High possibility of occurrence of earthquake with M9 or lager in the Nankai subduction zone was pointed out by Cabinet Office of Japanese government. Local governments along the area revised estimation of tsunami damages and attempt to reconstruct action plans for the disaster prevention. However, according to Cabinet office of Japanese government, the coastal area near the rupture zone receives huge tsunami within a few minutes after the earthquake happens. To take actions against the severe situation, we need a high-speed, and high-accurate tsunami prediction system. Baba et al. (2014) investigated the possibility for use of a concept of tsunami amplification in the early tsunami prediction. They computed tsunami waveforms at the twenty pressure gauges of dense ocean floor network system for earthquakes and tsunamis (DONET) and at prediction points near shore. They found clear correlations between them because tsunami height basically depends on the topography (bathymetry) during its propagation. In this study, an early tsunami prediction system using the concept of tsunami amplification was societally implemented in 6 regional areas in Wakayama Prefecture. We constructed a tsunami database that contains pre-computed tsunamis offshore and nearshore from 1506 earthquake scenarios may occur in the Nankai subduction zones. The new system detects first arrivals of earthquake and tsunami from DONET data in real time, and calculate average value of absolute observed pressure values among twenty DONET stations. The value is used to select an appropriate scenario from the tsunami database. Prediction accuracy of the system was also investigated by using cases of the 1944 Tonankai earthquake and the scenario earthquake provided by the Cabinet office. As a result, we found that predicted inundation area to be overestimated as the safety of the prediction.