Flexible Spatio-Temporal Models for the Analysis of River Network Water Quality Data: Recent Development and a Case Study
Olivier Thas1, Lieven Clement2
1BioStat, Ghent University, Gent, Belgium; 2L-BioStat, Katholieke Universiteit Leuven, Belgium

For many years river water quality in the region of Flanders (Belgium) suffers from high nitrate levels. Every few years a more stringent legislation is voted, aiming at reducing nitrate exposure from agricultural activity. An important task of the Flemish Environment Agency (VMM) is to estimate the change in nitrate concentration in Flemish surface waters. Data from a dense monitoring network is at their disposal (> 1000 monitoring sites, sampled about 10 times per year).

In Clement et al. (2008, Environmetrics,DOI: 10.1002/env.929) semiparametric regression methods were proposed for a spatio-temporal analysis of the monitoring data. Their method combined the directional river network spatial dependence structure with an AR(1) temporal autocorrelation. However, as their method was formulated as a state-space model, it relies on the assumption of equidistant sampling, both in time and in space.

In this presentation we present a more flexible modeling framework that can deal with irregularly spaced data. The statistical methods allow for local trend detection, as well as for trend detection at a larger regional scale. We develop the theory and demonstrate its usefulness on the nitrate concentration in Flemish rivers.

Keywords: Water quality; River networks; Spatio-temporal statistical model; Trend analysis

Biography: Olivier Thas is associate professor in biostatistics at the the BioStat research group of the Department of Applied Mathematics, Biometrics and Process Control, Ghent University. His research focuses mainly on nonparametrics (goodness-of-fit and lack-of-fit), but he is also active in more applied fields (e.g. environmental statistics, microarrays, genome analysis, and biomarkers).