The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
148 
No. 
Date of 
Acquisition 
Sensor 
Track 
1 
1992.09.11 
ERSI 
390 
2 
1992.09.30 
ERSI 
161 
3 
1993.01.29 
ERSI 
390 
4 
1993.02.17 
ERSI 
161 
5 
1993.03.05 
ERSI 
390 
6 
1993.04.09 
ERSI 
390 
7 
1996.01.07 
ERSI 
390 
8 
1996.07.01 
ERS2 
390 
9 
1998.05.16 
ERS2 
161 
10 
1998.06.20 
ERS2 
161 
11 
2004.01.10 
Envisat 
2161 
12 
2005.06.18 
Envisat 
2161 
13 
2005.08.27 
Envisat 
2161 
14 
2006.04.29 
Envisat 
2161 
15 
2007.03.10 
Envisat 
2161 
Table 1. SAR data selected for the research 
Pair 
No. 
Master 
Slave 
B x (m) 
Temporal 
Baseline 
Sensor 
1 
92.09.11 
93.04.09 
30.4 
210 
ERS 
2 
92.09.30 
93.02.17 
124.1 
140 
ERS 
3 
96.01.07 
96.07.01 
-37.2 
175 
ERS 
4 
04.01.10 
05.08.27 
119.5 
595 
Envisat 
5 
05.06.18 
06.04.29 
27.3 
315 
Envisat 
6 
06.04.29 
07.03.10 
175.7 
315 
Envisat 
Table 2. InSAR pairs formed 
Two-pass differential interferometric synthetic aperture radar 
(DlnSAR) method is applied with the 3 arc second SRTM 
DEM used for removing the topographic phase. The maximum 
elevation difference in the study area is within 100 m so that 
the effects of the errors in the DEM are negligible for the short 
normal baselines of the InSAR pairs. Precise ERS and Envisat 
orbit data from the Delft University are applied to mitigate the 
satellite orbit errors (Scharroo et al., 1998). The adaptive 
filtering algorithm (Li et al., 2006) is applied to reduce the 
phase noise in the interferograms. The minimum cost flow 
method is applied (Chen and Zebker, 2001) to unwrap the 
interferograms. The results are then geocoded and projected 
into the vertical direction considering the fact that the main 
deformation in the area is subsidence. Finally we select a 
square window of 20*20 pixels in a stable area to use their 
median value as the stable point value and then adjust the 
results from each InSAR pair to obtain the absolute subsidence 
values between the image acquisitions. 
Seventeen GPS stations were set up in 2005 and two 
measurement campaigns were carried out on June 2005 and 
June 2006. The locations of the GPS stations are shown in 
some of the plots in Figure 2. To compare the InSAR and GPS 
measurement results, the subsidence values at the discrete GPS 
stations are extracted from the subsidence map of InSAR Pair 5 
(050618-060429) and converted into annual subsidence rates. 
The correlation between the InSAR and GPS measurements at 
the GPS stations is then calculated. In addition, to depict the 
evolution of land subsidence over 1992 - 2007 and to detect the 
differential subsidence rates across the ground fissures, two 
profiles (see Figure 2) of subsidence, AA’ that starts from the 
stable point (Point A) and extends to the CAF and BB’ that is 
along one of the main streets in the city are generated and are 
shown in Figure 3. 
3. RESULTS AND DISCUSSIONS 
3.1 Results 
The six geocoded subsidence maps are shown in Figure 2 with 
the two profiles of subsidence AA’ and BB’ marked on the 
maps. The profiles of subsidence from five of the maps are 
shown in Figures 3(a) and 3(b) respectively. Figure 4 gives the 
correlation between the GPS and InSAR results from Pairs 5. 
3.2 Discussions 
It can be seen that the large subsidence gradients in some of the 
areas (Figure 2). The overall coherence of the interferograms 
also varied in time with the expansion of the city areas. For 
example, the coherence in the northeastern and southwestern 
suburbs of the city was getting better as large Hi-tech zones 
have been gradually built up in the areas since 1996. 
The accuracy of the InSAR results is assessed against the GPS 
results at the GPS stations. Figure 4 shows the annual 
subsidence rates at the GPS stations determined from the 
InSAR and the GPS measurements. The overall correlation 
between the two types of measurements is 0.725, and this 
increases to 0.867 if the outlying ground fissure point XJ03 is 
removed. Most of the residuals from the linear regression are 
within lcm/a. 
All the subsidence maps have shown that the northwestern part 
of the city was stable, while the southeastern and southwestern 
suburbs have been experiencing the most significant subsidence. 
It can also be seen that the subsidence was highly correlated to 
the locations and directions of the ground fissures that are 
mainly in the NEN direction. Some ellipsoidal subsidence 
cones can be identified between the fissures with their major 
axes being parallel to the directions of the fissures. The 
differential subsidence values on the two sides of the ground 
fissures are up to about 2 cm as seen in Figure 3. 
The subsidence rate has evolved over the past 15 years as seen 
in Figure 2. First, both the subsidence rate and the area affected 
increased from 1992 to 1996. By examining the results in 
Figures 2(a) and 2(c), it can be seen that the maximum 
subsidence rate increased from 16cm/a to 20cm/a over this 
period of time. In the same time, the subsidence rate in the 
eastern suburbs also increased. These can also be seen in the 
subsidence profiles given in Figure 3. Second, the subsidence 
cones shifted to the western and southeastern parts of the city 
after 2004, while the northeastern part became stable after 2006. 
The maximum subsidence rate also slowed down to about 
7.5cm/a by 2005 and 2006, as shown in Figures 2(e) and 2(f).