PE&RS February 2018 Full - page 96

Additionally, if the landscape pattern is more heteroge-
neous, the OC was reduced much more when the base map
was upscaled. For example, Table 2 shows that the ASD 4580
is the most homogenous of all the study areas, and ASD1830
is the most heterogeneous area. The OC in ASD 4580 was
reduced about 1.89 percent (Figure 3f) and about 8.69 percent
in ASD1830 (Figure 3c). These results demonstrate that the
degree of heterogeneity impacts the performance of FMC.
Additionally, considering the results discussed above, we
recommend that FMC should be employed to conduct the up-
scaling, despite that it is impacted by the landscape pattern.
Since both FMC and MRB are equally influenced by landscape
pattern, the advantages of FMC make it the approach of choice
when upscaling.
Influence of Class Proportion on FMC
Several researches (e.g., Raj et al., 2013) reported that MRB
increased the proportion of dominant classes and decreased
the proportion of non-dominant classes. In this paper, non-
crop is the dominant class, and others are the non-dominant
classes. Our analysis of MRB shows consistent results with
these previous studies. Since the CMP is calculated from MRB,
it is necessary to detect whether the class proportion has an
influence on FMC’s performance. Additionally, FMC shows that
the proportion of the dominant class increased when imple-
menting upscaling, while the proportion of non-dominant
class decreased (Figure 8).
Figure 8 showed that FMC reduced the influence of class
proportion on upscaling to some degree. For example, in
ASD1830, the proportion of the dominant class (i.e., non-
crop) increased about 6.33 percent when implementing MRB,
while only about 4.13 percent when implementing FMC. The
proportion of corn (non-dominant class) was reduced about
3.05 percent when implementing MRB in ASD1830, while only
about 2.03 percent when implementing FMC. These results
demonstrate that FMC can mitigate the impact of proportion
of one class on the proportion of this class in the upscaled,
coarser map.
Influence of Exponent Values on FMC
As illustrated in the Methodology Section, the exponent val-
ues determine the importance of the information (i.e., CLP and
CMP) to calculate the fused probabilities for determination of
the map class type of the output coarse-pixels. To assess the
effects of the exponent values on the upscaled maps produced
by FMC, thirteen pairs of exponent values for τCLP =0, 0.01,
Figure 8. Proportion of each class in coarse maps produced by FMC and MRB. The upper panel is for relatively heterogeneous
study areas. The lower panel is for relatively homogeneous study areas.
96 February 2018 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING