A non-parametric model of ground motion parameters for shallow crustal earthquakes in Europe

The current study focuses on deriving ground motion models (GMMs) for 21 ground motion parameters derived from data sourced from the Engineering Strong Motion (ESM) database. These parameters include Peak Ground Acceleration (PGA), Peak Ground Velocity (PGV), Peak Ground Displacement (PGD), PGV-to-PGA ratio, (V/H) PGA ratio Predominant Frequency ($F_p$), Central Frequency ($\Omega$), Spectral Parameter ($q$), Significant Duration ($T_{Sig}$), Root Mean Square Acceleration ($A_{rms}$), Arias Intensity ($I_a$), Cumulative Absolute Velocity (CAV), Characteristic Intensity ($I_C$), Acceleration Spectrum Intensity (ASI), Velocity Spectrum Intensity (VSI), Total Energy ($E_{acc}$), Spectral Centroid ($E_w$), Spectral Standard Deviation ($S_w$), Temporal Centroid ($E_t$), Temporal Standard Deviation ($S_t$), and Correlation between time and frequency [$\rho \left(t,\omega \right)$]. Both horizontal and vertical components are considered in this study. The inherent random effects within ground motion regression, encompassing inter-event, inter-site, inter-locality, and inter-region variabilities, are addressed using cross-nested mixed effect regression utilizing a non-parametric GMM approach employing Artificial Neural Network (ANN). Quantitative assessment of the models involves correlation coefficients for regression through the origin and error measures like mean squared error and mean absolute error. These findings of the assessment confirm reliable estimates of Ground Motion Parameters (GMPs). A comparison of GMPs computed using the proposed model and those reported in the literature indicated model's superior performance. Furthermore, satisfactory performance of the proposed GMM in ground motion simulation for the ESM region is demonstrated.