IEEE Geoscience and Remote Sensing Magazine - March 2019 - 51

shows three ALOS PALSAR interferograms with different
normal baseline lengths. Figure 13(d) is the residue distribution of Figure 13(a) as obtained by TSPA using Figure 13(a) and (b). In Figure 13(d), there are 64,836 residues
with the polarity range from −4 to 5, and the total polarity
of the residues is 82,542. Figure 13(e) shows the residue
distribution of Figure 13(a) obtained by TSPA using Figure 13(a) and (c). In Figure 13(e), there are 2,064 residues
with the polarity range from −1 to 1, and the total polarity
of the residues is 2,064. It is not difficult to see the effect
of different normal baseline lengths (usually, the sparser
residue distribution and the lower polarity range provide
more accurate PU results).
Figure 14 compares SB and MB PUs using the ALOS PALSAR repeat-pass DB InSAR data set of the Himalayan mountain area. Figure 14(a) and (b) shows two interferograms
with different normal baseline lengths. We can see that the
phase fringes in Figure 14(a) are complicated and dense;
therefore, the Itoh condition cannot be satisfied. Figure 14(c)
and (d) illustrates the PU results of Figure 14(a) obtained
by SB MCF and TSPA, respectively. We can see that there is
a long, vertical, discontinuous topographic jump marked
by the rectangular box in Figure 14(c); this is an obvious
artifact. However, Figure 14(d) is seamless. Figure 14(e) is
the reference unwrapped phase of Figure  14(a) generated
by the PALSAR digital elevation model. Figure 14(f) shows
the errors between Figure 14(c) and (e), with a mean square
error (mse) of 24.92. We can see that the PU error spreads
over the upper-right part of Figure 14(f). Figure 14(g) shows
the errors between Figure 14(d) and (e) with an mse of 7.34.
Because the MB PU method does not need to obey the Itoh
condition, the PU result shown in Figure 14(d) is more accurate and credible than that in Figure 14(c).

(a)

3

3

2

2

90

1

1

80

0

0

70

-1

-1

60

-2

-2

-3

-3

50

90

(b)

80

80

75

70

70

60

65

50
(d)

REVIEW OF LARGE-SCALE PHASE UNWRAPPING
With the rapid advance of InSAR technology, interferog rams are becoming inc reasingly larger in size. For
e xample, TerraSAR-X, e.g. (launched in June 2007) and
its twin satellite, TanDEM-X (launched in June 2010),
reliably provide high-resolution SAR imagery with 1-m
spatial resolution [102]. In this case, an increase of the
InSAR data volume is unavoidable. Moreover, several of
the latest InSAR technologies (e.g., multidimensional
[103], multitemporal [104], and MB [81], [92]) require
that the PU algorithm be able to process multiple interferograms simultaneously, further aggravating the processing burden of the PU step. Hence, this increasingly
large interferometric data volume has opened new possibilities and challenges for PU in terms of computational and memory consumption. If these new challenges
are not well handled, we may fall into a data-rich but
information-poor situation.
If the scale of the input interferogram exceeds the
computer's hardware capability, a divide-and-conquer
criterion needs to be adopted in the PU process. In other
words, the entire input interferogram must be divided
into several smaller subinterferograms and processed
independently. The PU results of all subinterferograms
are then spliced together. Under this condition, if the
PU result of each subinterferogram is not consistent
with that of the input interferogram, the PU result will
exhibit mosaic phenomena (i.e., tiling artifacts), which
greatly reduce PU accuracy [19]. Figure 15 illustrates the
statistics for journal and conference publications on LS
PU (from Web of Science). It is apparent that research
achievements related to LS PU have been quite limited
until recently.

(e)

60

(f)

(c)
20

20

10

10

0

0

-10

-10

-20

-20
(g)

FIGURE 14. (a) The ALOS PALSAR interferogram with a long normal baseline. (b) The ALOS PALSAR interferogram with a short normal
baseline. (c) The PU solution of (a) obtained by the SB MCF PU method. (d) The PU solution of (a) obtained by the TSPA MB PU method. (e)
The reference unwrapped phase of (a). (f) The errors between (c) and (e). (g) The errors between (d) and (e).
march 2019

ieee Geoscience and remote sensing magazine

51
IEEE Geoscience and Remote Sensing Magazine - March 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2019
