ON THE SENSITIVITY OF GROUND MOTION PREDICTION EQUATIONS IN GREECE
Ground motion prediction equations, widely known as attenuation relations, are common input for probabilistic and deterministic seismic hazard studies. The construction of a ground motion model to describe such a complex phenomenon as the effects of seismic wave propagation is highly dependable on a number of parameters. The quality and the distribution of strong motion data, which is the original input for the calculation of any ground motion model, can be thought as one of the main parameters that heavily influence the form of ground motion prediction equations. The selected processing scheme, involving significant choices about a series of adjustments and filter specifications, implemented to remove low and high frequency noise, is related with the credibility of the calculated ground motion parameters such as the spectral ordinates. Once a set of response variables for a number of predictors is available, the researcher’s interest is related with the mathematical definition of the ground motion model, in terms of selecting the appropriate parameters and the determination of their coefficients of the equation. Another significant part involves the selection of the optimum solver in order to achieve high confidence level coefficients and a computationally inexpensive solution. Each method should be evaluated through statistics but the researcher should bear in mind that residual analysis and statistical errors, although they can adequately represent the efficiency of the mathematical equations, do not always provide information about where our efforts should lie in terms of further improvement. The scope of this paper is to point out the multi-parametric nature of the construction of ground motion prediction equations and how each of the aforementioned development stages influences the credibility of the proposed attenuation relations.