Water Quality and Sediment Behaviour of the Future: Predictions for the
21st Century (Proceedings of Symposium HS2005 at IUGG2007, Perugia, July
2007). IAHS Publ. 314, 2007, 125-134
Modelling
catchment-scale nitrate transport using a combined process-based and data-driven
approach
RAJESH RAJ SHRESTHA & MICHAEL RODE
Department of Hydrological Modelling, UFZ – Helmholtz Centre for Environmental Research, Brueckstrasse 3a, D-39114 Magdeburg, Germany
rajesh.shrestha@ufz.de
Abstract Diffuse nitrate pollution in catchments is mainly driven by hydrological flow components characterised by complex relationships with streamflow nitrate concentration. This paper demonstrates a combined process-based–artificial neural network (ANN) approach for the simulation of streamflow nitrate concentration based on the relationships between driving and resultant variables. The simulated hydrological flow components from a process based WaSiM-ETH model, together with observations, are used to train two different ANNs. The results show a reasonable match between observed and simulated streamflow and nitrate-N concentration. The ANN with temperature as an input performed better than the ANN without it, indicating the effect of seasonal variability. Nash-Sutcliffe coefficients of 0.746 and 0.856 were obtained for streamflow in calibration and test periods, respectively, while these coefficients were 0.819, 0.629 and 0.627 for nitrate-N concentration in training, cross-validation and test periods, respectively. Hence, the combined approach offers an effective and efficient methodology for modelling catchment scale nitrate dynamics.
Key
words artificial
neural network; hybrid model; nitrate transport; water balance