Each year, earthquakes cause thousands of deaths and billions of dollars in property damage worldwide. These numbers are bound to grow as populations swell and are increa-singly concentrated along coast lines and mountain ranges, which often represent tectonically active plate boundaries zones (Bilham, 1991). Concerns about earthquake hazards have been formulated time and again and documented in the U.S. and throughout the world. Especially at risk are critical facilities, such as conventional and nuclear power plants, hazardous waste repositories, power transmission grids, transportation systems, schools, hospitals, and so forth.
The January 17, 1994 Northridge, California earthquake is now estimated to have been the most costly seismic disaster in U.S. history. It caused $20 billion in immediate property damage and in lost business during the months that followed. Lost business was due to the reduced transportation infrastructure (mainly bridge and freeway collapse) and associated delays, closed buildings and factories, lost production, increased insurance rates, and the uncertainty and loss of confidence by the business community. Even a moderate event near downtown Los Angeles during business hours, or a great earthquake along the San Andreas Fault in southern California could lead to much more severe losses of life and property, dealing a devastating blow to the socioeconomic structure of the state of California. The recent Kobe earthquake in Japan hints at the magnitude of disaster when a large earthquake strikes an urban center. With over 5000 fatalities and damage estimates exceeding $100 billion, it may rank as the most costly earthquake in modern history.
Even though seismic zones have been fairly well mapped in this country and worldwide over the past half century, society as a whole remains poorly prepared. Even California, which has fairly rigorous and well-enforced building codes, has suffered devastating long-term economic losses from moderate-size earthquakes in the last five years (Loma Prieta and Northridge). One problem is that earthquakes occur infrequently in any given location, confounding efforts to maintain a high level of preparedness. While this is largely a political/social problem, there are scientific issues as well.
In spite of decades of study and million of dollars of research funding, we do not know in advance when or even where the next earthquake will strike. In some areas we know where they will probably occur, but in many places we do not, either because the faults are hidden (such as the "blind thrusts" associated with the 1983 Coalinga, 1987 Whittier Narrows, or 1994 Northridge earthquakes) or because the tectonic regime is poorly understood (such as the New Madrid seismic zone in the midwestern U.S.). In some instances, earthquakes strike where there is no historic record of any significant seismicity--for example, the M = 4.6 earthquake near Reading, Pennsylvania in January, 1994 and the September, 1993 earthquake near Killari, India, which killed approximately 10,000 people. In order to reduce the loss of life and property in earthquakes, we need to learn as much as possible from every earthquake that occurs. This implies the need to study earthquakes on a global basis, and also suggests the importance of space-based techniques.
SAR can provide high-resolution imagery of earthquake-prone areas, high-resolution topographic data, and a high-resolution map of coseismic deformation generated by an earthquake. Of these, the last one is probably the most useful, primarily because it is unique. Other techniques are capable of generating images of the Earth's surface and topographic data, but no other technique provides high-spatial-resolution maps of earthquake deformation.
Crustal deformation is a direct manifestation of the processes that lead to earthquakes. Consequently, it is one of the most useful physical measurements we can make to improve estimates of earthquake potential. Figure 3.1 summarizes the temporal and spatial scales of deformation attached to various earthquake-related processes. In particular, it shows that the temporal scales associated with the pre-, co-, and interseismic portions of the earthquake cycle are very different. Figure 3.2 compares the operative temporal and spatial scales accessible through some of the techniques used to measure deformation. Space geodetic measurements initiated under NASA's Crustal Dynamics Project (CDP) continue to be collected as part of a follow-on program, named Dynamics of the Solid Earth (DOSE). Many of these measure-ments span tectonic plates and the plate boundary zones where strain accumulates and earthquakes occur (e.g., Dixon, 1993). In terms of hazard prevention, preparedness, and/or relief, geodetic measurements can be used most directly to assess the hazard of a given region.
Unfortunately, current geodetic networks woefully undersample the deformation field, especially in terms of spatial distribution. A critical aspect of crustal deformation is that, as we are able to refine our spatial sampling, we find that deformation is spatially inhomogeneous, with most of the deformation concentrated across relatively narrow zones. Deformation also varies with time, although the only well-documented time dependence to date is directly associated with co- and postseismic phenomena. In seismic zones, the spatial density of points required to explain the sources of crustal deformation is strongly influenced by the thickness of the elastic layer. In the western U.S., where the seismogenic zone extends to about 20 km, this translates into a desired mean station spacing of no more than 10 km and probably less across rapidly deforming basins and fault zones. To shorten the time required to estimate the deformation rate, precise repeated measurements are required; in some cases, where time dependence of strain accumulation is suspected, very frequent or even continuous measurements are needed. Better spatial sampling will allow better resolution of short wave-length features of the deformation field. It will also help to solve the problem of nonunique-ness, inherent in estimating a three-dimensional displacement field from two-dimensional surface measurements.
Fig. 3.1. Temporal and spatial scales of Earth deformation related to earthquake processes.
In summary, we need a global data set of densely sampled preseismic observations to study earthquakes. At the present time, no such data sets exists, and no system is in place to provide it. SAR interferometry has the potential to provide that data set.
After an earthquake, SAR interferometry can be used to characterize the slip that occurred at the surface (Fig. 3.3b), slip from aftershocks (Fig. 3.4), and slip distribution at depth. (Fig. 3.5). In terms of disaster relief, space-based observations--in the form of images or other data indicating areas requiring the most attention--can also be used by relief workers. SAR interferometry has a potentially important role to play immediately after an earthquake if processing times can be signficantly reduced.
Fig. 3.2. Temporal and spatial scales of various observation techniques relevant to earthquake studies.
The most important scientific objective is to understand the earthquake cycle, including the local peculiarities that arise because of different tectonic environments. Since earthquakes do not recur with sufficient frequency to acquire the necessary data in a single location, we must resort to what might be termed an "ergodic" hypothesis and study earthquakes worldwide, with the idea of recognizing patterns associated with broad sets of geological and tectonic circumstances, which presumably apply to other similar areas.
An essential piece of information that still eludes us, even in thoroughly studied and instrumented areas such as southern California, pertains to the correct prediction of the locations of future earthquakes. Satellite observations can help in several ways. First, remote sensing can help identify subtle surface features that are not easily recognized from the ground (e.g., Blom et al., 1984). Second, if the inhomogeneous character of surface deformation can be explained, potentially dangerous geological structures can be identified that might otherwise go undetected except through detailed seismic imaging of the subsurface. Perhaps the most important example pertains to blind-thrust structures, which are difficult to recognize from surface observations, yet give rise to damaging urban earthquakes such as the 1971 San Fernando, 1985 Coalinga, 1987 Whittier Narrows, and 1994 Northridge events. If precise geodetic data can shed light on the spatial distribution of accumulating strain--either through dense permanent GPS networks or SAR interferometry images--then any hint that strain is accumulating faster in a given area (e.g., Donnellan et al., 1993) would be important for devising better observing strategies, selecting target areas for intensive geological and geophysical study, and generally deploying our resources more effectively.
Fig. 3.3b. Synthetic interferogram calculated with an elastic half-space dislocation model for the Landers earthquake [Massonnet et al., 1993] . One cycle of shading represents 28 mm of change in range. White segments depict the fault geometry used in the model. Both this image and Fig. 3.3a cover a 90-by-110-km area from April 24-August 7, 1992.
Finally, an issue has only recently emerged as a critical one--the need for properly georeferenced data sets at all stages. The use of Geographical Information Systems (GIS) is becoming commonplace in the Earth sciences, as well as in all aspects of land use management. It is playing an increasingly important role in emergency management and disaster relief activities and offers the potential for much improved efficiency. Again, SAR imaging with precise orbit determination and attitude control permit rapid, detailed, accurately georeferenced, and repeated mapping of large areas of the Earth in all weather, day or night, even under circumstances where ground access is difficult or impossible.
SAR interferometry offers geophysicists their first opportunity to measure surface deformation with continuous spatial coverage, yielding a completely new view of earthquake-related ground deformation. In the case of Landers-related surface deformation (Massonnet et al., 1993, 1994; Peltzer et al., 1993; Zebker et al., 1994a), several features have been identi-fied that would be very difficult, perhaps impossible, to detect on the ground. In particular, there is evidence for a small (a few millimeters), heretofore unsuspected, amount of slip along the Garlock and Lenwood faults (Fig. 3.6). In addition, surface deformation patterns associated with individual Landers aftershocks can be seen in certain interferograms (Fig. 3.4). Such evidence could not be obtained from ground measurements without considerable luck.
Several aspects of the technique are worth mentioning:
(1) There is no evidence that the technique is more difficult to apply in urban areas. In fact, the town of Victorville, California lies in the middle of the fringe pattern published by Massonnet et al. (1993) (Fig. 3.3a, Fig. 3.3b ) and does not cause any degradation in the fringe quality. This bodes well for application of the technique to urban areas elsewhere, such as the San Fernando Valley area affected by the Northridge earthquake.
(2) Detection of surface offsets, seen as fringe offsets, benefits from the considerable areal averaging one can perform in processing the image. This means it is possible to map small integrated offsets that are distributed across a zone several hundred meters wide. Such offsets might be detected with ground-based GPS networks, but could not be mapped without extremely dense deployments, which are not economic-ally feasible at present. We have already mentioned the small amount of slip across the Garlock fault, perhaps associated with the Landers earthquake, but more important is the unanswered question of whether there was an offset across the surface trace of the Santa Susana fault associated with the Northridge earthquake.
(3) A combination of techniques may be helpful to facilitate data processing. For instance, the presence of precisely located GPS receivers or accurate topographic data within the SAR swath, could help calibrate the image and find the interferometric fringes with more ease and reliability.
Fig. 3.6. Detail of a coseismic interferogram near Landers constructed from ERS-1 radar images acquired on April 24, 1992 and June 18, 1993 [Massonnet et al., 1994] . Note the four closed fringes around the epicenter (star) of the December 4, 1992 magnitude 5.1 aftershock. Previously mapped faults are shown by dotted black lines; the rupture of the Camp Rock fault segment mapped in the field is shown as a solid white line; Soggy Lake is a white square; and secondary faults with slip observed in the radar interferogram are shown as solid white lines. These secondary faults include segments of the Garlock, Lenwood, Old Woman (OW), and possibly the Upper Johnson Valley (UJV) faults. One fringe represents 28 mm of range change.
(4) Most earthquake studies to date using SAR interferometry have focused on the well-studied Landers event. Peltzer and Rosen (1995) and Massonnet and Feigl (1995b) describe surface deformation associated with the M = 6.1 Eureka Valley earthquake, which occurred in a remote region of eastern California on May 17, 1993 (Figs. 3.7 and 3.8a, 3.8b, and 3.8c). In the future, SAR interferometry is likely to play an increasing role in the study of earthquakes in remote regions where no other surface deformation data are available.
(5) Imaging geometry may not be favorable for study of a particular earthquake, resulting in low SNR. However it is possible to "stack" several such images in order to improve SNR (Massonnet and Feigl, 1995b; Fig. 3.8a, Fig. 3.8b, Fig. 3.8c).
(6) Not yet achieved is the detection of secular strain in tectonic regions. Recent results using image pairs taken as long as 14 months apart suggest that such a goal is attainable, but the technique is likely to be valuable only in rare cases, e.g., in arid regions where deformation is rapid and concentrated in relatively narrow zones.
Fig. 3.7. Surface displacement associated with the May 17, 1993 Eureka Valley earth-quake in a remote section of eastern California as measured by ERS-1. One fringe repre-sents 28 mm of range change. In this image, color-coded range change is superimposed on a radar-amplitude image to show its relation to surface topography. Black areas indicate low radar correlation. Concentric rings show subsidence of hanging wall, up to 9.5 cm. Surface rupture occurred along southeast edge of basin. From Peltzer and Rosen (1995).
Fig. 3.8a. Composite interferograms may be produced by "stacking" two or more interferograms with unfavorable geometry. In this first of three images, observed fringes for the Eureka Valley earthquake are formed by stacking two interferograms; one fringe represents 14 mm of range change in all three panels because the stacking procedure doubles the earthquake signal.
Fig. 3.8b. In this second image, a modeled interferogram shows the fringes calculated from the focal mechanism estimated from the observed fringes in Fig. 3.8a.
Fig. 3.8c. In this third image, a residual interferogram represents the difference between the observed fringes in Fig. 3.8a and the model in Fig. 3.8b. Less than one fringe (or 14 mm) of unmodeled range change remains. From Massonnet and Feigl (1995b).
The Need for Global Coverage
The damaging effects of earthquakes occur more or less randomly along the seismic zones that girdle the planet. Because they do not occur often enough in any one place to allow us to collect a large sample over a short period of time, a global approach is required for their study. One can estimate the number of "interesting" events anywhere in the world (i.e., those that can be studied with surface geodetic techniques such as SAR interferometry) from the numbers of shallow (less than 70-km depth) earthquakes. Using the worldwide catalog for the period 1970-1991, and making the approximation that 30 percent are on land, we arrive at the following estimates of events that could be studied from satellites:
(1) There are about ten great earthquakes per year worldwide (with a surface wave magnitude greater than 7), so we can assume there are three on land. These earthquakes would typically have slip in excess of ~1 m on faults longer than about 25 km. Most would have surface rupture.
(2) There are about 60 to 70 large earthquakes per year worldwide (with a surface wave magnitude greater than 6), depending on whether the reported maximum depth used is 40 or 70 km. Let us assume that about 20 occur on land. These earthquakes would typically have slip in excess of ~0.5 m on faults longer than about 10 km. Some, perhaps half, would have surface rupture.
Only a few of these events would result in a disaster, but all would be of essential scientific interest if we are to make significant progress in understanding the earthquake cycle. This means that broad-based multinational cooperation is critical and global coverage is required. Clearly, space-based observations would be valuable.
Recommendations and Priorities
(1) The SAR interferometry technique delivers unparalleled spatial density, but only in areas where suitable scenes are available. In order to contribute significantly to earthquake and natural-hazard studies, we need much better global coverage than is available through existing spacecraft, with a focus on tectonic areas and denser temporal sampling.
We envision a SAR mission capable of two separate modes of operation: (a) a monitoring mode in which global coverage is offered to guarantee that "before" pictures are available in all areas of interest with reasonable repeat times; and (b) a rapid-response mode to allow acquisition of "after" pictures shortly after an event of interest. The latter mode may require considerable maneuverability of the spacecraft. We note that precise orbit control (using GPS) is a critical ingredient to any successful SAR interferometry mission.
(2) In spite of recent spectacular successes, there is still a need to continue developing SAR interferometry as a technique to support systematic studies of the Earth. Much of this effort can rely on existing spacecraft and aircraft capabilities. For instance, preliminary analyses indicate that L-band radar is preferable for such studies since the constraints on the interferometric baseline are not as rigid, the instrument is cheaper to build, the power requirements are not as severe, the accuracy is about the same, and temporal decorrelation is reduced. However, a full demonstration of these advantages is desirable. In particular, selected test sites with differ-ent ground covers should be examined over time with different frequen-cies (L-, C-, and X-band) to better understand temporal decorrelation.
(3) As soon as possible, priority areas should be established where existing data are examined and new data are acquired with existing systems. A high priority should be given to acquiring and archiving preseismic data.
(4) Laser altimetery can complement these studies by providing relatively high-precision (~10 cm) independent measurement of the vertical component of displacement.
(5) Existing topographic data should be declassified to provide good worldwide coverage, allowing change detection with two-pass interferometry.
Specific Space Missions
Specific space missions, including missions already planned for the near future, must be considered within the context of existing capabilities. Spectacular as they are for earthquake studies, the ERS-1 data used by Massonnet et al. (1993, 1994) have several shortcomings:
(1) Current orbit control is inadequate, making it difficult to obtain image pairs suitable for interferometric processing and leaving the spacecraft somewhat at the mercy of orbit fluctuations. Also, with the failure of the Precise Range and Range Rate Experiment (PRARE) system, orbit determination is not as good as desired.
(2) The areas available for study are too restricted, reflecting limitations in data-recording capability (i.e., there are none), distribution of ground data-reception facilities, budgets for replicating additional ground stations, and spacecraft operational limitations that preclude full-time SAR operation.
(3) For earthquake studies, an L-band radar would be preferable.
The SIR-C instrument addresses some of these concerns, although the low orbit inclination (<57deg.) precludes global coverage. Also, the short duration of the flights precludes serious monitoring capability except as a "proof of concept." In particular, it may be difficult to separate tectonic changes from atmospheric or other effects. Radarsat has the disadvantage that it does not incorporate a precise orbit-determination capability, so interferometric processing will be difficult at best. Also its C-band wavelength, while adequate for many studies, again falls short of an optimum system for change detection.
In summary, existing systems, while useful, fall short of the ideal system for monitoring the Earth's surface for earthquake-related deformation.
We therefore recommend that a dedicated SAR interferometry mission be launched by the end of the century, operate in L-band, include a laser altimeter, and offer global coverage.
Previous Section
Next Section
References
Imaging Radar Home Page
Return to Table of Contents
