NEURAL NETWORK APPROACH IN RISK ASSESSMENT OF PHOSPHORUS LOSS

The main objective of this study is to demonstrate the use of artificial neural network (ANN) modeling tool to predict the risk of phosphorus (P) loss from the fields to nearest water body. The attention is drawn to AN as an alternative approach to the P index calculation for prediction of the P losses. The specific tasks of this study were to determine risk classes of P loss by linking together source and transport factors that accelerate P losses and to evaluate AN model performance for predicting risk classes via nutrient transport. AN was trained with a Levenberg-Marquardt algorithm, and Scaled Conjugate Gradient algorithm was used to estimate the possible risk of P losses from agricultural land. Two small agricultural watersheds in Auce and Bauska were chosen to determine field parameters, and expert's evaluation was used for description of the risk classes' of P loss. Finally these values were used as inputs for the neural network model. The model was trained and validated by assessing its predictive performance on a testing set of data excluded from the training set. The research results highlight the capabilities of AN to predict risk for a particular field and suggest that future research on application of other algorithms is required.