Documento Interno — Confidencial

GEM Integration Research — Strategic Intelligence Report

PSHA Intelligence Platform  |  February 2026  |  Confidential

Section 1

Executive Summary

Key findings from 5 parallel research tasks covering GEM data products, licensing, competitor analysis, and engine evaluation.

5 Research Reports Completed
~2,000 Lines of Analysis
5 Parallel Research Agents
100+ Sources Analyzed

Key Findings

Finding 1
Licensing is Clear-Cut
Only GAF-DB (fault database) can be used commercially (CC BY-SA 4.0). Everything else requires paid GEM license (CC BY-NC-SA 4.0). Contact: [email protected]
Finding 2
Hybrid Architecture Recommended
Custom PSHA engine (NumPy/SciPy) for production + OpenQuake in separate Docker container for validation via REST API. Avoids AGPL contamination while gaining scientific credibility.
Finding 3
Competitor Validates GEM Data Approach
Langan International uses GEM fault sources directly + OpenQuake in production PSHA reports (Philippines). This proves GEM data is reliable and accepted professionally.
Finding 4
GMPE Implementation Priority
NGA-West 2 (ASK14, BSSA14, CB14, CY14) with equal 0.25 weights is the universal standard for shallow crustal. Implement these 4 first.
Finding 5
GEM Hazard Map for Benchmarking
Downloaded 165MB GeoTIFF with PGA values globally (~5.6km resolution). Example: Madrid = 0.016g (Low), Granada = 0.201g (High), Tokyo = 0.605g (Very High).

Strategic Decision Matrix

Decision Recommendation Risk Priority
PSHA Engine Custom (NumPy/SciPy) None Phase 1
OpenQuake Usage Validation only (Docker REST) Low Phase 0
Fault Data Source GEM GAF-DB (CC BY-SA 4.0) None Immediate
Earthquake Catalog USGS FDSN (public domain) None Immediate
GEM Hazard Map Internal benchmarking only Medium Phase 1
GEM Risk Profiles Contact GEM for license High Phase 2
GMPE Library Start with 4 NGA-West 2 None Phase 1
Report Format Follow Dynamis structure None Phase 1
NRML Export Parse + generate NRML v0.5 None Phase 2
Tier 0 Product Hybrid: own PSHA + GEM exposure Medium Phase 3
Section 2

Dynamis vs Langan — Professional PSHA Report Comparison

Comparative analysis of two professional PSHA reports for the Philippines, extracting implementation requirements for our platform.

Aspect Dynamis Langan
Pages 91 ~45
Software R-CRISIS v20.0.0 OpenQuake Engine
Theory depth Full derivation (46 equations) References only
Catalog 58,000+ events Not specified
GEM Data Not used Used directly (Figure 2)
Disaggregation Full M-R-ε plots Not included

PSHA Equations to Implement

Core PSHA Mathematics
// Cornell-McGuire Integral
λ(IM > im) = Σi νi ∫∫ P(IM > im | m, r) · fM(m) · fR(r) · dm · dr

// Gutenberg-Richter (truncated exponential)
fM(m) = β · exp(-β(m - Mmin)) / (1 - exp(-β(Mmax - Mmin)))

// Ground Motion Prediction Equation (general form)
ln(IM) = f(M, R, Vs30, ...) + ε · σ

NGA-West 2 GMPEs — Shallow Crustal Standard

GMPE Abbreviation Weight Used By
Abrahamson, Silva & Kamai (2014) ASK14 0.25 Dynamis + Langan
Boore, Stewart, Seyhan & Atkinson (2014) BSSA14 0.25 Dynamis + Langan
Campbell & Bozorgnia (2014) CB14 0.25 Dynamis + Langan
Chiou & Youngs (2014) CY14 0.25 Dynamis + Langan

Logic Tree Structure

Fault Sources
Branches: Fault Dip (3) × Recurrence Model (2) × Mmax (3) × Slip Rate (3) × GMPE (4)
= 216 branches
Area Sources
Branches: b-value (3) × Mmax (3) × GMPE (4)
= 36 branches

Implementation Priority Roadmap

Phase 1 — Core
Foundation
  • Cornell-McGuire integral
  • Source geometry processing
  • Distance calculations (Rrup, Rjb, Rx)
  • NGA-West 2 (4 GMPEs)
  • Gutenberg-Richter fitting
  • Basic logic tree
  • UHS & hazard curves
Phase 2 — Extended
Professional Features
  • Characteristic earthquake model
  • Full logic tree enumeration
  • Disaggregation (M-R-ε)
  • MCEr computation
  • Subduction GMPEs
  • Site amplification (Vs30)
  • Spectral matching
  • PDF report generation
Phase 3 — Advanced
Expert Capabilities
  • 3D complex fault sources
  • Directivity effects
  • Eurocode 8 compliance
  • Bayesian magnitude estimation
Section 3

GEM Data Products — Commercial Licensing Analysis

Comprehensive licensing matrix for all GEM Foundation data products relevant to our commercial SaaS platform.

Product License Commercial Use? Our Action
GAF-DB (Active Faults) CC BY-SA 4.0 YES Use with attribution
Global Hazard Map CC BY-NC-SA 4.0 NO Internal benchmark only
Regional Mosaic (34 models) CC BY-NC-SA 4.0 NO Contact GEM
Risk Profiles CC BY-NC-SA 4.0 NO Contact GEM
Vulnerability Model CC BY-NC-SA 4.0 NO Not needed (Tier 1-3)
Exposure Model CC BY-NC-SA 4.0 NO Contact GEM for Tier 0
OpenQuake Engine AGPL-3.0 Conditional Docker REST only
What We CAN Do
  • Use GAF-DB fault geometries in commercial products
  • Use USGS earthquake catalog (public domain)
  • Use USGS Vs30 raster (public domain)
  • Build custom PSHA engine with own code
  • Run OpenQuake for internal validation (Docker REST)
  • Display fault maps with proper GEM attribution
What We CANNOT Do
  • Use GEM hazard map values in customer reports
  • Use Mosaic model parameters commercially
  • Include risk profile data in SaaS output
  • Redistribute any NC-licensed data to customers
Attribution Requirements for GAF-DB
  • Credit: "GEM Global Active Faults Database, GEM Foundation"
  • License: CC BY-SA 4.0
  • Reference: Styron & Pagani (2020), Earthquake Spectra
GEM Foundation Contact for Commercial Licensing
[email protected] [email protected] [email protected]
Section 4

GEM Global Seismic Hazard Map v2023.1 — Data Analysis

Analysis of the GEM GSHM raster dataset: PGA values at 475-year return period for reference rock conditions.

Format
GeoTIFF
Dimensions
7200 x 3000 px
Data Type
float64
Resolution
0.05 deg (~5.57 km)
Coverage
-180 to +180 lon, -60 to +90 lat
Size
33 MB (ZIP) / 165 MB
DOI
10.5281/zenodo.8409647
License
CC BY-NC-SA 4.0

PGA Values — Major Cities

City PGA (g) Hazard Level
Tokyo 0.605 Very High
Lima 0.573 Very High
Los Angeles 0.429 High
San Francisco 0.414 High
Istanbul 0.369 High
Granada 0.201 High
Murcia 0.199 Moderate-High
Lisbon 0.157 Moderate-High
Barcelona 0.043 Low-Moderate
Madrid 0.016 Low

GEM Hazard Color Palette

0.00 0.01 0.02 0.04 0.08 0.15 0.30 0.50 0.70 0.90 1.50g

30+ Regional Models in the Mosaic

ALSARBAUS CANCCACEA CHNEURHAW IDNINDJPN KORMEXMIE NAFNEANWA NZLOATOIN OPAPACPHL PNGSAMSEA SSATEMUSA WAFZAF

Integration Opportunities

1
Benchmark PGA Lookup
Instant PGA reference value for any coordinate worldwide. Use as a quick sanity check before running full PSHA.
2
NRML Export Compatibility
Philippines regional model provides complete NRML format specification. Parse these files as templates for our own NRML generation.
3
Color Palette for Maps
Adopt the GEM standard color palette for Mapbox hazard overlays. Ensures familiarity for seismic engineers.
4
Validation Benchmark
Compare our custom PSHA results against GEM values. Any significant deviation flags potential errors in our engine.
Section 5

GEM Risk Profiles & Related Repositories

Analysis of three interconnected GEM repositories for potential Tier 0 Regional Hazard Explorer product.

All three repositories are licensed CC BY-NC-SA 4.0 — No commercial use without paid license
gem/risk-profiles
NC
190 countries · ~1.9 GB · Updated Dec 2025
PNG ONLY! No structured data. AAL, loss curves locked in images.
gem/global_exposure_model
NC
215 countries · ~498 MB · Updated Nov 2025
THE GOLD MINE. 1,341 CSV files. 82-column world summary.
gem/global_vulnerability_model
NC
215 countries · ~9 MB · Updated Feb 2026
860 NRML XML files with vulnerability functions.

World Exposure Summary Highlight

World/Exposure_Adm0_Summary.csv
82 columns × 215 rows. Includes: ISO3, population, buildings (Res/Com/Ind), replacement costs, GDP, HDI, building taxonomy...
Spain — Residential Buildings
9.3M
Spain — Replacement Cost
$4.3T
Spain — Avg Building Area
244 sqm
Spain — Avg Cost/Area
$1,451/sqm

Tier 0 Feasibility Assessment

Verdict: Feasible but Constrained
  • Exposure context (building stock, costs) = available as CSV, structured and rich
  • Risk metrics (AAL, loss curves) = PNG only, need GEM license or own computation
  • Recommended: Hybrid approach — own PSHA computation + GEM exposure data with a commercial license agreement
Section 6

OpenQuake Engine v3.24.1 — Evaluation Report

Technical evaluation of OpenQuake as a validation backend, including AGPL license risk analysis and integration strategies.

Version
3.24.1
License
AGPL-3.0
Python
3.11 required
GMPEs
200+
Source Types
7
Release
Nov 2025

AGPL License Risk Assessment

Integration Method Risk Level Source Disclosure Verdict
Python import HIGH Full codebase DO NOT USE
Subprocess call MEDIUM Possibly full codebase AVOID
REST API (Docker) LOW OQ modifications only VIABLE
Internal validation SAFE None RECOMMENDED

OpenQuake REST API Endpoints

Key REST Endpoints
POST  /v1/calc/run          // Submit PSHA calculation
POST  /v1/calc/aelo_run     // Single-site hazard (lon, lat, vs30)
GET   /v1/calc/:id/results  // Get calculation results
GET   /v1/available_gsims   // List all 200+ GMPEs

Integration Strategies

RECOMMENDED
Strategy 1
Hybrid Architecture
Custom PSHA engine (NumPy/SciPy) for production computation + OpenQuake in Docker for cross-validation via REST API. Best of both worlds.
FAST PATH
Strategy 2
OpenQuake-Only via REST
Use OpenQuake as the sole computation engine via Docker REST API. Fastest to market but creates dependency on OQ.
INCOMPATIBLE
Strategy 3
Direct hazardlib Import
Import OpenQuake libraries directly into Python code. Forces AGPL-3.0 on entire codebase. Unacceptable for commercial SaaS.
MAX CONTROL
Strategy 4
Pure Custom Engine
Build everything from scratch. Maximum control and IP ownership. Longer development time but zero licensing risk.

Implementation Roadmap

Phase 0 — Weeks 1-3
Infrastructure
  • Deploy OpenQuake Docker container
  • Build NRML export module
  • REST API client wrapper
  • Basic validation pipeline
Phase 1 — Months 1-3
Custom Engine Core
  • Cornell-McGuire implementation
  • 5 core GMPEs (NGA-West 2 + 1)
  • Source geometry processing
  • Cross-validation vs OpenQuake
Phase 2 — Months 3-6
Extended Features
  • 15-20 GMPEs library
  • Disaggregation module
  • Full logic tree support
  • Tier 2 validation against OQ
Phase 3 — Months 6-9
Full PSHA Platform
  • Complete Tier 3 PSHA
  • Advanced logic trees
  • UHS generation
  • Benchmark suite vs OQ + R-CRISIS