Modeling Of Algal Blooms In FreshwatersUsing Artificial Neural Networks

The development of a neural network model for predicting algal blooms is described. The neural network consists of a 3 layer structure with input, hidden, and output layers. Training is conducted using back-propagation where the data are presented as a series of learning sets such that the inputs are observable water quality parameters and outputs are the biomass quantities of specific algal groups. Training is conducted using three years of daily values of water quality parameters and validation is performed using two years of independent daily values to predict the magnitude and timing of blooms of 7 different algae groups with a lead time of 1 day using only the current day water quality parameters. The water quality data represent physical and limnological characteristics of a drinking water reservoir in Germany. Results indicate that the neural network model is capable of learning the complex relationships describing the seasonal succession of phytoplankton in freshwaters. The neural network is shown to perform well for predicting both the timing and magnitude of algae blooms for data in used in training set and to accurately predict the timing and typically overor under-estimate the magnitude of blooms when applied to the independent data.