Improved Event Location Uncertainty Estimates

Abstract

Abstract : The objective of this project was to develop methodologies that improve location uncertainties in the presence of correlated, systematic model errors and non-Gaussian measurement errors. We have developed a methodology based on copula theory to obtain robust estimates of variogram models for travel-time error. Using this methodology, we produced generic, transportable variogram models for regional Pn and teleseismic P phases. These models are employed to estimate the network covariance matrix that describes the spatial correlation structure of travel-time predictions errors. We have developed and validated models of measurement errors that map phase picking delay and variance as a function of SNR, as well as a function of epicentral distance and body wave magnitude considered as a surrogate for SNR. The phase pick delays are implemented as travel-time corrections, while the measurement error variances add to the diagonal of the network covariance matrix to obtain the full, non-diagonal data covariance matrix. Our representation of model and measurement errors assumes that the bulk of 3D velocity heterogeneities in the Earth are accounted for by calibrated travel-time predictions. We have developed a linearized iterative location algorithm that makes use of the full data covariance matrix, hence accounts for the correlated model error structure. By relocating a large number of GT events (GT0-2 nuclear explosions, and GT5 earthquakes produced by Reciprocal Cluster Analysis we show that ignoring the correlated error structure leads to rapidly deteriorating error ellipse coverage and location bias with increasingly correlated networks. Calibrated travel-times are responsible for the bulk of location improvements, but they do little to improve error ellipse coverage. While the effect of measurement errors on event locations is small on the order of 1-2 km), they tend to tighten event clusters by reducing the systematic separation between large and small events.