Data Assimilation and Ensemble Methods

Automating the Match

History matching by hand does not scale to hundreds of uncertain parameters and a stream of new data. Ensemble methods, the ensemble Kalman filter and its smoothers, automate it.

The Ensemble Update

Carry a cloud of models sampled from the prior. As each observation arrives, nudge every member toward the data by a gain set from the ensemble spread. The cloud both shifts toward the truth and tightens, so the estimate sharpens with every assimilation.

Bayesian Updating at Scale

This is Bayesian updating run on an ensemble, the workhorse of modern assisted history matching. It also delivers the posterior spread directly, so the forecast uncertainty of the previous chapter falls out of the same machinery.