Influence of site effect modelling approaches in seismic risk assessment

For decades, both qualitative and quantitative observations of earthquake intensities have highlighted the impact of surface geology on the characteristics of the ground shaking. However, site effects are often inadequately incorporated into seismic risk assessments. Many hazard and risk models simplify the consideration of site effects by using average soil properties in the upper 30 meters (Vs30), even though this approach neglects the influence of entire basins, as well as additional factors such as lateral heterogeneities, geomorphological features, and topographic properties of the landscape.

While various proxies have been developed to streamline the estimation of site effects with high accuracy, the adoption of site-response proxies based on Vs30 over the past two decades favoured practicality over rigorous validation. These generic proxies, often embedded in seismic codes, are often not subjected to thorough evaluations. In this study, we investigate how the use of simplified approaches impacts on earthquake loss assessments.

We evaluated over 20 approaches for modelling site effects, incorporating diverse proxies, methodologies, and geographical scales. Assuming seismic zonation studies (SZS) as the most accurate methodology and the benchmark, we conducted probabilistic risk assessments for five cities in Colombia, with varying geological conditions and seismic sources. These case studies utilized detailed exposure models and SZS-derived amplification functions at different intensity levels, integrating geotechnical, geophysical, and geological characterizations validated through numerical modeling and seismic records from local and national networks.

Our findings indicate that while the geographical resolution of the soil parameter used as a proxy (e.g Vs30) used to estimate site effects has limited impact on risk metrics, the chosen methodology significantly influences results. Additionally, consistent with previous hazard studies, we observed that incorporating the non-linear behavior of soil is crucial to avoid overpredicting the impact. These conclusions were derived from comparative analyses of risk metrics, including risk curves for return periods up to 1,000 years and annual average losses.

Based on our results, we offer two key recommendations for earthquake risk modelers:

• To focus not only on improving spatial resolution of soil parameters used as a proxy to assess site effects but also on validating amplification values using local data and numerical modeling, as this step critically shapes the final outcomes.
• Evaluate site effect approaches on a case-by-case basis, as simplified models fail to fully capture the complexities of soil response, leading to varying degrees of over- or underprediction.