Optimal site hazard grid for probabilistic risk assessment: a two-step approach

The availability of high-resolution open databases detailing building and population distribution has enabled the development of detailed exposure models at regional, national, and global scales. These databases are often used alongside high-performance computing clusters to perform probabilistic seismic risk analyses, simulating thousands or even hundreds of thousands of years of seismicity. However, such analyses may be infeasible on standard laptops or under time constraints where quick results are needed.

To address this challenge, we propose and implement a methodology to determine optimal grids for hazard calculation sites without compromising the accuracy of risk metrics, such as loss exceedance curves and annual average losses. The methodology consists of two main steps: (i) identification of Homogeneous Amplification Zones (HAZ) and (ii) generation of an optimal hazard grid based on exposed elements and HAZ.

In step (i), the initial hazard grid is used to estimate the expected seismic amplification based on a target amplification function. Users have three options for incorporating amplification data: using pre-implemented amplification functions (covering both linear and nonlinear models from peer-reviewed studies), importing custom amplification functions in a CSV format compatible with the OQ framework, or directly inputting an initial grid of amplification functions. The estimated amplification values are then used to cluster hazard sites with similar amplification characteristics using the k-means algorithm, leading to a number of HAZ.

In step (ii), each HAZ identified in the first step is assigned a target number of hazard sites using a k-means weighted methodology considering target information from exposed values, such as exposed structural value. This process leads to integrating data from hazard and risk inputs. Finally, an optional coordinate-based aggregation step removes redundant sites based on a specified resolution, further optimizing the grid.

We tested the proposed methodology at both national and urban scales, applying various site effect methodologies and scales. Our findings demonstrate that the algorithm significantly reduces computational resource demands (both time and memory) with minimal impact on the final risk metrics. These results highlight the practical potential of our approach for large-scale probabilistic seismic risk assessments.