Rietvlei, Hartbeesport and Roodeplaat are warm-monomictic and hypertrophic South African reservoirs that experienced recurrent blooms of the dinoflagellate Ceratium hirundella over the past ten years.
Merged limnological monitoring data of 12 years from the 3 lakes were used to assemble rule-based agents for forecasting Ceratium hirundella by means of the hybrid evolutionary algorithm HEA. The best performing Ceratium agents for 7, 14, 21 and 28 days ahead forecasting from both split sample and bootstrapping training and testing were documented and compared.
The expected outcomes of this research are: 1) comparing potential of the split sample and bootstrapping methods utilised by HEA; 2) determining causal relationships between Ceratium biomass, water quality and other algal groups such as diatoms, green algae and blue-green algae; 3) generalising predictive rule-based agents for Ceratium hirundella for three similar lakes; 4) forecasting Ceratium blooms 1,2,3 or 4 weeks ahead as a prerequisite for early warning