11 
middle-resolution 
gure6. They were 
ry. Figure6 shows 
l the commercial 
Island was greater 
reas was less than 
ire 6 were quite 
the density urban 
pped. But the ISP 
vere fragmentized 
distortion in roll 
■ the use of high- 
i that the SPOT- 
low to middle ISP 
has been reported 
>P overestimation 
> similar spectral 
;etation and urban 
. However, this 
Stic use of SPOT 
InSAR_3; 
attributed to the 
n interferometric 
sh the two land- 
id in detail was 
demonstrated by the ISP results of the typical area selected in 
study area that located near the West Kowloon Waterfront 
Promenade and covered with bared soils and sparse grass. The 
spectral confusion resulted in a dramatically ISP overestimation 
in the mixed area using just SPOT imagery and ISP values were 
(c) 
SPOT-derived ISP images represented a higher performance 
(AE 15.51%, RE 0.58 and R2 0.71). As was expected, the 
addition of the InSAR feature information improved the 
accuracy of ISP estimate. Compared to these statistical 
indicators corresponding to SPOT-derived ISP image, AE value 
(d) 
Figure 6 Comparison of ISP estimated results at reclamation region in Yau Ma Tei (a) InSAR_3; 
(b) SPOT 4; (c)SPOT_4+InSAR_3 ; (d)SPOT imagery 
up to more than 78% (Figure 6(b)), but less than 60% after 
considering InSAR feature information (Figure 6 (c)). 
Dataset 
Statistical iiulictor 
Average error 
(AE) (%) 
Relative 
Error 
m 
Pearson 
coefficient 
m 
InSAR 3 
19.27 
0.69 
0.61 
SPOT4 
15.51 
0.58 
0.71 
InSAR J +SPOT4 
12.93 
0.51 
0.77 
Table 2 accuracy assessment of ISP estimation results 
In this study, a quantitative validation analysis was conducted 
for the accuracy assessment of ISP estimation results. 
Approximately 7252 samples were selected randomly 
independently from the reference data and ISP images that 
derived by using middle-resolution dataset. It should be noted 
that these samples did not include those used for the ISP 
prediction model development. The statistical indicators, 
Average Error (AE), Relative Error (RE) and Pearson 
coefficient (R2), were calculated for the three ISP images 
derived from the three groups of test datasets. The results of 
accuracy assessment were summarized in Table 2. The results 
indicated that InSAR-derived ISP estimations were comparable 
to those derived from SPOT multi-spectral images, although the 
decreased by about 3 percent to 12.93% and R2 value increased 
from 0.71 to 0.77. In addition, the difference between the 
estimated ISP vales and reference data were calculated and 
demonstrated by the error histogram. Compared with the error 
histograms of InSAR-derived and SPOT-derived ISP estimate, 
the one of fused data represented an apparent peak near the zero 
value and more than 80% of the estimate error fell near the peak. 
5. CONCLUSIONS 
This study focused on urban impervious surface mapping by 
using the combined multi-sources and multi-sensors remote 
sensing dataset. Because of information complementarities 
between optical and radar imagery, we have demonstrated that 
the synergistic use of this two data sources has the potential to 
improve the ISP estimate performance, especially in some areas 
coved with bared soils or sparse vegetation. 
In this paper, a case- study was conducted for impervious 
surface mapping in Hong Kong by applying the CART-based 
ISP estimate approach to the SPOT 5 HRG imagery and the 
ERS-1/2 SLC SAR data. In particular, three InSAR features, 
coherence, average amplitude and amplitude ratio, were 
extracted from the SAR data for the investigate objective. 
Validated by reference data derived from the high-resolution 
CIR aerial photographs, our results show that the addition of