INSURTECH • BEIRUT
Probabilistic Seismic Hazard Analysis Platform for Earthquake Risk Underwriting
When earthquake risk assessment costs $50,000 and takes 4 weeks, you build a platform that does it in 30 seconds.
- Client
- North Assurance
- Location
- Beirut, Lebanon
- Industry
- Insurance / Seismic Engineering
- Year
- 2025
- Duration
- Ongoing
- Team
- 2 engineers
Cost reduction vs. traditional PSHA consulting
Full hazard-to-loss analysis (was 2–4 weeks)
NGA-West2 GMPEs in logic tree ensemble
Deviation from USGS benchmark results
Earthquake insurance underwriting in Lebanon was effectively flying blind. Traditional Probabilistic Seismic Hazard Analysis required hiring specialized consultants at $30,000–60,000 per engagement, with 2–4 week turnaround times. The math is extraordinarily complex: you need to characterize fault sources using moment-balanced magnitude-frequency distributions, predict ground motion across dozens of earthquake scenarios using multiple competing scientific models (GMPEs), integrate over all possible magnitude–distance–epsilon combinations, and then translate shaking intensity into building damage and financial loss through multi-component vulnerability functions. North Assurance needed this capability in-house, automated, and fast enough to quote earthquake policies in real time.
The Science: PSHA from First Principles
We implemented the full USGS NSHM-standard PSHA methodology from scratch, integrated with the OpenQuake Engine for validated seismic computations. The pipeline starts with seismic source characterization — modelling the Mount Lebanon Thrust (159 km offshore fault, capable of M7.4) and the Yammouneh Fault (170 km strike-slip, capable of M7.5) using tectonic moment balancing: Ṁ₀ = μ × A × slip_rate, which determines the Gutenberg-Richter magnitude-frequency distribution. Ground motion is predicted using a logic-tree ensemble of four NGA-West2 models — Abrahamson-Silva-Kamai, Campbell-Bozorgnia, Chiou-Youngs, and Boore-Stewart-Seyhan-Atkinson — each weighted at 25% to capture epistemic uncertainty. The hazard integration combines all possible earthquake scenarios using the total probability theorem, producing annual exceedance probability curves for peak ground acceleration and spectral acceleration at standard return periods (72, 475, 975, 2475 years).
From Shaking to Dollars: Multi-Component Risk Engine
The risk engine translates seismic hazard into financial loss using a FEMA P-58 inspired multi-component framework. A single building is decomposed into four damage components, each responding differently to earthquake shaking: Structural elements (22% weight) respond to long-period motion at SA(1.0s). Drift-sensitive non-structural components like partitions and cladding (33%) also respond to SA(1.0s). Acceleration-sensitive components — ceilings, HVAC, mechanical systems (18%) — respond to short-period SA(0.3s). Building contents — equipment, furniture, inventory (27%) — also respond at SA(0.3s). Each component has its own vulnerability function mapping shaking intensity to damage ratio. The system computes Expected Annual Loss by integrating the loss exceedance curve, then derives insurance-grade metrics: Value at Risk, Tail Value at Risk, Probable Maximum Loss at configurable return periods, and full portfolio aggregation with spatial correlation.
Engineering for Speed: Real-Time Scientific Computing
The backend is a FastAPI application with domain-driven hexagonal architecture — seven API routers covering hazard, risk, insurance metrics, analysis workflows, seismic source presets, moment balance, and batch operations. Vectorized NumPy operations make GMPE computations 100–200x faster than sequential loops. A two-tier caching strategy (in-memory LRU + SQLite disk cache) achieves 70–85% hit rates for portfolio analyses. The async workflow engine handles long-running computations — batch analyses across hundreds of sites — via an AsyncIO worker pool with WebSocket progress streaming. The frontend is a React/TypeScript application with TanStack Query for server state, Zustand for UI state, Recharts for interactive hazard curves and loss distributions, and Leaflet for geospatial visualization of regional hazard maps. The analysis run system chains hazard → risk → insurance computations automatically with full data provenance.
- Single-site hazard analysis in 1–2 seconds — down from weeks of consulting engagement
- Full multi-component risk assessment (hazard + vulnerability + loss) in under 30 seconds
- Results validated within 2% of USGS benchmark computations
- Portfolio analysis across 100+ sites completes in 5–10 minutes with spatial correlation
- Insurance metrics suite: VaR, TVaR, PML, EML at any confidence level or return period
- Platform extensible beyond Lebanon — architecture supports global fault databases
This platform turned earthquake risk from a black box we paid consultants to open into a tool our underwriters use daily. We can now quote earthquake policies with the same speed and confidence as motor or property.