Forecasting of Microcystis aeruginosa seasonal dynamics using the fuzzy logic and fuzzy neural networks

Abstract

The retrieval of potential predictors of blue-green algae M. aeruginosa blooms and bloom prediction using fuzzy logic rules and fuzzy neural networks is discussed. Time series of seasonal dynamics of M. aeruginosa quantities and parameter values were obtained through field observations of biotic and abiotic parameters of the water environment held at Lake Smolino (Chelyabinsk) in the warm season of 2009 and 2011. The cross-correlation analysis of the data revealed the potential predictors of M. aeruginosa abundance quasiperiodic oscillations with a period of 12-20 days: algae P. duplexabundance, water temperature and the concentration of nitrates. According to the results of cross-correlation analysis a number of rules and membership functions in a range of changes from zero to one is set forward. Specially written program was used to train the fuzzy neural network in data on the values of predictant and selected predictors to apply the predictive rules and membership functions automatically. To compare the results additionally performed a linear extrapolation of the predictant abundance. Seasonal development of M. aeruginosa was well predicted by the extrapolation forecast only on quasilinear intervals of M. aeruginosa abundance evolution, whereas fuzzy logic theory was good to predict the M. aeruginosa intense outbreaks.