is to establish the fundamental mode (M0) dispersion curve as accurately as possible, which has been one of the key issues with data acquisition and processing in the history

of surface wave applications. Theoretical M0 curves are then calculated for different earth models by using a proper forward modeling scheme, the most common type of which

is the one by Schwab and Knopoff (1972), to be compared against the measured (experimental) curve (

measured dispersion curve represents the M0 curve only not influenced by any other modes of surface waves. On the other hand, the concept of composite (or apparent)

dispersion curve is used in the SASW method accounting for the multi-modal influence as the modal separation cannot be accomplished only with two receivers.

Key issue with this inversion approach has been the optimization technique to search for the

most probable earth model among many other candidates as much efficiently as possible. The

root-mean-square (R-M-S) error (**Fig. 2**) is usually used as an indicator of the closeness between

the two dispersion curves (measured and theoretical), and the final solution is chosen as the 1D

Vs pofile resulting in a preset (small) value of R-M-S error. Either a deterministic method such as

the least-squares method (Menke, 1989; Xia et al., 1999) or a random approach (Socco and

Boiero, 2008, for example) is taken for the optimization (**Fig. 3**). The former type is usually faster

than the latter type at the expense of the increased risk of finding a local, instead of global,

minimum. Research issues in each type therefore have been how to reduce the level of this risk

with the former type, whereas how to improve the speed with the latter type. Another pitfall

common to both types (perhaps intrinsic to all types of inversion) is the risk of numerical artifacts.

For example, although a solution with a smaller R-M-S error is numerically correct, it may not

necessarily represent a more realistic one (**Fig. 4**).

most probable earth model among many other candidates as much efficiently as possible. The

root-mean-square (R-M-S) error (

the two dispersion curves (measured and theoretical), and the final solution is chosen as the 1D

Vs pofile resulting in a preset (small) value of R-M-S error. Either a deterministic method such as

the least-squares method (Menke, 1989; Xia et al., 1999) or a random approach (Socco and

Boiero, 2008, for example) is taken for the optimization (

than the latter type at the expense of the increased risk of finding a local, instead of global,

minimum. Research issues in each type therefore have been how to reduce the level of this risk

with the former type, whereas how to improve the speed with the latter type. Another pitfall

common to both types (perhaps intrinsic to all types of inversion) is the risk of numerical artifacts.

For example, although a solution with a smaller R-M-S error is numerically correct, it may not

necessarily represent a more realistic one (

Fig. 1. |

Fig. 2. |

Fig. 3. |