Global sensitivity analysis on a hybrid geostatistical model using a distance-based approach

Hybrid geostatistical models aim at mimicking depositional events. The resulting models have the capability to simulate realistic stratigraphic structures for a variety of environments. However, this family of algorithms requires a high degree of parameterization. Therefore, having a good knowledge of the model parameters sensitivity is vital for understanding the behavior of such models. In this study, a distance-based approach using kernel Multidimensional Scaling (MDS) is used to investigate the effect of parameters variability on the outputs of a turbidite model. This study is conducted on realistic seafloor topography and evaluates the relative influence of the size of the turbiditic lobes, the sediment source location, the deposition model and the noise used to randomize the lobe thickness. The findings of this analysis have important implications for understanding, conditioning and uncertainty analysis of these newly developed hybrid geostatistical models.