A Study on Prediction of Saltwater Instrusion in Costal Aquifer using LSTM Algorithm

In Korea, groundwater is used as the main water resource, and there is a high possibility of groundwater pollution from saltwater intrusion caused by various groundwater developments and overexploitation. In this study, time series data such as daily average sea level, groundwater level, upper and lower electrical conductivity, rainfall, upper and lower water temperature, and LSTM algorithm was used to forecast the electrical conductivity, which is an indicator of seawater intrusion, for four stations in Kyungnam Haeun, Kyungnam Mokdo, Gangwon Joyang, and Incheon Sungyeo, which are severe level stations in the coastal areas of each area, in the rural groundwater management system. A time lag of 3 to 10 days was applied to each area, and out of a total of 3,438 univariate data, 2,406 days (70%) were trained, and 1,032 days (30%) were forecast and evaluated, with LSTM layers ranging from 8 to 256, batch size from 5 to 50 epochs, and parameters from 10 to 150. As one of the best forecasts, RMSE = 0.0066 and R2 = 0.9827 were performed in Gangwon Joyang. Afterward, RMSE = 0.0603 and R2 = 0.9856 were performed with the same parameters when predicting using the Shuffle technique.