Seven big strike-slip earthquakes

We examine seven large (M_w > 7) strike-slip earthquakes that occurred since the beginning of ERS 1 and 2 missions. We invert GPS observations and InSAR interferograms and azimuth offsets for coseismic slip distributions. We explore two refinements to the traditional least-squares inversion technique with roughness constraints. First, we diverge from the usual definition of ``roughness'' as the average roughness over the entire fault plane, and allow ``variable smoothing'' constraints. Variable smoothing allows our inversion to select models that are more complex in regions that are well-resolved by the data, while still damping regions that are poorly resolved. Second, we choose our smoothing parameters using the _{jR_i} criterion. The _{jR_i} criterion draws on the theory behind cross-validation and the bootstrap method. We examine the theoretical basis behind such methods and use an analytical approximation technique for linear problems. We provide maps of model variance and spatial averaging scale over the fault plane, to explicitly show which features in our slip models are robust. We examine the 1992 Landers (CA), 1995 Sakhalin (Russia), 1995 Kobe (Japan), 1997 Ardekul (Iran), 1997 Manyi (Tibet), 1999 Hector Mine (CA), and 2001 Kunlun (Tibet) earthquakes. We compare features of the slip distributions such as the depth distribution of slip, the inferred magnitude and the degree of heterogeneity of slip over the fault plane, as resolved by the available InSAR and GPS data. We end with a brief description of the data coverage required for future earthquakes of similar size if we want to infer some of the above quantities to within a given confidence interval. We describe both the number of InSAR scenes and the distribution of GPS points that would be required, based on theoretical treatments of the fault plane/data point geometry using the _{jR_i} method.