WIDE BASELINE INTERFEROMETRY WITH VERY LOW RESOLUTION SAR SYSTEMS

1 of 25 26/03/2008 22.35 ne previo WIDE BASELINE INTERFEROMETRY WITH VERY LOW RESOLUTION SAR SYSTEMS Abstract: A. Monti Guarnieri, C. Prati, F. Rocca and Y-L. Desnos (*) Dipartimento di Elettronica e Informazione - Politecnico di Milano Piazza L. da Vinci, 32-20133 Milano - Italy fax: +39-2-23993585, e-mail: monti@elet.polimi.it (**) European Space and Technology Centre (ESTEC) Keplerlaan 1 - Postbus 299-2200 AG Noordwijk Netherlands We propose a combination of low + high resolution SAR acquisitions for differential SAR interferometry. A technique has been introduced the exploits a DEM to remove the volume scattering decorrelation that would result from such a combination. The resulting interferogram has the quality of a conventional full resolution one, but reduced geometric resolution. As a possible application we propose a system that exploits the wide swath imaged by low-resolution ScanSAR modes for frequent revisiting of areas interested by crustal deformation phenomena. Introduction Applications of SAR interferometry (InSAR ) and more specifically in the field of differential interferometry (DInSAR ) are well known and described in the literature [1,2,3]. DInSAR surveys can be used to detect small surface deformations (like glacier motions, earthquakes, landslides [2,4]), or even atmospheric changes [5]. Usually, DInSAR surveys are generated by exploiting high resolution SAR images, but in many cases the differential interferogram, i.e. the phenomena to be monitored is smooth; therefore it can also be monitored by Low-Resolution SAR. The principal idea expressed here is to provide DInSAR surveys by combining a full resolution acquisition, like for ASAR Image Mode (IM ), with a

2 of 25 26/03/2008 22.35 low-resolution one like for ASAR ScanSAR modes (WSM / GMM ). A technique capable of synthesizing a low resolution, zero-baseline interferogram from such a combination is proposed. This technique removes most of the volume scattering decorrelation, and, at the same time, extends interferometric capabilities to wide baselines, e.g. to the same baseline range of combined IM / IM interferometry. As a possible application of the proposed technique, two frequent revisiting systems are discussed. They exploit the wide-swath accessed by the low resolution ScanSAR mode and the potentials of differential interferometry. One system allows any acquired IM image to give a DInSAR survey, and is thus suited for monitoring hazardous areas on users' demand, whereas the other system is intended for detecting unpredictable events (earthquakes, landslides etc.) be performing systematic acquisitions over the areas of interest. Finally, examples of combined full resolution - low resolution interferometry obtained by simulating ASAR IM, WSM and GMM modes by means of ERS data are given. Interesting results are obtained by processing the Landers earthquake dataset [2,6] in low-resolution mode. SAR-ScanSAR Differential Interferometry The paper deals with repeat-pass, satellite-borne differential SAR interferometry. The generation of DInSAR surveys has been summarized in Fig. 1. Figure 1: Block diagram of a typical DInSAR processor. An interferogram is first generated, as usual, by co-registering the two phase preserving focused images and then by multiplying one image by the conjugate of the other. The interferogram is then compensated for the topography by means of a ``synthetic interferogram'' derived by a DEM (or another interferogram), co-registered in the same reference. The result is a residual, differential interferogram, whose phase depends on the

3 of 25 26/03/2008 22.35 ``changes'', like terrain motions, and is used to detect small crustal deformations or even changes in the atmosphere [5]. Usually, DInSAR surveys are generated starting from two full resolution SAR images. Yet, it will be shown here that low-resolution, wide swath ScanSAR acquisitions (available on most of the recent sensors, like RADARSAT, SIR-C,ASAR, LightSAR) could be combined with full resolution ones to give high quality DInSAR images. ScanSAR Interferometry The typical ScanSAR geometry is sketched in Fig. 2, for the two modes foreseen for that sensor: the Wide Swath Mode (WSM, resolution 150 150 m) and much lower resolution Global Monitoring Mode (GMM, resolution 1 1 km). Figure 2:ASARScanSAR geometry. The most important parameters have been summarized in Tab. 1, and compared with the full resolution, Image Mode (IM ). As Fig. 2 shows, ScanSAR achieves a large swath coverage, 400 km (ground range) for the ASAR

4 of 25 26/03/2008 22.35 case, by periodically switching -on the fly- the antenna pointing in five sub-swaths, each on them of 80 Km in ground range. Such way of operating causes the data within one sub-swath to be acquired in short bursts whose azimuth extent is samples for GMM and samples for WSM. The loss of data with respect of full resolution SAR system, due to the non-acquired pulses, causes a proportional loss in the azimuth resolution. This is what ScanSAR trades off for getting an increased range coverage. Table 1: Parameter values for ASAR Image and ScanSAR modes. The generation of interferometric surveys starting from two ScanSAR images is known [7,8]. Also well known is the need to use a IM and a ScanSAR image when it is not possible to have a pair of ScanSAR images acquired with synchronized patterns. Furthermore, in the ASAR case poor quality is expected due to the ``volume scattering decorrelation'' [3]. In practice, whenever in a single resolution cell, a pixel, the scatterers are uniformly distributed within an altitude r 0 being the slant range, B n the normal baseline and the incidence angle, their total contribution to the interferogram is pure noise. For a typical ASAR geometry, and assuming a baseline B n = 300 m, the altitude variations with the single resolution cell (150 150 m for WSM, 1 1 km for GMM ) should be much less than m (depending on the sub-swath). As a further example, for a distributed target over a flat, horizontal terrain surface, the normal baseline should be: (1)

5 of 25 26/03/2008 22.35 (2) where B nc is the ``critical'' value (for which we expect complete decorrelation). If we take some margin over (2), say B n =B nc /4, we get a baseline limit of m, for GMM, and 110 m for WSM. The occurrence of such baselines will be so marginal to exclude repeat-pass ScanSAR interferometry for ASAR, particular for frequent monitoring applications. Celestial Footprint Steering A more feasible system that still allows the exploitation of low-resolution imagery for differential SAR interferometry, consists in a combination of ScanSAR images with IM ones. Let us assume the geometry of Fig. 3, and let v be the complex backscattering (or ``reflectivity'') of a distributed target on the azimuth, slant range plane of the first acquisition (the full resolution, IM one).

6 of 25 26/03/2008 22.35 Figure 3: Geometry of the combined low/high resolution ISAR. The focused signal from sensor operating in IM mode, v 1, is then a bandpass filtered version of the wide-bandwidth bidimensional reflectivity, v where denoted the 2D convolution. In (3), h 1 is an ideal bandpass filter whose Fourier Transform is (3) where f r, f x are the range and azimuth frequency, and B r1, B x the range and azimuth bandwidth. The same target observed by sensor operating in ScanSAR mode results in the imaged reflectivity v 2, that can be expressed as follows:

7 of 25 26/03/2008 22.35 (4) h 2 accounts for the narrow bandwidth of the ScanSAR acquisition: being the difference in the two sensor-target travel paths for the target at range, coordinates (r,x). In (4) the filter where and. The central azimuth frequency f x0 is azimuth varying and accounts for the spectral properties of ScanSAR (see [8]). For zero-baseline, and the ScanSAR reflectivity, v 2 in (4), can simply be derived by filtering the IM one, v 1 in (3). For non zero baseline, v 2 could theoretically be synthesized according to (4), once there is known both v and the geometry dependent term: In the actual case, the complex reflectivity v is not known, but rather its bandpass filtered version v 1, in (3); hence the idea is to use v 1 in place of v for the actual synthesis of reflectivity: (6) (5) The synthesized ScanSAR reflectivity, in (8) can be then combined with v 2 to give the differential zero-baseline interferogram. Note that the zero-baseline synthesis is not exact since both the bandlimited reflectivity v 1 is used in (8) place of the very large bandwidth one, v in (7), and the DEM used to derive the fringes term s(r,x) has limited resolution. However, these approximations are acceptable when the range bandwidth of the geometric term s in (6) is, (and the same holds for the azimuth bandwidth), e.g. the fringes are stationary within the full resolution cell. Therefore, even if the cfs technique does not completely remove volume scattering decorrelation (like for true zero-baseline acquisitions), it is yet capable to reduce it up to the same level experienced in full resolution mode. DEM accuracy A critical aspect is the accuracy of the DEM used to synthesize ScanSAR reflectivity in (8). We can assume that the DEM has vertical errors with normal distribution. Then the probability density of the interferometric phase is:

8 of 25 26/03/2008 22.35 where the phase standard deviation can be related to the DEM error standard deviation and the local terrain slope,, It can be shown that the coherence is reduced by a factor: (7) being the characteristic function for normal distribution. As an example, Fig. 4 plots the maximum to ensure 0.7, computed for different combinations of baselines and slopes. It appears that values of in the range of m can be accepted for baselines up to 300 m. This accuracy can be provided by multibaseline averaged DEM ( m, [5]), and also from currently available DEM ( m, for USGS level 1 DEM ).

9 of 25 26/03/2008 22.35 Figure 4: DEM accuracy required to get a decorrelation <30%. Implementation The generation of differential interferogram by combining low / high resolution SAR acquisitions is detailed in the block diagram of Fig. 5.

10 of 25 26/03/2008 22.35 Figure 5: Block diagram of a DInSAR processor that uses the cfs technique. First of all, the ScanSAR raw data should be phase preserving complex focused, e.g. by using an algorithm in literature ([7,8]). Azimuth looks can be formed either before or after focusing. Each look results from focusing only a single burst in the footprint. An example of two focused looks coming from adjacent bursts is shown in Fig. 6 together with their spectral supports.

11 of 25 26/03/2008 22.35 Figure 6: Time-varying ScanSAR azimuth power spectrum. The marked blocks highlight the spectral support of two adjacent looks. The following processing are performed to the SAR IM image: 1. 2. the fringes contribution due to the known topography is compensated, by demodulating times the synthetic interferogram, derived from the DEM, according to (6,8), the image is bandpass filtered and subsampled to extract looks that match with the ScanSAR ones [*]. The pass-band of the filter, 5, should be the same of ScanSAR, and the central frequency f x0 should be tuned according to the ScanSAR timeline, in order to select the desired look. The last step of the differential interferogram generation is performed -as usual- by conjugation and multiplication, for each look. All the single look interferograms are then coherently averaged to get the final multi-look interferogram.

12 of 25 26/03/2008 22.35 Continuous monitoring system performances The DInSAR technique just proposed could usefully exploit the wide swath-low-resolution ASAR modes for a frequent monitoring system. The system will first make a reflectivity data-base of the area to be monitored. Then, each time a new image is acquired, it is combined with a proper one in the data-base and with an existing DEM to provide the requested differential interferogram. The large swath coverage of ScanSAR system ensures frequent monitoring, with revisiting time: n r being the number of revolutions per day, the total swath that can be accessed within one pass (e.g. 400 + 400 km if we assume ascending + descending orbits coverage) and the latitude of the area to be observed. For ASAR the revisiting interval is distributed with latitude as follows: (8) If the system is intended to monitor unpredictable events over large areas, by using the WSM mode, the scheme of Fig. 7.a should be adopted.

13 of 25 26/03/2008 22.35 Figure 7: Frequent revisiting systems for monitoring hazardous areas. In that case the reflectivity data-base should contain full resolution IM images, acquired by different view angles, so that for each new WSM acquisition, there is at least one image in the data-base with a ``suitable'' baseline. The same baseline limit assumed for IM interferometry could be applied if the cfs technique is exploited. A critical aspect is the number of SAR images required to build up the minimal ``data-base''. This depends on the orbit span and the number of sub-swaths considered, i.e. the desired revisit time. As an example, Fig. 8 plots the number of images that should be in the data-base, computed for different combinations of revisiting time (at ) and orbit span, in the case of randomly spaced orbits. Note that for the combinations corresponding to the larger orbit span and the shorter revisiting time, the acquisition time for the whole data-base would require years.

14 of 25 26/03/2008 22.35 Figure 8: Number of images to be acquired to build the minimum database. The number should be doubled to include descending passes. As a different strategy, one can create a WSM/GMM data base and revisit, when necessary, the location by using the IM mode, interfering the acquired image with the proper low resolution one in the data base, as shown in Fig. 7.b. In this way there is the advantage of a smaller data-base (it is low resolution), and a short time to build it, due to the extended swath depth. For the ASAR case, 4+4 (ascending + descending) WSM images of the same location would ensure a revisit interval of 7 days with an orbit span of 1000 m, or a revisit time of 3 days with an orbit span of 500 m. Results from simulations Validation of combined IM / WSM (or GMM ) interferometry has been performed by exploiting ERS raw data, properly windowed (along azimuth) and low pass filtered to simulate ASAR ScanSAR modes. Two different test areas were assumed: The area of Mt. Etna, that is particularly interesting for the contemporary presence of arid and vegetated areas, both on the land and on the

15 of 25 26/03/2008 22.35 mountain. The area of the Landers earthquake, that was studied and reported first in paper [2]. This dataset offers the opportunity to validate the proposed DInSAR technique with a real case of coseismic motion. Table 2: Parameters of the interferometric images over the selected test areas. The test area of Mt. Etna is represented in Fig. 9 as it would be imaged respectively in IM, WSM and GMM modes.

16 of 25 26/03/2008 22.35 Figure 9: Multi-look detected values of Etna, obtained by simulating: (a) IM, (b) WSM, (c) GMM mode.

17 of 25 26/03/2008 22.35 Figure 10: Coherence maps of Mt. Etna:(a) IM/IM interferometry; (b) IM/WSM interferometry and compensation for flat earth only, (c)im-wsm interferometry with the proposed cfs technique. As a first test, a complex interferogram has been generated by simulating WSM / IM interferometry basing on dataset TS1. A multi-baseline derived DEM (see [5]), with accuracy 6 m was exploited to perform cfs. A comparison of the coherence maps obtained by compensating for flat earth and by applying the proposed cfs technique is shown in Fig. 10. The significant improvement of coherence achieved with the cfs technique, (see also the histograms in Fig. 11), is due to the large baseline. Note that the achieved coherence is close to the IM / IM case. As a second test, we have simulated GMM / IM interferometry, where a smooth terrain upswelling (10 km wide and with peak displacement of 2.8 cm) was introduced by adding a gaussian phase field to the SAR IM image. Instead of a DEM (that was not available over the whole image), a ``synthetic interferogram'' was generated by ``cleaning'' the full resolution fringes with an adaptive bandpass filtering.

18 of 25 26/03/2008 22.35 Figure 11: Histograms of the three coherence maps. The resulting, differential interferogram is shown in Fig. 12.a (one look) and Fig. 12.b (three looks averaged). Fig. 12.c draws a coherence map obtained by full-resolution processing the same dataset. Note, by comparing Fig. 12.b and c that the phase dispersion is low where the coherence is high. Furthermore, note the vertical strips visible in Fig. 12.a in correspondence of the junctures between azimuth looks. These discontinuities are rather reduced in the multi-look image. As a last test, we processed the Landers data set used in [6], where the tandem ERS-1/ERS-2 pair (L2 in Tab. 2) was unwrapped to get the DEM. The quality of tandem pass was so good that very few errors were generated by unwrapping.

19 of 25 26/03/2008 22.35 Figure 12:GMM / IM Differential interferometry over Mt. Etna. A gaussian displacement with amplitude 2.8 cm and radius 5 km was assumed: (a) 1 look, (b) 3 looks averaged. (c) : IM coherence map. The detection of the coseismic motion by combined ScanSAR / IM DInSAR image was achieved in two different ways: either according to the usual technique draws in Fig. 1 or by the cfs technique, e.g. by performing DEM compensation while getting the interferogram. In this specific case, where the baseline is low (compared to the system resolution) and the topography mostly flat, the two techniques provides comparable results almost everywhere. However, the cfs technique achieves better quality over sloped terrains. The fact is shown in Fig. 13 where the difference bewteen the two coherence maps is compared with the map of the local slopes (measured in the steepest direction and with the sign of range slopes).

20 of 25 26/03/2008 22.35 Figure 13:(a) Difference between the coherence maps obtained by the proposed csf technique and by performing flat earth compensation only. (b) Local slopes (degrees). The advantage achieved by cfs over sloped terrain has been evidenced also in Fig. 14, that plots the average coherence measured for each local slope (and by excluding the completely decorrelated areas).

21 of 25 26/03/2008 22.35 Figure 14:(a) Average coherence as a function of local slope; (b) number of samples found for each slope. Fringes of the earthquake's area are shown 15 in either high resolution and low-resolution modes. It must be noted that - if measured on a pixel-by-pixel basis, the qualities of the two interferograms are comparable, as is also shown by Fig. 14. Yet, when the IM / IM interferogram is averaged to put it in the same geometric resolution as the IM / WSM one, the SNR improves in proportion to the square root of the resolution ratio, and this effect is visible in Fig. 15.

22 of 25 26/03/2008 22.35 Figure 15:(a) Differential interferogram of Landers earthquake, obtained by combininq (a) IM / IM and (b) IM / WSM acquisitions Finally, Fig. 16 shows a comparison between GMM / IM fringes, achieved with and without cfs, in this second case the terrain slopes caused total decorrelation whenever the area in the large resolution cell was not flat. However, once these effects were removed, an adequate detection of the coseismic motion fringes was possible even at the very low GMM resolution (1 km).

23 of 25 26/03/2008 22.35 Figure 16:IM/GMM combined interferogram: (a) by compensating for flat earth and (b) by exploiting cfs technique. Acknowledgments This work has been done under ESA contract N. 12285/97/NL/GS. The authors would like to thank Dr. D. Derauw and J. Moxhet for providing the Landers data set and Dr. A. Ferretti for providing the multi-baseline DEM of Mt. Etna and Dr. Ing. F. Gatelli for processing the Landers data set. Conclusions A technique has been presented to combine a full resolution IM acquisition, a low resolution one (WSM or GMM ) and a DEM to get high quality differential interferograms. This technique removes completely the decorrelation due to topography and volume scattering effects and extends the baseline range up to the limit implied in the full resolution interferometry. The WSM / IM combination has led to the design of two frequent revisiting monitoring systems. The first one, suited for detecting unpredictable hazards, requires a large data base ( 10 IM images for each site) and a significant time to build it. The other, suitable for routinely monitoring a specific site, requires a very compact data base (as many WSM image of each site as many independent angles are considered) that can be built say within 35 days (ASAR revisit time). The major limits to these systems lie in the detail of the differential phenomena to be monitored (that should be very smooth), and in the presence of temporal decorrelation, that limits availability to long term stable targets. Validation of the technique by means of ERS-1 data has shown interesting

24 of 25 26/03/2008 22.35 results, particularly in the case of the Landers earthquake. References 1 A. K. Gabriel, R. M. Goldstein Crossed orbit interferometry: theory and experimental results from SIR-B, in J. Remote Sensing, Vol. 9, no. 8, pp. 857-872, 1988. 2 3 4 5 6 7 8 Massonet D., et. al., The displacement of the Landers earthquake mapped by radar interferometry, Nature, Vol. 384, Jul. 93, pp. 138-142. Gatelli F., Monti Guarnieri A., Parizzi F., Pasquali P., Prati C., Rocca F., The wavenumber shift in SAR interferometry IEEE Trans. on Geosci. and Remote Sens., Vol. 32, No. 4, pp. 855-865, Jul 1994. R. Kwok, M. A. Fahnestock, Ice Sheet Motion and Tomography from Radar Interferometry, in IEEE Trans. on Geosci. and Remote Sens., Vol. 34, No. 1, pp. 189-200, Jan 1996. A. Ferretti - C. Prati - F. Rocca Multibaseline InSAR DEM Reconstruction: the Wavelet Approach, IEEE Trans. on Geosci. and Remote Sens., submitted 9/1/97 [unpublished]. D. Derauw, J. Moxhet, Multiple Images SAR Interferometry in proc. Fringe 96 Workshop on ERS SAR Interferometry, 30 Sept. - 2 Oct. 1996 University of Zurich (CH), Vol. ESA-SP406, pp. 167-178. R. Bamler, Adapting Precision Standard SAR Processors to ScanSAR, in proc. IGARSS '95 (Florence, Italy), 10-14 Jul, 1995, vol. 3, pp. 2051-2053, 1995. A. Monti Guarnieri, C. Prati ScanSAR focusing and interferometry - in IEEE Trans. on Geosci. and Remote Sens., Vol. 34, No. 4, pp. 1029-1038, Jul. 1996. Footnotes

25 of 25 26/03/2008 22.35...ones Note that the looks extracted from SAR image should match in time and frequency those of ScanSAR, to simulate perfectly aligned acquisitions [7,8]. This implies the knowledge (or the estimate) of the offset between the scan pattern and the scene position. ne previo Andrea Monti Guarnieri 2/13/1998