoscience and Remote Sensing Magazine - June 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2022
