Noise-reduction in optical image correlation through time-series analysis : application to the 2013 Balouchistan earthquake
Encadrants : James Hollingsworth , Marie-Pierre Doin (ISTerre)
keywords : remote sensing, inverse modeling, image processing, earthquake cycle
Optical image correlation (OIC) is a modern geodetic technique which measures the relative spatial movement of common features represented by pixels within two or more optical satellite images. Depending on the quality and resolution of the images used in the analysis, pixel shifts as low as 1/10th of the input resolution can be detected. OIC thereby allows for the rapid retrieval of ground deformation over large areas using remotely sensed data. Nevertheless, despite increasing volumes of medium resolution (10-15m) satellite data now available, OIC is limited to retrieving deformation signals higher than the background noise level (1/10th pixel, or 1m for 10m resolution imagery). As a result, OIC has so far only been useful in proving information on the spatial pattern of earthquake deformation during the co-seismic phase of rupture, where deformation is large. To study the full seismic cycle (including the smaller post-seismic and inter-seismic phase) a new method is required to reduce the noise level capable with OIC.
Until now there has been little discussion of the factors which control the noise content in image correlations. One important source of noise which has not previously been addressed is the influence of sun position on the illumination (cast-shadow content) of a satellite image. During image acquisition, the sun illumination conditions (i.e. sun aspect and elevation) change throughout the year. We expect large differences in the illumination conditions between image pairs to produce shifts of the shadow content, particularly in areas of higher topography, which will influence the correlator and result in large artifacts in the displacement field.
In this project, we will first investigate the influence of changing sun illumination by correlating a variety of sun-shaded digital elevation models, thus providing a new way to account for illumination artifacts. We will apply this method to Landsat8 satellite images spanning the 2013 Balouchistan earthquake, which ruptured a 220km-long section of the Hoshab fault in Western Pakistan, slipping up to 12m in a left-lateral sense. The pattern of slip revealed strong segmentation of the fault, which curves along the southern edge of the Hoshab fold. Measurements of near-field displacement, coupled with a rapidly decaying signal away from the fault suggest the majority of slip occurred in the uppermost crust. This contrasts with several previously studied strike-slip earthquakes around the world which show a peak of slip at 3-5km depth, decreasing substantially towards the surface (known as ’Shallow Slip Deficit (SSD)’, see Fialko, et al., 2003). Inversion of optical correlation data to produce a time-series will allow a more acccurate retrieval of the co-seismic phase, thereby allowing a more effective assessment of the degree of SSD for the Balouchistan earthquake. Furthermore, the noise reduction offered by our time-series analysis will allow us to better address the nature of the post-seismic slip phase, which remains poorly characterized due to the lack of InSAR data available at this time. Initial correlation of single image pairs during the post-seismic phase suggest N-S extension may have occurred across the fault. Although such motion is unexpected due to the compressional tectonic setting of this earthquake, it could have occurred in response to dynamic over-shoot in the co-seismic slip phase. We will test this hypothesis by using optical time-series. The speed, magnitude, spatial pattern and localization of post-seismic slip all provide important constraints on the rheological properties of the crust, which are critical for understanding how faults slip in single events and over multiple earthquake cycles. This work will further refine the OIC technique, increasing its effectiveness for studying many processes shaping the Earth’s surface, including volcanic deformation, landlides, glacier flow, and the seismic cycle.