PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2016
459
The difference between the two 1 m DEMs as well as its
histogram are shown in Figure 5. The majority of the area in
Figure 5a appears yellow, suggesting the elevation difference
ranges from -0.5 to 0.5 m. The histogram in Figure 5b further
reveals the peak difference occurs at a mean of 0.21 m, i.e.,
the SPL DEM is in general higher by this amount than the
reference DEM. Comparing the two shaded reliefs in Figure 4
and Figure 3 for the forest area, we notice that many small
fractal, rugged terrain features distributed rather evenly over
the study area. With reference to the image shown in Figure 2,
it is believed they are small bushes or trees remained in the
filtering results. The difference is likely caused by imperfect
filtering in forested areas where highly dense multi-stop data
is processed. We therefore see a challenging data processing
task for new, advanced lidar developments, eg., SPL.
We note that the ground point density of the SPL data after
filtering is as high as 3.6 points per sq. meter and the average
point spacing reaches 0.52 meters. Such dense points make
it theoretically possible to generate DEMs with a resolution
higher than 1 meter, except areas under dense trees. To
Figure 4. Point cloud (left), filtered ground points (middle) and the 1 m shaded DEM (right) from the linear lidar over the Forest area.
explore the potential of the SPL data, Figure 6 provides the
DEMs with 0.5 m, 0.25 m and 0.1 m resolutions. We can see
that the DEM with higher resolution presents less noise than
the lower ones. For example, the height difference along the
forest roads becomes less obvious as the resolution increases.
This can also be observed if we compare the SPL DEMs in
Figure 6 with the 1 m linear lidar DEM in Figure 3, where the
most significant differences can be found along the roads or
trails under canopy.
Finally, we would like to look into the distribution of the point
density. After filtering, the linear lidar dataset has 37,512
(31.51%) ground points left, corresponding to a point density
0.26 points per sq. meter at a ground spacing 1.96 meters.
When compared to the SPL Forest dataset in Figure 3 and
Table 3, the raw SPL data is 12.2 times (10/0.82) denser
as the linear lidar data, but its ground point density after
filtering is as 13.8 times (3.6/0.26) higher. However, it should
be noted that the distribution of remaining ground points is
not even, as depicted by the point density in Figure 7. For
comparative purpose, the figure also depicts the distribution
Figure 5. Difference of the SPL and linear lidar DEM (in meters) and its histogram of the Forest dataset.
(a)
(b)
