Nonparametric trend detection in river monitoring network data: a spatio‐temporal approach: Fid-Bau Portal

Nonparametric trend detection in river monitoring network data: a spatio‐temporal approach

Clement, Lieven / Thas, Olivier

10.1002/env.929.abs

Since the introduction of the Water Framework Directive (WFD), regional authorities have to take actions to improve the aquatic environment. In Flanders (Belgium), for instance a Manure Decree (MD) has been introduced to reduce the impact of eutrophication. It is of course important to assess the evolution of the water quality. In this context, we develop a spatio‐temporal method for the assessment of trends in river monitoring network data. The aim is to detect trends on a more regional scale. In contrast with most existing methods used for analysing river data, our model incorporates the spatio‐temporal dependence structure explicitly. An AR(1) process is assumed for the temporal dependence. The spatial dependence is implied by the river topology and the flow direction. The dependence structure is modelled by a state variable which is embedded into an observation model that makes the correlation structure less rigid. An additive model incorporates a nonparametric long‐term trend and it corrects for seasonal effects. To detect when the local trend is beneficial, tests on the first derivative of the nonparametric long‐term trend are performed at each time step. This, however, results in a large number of simultaneous tests. To correct for multiplicity, we propose to use a variant of the free step‐down method of Westfall and Young 1993 to take these dependences into account. The methodology is applied to a case study on the river Yzer (Belgium). A significant decrease in the nitrate concentration is detected after the implementation of the MD. Copyright © 2008 John Wiley & Sons, Ltd.