Application of artificial neural networks for classifying lake eutrophication status

Several artificial neural network architectures were ‘trained’ on data from the Eastern Lake Survey – Phase I of the Environmental Protection Agency's National Surface Water Survey in order to investigate which physical–chemical parameters are possibly of greatest importance in determining the eutrophication status of lakes. From the 110 available lake parameters in the Survey, 60 were chosen as input to the neural networks. The traditional eutrophication classification scheme of Vollenweider was used for comparative purposes. The various artificial neural network simulations showed that, in addition to total phosphorus and inorganic nitrogen, turbidity, specific conductance, lake elevation and hydrogen ion concentration were identified as the most significant parameters affecting the classification of lakes in regard to their eutrophication status. These results suggest a conceivable association between these parameters and lake eutrophication, thereby indicating a need for further study on these relationships. A model simulation utilizing an unsupervised neural network did not provide much insight into the lake eutrophication status, but did show that the available physical–chemical lake data could be categorized according to physical region, thereby providing an indication that the lake data used in this study were region‐dependent.