PREDICTING PHOSPHORUS CONCENTRATION AND PHOSPHORUS LOAD FROM WATERSHED CHARACTERISTICS USING BACKPROPAGATION NEURAL NETWORKS

In this study the authors show that backpropagation neural networks (B NN) can be used to construct a predictive model for complex phenomena such as the relationships between land use and the concentration and e xport of total phosphorus and of orthophosphate by watersheds. Data ar e taken from 927 tributary sites throughout the USA. These were random ly separated into two sets, a training set and a testing set. The pred ictive quality of the model was judged with the testing set after lear ning with the training set. Despite the diversity of the territory (cl imate, soil, land use)? the BNN procedure gave good predictions. The c orrelation coefficients (R) reached 0.80.