Linearized Regional Seismic Loss Estimation with PPCA

This software provides Python code to simulate regional seismic loss estimation. The software is also available through DesignSafe: https://doi.org/10.17603/ds2-9hbb-c610

It uses the computational framework proposed by Houng and Ceferino (2026) (https://doi.org/10.31224/6578), which:

1. Linearizes the ground-motion–fragility coupling
2. Incorporates Probabilistic Principal Component Analysis (PPCA) for dimensionality reduction, enabling more efficient simulations.

This approach reduces the computational complexity from O(N³) in traditional frameworks to O(N²).

The framework has been validated using a building portfolio in San Francisco, demonstrating high accuracy within the stochastic error range of Monte Carlo (MC) simulations:

• >95% agreement in modal damage state
• <2.5% error in loss estimation

For large portfolios containing approximately 30,000 buildings, the method achieves about 110× faster simulations compared to conventional testbeds.

---

How to Run the Code

Below is a simple instruction to run the code.

Directory Structure

Input Data

data/ contains the input data for the simulation:

• SanFrancisco_buildings_ext.csv
  Target building inventory composed of ~15K buildings in downtown San Francisco.

• fragility_PGA.csv
  Fragility curves depending on structural types.

• consequence_repair_PGA.csv
  Repair cost and repair time associated with each damage state.

Code Files

code/ contains scripts for running simulations and generating figures:

• run_proposed.py
  Runs the proposed framework.

• run_traditional.py
  Runs the traditional framework.

• run_figure_preprocessing.py
  Preprocesses the output from run_proposed.py and run_traditional.py.

• run_figure.py
  Generates figures comparing the loss curves from the traditional and proposed frameworks.

Output

Simulation results and figures will be saved in: out/ and out/figs