In this work an integrated methodological and operational framework for diagnosis and calibration of Stratigraphic Forward Models (SFMs) which are typically employed for the characterization of sedimentary basins is presented. Model diagnosis rests on local and global sensitivity analysis tools and leads to quantification of the relative importance of uncertain model parameters on modeling goals of interest. Model calibration is performed in a stochastic framework, leading to estimates of distributions of model parameters (and ensuing spatial distributions of model outputs) conditional on available information. Starting from a considerable number of uncertain model parameters, which is typically associated with SFMs of the kind analyzed, the approach leads to the identification of a reduced set of parameters which are most influential to drive stratigraphic modeling results. Probability distributions of these model parameters conditional on available data are then evaluated through stochastic inverse modeling. To alleviate computational efforts, this step is performed through a combination of a surrogate model constructed through the Polynomial Chaos Expansion approach and a machine learning algorithm for efficient search of the parameter space during model inversion. As a test bed for the workflow, focus is on a realistic synthetic three-dimensional scenario which is modeled through a widely used SFM that enables one to perform three-dimensional numerical simulations of the accumulation of siliciclastic and carbonate sediments across geologic time scales. These results constitute a robust basis upon which further deployment of the approach to industrial field settings can be designed.
