Generally, the accuracy of our work is persuasive. The overall
RMSE
is 10.98% and overall
MAE
is 7.55% in pixel-based ex-
periment, while in parcel-based experiment, the overall
RMSE
is 10.90% and overall
MAE
is 7.21%. Compared with physical
features (initial impervious fractions), our proposed method
could well avoid the influence of tree crowns/shadows with a
promising accuracy.
Discussion
In this study, we explored the potential of integrating remote
sensing data and social knowledge in urban impervious
surface estimation. Remote sensing is an efficient approach
to map the biophysical characteristics of the Earth’s surface.
However, its performance strongly depends on the obser-
vation conditions. This is a critical issue in urbanization
monitoring, for instance: (1) rooftops covered by vegetation
canopies may be characterized as pervious in remote sensing
imagery, and (2) impervious fractions of small parks hiding
in the residential areas are easily influenced by the reflected
lights of buildings. These could largely result in improper
impervious fractions. While combined with social knowledge,
these effects could be well avoided. Figure 9 gives an example
of Yuexiu District, which is the old city town of Guangzhou.
This area is mostly covered by big tree crowns, which leads
to the low-albedo (varying from 0.3 to 0.5) in impervious
fractions generated from remote sensing imagery (Figure 9b).
Figure 7. (a) impervious fractions estimated from physical features and
POI
features, and (b) estimated IS map integrating
physical features,
POI
features and road distribution map; (c) and (d) are two zoomed-in view of (a) and (b), respectively.
Fitting curves of (e) reference data versus physical features (R-square = 0.6932), (f) reference data versus IS result integrating
physical and
POI
features (R-square = 0.8345), and (g) reference data versus impervious fractions integrating physical and all
social features (R-square=0.8452), support the effectiveness of our method.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2018
777
