Quantifying subsurface parameter and transport uncertainty using surrogate modelling and environmental tracers

We combine physics-based groundwater reactive transport modelling with machine-learning techniques to quantify hydrogeological model and solute predictive uncertainties. train an artificial neural network (ANN) on a dataset of hydraulic heads 3H concentrations generated using high-fidelity model. Using the trained ANN as surrogate reproduce input–output response model, we posterior distributions parameters forcing conditions Markov chain Monte Carlo calibration against field observations concentrations. demonstrate methodology application that predicts Chlorofluorocarbon-12 (CFC-12) at contaminated site in Wyoming, United States. Our results show including reduced uncertainty estimated permeability infiltration rates, compared alone. However, quantification shows CFC-12 predictions conditioned parameter cannot measurements. found calibrating head mean ages are too large explain observed The coupling allows us efficiently uncertainties, which is typically computationally intractable models