GROUND MOTION PREDICTION EQUATIONS AND SPATIAL CORRELATION
MODEL FOR ISTANBUL– DEVELOPED BY REAL AND SIMULATED GROUND MOTIONS

Hakan Süleyman

(Thesis Supervisor: Prof. Dr. Eser Çaktı)

ABSTRACT

This study develops regional ground-motion models for the Istanbul region by combining instrumental observations (MW 4.0–6.1) with a large suite of synthetic broadband simulations (MW 6.05–7.6). More than 170 stations in and around Istanbul were compiled and uniformly processed for model development. A stochastic simulation approach was first validated against five small-to-moderate Marmara earthquakes in both time and frequency domains. Following validation, broadband motions were generated for five historical Istanbul scenarios (1509 MW 7.6; 1766 MW 7.0–7.1; 1894 MW 6.8–6.9) and 13 additional adapted global scenarios (Mw 6.05–7.6) on predefined rupture models, by merging the stochastic simulations with available 3-D physics-based deterministic simulations to cover 0.1–20 Hz range. Three ground-motion prediction models (GMPEs) are constructed and evaluated: (i) an empirical GMPE calibrated on Marmara Sea events recorded in Istanbul (MW 4.0–6.1); (ii) a machine-learning–enhanced GMPE trained on the same set of observations; and (iii) a synthetic-only GMPE for MW 6.05–7.6 that addresses the lack of regional recordings for moderate-to-large events. Predictive performance is evaluated against well-established GMPEs through quantitative comparisons. To characterize spatial variability, empirical semivariograms of residuals are developed and fitted with exponential models, revealing a clear period-dependent correlation structure.

Overall, integrating observations with physically consistent broadband simulations improves predictive skill in magnitude–distance ranges that are poorly sampled by data and yields region-specific inputs suitable for engineering applications in Istanbul and the wider Marmara region.