Rapid City‐Scale Earthquake Assessment by Combining Numerical Simulation and Sparse Sensing

This study proposes a framework to assess the seismic risk by integrating city‐scale numerical simulations with sensor data prediction. The study begins with advanced numerical simulations using two primary methods: the integrated earthquake simulator (IES) and the stochastic Green's function method. The stochastic Green function method is used to generate the motion from the fault model to the outcropped engineering bedrock. The IES is used to simulate the ground motion on the surface, considering different scenarios, and the sumulation results are seperated into training and validation datasets. In this study, all scenarios are generated from a single fault model, and variability is introduced by changing the rupture initiation point, representing a limited dataset but practically relevant setting. Proper orthogonal decomposition (POD) is then applied to the training datasets to extract spatial modes, and those modes are used for identifying the most suitable distributions for sensors. In this study, the sensors are strategically placed to maximize data collection efficiency while minimizing the overall number of sensors used. Optimizing sensor distribution is crucial owing to the challenges associated with deploying numerous sensors in actual situations. The distribution of sensors obtained by this method is validated using a test set generated from IES simualation. The case study focusing on Sendai city (Japan), for scenarios associated with the Nagamachi–Rifu fault, finds an optimal solution for the distribution of sensors and allows for an overall prediction of the expected accuracy. The results show that seismic risk prediction and assessment on a city scale can be achieved by this method while maintaining accuracy and the number of sensors. The present work therefore demonstrates the feasibility of the proposed rapid prediction framework under limited dataset conditions for a specific, high impact seismic source, while the extension to multiple faults and magnitudes is left for future work.