The National Centers for Environmental Prediction (NCEP), administered by the National Oceanic and Atmospheric Administration (NOOA), provides a climate forecast system analysis model (CFSR accessed at https://www.hycom.org/dataserver/ncep-cfsr), which comprehensively represents the dynamic interactions between atmosphere, oceans, and land surfaces from 1992 to now. The CFSR model generates an hourly time-step dataset with a spatial resolution of 1/2°. French-McCay et al. (2021) applied CFSR wind fields to simulate the oil trajectory and fate of the Deepwater Horizon from April to September 2010. Moreover, the wind dataset derived from the CFSR model was used to conduct 896 oil spill simulations to establish a risk assessment framework in the Perdido region near the Gulf of Mexico (Meza-Padilla et al., 2021). In addition, the fifth-generation ECMWF Reanalysis-ERA5 (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview) provides hourly wind-forcing reanalysis data presented in regular longitude and latitude grids at a spatial resolution of 1/4°×1/4 °. The eastward and northward velocity fields measured 10 m above the still water level were collected from 1940 to the present and updated daily. Zhang et al. (2020) tested the ERA5 wind forcing to model the drift trajectories in the South China Sea. The ERA5 meteorological fields were also applied to simulate oil spill movement and behavior in the busiest shipping routes through the Gulf of Suez, Egypt (Abdallah and Chantsev, 2022). Moreover, the CMEMS proposed the global ocean hourly reprocessed sea surface wind fields at 1/8° and 1/4° horizontal spatial resolutions from scatterometer satellite observations and numerical models. The CMEMS dataset was used to correct the persistent biases in the ERA5 dataset. The lower resolution of 1/4° for the air density, eastward and northward wind fields are provided from 08/1999 to 10/2009. On the other hand, a 1/8° spatial resolution of wind fields can be extracted from 01/2009 to 09/2022 (https://data.marine.copernicus.eu/product/WIND_GLO_PHY_L4_MY_012_006/download?dataset=cmems_obs-wind_glo_phy_my_l4_0.25deg_PT1H). A very short-term forecast model, a Korean Local Analysis Prediction System (KLAPS) referred to a numerical forecasting system performed to predict the weather on the Korean peninsula provide hourly regional and high-resolution meteorological fields at a spatial resolution of 1/20° (https://data.kma.go.kr/data/rmt/rmtList.do?code=320&pgmNo=66). Kim et al. (2023) deployed the KLAPS wind components in predicting the drift trajectories for maritime search and rescue purposes using the OpenDrift framework. The study indicated that the numerical results in OpenDrift with KLAPS meteorological input parameters show good agreement with satellite observation drifter results. Table 1 provides detailed information on meteorological models.

Numerous ocean circulation models provide hydrodynamic ocean datasets on a global scale, but only datasets that can be applied in an oil slick transport simulation in the Korean peninsula are discussed, as shown in Table 2. Therefore, the CMEMS and HYCOM models are presented. First, the Global Ocean Physics Reanalysis and Global Ocean Physics Analysis and Forecast products (https://data.marine.copernicus.eu/product/GLOBAL_MULTIYEAR_PHY_001_030/description), featuring regular horizontal resolutions of 1/12° obtained from CMEMS for hourly, daily, and monthly for sea surface temperature, sea surface height, currents, and sea ice parameters spanning from the ocean bottom floor to still wave level. Sepp Neves et al. (2020) used CMEMS products for current fields to establish a coastal oil spill hazard map. Similarly, the sea surface temperature and current data derived from CMEMS products were obtained to simulate the potential oil spills from tanker accidents near the Fernando de Noronha Archipelago by the MEDSLIK-II model (Siqueira et al., 2022). The second hydrodynamic ocean current model, Global Ocean Forecasting System (GOFS) in the HYCOM model (https://www.hycom.org/dataserver/gofs-3pt1/reanalysis), provides a global-scale product with a horizontal resolution of 1/12° and current predictions every three hours at 32 vertical depth layers from 0 to 400 m with a distance interval of 50 m. The HYCOM ocean current data were assimilated using a 24-hour numerical model forecast and several observation instruments such as altimeter observation, satellite, and in situ temperature and salinity profiles. Zacharias et al. (2021) conducted a probabilistic approach using the Spill, Transport, and Fate Model (STFM) to identify the multiple potential sources of oil spill sources, in which the input parameters of hydrodynamic currents, including current field, temperature, and depth data were extracted using the HYCOM model. Kim et al. (2023) also applied the sea surface velocities extracted from the HYCOM model to predict the drift trajectories along the Korean coasts.

Several wave models are available for direct user access. In particular, the CMEMS and ECMWF models provide global geographical coverage, whereas UOM, IFREMER, PdE, CYCOFOS, and HCMR models are used only for the Mediterranean regions. On the Korean peninsula, wave information is sourced from the Korean Operation Oceanographic System (KOOS) database, administered by the Korea Institute of Ocean Science and Technology (KIOST). On the other hand, the KOOS system does not facilitate online access for users to download the dataset. In contrast, the primary objective of this study is to provide a rapid assessment of oil spill transport, enabling expeditious decision-making for authorities responding to oil spill accidents. Consequently, the study emphasizes prioritizing wave models that are easily accessible. Within the context of seven previous case studies involving oil spill transport simulations using the OpenOil model, only ECMWF and CMEMS were examined. Therefore, the reliance is on global wave models, with CMEMS being the sole viable option because the ECMWF database access is currently unavailable because of ongoing system maintenance. The CMEMS model provides a global wave reanalysis product called WAVERYS (https://data.marine.copernicus.eu/product/GLOBAL_MULTIYEAR_WAV_001_032) to describe the wave characteristics covering the period between 1993 and 2019. This product is based on the third-generation wave model of the Meteo-France Wave Model (MFWM) that calculates the wave spectrum on 1/5° of an irregular horizontal grid. The wave parameters obtained from this wave spectrum, including Stoke drift velocities, significant wave height, and average wave period, were proposed on a rectangular 1/5°–grid with a three-hourly integrated time step. This product was assimulated using the altimeter wave data and Sentinel-1-provided directional wave spectra. Law-Chune et al. (2021) comprehensively described the WAVERYS product in the CMEMS models.

OpenOil is a state-of-the-art model for simulating oil-spill transport and fate, which was recently developed by the Norwegian Meteorological Institute. The OpenOil model was based on the Lagrangian trajectory model, which simulates spilled oil movement as many particles moving under the interaction from ocean currents, winds, and waves. Nguyen et al. (2023) reported a detailed description of governing equations on the OpenOil model. The OpenOil model employs hydrodynamic ocean circulation, wave, and meteorological models as input parameters in the oil simulation. The OpenOil model has been applied in several case studies across different regions globally. In detail, Röhrs et al. (2018) examined the vertical and horizontal transport mechanisms of marine oil spills in the Norwegian Sea using the NorShelf model for ocean current fields and ECMWF models for both wind and wave datasets. Moreover, the OpenOil model coupling with GoM-HYCOM and ECMWF models for met-ocean forcing data was used to simulate the impact of river fronts on the oil slick transport in the 2010 DeepWater Horizon accident (Hole et al., 2019). Following this study, the pathways of potential oil spill scenarios were simulated to improve the awareness of planning and preparedness technologies for various offshore sites in Cuba (Androulidakis et al., 2020). Keramea et al. (2022) conducted an operational oil spill model to examine the oil dispersion characteristics using met-ocean datasets from NOAA-GFS and CMEMS for the North Aegean region. On the other hand, studies investigating the predictive performance of the OpenOil model coupling with several met-ocean forecast models for oil slick transport around the Korean peninsula are limited. Therefore, six combinations considering several input met-ocean forecast models used in the OpenOil model were tested to simulate spilled oil in Korea (Table 3).

Fig. 1 presents the flow chart of the proposed methodology. HYCOM and CMEMS were used as hydrodynamic models to provide ocean current parameters. The wind data are received from CMEMS, CFSR, ERA5, and KLAPS models, while CMEMS models are also used to extract the wave parameters. Six combinations were suggested to evaluate the most effective input forcing parameters for simulating spilled oil transport on the ocean surface. The characteristics of oil types were identified using the NOAA oil library. Finally, the met-ocean forcings and oil-type information were used to drive the OpenOil model. The performance of each combination is rigorously assessed by comparing the results with synthetic aperture radar (SAR) observation data at a specific time.

On the morning of December 7, 2007, at 7:15 (local time), the very large oil crude oil carrier (VLLC) Hebei Spirit collided with Samsung cranes near the latitude and longitude of 36° 49.93’ N, 126° 2.46’ E, located approximately 10 km off the Taean coasts, leaking approximately 10,900 t of crude oil. Fig. 2 presents the site location for the spilled oil slick. The Korean Coast Guard (KCG) report indicates that strong northwestern currents and winds blow oil slick transport along approximately 375 km of the west coast of Korean coastlines, heavily contaminating 70 km of the Taean peninsula with crude oil. Remarkably, the oil slick quickly reached the Taean shoreline at 13 h, which was significantly shorter than the predicted time of approximately 24 h, as suggested by the Ministry of Oceans and Fisheries (MOF) (Lee et al., 2020). In particular, oil slicks polluted large areas of the open sea and contaminated Jeju Island, the southern end of the Korean Peninsula (Kim et al., 2014).

The oil slick transport resulting from the Hebei Spirit accident was simulated for five days from 2007/12/07 at 07:15 to 2007/12/07 at 11:00 am. The oil type used in the OpenOil model was Kuwait export crude, with a density of 820.5 kg/m³ as the representative oil type. Approximately 10,900 oil particles were initially set up. The Openoil model was coupled with the real-time current circulation data from the HYCOM and CMEMS models, winds from CFSR, ERA5, CMEMS, and KLAPS, and waves from CMEMS. Table 4 lists the initial input parameters for the oil slick simulation in the Hebei Spirit accident.

The study area on the eastern Yellow Sea is characterized by relatively shallow waters with a water depth of less than 60 m. On the other hand, the topography near the Taean coast is approximately 25 m deep (Fig. 2). Fig. 3 shows the wind velocity fields obtained from CMEMS, CFSR, ERA5, and KLAPS models on December 7, 2007 at 7:15 am. The color map indicates the wind velocity magnitude measured 10 m above the sea water level, and the blue arrows represent the wind direction. The wind fields in the Korean Peninsula follow the characteristics of the seasonal wind areas of Asia. The spring and summer seasons witnessed south/southwestern winds, but the main north/northwestern winds primarily blow in the autumn and winter seasons. Therefore, in the Hebei Spirit accident, the north and northwestern wind fields were recorded with their speed reaching a maximum of approximately 14 m/s (Fig. 3). Similarly, for the wind dataset derived from four meteorological models, while approximately less than 6 m/s of wind velocity speed was measured generally close to the Tean coastlines, the wind velocity exceeds 10 m/s in the open sea.

Fig. 4 presents the current circulation parameters, including the current field velocity, sea surface salinity, and sea surface temperature extracted from the HYCOM and CMEMS models. The current velocity ranged from 0.1 m/s to 0.6 m/s in the northwest direction. A slight difference in current velocity was measured near the Taean peninsula. On the other hand, a significant difference is observed in the open sea because HYCOM measured the current parameters every three hours, while the daily dataset was recorded in the CMEMS model. Moreover, the sea surface temperature and salinity magnitude were up to 12°C and 32 psu, respectively.

Keramea et al. (2023) comprehensively reviewed the operation and forcing in oil spill models. The literature review showed that only seven case studies of oil spill simulations were conducted while considering adequate met-ocean data as input parameters. Most studies focused on forecasting hydrodynamic ocean current and meteorological models as met-ocean input parameters while neglecting the wave fields model, represented by the significant wave height, wave period, and Stoke drift velocities, in oil spill models. This limitation may influence the oil slick transport on the ocean surface. Fig. 5 shows the Stoke drift velocities, significant wave height, and period received from the CMEMS model. The maximum significant wave height was approximately 3 m with a wave period of 6.5 s when the accident occurred. Moreover, the Stoke drift velocities ranged from 0.03 m/s to 0.2 m/s in the northwest direction.

Fig. 6 shows the distribution of oil slick on the sea surface after 99.2 h from the accident. The movement of spilled oil showed good agreement with the current and wind directions presented in Figs. 3 and 4. There are two clear trends for oil diffusion using CMEMS (combinations 1 and 2) and HYCOM models (combinations 3, 4, 5, and 6) as hydrodynamic ocean circulation models in the oil spill model. In combinations 1 and 2, a slight difference is observed in the spilled oil distribution, in which the oil slick moved approximately 92 km from the Taean coast to the Maryang-ri port. By contrast, the combinations from 3 to 6 showed that the distribution of spilled oil slicks ranged from approximately [125.6°–126.1°E] for longitude and [36.4°–36.96°N] for latitude. Table 5 lists the specific dimensions in terms of width and length, as well as the projected area of oil slicks in each combination.

The numerical results are validated by comparing the simulated distribution of oil slicks with the observed results obtained from Envisat ASAR at 10:40 on December 11, 2007, for the Hebei Spirit accident. The observation satellite image was processed using the adaptive threshold method (Fig. 7). The significant oil slicks extended longitudinally from 125.6°E to 126.3°E and latitudinally from 36.43°N to 36.99°N, encompassing maximum widths and lengths of 62.27 km and 62.63 km, respectively, and spanning an approximate area of 1600 km². Generally, configurations incorporating the HYCOM model as a hydrodynamic input showed reasonable agreement with the observed oil slick distribution in both spatial dimensions and projected area, exhibiting a deviation of less than 20% compared to configurations using CMEMS models, which displayed a maximum difference of 50%.

Kim et al. (2014) proposed a thorough validation approach for oil slick movement, enabling users to evaluate the performance of each combination (Fig. 8). First, the study area was divided into grid cells, each measuring 0.5 km × 0.5 km. The second step was to identify each cell containing oil particles in both simulated and observed oil slicks, with blue for the numerical result and pink for the observation result. Subsequently, the overlapped cells between simulated and observed results are extracted. To calculate the percentage of similarity between numerical and SAR results, a matching percent is introduced, which is determined by the fraction between the total number of overlapped cells and the total number of observation cells containing oil particles. Fig. 9 compares the matching percent in each combination with observed results. In the presence of the CMEMS hydrodynamic models, the match percentage was approximately 53%, while the HYCOM hydrodynamic models induced matching rate values from 70% to 88%, indicating a much higher accuracy than the CMEMS combinations. Therefore, the HYCOM model is more effective than the CMEMS model in contributing hydrodynamic ocean circulation parameters for oil spill models. Moreover, significant improvements in matching percentages were observed when a high-resolution meteorological dataset was introduced for HYCOM combinations. For CFSR (spatial resolution ∼56 km), the matching percentage was only 70%, which increased to approximately 86% for the ERA5 and CMEMS meteorological models (~27 km) and 88% for the KLAPS model (~5 km). Therefore, increasing the resolution of forecasting met-ocean models is necessary to enhance the performance of numerical models to predict the spilled oil diffusion on the sea surface. Furthermore, the notable precision exhibited by the HYCOM model in simulating the transport of oil slicks underscores the potential for widespread adoption of HYCOM met-ocean datasets coupling with different atmospheric models across the Korean peninsula.

The spatial and temporal resolution of met-ocean models significantly influences the accuracy of numerical results in oil spill simulations. Most hydrodynamic, meteorological, and wave models are extracted from open-source global-scale models with coarse resolutions. This highlights the need for further investigation on implementing regional models for oil spill models. The KOOS was established at the KIOST to provide real-time observations and simulated datasets of ocean fields in multiple scales. The KOOS dataset provides waves, currents, sea surface salinity, water temperature, and wind fields, updated twice daily for 72 hours (Park et al., 2015). The system collects observation datasets from real-time marine monitoring platforms. This information was used to calibrate the numerical models. Park et al. (2015) also implemented a 300-m horizontal resolution of atmospheric variables (wind components and atmospheric temperature), current and wave variables (current velocity, water temperature, salinity, and elevation) into the oil spill model, MOHID to simulate the oil spill transport at a quay off Yeosu port. The numerical results effectively replicated the spilled oil transport and satisfied the general acceptable criteria. The KOOS data have been under development since August 2009 by KIOST, and the dataset has not been applied to simulate the Hebei Spirit accident. Nevertheless, the current study suggests that KOOS data can be used to simulate the oil slick transport released from recent accidents and in further research aimed at improving spill modeling.