105 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Figure 2. The subsidence-velocity map of the study area. P) and P 2 are marked as two PS points analyzed in Figure 4. 
Figure 3. Comparison between simulated (a) and observed (b) differential interferograms with time interval of about 4 years. 
The further data reduction concentrates on a patch of 27 km by 
15 km within the ERS SAR frame as shown in Figure 1. The 
14618 PS pixels detected out by ADI are superimposed onto the 
amplitude image by red points. 86 differential interferograms 
are generated by the “two-pass” method. The DEOS precise 
orbit state vectors and the SRTM DEM (about 10-m accuracy) 
are used to remove both flat-earth trend and topographic effect, 
thus highlighting land subsidence. 
A very strong network was created by freely connecting each 
PS and all the others less than 1 km apart, resulting in 1463306 
arcs. The increments of both linear motion velocities and 
elevation errors at each arc were then estimated by maximizing 
the model coherence with equation (3). The LOS deformation 
velocities and elevation errors at all the PSs are estimated by 
the weighted LS solution. Figure 2 reports the derived 
subsidence-velocity map in the study area. It can be seen that a 
subsiding bowl with a diameter of about 5 km appears in 
Glendale and has a peak subsidence rate of 54 mm/yr, while a 
wider subsiding bowl with a diameter of about 12 km spans 
Glendale, Peoria and Sun City and has a peak subsidence rate of 
30 mm/yr. It can be inferred that the linear subsidence 
magnitude accumulated during the maximum time span of SAR 
acquisitions (about 8 years) may be up to 43 and 24 cm, 
respectively, at the two peaks. The eastern parts of the study 
site present subtle or zero subsidence. The subsidence in 
farmlands cannot be estimated due to the lack of PSs. 
The fidelity of the estimated subsidence rates has been checked 
by visually comparing the observed differential interferograms 
with those simulated using the subsidence-velocity map. As an 
example, Figure 3 shows such comparison for the differential 
interferograms with time interval of about 4 years. It is evident 
that they are in good agreement. Some minor inconsistency in 
some areas can be ascribed to atmospheric artifacts, topographic 
errors, and nonlinear motion. It also can be seen that the small- 
extent but deeper subsiding bowl in Glendale can be completely 
recovered by the PS networking method. However, its complete 
shape and extent do not present in any observed individual 
differential interferograms due to temporal decorrelation. All 
these not only verify that the estimation approach is powerful 
and reliable, but also suggest that the linear subsidence in the 
study area dominates the nonlinear component. 
The nonlinear subsidence was separated from the atmospheric 
artifacts by both the SVD and EMD method. As examples, 
Figure 3 shows the temporal evolution of atmospheric delay in 
LOS direction, nonlinear and total subsidence at two PS points 
(PI and P2) near the centres of two subsiding bowls (see Figure 
2). The atmospheric variation is evidently random in time. The 
atmospheric artifacts at P2 range from -2.0 to 2.1 cm, which are 
slightly higher than those at PI. Point P2 presents a dynamic 
range of-2.5-2.2 cm nonlinear subsidence, while point PI has a 
narrower range of nonlinear subsidence (-2.0-1.4 cm). 
Additionally, it can be seen that point PI located near the 
deeper subsiding bowl exhibits more seasonal undulation than 
point P2 located near the shallow subsiding bowl. From the two 
profiles of total subsidence, we stress once again that the linear 
trend of subsidence dominates the nonlinear component in this 
study area.