2. STUDY AREA AND DATA SET 
2.1 Description of study area 
The study area covers about 100 km 2 along both sides of the 
Victoria Harbor, Hong Kong and is located within latitude 
22.267°N~22.343°N, longitude 114.122°E~114.240°E (Figure 
1). It includes Kowloon and Hong Kong Island, which are both 
well-known central business districts (CBD). The study area 
consists of dense urban area, bared filed, mixed vegetation and 
open water. The urban areas along Victoria Harbour are even, 
but the bottom and upper left parts in the study area are rolling 
piedmont terrain covered with dense and mixed vegetation. 
SAR 
sensor 
Orbit 
number 
Time 
Time 
interval 
(day) 
Spatial 
perp. 
baseline(m 
) 
ERS-2 
18795 
1998-11-24 
35 
-13 
ERS-2 
19296 
1998-12-29 
Table 1 the basic parameters of ERS SAR interferometric 
pairs in this study 
2.2 Datasets and pre-processing 
In this study, a pair of ERS-2 C-Band Single Look Complex 
(SLC) SAR data were chosen for generating feature images of 
interferometric SAR signature (Table 1). This InSAR pair was 
characterized with a short spatial perpendicular baseline (-13 m) 
and a relatively long-term interval (35 days). Unlike the 
Tandem coherence derived from ERS-1/2 images and used in 
standard approaches, a long-term coherence image computed 
from the considered InSAR pair can improve capacity to 
distinguish man-made features with other urban land- 
covers(Bruzzone, Marconcini et al. 2004; Liao 2007). This is 
due to the fact that coherence of man-made features remains 
high even between image pairs separated by a long time interval 
(months to years), however, most of natural land surfaces (e.g. 
farmland and forest field, etc. ) are significantly influenced by 
temporal decorrelation and lose coherence within a few days. In 
addition, average amplitude and amplitude ratio were obtained 
from the same InSAR pair and an RGB combination of the 
three resulted feature images was shown in Figure 2(a). 
In addition, a SPOT 5 HRG multi-spectral imagery acquired in 
November, 2002 was used in this study. The SPOT 5 image had 
three bands in visible and NIR bands (0.50-0.59 um, 0.61-0.68 
um, 0.79-0.89 um) with 10 meter spatial resolution and one in 
SWIR band (1.58 - 1.75 um) with 20 meter resolution (Figure 2 
(b)). In particular, a Colour-Infrared (CIR) aerial photography 
acquired in November, 2000 was utilized to derive training/test 
data for an ISP CART-based prediction model based and 
validation data for accuracy assessment of ISP mapping. The 
CIR aerial photograph was scanned color balanced into a digital 
format with 33-cm nominal spatial resolution. The SPOT 5 data 
and the CIR aerial photograph were geometrically corrected and 
ortho-rectified to Hong Kong 1980 Grid Map Projection using a 
1:5, 000 Digital Elevation Model (DEM) data of Hong Kong. 
For merging multi-sensors remote sensing data, several pre 
processing steps are necessary to establish a more direct 
relationship between image signals and physical phenomena. In 
this study, some standard InSAR pre-processing steps were 
applied to the two ERS-2 SLC SAR data above, mainly 
including radiometric calibration, coregistration (with 0.1 pixel 
accuracy), a temporal SAR speckle filtering applying to 
amplitude images, and deviation of InSAR features (coherence, 
average amplitude and amplitude ratio). In order to geocode the 
InSAR products, the satellite precision orbit data and the 
mentioned large-scale DEM data were utilized to generate a 
geocoding lookup table for the conversion between DEM 
projection coordinate and SAR imaging geometry. The 
geocoding process was completed with sup-pixel accuracy by 
using GAMMA DIFF&GEO module (www.gamma-rs.ch). The 
geocoded InSAR products, coherence image, average amplitude 
image and amplitude ratio image, finally were resampled at 10 
m spatial resolution as same as the SPOT data. 
3. METHODOLOGY 
In this case study, a CART-based approach developed by Yang 
(2003) was adopted to estimate sub-pixel percentage of 
impervious surface with the synergistic use of SPOT multi- 
spectral imagery and InSAR products. This approach used high- 
resolution imagery as a source of training data for representing 
urban land-cover heterogeneity, and medium-resolution 
imagery to extrapolate imperviousness over large spatial areas. 
Figure 2 The medium-resolution remote sensing data set in the study area (a)ERS-2 InSAR data (Red-Coherence, Green- 
Interferogram amplitude, Blue-Amplitude ratio); (b) SPOT5 HRG data(Red-Band4, Green-Band 1, Blue-Band3,)