인공지능 모델의 앙상블이 제주도 중산간지역 지하수위 예측 향상에 미치는 영향 = Impact of an ensemble of artificial intelligence models on improving groundwater level prediction in mid-mountainous region of Jeju Island

다국어 초록 (Multilingual Abstract)

Groundwater is a water resource that can be used for various purposes along with surface water. In the case of Jeju Island, groundwater is an essential water resource, so accurate groundwater level prediction and management for the distant future are necessary for sustainable groundwater use. In this study, groundwater levels were predicted using three artificial intelligence(LSTM, GRU, ANN) models for accurate long-term(1-3 months) future monthly groundwater level predictions for two groundwater level monitoring wells located in the mid-mountainous region of the Pyoseon watershed in Jeju Island. Afterwards, an ensemble model was used to analyze the improvement in groundwater level prediction for the entire data period and the low groundwater level period(November to May). As a result, the AI ​​models and the ensemble model appropriately predicted future groundwater levels(1 to 3 months) for the entire data period, and the ensemble model showed higher prediction performance than the individual AI models. The superiority of the groundwater level prediction performance of the three AI models varied by monitoring well and future prediction period, and a specific AI model did not always show the highest groundwater level prediction performance. Therefore, for more improved groundwater level prediction, an ensemble model that utilizes the results of different artificial intelligence models is needed. The groundwater level prediction performance for the low groundwater level period was higher than that for the entire data period. This means that the AI ​​models and the ensemble model are more suitable for groundwater level prediction during the low groundwater level period, which mostly corresponds to the groundwater level recession curve period. In particular, the ensemble model showed an appropriate NSE value of 0.7184 or higher for 3-month predictions, and this model produced prediction results with an NSE value that was improved by up to 0.1434 compared to individual AI models. This supports the importance and necessity of using ensemble models for accurate prediction of low groundwater levels in the distant future.