Community patterning and identification of predominant factors in algal bloom in Daechung Reservoir (Korea) using artificial neural networks

Abstract

The water quality and phytoplankton communities in the Daechung Reservoir, Korea, were monitored from summer to autumn in 1999, 2001, and 2003. The temporal patterns of cyanobacterial blooming caused by Microcystis were then elucidated using a combination of two artificial neural networks (ANNs): self-organizing map (SOM) and multilayer perceptron (MLP). The SOM was initially used to cluster the phytoplankton communities, then the MLP was applied to identify the major environmental factors causing the abundance of phytoplankton in the clustered communities. The SOM divided the phytoplankton communities into four clusters based on their algal composition (Cyanophyceae, Chlorophyceae, Bacillariophyceae, and others). In particular, cluster II was mostly composed of sampling times in August and September, and closely matched the period of severe cyanobacterial bloom dominated by Cyanophyceae. Meanwhile, cluster IV was mainly composed of the samples collected in the other periods, covering April, May, June, and October, and was mostly dominated by Bacillariophyceae. Cyanophyceae was the main component of the total algae, and its variation among the clusters showed a similar pattern to that of the changes in the chlorophyll-a concentration. Based on the MLP model, the water temperature, total particulate nitrogen, daily irradiance, and total nitrogen were highlighted as the four most important environmental variables predicting cyanobacterial abundance, yet quite different environmental variables were found to affect the chlorophyll-a concentration. The usage of sampled data and analyses by ANNs are also discussed with reference to an early alert system for algal bloom.