flow regulation for water quality (chlorophyll a) improvement

in this study a machine learning algorithm was applied in order to develop a predictive model for the changes in phytoplankton biomass (chlorophyll a) in the lower nakdong river, south korea. we used a “hybrid evolutionary algorithm (hea)” which generated model consists of three functions ‘if-thenelse’ on the basis of a 15-year, weekly monitored ecological database. we used the average monthly data, 12 years for the training and development of the rule-set model, and the remaining three years of data were used to validate the model performance. seven hydrological parameters (rainfall, discharge from four multi-purpose dams, the summed dam discharge, and river flow at the study site) were used in the modeling. the hea selected reasonable parameters among those 7 inputs and optimized the functions for the prediction of phytoplankton biomass during training. the developed model provided accurate predictability on the changes of chlorophyll a (determination coefficients for training data, 0.51; testing data, 0.54). sensitivity analyses for the model revealed negative relationship between dam discharge and changes in the chlorophyll a concentration. while decreased dam discharge for the testing data was applied; the model returned increased chlorophyll a by 17-95%, and vice versa (a 3-18% decrease). the results indicate the importance of water flow regulation as specific dam discharge is effective to chlorophyll a concentration in the lower nakdong river.