EVALUATING THE CONTRIBUTING ENVIRONMENTAL PARAMETERS ASSOCIATED WITH EUTROPHICATION IN A SHALLOW LAKE BY APPLYING ARTIFICIAL NEURAL NETWORKS TECHNIQUES

Eutrophication is a serious problem that affects water quality and it may cause Harmful Algal Bloom (HAB) events with many unpleasant consequences. For that purpose, an Artificial Neural Network (ANN) was developed able to forecast one month ahead the Chlorophyll-a (Chl-a) levels and by that way to act as a warning tool when a HAB event might follow. Sampling data from eleven monitoring stations were collected from Lake Pamvotis (Greece), a shallow hypereutrophic lake, affected by HABs. The created ANN managed with high accuracy to simulate the next month's Chl-a concentration, establishing it as a reliable predictor that represents well the non-linear relationships between the Chl-a and the environmental parameters. The significance of each environmental parameter associated with eutrophication was also examined. Focusing on the contribution of the environmental parameters three different methods that give the ANN model sensitivity are applied: (i) the 'Perturb' method; (ii) the 'Weights' method; (iii) the `PaD' ('Partial Derivatives') method. A combined parameter importance index was introduced, in order to overcome the computational differences resulted from the three methods. The combined interpretation of the results produced led to useful conclusions regarding the effect of each parameter on the eutrophication process.