Including land use information for the spatial estimation of groundwater quality parameters - 1. Local estimation based on neighbourhood composition

- Land use information of neighbourhood used to estimate locally averaged distribution.
- Local distributions improve estimates; can be used as drift for dependence model.
- Neighborhood size is optimized (cross-validation, information content over domain).
- The pure distributions of homogeneous land uses used for local mixing are estimated.
- Similar land use categories based on their concentration distributions are grouped.

SummaryMost groundwater recharge comes from the infiltration of water through the land surface. Data analysis shows that solute concentrations at the water table vary between land use categories and depending on the land use composition within a certain neighbourhood.Driven by these observations, the goal of this paper is to estimate the solute distribution at a location depending on the composition of land use in the neighbourhood, even though land use information is categorical. This goal is achieved by mixing pure distributions of homogeneous land use according to their frequency of occurrence in the vicinity of, and their distance from an estimation location. These pure distributions are jointly inverted using a maximum likelihood-based approach.The neighbourhood size is optimized using cross-validation. Measurements below detection limit are included via their probabilities of non-exceedance. A solute-specific, spatially distributed measure of information content of the secondary information is presented. The method is applicable for many types of secondary information and can be used as drift for spatial estimation of the primary variable. This estimation is a local estimation and does not include larger scale spatial information. The information of measurements is included via the optimized concentration distributions for land use groups, not via a model of spatial dependence. The global estimation is described in the companion paper.