Traces of Processes: Data & Stochastic Models for Improving Decisions
We need to assess change in human and natural systems quantitatively. Hydrology provides suitable assessment tools for planning, design, and societal issues. The basis for these informed decisions is data. Quantitative information about the subsurface is sparse. Given the
available data (e.g., measurements of hydraulic conductivity), we should tease out as much information as possible. I demonstrate copula-based methodology to describe and model non-linear spatial dependence in hydraulic conductivity data-sets. Subsequently, the effects of this non-linear spatial dependence structure on dependent processes (e.g., solute transport behaviour) are analysed. The characteristics of data are manifold: both direct and indirect measurements are useful data. Certain kinds of data (e.g., land cover,
snow equivalents) can be measured accurately with high spatial coverage.On the other hand, direct measurements are sparse. Thus, we need techniques to combine different kinds of information (censored,categorical, real-valued) as we need to estimate at unsampled times and locations. In the second part of this presentation, I will show an example where the goal is to predict groundwater quality parameters at
regional scale in the federal state of Baden-Württemberg, Germany.
For this purpose I am employing a stochastic copula-based model that is capable to mimic the following properties that were encountered in the data: the statistical distribution of the measurements, a varying degree of dependence in different quantiles, censored measurements (e.g., measurements below detection limit), the composition of categorical additional information in the neighbourhood (exhaustive secondary
information), and the spatial dependence of a dependent secondary variable, possibly measured with a different observation network
(non-exhaustive secondary data).