Prediction model for red tide occurrence by Heterocapsa circularisquama in Ago Bay (Mie Prefecture) was constructed by means of fuzzy neural network (FNN) or multiple regression analysis (MRA). Six-year-data sets on the occurrence of H. circularisquama in the observation point, Tategami-ura, and the hydrographic parameters of seawater and the wind condition of the bay area were used. Based on the collected data, 24 kinds of available variables were prepared for modeling by parameter increasing method (PIM). The constructed FNN model could predict the occurrence in each year, while prediction by MRA never succeeded; the recognition index for the threshold value of 0.5 was 70% and 0% in the FNN and MRA models, respectively. When the three variables, the maximum strength of south wind in the last week, the 2 weeks ago salinity, and the 2 weeks ago temperature difference between at the surface and at the 5-m depth of seawater, were selected, the occurrence of red tide could be predicted more correctly, and the recognition index of the FNN model increased to 80%. When the threshold value decreased to 0.1, 100% recognition index was obtained in both models. However, 38 false-positive answers were obtained in the MRA model, whereas that of FNN model was only 5. From the constructed FNN model, IF-THEN rules were extracted linguistically as follows; if the big south wind flowed when the salinity was high and poor mixing in vertical direction occured, the red tide will occur with high risk. The IF-THEN rule corresponded well to the empirical knowledge.
All Science Journal Classification (ASJC) codes
- Chemical Engineering(all)