Seismic Risk Atlas - Project Narrative

Inspiration

Most existing earthquake tools stop at shaking intensity. However, decision-makers and communities care about a more fundamental question:

What is the expected financial impact on households?

This gap inspired us to build Seismic Risk Atlas - a system that translates physics-based seismic simulations into tract-level economic risk.

We were particularly motivated by:

• The disconnect between scientific outputs (PGA, PGV) and real-world decisions
  
• The lack of accessible, interpretable risk tools for urban planners
  
• The opportunity to combine ML systems, geospatial data, and physics models into a unified pipeline
  

What We Built

We developed an end-to-end ML-driven system that estimates expected household earthquake loss across 2,498 census tracts in Los Angeles County.

At its core:

$Expected\ Loss = Damage\ Ratio \times Home\ Value \times Structural\ Fraction$

The platform includes:

• Physics-based seismic feature extraction (PGA, PGV, Arias intensity)
  
• HAZUS-based damage modeling
  
• Monte Carlo aggregation across 500 scenarios
  
• Interactive choropleth risk map
  
• Explainable API for natural language summaries
  

Built With

• brev
• databricks
• fastapi
• geojson
• geopandas
• javascript
• leaflet.js
• marimo
• numpy
• nvidia
• openai
• pandas
• pygris
• python
• scipy
• sphinx
• unittest
• xgboost