1. Introduction
Incheon International Airport (IIA) is the largest airport in South Korea and ranked first in Asia in terms of international passenger traffic in 2020. The National Aviation Safety Data Analysis Centre (NASDAC) Review of the National Transportation Safety Board (NTSB) showed that weather-related accidents and conditions that considerably affected near-surface aviation operations between 1994 and 2003 included wind and visibility (Gultepe et al. 2019). As per the criteria of the Korea Aviation Meteorological Agency, a strong wind warning is issued when the average wind speed is 25 kt (1 kt ≈ 0.51 m s−1) over a 10-min period and an aerodrome weather warning is issued when the maximum instantaneous wind speed exceeds 35 kt. Additionally, a low-visibility warning is issued when the visibility drops to ≤400 m. This is particularly applicable to IIA, located in Yeongjong Island, where sea fog could considerably affect visibility. Generally, a fog event is defined as horizontal visibility impairment of <1000 m (Klemm and Wrzesinsky 2007; Kim and Yum 2010; Ahrens 2009; Feigenwinter et al. 2020). Therefore, accurate simulation of near-surface moisture and temperature is a key factor in aviation operations. Between 1 July 2001 and 30 June 2005, 10 549 flight delays occurred at IIA, of which 1350 delays (∼13%) were due to poor weather conditions. Among weather-related delays (1350 delays), 56 delays (∼4%) were due to strong winds and 830 delays (∼62%) were due to fog events (Leem et al. 2005).
The forecast capabilities of the numerical weather prediction model are considerably affected by initial conditions/boundary conditions (ICs/BCs) as well as physics/dynamics enhancement of models (Gallus and Bresch 2006). Therefore, assessing the model predictability under varying ICs/BCs is a crucial step for improving weather forecasting accuracy. Gevorgyan (2018) showed that the initial and lateral boundary conditions affect the simulated spatial distribution and peak time of heavy rainfall events in regional model simulations. Kumar et al. (2017) used four global model analyses [i.e., European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP), Global Data Assimilation System (GDAS), NCEP Global Forecast System (GFS), and National Centre for Medium Range Weather Forecasting (NCMRWF)] as the initial conditions for a Weather Research and Forecasting (WRF) Model over a period of 1 month during winter. They found that the model initiated by the ECMWF analysis data simulated the most comparable Atmospheric Infrared Sounder (AIRS)-retrieved vertical profiles of temperature, specific humidity, and Tropical Rainfall Measuring Mission (TRMM) 3B42 product measured rainfall fields.
Recently, high-resolution large-eddy simulation (LES), which resolves the atmospheric mixing process by large turbulence eddies in the atmospheric boundary layer while parameterizing only small-scale turbulence in the inertial subrange, has been utilized for weather prediction around airports (Prasanna et al. 2018; Cui et al. 2019; Chen et al. 2022). Chen et al. (2022) conducted a case study that evaluated the simulated wind shear around Hong Kong International Airport using a high-resolution LES model based on forecast data from the WRF Model output and found that the WRF-LES system served as a useful tool for real-time forecasting around the airport region. Cui et al. (2019) simulated a radiation fog event and found that WRF-LES with 0.33-km grid spacing resulted in a smaller mean bias in surface meteorological elements such as 2-m temperature, 2-m relative humidity (RH), and 10-m wind speed compared to the WRF simulation with 1-km grid spacing. Moreover, WRF-LES exhibited improved reproducibility in terms of visibility fluctuations. Prasanna et al. (2018) designed a regional model using a Python-based rose nesting tool developed by the Met Office. This model utilized a 300-m grid spacing centered on IIA and was downscaled with the unified model (UM), a numerical weather prediction model developed by the Met Office (UKMO), to produce spatiotemporally high-resolution forecasts that reproduced the observed wind gust and provided better prediction of wind compared to the coarser-resolution model.
Recognizing the importance of ICs/BCs and precise wind field and fog forecasts in ensuring aircraft safety and improving model forecast ability, the current study aims to investigate the effects of different ICs/BCs on low-level forecast fields including temperature, water vapor, and wind by designing a high-resolution regional weather forecast model using a WRF that covered the area surrounding IIA. Additionally, the study also aimed to provide guidance on proper horizontal resolution for forecasting low-level meteorological fields around IIA by comparing the mesoscale model forecast outputs with those from LES or global model forecasts. Finally, the influence of postprocessing on LES experiment results is also assessed.
The contents of this paper are organized into sections, with section 2 including a description of the experimental setting; section 3 including simulation results and validation with observations; and section 4 including the study summary and conclusions.
2. Model and experimental design
The current study uses the WRF regional model, version 4.2.2 (Skamarock et al. 2019). Three nested domains centered on IIA (Fig. 1a) and having 10-km, 1-km, and 100-m horizontal resolutions are constructed for the simulations. A one-way interaction from coarser-resolution domains to finer-resolution domains is applied in the integrations. The top layer of the model is spaced at 50 hPa, and every domain has 75 vertical layers. The grid spacing of the first layer is set as 0.006 in the eta level for all domains. For physics parameterizations, the unified Noah land surface model (Chen and Dudhia 2001); the revised MM5 Monin–Obukhov (Jiménez et al. 2012); the Rapid Radiative Transfer Model for general circulation models (RRTMGs) short-wave and long-wave scheme (Iacono et al. 2008; Morcrette et al. 2008); the Shin–Hong scale-aware planetary boundary layer (PBL) scheme (Shin and Hong 2015); and the WRF double-moment 6-class (WDM6) microphysics scheme (Lim and Hong 2010) are used. For cumulus parameterization, the Kain–Fritsch scheme (Kain 2004; Kain and Fritsch 1990) is used for the 10-km resolution domain only. Table 1 shows a summary of the model domain configurations used and the physical parameterizations applied. Additionally, subgrid-scale (SGS) orography parameterization (Jiménez and Dudhia 2012) is applied to the 10- and 1-km resolution domains using the Shin–Hong PBL scheme. For domain 3 (d03 in Fig. 1b) with a 100-m horizontal resolution, three-dimensional turbulent mixing is simulated by the WRF-LES without PBL parameterization. Soil hydraulic parameters and vegetation reference tables are adjusted using the findings reported by Campbell et al. (2019).
Model domain with terrain height (m) for (a) the entire domain and (b) the innermost domain (d03), along with locations of observation stations. The circle, square, and diamond markers in (b) indicate AWS, ASOS, and AMOS, respectively. (c) The averaging method for LES-GFS (10 × 10) data; the black lines and orange crosses (A–D) represent the location of model grids and simulated fields for 1-km simulations; the gray lines represent the location of model grids for 100-m simulations. The average values around the 10 × 10 grids within the red dashed boxes are shown as green crosses.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
Summary of the WRF Model configuration.
For ICs/BCs, the 6-h-interval forecast fields of the global models [i.e., NCEP Global Forecast System (GFS) model (Global Climate and Weather Modeling Branch 2003; National Centers for Environmental Prediction 2015) and the Korean Integrated Model (KIM) of the Korea Meteorological Administration (KMA) (Hong et al. 2018)] are utilized. The GFS and KIM forecast fields are selected as ICs/BCs for this study since they are used as ICs/BCs for regional models by the Republic of Korea Air Force and KMA, respectively. The NCEP-GFS model forecast outputs are provided four times per day at 0000, 0600, 1200, and 1800 UTC at a horizontal resolution of ∼28 km (0.25°) comprising 127 vertical layers. The model provides 3-h-interval forecasts from 0 to 240 h and 12-h-interval forecasts from 240 to 384 h. The KIM, which has been used as an operational forecast model in KMA since April 2020, affords a horizontal resolution of ∼13 km (0.125°) comprising 91 vertical layers. The model is initialized four times per day (0000, 0600, 1200, and 1800 UTC), similar to the NCEP-GFS model, and provides 3-h-interval forecasts from 0 to 84 h and additional 6-h-interval forecasts from 84 to 288 h at 0000 and 1200 UTC. When the model is initialized at 0600 and 1800 UTC, it only offers 3-h-interval forecasts from 0 to 84 h. The National Oceanic and Atmospheric Administration Optimum Interpolation SST climatology data (Huang et al. 2021) interpolated into the daily dataset are used for sea surface temperature (SST), and the National Aeronautics and Space Administration Shuttle Radar Topography Mission 3-s (90 m) topography data and the Environmental Geographic Information Service 3-s land-use data by the Ministry of Environment, the environmental authority of the Republic of Korea, are applied for high-resolution terrain representation.
Both simulations and analyses are performed for 24 h starting at 0000 UTC [0900 Korean standard time (KST)] on each day of April 2021. All domains are initialized at the same time. The month of April is selected for numerical experiment as it has exhibited the highest average monthly wind speed in the Korean Peninsula over the past 10 years (2011–20), and the year 2021 is selected based on data availability (as the KIM started operations in April 2020). The mesoscale WRF experiments, run under the 1-km horizontal resolution utilizing KIM and GFS forecast fields as ICs/BCs for the integration of the WRF Model, are denoted as WRF-KIM and WRF-GFS, respectively. Additionally, the LES experiment, run under the 100-m horizontal resolution with GFS forecast fields as ICs/BCs, is named as LES-GFS. The temporal resolutions for each WRF domain are 4 and 0.4 s for the 1-km and 100-m grid-spacing domains, respectively. The model output was recorded at 10-min intervals and utilized to calculate the 1-month average. To our knowledge, there is a distinct lack of studies validating such high-resolution simulations over long periods near IIA.
LES model often exhibits high spatial variability when simulating low-level meteorological fields. This variability can result in different validation outcomes when comparing point observation values to model simulation results, depending on the utilized data processing methods. To propose a suitable verification method for high-resolution simulations that resolve spatiotemporal variability of turbulence, we designed the LES-GFS (10 × 10) data, which are postprocessed outputs of the LES-GFS experiment. To generate LES-GFS (10 × 10) data, meteorological values from the LES-GFS experiment at 10 × 10 grid points in subdomains (size: 1 km × 1 km; red dashed boxes in Fig. 1c) centered around the orange points (A, B, C, and D in Fig. 1c, equivalent to 1-km grid spacings) are averaged. The average values, indicated by the green crosses in Fig. 1c, are then interpolated to observation station points and compared with the observations. This comparison method is devised because LESs often exhibit high spatial variability in simulated low-level meteorological fields, which can lead to discrepancies when compared to point observations at coarse temporal and/or spatial resolutions.
The 6-hourly fifth major global reanalysis produced by ECMWF (ERA5) (Hersbach et al. 2020) 0.25° data are used for verification of synoptic fields simulated by the KIM and GFS, while the low-level meteorological fields including wind, temperature, and water vapor are verified using measurement data from the Automated Weather Station (AWS), Automated Surface Observing System (ASOS), and Aerodrome Meteorological Observation System (AMOS). Figure 1b shows the AWS (circles), ASOS (square), and AMOS (diamond) stations [i.e., IIA (L1), Incheon (L2), Jangbongdo (O1), Wangsan (O2), and Muuido (O3)]. The “L” corresponds to stations where the four grids adjacent to each observation point are classified as land in the WRF Model, while the “O” corresponds to stations where some of the four grids are classified as ocean.
3. Results
a. Effects of using the different global forecast fields as ICs/BCs
The monthly average synoptic charts of the 6-, 12-, 18-, and 24-h forecast fields from the global models (the KIM and GFS), which are used as ICs/BCs in the WRF regional model, are shown in Fig. 2. Both global models, which depict high mean sea level pressure (MSLP) over the Yellow Sea and East Sea (Figs. 2a,b), simulate lower MSLP over the majority of the domain when compared to the ERA5 reanalysis data. The KIM yields a larger negative bias than GFS over the Yellow Sea, Korean Peninsula, and East Sea (Figs. 2c,d). Both models simulate westerly winds from the Yellow Sea to the East Sea passing through the Korean Peninsula in April. Moreover, no significant differences in wind direction and speed over the Incheon and Seoul metropolitan areas are observed (Figs. 2e,f). The KIM shows higher humidity overall, except in the southern part of the domain, when compared to the ERA5 reanalysis data. The wet bias shown in the KIM forecast is relieved in GFS, particularly over the Korean Peninsula and the northern part of the domain (Figs. 2g,h). Therefore, forecast fields from the KIM exhibit lower MSLP and higher moisture in the low-level atmosphere over the Korean Peninsula, including Yeongjong Island where IIA is located, when compared to ERA5 reanalysis data and GFS forecast fields.
Spatial distributions of 1-month average MSLP (hPa) of the (a) KIM and (b) GFS and (c),(d) bias of KIM and GFS compared to ERA5, respectively. (e)–(h) As in (a)–(d), but for results about 850-hPa RH (%) and wind vector (m s−1). The red cross and blue circle in (a) indicate Yeongjong Island and Seoul metropolitan areas, respectively.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
Figure 3 compares the time series of the 30 simulations’ set-average 2-m temperature (T2m) from the global models (the KIM and GFS), regional models (the WRF-KIM and WRF-GFS), and observation data (left column) and root-mean-square error (RMSE) scores (right column) from each observation station over 24 forecast hours during 1 month. The regional model results are from domain 2 with 1-km grid spacing (d02 in Fig. 1a). The global model forecasts show lower T2m values at land (L1 and L2) stations, while the regional model simulations show higher T2m values and exhibit magnitude of diurnal cycle similar to the observations. The magnitude of diurnal variation in T2m is captured better by the regional models compared to the global model forecasts at all L stations, and this superiority is confirmed by a reduction of more than 1°C in the 1-month average RMSE of T2m in the regional model, as well as by the correlation coefficient r of the T2m time series between the observation and model simulations (Tables 2 and 3).
Time series of observed and simulated 1-month average of (a),(c),(e),(g),(i) T2m and (b),(d),(f),(h),(j) RMSE of T2m over 24 h in April 2021 at each observation station. (a),(b) L1; (c),(d) L2; (e),(f) O1; (g),(h) O2; and (i),(j) O3, as shown in Fig. 1b. The gray, red, and blue colors represent observations, the KIM (the WRF-KIM and KIM), and GFS (WRF-GFS and GFS), respectively. The solid and dashed lines with cross marks represent the WRF (WRF-KIM and WRF-GFS) and global model (KIM and GFS) forecast results, respectively.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
One-month average RMSE of T2m, Q2m, and WS10 over 24 h for the KIM, GFS, WRF-KIM, and WRF-GFS against observations at each station. Boldface indicates the best scoring experiment.
The mean value of observation, global model forecasts, and regional simulations for T2m, Q2m, and WS10 together with the correlation coefficient r between their time series of observations and the KIM, GFS, WRF-KIM, and WRF-GFS simulations at each station. These values are obtained from a set of 30 simulations over a 24-h period in April 2021. Boldface indicates the best scoring experiment.
At the ocean stations (O1, O2, and O3; Figs. 3e–j), the global models forecast considerably lower T2m values during the daytime and similar T2m values during the night compared to the observed T2m. Both the regional models simulate T2m values that are lower than the observations across all forecast periods at the O stations, and these simulation features can be confirmed using the statistical scores. The RMSE values of T2m are quite similar for the global and regional simulations (Table 2), although the r scores increase in regional models at all O stations (Table 3). When comparing the simulated T2m from the KIM and GFS at the O stations, the KIM is observed to simulate about 1°C higher T2m values during the daytime and 1°C lower T2m values during the night compared to the GFS forecasts, thus presenting better diurnal cycle amplitudes for T2m. However, the GFS and KIM exhibit a deficiency of excessively cooling the near surface during the night. In the regional simulation, T2m commonly shows lower RMSE at the L stations compared to the O stations, and the RMSEs of WRF-GFS are generally lower than those of the WRF-KIM. Overall, the WRF-GFS presents better T2m simulation and diurnal variation features compared to the WRF-KIM (Tables 2 and 3).
Figure 4 shows the time series of the 1-month-average 2-m water vapor mixing ratio (Q2m) from 30-set model simulations, observation data (left column), and their RMSE scores (right column). The regional models simulate drier environments than the global models, particularly over the L stations. WRF-KIM at L2 shows the largest difference of 0.56 g kg−1 compared to the global model simulation used as input data. This difference arises from the more detailed land–sea representation in the regional models, where certain areas, classified as land in the regional models, are categorized as ocean in the global models. Consequently, although the regional models accurately represent the land–sea categories, they exhibit a dry bias compared to the observations, as demonstrated in Table 3. The regional model simulations do not exhibit any superiority in terms of Q2m simulation when compared to the global model simulations, regardless of land or ocean stations (Table 2 and Figs. 4b,d,f,h,j). The magnitude of diurnal variations considerably increases in both regional model simulations compared to the global model simulations, resulting in better r scores over the L1, L2, O2, and O3 stations (Table 3). When comparing two regional model simulations, no one model is better than the other in simulating Q2m.
As in Fig. 3, but for Q2m.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
Land surface models are utilized to depict precise surface characteristics and exchange fluxes, ensuring a realistic lower boundary condition for numerical weather prediction models (Di et al. 2023). Although both KIM and GFS employ the same land surface model, namely, the Noah land surface model, modifications have been made during the model’s version updates (Global Climate and Weather Modeling Branch 2003; Hong et al. 2018). This could potentially impact the simulated lower-level meteorological environment. KIM’s simulation of a wetter environment than GFS affects the results of the regional model, resulting in a wetter environment in WRF-KIM compared to WRF-GFS (see Fig. 4). Enhanced diurnal cycles of T2m in WRF-KIM compared to WRF-GFS also reflect the results of global simulations, especially at L stations.
Figure 5 presents the results for 10-m wind speed (WS10). The global models simulate relatively lower and higher wind speeds over the L and O stations, respectively, compared to the observations. They also present very weak diurnal variations over all stations with relatively strong wind speeds during the night (Fig. 5). Both regional models simulate higher WS10 than the global models during the daytime, thus increasing the amplitude of diurnal variation in all stations. The overestimation of WS10 in regional models at the O stations is consistent with previous studies (Avolio et al. 2017; Solbakken et al. 2021). Jiménez and Dudhia (2012) stated that the overestimation of wind in the WRF Model could be attributed to the difficulties in accurately simulating subgrid surface roughness and the resulting induced turbulence at lower atmospheric levels. The regional models present better r scores than the global models for the majority of observation stations except O1. Particularly noteworthy is the high r score of 0.93 at L1 for both WRF-KIM and WRF-GFS (Table 3). However, they show deteriorated RMSE of wind speed compared to the global models (Table 2). The RMSE of WS10 is the smallest in GFS (except for L2), and the mean WS10 values from GFS are closest to the observations at all stations except O1 (Tables 2 and 3).
As in Fig. 3, but for WS10.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
Overall, the regional models show a merit in terms of capturing the time series of T2m, Q2m, and WS10 compared to the global model forecasts. The r scores of the time series for T2m, Q2m, and WS10 (except for Q2m and WS10 at O1 station) are improved in the regional models compared to the global models (Table 3). This result is consistent with the study of Falasca and Curci (2018) who showed that underestimation of T2m and overestimation of WS10 in the European-scale 36-km horizontal resolution model simulations were mitigated in finer model simulations with 4- and 1.33-km resolutions, respectively, resulting in improved ability of the high-resolution regional models to predict the magnitude of diurnal variations in these parameters. Furthermore, the normalized time series, obtained by subtracting the mean values of the meteorological variables from the 1-month averaging time series at each observation station, reveals that WRF-GFS simulates Q2m and WS10 time series better than WRF-KIM, except at the O2 and L2 stations, respectively (not shown).
b. Effects of LES modeling
The impact of increased horizontal resolution of the model on simulated low-level meteorological fields is examined using the LES model configuration (LES-GFS). Figure 6 shows the time series of the 1-month-average T2m with its RMSEs at each observation station over 24 forecast hours. The WRF-GFS and LES-GFS, which used the GFS forecast fields as ICs/BCs, are compared for this purpose. Only GFS is used as ICs/BCs for the LES simulations as GFS performed better than the KIM in terms of capturing synoptic fields (Fig. 2). Additionally, the LES-GFS (10 × 10) data are analyzed to investigate the impact of averaging methods for verification. The LES-GFS results and the LES-GFS (10 × 10) data are from the innermost domain (d03 in Fig. 1b) with a 100-m grid spacing. Figures 7 and 8 show the results for Q2m and WS10, respectively.
As in Fig. 3. The blue and light blue solid lines indicate WRF-GFS and LES-GFS simulation results, respectively; the green dashed line indicates LES-GFS (10 × 10) data.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
As in Fig. 6, but for Q2m.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
As in Fig. 6, but for WS10.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
In general, both the WRF-GFS and LES-GFS models simulate lower T2m compared to the observation, with the latter simulating even lower temperatures than the former during the night, except for O3. This results in the simulated T2m deviating further from the observation during nighttime at the L stations (Figs. 6a–d). The RMSEs of LES-GFS increase with the forecast times, resulting in deterioration of the average RMSE (Table 4). However, in terms of the diurnal cycle, WRF-GFS underestimates the diurnal cycle amplitude of T2m at L stations, and more nighttime cooling of T2m in LES-GFS results in the diurnal cycle amplitude increasing (Figs. 6a,c). At O stations, LES-GFS simulates higher T2m than WRF-GFS except early in the morning, and this is more consistent with the observation (Figs. 6e–j) and exhibits better RMSE scores. Especially, LES-GFS at O1 shows a reduction of 0.45°C in RMSE compared to WRF-GFS (Table 4). LES-GFS (10 × 10) data show lower and higher T2m in the daytime and nighttime at O1 and O2 than LES-GFS, respectively. These results mean that small-scale variability, resolved in LES-GFS, is smoothed out in LES-GFS (10 × 10) data. The r scores between the observed and simulated T2m time series are almost 1 in both experiments, with neither model exhibiting superiority over the other (Table 5).
Like T2m, Q2m is underestimated by both WRF-GFS and LES-GFS (Fig. 7), with the latter simulating more moisture with increased mean values (from 4.85 to 5.02 g kg−1 at L1 and from 4.67 to 4.82 g kg−1 at L2) and better RMSE scores at L stations compared to the former (Tables 4 and 5). LES-GFS simulates lower Q2m than WRF-GFS and LES-GFS (10 × 10) over O stations during the night, resulting in the deterioration of RMSEs. WRF-GFS simulates a Q2m time series that is more comparable with the observation than LES-GFS, except at the O2 station (Table 5). As previously shown by the verification results of T2m, LES-GFS (10 × 10) data exhibit a similar Q2m simulation feature to that of WRF-GFS at the O1 and O2 stations.
Figure 8 shows the results for WS10. LES-GFS simulates weaker WS10 than WRF-GFS, especially during the daytime (0000–0900 forecast hours; 0900–1800 KST) when the wind speed is strongest. Consequently, it also exhibits weaker diurnal variation and lower r scores than WRF-GFS for all stations (Fig. 8 and Table 5). Even though the time series of 30-set averaged LES-GFS results shows similar WS10 values to the observations, with differences from observations of 0.02 and 0.16 m s−1 at L2 and O2, respectively, compared to WRF-GFS (Figs. 8c,g), the RMSE is similar or even greater than that of WRF-GFS (Figs. 8d,h). These findings can be attributed to overestimation or underestimation of WS10 by LES-GFS for certain simulation dates (i.e., overestimation at L2: 13 and 29 April; underestimation at L2: 9 and 25 April; overestimation at O2: 9 and 25 April; and underestimation at O2: 6, 7, and 15 April). This variability results in the simulated WS10 appearing closer to the observed value on average despite increasing daily RMSE. The LES-GFS (10 × 10) data show higher WS10 than LES-GFS at the O2 and O3 stations, and these values are similar to those of WRF-GFS in the nighttime.
The LES-GFS (10 × 10) data exhibit similar T2m and Q2m time series as WRF-GFS at the O1 and O2 stations, while the WS10 trend from LES-GFS (10 × 10) data is similar to that of WRF-GFS at the O2 and O3 stations. Figure 9 presents the horizontal distributions of low-level atmospheric fields (e.g., T2m, Q2m, and WS10) shown by WRF-GFS and LES-GFS at 1800 UTC 1 April 2021 (0300 KST 2 April 2021). The T2m, Q2m, and WS10 time series shown by LES-GFS (10 × 10) data are similar to those of WRF-GFS, especially at the O stations where the horizontal variability is greater than that of L stations. Meanwhile, the time series for the thermal and dynamic fields over land shown by the LES-GFS (10 × 10) data closely resembles that shown by LES-GFS. This suggests that caution should be exercised when comparing LES results with observation values at specific stations as the model verification methods involve interpolation and averaging, which can lead to differences in outcomes. For example, the temporal and/or spatial scales of traditional observations that have been used to verify mesoscale model outputs (e.g., surface observations at 10-min or 1-h intervals) may be insufficient to capture the variability of microscale fields that are resolved in LES.
Spatial distributions of (a),(b) T2m (°C), (c),(d) Q2m (g kg−1), and (e),(f) WS10 (m s−1) at 1800 UTC 1 Apr 2021 (0300 KST 2 Apr 2021). (left) WRF-GFS and (right) LES-GFS.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
The probability density functions (PDFs) of T2m, Q2m, and WS10 over all L and O stations are compared between the simulations and observations in Fig. 10. For T2m, while the frequency of both WRF-GFS and LES-GFS near 17°C is lower than the observations at the L stations, both accurately reproduce the observed PDF of T2m (Fig. 10a). However, at the O stations, the regional WRF Model (WRF-GFS) shifts the PDF toward colder temperatures. This deficiency in WRF-GFS is rectified by LES-GFS, which simulates more occurrences of higher temperatures and lowers the peak of the PDF, thereby presenting similar features to the observations (Fig. 10b). WRF-GFS tends to simulate a drier environment than observed at the L stations. This dry bias is mitigated in LES-GFS, as evidenced by an increase in frequency around 4.5 g kg−1 rather than 3.0 g kg−1 (Fig. 10c). As for Q2m at the O stations, both WRF-GFS and LES-GFS depict a similarly drier environment compared to observations (Fig. 10d). For WS10, at the L stations, WRF-GFS simulates a higher frequency of weak winds compared to observations, while LES-GFS shifts the PDF to the right, indicating stronger winds than those simulated by WRF-GFS, which better align with the observations (Fig. 10e). At the O stations, WRF-GFS simulates stronger winds compared to observations, and the LES-GFS simulations mitigate this positive wind bias observed in WRF-GFS. Furthermore, when considering the top 10% of observed PDFs of WS10 [over 6 (5) m s−1 at the L (O) stations], the PDFs of LES-GFS at L (O) stations become more similar to observations with increasing (decreasing) frequency. As shown in the time series of the 1-month average (Figs. 6–8), the PDFs of T2m, Q2m, and WS10 from LES-GFS (10 × 10) data are similar to those from WRF-GFS at the O stations, and the ones from LES-GFS (10 × 10) data closely resemble those shown by LES-GFS at the L stations, reflecting the horizontal variability.
PDFs of (a),(b) T2m, (c),(d) Q2m, and (e),(f) WS10. (left) Results for L stations and (right) results for O stations.
Citation: Journal of Applied Meteorology and Climatology 63, 8; 10.1175/JAMC-D-24-0011.1
Overall, LES-GFS shows some merits in simulating time series for low-level meteorological fields. For example, LES shows superiority over the mesoscale WRF Model in terms of simulating time series of T2m over ocean and Q2m over land as well as PDFs of T2m, Q2m, and WS10. Cécé et al. (2016) also showed WRF-LES with a 111-m grid spacing predicted T2m and Q2m better than coarser grid simulations. Crosman and Horel (2017) demonstrated that, in cold-pool case simulations, the simulated T2m in the LES experiment closely matched the observations, and this was in contrast to the coarser grid simulations.
4. Summary and discussion
The current study investigates the effects of different horizontal resolutions and initial/boundary conditions from global forecasts on simulated low-level meteorological variables such as temperature, water vapor, and wind speed around Incheon International Airport. For these purposes, numerical experiments are conducted using the WRF Model over a period of 24 h starting at 0000 UTC on each day of April 2021, resulting in 30 sets of forecasts. The study primarily focuses on evaluating monthly averaged forecast fields, as they help validate the accuracy of atmospheric models in general. The period of April 2021 is selected for numerical experiments because this month presents the highest average wind speed for 10 years, from 2011 to 2020. The model domain comprises two domains having 10- and 1-km grid spacings centered at Incheon International Airport. The forecast fields from the KIM and GFS global models are used as initial/boundary conditions for the model. Additional model domain having 100-m horizontal grid spacing is inserted as the innermost domain to investigate the impact of model horizontal resolution on the simulated low-level meteorological fields.
A comparison of the two global models, the KIM and GFS, reveals high-pressure patterns at mean sea level over the Yellow Sea and East Sea, with the former simulating lower mean sea level pressure over the Korean Peninsula compared to the latter, deviating from the reanalysis data. Overall, the KIM simulates wetter conditions than GFS over the Yellow Sea and the western part of the Korean Peninsula, including Yeongjong Island where Incheon International Airport is located. For low-level fields, both models simulate lower 2-m temperature than the observations, with the former in particular simulating warmer temperatures during the daytime and cooler temperature during the night than the latter. The KIM consistently simulates higher 2-m water vapor mixing ratio and smaller diurnal variation in 10-m wind speed at all observation stations compared to GFS.
A comparison of the global model forecasts and the 1-km grid-spacing WRF Model simulation shows that the regional model captures 2-m temperature time series relatively well compared to the global forecasts. Additionally, the magnitude of diurnal variation in 2-m temperature and 2-m water vapor mixing ratio increases in the regional model simulations compared to the global forecasts, and this is comparable with the observations. Although the regional models underestimated 10-m wind speed over land stations and overestimated it over ocean stations compared to the observations, the regional models exhibit superiority in terms of capturing time series compared to the global model forecasts. Two sets of regional WRF Model simulations forced by the KIM and GFS forecasts are also compared to examine the impact of initial and boundary conditions on simulating low-level atmospheric fields. The WRF simulations forced by GFS forecasts perform better in terms of forecasting 2-m temperature and capturing its time series than the WRF simulation forced by KIM forecasts. Meanwhile, the WRF Model simulations forced by KIM forecast fields simulate 2-m water vapor mixing ratio better in terms of mean values of 24-h forecasts over a period of 1 month, while those forced by GFS show superior correlation coefficients for the time series of 2-m water vapor mixing ratio. In the simulation of 10-m wind speed, the performance of the two models is similar.
The impact of the horizontal resolutions of the model on simulated low-level atmospheric fields is examined by comparing WRF-LES with a 100-m horizontal grid spacing and high-resolution WRF simulations with 1-km grid spacing. The former employs a three-dimensional LES subgrid-scale model for subgrid-scale turbulence mixing, while the latter uses a one-dimensional PBL scheme. The initial and boundary conditions for both simulations are forced using GFS forecasts, as they perform better in capturing synoptic fields compared to those using KIM forecasts. The 100-m grid-spacing experiment simulates better 2-m temperature time series over ocean stations and 2-m water vapor mixing ratio ones over land stations compared to the 1-km grid-spacing experiment. The analysis of the PDFs for low-level weather variables also highlights the superiority of the 100-m grid-spacing experiment over the 1-km grid-spacing experiment. Both the 100-m grid-spacing and 1-km grid-spacing experiments well capture the diurnal variation in 2-m temperature well, with the correlation coefficients between the time series and observations being ∼1. However, the 1-km grid-spacing experiments show higher correlation coefficients with the observed 2-m water vapor mixing ratio and 10-m wind speed time series at all the stations (excluding 2-m water vapor mixing ratio at the O2 station).
Overall, the 100-m high-resolution simulations offer some benefits in terms of forecasting low-level atmospheric fields compared to the 1-km resolution simulations. The WRF experiment using a 1-km grid spacing as the innermost domain requires ∼83 min for a 24-h forecast; in contrast, the 100-m grid-spacing experiment takes ∼2224 min (∼27 times longer) for the same forecast duration using identical computational resources, which is roughly 27 times longer. Additionally, we perform downsampling of WRF-LES by averaging meteorological values across 10 × 10 grid points, transitioning from a 100-m horizontal resolution to 1 km × 1 km subdomains. The resulting averages are subsequently interpolated to observation station points. The downsampled data present the smoothed-out small-scale variability over the ocean, resolved in LES-GFS. Importantly, this study underscores the significance of comparing high-resolution simulations with point observational data in areas adjacent to airports located near the ocean.
Acknowledgments.
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (RS-2023-00208394) and by the Korean Meteorological Administration Research and Development Program under Grant KMI2022-00310.
Data availability statement.
The source code of the WRF Model, version 4.2.2, is available at https://github.com/wrf-model/WRF/releases. The GFS and KIM forecast data utilized for the initial and boundary conditions are available at https://rda.ucar.edu/datasets/ds084.1/ and https://www.data.go.kr/, respectively. The ERA5 data for the GFS and KIM forecast model validations are available at https://cds.climate.copernicus.eu/cdsapp#!/home. The scripts for figures reported in this manuscript are available at https://doi.org/10.5281/zenodo.11071736. Model output and AWS data are available upon request (Yujeong Do via doyujeong@knu.ac.kr).
REFERENCES
Ahrens, C. D., 2009: Meteorology Today: An Introduction to Weather, Climate, and the Environment. 9th ed. Brooks/Cole, 599 pp.
Avolio, E., S. Federico, M. M. Miglietta, T. L. Feudo, C. R. Calidonna, and A. M. Sempreviva, 2017: Sensitivity analysis of WRF Model PBL schemes in simulating boundary-layer variables in southern Italy: An experimental campaign. Atmos. Res., 192, 58–71, https://doi.org/10.1016/j.atmosres.2017.04.003.
Campbell, P. C., J. O. Bash, and T. L. Spero, 2019: Updates to the Noah land surface model in WRF‐CMAQ to improve simulated meteorology, air quality, and deposition. J. Adv. Model. Earth Syst., 11, 231–256, https://doi.org/10.1029/2018MS001422.
Cécé, R., D. Bernard, J. Brioude, and N. Zahibo, 2016: Microscale anthropogenic pollution modelling in a small tropical island during weak trade winds: Lagrangian particle dispersion simulations using real nested LES meteorological fields. Atmos. Environ., 139, 98–112, https://doi.org/10.1016/j.atmosenv.2016.05.028.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Chen, F., H. Peng, P. Chan, Y. Huang, and K.-K. Hon, 2022: Identification and analysis of terrain-induced low-level windshear at Hong Kong International Airport based on WRF–LES combining method. Meteor. Atmos. Phys., 134, 60, https://doi.org/10.1007/s00703-022-00899-1.
Crosman, E. T., and J. D. Horel, 2017: Large-eddy simulations of a Salt Lake Valley cold-air pool. Atmos. Res., 193, 10–25, https://doi.org/10.1016/j.atmosres.2017.04.010.
Cui, C., Y. Bao, C. Yuan, Z. Li, and C. Zong, 2019: Comparison of the performances between the WRF and WRF-LES models in radiation fog–A case study. Atmos. Res., 226, 76–86, https://doi.org/10.1016/j.atmosres.2019.04.003.
Di, Z., S. Zhang, J. Quan, Q. Ma, P. Qin, and J. Li, 2023: Performance of seven land surface schemes in the WRFv4. 3 model for simulating precipitation in the record‐breaking meiyu season over the Yangtze–Huaihe River Valley in China. GeoHealth, 7, e2022GH000757, https://doi.org/10.1029/2022GH000757.
Falasca, S., and G. Curci, 2018: High-resolution air quality modeling: Sensitivity tests to horizontal resolution and urban canopy with WRF-CHIMERE. Atmos. Environ., 187, 241–254, https://doi.org/10.1016/j.atmosenv.2018.05.048.
Feigenwinter, C., J. Franceschi, J. A. Larsen, R. Spirig, and R. Vogt, 2020: On the performance of microlysimeters to measure non-rainfall water input in a hyper-arid environment with focus on fog contribution. J. Arid Environ., 182, 104260, https://doi.org/10.1016/j.jaridenv.2020.104260.
Gallus, W. A., Jr., and J. F. Bresch, 2006: Comparison of impacts of WRF dynamic core, physics package, and initial conditions on warm season rainfall forecasts. Mon. Wea. Rev., 134, 2632–2641, https://doi.org/10.1175/MWR3198.1.
Gevorgyan, A., 2018: Convection-permitting simulation of a heavy rainfall event in Armenia using the WRF model. J. Geophys. Res. Atmos., 123, 11 008–11 029, https://doi.org/10.1029/2017JD028247.
Global Climate and Weather Modeling Branch, 2003: The GFS atmospheric model. NCEP Office Note 442, 14 pp., https://repository.library.noaa.gov/view/noaa/11406.
Gultepe, I., and Coauthors, 2019: A review of high impact weather for aviation meteorology. Pure Appl. Geophys., 176, 1869–1921, https://doi.org/10.1007/s00024-019-02168-6.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Hong, S.-Y., and Coauthors, 2018: The Korean Integrated Model (KIM) system for global weather forecasting. Asia-Pac. J. Atmos. Sci., 54, 267–292, https://doi.org/10.1007/s13143-018-0028-9.
Huang, B., C. Liu, V. Banzon, E. Freeman, G. Graham, B. Hankins, T. Smith, and H.-M. Zhang, 2021: Improvements of the Daily Optimum Interpolation Sea Surface Temperature (DOISST) version 2.1. J. Climate, 34, 2923–2939, https://doi.org/10.1175/JCLI-D-20-0166.1.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long‐lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, https://doi.org/10.1029/2008JD009944.
Jiménez, P. A., and J. Dudhia, 2012: Improving the representation of resolved and unresolved topographic effects on surface wind in the WRF model. J. Appl. Meteor. Climatol., 51, 300–316, https://doi.org/10.1175/JAMC-D-11-084.1.
Jiménez, P. A., J. Dudhia, J. F. González-Rouco, J. Navarro, J. P. Montávez, and E. García-Bustamante, 2012: A revised scheme for the WRF surface layer formulation. Mon. Wea. Rev., 140, 898–918, https://doi.org/10.1175/MWR-D-11-00056.1.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181, https://doi.org/10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47, 2784–2802, https://doi.org/10.1175/1520-0469(1990)047<2784:AODEPM>2.0.CO;2.
Kim, C. K., and S. S. Yum, 2010: Local meteorological and synoptic characteristics of fogs formed over Incheon international airport in the west coast of Korea. Adv. Atmos. Sci., 27, 761–776, https://doi.org/10.1007/s00376-009-9090-7.
Klemm, O., and T. Wrzesinsky, 2007: Fog deposition fluxes of water and ions to a mountainous site in Central Europe. Tellus, 59B, 705–714, https://doi.org/10.1111/j.1600-0889.2007.00287.x.
Kumar, P., C. M. Kishtawal, and P. K. Pal, 2017: Impact of ECMWF, NCEP, and NCMRWF global model analysis on the WRF model forecast over Indian region. Theor. Appl. Climatol., 127, 143–151, https://doi.org/10.1007/s00704-015-1629-1.
Leem, H. H., D. H. Kim, and K. W. Song, 2005: The importance of fog prediction at the Incheon Int’l Airport through flight delay and cancel statistics (in Korean with English abstract). J. Korean Soc. Aviat. Aeronaut., 13, 1–10.
Lim, K.-S. S., and S.-Y. Hong, 2010: Development of an effective double-moment cloud microphysics scheme with prognostic Cloud Condensation Nuclei (CCN) for weather and climate models. Mon. Wea. Rev., 138, 1587–1612, https://doi.org/10.1175/2009MWR2968.1.
Morcrette, J.-J., H. W. Barker, J. N. S. Cole, M. J. Iacono, and R. Pincus, 2008: Impact of a new radiation package, McRad, in the ECMWF Integrated Forecasting System. Mon. Wea. Rev., 136, 4773–4798, https://doi.org/10.1175/2008MWR2363.1.
National Centers for Environmental Prediction, 2015: NCEP GFS 0.25 degree global forecast grids historical archive. National Center for Atmospheric Research, Computational and Information Systems Laboratory, accessed 26 December 2022, https://doi.org/10.5065/D65D8PWK.
Prasanna, V., H. W. Choi, J. Jung, Y. G. Lee, and B. J. Kim, 2018: High-resolution wind simulation over Incheon International Airport with the Unified Model’s rose nesting suite from KMA operational forecasts. Asia-Pac. J. Atmos. Sci., 54, 187–203, https://doi.org/10.1007/s13143-018-0003-5.
Shin, H. H., and S.-Y. Hong, 2015: Representation of the subgrid-scale turbulent transport in convective boundary layers at gray-zone resolutions. Mon. Wea. Rev., 143, 250–271, https://doi.org/10.1175/MWR-D-14-00116.1.
Skamarock, W. C., and Coauthors, 2019: A description of the Advanced Research WRF version 4. NCAR Tech. Note NCAR/TN-556+STR, 145 pp., https://doi.org/10.5065/1dfh-6p97.
Solbakken, K., Y. Birkelund, and E. M. Samuelsen, 2021: Evaluation of surface wind using WRF in complex terrain: Atmospheric input data and grid spacing. Environ. Modell. Software, 145, 105182, https://doi.org/10.1016/j.envsoft.2021.105182.
