COMPARATIVE ANALYSIS OF INSAR DIGITAL SURFACE MODELS FOR TEST AREA BUCHAREST

Transcription

1 COMPARATIVE ANALYSIS OF INSAR DIGITAL SURFACE MODELS FOR TEST AREA BUCHAREST Iulia Dana (1), Valentin Poncos (2), Delia Teleaga (2) (1) Romanian Space Agency, Mendeleev Street, , Bucharest, Romania, (2) Advanced Studies and Research Center, 19 Ion Luca Caragiale Street, , Bucharest, Romania, ABSTRACT This paper presents the results of the interferometric processing of ERS Tandem, ENVISAT and TerraSAR- X for digital surface model (DSM) generation. The selected test site is Bucharest (Romania), a built-up area characterized by the usual urban complex pattern: mixture of buildings with different height levels, paved roads, vegetation, and water bodies. First, the DSMs were generated following the standard interferometric processing chain. Then, the accuracy of the DSMs was analyzed against the SPOT HRS model (30 m resolution at the equator). A DSM derived by optical stereoscopic processing of SPOT 5 HRG data and also the SRTM (3 arc seconds resolution at the equator) DSM have been included in the comparative analysis. 1. INTRODUCTION A digital elevation model (DEM) consists of a number of points with X, Y and Z coordinates describing the bare soil. The generation of DEMs based on remotely sensed data can be efficient and cost effective. Basically, the methods used for DEM generation can be divided in two main categories: (1) stereogrammetry techniques, using aerial / satellite imagery or radar data and (2) radar interferometry [1]. Initially, DEMs are digital surface models (DSM) containing points located on top of the visible surface, including buildings and vegetation. These DSMs have to be filtered in order to remove all the points that do not belong to the bare ground [2] and to obtain the final DEM. DEMs are used in a wide range of scientific, commercial, industrial, and military applications. Based on ERS Tandem, ENVISAT ASAR and TerraSAR-X data and using the interferometry technique, DSMs were generated over test area Bucharest. Using SPOT HRS as reference model, the accuracy of the DSMs was analyzed. 2. TEST AREA Test area is represented by the city of Bucharest (Fig. 1 yellow rectangle). The DSMs analysis was performed over a subset test area of 5 km x 4 km (Fig. 1 red rectangle). The coordinates of the subset test area in UTM projection, zone 35N, WGS84 ellipsoid are: E, N for the upper left point and E, N for the lower right point. Figure 1. Multi-temporal image of Bucharest created based on 3 ENVISAT images ESA 2009 (red: 17/09/2005, green: 29/06/2008, blue: 07/12/2002) 3. METHODOLOGY 3.1. Input data ERS images used in this study were acquired during the Tandem mission (1 day time interval), while the ENVISAT ASAR I2 images present a time interval of 735 days between the two acquisitions. Both ERS and ENVISAT data are from track 465 (VV polarization). ERS images are acquired on a descending orbit while ENVISAT images from an ascending one. TerraSAR-X data (one pair of StripMap images and one pair of High Resolution Spotlight images) were acquired from an ascending orbit, track 9, having HH polarization. In case of TerraSAR-X StripMap (TSX SM) images the range bandwidth is 100 MHz and the resolution is 1.8 m (slant range) x 3.0 m (azimuth). TerraSAR-X High Resolution Spotlight (TSX HS) images were acquired using a range bandwidth of 300 MHz. TSX HS images have 0.6 m resolution in slant range and 1.1 m resolution in azimuth. The characteristics of these interferometric pairs are presented in Tab. 1. Proc. Fringe 2009 Workshop, Frascati, Italy, 30 November 4 December 2009 (ESA SP-677, March 2010)

2 Interferometric pair ERS Tandem ENVISAT ASAR TerraSAR-X SM TerraSAR-X HS Table 1. Characteristics of ERS, ENVISAT and TerraSAR-X data Date Track Orbit Polarization Incidence Perpendicular Height of angle baseline [m] ambiguity [m] desc. VV desc. VV asc. VV asc. VV asc. HH asc. HH asc. HH asc. HH The height of ambiguity is defined as the altitude difference that generates an interferometric phase change of 2π and it is inversely proportional to the perpendicular baseline [3]. The height of ambiguity was calculated for each interferometric pair using the wavelength of the carrier wave, the range of the target, the look angle and the perpendicular baseline Interferometric processing The interferometric processing of ERS, ENVISAT and TerraSAR-X data has basically followed the same steps, with few exceptions: application of precise orbits (ESA FTP site) only in the case of ENVISAT data. For ERS Tandem images, the precise orbits were not available. coarse and fine coregistration - was performed by evaluation of the cross-correlation measurements based on the complex input data. The slave image was resampled over the master image. The precision of coregistration is very important for the phase quality of the interferogram [4]. synthetic interferogram generation - based on the orbital parameters and the reference DSM (SPOT HRS); the resolution of the reference DSM is 22 m at the mean latitude of the test area (30 m resolution at the equator). interferogram calculation - by cross multiplying, pixel by pixel, the first SAR image with the complex conjugate of the second [3]. The noise that affects the interferogram was reduced by averaging adjacent pixel in the complex interferogram using a multi-looking factor of 2x10 (2 looks in range, 10 looks in azimuth) for ERS and ENVISAT data, 12x12 in case of TSX SM and 20x20 for TSX HS data. interferogram filtering - in order to improve the phase signal-to-noise ratio (SNR); a filter of 0.4 (where 0 represents no filtering and 1 maximum filtering) was applied using a window of 64 pixels. differential interferogram generation - by subtracting the synthetic interferogram from the filtered one. generation and analysis of the coherence maps - within a pixel, the difference in phase between two complex SAR images can be translated into a combination of contributing factors like topography, ground displacement, atmosphere and noise [5]. Coherence is a self-validating indicator of the phase measurement which depends on the proportion of useful signal to non-useful signal [4]. Thus, the phase noise can be estimated by means of the local coherence γ and it represents the crosscorrelation coefficient of the SAR image pair estimated over a small window once all the deterministic phase components are compensated for [3]. The highest coherence values were obtained in case of TerraSAR-X data (0.44 average value) and the lowest in the case of ENVISAT data (0.14 average value). phase unwrapping - was executed using the MCF (Minimum Cost Flow) algorithm. geometry optimization - by measuring a number of ground control points (GCPs). The (x, y) coordinates of the GCPs were extracted from the digital orthophotos (0.5m spatial resolution), Stereographic '70 Projection, Pulkovo 1942 datum, Krassovski 1940 ellipsoid. These coordinates were transformed from the Stereographic '70 Projection into the CRS ETRS89 system (Coordinate Reference System - European Terrestrial Reference System), GRS80 ellipsoid. The CRS ETRS89 coordinates, ellipsoid GRS80 were considered to be identical with the geographic coordinates, WGS84 datum, and WGS84 ellipsoid. Next, these coordinates were transformed into UTM projection system, zone 35 0 N, WGS84 datum, and WGS84 ellipsoid. For the Z coordinate, a change of the SPOT HRS DSM vertical datum was performed from EGM96 (World Wide 15-Minute Geoid Height) to WGS84. That means that the orthometric heights (EGM96 datum) were transformed into ellipsoidal heights (WGS84 datum). The interferometric baseline was adjusted based on the GCPs and the differential interferogram and the unwrapped phase were computed again. phase to height conversion the generated DSMs are in UTM projection system, zone 35 0 N, WGS84 datum, WGS84 ellipsoid (ellipsoidal heights). In order to fill in the gaps of information, the DSMs were interpolated (method: bilinear interpolation)

3 using a regular grid with a spacing of 25 m for ERS and ENVISAT, 12 m for TSX SM (accordingly to the specifications of the future TanDEM-X Mission) and 5 m for TSX HS Editing and filtering of DSMs The DSMs generated in the previous paragraph were edited for the purpose of removing their artefacts. For each DSM, an image containing the height difference between the analyzed model and the reference model (SPOT HRS) was generated. The values of this differential model that were not belonging to the normal distribution interval were replaced with values from the reference model. This operation was performed using band mathematics. Next, a median filter was applied to each DSM in order to obtain a smooth surface. The median filter smoothes the surface and preserves the edges. Within a window, the value of each center pixel is replaced with the mean value of the neighbored pixels DSM analysis The accuracy of the DSMs was analyzed against the SPOT HRS reference DSM (22 m pixel spacing). The analysis was executed using the DEMANAL software, Program System BLUH, using two iterations. The test area (illustrated in the colored orthophoto, copyright National Agency for Cadastre and Land Registration - NACLR) and the DSMs used in the comparative analysis are presented in Fig. 2 - Fig. 7. In the comparative study, a DSM (15 m resolution) derived by optical stereoscopic processing of SPOT 5 HRG data (acquired in a time interval of 1 day, 0.51 base-to-height ratio) and also the SRTM DSM (71 m pixel spacing at the latitude of the test area) have been added (Fig. 8 Fig. 9). Figure 3. SPOT HRS DSM ( SPOT IMAGE 2007) 22 m resolution, overlaid on the orthophoto Figure 4. ERS TANDEM DSM 25 m resolution, generated based on ERS data ( ESA 2009), overlaid on the orthophoto Figure 2. Subset test area Bucharest (colored orthophoto all rights reserved to NACLR) Figure 5. ENVISAT DSM 25 m resolution, generated based on ENVISAT data ( ESA 2009), overlaid on the

4 Figure 6. TSX SM DSM - 12 m resolution, generated based on TSX SM data ( DLR 2008), overlaid on the Figure 7. TSX HS DSM - 5 m resolution, generated based on TSX HS data ( DLR 2008), overlaid on the Figure 8. SPOT HRG DSM - 15 m resolution, generated based on SPOT HRG data ( CNES 2009, distribution SPOT IMAGE S.A.), overlaid on the orthophoto Figure 9. SRTM DSM ( Jarvis A., H.I. Reuter, A. Nelson, E. Guevara, 2008, Hole-filled seamless SRTM data V4, International Centre for Tropical Agriculture, available from cgiar.org) 71 m resolution, overlaid on the orthophoto 4. RESULTS The results (DEMANAL second iteration) of the DSMs analysis are presented in Tab. 2. The analysis of the DSMs generated through interferometry shows that best results were obtained in case of ERS Tandem DSM (25 m resolution), with a ± 4.75 m root mean square error of the Z heights (RMSZ). These very good results are due to the very short acquisition interval of only one day and the value of perpendicular baseline ( 229 m) which is optimal for DEM generation. The analysis of TerraSAR-X SM DSM reveals a RMSZ of ± 6.60 m. The results are slightly better in case of ERS Tandem DSM because the perpendicular baseline of the TSX SM interferometric pair is very short ( 55 m), being more suitable for differential interferometry. Consequently, the height of ambiguity is very large ( 125 m), thus the phase is not very sensitive to the topography. Also, TerraSAR-X SM data were acquired in a time interval of 44 days. The case of TerraSAR-X HS DSM data is very similar with the one of TerraSAR- X SM DSM: a RMSZ of ± m due to a very short perpendicular baseline ( 46 m) and a very high height of ambiguity ( 144 m). The advantage of the TerraSAR-X DSMs is given by their higher resolution: 12 m for TerraSAR-X SM DSM and only 5 m for TerraSAR-X SM HS. In case of ENVISAT DSM (± 7.93 m RMSZ) the perpendicular baseline ( 153 m) is optimal for DEM generation, but the time interval of more than 2 years led to coherence loss and reduced the accuracy of the height values. Moreover, the model does not accurately represent the characteristics of the terrain and its artefacts could not be eliminated even after editing and filtering. The resolution of ENVISAT DSM is 25 m.

5 The results obtained in the case of the DSM generated by means of optical stereoscopy (using SPOT HRG data) indicate a RMSZ of ± 3.46 m. The results are better than the ones obtained using SAR interferometry because there are no geometric effects like shadow, layover or foreshortening. Moreover, the SPOT HRG stereoscopic pair has only one day time interval between the acquisitions, thus very good results were obtained with the automatic image matching. SRTM DSM analysis in comparison with SPOT HRS reference DSM shows a RMSZ of ± 2.20 m. Table 2. Results of DSMs analysis DSM analysis RMSZ [m] Bias [m] RMSZ without bias [m] HRS ERS TANDEM 4,75-0,08 4,75 HRS ENVISAT 7,93-3,71 7,01 HRS TSX SM 6,60 0,65 6,57 HRS TSX HS 11,80-1,15 11,74 HRS SPOT 5 HRG 3,46-2,09 2,76 HRS SRTM 2,20-0,18 2,19 5. CONCLUSIONS A comparative study regarding the accuracy of the interferometric DSMs have been performed. Best results were obtained with ERS Tandem data due to the very short revisiting time and a perpendicular baseline optimal for interferometry (InSAR). As expected, due to the very large acquisition period of almost two years for the ENVISAT interferometric pair, the DSM generated based on these data is not accurate and it presents strong artefacts. The suitability of TerraSAR-X data for DSM generation over an urban area has been tested. The small perpendicular baselines of both StripMap and High Resolution Spotlight interferometric pairs suggested that TerraSAR-X data were more suitable for an application of differential interferometry (DInSAR) or persistent scattereres interferometry (PSI). Moreover, the High Resolution Spotlight images showed a very strong pattern of point scatterers that could be successfully used in a PSI application. For DSM generation, the future mission TanDEM-X will acquire interferometric pairs of images with optimal perpendicular baselines and no temporal decorrelation. ACKNOWLEDGMENTS The ERS and ENVISAT data used in this study were kindly provided by the European Space Agency (ESA) under the ESA Category-1 Proposal ID The TerraSAR-X images were acquired under the German Aerospace Center (DLR) TerraSAR-X Project, Pre-launch Proposal ID LAN_0130 and SPOT data in the framework of the Centre National d Études Spatiales (CNES) ISIS Project no 181. DSM analysis was performed with DEMANAL software (Program System BLUH), copyright of Prof. Dr. Eng. Karsten Jacobsen, Institute of Photogrammetry and GeoInformation, Leibniz University Hannover. REFERENCES 1. Levin, N. (1999). Fundamentals of Remote Sensing, 1 st Hydrographic Data Management Course, Imo International Maritime Academy, Trieste, Italy; Remote Sensing Laboratory, Geography Department, Tel Aviv University, Israel 2. Jacobsen, K. (2003). DEM Generation from Satellite Data, Online at Jac _03DEMGhent_red.pdf 3. Ferretti, A.; Monti-Guarnieri, A.; Prati, C.; Rocca, F. & Massonnet, D. (2007). InSAR Principles: Guidelines for SAR Interferometry Processing and Interpretation, ESA Publications, ISBN , ESTEC, The Netherlands 4. Massonnet, D. & Souyris, J. C. (2008). Imaging with Synthetic Aperture Radar, EPFL Press / CRC Press, ISBN (EPFL Press), ISBN (CRC Press), USA 5. Teleaga, D.; Poncos, V.; Dana, I. F.; Nedelcu, I. & Olteanu, V. G. (2009). Urban Infrastructure Monitoring Using Spaceborne Interferometric Synthetic Aperture Radar Techniques. In Proc. of the 1 st International Conference on Space Technology, Thessaloniki, Greece