Remote Sensing: Mapping a Broken Land
Using repeat pass, space-based SAR interferometry, researchers measure
effects of the Kobe, Japan earthquake.
By Jim Ehrismann, Bernard Armour, Marco van der Kooij and Harald Schwichow

Imagine a remote sensing technique so sensitive that it can measure centimeter shifts in the Earth's surface from a distance of over 800 km in space. What if this technique were so versatile that it could routinely produce topographic maps on a regional scale with a spatial resolution of tens of meters and a height accuracy of a few meters, as well as measure thematic information and super change detection? Is this just dreaming? Recent advances in repeat pass space-based synthetic aperture radar (SAR) interferometry have made these goals not only possible, but accessible to the remote sensing community.
      There is a broad range of SAR interferometry applications. These may be grouped into four areas: SAR imagery corrected for all geometry and brightness errors (precision calibration), elevation mapping creating digital elevation models (DEMs), changes in elevation mapping (deformation), and super-sensitive change detection. Among the most exciting new applications of SAR interferometry is the two-dimensional mapping of large scale surface deformation with very high sensitivity (to within millimeters!). These deformation maps are used to detect and monitor Earth disasters including land deformation due to earthquakes, crustal movements, land slides, volcanic activity (subsidence and absidence), and mining and water extraction activities (land subsidence).
      To exploit SAR interferometry, two basic ingredients are required. The first is the acquisition of two phase preserved SAR images of the same region but obtained at slightly different viewing angles. These images are produced by the space-based SARs RADARSAT, ERS-1, ERS-2 and JERS-1 and may be purchased at a nominal cost. The second requirement is a set of image processing tools that combine the two SAR images to create an interferogram and extract information from it. This tool set may be developed in-house or purchased from a commercial supplier.

Repeat-Pass SAR Interferometry
Interferometry is a process of using interference effects to determine lengths or changes in lengths very accurately. This technique has some similarities to stereo-optical imaging in that two images of the same area, viewed from different angles, are appropriately combined to extract the topographical information. A major difference between interferometry and stereo imaging lies in the painstaking manual effort required in obtaining topography from stereo-optical images. Inherent in SAR data is distance information that enables the automatic generation of topography using interferometry. This means that digital elevation models (DEMs) can be generated using SAR interferometry with greater automation than optical techniques. Also, using a technique known as differential SAR interferometry, surface deformations can be measured with an accuracy and spatial measurement density which is unprecedented. Add to this the ability of a SAR to penetrate clouds and provide day and night operation, and it becomes clear that SAR interferometry has definite advantages over conventional mapping techniques.
      To understand SAR interferometry we must first understand that SAR measures distance information as well as radar "brightness" information. The distance information is encoded in something called phase. Broadly speaking, phase is just a fine scale measurement of distance which we measure twice in two separate SAR passes. Technically speaking, the phase difference between the two passes at corresponding pixels, allow a measurement of the incidence angle of the incoming radiation. Distance combined with incidence angle and with the location of the SAR platform on each of the two passes gives a three-dimensional localization of points on the Earth's surface. Either way, we require two SAR images, each containing brightness and phase information, to produce one interferometric data set from which height and other information is extracted.
      In repeat-pass SAR interferometry a spaceborne SAR orbits the Earth repeatedly. With respect to the Earth, the satellite goes through a cycle of orbits, for example, every 24 days in the case of RADARSAT. This means the satellite returns almost exactly to the same position with respect to the Earth after 24, 48, 72, etc. days. On the first pass of the satellite, a SAR image is acquired. After a period of time (one or more orbit repeat cycles), the same area may be imaged again, thus acquiring the second SAR image.
      The two SAR images are individually processed by a precision, phase preserving, SAR processor to convert the "raw" radar signals into the familiar black and white SAR images with the phase information carefully preserved. Sophisticated SAR interferometric processing is then applied to register the two SAR images to an accuracy of 1/8th of a pixel. Next, the phase differences in the two images are calculated by subtracting the phase in one image from the phase in the other. The result is called an interferogram. Theses phase differences in the interferogram wrap around in cycles of 360 degrees and must be unwrapped to obtain the absolute phase. The problem is similar to knowing the minutes in the hour, but not hours in the day, and we must determine the time of day. It suffices to say that phase unwrapping is a complicated process. After phase unwrapping, the absolute phase is converted to the various data products through further processing.

Special Considerations
Four important facts should be noted. First, the two SAR acquisitions should be repeated as closely as possible spatially, temporally and physically. As a result it is usually only possible to combine two SAR images from the same sensor, or from two very similar sensors (e.g. ERS-1 and ERS-2).
      Second, from an orbit geometry point of view, satellite orbits exhibit a small degree of drift such that they do not return to the exact location on subsequent orbit repeats. These repeats are generally parallel and separated by a distance (called the baseline) on the order of a few hundred meters. This baseline between passes provides the different viewing angles necessary for interferometry to work.
      Amazingly, this inaccuracy in the satellite orbit repeatability provides the ideal condition for interferometry!
      Third, the terrain being imaged may change between passes of the SAR. For example, ice floes may drift, Earth tectonic plates may shift, a city may subside, a rainfall or snowfall may occur, or simply a farmer may harvest some crops. These variations result in a change in the phase information, called temporal decorrelation. Temporal decorrelation implies that the phase contains not only distance information but also thematic change information. We handle this in two ways. We subtract out the previously known distance information (using for example a pre-existing digital elevation model), to measure the fine changes in height. To obtain thematic change information, we measure the random differences in the phase between the two SAR images using simple statistical correlation techniques. If we do not want to have any thematic change information, then the two SAR images must be repeated very closely in time. The ERS-1/ERS-2 tandem mission operating until May 16, 1996 has a very short time between orbit passes (about 24 hours) and is ideal for minimizing temporal decorrelation.
      Fourth and finally, as with all SAR imagery, high spatial resolution and control of incidence angle greatly influences the utility of the imagery for a particular application. The high resolution of the RADARSAT fine beam mode, combined with the ability to select a suitable incidence angle, makes this mode the most promising of all the RADARSAT modes for producing interferometric SAR data products.

The Kobe Earthquake: Jan. 17, 1995
On Jan. 17, 1995 an earthquake struck the city of Kobe, Japan and the surrounding area. This incident resulted in over 40,000 injuries and 5,000 deaths, devastating the people of that region. In general, Japan suffers a large number of geological and other natural disturbances. The National Research Institute for Earth Science and Disaster Prevention (NIED) is one of several Japanese government agencies responsible for monitoring and investigating such occurrences in order to better prepare and cope with these disasters.
      Given Atlantis Scientific's interest in SAR processing and SAR interferometry, NIED asked Atlantis to measure the effects of the Kobe earthquake using differential interferometry. NIED provided two JERS-1 SAR data sets covering the Kobe area, acquired approximately one and a half years apart. The earthquake struck shortly before the second data set was collected. Also, a digital elevation model (DEM) of the Kobe area, obtained from the Geological Survey Institute of Japan, was used to subtract out the topographic elevation information.
      The JERS-1 SAR data sets were carefully processed to form phase preserved SAR images, as described in "Desk-Top SAR Processing: No Longer the Best Kept Secret," EOM March 1996. Figure 1 shows the interferogram of Awaji Island, located south of Kobe. The earthquake epicenter was located just to the northeast of the island. Although there are many noisy regions due to thematic changes, definite patterns can be seen on the island. This is quite remarkable given the fact that the images were taken nearly one and a half years apart and represent different seasons. Many changes have likely occurred which contribute to information loss yet accurate patterns still persist. Toward the bottom of the interferogram, away from the affected region of the earthquake, the patterns can be interpreted as elevation contours. The colors cycle from red to blue to green and back to red with increasing height. Near the tip of the island it is suspected that the patterns represent a combination of topography and earthquake effects.
      To disentangle the topographic patterns from the surface deformation patterns, the DEM was used to simulate an interferogram (Figure 2). The elevation contours are quite clear and similarities to Figure 1 are evident. A simple combination of this interferogram and that of Figure 1 produced the differential interferogram shown in Figure 3. The major patterns which remain represent land deformation due to the earthquake, including a major fault parallel to the coastline in the northeast corner of the image. The maximum deformation near Kobe is estimated to be a horizontal shift of about 1.5 meters toward the southwest, the result of a shift along the major fault running in the direction of the fringe lines. The maximum deformation on Awaji Island is estimated as nine color cycles (measured at near the northern tip relative to the homogeneous midsection of the island) corresponding to a displacement of 1 meter towards the satellite.

Interferometric SAR (InSAR) Processing Tools
Easy to use software tools are required to facilitate the production of high quality map products using interferometric techniques. This software would provide researchers and operational users with the capability of easily creating interferograms as well as generating DEMs, surface deformation maps, and more advanced data products. Features such as DEM mosaicking are needed to produce large and accurate topographical maps that can be regularly updated. As well, InSAR processing tools are needed to produce calibrated geocoded SAR images, terrain slope maps, incidence angle maps, and coherence images.
      The Kobe differential interferograms were produced using specialized software tools. Atlantis is currently developing InSAR processing software intended for desktop computers. The development of the InSAR Workstation is being funded by the RADARSAT User Development Program of the Canadian Space Agency with technical supervision by the Canada Centre for Remote Sensing, Geomatics Canada.
      SAR interferometry has the potential to lead to a revolution in the process of map-making and surface deformation monitoring. The availability of near global coverage of spaceborne SAR data, the low cost of the data as compared to airborne sensors, and the near automated procedures available to create InSAR products are certain steps toward satisfying the needs of the geographic and Earth observation users worldwide.

About the Authors:
Jim Ehrismann, Bernard Armour, Marco van der Kooij and Harald Schwichow are members of the InSAR Workstation development team at Atlantis, Ottawa, Ontario. They share an active interest in developing new applications for spaceborne SAR imagery and making these applications accessible to the user community. They may be reached at [email protected]

Back