Process-based numerical models of flow, transport, and reactive turnover are necessary tools to understand major influences on water quality, nutrient cycling, and fate of pollutants. These models require spatially distributed parameters which, unfortunately, are often quite uncertain.

The main objective of project P7 is to develop a modeling framework for flow and reactive transport on the catchment scale that is

• as mechanistic as necessary to account for the spatial structure of the domain and the major processes determining water-quality metrics,
• as efficient as possible to reduce computational effort and facilitate ensemble runs for stochastic approaches towards data interpretation and predictions.

The model framework is based on simulating the strong feedbacks between soils and vegetation at the land-surface with fully coupled 1-D, vertical soil-crop models (Expert-N) that are weekly coupled to an underlying 3-D flow model (HydroGeoSphere) of the deeper subsurface and streamline-based models of reactive transport therein.

While the individual model components honor existing information about soil types, land use, topography/bathymetry, and geology, we address the uncertainty of parameters and geometries by treating all coefficients as random distributions, requiring ensemble calculations of the full system.

For predicting flow and reactive transport in the environment, many scientific models concepts compete with each other. This adds an additional level of uncertainty to system understanding and prediction. Among the competing models, more accurate models are typically more complex. Unless they are fed with more data for calibration, their predictions are affected by more uncertainty.

To address this problem, we investigate:

• What are the key uncertainties in model choice for CAMPOS-related applications?
• What model concepts (lower versus higher complexity) are legitimate to use, given the limited amount of data that is affordable to collect?
• How can we identify optimal instrumentation strategies that minimize dominant uncertainties (both parametric and model choice)?
• With these insights, can we provide feedback to the overall CAMPOS research strategy?

Project P8 works with multi-model ensembles to address the uncertainty of predictions, including uncertainty in parameters, forcings, and model formulations. Then, we develop methods and apply them to perform

• Bayesian multi-model analysis and cross-validation techniques to assess the legitimacy and predictive skills of our (multi-)models,
• Optimal design of experiments, maximizing the confidence in model choice and model predictions.

We do this in close collaboration with projects that generate data and develop models, and provide our expertise in Bayesian analysis, computational statistics and mathematical optimization.