Comparison of three recent discrete stochastic inversion methods and influence of the prior choice

Groundwater flow depends on subsurface heterogeneity, which often calls for categorical fields to represent different geological facies. The knowledge about is however limited and provided indirectly by state variables, such as hydraulic heads of contaminant concentrations. In cases, solving a inverse problem an important step in modeling. this work, we present compare three recent frameworks: Posterior Population Expansion (PoPEx), Ensemble Smoother with Multiple Data Assimilation (ESMDA), DREAM-ZS (a Markov chain Monte Carlo sampler). PoPEx ESDMA are used Multiple-point statistics (MPS) geostatistical engines, Wasserstein generative adversarial network (WGAN). inversion methods tested synthetic example pumping test fluvial channelized aquifer. Moreover, the solved times each method, time using training image check performance priors. To assess quality results, propose framework based continuous ranked probability score (CRPS), compares single true values predictive distributions. All performed well when create reference, but their performances were degraded alternative images. produced least artifacts presented rather slow convergence. ESMDA showed initially very fast convergence reaches plateau, contrary remaining methods. was overly confident placing some incorrect features outperformed other terms