Modelling Approaches for Subsurface Drainage Water Quality Management

Abstract

Abstract Contamination of surface waters by agricultural activities is a serious problem. Two different modelling approaches to simulate nutrient and pesticide transport in subsurface drained soils were investigated in this study. First, artificial neural network (ANN) models, a trainable fast back-propagation (FBP) network and a self-organizing radial basis function (RBF) network, were developed for simulation of NO3--N concentration in tile effluent. Second, a hydrologic model, DRAINMOD, was linked with a chemical transport model, GLEAMS, to simulate chemical transport of atrazine through the soil into subsurface drain outflow. The ANN models and linked DRAINMOD-GLEAMS model were calibrated and validated against experimental data collected at the Greenbelt Research Farm of Agriculture Canada during the years 1988, 1989 and from 1991 to 1994. Several statistical parameters were calculated to evaluate model performance. A comparison of results indicated that the RBF neural network model was superior to the FBP model in predicting drain outflow and NO3--N concentration. Results obtained from the linked DRAINMOD-GLEAMS model demonstrate that atrazine simulations were underpredicted in subsurface drain outflows for spring and fall seasons. Both modelling approaches provide a useful tool for management of fertilizer/manure and pesticides transport through soil and crop root zones into surface water.