Application of Machine Learning Methods for Groundwater Level Forecasting in the eastern region of Jeju Island

This study applies ANN, ELM, LSSVR, and LightGBM as machine learning models for groundwater level forecasting in the eastern region of Jeju Island, South Korea. The groundwater level forecasting performance of the applied models is evaluated through model evaluation indices. In forecasting the groundwater levels with strong autocorrelation, the comparative models all show excellent performance in the 1-day forecasting. In contrast, the forecasting performance deteriorates in the 7-day forecasting. The results also show that the forecasting performances among the comparative models are almost similar. In addition, it is confirmed that in coastal areas, groundwater level forecasting performance deteriorates compared to inland areas due to the influence of tidal variation. When forecasting results are similar in forecasting the groundwater levels with strong autocorrelation, models that do not require relatively complex detailed tuning, including ANN and ELM, can have excellent applicability. Therefore, groundwater level forecasting using machine learning models can be an effective tool for groundwater resources in Jeju Island.