Simulation of sub-surface tile drain water quality (nitrates) using artificial neural networks

Artificial neural network (ANN) models namely, (i) a multilayer perceptron (MLP) using fast backpropagation (FBP) learning algorithm, and (ii) radial basis function (RBF) network using orthogonal least squares (OLS) learning algorithm, were used for simulation of tile outflow and nitrate nitrogen concentration in tile drains at Greenbelt Research Farm of Agriculture and Agri-Food Canada, near Ottawa, ON. Sensitivity analysis to find the optimum network parameters was done. Simulation results of the ANNs were compared on daily and monthly basis. Further, a preliminary comparison was done with results from a deterministic model, DRAINMOD-N. The results indicate that the MLP ANN with FBP algorithm could be used to simulate tile outflow and nitrate nitrogen concentrations on monthly basis while RBF ANN with OLS algorithm can be used as an effective tool to simulate the said parameters on daily as well as monthly basis.