A hybrid deterministic-fuzzy rule based model for catchment scale nitrate dynamics

Current understanding of nitrate export from catchments indicates that the transport dynamics are mainly driven by hydrological processes characterised by complex nonlinear relationships. The aim of this paper is to develop a hybrid deterministic-fuzzy rule based model capable of simulating catchment scale nitrate transport on the basis of the relationships between driving and resultant variables. The deterministic water balance model WaSiM-ETH is used for the simulation of hydrological flow components. The simulated flow components from the WaSiM-ETH model together with observations are used to develop a fuzzy rule based nitrate transport model. The fuzzy rules are derived using a simulated annealing optimisation procedure supplemented by knowledge about data relationships. The study is undertaken using daily time step data from the Weida catchment. in the North-Eastern Germany, which is a 100 km(2) subcatchment of the Weisse Elster river in the Elbe river basin. The models show reasonable performance with regards to the magnitude and dynamics of the streamflow, and nitrate-N concentration and load. The superior performance of the fuzzy rule based model in comparison to a multiple linear regression model indicates a complex nonlinear relationship between driving and resultant variables. The assessment of the rules provides explicit insights on the qualitative and quantitative relationships between different variables and their relative importance. The subsurface flow is found to be the most important variable which corresponds to the prevailing understanding that the nitrate transport processes are mainly driven by it. The relative importance of temperature as an input variable indicates the effect of seasonal. variability. The hybrid model is valid for present land use characteristics and management practices, which can be extended to include additional variables that affect nitrate entry to subsurface flow. (c) 2007 Elsevier B.V. All rights reserved.