Model fitting and testing in near surface seismics using maximum likelihood in frequency domain

Imaging the subsurface of seismic near-field scenarios by an environmental parameter inversion technique is considered. The wave propagation along a geophone line array after exciting the elastic wavefield by a seismic source (e.g. a falling weight) is modelled by a suitable forward model. The important forward model parameters, e.g. P-, S-wave velocities or the depths of the Earth-layers are estimated using a broadband maximum-likelihood estimation procedure and a genetic algorithm for solving the complicated optimization problem. An important question is the required number of parameters to describe the observed data accurately. To solve this problem a hypothesis testing procedure with a sequence of models with an increasing number of parameters is used to verify model order against a set of alternatives. The sequence of models is of hierarchical order in which each particular model contains all previous ones as special cases. The technique is applied to broadband data measured by a horizontal geophone line array in a former coal-mining area near Bonn, Germany.