2.1 Tide-surge-wave prediction model

We used a tightly two-way coupled tide-surge-wave model, i.e., the ADCIRC and Simulating Waves Nearshore (ADCIRC+SWAN) models (Dietrich et al., 2010) to predict waves and storm surges caused by Typhoon Chaba. The elevation and velocity were computed from the continuity equation in a two-dimensional depth integrated model (ADCIRC-2DDI) and the 2DDI momentum equations, respectively (Luettich and Westerink, 2004). The third-generation spectrum wave model, SWAN model (Booji et al., 1999; Ris et al., 1999) obtains the wave parameters by solving the wave action balance equation and integrating the two-dimensional wave energy spectrum in the frequency and direction domains. The ADCIRC and SWAN models exchange information with each other with the same unstructured grid. ADCIRC passes the input wind, water level, and currents processed by itself to SWAN to gain the wave spectrum by solving the wave action balance equation. Then, the ADCIRC model is driven by radiation stress gradients (Longuet-Higgins and Stewart, 1964) obtained from the SWAN model to compute the water levels and currents. Since there are many studies using ADCIRC+SWAN (Choi et al., 2013; Choi et al., 2018; Kerr et al., 2013; Xie et al., 2016; Yuk et al., 2015), a detailed model description is not be provided here.

The model domain covers the East Sea, Yellow Sea, and a part of the East China Sea. The model grid is unstructured, and thus can be made fine in the target areas and coarse in the other areas. Fig. 1 shows the grid in a part of the model domain. The grid size is at least 10 m in the south coast of Korea (Fig. 1(b)) and gradually increases, reaching about 30 km at the open boundary. The amplitudes and phases of eight tidal constituents (M₂, S₂, K₁, O₁, N₂, K₂, P₁, and Q₁) were assigned to the open boundary, which were obtained from Oregon State University (OSU) tidal databases (Egbert and Ray, 2003). As the main setting of ADCIRC, we used the default settings of the model, and previous studies (Choi et al., 2013; Choi et al., 2018; Yuk et al., 2015) using these settings have shown reasonable results. As the main setting of SWAN, for all experiments, we set the frequency increment factor, first frequency, and the number of frequencies and directions 1.1, 0.031384 Hz, 31, and 36, respectively. In this study, for the values of other variables, the default settings of SWAN were used.

Figure 1.

Unstructured grids with observation station locations with regard to the Typhoon Chaba track. (a) Wave buoys with blue stars; (b) tide observation stations with red squares.

2.2 Meteorological forcing

Storm surges and waves are caused by meteorological forcing such as wind and surface air pressure. Therefore, wind and surface pressure are used as meteorological inputs for the ADCIRC+SWAN model. The ADCIRC+SWAN model has the several types of meteorological forcing input, e.g., the wind and surface pressure to all nodes at the time specified in the model. Also, this model can be driven by the parametric tropical cyclone models such as symmetric / asymmetric Holland vortex model and generalized asymmetric Holland vortex model (GAHM) (Holland, 1980; Luettich et al., 2018). These parametric models can directly use typhoon track information in ATCF file format. GAHM (Gao et al., 2013; Musinguzi et al., 2019) was developed to realistically consider the asymmetric vortex and intensity of a tropical cyclone. Therefore, this model uses all available isotachs of R34, R50, and R64 (radii of 34 kt, 50 kt, and 64 kt wind) in the four quadrants (NE, SE, SW, and NW). Both hindcast and forecast experiments were performed and compared to evaluate the prediction performance. Table 1 shows the meteorological forcing inputs of the forecast and hindcast experiments.

Table 1.

Hindcast and forecast simulation cases

For the forecast experiments, we selected the parametric tropical cyclone models with the symmetric Holland vortex model and GAHM using the forecast track of the MPAS in ATCF file format. Fig. 1(a) shows the Typhoon Chaba track predicted by the MPAS from 00 UTC on 3 October. Compared to the best track of JWTC, the MPAS underestimated the intensity until about the 1-day forecast (18 UTC on 3 October); however, the predicted typhoon followed the best track well until the 2-day forecast (00 UTC on 5 October). The parametric model with the tropical cyclone track can produce the wind and surface pressure following its track. However, for a more accurate forecast, the wind and surface pressure generated along the track as well as in other areas are required (Choi et al., 2013; Yuk and Joh, 2019). Therefore, we tested the parametric models of the symmetric Holland vortex model and GAHM, and blended the same parametric models and forecasting model for our forecast experiments. Here, the forecasting model used was the MPAS model, and the typhoon track in the parametric model was extracted from the same MPAS model, thus allowing for the two to be blended. We generated the blended meteorological field by combining the wind and surface pressure fields along typhoon eye obtained from a parametric tropical cyclone model with MPAS track and those fields surrounding the typhoon produced by MPAS. According to the previous studies (Kawaguchi and Kawai, 2007; Tanemoto and Ishihara, 2013), the combined variables v_c are obtained from Eq. (1):

where v_p and v_m are the variables by the parametric model and forecast model, respectively. W is the weight at any point shown in Eqs. (2) and (3), as follows:

where R₁₅ is the radius at 15 m/s wind, f is the Coriolis parameter, and r is the distance from the center of the typhoon. Fig. 2 shows the blended wind fields 00 UTC on 5 October, using the symmetric Holland vortex (Simulation name of MT-Sy-B) and GAHM (MT-Asy-B). The wind near the center of typhoon was stronger for the asymmetric vortex considering the information of all radii (R34, R50, and R64).

Figure 2.

Wind field at 00 UTC 5 Oct. blended using the symmetric Holland vortex model (MT-Sy-B) (a) and GAHM (MT-Asy-B) (b) with the MPAS-forecasted track.

For the hindcast experiments, the parametric models with symmetric and asymmetric vortex methods identical to those used in the forecast experiments were implemented with the typhoon best track of the JTWC including R34, R50, and R64 in the four quadrants. In the hindcast experiments, the blending of the forecasting model and parametric model could not be performed because of the nonexistence of ambient wind and surface pressure data from the forecasting model.

4.1 Wind field

The GAHM simulated stronger wind near the typhoon center than the symmetric Holland vortex model (Fig. 2). Since the MPAS typhoon track information at 00 UTC on 5 October showed that the radii of 50 kt and 64 kt in the SE quadrant were larger than those in the other quadrants, GAHM determined a larger intensity in the SE quadrant from the typhoon center. In addition, since the wind field computed by the parametric model was inserted into the surrounding wind field, which was the forecast model result, the area near the center of the typhoon represents the typhoon-induced wind distribution from the parametric model, and the area outside the center of the typhoon indicates the wind distribution from the forecast model result. This means that a meteorological situation similar to the real one can be produced.

4.2 Significant wave height

We compared the significant wave heights of the 4 forecast experiments and observations at the 5 buoys, as well as the 2 hindcast experiments to find an improvement method that can predict waves more accurately. Regarding the distribution of significant wave heights, the four-parametric model (BT/MT-Sy/Asy) adequately predicted the maximum significant wave height without showing a significant difference from the observed value. Compared to using the MPAS-forecasted track (MT-Sy/Asy), the best track experiments (BT-Sy/Asy) predicted the peak significant wave height better. For the entire period, the RMSEs of the forecasts using the MPAS track were not significantly different from the RMSEs of the hindcasts, indicating substantial predictive performance of the significant wave height using the MPAS-forecast track.

As for the difference between symmetric and asymmetric vortices, the symmetric vortex case was generally better for both the best track and MPAS track, but the difference between the two was not significant. In addition, when only the parametric model is used, since there is only a meteorological field near the center of the typhoon, the significant wave height due to the meteorological field outside the typhoon influence area cannot be simulated. Therefore, the parametric model did not simulate a significant wave height that slowly rises before the peak significant wave height (Fig. 3). When blending was applied (comparison of MT-Sy/Asy and MT-Sy/Asy-B), RMSEs became smaller regardless of the symmetric and asymmetric condition. Regarding the significant wave height change during the typhoon, the case (MT-Sy/Asy-B) employing the track and blending method was closely correlated with the observed value. Also, in terms of the highest significant wave heights, the experiment where the GAHM with MPAS track and the blending (MT-Asy-B) were applied was almost identical to the highest observation. Here, the asymmetric vortex experiment showed better scores with 2 buoys, and the symmetric experiment with 3 buoys; therefore, it seems that the effects of symmetry and asymmetry are inconsequential for predicting the significant wave height. To accurately predict significant wave heights caused by typhoons, it is necessary to use not only the wind field showing rapid changes near the center of the typhoon, but also the wind field outside the typhoon influence area.

4.3 Storm surge

Since Typhoon Chaba caused severe damage to the southern coast, storm surge prediction performance was evaluated through the same method as the significant wave height analysis at the five representative tide stations in the southern coast.

Regarding the storm surge distribution, the four parametric models (BT/MT-Sy/Asy) show good agreement with the observations for the positively maximum storm surge. For the entire period at the five stations, the RMSE of the forecast using the MPAS track was slightly smaller than the RMSE of the hindcast; thus, the performance of predicting storm surges using only the MPAS track is slightly better. Regarding the effects of the symmetric and asymmetric vortices, in the experiment using the best track, the experiment using the symmetric method generally shows better results, whereas in the experiment using the MPAS track, the experiment using the asymmetric effect shows remarkably improved results (Table 3). When blending is applied, i.e., when MT-Sy/Asy and MT-Sy/Asy-B are compared, the RMSE is higher for both symmetric and asymmetric experiments. Therefore, in this study, it can be understood that storm surge prediction performance is reduced when blending is used. However, comparing the RMSE of the storm surge predicted by blending (MT-Sy/Asy-B) with that of best track (BT-Sy/Asy), MPAS is generally slightly smaller than the RMSE of the storm surge reproduced using the best track. Therefore, using blending does not significantly degrade storm surge prediction performance.

The storm surge prediction performance is the best when using the GAHM with the MPAS-forecasted track. In addition, even if the GAHM-based meteorological field is blended with the meteorological field derived from the forecasting model, it is comparable to the hindcast using the best track. Therefore, storm surge prediction using the typhoon track from the meteorological model, MPAS and meteorological data from that model are reliable.