Predicting dominant phytoplankton quantities in a reservoir by using neural networks

Abstract

The Levenberg-Marquardt algorithm was used to train artificial neural networks to predict the abundance of Cyclotella ocellata Pant. and Cyclotella kutzingiana Thwaites using time, depth, temperature, pH, dissolved oxygen, and electrical conductivity as input parameters for the oligo-mesotrophic Kuzgun Dam Reservoir, Turkey. The data were collected in monthly intervals during two ice-free seasons: between April 2000-November 2000 and April 2001-November 2001. To reduce over-fitting of the neural network based models, we employed single hidden layer networks with early stopping of training. Correlation coefficients, of neural network predictions with measurements of abundance of Cyclotella ocellata Pant. and Cyclotella kutzingiana Thwaites were 0.88 and 0.86, respectively.