462
July 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
for all points, water surface points and non-water points are
shown in Figure 10a. We then generated a 1 m DEM (Figure
10b) from the water surface points. Several noisy spots are
visible in the north, since they are shallow waterbed and
close to the shore, leading to an imperfect filtering. The mean
height of the filtered water surface points is 0.13 m above
the vertical datum and the standard deviation is 0.12 m. If
we compare this with our previous observation on the pond,
we are able to conclude that the height sensitivity of SPL is
in the range of 0.09 – 0.12 m. It should be noted that the
pond water should be quite still, while the river water surface
in the Bathy dataset connects to the Chesapeake Bay, with
possible minor water waves; hence a slightly larger height
variation of 0.12 m is observed.
C
onclusion
Single photon laser has opened an exciting opportunity for
lidar topographic mapping. It can simultaneously transmit
hundreds of beams and record single photon returns. This
allows data collection to be carried out in a much higher
altitude and a much higher rate than the conventional linear
lidar. The SPL datasets used in this paper were collected
at an altitude of 2,286 m (7,500 ft). The SPL was able to
generate a spatial resolution over 10 times denser than what
Figure 9. Google Street View (left), SPL points (middle) and height histogram (right) of the selected water pond area (the small red box).
a conventional lidar can possibly achieve even at a much
lower altitude (a few hundred meters), thus causing a smaller
swath and lower productivity for the linear systems.
The high density of single photon lidar poses a new challenge
to the ground filtering process for digital elevation model
generation. It is shown majority of the returns in forest are
from tree canopy, while only a small portion of the returns
are actually from the ground. Taking the forest dataset as an
example, the filtered ground returns under canopy are only
as 3 – 7 times more as the linear lidar. As a result, we may
not optimistically expect that terrain generation will benefit
from the new systems as much as tree canopy inventory
would, though this statement depends on the canopy density
in general. Because of higher
density over canopy area, local
canopy tends to be treated as flat
and therefore wrongly classified
as ground, leading to visible
artifacts in the resultant digital
elevation model. This study has
demonstrated a critical need to
revisit existing popular filtering
methods and explore new
ones to accommodate recent
developments in lidar systems.
Similarly, the ‘penetration’
capability of such new systems
in leaf-on seasons is worthwhile
of further investigation.
The intrinsic uncertainty of
elevation measurements for
the SPL system is another
factor that we looked into. This
was an independent effort in additional to using laboratory
calibration or ground control. Instead, we intended to use
natural features that have known geometric properties. The
water pond and the river water examples demonstrated that
the SPL system has an intrinsic elevation uncertainty of 0.09
- 0.12 m. Besides, this was supported by the fact that small,
low roof objects at a height of 0.20 m could be easily detected.
Finally, although the capability of acquiring highly dense
Figure 10. The height distribution of the SPL points (a) and the 1 m water surface shaded DEM (b) for the
selected river water area in the Bathy dataset.
(a)
(b)
