accuracy of upscaling, especially when the use of the upscaled
map is concerned about class area information.
Despite the improved performance of
FMC
as described
above, it does not effectively maintain a consistent landscape
pattern between the base map and the upscaling. Instead, the
performance of
FMC
as assessed by
CL
is almost the same as
MRB
’s performance. Figure 6 shows that upscaling resulted
in increasing the
CL
for
FMC
. This result is very similar with
the
CL
of
MRB
for each coarse map for each study area. These
results demonstrate that if the upscaled maps were to be used
for an application that was only concerned about landscape
pattern, then either
FMC
or
MRB
could be applied as there is no
improvement using
FMC
here.
Relationship between Upscale, Heterogeneity, and Performance of FMC
Previous studies (e.g., Moody and Woodcock, 1995) showed
that landscape pattern and scale impacted the performance of
upscaling. To analyze the relationship between upscale, het-
erogeneity and the performance of
FMC
, we conducted these
experiments on different areas with various heterogeneities.
OC
,
PE
, and
CL
, derived from
FMC
, showed similar tendencies
with
MRB
when the spatial resolution becomes coarser (Figures
3, 5, and 6). These similar tendencies occur because
CMP
is
estimated from the results of
MRB
. Hence, the results of
MRB
impacted the results of
FMC
. With increased upscaling,
OC
is
reduced, while
PE
and
CL
are increased. This is due to the mix
pixels in the coarse-resolution maps. Although
FMC
is impacted
by the upscaling (resolution of the coarse maps), it can reduce
the
PE
of the map classes and obtained higher
OC
in each study
area as the section “comparison of
FMC
and
MRB
” illustrated.
Figure 6.
CL
values when conducting upscaling: (a) is the results of FMC, (b) is the results of
MRB
.
Figure 7. The relative error of land cover change detection: (a) and (b) are the relative error derived from
MRB
and
FMC
,
respectively, for
ASD
1870; (c) and (d) are the relative error derived from
MRB
and
FMC
, respectively, for
ASD
4550.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2018
95
