|
|
 |
Long time scale INSAR by means of high coherence features
| Stefania Usai and Ramon Hanssen |
|
Delft Institute for Earth-Oriented Space Research (DEOS)
Delft University of Technology, Faculty of Geodetic Engineering
Thijsseweg 11, 2629JA Delft, The Netherlands
email: usai@geo.tudelft.nl |
|
Abstract
-
The monitoring of slow deformation processes by means of SAR Interferometry
requests the observed area to mantain its correlation for more years.
This usually happens only for those areas having particularly favourable
characteristics, such as poor vegetation and dry and no windy climate.
However, even in an area which decorrelates in a few weeks, a certain number of
features has been observed, which appear to mantain high coherence values.
The aim of this paper is therefore to investigate to which extent these structures,
mainly man-made features, remain coherent on long time scales.
For this purpose, a time series in a test area in the Northern Netherlands has been
generated, with a maximum temporal extension of 3 1/2 years. Even on such a long
time span, a considerable number of highly coherent features has been found.
A sample of such features has then been selected and the coherence values have been
traced in the whole time series. This technique has revealed to be useful in order
to check the "goodness" in terms of coherence of an interferogram: in particular,
in this way an interferogram could be identified having generally lower
coherence, probably due to different seasonal conditions. Other tests have been
performed in order to study the coherence as a function of time and of the baseline
length.
Keywords: SAR Interferometry, coherence, time series
Note: for a PostScript version of the paper click here
Introduction
It is well-known that a strong limitation for time range applicability of
repeat-pass differential SAR interferometry is temporal decorrelation.
Therefore, we still cannot perform longterm studies of slow deformation
processes like land subsidence and plate tectonics, in spite of the availability
of SAR images over several years. However, we noticed that even on very long
time spans, highly coherent features are still present, mainly man-made
features. We thus want to investigate whether or not the coherence stability of
these structures could be used for long time-scale monitoring of slow
deformation processes. As a first step, we generated a time series of
interferograms, which constitutes the database of the present work, spanning
different time intervals between 1-day up to more than 3 years. The details
and some remarks about the construction of such time series are shown in section 2.
We then selected a sample of features showing high coherence on long time scale
and we performed some tests on it. The first results are presented in section 3.
The 1992-1996 time series
The database is a time series of interferograms of the area around the city of
Groningen, in the northern part of The Netherlands. The area is well known for
its land subsidence, caused by the extraction of natural gas: the rate of land
subsidence is up to 1 cm/yr. In order to be able to detect it by
means of the INSAR technique, monitoring of the area for more years would be
necessary. As a first step to assess whether this is possible, we want to
study the features (mostly man-made structures) which preserve their
coherence on such long periods.
Figure 1: The Groningen dataset. The 5th column shows the time
span in days, the 6th the serial number of the interferogram.
| master | slave | B_par | B_perp | days | no |
| 16-3-96 | 17-3-96 | -17 | 24 | 1 | 9 |
| 16-3-96 | 11-2-96 | 36 | 212 | 34 | 8 |
| 16-3-96 | 20-4-96 | 50 | 145 | 35 | 7 |
| 16-3-96 | 21-4-96 | 18 | 79 | 36 | 6 |
| 16-3-96 | 6-1-96 | -109 | -129 | 70 | 5 |
| 16-3-96 | 20-8-95 | -62 | -272 | 209 | 4 |
| 16-3-96 | 19-8-95 | -26 | -190 | 210 | 3 |
| 16-3-96 | 15-10-92 | 5 | 25 | 1248 | 2 |
| 16-3-96 | 10-9-92 | 43 | 52 | 1283 | 1
|
The time series generated for this purpose has
the following characteristics:
The same image is used as master for all the interferometric pairs, in
order to guarantee that all the slave images are interpolated on the same grid,
namely the master. In this way, a given pixel represents the same area in all the
interferograms.
Azimuth filtering has been applied: our tests confirmed what was already
highlighted in the literature (Geudtner, 1994, and Schwabisch, 1995), i.e. that
azimuth filtering highly improves the coherence.
This is particularly important for the long time scale interferograms, where the
coregistration is more difficult because of the generally low coherence.
Figure 1 shows the interferograms generated, and their baseline
components, column 5 contains the interferogram serial numbers, which are used
as reference in the plots. In Figure 2 the distribution of the time
spans and of the perpendicular component of the baselines is visualized.
The final products, i.e. the coherence and phase images, are mediated over
2 pixels x 10 lines.
Figure 2: Perpendicular component of the baseline vs. temporal gap between the
two images for each of the interferograms.
|
Analysis of the coherence
The coherence is estimated on a 2 x 10 window. Figure 3 and
figure 4 are the coherence images respectively on the shortest (tandem
pair, no.9) and on the longest (about 3 1/2 years, no.1) time interval
considered.
Figure 3: Coherence image of interferogram no.1 (16-3-96/10-9-92).
|
Figure 4: Coherence image of interferogram no.1 (16-3-96/10-9-92).
|
We selected in the coherence image no.1 those pixels having coherence higher
than 0.8. We concentrated then for our tests on the area of the city of Assen
(low left in figure 4), which contains a statistically significant
number of such pixels. The area has an extension of 200 pixels x 300 lines.
The coherence of those pixels has been traced in the whole time
series. Figure 5 represents the coherence of these points as results
in all the interferograms: each column represents an interferogram of the
series, each horizontal line contains the values of the coherence for a given
point.
Figure 5:Coherence values for each point of the dataset and for each interferogram.
|
Note that high coherence in the longest time span interferogram (no.1)
doesn't seem to imply high coherence in other interferograms on shorter time
intervals. As we could expect, in fact, the time gap doesn't seem to have any
influence on the coherence values. This is also confirmed by the fact that any
attempt to find a linear trend in the coherence as a function of time didn't
give any significant result. It is also evident from figure 5 that
interferogram no.8 presents a generally lower coherence. We don't know the
reason for this; it can be possibly due to different weather conditions (the
master image is taken in March, the second in February, during a period of snow
precipitation) and to the long baseline.Figure 6 shows the
coherence histograms of the time series.Figure 7 is the plot
of the coherence values along the series for the ten highest-coherence points.
Figure 6:Histograms of the coherence values for each interferogram.
|
Figure 7:Coherence values in the series for the ten highest-coherence points.
|
We also considered the possibility of a dependence of the coherence on the
baseline length. For this purpose, the coherence has been plotted against the
baseline length (perpendicular component) in figure 8, where under each
baseline value, along the x-axis, also the corresponding interferogram serial number
has been indicated.
Figure 8:Coherence vs. baseline length (perpendicular component)
in the series. Under the baseline values, the corresponding interferogram
numbers are given.
|
The interferogram having coherence generally lower than the others (no.8) has
not been considered here. We also did not consider the two long-period
interferograms (no.1 and no.2): we restricted ourselves to the temporally
nearest interferograms. From the figure there seems to be some trend, but we
have to perform more detailed tests in order to be able to assess it. We note also
that for interferogram no.5 the coherence is worse than for the other
interferograms with higher baseline. Since the slave in this pair is taken in
January, this could be due to different weather conditions with respect to
the time the master has been taken.
Conclusions
A time series of interferograms has been generated covering time spans up to 3
1/2 years. Even on such a long time span, highly coherent features could be
identified. The analysis of the series in its whole permitted the identification
of an interferogram showing significantly lower coherence than the others.
No significant signatures of dependence of the coherence values from the time
have been found. Some dependence of the coherence from the baselines could be
present, but more tests are necessary in order to assess it.
Acknowledgements
We would like to thank P.Visser and R.Scharroo of DEOS for providing the ERS
precise orbits.
References
- Geudtner, D., Die Interferometrische Verarbeitung von SAR-daten des ERS-1,
PhD Thesis, University of Stuttgart,1994.
- Schwabisch, M., Die SAR-interferometrie zur Erzeugung Digitaler Gelaendemodelle,
PhD Thesis, University of Stuttgart,1995.
Keywords: ESA European
Space Agency - Agence spatiale europeenne,
observation de la terre, earth observation,
satellite remote sensing,
teledetection, geophysique, altimetrie, radar,
chimique atmospherique, geophysics, altimetry, radar,
atmospheric chemistry
|
EO Data Access
