The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B7. Beijing 2008 
175 
artifacts) are compensated for at PSs by exploiting their 
different spectral behaviors in a multidimensional framework of 
time, space and normal baseline (i.e., acquisition geometry). 
4.1 Processing Result 
The 26 scenes of archived ERS-1/2 SLC images are processed 
by TRE (In the framework of Dragon I Project). 22,139 PSs are 
identified in the test area of about 1,000km 2 , i.e. with a density 
of 22 points/ km 2 . The subsidence velocities on the PSs from 
1993 to 2000 are derived to produce the subsidence map, as 
showed in figure.4. 
4.2 Validation 
In the test area, 126 LBs were measured from 1993 to 2000 year, 
as showed in figure.5. The levelling campaign was made each 
year in this period, LBs rank Level 1, Level 2 as the levelling 
standards of China, with standard deviation less than 2mm. The 
average of multiple surveys at each LB is used as the criterion 
to validate the result in long time series. 
Nearest neighboring method is used to validate PSs computed 
by PS. The distance between PSs and LBs is limited to range of 
100m. Within the distance, PS closest to LP will be selected for 
comparison. In figure.8, 115 pairs of PSs and LBs are chosen 
for comparison, and the comparison result shows good 
agreement between PSs and LBs. The mean and standard 
deviation of errors (i.e. the difference between PSs and LBs) are 
respectively -2.02mm/yr and 3.44mm/yr. 85.22 % absolute 
value of errors are centralized in the range of 6mm, and the 
number and percentage of errors decrease with the increase of 
errors, as shown in figure. 10. 
5. STUN TECHNIQUE 
In the processing system of STUN technique, the PSs are 
initially identified with sublook correlation method (Lijun Lu et 
al., 2008) instead of amplitude dispersion threshold method in 
the primitive STUN as it can identify PSs with a limited number 
of observations or even without temporal baseline. The PSs 
identification method is applicable for small dataset. STUN is 
based on the refined differential interferometric model 
(functional model and stochastic model). The functional model 
is established similar to interferometric phase model at 
identified PSs. And the stochastic model is also introduced to 
adjust the different weight for interferometric data, which 
contains disturbing phase terms (e.g. atmospheric artifacts and 
noises). On the basis of functional and stochastic model, 
weighted integer least-square estimator is utilized to estimate 
the integer ambiguity with the highest probability and the 
unknown key parameters including DEM error and subsidence 
velocity. 
5.1 Processing Result 
We processed the 9 latest scenes of ASAR SLC images. 83,044 
PSs are identified in the test area of about 1,000km 2 ,i.e. with a 
density of 83 points/ km 2 . The density of PSs in short time 
series is much bigger than ones in long time series as more 
points exhibiting phase stability in the shorter time span are 
identified as PSs. The subsidence velocities on the PSs from 
2003 to 2005 are derived to produce the subsidence map, as 
showed in figure.6. 
5.2 Validation 
In the test area, 177 LBs ranked Level 1 and Level 2 are 
measured from 2003 to 2005 year, as showed in figure.7. There 
are two levelling campaign in this period. The average of two 
surveys is used as the criterion to validate the result in short 
time series. 
Nearest neighboring method is also used to validate PSs 
computed by STUN. As shown in figure.9, there are 140 pairs 
of PSs and LBs chosen for comparison, and the result shows 
good agreement between PSs and LBs. However, we also notice 
that a few bigger errors appear and the possible cause may be 
that subsidence variation does not fit the referred the linear 
model. The mean and standard deviation of errors are - 
2.01mm/yr and 4.30mm/yr respectively. Seen from figure.ll, 
79.29% absolute value of errors are centralized in the range of 
6mm, and the number and percentage of errors decrease with 
the increase of errors. 
Figure 4. Subsidence velocity map in 1993-2000 
Figure 5. The Leveling Benchmarks surveyed in 1993-2000