IMPROVED SENSITIVITY CALCULATIONS

Abstract

Model sensitivities are a measure of how specific model results vary with selected model parameters (such as permeability and porosity). They are important for parameterizing numerical models, automatic history matching and uncertainty quantification of model predictions. However, the current way of evaluating model sensitivities using finite differencing does not scale well for large and highly nonlinear geothermal reservoir simulations.
Alternatively model sensitivities may be generated by solving related linear problems forward or backward in time. The method of choice depends on the dimensionality of the parameter and observation spaces. The proposed linear methods were tested on one-dimensional transient two-phase flow problems, solved using the TOUGH2 simulator. Results of the forward and backward propagating methods were compared with those found by using simple, but computationally expensive, finite differencing.