In this study, the FORMOSAT-2 satellite data was used to estimate the bathymetry directly or through unsupervised classification first. Both the satellite retrieved water depths were corrected by the in-situ tidal data. The no-classified results show that more than 82% satellite-derived water depths have the error below 20% comparing with the in-situ data. Meanwhile, its root mean square error is 2.5 m. After classification, it is found that class 4 and 5 are located on the water region. Therefore, the retrieved water depth can be calculated just in class 4 and 5 regions. At class 4, there are 77% of data points having the error less than 20%. Its root mean square error is 2.7m. At class 5, more than 81% data points with the error below 20% and its root mean square error is 1.0m. The results indicate that the shallow water depth could be mapped accurately from FORMOSAT-2 image. Moreover, the satellite data has the potential to become an important tool for bathymetry map based on its spatial coverage, frequent interval, and safety.