Tectonic motion of the continents is not confined to narrow zones at plate boundaries but is characterised by wide regions of distributed seismic deformation. This arises as a result of the differing strengths and depth-dependent behaviour of oceanic and continental lithospheric materials. In addition, the presence of a ductile zone in the continental lower crust, beneath the brittle seismogenic upper crust, may decouple surface displacements from those of the lithosphere as a whole and permit post-seismic stresses to relax, and thus influence the seismic cycle.
The Aegean region is the most seismically active part of Europe and one of the most rapidly extending basins of the world. Tectonic motion is related to the gravitational spreading of thickened Anatolian and Aegean continental crust as the interface with the subducting African oceanic crust migrates southward. Normal fault systems in the area provide the opportunity to study crustal deformation at various stages of the seismic cycle. The balance between inter-seismic geodetic build-up of regional strain and co-seismic localisation of strain has important consequences in terms of seismic hazard. Also, direct comparison between the co-seismic deformation inferred from seismology and that measured by geodesy assists with the understanding of historical seismological evidence. Post-seismic transient strains are expected to be small, but may have important effects and so must be quantified and modelled.
Geodetic surveying methods have been used to detect tectonic motions for many years, but terrestrial surveying techniques are limited by poor precision over long distances and so many years must elapse before a detectable signal has built up. Space geodetic techniques have recently been developed that are precise over long distances, but the equipment for SLR and VLBI measurements is bulky and expensive. The differential GPS carrier phase surveying technique is precise, portable and relatively cheap, and so offers the possibility to detect crustal tectonic motions over the time-scale of a few years by repeated temporary occupation of markers situated in bedrock. The precision of earlier GPS surveys is limited by poor satellite orbit information, but can be improved using the technique of fiducial GPS, in which orbit parameters are estimated along with site coordinates, while the coordinates of sites which are well constrained by other space geodetic methods and have velocities constrained similarly or by geological information are held fixed. Later surveys can be processed using readily-available post-processed `precise' orbits.
Version 3.4 of the Bernese GPS Software is used in this study. The station-to-station, satellite-to-satellite double-difference observable is used, with integer ambiguity parameters estimated and fixed where possible, with the aid of a model of ionospheric total electron content when necessary. Zenith tropospheric signal delays are estimated using a stochastic model. Baselines are chosen to maximise the number of common observations and minimise overall baseline length and receiver type mixing. GPS observations are combined in daily networks to yield daily coordinate sets and their covariances, which are then combined using network adjustment software to generate a campaign coordinate set.
Differential GPS surveys yield site coordinates in a global reference frame, but relative to the base station at each epoch. An error in positioning over the mark at the base station will cause a translational error in the realisation of the reference frame at that epoch, but these offsets can be estimated if there are sites occupied three or more times, and the assumption of constant site velocities is made. To justify this assumption, temporal displacement discontinuities such as co-seismic deformation must first be eliminated by modelling, leaving only inter-seismic accumulation of strain. Translational errors will affect attempts to fit the time series of site coordinates by a low-order polynomial velocity field.
Sites in a 66-station network covering central Greece have been occupied with GPS up to six times over the interval 1989 - 1996. The network bounds the two major extensional features of the region, the Gulfs of Korinthos and Evvia. Data from the first three epochs (June 1989, October 1991 and May 1993) are processed using fiducially-improved satellite orbits, whereas later epochs (June 1995, October 1995 and May 1996) are processed using CODE precise orbits. Unfortunately, coordinates from the first epoch, even from the subset of sites which was occupied with dual-frequency GPS receivers, are too unreliable to be of use in short-term deformation studies. Later surveys of the Central Greece Network utilised dual-frequency receivers throughout and yield excellent results, although the three latest surveys only cover the sites around the Gulf of Korinthos.
The Ms=6.2 15 June 1995 Egion earthquake is the only event to cause significant co-seismic displacements of any sites in the Central Greece Network, and these are removed from coordinate sets obtained after this date using a forward model based on the source mechanism of Bernard et al. (1996), leaving inter-seismic displacements only. Inter-seismic baseline length changes are smooth with time, demonstrating that the scale of the GPS reference frame is maintained through time. Whole-network translations at each epoch are estimated and attributed to possible errors in base station positioning. The translated coordinate sets can be fitted by temporally uniform site velocities, within expected the error bounds of GPS-derived coordinates. Whole-network rotations do not significantly improve the fit and so have not been applied.
The site velocities are expressed in a Europe-fixed reference frame using the \nv1nnr and \itrf plate and site velocity models, which include the base station of the Central Greece Network (DION). All sites in the network show significant motion with respect to `stable' Europe, indicating that deformation must occur to the northwest of the network in northwest Greece and Albania.
Polynomial velocity fields of up to 4th order are unable to fit the time series of translated site coordinates, indicating that deformation is localised at a scale much smaller than the region. Analysis of uniform strain rates within small polygonal regions reveals significant high strains within the western Gulf of Korinthos which are not matched elsewhere, even in the eastern Gulf of Korinthos or Gulf of Evvia. Overall, the pattern of inter-seismic strain resembles that observed over the interval 1892 - 1992 by Davies et al. (1996), indicating that secular strain may be measured equally well over short time-scales in which no earthquakes take place or longer time-scales in which the effects of several earthquakes are averaged out.
The Gulf of Korinthos is the best-constrained part of the Central Greece Network by virtue of its longer span of occupation, and also exhibits the highest strain rates. As one of the most densely-populated regions of Greece, it has a good record of historical seismicity which can be related to the geodetic extension using the method of Kostrov (1974). Uncertainty in the scalar moment of earthquakes arises principally from instrumental imperfection and errors in the Ms - M0 relationship for older events, and in the amount of strain released in aftershocks and pre- or post-seismic creep for all events. Additional strain release may have occurred in earthquakes too small to be included in the historical record, but previous workers have shown that this is unlikely to affect the total strain by more than 50% (Ambraseys & Jackson, 1990).
Sites in the northern Peleponnessos show very little relative motion, so across-Gulf extension can be easily studied in a reference frame in which these sites are fixed. Strike-perpendicular site velocities observed with GPS over the interval 1991 - 1996 increase smoothly from east to west, and the velocities obtained from triangulation / GPS data over the interval 1892 - 1992 by Davies et al. (1996) are commensurate with this trend. On average, geodetic extension rates in the western Gulf (12.7 +- 1.0 mm/yr) are approximately twice those in the east (6.4 +- 1.0 mm/yr).
In contrast, the rate of seismic moment release this century is marginally higher in the eastern Gulf of Korinthos than in the west, and the `seismic' extension rates calculated in the east on the basis of a 10 km or 15 km seismogenic layer (6.0 +- 2.4 mm/yr or 4.0 +- 1.6 mm/yr) agree with the geodetic extension rate, with a closer match for the higher rate based on a 10 km layer. In the western Gulf, the rate of seismic moment release implies extension rates of 3.0 +- 1.2 mm/yr (15 km layer) to 4.5 +- 1.8 mm/yr (10 km layer), significantly smaller even than the lowest geodetic extension rate observed at this end of the Gulf. The frequency of large earthquakes in the western Gulf during the period 1690 - 1890 has been higher than this century, and so it seems likely that the deficit of seismic strain release in the western Gulf will be met by several such earthquakes in the medium term. For a 10 km seismogenic layer, the total moment release required to eliminate the deficit of seismic strain is 22 x 10^18 N m (for the maximum possible seismic strain that has already occurred), and for a 15 km layer the required moment release may at the extremes of possibility be as high as 52 x 10^18 N m, equivalent to more than six earthquakes the size of the 24 February 1981 Alkyonides mainshock.
The Ms=6.6 13 May 1995 Kozani - Grevena earthquake struck a region of low historical seismicity in northwestern Greece in which geodetic strain has not previously been quantified. The epicentre is surrounded by a recently-occupied Hellenic Army Geodetic Service triangulation / trilateration network, from which pillars selected on the basis of a forward model of the expected co-seismic deformation were occupied with GPS immediately after the event. Co-seismic horizontal displacements are obtained by differencing the pre-seismic conventionally-surveyed and post-seismic GPS coordinate sets.
Surface displacements caused by uniform slip on a fault plane can be computed using the elastic model of Okada (1985), but the inverse problem is highly non-linear. A variation on previous inversion algorithms is proposed, in which the downhill simplex method of Nelder & Mead (1965) is augmented by the performance of several hundred inversions from randomly-selected starting points. The need for good a priori parameters is thus removed, and a more complete understanding of confidence limits is built up. The algorithm is tested using data from the 1981 Alkyonides earthquake sequence.
The algorithm is then used to estimate the full set of source parameters of the 1995 Kozani - Grevena earthquake from the geodetic displacements obtained as above. Because the pre-seismic and post-seismic coordinates are expressed in different reference frames, scale, rotation and translation parameters are also estimated. Extensive stability tests are conducted on the solution using the actual data, synthetic data based on the actual site distribution, and synthetic gridded data.
The geodetic earthquake source mechanism agrees well with the Harvard CMT solution and studies of local aftershocks (Hatzfeld et al., 1996) in all respects save the scalar moment corresponding to the geodetic solution (16.3 x 10^18 N m for 1.2 m of displacement on the fault), which is over twice that of the CMT solution (7.6 x 10^18 N m). This discrepancy cannot be explained purely by the cumulative effect of aftershocks, so pre- or post-seismic creep may be significant. Such a high proportion of aseismic creep relative to the seismic moment, if repeated for all earthquakes in the western Gulf of Korinthos, would not eliminate the discrepancy between geodetic and seismic strain that occurs in this area.
The minimum and maximum depth extents of faulting in the model are 2.8 km and 13.5 km respectively. Faulting does not propagate to the surface because near-surface deformation is distributed within a layer of unconsolidated sediments of mid-Pliocene and younger ages, which cover much of the region. The maximum depth of faulting is consistent with the aftershock distribution and expected values for the seismogenic layer thickness.
The model fault scarp does not correspond to any of the clear-cut surface features in the region, but to a poorly-defined scarp within the zone of soft surficial sediments, close to the location of the most prominent co-seismic ground cracking. Without geodetic observation and modelling it is likely that such a feature would have been overlooked in the search for faults associated with possible seismic hazard.
Post-seismic transient deformation completes the seismic cycle. Surface deformation caused by relaxation within a visco-elastic layer of the stresses induced by co-seismic elastic deformation is calculated using the model of Rundle (1982) and later co-workers, for the case of co-seismic motion in an elastic layer overlying a visco-elastic half-space. Post-seismic motion after the 1981 Alkyonides earthquakes is used as an example, and motion after the 1995 Grevena and Egion earthquakes is computed to aid geodetic network design. Relaxation by aseismic creep localised on the fault plane or its down-dip extension as a shear zone may also occur, and can be distinguished from visco-elastic distributed deformation by its temporal and spatial variation. When designing networks for the study of post-seismic displacements, consideration must be given to monument stability and the independent measurement of secular strain, because the post-seismic signal is small.
This thesis demonstrates that multiple-occupation GPS studies can quantify inter-seismic crustal deformation as precisely as and in agreement with medium-term GPS / triangulation studies, but in a far shorter time-scale, providing that co-seismic displacements of sites can be accurately modelled. Post-seismic transient deformation after normal-faulting earthquakes such as those studied is smaller in magnitude and can only be detected with GPS or more accurate techniques. Future work will include measurement and further modelling of post-seismic motion after the 1995 Grevena and Egion earthquakes, and integration of GPS and SAR data concerning the two. Also, the strain field in northwestern Greece and the northern Aegean Sea will be quantified by comparison of future GPS observations with existing terrestrial surveying measurements to determine the limits of the Aegean deformational zone and to estimate seismic hazard in the region.