DID YOU KNOW? DTIC has over 3.5 million final reports on DoD funded research, development, test, and evaluation activities available to our registered users. Click HERE
to register or log in.
Estimating the Uncertainty and Predictive Capabilities of Three-Dimensional Earth Models
WESTON GEOPHYSICAL LEXINGTON MA
Pagination or Media Count:
Many three-dimensional models of seismic velocity structure in Eurasia have been developed in recent years by the seismic nuclear monitoring community. Most of these models are not accompanied by quantitative estimates of uncertainty, either in the model velocities themselves or in geophysical observables predicted by the models e.g., body-wave travel times. Moreover, the various 3D models produced by these studies have not been compared to one another for their predictive capabilities in any meaningful way. We have recently begun a new effort to address these issues, which will culminate in a comprehensive evaluation of the current generation of 3D seismic velocity models. In this paper we show the results of applying two familiar validation techniques, or model evaluation metrics, to three seismic velocity models. The evaluation metrics are regional travel-time prediction and event relocation, each using a ground-truth GT dataset that includes events with epicenters known to 7 km or better and regional P and S arrivals within the footprint of the model region. The models include the Joint WestonMIT JWM crust and upper-mantle velocity model for south-central Asia, which was derived by jointly inverting a large set of body-wave Q travel times and surface-wave group velocities in a coupled nonlinear procedure. We also derived models from the body-wave and surface-wave datasets separately, using the same initial model, inversion grids, constraints and regularization employed in the joint inversion. To make comparisons with the JWM model possible, we applied the Poissons ratio of the initial model to convert the P velocity model constructed with travel times to an S velocity model and vice versa. The results of these exercises reveal many factors that complicate the straightforward evaluation of the models.
APPROVED FOR PUBLIC RELEASE