I.D.I.O.T.: A FREE AND EASY-TO-USE SOFTWARE TOOL FOR DINSAR ANALYSIS A. Reigber, E. Erten, S. Guillaso, and O. Hellwich Berlin University of Technology, Computer Vision and Remote Sensing Franklinstr. 28/29 (FR3-1), D-10587 Berlin, Germany Tel.: ++49-30314-23276, Fax: ++49-30314-21104, E-mail: anderl@fpk.tu-berlin.de ABSTRACT This paper presents the I.D.I.O.T. (InSAR Deformation Inspection and Observation Tool) software, developed by the Computer Vision and Remote Sensing Group of the Berlin University of Technology. The purpose of I.D.I.O.T. is to ease as much as possible the generation of differential SAR interferograms, even for complete novices in SAR interferometry. The generation of differential interferograms from ENVISAT-ASAR data is simplified to choice of files of appropriate IMS (single look complex) files. From these files, displacement maps are produced without further user interaction. I.D.I.O.T. is programmed in IDL (Interactive Data Language); a binary edition for non-commercial purposes is provided free-of-charge via the Internet. The current version of I.D.I.O.T. supports ENVISAT-IMS as input data and generates output in png and rat format. Several examples of deformation maps derived with I.D.I.O.T. over test sites in Iran, Kyrgyzstan and Saudi- Arabia will be presented, demonstrating the processing accuracy of I.D.I.O.T. Key words: SAR, DInSAR, subsidence. 1. INTRODUCTION In recent years, differential interferometry (DInSAR) using space-borne synthetic aperture radar (SAR) sensors has become an established technique for detecting and monitoring centimetre-scale deformations of the earth s surface, as well as glacier flows and land slides [1]. DIn- SAR is the study of interference patterns between two SAR images acquired from relatively similar positions. After precise co-registration of the two images, the measured phase difference between two corresponding pixels can be expressed as Φ = Φ topo +Φ flat +Φ diff +Φ atm +Φ err +Φ noise (1) where Φ topo denotes the interferometric phase caused by terrain topography, Φ flat the so-called flat-earth phase generated by the imaging geometry, Φ diff the differential deformation pattern, Φ atm the atmospheric effects, Φ err errors due to inaccurate orbit and topographic height information, and Φ noise the noise contribution. In DIn- SAR analysis, one is usually interested in the deformation pattern Φ diff, which requires to eliminate all the other contributions. The topography and orbit error as well as the atmospheric effect, described by Φ err and Φ atm are usually unknown and are neglected. The flatearth component Φ flat can be calculated from the imaging geometry, while the elimination of the topographic term Φ topo requires precise knowledge of imaging geometry and ground topography. Although the above expression appears simple, it is a common assumption that DInSAR analyses are complicated and require an experienced user, as well as expensive specialised software packages. Consequently, in many cases, potential users are scared off by the alleged complexity and high demands of differential SAR interferometry. However, in practice this is not really true: As long as the imaging geometry is precisely known, differential SAR interferometry is an extremely straightforward technique. With the availability of precise orbit information as well as the global digital elevation model (DEM) of the Shuttle Radar Topography Mission (SRTM), control of imaging geometry became a rather simple task, which can be automatised to a high degree. I.D.I.O.T. is such a software package for fully automatic generation of differential SAR interferograms with a minimum amount of user interaction. Its main purpose is to simplify as much as possible the generation of differential SAR interferograms. This is achieved by a fully automatic handling of the SAR data, orbits and topography information. The user has only to specify a reasonable pair of SAR data sets. While processing the data, all steps are implemented having maximum interferogram quality and precision in mind. This includes in particular a topography adaptive co-registration, range-adaptive DOPPLERfiltering and topography adaptive filtering of the baseline decorrelation. From the derived interferograms, several meaningful maps and colour overlays for visual interpretation are automatically generated. In the following, the processing strategy is I.D.I.O.T. is briefly outlined. Proc. Envisat Symposium 2007, Montreux, Switzerland 23 27 April 2007 (ESA SP-636, July 2007)
Figure 2. Synchronisation of DEM and master image Figure 1. chain Block diagram of the I.D.I.O.T. processing in Fig. 2, for each pixel of the resulting DEM, the shortest line-of-sight (LOS) distance r min between the DEM pixel and the master orbit is calculated. The respective slant-range pixel number p rg of this DEM element can then be derived using p rg = (r min r 0 )/δ sr (2) 2. PROCESSING STRATEGY The quality of differential interferograms is highly based on geometric configuration of satellite orbits and the ground topography. Therefore, I.D.I.O.T. uses a rigorous geometric approach based on precise orbit information and SRTM (Shuttle Radar Topography Mission) elevation data. From these data, precise co-registration information, as well as the topographic phase components, can directly be estimated and deformation maps are computed without further inputs. In order to achieve the maximum coherence, I.D.I.O.T. includes various aspects of advanced InSAR processing, including range adaptive DOPPLER-filtering, topography adaptive range filtering, residual sub-pixel image registration and high-precision interferogram generation. Output is generated as raster images (png format) in reduced resolution and as binary images (RAT format) in full-resolution. The entire block diagram of I.D.I.O.T. is shown in Fig. 1 In the following, the different tasks of data processing are described more in detail. 2.1. Master file processing The first step in the processing is the extraction of the complex image data of the master track. Additionally, the orbit information is extracted from the data header, or, alternatively, using external precise orbit files. After reading the SLC data, for each image pixel the corresponding geographical coordinates are calculated, which requires to project the DEM information onto the slant-range geometry of the master image. To do so, first the required SRTM patches are mosaicked together. Then, as depicted with r 0 denoting the range delay distance and δ sr the slant-range resolution of the sensor. The azimuth pixel number p az is directly determined by the intersection of the estimated LOS direction with the master orbit. Once all the p rg, p az and corresponding topographic heights are known, the DEM is projected to slant-range geometry at full-resolution using a triangulation between the known data points. 2.2. Slave file processing The second step in the processing is the precise coregistration of the slave image onto the geometry of the master image. At this stage of the processing, the master geometry is already fully determined; for each image pixel the precise height is known from the backward-geocoded DEM. Therefore, the slave image co-registration in range can be reduced to a measure of the difference in slant-range distances between a given pixel of the backward-geocoded DEM and the two orbits. Similarly, the co-registration in azimuth is the difference in the intersection points of the two LOS directions with master orbit and slave orbit, respectively [2]. In Fig. 3, this technique is illustrated: First, the precise slant-range distances of the pixels in the master image are calculated from their 3D position in space, known from the backward-geocoded DEM. Then, using the slave orbit, it is calculated which slant-range distance the points possess in the slave image and where the slave LOS intersects with the slave orbit. From this information, the entire vector field of image offsets can be derived without touching the image data itself. On minor problem of this approach is that one depends on the quality of the orbits and on precise timing infor-
Figure 3. Coregistration of master and slave mation in the data headers. In practice, to correct for such errors, it is necessary to perform a calibration of a global offset between the images. In I.D.I.O.T. is done by an additional residual offset correction based on a single cross-correlation measurement between the image amplitudes, although it has been found that in most cases the estimated residual error is in the order of only 1/10th of a pixel. After the co-registration parameters are known, the slave image is resampled to the master geometry by an interpolation based on oversampled cubic-convolution. 2.3. Interferometric processing The most important step in interferometric processing is the generation of the interferogram itself, which has to be performed with the maximum quality in mind. In I.D.I.O.T., first a range dependent filtering of the different DOPPLER-centroids is performed. Then the image is range-filtered using a topography adaptive approach [3]. This ensures an optimal coherence, even in case of very large baselines and in the presence of steep topography. In addition to spectral filtering, the interferogram is filtered in time-domain by a low-pass filter in azimuth, which compensates for the different ground resolution in range ( 20m) and azimuth ( 4.5m). During processing, several versions of the interferometric phase are generated: The pure one, containing flat-earth, topography and deformation, the topographic phase after removal of the phase component due to the earth s ellipsoid, and the differential phase after correcting for ellipsoid and topography. For topography compensation, again the backward-geocoded DEM derived from the SRTM data is used. Additionally to the phase images, a coherence map is calculated from the filtered topography corrected images. 2.4. Generation of output images After finishing the data processing, I.D.I.O.T. automatically generates several images with reduced resolution in Figure 4. The graphical user interface of I.D.I.O.T. png format for direct control of the output and/or inclusion in presentations. Additionally, a binary version of the results is provided in RAT format [4][5]. An overview of all generated files is given in Tab. 1. For the moment, I.D.I.O.T. does not perform phase unwrapping, i.e. the derived deformation maps are mainly thought for visual interpretation. Geocoding is also not performed; all output images are in the slant-range geometry of the master image. 3. GRAPHICAL USER INTERFACE The purpose of I.D.I.O.T. is to ease as much as possible the generation of differential SAR interferograms, even for complete novices in SAR interferometry. Therefore, the graphical user interface is very simple and basically provide only selectors for master and slave files. Additionally, one has to set the output directory and the directory containing the SRTM patches. If the usage of precise orbits is desired, also the directory with the TU Delft orbit file and the path to their getorb tool has to be provided. After starting the process, all processing is performed automatically. In Fig. 4, the graphical user interface of I.D.I.O.T. is depicted. 4. EXPERIMENTAL RESULTS I.D.I.O.T. in its current form has been tested with several ENVISAT data sets. In the following, some examples should be given, demonstrating the the processing accuracy of I.D.I.O.T. and the look of the automatically generated output. The first test-site is the region of Bam (Iran), where a major earthquake occurred on December 26 th 2003. The
Table 1. Overview of the generated output files filename amplitude_master.* amplitude_slave.* amplitude_rgb.* amplitude_coherence.* phase.* phase_flat.* phase_dinsar.* amplitude_dinsar.* amplitude_phase.* amplitude_topography.* coherence_dinsar.* description image amplitude master image amplitude slave RGB composite (amplitude master, amplitude slave, mean amplitude) interferometric coherence CMY composite (amplitude master, amplitude slave, coherence) interferometric phase interferometric phase ellipsoid corrected differential (DInSAR) phase overlay mean amplitude with differential phase overlay mean amplitude with interferometric phase ellipsoid corrected overlay mean amplitude with topography differential (DInSAR) phase darkened by coherence values two investigated image pairs have temporal baselines of 35 and 70 days. The first image pair has a very large spatial baseline of 521m, which is considered not to be ideal for DInSAR analyses. Nevertheless, as shown in Fig. 5, I.D.I.O.T. was capable of producing highly coherent fringe patterns in both cases. The deformation pattern of the earthquake is clearly visible in the overlay of mean amplitude with the differential phase. Despite of the different imaging geometry, both deformation patterns are very similar 1, which demonstrates the accurate compensation of the topographic term. Visually compared to results of this test-site found in literature [6], the derived results appear practically identical. The second test-site is located in Kyrgyzstan and shows parts of the Inyltshik glacier system located in the Tien- Shan mountain range with topographic heights up to 7000m. In Fig. 6, some examples of I.D.I.O.T. derived output is shown. On the left side, two overlays of image amplitude and deformation patterns are depicted, acquired with different baselines and temporal offsets. In both cases, the glacier in the top part of the image appears decorrelated, while a deformation pattern can be observed in the lower part of the image. This example demonstrates, that I.D.I.O.T. is able to reliably derive interferometric information even in case of steep topography and very large baseline (here 747m). On the right side of Fig. 6, an automatically generated CMY overlay of image amplitude and interferometric coherence is shown. In this case, it allows to clearly locate the position of the glacier tongue, since the melting and moving ice of the glacier appears decorrelated due to temporal decorrelation. The third test-site is located around the Abqaiq oil field in the Ghawar region of Saudi Arabia. In Fig. 7, some results from a study carried out using I.D.I.O.T. is shown. Here it has been tried to generate an interferogram from data with a very large temporal baseline of about 1.5 years in order to analyse presumed ground subsidence due to oil extraction. As expected, the derived coherence is relatively low in most part of the image, which is 1 the inverse colouring results from an inverse baseline direction due to temporal decorrelation in the sandy desert around the oil production sites. Over stable areas, a good coherence could be achieved. The DInSAR phase shows some areas, where ground subsidence is probably happening. However, more investigations are necessary for a complete analysis of the observed effects. 5. AVAILABILITY / TERMS OF USE I.D.I.O.T. represents a subset of the internal InSAR software developed at the Berlin University of Technology, which is released as free software. It has certain limitation compared to commercial InSAR packages: As mentioned above, the free version accepts only ENVISAT- IMS data as input and generates only output in png format. Geocoding and phase unwrapping are missing. I.D.I.O.T. runs on UNIX (Linux, Mac OS X, UNIX) and Windows platforms. Further functionality will probably be provided in a commercial version. I.D.I.O.T. has been entirely programmed in IDL (interactive data language) and will be distributed in a precompiled version (sav-file). To run it, the freely available IDL virtual machine is necessary [7]. I.D.I.O.T. will be provided free-of-charge via the internet: http://www.cv.tu-berlin.de/idiot Its usage is completely free for non-commercial and educational purposes. Detailed license regulation will be available when I.D.I.O.T. is released. 6. CONCLUSIONS With I.D.I.O.T. for the first time a free and easy-touse DInSAR tool becomes available, which requires almost no knowledge about InSAR. It is fully automatic and needs the user only to select the correct input files.
Figure 5. Automatically generated results of the Bam earthquake in Iran. Top left: Overlay of image amplitude with topography. Top middle and right: Derived coherence map and deformation pattern at 521m spatial baseline. Bottom left: Overlay of image amplitude with ellipsoid corrected interferometric phase. Bottom middle and right: Derived coherence map and deformation pattern at 8m spatial baseline. Figure 6. Automatically generated DInSAR results of the Inyltshik glacier in Kyrgystan: Both pairs with very large baselines yield similar results with good quality.
Figure 7. Presumed ground subsidence in the Ghawar region of Saudi-Arabia.
I.D.I.O.T. achieves very high coherence even in case of non-optimal baseline constellations and delivers highly accurate interferometric results. As a result, I.D.I.O.T. might be the ideal tool for everybody interested in DIn- SAR deformation maps, but uninterested in the technical details of SAR interferometry and not willing to purchase expensive software for just a few trials. [8] A. Ferretti, C. Prati and F. Rocca: Nonlinear Subsidence Rate Estimation Using Permanent Scatterers in Differential SAR Interferometry, Transactions on Geoscience and Remote Sensing, Vol. 38, No. 5, pp. 2202-2212, 2000 It is important to note that standard DInSAR analyses, as they can be performed with I.D.I.O.T., might be significantly influenced by atmospheric errors. Therefore, in many cases the derived deformation maps can only serve for visual analysis. For precise measurement of subsidence rates, advanced DInSAR techniques are required [8]. I.D.I.O.T. has also several disadvantages. First of all, it is relatively slow, which is a result of a very precise handling of the data. Depending on the image size, typical processing times are about 2h on a 2GHz standard PC. Due to its fully automatic nature, I.D.I.O.T. is also very inflexible. The generated output images are of good quality, but the way I.D.I.O.T. is preparing them is fixed and there is no way of influencing the processing. In its current form, I.D.I.O.T. is specific for ENVISAT and not supporting any other sensor. However, support for ERS- 1/2 is in development. ACKNOWLEDGEMENTS The authors would like to thank Prof. Bitzer, who provided the results of the Ghawar region. His project is supported by the German Research Foundation (DFG) under project Bi 1074/2-1 and by the ESA category-1 project 3781. REFERENCES [1] R. Bamler and P. Hartl: Synthetic aperture radar interferometry, Inverse Problems, Vol. 14, R1-R54, 1998. [2] G. Fornaro, M. Manunta, F. Serafino, P. Berardino and E. Sansosti: Advances in multipass SAR image registration, Proceedings of IGARSS 05, Seoul, South-Korea, 2005 [3] A. Reigber: Range Dependent Spectral Filtering to Minimize the Baseline Decorrelation in Airborne SAR Interferometry, Proceeding of IGARSS 99, Hamburg, Germany, pp. 1721-1723, 1999 [4] http://www.cv.tu-berlin.de/rat [5] A. Reigber and O. Hellwich: RAT (Radar Tools): A free SAR Image Analysis Software Package, Proceedings of EUSAR 04, Ulm, pp. 997-1000, 2004 [6] Y. Xia: Bam earthquake: Surface deformation measurement using radar interferometry, Acta Seismologica Sinica, Vol. 18, No. 4, pp. 451-459. 2005. [7] http://www.ittvis.com/idlvm