About the BBP API

Why a Simulation, Not Just a Prediction Equation

A ground motion prediction equation (GMPE) answers a narrow question: given a magnitude, a distance and a site condition, what is the median peak acceleration — or spectral acceleration at one period — and how much scatter is there around it? The answer is a handful of numbers at a point. That is enough to shade a map, and it is what most rapid shaking products report.

A ground motion simulation answers a larger one. It returns a full three-component time series — the actual motion of the ground, second by second, in all three directions. Everything a scalar leaves out is in there:

• Duration. Two sites can share a peak acceleration and shake for ten seconds or for sixty. Damage does not care only about the largest cycle.
• Frequency content. A structure responds at its own period. A time series says what energy arrived near that period; a single peak value cannot.
• Arrival and sequence. P-wave, S-wave, surface waves, and the gap between them — which is what makes a simulated record look and behave like a recording rather than a bar chart.
• Near-field and directivity effects. Rupture propagating toward a site produces large velocity pulses that a distance-based regression averages away.
• Basin response. Sedimentary basins trap and amplify long-period energy in a way that depends on the wave path, not just on the site's V_S30.

That output is what an engineer needs for a response-history analysis, and it is what drives the shaking animations published here — neither is possible from a peak value alone.

There is a second difference, about where each method's confidence comes from. A GMPE is a regression fitted to recorded earthquakes, so it is best constrained where recordings are plentiful — moderate magnitudes at moderate distances — and thinnest exactly where the stakes are highest: large ruptures, close in. A Broadband Platform seismogram is instead computed, propagating a kinematic rupture model through a regional crustal velocity model, so it is not limited to the magnitude-distance range that happens to be well sampled.

What the BBP API Is

Running the SCEC Broadband Platform yourself means installing a Fortran scientific code, downloading tens of gigabytes of Green's functions per region, and finding a machine to run it on. BBP Services hosts the Broadband Platform as a metered REST API so you do not have to. The exact build a job runs — its version, the source commit it was compiled from, and the SCEC data release its Green's functions come from — is shown in the simulation builder and returned by GET /v1/catalog/versions.

Configure a scenario from a real earthquake or design a hypothetical rupture by hand, place your sites, see what the compute will cost before you commit to it, then submit the job and download the resulting simulated seismograms — no cluster, no Fortran, no Green's functions to install. The Simulation Lab follows the same sequence of questions as the Broadband Platform's own command-line program, so a seismologist who already runs BBP from a terminal can express the same run here.

The hosted service is being prepared for launch; running metered simulations requires an approved account. The API reference documents the interface so you can plan an integration, and the System Update Log carries availability news.

How We Check It

A simulation that looks convincing is not necessarily accurate, so each qualifying event is measured against the real earthquake: recorded accelerations are retrieved from seismic stations around the epicenter and compared with the simulation of the same event. We publish the resulting bias — period by period, not as a single average, because a simulation can pass on the mean while being well off at the period a particular structure cares about — along with the spectra and the caveats. See Verification & Validation.

What These Results Are Not

• Not design values. These are research and visualization products illustrating the character of shaking, not site-specific engineering ground motions for design.
• The rupture is derived, not observed. Fault dimensions come from magnitude scaling, and a moment tensor's second nodal plane is equally consistent with the data — choosing it would produce different waveforms.
• Calibration range. The Graves & Pitarka method is validated for finite-fault ruptures of roughly M5.5 and above; smaller qualifying events run below that range and carry the caveat on their report.
• A velocity model is a 1-D crustal profile, not a map. An event outside every installed region's native extent is simulated against the nearest one and the result is labeled an approximation, and nothing in the platform checks that a rupture's tectonic style matches the crustal profile it runs against.

The full list of standing limitations lives on Physics-Based Simulation (BBP).