Error models for non-resolved structure or processes in simplified geothermal reservoir models

Predicting stresses or water and heat fluxes in geothermal reservoirs is challenging due to the lack of knowledge about the subsurface structure, but also due to the extensive computational costs for the models that describe complex flow processes in complex 3d structures. Model parameters are often obtained from model calibration using different indirect observations. Addressing also model uncertainty often requires a large number of forward runs of the model, which is mostly unfeasible due to the computational costs. It can be practical to represent the system in a simplified way. This may mean that specific structural details are not resolved, but it may also mean that processes are neglected or that the reservoir model is set up in 2d. To avoid biased parameter estimations, it can be beneficial to compensate the mismatch due to non-resolved structure by error models, which express the mismatch due to the simplification of the real system. It is a challenge to derive appropriate error models, in particular if processes are slow and quasi-steady state reactions cannot be expected, as this requires time-dependent error models. A model of a geothermal reservoir will be set up and the potential of error models for different types of simplifications (coarsening, fracture representation, spatial dimensionality) will be investigated in a setting where well pressure data are used for parameter identification. An Iterative Ensemble Kalman Filter will be used for parameter estimation. It will be tested if machine learning algorithms can be useful for derivation of appropriate error models. It will also be tested how the impact of using an error model is on prediction uncertainty. 

Doctoral Researcher: Wansheng Gao

Scientific Advisors: Insa Neuweiler, Ludovic Chamoin