Assessment of nitrate contamination vulnerability using SVR-kernel models: A case study of Miryang City, Korea

The assessment of groundwater contamination vulnerability is crucial for the effective management and conservation of groundwater. The purpose of this study is to determine the most accurate SVR-kernel model among 4 kinds of SVR-kernel models (Linear, Radial, Polynomial and Sigmoid models) for the assessment of groundwater contamination vulnerability to nitrate components in Miryang City of Korea (South) which has two functions of urban and rural activities. The split ratio for training and testing of input data was determined as 7:3, using 4 kernel models on basis of statistical analyses (Mean absolute error (MAE), RMSE (Root mean square error) and Correlation Coefficient ()). Some statistical data of MAE, RMSE and , ROC/AUC, graphical comparisons between target nitrate values and predicted nitrate values for training and testing data, and spatial maps of predicted nitrates were used for the evaluation of 4 kinds of SVR-kernel models. SVR- Polynomial model made the least MAE (0.045) and RMSE (0.063), and largest Correlation Coefficient (0.887) and ROC/AUC (0.83) for the test data. Graphical comparisons and spatial maps also represented the superiority of SVR- Polynomial model. DRASTIC thematic maps and predicted nitrate maps indicated the high groundwater contamination vulnerability around Nakdong River and Miryang Stream. Thus, the effective countermeasure was required for the management and conservation of groundwater in Miryang City of Korea (South). It is considered that this study can contribute to the determination of sustainable utilization of groundwater in the world.