利用星载SAR差分干涉测量改进变形含水层系统的绘图、监测和分析.doc

利用星载SAR差分干涉测量改进变形含水层系统的绘图、监测和分析美国国家研究委员会地面沉降座谈会（NRC；1991）就3种信息需求达成了共识：“第一，有关地面沉降大小和分布的基本的地球科学数据和信息要得到认可并用来评价未来的问题。这些数据不仅能够帮助研究局部地区的沉降问题，也能识别国家范围内的问题。第二，针对地面沉降开展沉降治理和工程方法的研究为了有效阻止或控制破坏第三，尽管美国现行的地面沉降减轻方法有很多种，但是对这些方法的成本效益进行研究将有助于决策者做出更好的选择。”有各种基于地面和卫星的方法可用来测量含水层系统的压缩和地面沉降（表1）。SAR干涉测量理论上适合测量与含水层系统压缩相关的地面变形的空间范围和大小。InSAR可以提供一个区域内覆盖整个含水层系统的数百万个数据点，与使用大量人力而只能获得有限个点测量数据的水准测量，和GPS测量相比，通常而言，花费要更低一些。通过识别研究区内某一变形的特定区域，SAR干涉测量也可以用于定点测量并同时监测局部和区域尺度上的地面沉降（如钻孔伸长计、GPS监测网络、水准路线；Bawden等，2003）。SAR干涉测量的这些优势，尤其是InSAR，能够满足NRC提出的每一种信息需求。SAR干涉测量的另一个重要优势就是SAR历史数据的存档文件越来越多。在很多地区，从上世纪90年代初开始，就已经有了大量的数据集，因而这一时期的地面形变历史测量数据即可应用。此外，为满足新需求可以定制新数据。详细的过程和费用要依赖于使用的传感器。Space-based Tectonic Modeling in Subduction Areas Using PSInSARR. M. W. Musson British Geological Survey M. Haynes NPA Group A. Ferretti TeleRilevamento EuropaINTRODUCTIONWhile the application of InSAR (INterferometric Synthetic Aperture Radar) techniques to seismology has been well known since the mid-1990s (Massonnet et al., 1993; Massonnet et al., 1996), PSInSAR is generally unfamiliar to the Earth science community. The PS stands for permanent scatterer, and it is the use of these (along with the volume of scenes employed) that distinguishes the method from more familiar InSAR techniques. A permanent scatterer is any persistently reflective pre-existing ground feature, such as building roofs, metallic structures, and even large boulders. The use of these features offers the possibility of measurements of ground displacements to a degree of accuracy, and over periods of time, previously unobtainable from conventional interferometry. Furthermore, it is possible to construct histories of displacements over the full temporal extent of the SAR data archive (started in 1991) for any part of the globe with data coverage. PSInSAR therefore represents the equivalent of a newly discovered, superaccurate, extremely dense GPS network that has been in existence for the last twelve years. The high resolution of PSInSAR data, coupled with its being particularly suited to urbanized areas (numerous buildings, therefore many PS points), makes it an excellent tool for studying things such as urban subsidence (Ferretti et al., 2000; Mizuno and Kuzuoka, 2003; Dehls and Nordgulen, 2004). It also has applications in seismology: as a substitute for GPS data where these do not exist, and as an enhancement where they do. In this paper we report on a pilot project in Japan, the principal aim of which was to calibrate and test the PSInSAR measurements in an area where ground truth is very well established from GPS and leveling data. This work results from a European Space Agency (ESA) Earth Observation Market Development project entitled Developing markets for EO-derived land motion measurement products, involving NPA (lead), the British Geological Survey (UK), Imperial College (UK), TeleRilevamento Europa (Italy), ImageOne (Japan), the Geographic Survey Institute (Japan), Oyo Corporation (Japan), Fugro (Netherlands), and SARCOM (the ESA data distributing entity). TECHNICAL BASIS OF PSInSARConventional satellite radar interferometry involves the phase comparison of synthetic aperture radar (SAR) images gathered at different times (Massonnet et al., 1993; Massonnet et al., 1994; Zebker et al., 1994; Gens and van Genderen, 1996; Massonnet and Feigl, 1998). This technique has the potential to detect millimeter-level target displacements along the line-of-sight (LOS) direction. The aim of the interferometric techniques is to highlight possible range variations of the target by means of a simple phase difference between two images gathered at different times. If the local reflectivity remains unchanged in time, its phase contribution disappears in the differentiation and possible range variations can then be detected. Since the wavelength of the illuminating radiation is usually a few centimeters (satellite SAR operates in the microwave domain), even a millimetric range variation translates to a phase change that can be detected. Due to low signal-to-noise ratio values typically present in SAR phase values (never greater than 12 dB), however, the monitoring of subsidence rates of more than 6-7 cm/year is not feasible. Apart from cycle ambiguity problems, other limitations are due to temporal and geometrical decorrelation and to atmospheric artifacts. Temporal decorrelation makes interferometric measurements unachievable where the electromagnetic profiles and/or the positions of the scatterers change with time within the resolution cell, so that the reflectivity phase contribution cannot be assumed constant with time. The use of short revisiting times proves to be an unsuitable solution, since very slow terrain motion (e.g., seismic creep) cannot be detected. Reflectivity variations as a function of the incidence angle (i.e., geometrical decorrelation) further limit the number of image pairs suitable for interferometric applications, unless the change is confined to a pointwise character of the target (e.g., a corner reflector). In areas affected by either kind of decorrelation, generation of the interferogram no longer compensates the reflectivity phase contribution, and possible phase variations due to target motion cannot be highlighted. Finally, atmospheric heterogeneity creates an atmospheric phase screen superimposed on each SAR image that can seriously compromise accurate deformation monitoring. Indeed, even considering areas slightly affected by decorrelation, it may prove extremely difficult to discriminate the signal of interest from the atmospheric signature, at least using individual interferograms. The PSInSAR method, developed by TeleRilevamento Europa of the Politecnico di Milano in Italy, provides a way to overcome these limitations. Although temporal decorrelation and atmospheric disturbances still strongly affect interferogram quality, reliable deformation measurements can be obtained in a multi-image framework on a small subset of image pixels corresponding to stable areas. These points, the permanent scatterers (PS), can be used as a natural GPS network to monitor terrain motion, by analyzing the phase history of each one. Atmospheric artifacts show a strong spatial correlation within every single SAR acquisition (Hanssen, 1998) but are uncorrelated in time. Conversely, target motion is usually strongly correlated in time and can exhibit different degrees of spatial correlation depending on the phenomenon at hand (e.g., subsidence due to water pumping, fault displacements, localized sliding areas, collapsing buildings, etc.). Atmospheric effects can therefore be estimated and removed by combining data from long-time series of SAR images, such as those available in the ESA ERS archive, which has been gathering data since late 1991. To exploit all the available images, and improve the accuracy of the estimation, only scatterers that are not greatly affected by temporal and geometrical decorrelation are selected. Possible stable and point targets, known as permanent scatterers (PS), are detected on the grounds of the stability of their amplitude returns (Ferretti et al., 2001): i.e., how constant their brightness or intensity remains from one SAR image to the next. This allows pixel-by-pixel selection with no spatial averaging. Due to the high spatial correlation of the atmospheric contribution, proper sampling of the atmospheric components can be achieved with a sparse grid of measurements, provided that the PS density is high enough (greater than 4-5 PS/km2; Ferretti et al., 2000, 2001). A sufficient number of images is needed (usually more than 30) to identify PS and separate the different phase contributions. Even though precise satellite position and velocity state vectors are available for ERS satellites, orbit ambiguities and their impact on the interferograms cannot be neglected. The estimated atmospheric phase screen is actually the sum of two contributions: atmospheric effects and fringes due to orbital errors. The latter correspond to low-order phase polynomials, however, and do not change the low wave-number character of the signal to be estimated on the sparse PS grid. At the PS point, submeter accuracy elevation and millimetric terrain motion detection (due to the high phase coherence of these scatterers) can be achieved once atmospheric contributions are estimated and removed. Relative target LOS velocity can be estimated with unprecedented accuracy, sometimes even better than 0.1 mm/year, due to the long time span of the data used. The higher the accuracy of the measurements, the more reliable the differentiations between models of the deformation process under study. PILOT PROJECTThe area selected for the pilot project was the Tokai area in Japan, initially around Hamamatsu and then extended to cover the rest of the west side of Suruga Bay and the northern part of the Izu Peninsula (Figure 1). This area was attractive for the project for several reasons. It is one of the most intensively studied areas in Japan, because it was identified as the likely location of the next major earthquake as long ago as the 1970s. It is an area of active tectonics in a complex structural setting. The principal component of the tectonic structure in the Tokai district is the collision of the northward-moving Philippine Sea Plate (PSP) with Japan (Figure 1). This collisional process started about 6-7 million years ago (Niitsuma, 1982) and has been responsible for most of the seismicity of southern Japan as the PSP subducts under the overlying Eurasian Plate. In southern Honshu the situation is relatively simple and follows the conventional subduction model. The plate boundary geometry around the Tokai district is very much more complex, however (Takahashi, 1994). The subduction trench, which, as the Nankai Trough, is oriented northeast-southwest to the south of Honshu, bends to an almost northsouth orientation as the Suruga Trough in Suruga Bay. The most northerly point of the PSP is occupied by the Izu Peninsula, which is colliding with Honshu rather than being subducted beneath it. The reason for this is believed to be the relative lightness of the volcanic rocks of the Izu Peninsula, the buoyancy of which prevents subduction (Takahashi, 1994). The northward movement of the PSP in the Izu area is therefore a process of collision tectonics akin to continental collision rather than normal subduction (Niitsuma and Matsuda, 1985; Koyama, 1991). Prior to the collision of the Izu Peninsula, the Tanawa Block collided with the Honshu mainland during the Miocene, and the process by which the Tanawa Block accreted to the mainland is now being repeated with the Izu Peninsula (Amano, 1991). The process is described in detail by Takahashi (1994). The possibility, or even the probability, of a large and disastrous earthquake in the Tokai district has been of concern since the area was identified as a danger area and seismic gap by Mogi (1970) and Ishibashi (1976). The subduction front from Shikoku to Hamamatsu has been identified as being partitioned into several principal fault planes that appear to rupture in characteristic earthquakes. These segments were labeled A to D (from west to east) by Ando (1975), and this system was expanded by Sugiyama (1994) to include segment Z (Bungo Channel) in the west and segment E (Suruga Bay) in the east (Figure 1). Any large earthquake may rupture one of these segments entirely, or in the worst case all six at once, as apparently occurred in the 1707 Hoei earthquake (Sugiyama, 1994). For histori