Prediction of fecal coliform using logistic regression and tree-based classification models in the North Han River, South Korea

In this study, data-based classification models were developed for real-time prediction of the exceedance of the safety level on fecal coliform in Daesung-ri site of North Han River. The prediction models were developed using the logistic regression model (LRM) and the tree-based models such as classification and regression model (CART), bagging model (BGM), and random forest model (RFM). For model development, rainfall, water quality, and dam discharge data from 2010 to 2015 were collected from the study site. Clustering methods were applied to reduce the sampling bias of training and test datasets and to improve the prediction accuracy. The developed four models were compared with each other in terms of prediction accuracy and applicability. The test results of developed models showed that the total correct classification rate of the four models ranged from 83.7% to 93.0%. Each classification model showed its own strengths; LRM offered flexibility by tuning cutoff values, while RFM showed the highest accuracy among the four models. The hydro-ecological process on fecal coliform could be explained by analyzing important variables in the prediction models and identifying the impacting factors through the field monitoring. The important factors both in the models and field monitoring were revealed as the rainfall-related variables, dam discharge and total phosphorus, which imply that the fecal pollution in North Han River came mainly from the rainfall events and runoff including nutrients from farmland and livestock farming in the upstream basin of Guwoon Creek and Chungpyung Dam.