3rd Croatian Conference on Earthquake Engineering 3CroCEE

Earthquakes pose a significant threat to the life safety and well-being of a large portion of the world’s population. The effects of sedimentary basins on infrastructure damage have been observed in numerous earthquakes that have occurred in the past few of decades. The United States Geologic Survey (USGS) incorporated basin effects into the development of the recent National Seismic Hazard Model (NSHM) and found that ground shaking levels increased by as much as 50% in areas overlying deep sedimentary basins. The state-of-the-art in assessing the effects of sedimentary basins in ground shaking is rooted in traditional statistical methods and assumptions regarding geomorphological homogeneity. Specifically, quantifying the effect of sedimentary basins on ground shaking intensity and duration is currently addressed using ergodic procedures, which employ a median derived from the site response component of the Next Generation Attenuation-West2 (NGA-W2) ground motion models. These ground motion models (GMMs) relate the amplification due to one-, two-, and three-dimensional effects to the time averaged shear wave and the isosurface depth. Because they adopt a statistical approach, the challenge of having multiple confounding variables (many of which are correlated) is not rigorously addressed in these studies. We take an interdisciplinary approach to quantifying the effect of basin geomorphology on ground shaking. More specifically, we will use causal machine learning to isolate the effect of multiple geomorphological features (individually and collectively) on ground shaking and then estimate the causal relationship. The goal is to use principles and methods of causal inference to elucidate and quantify the effect of multiple geomorphological features (individually and collectively) on seismic ground shaking to structural response and damage. The proposed work represents a potential paradigm shift where drawing insights from earthquake scientific and engineering data moves “beyond statistics” to more explicit consideration of cause and effect.