About GeoRisk

GeoRisk uses the National Liquefaction Hazard Maps by Zdanovski, Sanger and Maurer (2026). Their research combines a geotechnical model with the 2023 U.S. National Seismic Hazard Model to estimate liquefaction-related deformation at the ground surface. The national mapping study was released as a research preprint; it is not an official California hazard-zone designation.

The source supplies an ensemble return period, T, for ground failure on a roughly 90 metre grid. GeoRisk calculates a 15-year probability using p = 1 − exp(−15/T), which assumes a steady annual event rate. A return period is an average rate, not a schedule.

We combine 12 × 12 source pixels into cells about 1 km across and retain the lowest and highest probabilities, rounded outward. A ZIP result spans cells whose centers fall within its Census boundary; very small ZIPs use one overlapping cell. An address result uses the cell containing the estimated address point. Cells can include land outside a ZIP, and missing source pixels stay missing. The source excludes some terrain, including slopes above 5 degrees, so a statewide dataset does not provide a probability for every location.

The displayed range describes differences across the modeled area. It does not include all model uncertainty and is not the chance that a particular home will be damaged. The model is less certain where it relies on estimated ground conditions. Use detailed local studies and qualified site evaluations for property or construction decisions.

Derived from Zdanovski, V., Sanger, M., and Maurer, B. (2026), U.S. National Liquefaction Hazard Maps, under the Open Data Commons Attribution License. The calculation and regional aggregation are GeoRisk modifications.

A ZIP result covers the whole area. An address result checks an approximate point and zones within 100 metres. Check that the dot is in the right place before using the result for a property.

A hazard zone marks an area to investigate, not a prediction that damage will occur. Blank areas may not have been mapped. The number of map shapes tells you how many overlap your search, not the chance of an event or the amount of land affected.

These maps cannot assess a building’s condition. Local ground conditions and construction affect how a particular home performs.

This estimate uses the original December 2023 USGS shaking dataset. USGS released revised hazard curves in August 2026; those revisions are not included here.

Strong shaking means MMI VI or higher: heavy furniture can move and minor damage can occur. This describes shaking at a place, not the magnitude of an earthquake.

The USGS source combines two shaking measures into a 100-year average. That average alone cannot give one exact 15-year percentage. GeoRisk calculates a range, assuming the annual rates stay constant. It is an estimate of shaking in the area, not a prediction of when an earthquake will happen.

For each measure, p₁₅ = 1 − (1 − p₁₀₀)⁰·¹⁵, with probabilities written as fractions. The bounds cover the pairs of measures that could produce the published average. They are not a confidence interval and do not capture all model uncertainty. Each result includes the source value, model-point distance, and range before rounding. USGS probability background.

We run MiniLM, a pretrained language model, on our server. It ranks a fixed set of preparation steps; it doesn’t write new safety instructions or calculate hazard probabilities. Your location is not sent to it.

Rules keep essential steps for your selected household needs and mapped tsunami zones in the plan. Tasks you set aside remain available. If the model is offline, a general checklist still works.

Census locates your ZIP code or address. Hazard services receive the area needed for the check. OpenStreetMap and Esri supply the map tiles and receive the area you view and your IP address.

When you make a plan, the AI server receives household choices, hazard categories, and IDs for completed or deferred steps. It receives no address, ZIP code, coordinates, or notes.

To limit abuse, we use daily-changing hashes of IP addresses; request counters expire within two days. Hosting providers may keep access and security logs.

We count checks, failed requests, downloads, shares, and feedback to see what works and what needs fixing. We store daily totals, without locations, household answers, or visitor identifiers. These count events, not unique people, and are removed after about a year.

Reports download directly as PDFs created in your browser. A ZIP report includes the ZIP code. An address report leaves out the address, coordinates, and map. Neither includes household answers or notes.

Saved plans include your progress and notes, without your search location. Download your plan before leaving to keep a copy. Share links include the ZIP for a ZIP search; address searches share only the homepage.