Hurricane Irma occurred in the Atlantic Ocean in 2017. Because there is concern that the risk of storm surge is increasing, it is important to predict storm surge for prevention of the disasters. For that purpose, it is important to construct a method that can predict local area phenomena. Therefore, in this study, to achieve that purpose, a methodology of downscaling the mesoscale data is developed by using STOC-ML and WRF, and the applicability is verified to the storm surge due to Irma. In addition, wave set up is also estimated using SWAN and the characteristics of Irma is considered.