GPS and GIS Assisted Radar Interferometry

Transcription

1 GPS and GIS Assisted Radar Interferometry Linlin Ge, Xiaojing Li, Chris Rizos, and Makoto Omura Abstract Error in radar satellite orbit determination is a common problem in radar interferometry (INSAR). For example, when we try to locate a radar test site with known geographic coordinates using the geocoding information in SLC (the latitude and longitude of the four image corners), the location is well away from the true position. Another example is when there is a significant disturbance in the differential INSAR result, we sometimes are not sure whether it is from ground deformation or atmospheric heterogeneity. Even after these are corrected, we need to export the INSAR results to a GIS format so that they can be overlaid as layers over orthophotos and mine plans (in the case of mining subsidence) in order to interpret the results. Therefore, it is proposed to use both GPS and GIS to assist radar interferometry. Results are presented with an application to monitoring subsidence due to underground mining southwest of Sydney, Australia. Introduction The fact that no satellite has been launched so far for the purpose of radar interferometry (Graham, 1974) has lead to the inaccurate orbit determination for satellites such as JERS-1, ERS-1 and ERS-2, inorder to derive topography and topographic change information from radar images. For example, we had to locate our radar test site southwest of Sydney (Figure 1), using both the known GPS coordinates of the site and the geocoding information in the single look complex (SLC) data (i.e., the latitude and longitude of the four image corners). There are three collieries at the test site. Because the geodetic coordinates of the test site and the SAR images are acquired independently, especially the coordinates of the test site are obtained using e.g., GPS, which is external to the SAR imaging process, this method is termed as external location. Results of the external location on both C- (ERS-2) and L- (JERS-1) band SAR images are shown in Figure 2. On the other hand, the test site was also located based on the ground features (e.g., rivers, railways, and small towns) in the SAR image and on the map for the same region. Since the relative locations of the features within the same radar image are used, this method is termed as internal location. The location of a colliery using this method is given in Figure 3. Comparing Figures 2 and 3, it can be seen that the location determined by the external method is well away from the true position given by the internal method, as can be told from the features in the images. Also, when we check the coordinates of the same feature in the INSAR results and on the map, there is significant discrepancy. This is because the SLC data is produced in the data transcription using the satellite Linlin Ge and Chris Rizos are with the School of Surveying & Spatial Information Systems, University of New South Wales, Sydney NSW 2052, Australia (l.ge@unsw.edu.au). Xiaojing Li is with the School of Electrical Engineering & Telecommunications, University of New South Wales, Sydney, NSW 2052, Australia. Makoto Omura is with the Department of Environmental Science, Kochi Women s University, Kochi , Japan. orbit information without any information of ground control point. Still another example is, when there is a significant disturbance in the differential INSAR (DINSAR) result, we sometimes are not sure whether it is from ground deformation or atmospheric heterogeneity. Figure 4 shows a significant disturbance detected in the differential INSAR result, which was proved to be due to a cold front by comparing with the observation of a weather radar system (Hannssen, et al., 1999). Even after we get these all corrected, we often find ourselves busy either swapping between the geocoded master image and the interferogram (or height/height change image) or reading the latitude and longitude of a pixel in order to link the INSAR result to the real world. Hence, there is an urgent need to export the INSAR results to a GIS format so that we can overlay them as layers over GIS data such as orthophotos and mine plans (in the case of mining subsidence), in order to interpret the results. Therefore, it is proposed in this paper to use both GPS and GIS to assist radar interferometry. Using GPS to Georeference INSAR Results Corner reflectors (Figure 5) or permanent scatterers (natural reflectors such as the twin towers in Figure 6) are surveyed using GPS receivers so that they are well-defined not only in the image domain with precise image coordinates (row/column), because they are very bright in the radar image, but also in the object domain with precise geographic coordinates (latitude/longitude/height). These image and geographic coordinates are used to establish a more accurate and realistic georeferencing for the geocoded INSAR products (i.e., DEM and deformation image). Assume the image coordinates of a reflector are (Irow, Icol) while its geographic coordinates are (Glat, Glon). Then, the transformation between the two coordinates in the same projection referenced to the same datum can be expressed as [Glat Glon] [Irow Icol 1] (1) or L BX, (2) where, for multiple reflectors, L can be extended to a matrix of GPS coordinates, B a matrix of image coordinates, and X remains to be a 2 3 matrix of transformation coefficients. Therefore, at least three (corner or natural) reflectors have to be identified in the radar image in order to solve for the six transformation coefficients. As a matter of fact, six corner reflectors as shown in Figure 5 were deployed in our test site. When there are more reflectors available, least square estimation can be introduced into the data analysis. Assuming a 1 b 1 c 1 a 2 b 2 c 2 Photogrammetric Engineering & Remote Sensing Vol. 70, No. 10, October 2004, pp /04/ /$3.00/ American Society for Photogrammetry and Remote Sensing PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING October

2 Figure 1. Location of Test site. Figure 4. The effect of a cold front on: (a) the DINSAR result; and as measured by a weather radar system. Figure 2. Results of the external location on both C- (ERS-2) and L- (JERS-1) band SAR images. (a) Figure 3. Results of the internal location on both C- and L- band (images copyright of ESA and NASDA). the vector of observation errors is, then the observation equation is L B X. (3) Therefore, the least square estimation (Giordano and Hsu, 1985) of the transformation coefficients is Xˆ (B T P B) 1 B T P L, (4) where P D 1, D is the variance matrix of. Figure 5. Corner reflector (a) imaged by ERS-2 radar (18 July 2002) October 2004 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

3 (a) Figure 7. An INSAR-derived DEM after applying GPS corrections. Figure 6. Natural reflector as a GPS ground control point, the twin tower on: (a) aerial photo and geocoded master image. The next step using GPS is to check and correct the INSAR product. With the horizontal coordinates (latitude and longitude) corrected in the last step, we focus on the third component, i.e., the height in DEM. In addition to the reflectors, GPSsurveyed geodetic points are also used. Both reflectors and geodetic control points are located or identified in the DEM using their horizontal coordinates. Their heights are then read from the DEM and compared with GPS heights. The DEM is then shifted or rotated to minimize the difference between the two heights. Figure 7 gives an INSAR-DEM derived from ERS tandem data after applying such corrections (master image: (acquired by ERS-1 on 29 October 1995); slave image: 2761 (acquired by ERS-2 on 30 October 1995)). The approach to deal with the third component in differential INSAR (DINSAR) result (height change) is very different. As shown in Figure 4, the tropospheric delay heterogeneity could be very misleading in the DINSAR result. Hence, GPS data are used to estimate the differential tropospheric delay among the GPS stations in the same radar image. These differential delays are then interpolated to generate an image that can be used to correct the atmospheric disturbance in DINSAR result on a pixel-by-pixel basis (Ge, 2000). GIS Assisted Interpretation of INSAR Results When GPS corrections have been done, the INSAR/DINSAR products can be exported to GIS format for interpretation. It is also useful to import some GIS data (e.g., orthophotos) to the INSAR format, for example, the multiple file format (MFF) for Atlantis EarthView INSAR software, so that the GIS data can be used to assist GPS corrections. Figure 8 shows the display in ArcMap (a core application of the ArcGIS Desktop) of a differential INSAR result over a period of 132 days from 9 November 1993 to 21 March Both the master and slave images were acquired by the L- JERS-1 satellite. The INSAR-DEM derived from ERS tandem data as shown in Figure 7 was used to remove the topographic fringes in the interferogram. The interferogram was then phase-unwrapped and converted to a height-change image. This image was enhanced and imported to GIS as a layer. The geocoded master image was also imported to GIS in order to double check the georeferencing. In addition to these two layers, mine plan and aerial photo of the area have also been included in the GIS. From Figure 8 it can be seen that four regions have experienced significant subsidence, two in the first colliery (Figure 8b), one in the second colliery (Figure 8c), and one in the third colliery where the mine plan is not available. It seems the extent and magnitude of subsidence in the second colliery is much larger that those in the first colliery, probably due to different geological settings. By checking the subsidence against mining progress given in Tables 1 and 2 (for example, in Figure 8b the longwall LW25 start date is 27 January 1994, and the slave imaging date is 21 March 1994), it can be concluded that the delay between mining and subsidence will be less than two months, if any. Figure 9 shows the display in ArcMap of a differential INSAR result over a period of 44 days from 21 April 1995 to 04 June Both the master and slave images were acquired by the JERS-1 satellite. Again the INSAR-DEM derived from ERS tandem data was used to remove the topographic information. Two regions have experienced significant subsidence, one in the first colliery (Figure 9b), and one in the second colliery (Figure 9c). Again it seems the extent and magnitude of subsidence in the second colliery is much larger that those in the first. By checking the subsidence against mining progress given in Tables 1 and 2 (for example, in Figure 9b the master and slave imaging dates 21 April 1995 and 4 June 1995 fall between the longwall LW26 start and finish dates 8 December 1994 and 5 August 1995, respectively), it can be concluded that the DINSAR result has picked up the correct longwall responsible for the subsidence. Comparing the results between Figures 8 (height change in mm) and 9 (height change in cm), the extent of subsidence is larger for the longer time span (132 days). The magnitude, PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING October

4 (a) (a) (c) Figure 8. Differential INSAR result over the period of (JERS-1, 132 days): (a) Overall result Zoom-in the first Colliery (c) Zoom-in the second Colliery. (c) Figure 9. Differential INSAR result over the period of (JERS-1, 44 days): (a) overall result Zoom-in the first Colliery (c) Zoom-in the second Colliery October 2004 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

5 TABLE 1. MINING ACTIVITIES AT THE FIRST COLLIERY Longwall Start (YYYYMMDD) Finish (YYYYMMDD) LW LW LW LW LW28a LW28b TABLE 2. MINING ACTIVITIES AT THE SECOND COLLIERY Longwall Start (YYYYMMDD) Finish (YYYYMMDD) however, is almost the same, especially in the second colliery which means the subsiding process can finish in about 44 days directly above the longwall. If the mining methods used at the two collieries are the same, the DINSAR results also imply second colliery is more susceptible to subsidence. From the induced subsidence, it can be deducted that in the first colliery mining is progressing from top to bottom one longwall after another (Figure 8b versus Figure 9b) while mining in each longwall starts at the top-right (northeast) end and finishes at the bottom-left (southwest) end (Figure 8b), even if information about mining activities given in Tables 1 and 2 is not available. In the second colliery, mining is progressing from bottom to top one longwall after another (Figure 8c versus Figure 9c) while the mining direction within each longwall cannot be inferred. With the DINSAR results already in the GIS format, many more map-quality products can be generated. For example, the subsidence can be displayed in 3D; many profiles across the subsidence area can be produced. Figure 10 shows subsidence profiles derived from the JERS-1 DINSAR result for the second colliery. Such precisely georeferenced and well-archived DINSAR results will have a big impact on the mining industry. In the established coal mine fields of eastern Australia, it is becoming increasingly difficult to select underground mine sites which avoid major engineering structures, both on the surface and underground (highways, bridges, buildings, abandoned underground workings). The third colliery, also covered in the radar image, is a representative example, where the surface topography overlying the mine consists of several steep-sided river gorges. The surface is traversed by a freeway which crosses one of the gorges. Consequently, a major surface subsidence monitoring program has been in place for several years, including intensive conventional, GPS and EDM surveying (once or more than once a week) plus real-time monitoring of critical components of the bridge structure. However, current subsidence monitoring techniques are relatively time-consuming and costly. Hence, the monitoring Figure 10. Subsidence profiles derived from the JERS-1 DINSAR result for the second colliery. LW LW LW LW LW is usually constrained to very localized areas, and it is very difficult to monitor any regional deformation induced by underground mining. In addition, even in the localized area, the monitoring points are not usually close enough to assist in understanding the mechanisms involved in ground subsidence. The integrated INSAR-GPS-GIS technique demonstrated here can both be accurate and give a fine spatial characterization of the ground deformation so that it can be used to complement current techniques to measure on a cost-effective basis, on a regular basis, the ground subsidence due to underground mining. Conclusions The importance of using GPS and GIS to assist INSAR data analysis is discussed. An application of the integrated INSAR-GPS-GIS technique to monitoring subsidence due to underground mining southwest of Sydney, has been demonstrated. The integrated technique can be used as a viable operational tool for ground subsidence monitoring to complement current geodetic techniques. Acknowledgments The authors thank Messrs. Yufei Wang and Lijiong Qin of UNSW for assistance in GIS, and the ACRES (the Australian Centre for Remote Sensing) for providing SAR images and satellite programming. References Ge, L., Development and Testing of Augmentations of Continuously-Operating GPS Networks to Improve their Spatial and Temporal Resolution, Ph.D. Dissertation, University of New South Wales, Sydney, NSW 2052, Australia, 230 p. Giordano, A., and F. Hsu, Least Square Estimation with Applications to Digital Signal Processing, New York, Wiley, 412 p. Graham, L.C., Synthetic interferometer radar for topographic mapping, IEEE Transactions on Geoscience and Remote Sensing, 62: Hanssen, R.F., T.M. Weckwerth, H.A. Zebker, and R. Klees, High-resolution water vapor mapping from interferometric radar measurements, Science, 283: PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING October