수치해석 접근법에 따른 지하수 공간분포 예측

Abstract

This study applied the MODFLOW model using finite difference method and the NNRK model combining neural network and residual kriging for spatial groundwater level prediction in Jeju Island, South Korea. For the MODFLOW model, the differences between observed and predicted groundwater levels were 0.014-4.327 m and the relative errors were 0.50-50.82 %. For the NNRK model, the differences between observed and predicted groundwater levels were 0.022-6.185 m and the relative errors were 0.66-49.35 %. Based on MSRE, IDW model, which is a non-spatial statistical model, showed better spatial prediction results than SK and OK models, whereas RK and NNRK models outperformed the IDW model. The MODFLOW model showed relatively better results than the NNRK model in dense areas considering recharge and geological structure although the MODFLOW and NNRK models produced similar results. Furthermore, the MODFLOW model was found to effectively reflect groundwater flow by strata and temporal groundwater head fluctuations due to pumping compared to the NNRK model.