PREDICTION OF ANNUAL NITRATE-N LOSSES IN DRAIN OUTFLOWS WITH ARTIFICIAL NEURAL NETWORKS

Abstract

An artificial neural networks (ANN) model was developed for the prediction of annual nitrate-N (NO3-N) losses into the drain flow at Eugene F. Whelan Experimental Farm (Agriculture Canada, Woodslee, Ontario, Canada). Data consisted of daily measurements of nitrate-N taken from eight different soil conservation treatments during 1992- 1994. The experiment consisted of four crop/tillage and two water table management systems. Due to the moderate size of the data set, a tenfold cross validation method was used for model validation. A sensitivity analysis was also performed to assess the effect of the input variables on the performance of the networks. The results of this study indicated that the performance of network predictions of nitrate-N was highly satisfactory for 6 of the treatments and acceptable for the remaining two. The sensitivity analysis demonstrated that network predictions of nitrate-N were not affected when either drain flow or evapotranspiration data were excluded from the network training files. Overall, this study reveals that, from adequate input information, Artificial Neural Networks could effectively predict loss of nitrate-N in drain outflows. While the ANN model itself is not transportable to any other site, it does provide another method of estimating nitrate-N losses from agricultural fields with fewer input parameters. In addition, they could also be used to identify the unnecessary parameters for ANN modeling and thus save valuable time and resources in data collection.