Novel application of a back-propagation artificial neural network model formulated to predict algal bloom

This paper describes the novel application of an artificial neural network (ANN) model based on the back-propagation method formulated to predict algal bloom by simulating the future growth of five phytoplankton species and the chlorophyll a concentration in the second largest lake in Japan: eutrophic freshwater Lake Kasumigaura. Comparison of observed and calculated values showed that (i) seasonal variations in the biomass of Microcystis spp. were well-predicted with respect to the timing and magnitude of algal bloom, and (ii) the concentration of chlorophyll a, as an indicator of the total biomass of phytoplankton, was well predicted in general. The resultant correlations for the other species, however, showed that model learning was insufficient to effectively predict species biomass; thereby indicating that some unknown factors which are not represented by the set of water quality parameters used as model input data affect phytoplankton growth. A sensitivity analysis performed on input parameters showed that chlorophyll a concentration was mainly affected by PO4-P concentration, while cyanobacteria and diatom species were affected by NO3-N and NH4-N concentrations, respectively. These results indicate that the “algal bloom” ANN model achieved reasonable effectiveness with respect to learning the relationship between the selected water quality parameters and algal bloom.