Uncertainty Re-parameterization Approach to Calibrate Simulation Models with Production data in a WAG Injection Field

The development of robust methodologies to assess and mitigate uncertainties is essential. This work proposes a semi-automated approach to build an ensemble of prior-models calibrated with the observed production data while maintaining geological consistency. The methodology consists of: (1) generating the first ensemble of models (base ensemble) - not calibrated with observed production data - with enough variability and encompassing the observed production data, (2) re-parameterizing uncertainties by combining the Normalized Quadratic Distance with Signal (NQDS) indicator with Gaussian Kernel Density Estimation (KDE) and (3) iterating the previous step until a considerable number of approved models (models within the confidence interval for the NQDS of each objective function) is achieved within the full ensemble of models. The methodology is applied to a giant Brazilian pre-salt field. The base ensemble consists of 200 models combining static and dynamic uncertainties, and reproducing models with enough variability to encompass the observed production data. After two iterations, by combining the NQDS indicator with Gaussian KDE, 155 models calibrated with observed production data were achieved in an ensemble of 200 models. For the base ensemble, only 55 models were approved. Therefore, a much higher number of accurate models was obtained by rebuilding the probabilistic distribution functions (PDFs) for each uncertain variable based on observed production data. The mean permeability showed the greatest improvement as the well-log-derived permeability was based on empirical correlations with pore size. Based on a multidisciplinary effort, this work successfully improved the accuracy of prior-models.