Ensemble simulations in MRST using the ensemble module

Simulation of subsurface flow is always subject to a high degree of uncertainty. Typically, we only have partial and noisy knowledge about the geology and state of the reservoir, meaning that the computational model, grid, and its parameters are uncertain. A common way to express this uncertainty is to run an ensemble of simulations using parameters sampled from a probability distribution that reflects the uncertainty in the reservoir model. By simulating the ensemble members independently, we can analyse the corresponding uncertainty in the future well responses, giving us richer knowledge of the reservoir compared to running only a single simulation. Ensembles have several applications, such as quantifying the uncertainty in production data through Monte Carlo simulations, calibrating the uncertain reservoir parameters through history matching, or doing optimization under uncertainty.

This poster presents the recent work in creating a module within MRST for doing ensemble simulations. We present the MRSTEnsemble class and show the different components it is built around. We also highlight some of the options you have when running ensembles, such as parallel execution both with and without the parallel toolbox installed. Finally, we give a quick demonstration to how to use the module for history matching.