19 Inverse Modeling and Related Analyses: Methods and Applications

Thursday, 10 June 2010
8:00 AM - 12:00 PM
Pine

QUANTIFYING UNCERTAINTY IN EARTHQUAKE SOURCE INVERSIONS

Paper number: 19-8
Program: Special Sessions
Time: 10:30 AM – 10:55 AM

Authors

Abstract

Finite-fault source inversions are routinely used to image the earthquake rupture process at depth. Differences between slip models produced by different research groups, however, make it clear that uncertainties in these models can be quite large. There are several factors that can contribute to these uncertainties: 1) the resolving power of the data is limited, particularly at depth, 2) errors in the data (or more typically, the Green’s function) propagate into the solution, and 3) modeling assumptions used to regularize and parameterize the problem may be oversimplified or even false.

We discuss ways to quantify the uncertainty in source inversions and reduce artifacts in poorly resolved areas of the fault plane. We examine the resolving power of GPS data from the 2004 Mw6.0 Parkfield Earthquake, which was recorded by 13 near-field 1-Hz receivers. We find the resolution of our inverted slip model to be very poor at depth and near the edges of the modeled fault plane. The extreme spatial heterogeneity of the model resolution leads to artifacts in poorly resolved areas. To remove these artifacts, we limit the free parameters of the inversion using a variable subfault size that matches the local resolution length of the GPS data. We show via synthetic tests that this formulation reduces spurious structure in poorly resolved areas while recovering structure on a fine scale near the surface. We also use Monte Carlo sampling to quantify the effect of data errors in the final slip model. In this way, our final model captures both resolution errors (via a spatially nonuniform gridding of the fault plane) and data errors.

In addition, we discuss the on-going Source Inversion Validation (SIV) collaboration, which seeks to better determine which features of source inversions are robust and which inversion techniques are best able to recover details of the source process.