Prediction-focused approaches: an opportunity for hydrology

Our ability to predict the evolution of complex hydrological system is fundamental. For decades, such problems have been solved by calibrating a conceptual model of the subsurface to fit data. Unfortunately, model calibration does not allow a realistic uncertainty quantification, whereas stochastic inversion is often computationally prohibitive. In this contribution, prediction-focused approaches (PFAs) are introduced to overcome those main shortcomings. This new paradigm focused on generating predictions directly from the data instead of generating models. A group of prior models is used to generate the data and the prediction in order to derive a direct relationship between both types of variables. The advantages, limitations and research perspectives are discussed.