HR: 11:05h
AN: S42A-04    [Abstracts]
TI: Seismic Tomography Using Adjoint Methods: Experiments Using Membrane Surface Waves
AU: * Tape, C
EM: carltape@gps.caltech.edu
AF: Caltech Seismological Lab, 1200 E California Blvd. MC 252-21, Pasadena, CA 91125
AU: Liu, Q
EM: lqy@gps.caltech.edu
AF: Caltech Seismological Lab, 1200 E California Blvd. MC 252-21, Pasadena, CA 91125
AU: Maggi, A
EM: alessia@gps.caltech.edu
AF: Caltech Seismological Lab, 1200 E California Blvd. MC 252-21, Pasadena, CA 91125
AU: Zhou, Y
EM: yingz@gps.caltech.edu
AF: Caltech Seismological Lab, 1200 E California Blvd. MC 252-21, Pasadena, CA 91125
AU: Tromp, J
EM: jtromp@gps.caltech.edu
AF: Caltech Seismological Lab, 1200 E California Blvd. MC 252-21, Pasadena, CA 91125
AB: We employ adjoint methods in a series of seismic tomography experiments to recover surface wave phase velocity maps of southern California. Our technique involves computing the Fréchet derivatives for tomographic inversions via the simultaneous interaction between a forward wavefield, propagating from source to receivers, and an adjoint wavefield, propagating from receivers to source. The forward wavefield is computed using a 2-D spectral element method and a phase velocity map for southern California. A reference model, m(r) (not necessarily homogeneous), is used to generate the synthetic wavefield, s. A (frequency-dependent) phase velocity map is used to generate the `data' wavefield, d, which is saved at the receivers. We then specify an objective function, χ(m), that defines the measure of misfit between the data and synthetics. This misfit, defined at each receiver, is time-reversed and then used as the source of the adjoint wavefield, s^†. For each event, the interaction between s and s^† is used to construct finite-frequency sensitivity kernels, K_m^e(x). The overall sensitivity is simply the sum over all events: K_m(x) = ∑_e K_m^e(x). The summed kernel is used to compute the Fréchet derivatives of the objective function via δχ=∫_V K_m(x) δln m(x) d³x. Using K_m(x), we generate basis functions that are optimally concentrated where the kernel is large, which reduces the number of parameters needed in the inversion. The tomographic inversion generates a new reference model, from which new kernels are computed, and thus the model iteratively improves. Thus, using a 2-D numerical model, we investigate several aspects -- e.g., the type of measurement, the source-receiver geometry, the weighting of the measurements, the parameterization of the model -- in preparation for 3-D tomographic inversions at the scale of southern California and the globe.
DE: 7255 Surface waves and free oscillations
DE: 7260 Theory
DE: 7270 Tomography (6982, 8180)
DE: 7290 Computational seismology
SC: Seismology [S]
MN: Fall Meeting 2005