Forecasting performance of algae blooms based on artificial neural networks and automatic observation system

In recent, South Korea has faced on water quality management problems in reservoir and river because of increasing water temperature (Tw) and rainfall frequency caused by climate change. For these reasons, South Korea has set up automatic water quality monitoring system for preventing early algae blooms in five representative watershed. Also, Government makes a greater effort for preparing remedy with numerical models which handle water quality accidents quality in advance by predicting variation of water quality factors on account of change of weather conditions and source of pollutants in the future. Many countries have conducted various studies on forecasting water quality by artificial neural network which has a number of advantages, as opposed to the traditional models based on methods like data driven self-adaptive methods, generalization through learning the data presented, universal functional approximators, and nonlinear for forecasting. Daecheong reservoir located in Geum river has suitable environment for algae blooms because it has lots of contaminants that are flowed by rainfall in Daejeon and Chungcheong area. This study selected Daecheong reservoir in the Geum river and used the data of the automatic water quality observation system. By using back propagation algorithm (BPNN) of feed forward neural networks, a model has been built to forecast the algae blooms over short periods (1, 3 and 7 d). In this model, input parameters considered the hydrologic and water quality factors as following: inflow, outflow, average areal precipitation, air temperature (Ta), Tw, dissolved oxygen (DO), total organic carbon (TOC), pH, chlorophyll-a (chl-a), total nitrogen (TN), and total phosphorous (TP) in Daecheong reservoir. However, the chl-a data of automatic water quality observation system has some missing data caused by defect and maintenance in the system. Through carrying out correlation analysis, interpolation has been implemented by neural network with BPNN. Correlation analysis has been implemented to analyze lag time and components that sensitively responded to chl-a by referring the interpolated data and water quality and hydrologic factors in all. Based on the results of the data, after selecting input parameters for algae bloom prediction model, each case has been verified along with making various models. As a result of this research, the short term algae bloom prediction models showed minor errors in the prediction of the 1 d and the 3 d. Comp