Bayesian Parameter Inversion

Fit physics forward models to observed well-log, seismic or pressure data and get full posterior distributions, not point estimates. Independent chains run in parallel in your browser; convergence diagnostics and posterior predictive checks tell you whether to trust the answer.

Three forward models, one inference engine

• Well-log petrophysical inversion — porosity and shale volume per zone from GR, DT, RHOB and NPHI, with documented matrix, shale and fluid endpoints and automatic zone detection
• Post-stack acoustic seismic inversion — layer acoustic impedance, wavelet frequency and time shift from an observed seismic trace, with a well log as the reference model and prior
• Pipeline transient inversion — wave speed, friction and leak or blockage location and size from a hammer-test trace using a method-of-characteristics forward model

A real inversion workflow

Upload data → the tool recognises the dataset and maps channels → compatible forward models are offered → every parameter is shown with its prior, bounds and units → the observation model is configured → four independent adaptive-Metropolis chains run in Web Workers → results report split-R̂, bulk and tail effective sample size, Monte-Carlo standard error, posterior predictive fit and residuals. An INFERENCE STATUS of CONVERGED or REVIEW REQUIRED tells you whether the posterior can be trusted without reading the diagnostics.

Reproducible and auditable

Every run is seeded and reopens exactly from the saved record, which stores the model specification, priors, dataset SHA-256, chains, summary and diagnostics. Export posterior samples as CSV and a run report as Markdown. A separate Validation Lab scores recovery against known synthetic models.

Frequently asked questions

What is Bayesian inversion of well logs?

It treats porosity, shale volume and other rock properties as unknowns, passes them through a petrophysical forward model (density, sonic, gamma-ray and neutron responses), and finds the distribution of values consistent with the measured logs. The result is a posterior with credible intervals per zone rather than a single deterministic value.

How do I know the MCMC has converged?

The tool runs several independent chains and reports split R-hat and bulk and tail effective sample sizes for every parameter. CONVERGED requires R-hat below 1.01 and ESS above 400; otherwise the status reads REVIEW REQUIRED or NOT CONVERGED and explains why.

Can I invert my own seismic trace?

Yes. Upload a time-amplitude trace, optionally a well log with DT and RHOB as the reference model, choose the number of layers and wavelet, and run. The well log sets the prior; the top-layer impedance is fixed because reflectivity only constrains ratios.

Does it run on my server or in the cloud?

Neither — the sampler runs in your browser using Web Workers, so your data does not leave your machine unless you choose server-side storage on your own IIS host.

What file formats are supported?

LAS 2.0 well logs, CSV/TSV/TXT tables with headers and units, and JSON.