Motivated by the challenging task to model extreme flood events along rivers caused by heavy precipitation and the resulting runoff, this project aims at developing a Bayesian framework to describe extreme events by the interplay of statistical and physically-driven models. More precisely, in a first step, we will establish a Bayesian statistical model for single extreme events in spatio-temporal random fields, such as heavy precipitation. In a second step, simulations from this model will serve as input for physically-driven models describing, for instance, the runoff. The flexibility of our modelling framework will allow us to tackle various challenges such as modelling on various spatio-temporal scales, inclusion of multiple data sources, combination of different models, and inclusion of various types of flexibly defined extreme events – coming at the price of high computational costs. Varying the model specification within this framework will enable us to navigate in the PN5 triangle of model accuracy, precision and the required computational resources. This will be demonstrated in a showcase example on extreme flooding in the upper Neckar basin, linked to the SimTech vision “Engineered Geosystems”. In particular, we will compare the benefits and costs of our combined approach to purely statistical benchmark models directly built from river discharge data.