Observed vs Simulated — Goodness of Fit

A physics-based simulation of this earthquake, compared against the seismograms the network actually recorded. M6.4 15km WSW of Ferndale, CA · 2022-12-20 10:34:24 UTC

Download the full report (PDF) Includes per-station waveform and displacement pages. Generated 2026-08-04T07:24:38Z · BBP job job_d6c24a71669e4cc9b1ff

Before reading the result

What this measurement does and does not establish.

Emulated hardware. The simulation ran on arm64 under emulation. Amplitudes are not SCEC-reference-accurate, and this measurement cannot separate the platform from the model.

Velocity model. Simulated with NoCal500. No installed region is native to this epicentre, so the crustal profile is an approximation.

One realization. A single stochastic rupture, not an ensemble. The bias below is one draw; a second run with a different seed would give a different number.

Defaulted Vs30 at 4 of 31 sites. Those sites were simulated with a Vs30 value repeated across three or more stations — a lookup default rather than a site measurement. Vs30 drives the site term, so a defaulted value produces bias at intermediate and long period that is not the model's fault.

What a pass means. Consistency with recordings at these 31 stations, for this event, under this velocity model, on this hardware, within a factor of 1.99. It is not a general statement about BBP and not a prediction of accuracy at an unrecorded site.

Result

Mean within pass band 31 stations · 63 spectral periods · BBP simulated

mean |bias|: 0.165; a factor of 1.18×
pass band: 0.69; a factor of 1.99×, declared before the run
sigma: 1.244; station-to-station scatter
cross-check: 0.327; recomputed here from the waveforms

The metric is ln(observed / simulated) RotD50 spectral acceleration, averaged over stations. Positive means the simulation under-predicts — real shaking was larger. Basis: BBP RotD50 residuals (authoritative).

Residual by period

Mean ln(obs/sim) at each spectral period. A structure responds at its own period, so agreement at one period says nothing about another.

Period	Bias	Simulation	Factor	vs. band
0.1s	-0.512	over-predicts by 67%	1.67×	within
0.3s	-0.283	over-predicts by 33%	1.33×	within
1s	-0.135	over-predicts by 14%	1.14×	within
3s	+0.178	under-predicts by 19%	1.19×	within

Every period compared falls within the 0.69 band.

Goodness of fit across the spectrum

Mean residual with its ±1σ band, against the pass band. Diamonds are the Broadband Platform’s own RotD50 residuals; the line is this report’s independent recomputation from the waveforms.

Where the stations are

Each station coloured by its PGA residual. Red means the simulation under-predicted the recording; blue means it over-predicted.

Map of compared stations coloured by residual

Station summary — three independent sources

31 stations with both a recording and a simulated seismogram. Scroll sideways for the full set of columns.

Station	Dist (km)	Vs30 BBP	Vs30 ShakeMap	Chan	PGA (g) ShakeMap	PGA (g) obspy	PGA (g) BBP	PGV (cm/s) ShakeMap	PGV (cm/s) obspy	PGV (cm/s) BBP	PGD (cm) obspy	PGD (cm) BBP	PGD obs/sim	ln(obs/ sim)	Flags
NC.KCT	9.1	466	466	HNE	0.1239	0.1189	0.3334	30.0	30.2	27.1	5.70	8.08	0.71x	-1.031	—
NP.1584B	11.4	458	458	HNE	0.2212	0.0190	0.2577	10.3	0.7	17.5	—	7.06	—	-2.608	—
NP.1023	14.6	296	296	HNE	0.2561	0.0465	0.3508	22.3	1.5	43.0	—	15.34	—	-2.022	—
NP.1586	21.6	384	384	HNE	0.3000	0.2968	0.1976	25.8	27.3	19.6	6.30	7.74	0.81x	+0.407	—
CE.89486	24.3	339	339	HNE	0.3938	0.3974	0.1934	28.7	34.9	22.3	7.45	7.44	1.00x	+0.720	—
CE.89255	24.5	382	382	HNE	0.8010	0.6483	0.1918	34.8	59.7	23.4	19.17	6.65	2.88x	+1.218	—
BK.PETL	27.2	347	347	HNE	0.1236	0.1232	0.2198	8.5	8.5	48.5	2.15	12.82	0.17x	-0.579	—
CE.89462	27.3	302	302	HNE	1.4387	1.3867	0.1864	63.4	178.6	21.5	124.56	6.63	18.78x	+2.007	—
NP.1582	27.4	334	334	HNE	0.4536	0.0719	0.1724	34.1	2.3	24.2	—	8.10	—	-0.875	—
NC.KMPB	28.2	490	490	HNE	0.1237	0.1226	0.1539	11.7	12.5	10.5	3.56	3.81	0.93x	-0.228	—
NP.1581	29.6	326	326	HNE	0.2748	0.2718	0.2098	32.5	31.5	21.4	7.87	8.08	0.97x	+0.259	—
NC.KCO	32.7	487	487	HNE	0.1201	0.1201	0.1240	15.3	15.6	19.6	3.33	7.33	0.45x	-0.033	—
BK.RBOW	33.8	490	490	HNE	0.0795	0.0802	0.1075	10.5	11.0	10.5	3.63	3.90	0.93x	-0.293	—
CE.89641	38.3	290	290	HNE	0.2959	0.3021	0.2068	21.4	24.8	17.9	6.16	6.19	1.00x	+0.379	—
BK.BJES	45.9	744	744	HNE	0.1254	0.1238	0.0723	22.0	20.8	6.0	6.89	2.53	2.72x	+0.538	—
NC.KPR	46.0	492	492	HNE	0.1402	0.1397	0.0900	15.6	15.6	6.9	3.76	3.27	1.15x	+0.440	—
BK.KNEE	48.9	730	730	HNE	0.1636	0.1644	0.0583	39.2	38.2	5.3	11.13	2.45	4.55x	+1.037	—
NC.KCR	52.1	715	715	HNE	0.0585	0.0586	0.0563	6.8	6.7	4.0	1.67	1.69	0.99x	+0.040	—
BK.PRDS	56.6	493*	493	HNE	0.1806	0.1786	0.0742	9.7	10.6	6.0	2.61	2.96	0.88x	+0.879	Vs30d
NC.KBU	57.4	493*	493	HNE	0.1382	0.1406	0.0608	18.5	17.6	3.6	4.08	1.89	2.15x	+0.838	Vs30d
CE.89688	62.1	493*	493	HNE	0.0506	0.0511	0.0630	2.4	2.5	5.5	0.59	3.18	0.18x	-0.209	Vs30d
NC.KTD	63.1	701	701	HNE	0.0195	0.0188	0.0449	3.1	3.3	3.0	1.42	1.19	1.19x	-0.871	—
BK.BRIC	64.0	493*	493	HNE	0.0752	0.0743	0.0600	7.8	8.8	4.6	3.51	1.84	1.91x	+0.214	Vs30d
BK.DMOR	64.1	743	743	HNE	0.2033	0.2086	0.0458	47.8	45.6	4.2	13.77	1.90	7.25x	+1.517	Manu,Outl
NC.KHMB	70.0	628	628	HNE	0.1253	0.1248	0.0452	15.1	16.0	4.1	4.03	1.46	2.75x	+1.016	—
NC.KMR	70.4	748	748	HNE	0.0721	0.0719	0.0393	9.5	9.6	2.7	1.45	0.95	1.52x	+0.603	—
NP.1591	71.1	453	453	HNE	0.0950	0.0003	0.0606	9.0	0.1	4.3	—	1.50	—	-5.250	—
CE.99640	71.8	501	501	HNE	0.0186	0.0188	0.0483	1.2	1.4	2.9	0.63	1.06	0.60x	-0.945	—
NC.KCSB	76.8	749	749	HNE	0.0868	0.0856	0.0374	12.7	13.5	4.0	3.29	1.60	2.06x	+0.828	—
NC.KRP	78.0	732	732	HHE	0.0199	0.0192	0.0355	1.9	1.8	1.9	1.00	0.59	1.69x	-0.616	1,Outl
CE.99700	90.5	524	524	HNE	0.0182	0.0183	0.0419	1.2	1.3	2.6	0.34	0.67	0.51x	-0.828	—

ShakeMap = amplitudes the seismic network reported in stationlist.json. obspy = the same recordings fetched from FDSN and processed here. BBP = simulated. The first two should agree closely; where they do not, the ShakeMap quality flags usually explain why.
PGD is the animation pipeline's own number on both sides, read from the displacement each side hands to Blender — a different processing chain from the PGA and PGV columns, and from each other. See the displacement pages at the end.
* / Vs30d: 4 of 31 sites were simulated with a Vs30 value repeated across three or more stations — a lookup default rather than a site measurement. Vs30 drives the site term, so a defaulted value produces bias at intermediate and long period that is not the model's fault. Note also that the two Vs30 columns come from different sources and need not agree.

Cross-check — how far apart are the two ‘observed’ sources?

The amplitudes the network reported in ShakeMap against the same recordings re-fetched and processed here. These should agree closely; where they do not, the ShakeMap quality flags usually explain why. A systematic offset here would be a processing fault, not physics.

Observed amplitude cross-check versus distance

Residual against distance

A trend with distance points at the attenuation model rather than the source: the simulation losing energy faster or slower than the real crust does.

Residual versus distance at four periods

Response spectra, station by station

Observed and simulated response spectra at every compared station.

Per-station detail

Nearest first. Open a station for its three-component waveform overlay and the displacement record that drives the shaking animations. Figures load when opened.

NC.KCT 9.1 km PGA obs 0.1189 g · sim 0.3334 g ln(obs/sim) -1.031

Observed versus simulated acceleration at NC.KCT

NP.1584B 11.4 km PGA obs 0.0190 g · sim 0.2577 g ln(obs/sim) -2.608

Observed versus simulated acceleration at NP.1584B

NP.1023 14.6 km PGA obs 0.0465 g · sim 0.3508 g ln(obs/sim) -2.022

Observed versus simulated acceleration at NP.1023

NP.1586 21.6 km PGA obs 0.2968 g · sim 0.1976 g ln(obs/sim) +0.407

Observed versus simulated acceleration at NP.1586

CE.89486 24.3 km PGA obs 0.3974 g · sim 0.1934 g ln(obs/sim) +0.720

Observed versus simulated acceleration at CE.89486

CE.89255 24.5 km PGA obs 0.6483 g · sim 0.1918 g ln(obs/sim) +1.218

Observed versus simulated acceleration at CE.89255

BK.PETL 27.2 km PGA obs 0.1232 g · sim 0.2198 g ln(obs/sim) -0.579

Observed versus simulated acceleration at BK.PETL

CE.89462 27.3 km PGA obs 1.3867 g · sim 0.1864 g ln(obs/sim) +2.007

Observed versus simulated acceleration at CE.89462

NP.1582 27.4 km PGA obs 0.0719 g · sim 0.1724 g ln(obs/sim) -0.875

Observed versus simulated acceleration at NP.1582

NC.KMPB 28.2 km PGA obs 0.1226 g · sim 0.1539 g ln(obs/sim) -0.228

Observed versus simulated acceleration at NC.KMPB

NP.1581 29.6 km PGA obs 0.2718 g · sim 0.2098 g ln(obs/sim) +0.259

Observed versus simulated acceleration at NP.1581

NC.KCO 32.7 km PGA obs 0.1201 g · sim 0.1240 g ln(obs/sim) -0.033

Observed versus simulated acceleration at NC.KCO

BK.RBOW 33.8 km PGA obs 0.0802 g · sim 0.1075 g ln(obs/sim) -0.293

Observed versus simulated acceleration at BK.RBOW

CE.89641 38.3 km PGA obs 0.3021 g · sim 0.2068 g ln(obs/sim) +0.379

Observed versus simulated acceleration at CE.89641

BK.BJES 45.9 km PGA obs 0.1238 g · sim 0.0723 g ln(obs/sim) +0.538

Observed versus simulated acceleration at BK.BJES

NC.KPR 46.0 km PGA obs 0.1397 g · sim 0.0900 g ln(obs/sim) +0.440

Observed versus simulated acceleration at NC.KPR

BK.KNEE 48.9 km PGA obs 0.1644 g · sim 0.0583 g ln(obs/sim) +1.037

Observed versus simulated acceleration at BK.KNEE

NC.KCR 52.1 km PGA obs 0.0586 g · sim 0.0563 g ln(obs/sim) +0.040

Observed versus simulated acceleration at NC.KCR

BK.PRDS 56.6 km PGA obs 0.1786 g · sim 0.0742 g ln(obs/sim) +0.879

Observed versus simulated acceleration at BK.PRDS

NC.KBU 57.4 km PGA obs 0.1406 g · sim 0.0608 g ln(obs/sim) +0.838

Observed versus simulated acceleration at NC.KBU

CE.89688 62.1 km PGA obs 0.0511 g · sim 0.0630 g ln(obs/sim) -0.209

Observed versus simulated acceleration at CE.89688

NC.KTD 63.1 km PGA obs 0.0188 g · sim 0.0449 g ln(obs/sim) -0.871

Observed versus simulated acceleration at NC.KTD

BK.BRIC 64.0 km PGA obs 0.0743 g · sim 0.0600 g ln(obs/sim) +0.214

Observed versus simulated acceleration at BK.BRIC

BK.DMOR 64.1 km PGA obs 0.2086 g · sim 0.0458 g ln(obs/sim) +1.517

Observed versus simulated acceleration at BK.DMOR

NC.KHMB 70.0 km PGA obs 0.1248 g · sim 0.0452 g ln(obs/sim) +1.016

Observed versus simulated acceleration at NC.KHMB

NC.KMR 70.4 km PGA obs 0.0719 g · sim 0.0393 g ln(obs/sim) +0.603

Observed versus simulated acceleration at NC.KMR

NP.1591 71.1 km PGA obs 0.0003 g · sim 0.0606 g ln(obs/sim) -5.250

Observed versus simulated acceleration at NP.1591

CE.99640 71.8 km PGA obs 0.0188 g · sim 0.0483 g ln(obs/sim) -0.945

Observed versus simulated acceleration at CE.99640

NC.KCSB 76.8 km PGA obs 0.0856 g · sim 0.0374 g ln(obs/sim) +0.828

Observed versus simulated acceleration at NC.KCSB

NC.KRP 78.0 km PGA obs 0.0192 g · sim 0.0355 g ln(obs/sim) -0.616

Observed versus simulated acceleration at NC.KRP

CE.99700 90.5 km PGA obs 0.0183 g · sim 0.0419 g ln(obs/sim) -0.828

Observed versus simulated acceleration at CE.99700

Provenance

Simulation

produced by: BBP simulated
BBP job id: job_d6c24a71669e4cc9b1ff
BBP version: 22.4.0
method: GP
realizations: 1 of 1 succeeded — a single stochastic rupture, not an ensemble; the bias below is one draw
realization used: r0001
sites in job spec: —
job submitted: 2026-08-04T07:15:14Z
per-site Vs30 from: submit-time record (/app/plots/gof_work/nc73821036/compare/station_specs.json)

Source parameters

magnitude used: —
hypocentre: —
depth: —
strike / dip / rake: —
mechanism source: USGS moment tensor, nodal plane 1 — the conjugate plane is equally consistent with the tensor and gives different amplitudes

Velocity model

region used: NoCal500
native for this location: Mojave500 — not installed, so the run is approximate. LABasin500 carries a deep sedimentary basin that Ridgecrest does not have, which would amplify 0.3-1 s motion.
calibration: M6.4 against GP's ~M5.5 floor
platform: arm64 under emulation — amplitudes are not SCEC-reference-accurate; this number measures BBP plus Rosetta and cannot separate them

Observed data and processing

Station selection

search radius: 100.0 km
policy: all usable within radius
stations on event: 596
without FDSN code: 0 — cannot be queried at all
candidates: 31
usable records: 31 (0 returned no usable 3-component set)
in this report: 31 with both observed and simulated
distance range: 9.1 – 90.5 km

Instrumentation

channel priority: HN?, EN?, HH?, BH?, EH?
channels used: HH, HN
data centres: NCEDC
mixed instrument types: yes — strong-motion and broadband are both present, which puts a second variable into the residual
Vs30 source: ShakeMap stationlist (per-station)

Signal processing

window: origin −60.0 s to +300.0 s (requested)
records not starting as requested: 11 of 31 — triggered instruments whose data begins later. Each trace's true start is measured and used; assuming the requested lead would misplace them.
response removal: to acceleration and velocity separately, pre-filter (0.05, 0.1, 20.0, 25.0) Hz
resampled to: 100.0 Hz
time origin: t = 0 is the rupture origin for both observed and simulated traces on every overlay

How to read this report

The metric

definition: ln(observed / simulated) RotD50 spectral acceleration, averaged across stations at each period.
sign: POSITIVE means the simulation under-predicts — real shaking was larger. NEGATIVE means it over-predicts.
scale: It is a log ratio: ±0.10 is about ±10%, ±0.69 is a factor of 2, ±1.10 is a factor of 3.
sigma: The station-to-station spread around that average. A large sigma with a small bias means the model is right on average and wrong site by site.

Reading the pattern

short-period poor, long-period good: Points at high-frequency attenuation or near-surface site response rather than the source.
long-period poor, short-period good: Points at the source — corner frequency, stress drop — or at basin response in the velocity model.
one station far off: Usually 3D site or directivity effects that a 1D layered velocity model cannot reproduce by construction. Check whether it is driving the sigma before treating it as model error.

What this does not tell you

: A pass means the simulation is consistent with recordings at these stations, for this event, under this velocity model, on this hardware. It is not a general statement about BBP, and it is not a prediction of accuracy at an unrecorded site.