Liquefaction & Ground Failure

Scientific framing¶

Earthquake shaking can drive saturated, loose granular soils to lose strength and behave as a fluid, producing settlement, lateral spreading, and surface ejecta. The trigger is seismic, but susceptibility is set by the ground itself — how loose it is, how deep the water table sits, and what the deposit is made of.

That split is what makes liquefaction a coupling problem for GAIA rather than a purely seismic one. Shaking comes from the earthquake side; saturation comes from the same Pillar 1 hydromechanical state the landslide models use. A wetter season, a drought, or a rising sea level changes the hazard without changing the earthquake.

State variables & observables¶

Water-table depth is the dominant control and the one that varies in time.

Data — what we ingest¶

Cone penetration tests are the backbone: ~37,000 of them, compiled globally, are what the GAIA model is trained against and what anchors its predictions where they exist Sanger et al., 2024Rasanen et al., 2024. Around them sit geospatial proxies for soil thickness, saturation, and typology — terrain and hydrologic derivatives, modeled groundwater, $V_{s30}$, surface geology — plus ground motion from USGS ShakeMaps. Sources, resolutions, and caveats are in the Data Inventory.

Models¶

GAIA uses the mechanics-informed geospatial surrogate of Sanger et al., 2025, which predicts what a CPT-based geotechnical analysis would say at locations with no CPT, and updates those predictions with subsurface data where it exists. Full treatment on the Liquefaction Model page.

Evaluation & metrics¶

Scored against observed liquefaction from past earthquakes using Brier score and calibration curves; the 2001 Nisqually record Rasanen et al., 2023 is the regional target. Definitions in HazEvalHub.

Connection to use cases¶

Central to the 2001–2031 Nisqually earthquake use case, and to Cascadia scenario planning.

Open questions & roadmap¶

• Make the water table a live model input rather than a static training value — the step that turns a static hazard map into a twin that responds to season, drought, and sea-level rise.

References¶