Mountain tunnel excavations using the New Australian Tunneling Method should be assessed in light of the method's impact on spring water in the excavation and the surrounding water environment. Water flow across a tunnel (excluding watertight tunnels) has long been modeled as a well spring, which can be predicted by employing Kogiso's Sequential Data Assimilation -System on Water Information of Ground (SDA-SWING) method, a spring prediction method using Bear's formula. However, Bear's formula is a well formula based on Dupuit's quasi-uniform flow that, when applied to tunnel excavation, always yields an error in the results due to the flow dimension. SDA-SWING method solves this error by applying an ensemble Kalman filter to correct the hydraulic conductivity so that the calculation result matches the actual measurement. Nevertheless, such an error positions a great influence on the future prediction accuracy. Thus, in this study, we have improved the SDA-SWING method to consider three-dimensional flow. We used the method in simulating 3D seepage flow analysis for a simple tunnel excavation model and verified its ability to reproduce 3D flow and improve the future prediction accuracy.