Project Topics
Get Complete Project Material File(s) Now! »
Eruptive activity on Piton de la Fournaise (1989-1997)
Three basaltic eruptions took place in the period 1989-1997 at Piton de la Fournaise volcano in 1990, in 1991 and in 1992 (Toutain et al. 1990; Toutain et al. 1991; Toutain et al. 1992a) (Fig. 3.2). These eruptions occurred in an unpopulated area and did not induce any threat to property. The 1990 eruption began on January 18th from both the E interior of the Dolomieu crater and SE Dolomieu upper flank where it propagated S-SE for several hundreds of meters. During the dyke injection, tilt-meters recorded migration of the deformation centres in the same direction as the feeding fracture (Toutain et al. 1990). The 1991 eruption started on July 19th. Lava was emitted from two fissures along an eruptive fissure trending ENE. One fissure was localised inside Dolomieu crater, erupting some lava along its SE border. The other fissure was localised outside The 1992 eruption began on August 27th from a fissure within the S-W Dolomieu crater area. The fissure propagated rapidly southward, crossing the rim. Four additional fissures opened SE of the first fissure. The seismic pattern suggested a small magma pocket fed the intrusion moving up towards Dolomieu (Toutain et al. 1992a). The tilt-meter pattern suggested general inflation of Dolomieu crater (Toutain et al. 1992b).
Data processing and accuracy assessment
The image correlation technique relies on the statistical analysis of two sets of data (e.g. digital panchromatic imagery). This technique matches the ‘before’ image (1989) and the ‘after’ image (1997) at each point on a grid, analysing the degrees of local correlation at each step. Differences in the local instantaneous frequency of the images result in sub-pixel spatial differences in ground patterns (Crippen, 1992). The results are an expression of both movements in the ground surface and image distortions. Sources of image distortions are very well discussed in Michel & Avouac (2006). Distortions mainly depend on the B/H, where B is the perpendicular distance between the two airplane positions and H is the altitude above the datum. The B/H controls the stereoscopic effect and therefore the degree of local differences in both scale and feature position between the two air-photos. Moreover, when correlating two images, the B/H determines the level of quality the DEM has to have, given that the higher the B/H the higher the apparent horizontal displacement that would be induced by an error in the DEM (Van Puymbroeck et al. 2000). In a hypothetical case
where both the images have been acquired from exactly the same position (B/H=0), one would not need to perform any ortho-rectification since there would not be any stereoscopic effect. In our case (B/H≈0.25), an error of 1.5 meter in the DEM would create an apparent horizontal displacement of ~0.3 meters (~0.4 pixels). Since a finer DEM is not available, 0.3 meters represents one significant limit in the accuracy of our measurements. Another source of distortion might come from the camera lens.
This distortion is virtually zero at the centre of the scene and increase towards the edges. Our photos have been acquired by a metric camera so that the degree of distortion rate versus the lens radius is a known parameter that can be compensated during resampling.
The image correlation technique requires both the airborne images to be re-projected into the same geometry before calculating sub-pixel cross-correlation. Therefore, images are first projected into the DEM geometry (Orthorectification). In the orthorectification process we used both the metric camera parameters and 40 ground control points (GCPs). The viewing camera parameters have been provided by IGN while the GCPs have been retrieved from both the GPS network at Piton de La Fournaise and selected points on kinematic GPS profiles from Trembley & Briole (2005). Moreover, 20 tie points identified on both the images have been used to improve co-registration. Both the images have then been resampled to 0.7 meters pixel size using a cubic convolution in order to preserve radiometric quality. Uncertainties on the viewing camera parameters are minimised in the process of orthorectification. However, a residual low wavelength error might bias the results. An empirical estimate of the cumulative effects of all un-modelled sources of errors could not be performed in this study due to the lack of quality data in non-deforming periods prior to 1989. Therefore, somehow our error assessment is underestimated. Michel and Avouac (2006) have shown that un-modelled errors other than the DEM uncertainty might account for another ~1/3 of the overall error. This yields a comparative estimate which we use to reassess the accuracy of our offset measurement to ±0.42 meters.
The correlation analysis has been performed by the use of MEDICIS software. MEDICIS allows sub-pixel offset measurements by means of a sliding NxM window (Centre National d’Etude Spatiales, 2000) with an accuracy of 1/10 pixel. The size of the sliding window is chosen according to local differences in the radiometry of the input images. Each pixel is a measure of the radiance and each digital number changes according to differences in either solar illumination or surface changes, mainly. Therefore, on the one hand, a small window would be too sensitive to noise (i.e. random radiometric changes). On the other hand, a large window would be less sensitive to non-uniform offsets. In this case study, a 31x31pixel sliding window has been used. It yields independent measurements every 16 pixels (11.2 metres). The local correlation coefficient (ρ) is calculated between the two images at each step of the grid. The local correlation coefficient is a normalised value that varies between 0 and 1. It gives an assessment of the radiometric differences between the images and therefore provides us with an estimate of the uncertainties on the measurements; i.e., below a particular threshold the measurement is not reliable. In this study, we do not use offset values with ρ≤0.4. Radiometric changes typically reflect physical variations occurring on the ground between the acquisition of the two photos. Differences in solar illumination and random surface changes (like erosion phenomena or lava flows) result in poorly correlated pixels.
Measuring coseismic deformation on the northern segment of the Bam-Baravat escarpment associated with the 2003 Bam (Iran) earthquake, by correlation of very-high-resolution satellite imagery
Foreword – This chapter resumes part of the research work that I carried out in 2005/2006 at BRGM and that has been published as de Michele et al., 2008 in Geophysical Journal International. This research started from the need of using very high-resolution imagery data such as Quickbird data to retrieve co-seismic surface displacement without having access to the images ancillary information. So basically, we tried to answer the question «what can you do when you do not have access to the image geometry data and need to measure ground displacement? ». Also we try to address the issue of using high spatial resolution image data without having high spatial resolution topography data. This paper has been the object of some critics, as we should have gone into some more details on the technique given that this is the first time Quickbird is used along with the image correlation technique. We published this work as a four pages letter as we thought that the displacement on the Bam-Baravat escarpment implied some important issues for seismic hazard in the region. We did go further in this research direction (see chapter 7 and 9) and a more detailed study is ongoing at the moment. Despite the critics we like to think that with this work we have introduced a kind of «three pass» image correlation method.
Using InSAR for seismotectonic observations over the Mw 6.3 Parkfield earthquake (29/09/2004), California.
Foreword – This chapter represents our first approach on the use of InSAR on the San Andreas Fault at Parkfield. The use of InSAR on the Parkfield segment of the San Andreas Fault is the object on a ESA Category-1 research proposal.
This chapter mainly resumes the work we made during 2007 at BRGM, that I presented at the 37th congress of the International Society of Photogrammetry and Remote Sensing held in Beijing in July 2008. The results are published as de Michele et al., 2008 in the ISPRS proceedings. This work marks my first publication in the InSAR domain, which theoretical basis I learnt at college and improved practically at Institute de Physique du Globe de Paris. In this chapter I chose to show only the published results that are constrained by the small allotted space. Consequently I don’t show the numerical modeling efforts made by Jerome Salichon and Anne Lemoine whose intensive work I wish to acknowledge.
The Parkfield segment of the San Andreas Fault is one of the main topics of this dissertation; I chose to show a further detailed InSAR study over Parkfield in the next chapter.
InSAR evidence of pre-seismic strain accumulation
We have examined the preseismic deformation field recorded by space-based InSAR (Interferometric Synthetic Aperture Radar) in the Parkfield area. InSAR can map ground deformation at a decameter spatial resolution with sub-centimeters precision in the line of sight direction (LOS) (Massonnet and Feigl, 1998).
The San Andreas fault north of Parkfield undergoes continuous creep at an average rate of ~3 cm/yr (Toké and Arrowsmith, 2006). Atmospheric delays in the radar scenes of the interferometric pair (Zebker et al, 1997) could mask this kind of signal in a single interferogram. We therefore constructed the InSAR velocity field by averaging 30 selected unwrapped interferograms (Raucoules et al., 2003; Le Mouélic et al., 2005) calculated over the preseismic period 1993-2004, prior to the September 28 2004 earthquake. With this procedure, the effect of atmospheric delay is minimized as it is assumed to be uncorrelated with time.
Considering the simple mechanism of the strike-slip faulting regime of this region, we can rule out substantial contribution of vertical fault slip in this area of the SAFZ (Titus et al., 2006; Murray and Langbein, 2006 ) and assume that the InSAR signal is only due to horizontal surface fault slip. The result is, therefore, a velocity map (Fig. 5.2) showing right lateral shear distribution over 141 km of the San Andreas Fault centred in the Parkfield area prior to the 28 September 2004 earthquake. A prominent feature of strain accumulation in the Gold Hill area highlighted in the preseismic interferogram stack is the abrupt spatial decrease of creeping rate along the SAFZ towards Gold Hill, where relative creep velocity reaches a local minimum (Fig. 5.3). This suggests either a change in the rheological material properties or a bend in the fault strike resulting in increasing friction towards Gold Hill. 30 km further south-east of Gold Hill, at the beginning of the Cholame segment, relative surface velocity increases to moderate values (~0.7 cm/yr) indicating the re-starting of surface creep. This spatially discontinuous change in the mode of slip suggests the presence of a stronger section of the fault between Gold Hill and Cholame that acts as a barrier locking the fault sub-segment south of Gold Hill. We will see further how this feature played a significant role in the nucleation of the 2004 Parkfield earthquake. Subsidence due to water pumping in the Paso Robles sub-unit (Valentine et al., 1997) manifested as a bull eye shaped range change pattern south of Parkfiel is a marked feature in the presismic interferogram. Similar features though smaller in area can be observed in the northern sector of the preseismic interferogram (Fig 5.2). These correspond to petroleum and gas withdrawal from a shallow reservoir in the Lost Hills field and neighbouring reservoirs (Fielding et al., 1998, Brink et al., 2002).
InSAR evidences of co-seismic strain release
We then used radar data from ERS-2 satellite to form interferograms that records coseismic surface deformation of the September 28 2004 Parkfield event. The ERS-2 satellite experienced gyroscope failure in 2001. However, we could use 4 selected radar scenes among the ERS-2 dataset to construct coherent interferograms. Atmospheric phase contribution is a major limiting factor in the precision of a single interferogram.
We minimize the atmospheric contribution to the coseismic signal by averaging two coherent coseismic interferograms spanning 950 and 490 days, respectively. It has to be noted that with such an important time lapse, aseismic deformation contributes significantly to interferometric signal in masking co-seismic deformation.
Thus, we estimate aseismic phase contribution according to each time lapse from the aforementioned pre-seismic interferogram and remove pertinent aseismic contribution from each individual co-seismic interferogram before stacking. The results show a LOS spatially detailed coseismic surface displacement and provide a map of the surface that ruptured during the September 28 2004 Parkfield earthquake (Fig. 5.4). We observe that the earthquake rupture extends from south of Gold Hill and ruptured North West toward Middle Mountain forming discontinuous breaks on the north western side of Cholame Valley, which is consistent with field observations (Rymer et al., 2006). We observe that the rupture developed along at least 35 km of the SAFZ. Coseismic surface deformation is consistent with a dextral strike slip mechanism. We measure maximum coseismic slip of up to 15 cm LOS, which makes ~21 cm horizontal displacement assuming that the vertical coseismic slip component is negligible. We compared our coseismic InSAR results with permanent GPS solutions GPS permanent stations were installed in the framework of the Earthquake Prediction Experiment at Parkfield (Roeloffs and Langbein, 1994). InSAR results are briefly in good agreement with continuous coseismic GPS solutions that captured the coseismic signal about 20 km north-west of the epicenter.
Table of contents :
Chapter 1 Introduction and background.
Context of the Thesis and structure of the manuscript
Methods
Synthetic Aperture Radar Interferometry
Subpixel correlation technique
Chapter 2 Early detection of eruptive dykes revealed by normalized difference vegetation index (NDVI) on Mt. Etna and Mt. Nyiragongo
Summary
Introduction
Analytical method
Results
Discussion
Acknowledgements
Chapter 3 Deformation between 1989 and 1997 at Piton de la Fournaise volcano retrieved from correlation of panchromatic airborne images
Summary
Introduction
Eruptive activity on Piton de la Fournaise (1989-1997)
Data
Data processing and accuracy assessment
Quantification of displacement
Mapping lava flows
Discussion and conclusions
Acknowledgements
Chapter 4 Measuring coseismic deformation on the northern segment of the Bam-Baravat escarpment associated with the 2003 Bam (Iran) earthquake, by correlation of very-high-resolution satellite imagery
Summary
Introduction
Data
Methodology
Results
Discussion
Acknowledgments
Chapter 5 Using InSAR for seismotectonic observations over the Mw 6.3 Parkfield earth- quake (29/09/2004), California.
Summary
Introduction
InSAR evidence of pre-seismic strain accumulation
InSAR evidences of co-seismic strain release
Discussion and Conclusions
Acknowledgements
Chapter 6 Spatiotemporal evolution of surface creep in the Parkfield region of the San Andreas Fault (1993-2004) from Synthetic Aperture Radar.
Introduction
Methodology
Results and discussion
Conclusions
Acknowledgements
Chapter 7 The Mw 7.9, 12 May 2008 Sichuan earthquake rupture measured by sub-pixel correlation of ALOS PALSAR amplitude images.
Summary
Introduction
Data analysis
Results
Discussion and conclusion
Acknowledgement
Chapter 8 Assessing ionospheric influence on L-band SAR data: Implications on co- seismic displacement measurements of the 2008 Sichuan Earthquake.
Summary
Introduction
Ionospheric effects on InSAR data
Azimuth correction
Computation of the phase derivative
Integration of the azimuth correction: the IPS
Discussion
Conclusions
Acknowledgement
Chapter 9 Three-dimensional surface displacement of the 12 May 2008 Sichuan earthquake (China) derived from Synthetic Aperture Radar: evidence for rupture on a blind thrust.
Summary
Introduction
Methodology
Results
Interpretation and modelling
Discussion and conclusions
Acknowledgements
Chapter 10 Surface displacement of the Mw 7 Machaze earthquake (Mozambique): Complementary use of 2 multiband InSAR and radar amplitude image correlation
with elastic modelling.
Summary
Introduction
Data
Data Processing
Results
Co-seismic deformation modelling
Discussion and Conclusion
Acknowledgements
Chapter 11. Conclusions and perspectives
