Author(s): Yuqing Lin; Min Wang; Qiuwen Chen; Arthur E. Mynett
Keywords: Cellular automata; Cyanobacteria blooms; Genetic programming; Thiessen polygons
Abstract: Cyanobacteria blooms are a serious problem around the world and have caused severe ecological and social-economic damage. Data-mining models have proven effective at predicting such blooms; however, few of them can provide the spatial dynamics. Although process-based numerical models do provide a way to model, such approach often involves too many parameters to be calibrated. This study took the case of Taihu Lake and developed a model that embedded a data-mining technique into a cellular automata configuration, so as to obtain the spatial-temporal dynamics of cyanobacteria blooms. The lake was divided into polygons in accordance with the monitoring stations by using the Voronoi method. Data on flows, water quality and phytoplankton were collected from monitoring stations inside Taihu Lake. Genetic programming was applied to establish predictive formulations for the cyanobacteria population dynamics in relation to flow and water quality variables by using the collected data. The formulations accounted for local interactions between polygons as the evolution rules for the cellular automata. The results show that in this way CA models are able to predict both approximate magnitudes as well as accurate timing of cyanobacteria blooms quite well for all areas except for regions with lower cyanobacteria population. Overall the CA model shows very promising performance in capturing the spatial-temporal dynamics of algal abundance in lakes.