Quantitative understanding of pollutant fluxes from diffuse input and turnover of pollutants at catchment scale requires process-based numerical models that can explain observed time series of heads, fluxes, and concentrations under current conditions and predict future states under changing conditions. The uncertainty of forcing, parameters, and conceptual assumptions as well as the unresolved internal variability entails a probabilistic framework, predicting probabilities of reactive-species concentrations rather than single concentration values. Due to high computational effort, such evaluations cannot be done with a fully coupled, multi-dimensional, spatially explicit reactive-transport model overarching all components. Conceptual simplifications are needed, keeping spatially explicit calculations whenever needed and computationally manageable, but simplifying reactive-transport computations without sacrificing mechanistic understanding.
The aim of this project is to develop a probabilistic modeling approach in order to assess metrics of water quality at catchment scale. The statistical parameter distributions are conditioned on measurements using Markov-Chain Monte-Carlo and ensemble Kalman methods. Afterwards the key sources of uncertainty and their relative importance in simulating reactive-species concentrations at different scales can be determined. The approach expresses a paradigm shift from deterministic to probabilistic models even for complex reactive-transport systems. The reactive transport is based on partial differential equations wherever it is computationally feasible and hydromechanically necessary to obey structural information of the catchment. However, it switches to travel- and exposure-time formulations when spatially explicit simulation of reactive transport is not needed and computationally very expensive. The statistical parameter distributions are conditioned on measurements using Markov-Chain Monte-Carlo and ensemble Kalman methods. Afterwards the key sources of uncertainty and their relative importance in simulating reactive-species concentrations at different scales can be determined.
