1 Introduction

On 31 March 2007, a storm surge struck Yeonggwang (35.3°N, 126.5°E). The storm surge that originated in the Yellow Sea resulted in flooding and significant coastal damage in the vicinity of Yeonggwang. As a result of the surge, about 38 families or approximately 87 people were made homeless and 4 died in the area. Property damage in the Yeonggwang area was estimated to exceed 1.6 million dollars. Seventy-two fishing boats were destroyed and 79 marine product facilities were submerged. The intensity of the disaster, calculated using a factor of 1.8 by the method of Feng and Hong (2008), indicates that it was a very small disaster. The invading storm surge was accompanied by a high tide and the recorded sea level reached 703 cm. This corresponds to a sea level that is 102.6 cm higher than the Yeonggwang high tide (Fig. 1) on 1631UTC 30 March 2007. The wind speed increased substantially about 1 h prior to the storm surge. Moon et al. (2003) noted that Yeonggwang is located on the west coast of Korea, which is one of the strongest tidal areas in the world. They also concluded that the sea levels can be higher than when the tides are average if the astronomically enhanced tide levels coincide with the passage of a cyclone.

Fig. 1
figure 1

Temporal variation in observed sea level height (solid line, cm) and wind speed (dashed line, m s−1) at the Yeonggwang station

Full size image

Figure 2 shows the surface weather chart and the QuikSCAT sea surface wind (SSW) for this time period. On 09 UTC 30 March 2007, a mesoscale cyclone and an anticyclone were located off the west coast of Korea (Fig. 2a). In the weather map, the storm surge is assumed to be developing with the mesoscale cyclone. The cyclone was generated rapidly, as shown in the black box of Fig. 2b on 15 UTC 30 March 2007 when the storm surge occurred. Storm surges occur frequently, due to the presence of typhoons over the west or south coast of the Korean Peninsula. However, this is an exceptional case, in that a meso-β scale cyclone was generated, with a centre pressure of 1010 hPa at Yeonggwang, near the west coast of Korea (Fig. 2b). In order to accurately predict a storm surge, precise data on sea level pressure (SLP) and SSWs are required. Figure 2c and d shows the SSW obtained from the QuikSCAT satellite at 1012 UTC and 2124 UTC, respectively. The high-wind speed over the Yellow Sea had strengthened, which corresponds with the enhanced pressure gradient between a cyclone and an anticyclone (Fig. 2b). Cyclonic circulation occurs around 35°N, 125°E in Fig. 2b.

Fig. 2
figure 2

Observed surface weather map on a 0900 UTC 30 March 2007, b 1500 UTC 30 March 2007, and observed sea surface wind by QuikSCAT on c 1012 UTC 30 March 2007 and d 2124 UTC 30 March 2007

Full size image

Very few studies can be found in the literature (Seo and Chang 2003; Kim et al. 2006; Heo et al. 2008) concerning the simulation of SSWs using regional models near the Korean coastline. Seo and Chang (2003) analyzed the characteristics of monthly mean SSWs and wind waves near the Korean marginal seas in the year 2002, based on predictions of SSWs using the MM5/KMA (Fifth-generation Mesoscale Model/Korean Meteorological Administration) model. Kim et al. (2006) carried out a sensitivity experiment for SSWs based on PBL schemes (Medium-range Forecast, MRF and Mellor-Yamada-Janjic, MYJ) and dynamic frames of MM5 and Weather Research and Forecasting model (WRF). Heo et al. (2008) reported on optimal combinations for parameterization for simulating SSWs using MM5 and WRF for cases of strong winds, such as the Typhoon Shanshan (0613) and the development of mesocyclones in the winter of 2006. It was also found that PBL parameterization plays a crucial role in mesoscale simulation and has a significant influence on the magnitude of simulated wind speeds.

In this study, we attempted to select the optimal physics combination for a high wind (Davis and Low-Nam 2001; Ivanov and Palamarchuk 2007) using the MM5 and the WRF. In order to investigate the atmospheric states that lead to a storm surge, MM5 and WRF the optimal physics, and the Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS) model were compared in terms of the time evolution of a surface wind and SLP. Finally, we investigate the causes of mesocyclone development and the role of this phenomenon in the generation of a storm surge along the west coast of Korea using the COHERENS (COastal Hydrodynamical Ecological Model for REgioNal Shelf seas) model.

2 Models

2.1 Atmospheric model

MM5 version 3.7, WRF version 2.2 and COAMPS version 3.1 were used in this study. The nested domains over the centre of 35.0°N and 129.0°E are covered with 115 (in longitude) × 124 (in latitude) grids in the 27-km mesh and 175 (in longitude) × 238 (in latitude) grids in the 9-km mesh systems, respectively. The initial and boundary conditions are based on the NCEP Final Analysis (FNL) for MM5 and WRF. Initial and boundary conditions for COAMPS were obtained from NOGAPS and several sets of GOES satellite data, QuikSCAT and SSMI. The simulations were executed from 00 UTC 30 March 2007.

In order to select the physics options for MM5 and WRF, three sets of experiments (EXP1, EXP2 and EXP3) were carried out for high-wind conditions. EXP1 and EXP2 used Eta PBL (Janjic 1990; Mellor and Yamada, 1982) and EXP3 used MRF PBL (Hong and Pan, 1996). EXP1 and EXP2 are recommended to be the best physical parameterizations for tropical cyclone prediction (Davis and Low-Nam 2001) and EXP3 arose from the best optimal parameterization scheme sets for atmospheric variables such as temperature and geopotential height (Ivanov and Palamarchuk 2007). It has been reported that the MRF PBL scheme (Hong and Pan 1996) tends to produce boundary layers that are too deep and dry outside the eye wall of mature hurricanes (Braun and Tao 2000) but may be adequate if the storms are weak. For the cumulus scheme, EXP1, EXP2 and EXP3 used the Betts–Miller scheme (Betts 1986; Betts and Miller 1993; Janjic 1994), Grell scheme (Grell 1993) and Kain–Fritsch scheme (Kain and Fritsch 1993), respectively. The Betts–Miller (Betts and Miller 1993) scheme is the most popular for tropical systems, and the Grell scheme is routinely run near 10-km grid spacings. Davis and Low-Nam (2001) reported that physical schemes are very sensitive to regional and synoptic situations. The selected physics schemes were applied to the MM5 and WRF model, and the results were then compared with the output from the COAMPS model.

2.2 Storm surge model

We used COHERENS, an integrated model for coastal and regional seas. COHERENS is a multi-purpose three-dimensional hydrodynamic model for coastal and shelf seas, which solves the continuity, momentum, temperature and salinity transport equations using finite differences on an Arakawa-C staggered Cartesian, sigma-coordinate grid. The model is coupled to biological, resuspension and contaminant transport models and can resolve mesoscale to seasonal scale processes (Luyten et al. 1999). The model allows the choice of several advection schemes, for momentum and scalars, and turbulence closure schemes. The model has been systematically tested and applied in several studies (Umgiesser et al. 2002; Luyten et al. 2003; Lacroix et al. 2004; Marinov et al. 2006).

We used the above model to calculate storm surges in this study. A forcing term in the model takes variations in tide, air pressure and wind stress into consideration. The domain covered the entire Yellow Sea and the East China Sea (24°N–43°N, 117°E–131°E) with a 1/12° × 1/12° grid resolution. Ten layers were used in the vertical (Δσ = 0.1). SSW, SLP and oceanic currents are considered to be external forcing in the surge model.

3 Optimal combination of parameterization

In order to select the optimal physics combination for a high wind, we executed three experiments (EXP1, EXP2 and EXP3) with different parameterizations. In order to evaluate the horizontal distribution of simulations, BIAS (mean error), RMSE (root mean square error), SI (scatter index) and CORR (correlation coefficient) statistics were used (given in Table 1). In this case, N is the total number of grid points except for land and missing data. \( W_{\bmod } \left( {W_{\text{obs}} } \right) \) is the wind speed of the model (observation) and \( \overline{{W_{\bmod } }} \left( {\overline{{W_{\text{obs}} }} } \right) \) is the mean wind speed of the model (observation) for the area.

Table 1 Equations for validation scores
Full size table

Table 2 shows the result of the statistics over the entire domain on 09 UTC 30 March 2007 and 21 UTC 30 March 2007 when the QuiKSCAT wind was observed. All experiments have negative values on 09 UTC 30 March, while all experiments have positive values on 21 UTC 30 March for the BIAS. That is, the wind speeds of the models are underestimated on 09 UTC and overestimated on 21 UTC. RMSE and SI have lower values and CORR has a higher value for EXP3 than the other experiments on 09 UTC 30 March 2007. EXP1 and EXP2 have better values in terms of SI and CORR on 21 UTC 30 March 2007 for the cases of MM5 and WRF, respectively. Overall, EXP2 showed good performance in terms of the distribution and the magnitude of the wind speed in this case.

Table 2 BIAS, RMSE, SI statistics and the correlation of simulated wind speed by the MM5 and WRF models over the entire domain on 09 UTC 30 March 2007 and 21 UTC 30 March 2007
Full size table

4 Model comparison—WRF, MM5 and COAMPS

In order to examine the simulations of cyclone development, the runs for MM5 with EXP2, WRF with EXP2, COAMPS were compared. The horizontal distribution of SLP simulated by the three models on 1600 UTC 30 March 2007 at peak amplitude is shown in Fig. 3. The models simulate the cyclone in the southern part of western Korean Peninsula reasonably well, and the results show and anticyclone to the north of the cyclone as shown in the black box of the figure, even though their intensities are different. As a result, the WRF simulates the infrastructure of the developed meso-β scale cyclone, whereas MM5 and COAMPS show a weak trough westward from the south-western part of the Korean Peninsula.

Fig. 3
figure 3

Horizontal distribution of sea level pressure simulated using a MM5 EXP2, b WRF EXP2 and c COAMPS on 1600 UTC 30 March 2007

Full size image

In order to compare the observed maximum wind speed and the simulated wind speed on 16 UTC, Fig. 4 shows the wind vector and wind speed at a level of 10 m, as simulated by the three models at 16 UTC. The wind speed simulated by the three models corresponds to the pressure gradient shown in Fig. 3, reasonably well. In Fig. 4a–c, the WRF simulation indicates that the wind speed is in excess of 13 m s−1 at Yeonggwang shown in Fig. 1. However, the wind speeds simulated by MM5 and COAMPS tend to be weaker than the observed speeds. In order to investigate the potential role of the atmospheric states in mesocyclone development, we calculated the power dissipation indexes (PDIs) proposed by Emanuel (2005) around the mesocyclone, based on the models, defined as the time integral of the cube of the maximum wind. In order to compare them easily, we divided the PDIs based on WRF and the COAMPS by PDI based on MM5. The PDI ratios of WRF and COAMPS to MM5 were found to be 1.8 and 1.2, respectively. The influence of an 80% increase in PDI of WRF and a 20% increase in PDI for COAMPS are presented in Table 3. These data were obtained using the method for calculating the magnitude of a storm surge proposed by Feng and Hong (2008). The magnitude of the storm surge was 0.7, as obtained using the WRF model, which indicates a very small storm surge, where the observed maximum elevation in sea level is 1.0 m; the simulated maximum wind speed is 19.2 m s−1; and the simulated minimum pressure in the centre of the mesocyclone is 1003.2 hPa. In the results using the MM5 and COAMPS models, the magnitudes of the storm surges are −0.3 and 0.0, respectively, indicating that the storm surge is insignificant. These results suggest that it is important to accurately simulate the mesocyclone for fostering a strong pressure gradient because the response of this situation to a storm surge is sensitive.

Fig. 4
figure 4

Horizontal distribution of wind speed and wind vector simulated using a MM5 EXP2, b WRF EXP2 and c COAMPS on 1600 UTC 30 March 2007

Full size image
Table 3 Wind forces and quantitative calculations for the magnitude of the storm surge reported by Feng and Hong (2008)
Full size table

For the analysis of the atmospheric state in mesoscale cyclone development, Fig. 5 shows the air temperature at 900 hPa and the wind vector, in an attempt to examine the warm advection by low-level winds, which is related to the induction of convective instability. The analysis data were obtained from objectively reanalyzed data of the final analysis (FNL) issued by the National Center for Environmental Prediction (NCEP). The data are available four times daily at 00, 06, 12 and 18 UTC, and have a 1° × 1° horizontal resolution and 26 vertical levels. In Fig. 5, the warm advection is located to the north of east China on 0600 UTC. As time passes, the warm advection regions are extended into the Yellow Sea to the west of Jeju Island (around 33°N, 126°E). Finally, on 18 UTC, a strong warm advection develops over the western coast of Korea, which would enhance the development of a mesoscale cyclone.

Fig. 5
figure 5

Horizontal wind vector and temperature (shaded area) at 900 hPa level in the NCEP final analysis data (FNL) on a 06 UTC, b 12 UTC and c 18 UTC 30 March 2007

Full size image

In order to examine warm advection in the low-level atmosphere, Fig. 6 displays the 900-hPa air temperature and wind vector simulated by the MM5, WRF and COAMPS models. In relation to the time evolution of the warm area and low-level flows, all of the models simulate the abrupt enhancement on 12 UTC, for both location and magnitude, compared with the data shown in Fig. 5. As time passes, the simulation results show a northeastward extension of the warm advection region. The abrupt warm advection can induce destabilization over the southern part of the Yellow Sea, which extends to the western coast of Korea.

Fig. 6
figure 6

Horizontal wind vector and temperature (shaded area) at 900 hPa level simulated using MM5, WRF and COAMPS on 06 UTC, 12 UTC and 18 UTC 30 March 2007

Full size image

Among the three models, MM5 and WRF tend to underestimate the magnitude of the warm tongue in the East China Sea, while the WRF simulation is in better agreement with the observations than the other simulations. However, the wind speeds using MM5 and WRF are overestimated. The temperature gradient due to warm tongue advection is simulated in the COAMPS model reasonably well. On 18 UTC, the simulation results indicate enhanced unstable conditions, corresponding to the development of a mesoscale cyclone.

5 Storm surge

In order to examine the effect of different atmospheric forcings on the surge, including the SSW and SLP, the surge model was implemented using simulation results obtained using MM5, WRF and COAMPS. In order to focus on the response to atmospheric forcings, tides are not considered in the surge model as external forcings. Figure 7 shows the time series for the simulated surge using results obtained using MM5, WRF and COAMPS at Yeonggwang station during the period from 00 UTC 30 March to 00 UTC 31 March 2007. Maximum surge heights (MSHs) calculated using the atmospheric forcings of MM5, WRF and COAMPS were found to be 35, 29 and 20 cm, respectively. Although the wind forces and the magnitude of the storm surge in WRF are higher than those obtained using MM5, the MSH simulated using MM5 results is highest but the duration of the surge and the onset time, simulated using WRF results are relatively more accurate than the other models. The MSH simulated using the COAMPS results is small because a relatively weak southerly wind is simulated around the mesocyclone and the coastal region (see Fig. 4c). The difference for MSH and its onset time along the three models is the result of differences in southerly wind speed south of the mesocyclone among the three models which affects the propagation of long ocean waves. In addition, the surge model underestimates the MSHs, compared to observations. The observed maximum positive and negative surge elevations are approximately +100 and −100 cm, respectively, and the duration is about 20 min (see Fig. 1). In the case of the Yeonggwang storm surge, a long ocean wave generated by ocean effects including Proudman resonance (resonant excitation of sea waves by atmospheric waves), and the amplitude of the long ocean wave increases due to shoaling effects and reflections at the shore (Choi et al. 2008). The surge is the result of the combined effect of strong winds, the development of a mesocyclone and ocean effects. Further study will be required to develop a more complete understanding of the causes of the Yeonggwang storm surge.

Fig. 7
figure 7

Comparison of surge heights simulated using the atmospheric forcings of MM5 (dashed line), WRF (solid line) and COAMPS (dotted line) at the Yeonggwang station during the period from 00 UTC 30 March to 00 UTC 31 March 2007

Full size image

6 Discussion

The Yeonggwang storm surge that occurred on 31 March 2007 is an exceptional one that was generated by a meso-β scale cyclone with a centre pressure of 1010 hPa. In order to accurately predict a storm surge, accurate data on SLP and SSWs are required. In this study, the atmospheric conditions of the storm surge case were simulated using different models, namely MM5, WRF and COAMPS and physics options. We first selected the physics options as the optimal parameterizations for the high-wind conditions and then investigated the causes of cyclone development.

EXP1 (Eta PBL and Betts-Miller cumulus), EXP2 (Eta PBL and Grell cumulus) and EXP3 (MRF PBL and Kain-Fritsch cumulus) with MM5 and WRF were initially designed for the optimal combination of physics. According to the statistics for evaluating the horizontal distribution of the simulations, the performance of EXP2 was good, in terms of predicting the distribution and magnitude of wind speed.

It has been shown that different models can have a great impact on the magnitude and evolution of a cyclone. The results of the numerical simulations performed in this study showed that the models simulated the anticyclone and mesocyclone in the southern part of the western coast in causing the storm surge, WRF simulated the position and amplitude of the developed meso-β scale cyclone particularly well.

In order to investigate the causal mechanism for generating a storm surge, low-level warm advection was examined using the three simulations. As a result, a low-level positive temperature advection is the main factor in terms of enhancing the development of a mesoscale cyclone. The three models simulated the enhanced unstable conditions required for the development of a mesoscale cyclone relatively well. The three simulations showed the abrupt evolution of warm temperature advection in the time evolution. However, MM5 and WRF tended to underestimate the warm tongue and overestimate the wind speed in the East China Sea, while the WRF simulation is closer to the observed situation than the other simulations for temperature gradient and advection speed. The wind forces and the magnitude of the storm surge for WRF are higher than those for MM5 and COAMPS. The MSH simulated using the MM5 results was the highest but the duration of the surge and the onset time simulated using the WRF results are relatively more accurate than the other models. Therefore, it is important to accurately simulate the mesocyclone accompanied by a strong pressure gradient because the response of this situation to a storm surge is sensitive. The difference between the simulated and observed surge heights appears to be the result of ocean effects.