GMDH algorithms applied to turbidity forecasting

By applying the group method of data handling algorithm to self-organization networks, we design a turbidity prediction model based on simple input/output observations of daily hydrological data (rainfall, discharge, and turbidity). The data are from a field test site at the Chiahsien Weir and its upper stream in Taiwan, and were recorded from May 2000 to December 2008. The model has a regressive mode that can assess the estimated error, i.e., whether a threshold has been exceeded, and can be adjusted by updating the field input data. Consequently, the model can achieve accurate estimations over long-term periods. Test results demonstrate that the 2006 turbidity prediction model was selected as the best predictive model (RMSE = 5.787 and CC = 0.975) because of its ability to predict turbidity within the acceptable error range and 90 % required confidence interval (50NTU). 70(3,1,1) is the optimum modeling data length and variable combinations.