Fig 1.
The red polygon indicates the study boundaries.
Fig 2.
Fault map of the study area.
Fig 3.
Seismicity-based identification of active faults.
Fig 4.
The maximum magnitude distribution of the study area.
Adapted from [52].
Fig 5.
The white polygon indicates the study boundaries.
Table 1.
Site classification in AzDTN based on VS30 values.
Fig 6.
The site classification map of the study area.
A negligible extent of Site Class IV is observed and therefore excluded from the synthetic dataset generation.
Fig 7.
Schematic representation of the logical sequence of the four accelerogram generation methods implemented in the Seismoartif software.
(a) “Synthetic accelerogram generation & adjustment” method. (b) “Artificial accelerogram generation” method. (c) “Artificial accelerogram generation & adjustment” method. (d) “Real accelerogram adjustment” method.
Fig 8.
Acceleration spectrum functions defined by AzDTN for the site classes.
(a) II and (b) III.
Fig 9.
The validation plots of the synthetic acceleration time series.
(a) Mean Error. (b) Coefficient of Variation.
Fig 10.
Distributions of the synthetic earthquake dataset.
(a) Magnitude variation with respect to hypocentral distance. (b) Number of synthetic earthquakes by magnitude. (c) Number of synthetic earthquakes by hypocentral distance. (d) Hypocentral distribution of PGA values.
Table 2.
ANN Performance Metrics in Linear Space (Gals).
Table 3.
XGBoost Performance Metrics in Linear Space (Gals).
Table 4.
RF Performance Metrics in Linear Space (Gals).
Table 5.
SVM Performance Metrics in Linear Space (Gals).
Table 6.
A&K-1979-1 Performance Metrics in Linear Space (Gals).
Table 7.
Comparative Performance Leadership Table (Gals).
Fig 11.
Comparison of models-predicted and synthetic PGA values and corresponding residuals.
(a) Models-predicted versus synthetic PGA values. (b) Residuals of PGA predictions for each model.
Fig 12.
Permutation feature importance analysis for the developed ML-GMMs.
(a) ANN. (b) XGBoost. (c) RF. (d) SVM.