1. Introduction
The occurrence of fog could not only pose severe threats to all modes of transportation systems (land, sea, and air) by lowering horizontal visibility but also exacerbate air pollution and pose health hazards to human beings (Hao et al. 2017; Li et al. 2019; Gultepe et al. 2019). With the rapid development of high-performance computation, numerical weather prediction (NWP) models have become powerful tools for studying and forecasting fog processes (Li and Pu 2022, 2024). However, the current skill of fog forecasts by NWP models is still poor compared to the forecasts of other weather elements such as precipitation (Zhou et al. 2012; Pu et al. 2016). There are lots of factors influencing the prediction capability of fog, within which the lack of sufficient horizontal grid resolution (HGR) in NWP models is thought to be one of the major reasons for the low accuracy of fog forecasts (Shi et al. 2010; Bergot and Koracin 2021).
How to choose the most appropriate HGR for different research objects remains one of the key scientific issues in atmospheric boundary layer modeling (Wang et al. 2020). Fog is a weather phenomenon influenced by multiscale systems, and the meso- and small-scale forcing always lead to distinct local features in fog. Enhancing the model HGR is an important direction for improving the accuracy of fog numerical forecasts (Pu et al. 2023). However, studies suggested that when the HGR of mesoscale models rises to a certain extent, the applicability of parameterization schemes may be compromised, resulting in the so-called gray-zone problem mainly associated with convection and boundary layer turbulence processes (Zhou et al. 2014; Doubrawa and Muñoz-Esparza 2020; Lakra and Avishek 2022). The gray-zone problem can substantially downgrade the forecast skill of NWP models even with finer resolutions (Dyer et al. 2016), and to address this issue, several revised approaches have been proposed for appropriately representing the gray-zone turbulence, mainly focusing on idealized convective boundary layers (Honnert et al. 2011; Boutle et al. 2014; Ito et al. 2015). Nonetheless, studies concerning the performance of currently widely used turbulence parameterizations at different horizontal resolutions (involving gray-zone scales) are relatively scarce, especially under foggy or stable stratified conditions.
Existing research studies on the impact of HGR on numerical forecasting skills mainly focused on severe convective systems such as tropical cyclone (TC) and heavy precipitation events (Dyer et al. 2016; Schwartz and Sobash 2019; Weisman et al. 2023). Lynn et al. (2020) examined the impact of different HGRs on simulations of winter precipitation in the United States and indicated that the 4-km HGR significantly outperformed the 12-km HGR in precipitation forecasting, whereas the 1.3-km HGR did not show better skills than the 4-km HGR. Xu and Wang (2021) pointed out that the model HGR played an important role in the simulation of TC intensities and inner-core structures. Enhancing the model HGR from 2 km to 500 m tends to produce a stronger TC with stronger surface winds and a smaller TC inner-core size. In comparison with convective conditions, research studies on the influence of different HGRs on radiation fog forecasting are relatively scarce. Shi et al. (2005) and da Rocha et al. (2015) investigated the impact of different HGRs (30 km vs 10 km and 50 km vs 20 km) on radiation fog simulation, suggesting that some improvement in forecasting skills could be achieved by increasing the model HGR, but both of their research studies only focused on relatively coarser HGRs within the range of 10-km scales. Boutle et al. (2016) nested a downscaling model, the London Model (LM), within the operational forecast model, the U.K. variable-resolution (UKV) model (the U.K. forecast model), to investigate whether further enhancements to HGR (from 1.5 km of the parent grid to 333 m of the nested grid) provide any benefit for fog forecasting in the greater London area. Their results indicated that similar performance (between 1.5-km and 333-m HGRs) was observed at short lead times, whereas improved performance was mainly exhibited at longer lead times. Smith et al. (2021) evaluated the performance of the Met Office Unified Model (MetUM) with three HGRs, 1.5 km, 333 m, and 100 m (333- and 100-m grids are one-way nested within the 1.5-km parent grids) in simulating the radiation fog. They found that the 100-m HGR model performed better in fog formation and stability transition timing, whereas the overall accuracy is still limited. Little research was conducted to examine the role and influence of model HGRs, covering the range from ∼10-km to ∼100-m scales, on the simulation of radiation fog.
The North China Plain (NCP) region is one of the key areas in China with frequent fog occurrences (Sun et al. 2013). Moreover, both the frequency and duration of fog events have shown a clear increasing trend in recent decades (Han et al. 2015). Tianjin, as the second-largest megacity in the NCP, also serves as a major international shipping center in North China with the Bohai Sea to its east. The concentrated economic activities and well-developed transportation systems in Tianjin both have an urgent need in fog forecasting with higher precision and resolution. However, enhancing the model HGR would inevitably accompany the exponentially increased computing costs as well as the abovementioned “gray-zone” problems. As Lawson et al. (2021) estimated, raising the model HGR from 3 to 1 km requires about 50 times the computing resources, so one must carefully evaluate the benefits against the cost. Therefore, a comprehensive understanding of the impact of different model HGRs on fog forecasting, and finding an optimal configuration balancing the HGR, computational cost, and forecasting accuracy, is crucial for further improving the forecasting skills of fog events.
The purpose of this study is to examine the impact of different model HGRs, covering the range between ten-kilometric and hectometric scales, on the forecasting performance of radiation fog in Tianjin, a core city of the NCP region in China. By utilizing the densely distributed automatic weather stations (AWSs), the sensitivities of simulated fog to model HGRs are studied in detail, including the statistical scores of forecasting skills, the simulated spatial fog coverage, and the near-surface meteorological conditions during foggy conditions. The remainder of this paper is organized as follows. The observation data, model configuration, and experimental design are described in section 2. Section 3a statistically analyzes the forecasting skills of four fog events with HGRs of 15, 5, 2.5, and 500 m, respectively. In section 3b, the impacts of model HGR on the simulated time evolutions of spatial fog coverage are investigated through the study of a radiation fog case over the NCP region. Additionally, section 3c further discusses the reasons for the HGR impact on fog simulation through comparing the hourly threat scores (TSs) for fog forecasting at different HGRs as well as assessing the model performances with different HGRs on predicting the temporal and spatial variations of near-surface meteorological elements under radiation fog conditions. Finally, the main conclusions are summarized in the last section.
2. Data and model configuration
a. Source and processing of observational data
To comprehensively analyze the impact of different HGRs on the simulation of fog and corresponding meteorological conditions, this study utilized the observational data from the China Meteorological Administration (CMA; https://data.cma.cn). The surface observational data mainly included hourly temperature, relative humidity, wind speed, wind direction, and visibility, which are obtained from national standard surface AWSs located in Tianjin and its surrounding areas (as shown in Figs. 1a,b). These data are used for assessing the model performance in predicting the spatial extent of fog areas and surface meteorological elements. A Cressman interpolation method is performed on the site-observation data to yield their corresponding contour maps.
(a) The model domain configurations and (b) the locations of all AWSs within the TJS area (as indicated by the black symbols, wherein the star symbol represents the Hangu station) and the spatial distributions of surface temperature (°C) derived from (c) MODIS myd11a2 product (the locations of Tianjin and two reservoirs are outlined by the coarse lines in red and blue colors, respectively) and (d) ERA5 reanalysis data.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
b. Model configuration and experimental design
To investigate the impact of different HGRs on fog simulations, four sets of numerical experiments using single-domain simulations (without nesting) are designed with HGRs of 15, 5, and 2.5 km and 500 m, as detailed in Table 1. As indicated in Steeneveld et al. (2015), the employment of grid nesting appears to deteriorate the simulation of fog formation in the model. Hence, following their recommendation, a single-domain simulation (without nesting) is conducted in the present study. The model domain for all four sets of experiments is shown in Fig. 1, with a central latitude and longitude of 39.6°N and 115.4°E, respectively. The number of grid points is chosen such that the model domain remains constant for all simulations. The initial and lateral boundary conditions were derived from the National Centers for Environmental Prediction (NCEP) final (FNL) analysis data with a horizontal resolution of 1° × 1°. The number of vertical layers was 53, with 23 layers below 1 km. The lowest model layer is typically around 10 m above ground level in the NCP region. The simulation duration for all experiments was 72 h.
Configuration of the numerical experiments.
The Tianjin and its surrounding (TJS) area (Fig. 1b, 38.3°–40.3°N, 116.3°–118.3°E) is characterized by a relatively complex underlying surface, with the Bohai Sea to its southeast side, the Yuqiao Reservoir in its north side, and the Beidagang Reservoir in its south side (both with an area of approximately 150 km2). During winter, the water body (including reservoirs and sea) is generally a warm and humid region at a local scale (as shown in Figs. 1c,d). With the rapid development of urbanization in the Tianjin area, the increase of high-rise buildings (namely, the increase in surface roughness lengths) would affect the local wind flows (both speed and direction) to a large extent. Therefore, Hangu station, located far from urban areas with flat surroundings, is selected to indicate the influence of background synoptic wind flows (as marked by the star symbol in Fig. 1b). Note that the following statistical assessment in this study only focuses on the observation stations within the TJS area, which included a total of 87 AWSs.
c. Identification criteria of the observed and simulated fog areas
Hourly meteorological and visibility observations from AWSs are utilized for the determination of observed fog areas. Following Ye et al. (2015), when the visibility is less than 1 km and the relative humidity is ≥90%, excluding visual range obstacles caused by precipitation, sandstorms, and other weather phenomena would be identified as fog.
Regarding the diagnosis of simulated fog areas, as the conversion algorithm from cloud liquid water content (QC) to visibility is still full of controversy (Gultepe et al. 2006), this study adopts the method suggested by Tian et al. (2019). According to this method, regions where the model’s lowest layer met the criterion of 0.005 g kg−1 ≤ QC < 0.7 g kg−1 and the height of continuous vertical distribution is below 400 m are diagnosed as simulated fog areas.
d. Statistical measures for numerical forecasting skills
3. Results and discussion
a. Statistical analysis of the impact of HGR on fog forecasting skills
To quantitatively assess the impact of model HGR on fog forecasting skills, this study selects four radiation fog cases (involving 10 fog days) in the NCP region during the 2016–19 period (as detailed in Table 2). Analysis on the background meteorological conditions shows that all fog days considered in the present study formed under weak-wind conditions at night, during which the radiative cooling effect dominates the formation of fog. Nonetheless, the dominant mechanism led to the dissipation of fog differs and could be categorized into two types. Out of 10 total fog days, eight dissipated mainly due to the shortwave radiative heating after sunrise when the prevailing wind is still weak (domain-averaged simulated wind speed less than 3 m s−1). Accordingly, these fog days could be categorized as type I. The dissipation of type-II fog days (two out of 10 total fog days) is driven by both radiative heating and wind force. That is to say, the dissipation of fog is along with the invasion of cold airflow, which strengthens the near-surface wind speed (domain-averaged simulated wind speed larger than 3 m s−1), namely, the horizontal diffusion effect, and finally leads to the enhancement of fog dissipation. According to the different dissipation mechanisms, Table 3 presents the comparison of fog forecasting skills (i.e., TSs) among four applied HGR simulations in two types of fog days. The mean TSs for different HGR runs are also presented in Table 3, which are calculated based on the hourly values in all fog cases.
The duration period and model initial time for the four fog cases (LST stands for local standard time, i.e., Beijing time. LST = UTC + 8).
Statistical scores (TS) for the forecasting skills of each fog day using different HGRs (the bold font styles indicate the optimal results in each row).
As shown in Table 3, for type-I fog days (with no obvious impact of the background synoptic wind), differences in fog forecasting skills (TS values) among different HGR simulations are relatively smaller, with the 5-km HGR simulation generally showing slightly better performance than the 15- and 2.5-km HGRs, while for type-II fog days (with the dissipation of fog strongly affected by the invasion of cold airflow), the deviations in TS values among simulations using different HGRs become more evident, with the 2.5-km HGR providing better performance, followed by the 5-km and 500-m HGRs. Moreover, it is also noteworthy that, for type-II fog days, the coarse-HGR (15-km) simulation shows a noticeable degradation in the fog forecasting skills and provides the lowest TSs compared to finer-HGR (5-km, 2.5-km, and 500-m) simulations.
Based on the analysis mentioned above, we found that fog forecasting using different HGRs behaves differently for the two types of fog days, which are primarily differentiated by whether or not the dissipation of fog is substantially influenced by the background synoptic wind flow (e.g., the invasion of cold air masses). Accordingly, in order to further investigate the role of wind intensity on fog forecasting skills at different HGRs, Table 4 presents the comparison of TSs of different HGR simulations under different wind conditions (i.e., domain-averaged simulated wind speed below or above 3 m s−1). The wind threshold (3 m s−1) represents the domain-averaged simulated wind speed above which the simulated fog would enter into the dissipation stage and the deviations in TSs among different HGR simulations were substantially enlarged. As shown in Table 4, when the simulated average wind speed is lower than 3 m s−1, deviations in TSs among different HGR simulations are relatively small, with the 5-km HGR providing slightly better TS performance followed by the 15- and 2.5-km HGRs. However, when the average wind speed increased to greater than 3 m s−1 (namely, the dissipation of fog is greatly influenced by wind flow), deviations in TSs among different HGR simulations become more pronounced. Under this condition, the 2.5-km HGR shows the best fog forecasting skills, followed by the 5-km and 500-m HGRs. Meanwhile, notable performance degradation is seen in the 15-km simulation, producing a much lower TS compared to finer-HGR simulations.
Statistical scores (TS) for fog forecasting skills under different wind conditions (the bold font styles indicate the optimal results in each row).
In general, the analysis mentioned above reveals that changes in model resolution (15, 5, and 2.5 km and 500 m) have a notable impact on the forecasting skills of radiation fog processes. The enhancement of model HGR to a certain extent would surely improve the model performance of fog forecasting. However, the model performance in forecasting fog would not monotonically improve with the enhancement of model HGR and simulations using different HGRs behave differently for the two types of fog days. When the near-surface wind intensity remains weak during the whole process of fog (i.e., type-I fog days), deviations in fog forecasting skills at different HGRs are relatively small, with the 5-km HGR providing slightly better TS performance than the 15-km, 2.5-km, and 500-m HGRs. When the dissipation of fog is greatly influenced by wind flow (i.e., type-II fog days), deviations in TSs among different HGR simulations become more evident, with the 2.5-km HGR showing the best fog forecasting performance, followed by the 5-km and 500-m HGRs, while a notable performance degradation is seen in the 15-km simulation. Generally speaking, the 5-km HGR provides an overall better performance in forecasting the two different types of fog over the North China Plain, followed by the 2.5-km HGR, whereas simulation using the finest hectometric-scale (500-m) HGR shows no superiority (or somewhat degradation) in fog forecasting skills for both of the two types of fog days and also would inevitably lead to the dramatic growth in computational costs. Additionally, it should be noted that, given the relatively poor capability of forecasting radiation fog by mesoscale models (especially for the scattered, patchy radiation fog which is greatly influenced by small-scale topography), this study only focuses on the simulation of relatively widespread radiation fog events; hence, the conclusion drawn here may be only applicable for the relatively widespread mesoscale radiation fog events.
In the next section, further analysis will be conducted to examine the differences in the forecasting skills of near-surface meteorological elements (temperature, humidity, wind) and fog weather at different model HGRs, taking the radiation fog event occurred in November 2018 as a typical case (for this case including both type-I and type-II fog days), and discuss their potential impacts on the formation and dissipation of fog.
b. Case study of the impact of HGR on fog simulation
1) Synoptic background during fog event
From 25 to 27 November 2018, a widespread and persistent fog event occurred in the NCP region. The occurrence of dense fog events is closely related to favorable synoptic background conditions. Figure 2 presents the synoptic patterns based on ERA5 reanalysis data at the formation stage of this fog event (2000 LST 25 November 2018). At a height of 500 hPa (Fig. 2a), the East Asian continent exhibited a south trough and north ridge circulation pattern, with the NCP region predominantly influenced by straight westerlies ahead of a shallow trough. The 700-hPa geopotential height (GPH) field (Fig. 2b) was mainly under the control of the rear of a weak high-pressure ridge with sparse isobars. At 850 hPa (Fig. 2c) and the surface (Fig. 2d), the NCP region was located within the uniform pressure field between the bottom of the northeast low pressure and the rear of the marine high pressure system. The pressure gradient over the NCP region was weak, and the prevailing flows were from the south with low-wind velocity. Both the circulation pattern at the upper atmosphere and the surface synoptic system controlling the NCP region were quite stable, hence favoring the formation of dense fog. After 1800 LST 26 November, a weak cold front began to intrude, leading to a shift in wind direction to a weak northerly flow within the research domain. By 1000 LST 27 November, the northerly wind intensified to over 3 m s−1 (figures not shown), leading to the dissipation of this fog process. Considering the characteristics of weak-wind conditions, nocturnal formation, and diurnal dissipation, this fog process could be classified as radiation fog.
The GPH and wind field at the (a) upper (500 hPa), (b) middle (700 hPa), and (c) lower (850 hPa) atmosphere and the (d) sea level pressure field at 2000 LST 25 Nov 2018 over eastern-central China.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
Figure 3 exhibits the time evolution of observed fog areas interpolated from the surface AWS data. It can be observed that at 2100 LST 25 November, this fog event formed in the uniform pressure field at the rear of a high pressure system, first appearing as isolated patchy structures over the western part of the Shandong Peninsula. Later, with the strengthening of nighttime radiative cooling, the fog extent continuously expanded in the northeast direction, with increasing intensity (Fig. 3b). By 0500 LST 26 November (Fig. 3c), the fog areas over the western part of the Shandong Peninsula and the central and southern NCP region continued to strengthen and connect, forming more widespread fog areas. The maximum fog intensity reached the level of extremely dense fog (with visibility less than 50 m). After sunrise (Figs. 3d,e), with the enhanced heating from solar radiation, the fog area diminished rapidly. Note that the overall wind intensity throughout this fog day (26 November 2018) remained at a low level; hence, it should be categorized as type-I fog day. During the night of 26 November (Fig. 3f), under the influence of longwave radiative cooling and weak cold air, fog areas reappeared and expanded rapidly. By 0200 LST 27 November (Fig. 3g), dense regional fog developed over a large part of the NCP region, with part areas in Tianjin and Hebei reached the level of extremely dense fog. After sunrise on 27 November (Fig. 3h), with the intensification of both shortwave radiation and cold air (by 1000 LST, the northerly wind increased to over 3 m s−1), the fog areas shrank once again. Until 1200 LST (figures not shown), all fog areas had dissipated. Because of the great influence of synoptic wind flow on fog processes, the fog day of 27 November 2018 should be categorized as type-II fog day.
Spatial distribution of the observed fog areas based on the interpolations from surface AWS observations at (a) 2100 LST 25 Nov, (b) 0000, (c) 0500, (d) 1000, (e) 1400, and (f) 2000 LST 26 Nov, and (g) 2000 and (h) 0800 LST 27 Nov 2018 [the color bar represents different levels of visibility (m); the dashed rectangle denotes the TJS area].
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
2) Impact of HGR on the simulation of horizontal fog area
Figure 4 presents a comparison between the observed horizontal fog area and the corresponding simulation results under different HGRs (15, 5, and 2.5 km and 500 m) at different times during this fog process. As shown in Fig. 4, at 0000 LST 26 November during the first fog day (which belongs to the type-I fog day), the spatial coverage and intensity of fog simulated using different HGRs exhibit obvious discrepancies. For the simulation at 15-km HGR, most of the central and southern regions of Tianjin are covered by fog, which means the coarse HGR overestimates the fog area in the south of Tianjin at the formation stage of fog, with the maximum value of QC reaching approximately 0.2 g kg−1. The simulated fog coverage diminishes at the HGR of 5 km, mainly concentrating in the central and eastern regions of Tianjin, whereas the spatial extent and magnitude of high-QC areas (QC > 0.3 g kg−1) show increasing trends (indicating an expansion of the spatial extent of denser fog and an increase in its intensity). When the HGR is enhanced to 2.5 km and 500 m, the simulated distribution of fog area becomes even finer and captures the small fog area in the northeast of Tianjin, whereas some false-alarmed fog areas appear in the north of Tianjin. With the evolution of fog process (by 0200 LST 26 November), the simulated fog areas under coarse HGRs still exhibit blocky and continuous spatial distribution patterns, while with the enhancement of model HGR, the simulated fog areas begin to show discrete distribution characteristics. At 0500 LST 26 November, as shown by the visibility observations interpolated from AWS data [Fig. 4c(0)], most of the Tianjin region is covered by fog, with visibility dropping below 50 m in the northeast of Tianjin, reaching the criteria for extremely dense fog. A comparison of the simulation results with different HGRs revealed that, at an HGR of 15 km, the simulated fog extent spread over nearly the whole land area within the model domain, indicating an overestimation of the horizontal extent of fog area at this time [Fig. 4c(1)]. With the enhancement of model HGR [Figs. 4c(2)–c(4)], the simulated fog area shows some degree of reduction. Furthermore, the 5- and 2.5-km HGR simulations could better reproduce the large-QC area (approximately 0.7 g kg−1) in the northeast of Tianjin, which corresponds to the location of minimum visibility (<50 m) in the observational results [Fig. 4c(0)], whereas the 15-km HGR simulation shows an overestimation and the 500-m HGR simulation shows an underestimation of this area to some extent.
Time series of the (first column) observed and simulated fog areas using HGRs of (second column) 15 km, (third column) 5 km, (fourth column) 2.5 km, and (fifth column) 500 m during 26–27 Nov 2018. (a0)–(a4) 0000 LST 26 Nov; (b0)–(b4) 0200 LST 26 Nov; (c0)–(c4) 0500 LST 26 Nov; (d0)–(d4) 2300 LST 26 Nov; (e0)–(e4) 0800 LST 27 Nov.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
By 2300 LST 26 November during the second fog day (which belongs to the type-II fog day), the simulated fog area at different HGRs shows marked discrepancies [Figs. 4d(1)–d(4)]. Under coarser HGR (15 km), the simulated fog tends to dissipate too early, with only very small and isolated fog areas existed, resulting in a significant underestimation compared to the observed fog areas. Under finer HGRs (5 km, 2.5 km, and 500 m), the simulated fog areas are still underestimated to varying degrees. Nonetheless, simulations with finer HGRs could capture the fog area over the central and northeast of Tianjin, bringing them closer to the observations. By 0800 LST 27 November, only a small area of extremely dense fog (with horizontal visibility below 50 m) remained in the northeastern part of Tianjin [Fig. 4e(0)]. The simulated fog over the Tianjin area had completely dissipated, indicating a missed forecast of fog at this time under HGR of 15 km. Comparatively, simulation at 5-km, 2.5-km, and 500-m HGRs could better capture the detailed distribution patterns of different fog intensities and the location of simulated maximum-QC area showed good consistency with the observed dense fog area, among which the 2.5-km simulation appears to be the closest to observations.
c. Discussion of the reasons for the HGR impact on fog simulation
Fog is a meteorological phenomenon mainly occurring under the circumstances of clear sky, high humidity, low-wind speed, and stable stratifications, and its formation and dissipation processes are closely related to changes in the near-surface meteorological field. Therefore, in the following sections, the impacts of different HGRs on the simulations of meteorological field will be analyzed to gain insight into the reasons for differences in fog simulations. The analyses are based on the simulations of fog events during 25–27 November 2018.
1) Impact of HGR on the temporal evolution of TSs for fog forecasting
To further investigate the reason for the difference in the fog forecasting capabilities with different HGRs, Fig. 5 exhibits the temporal evolutions of TSs with HGRs of 15, 5, and 2.5 km and 500 m. As shown in Fig. 5 (and mentioned above), this fog case mainly involves two fog days. During the first fog day (categorized as type-I fog day), TS values are in close agreement among different HGR simulations, whereas during the second fog day (categorized as type-II fog day), differences in TS become more evident.
Time series of TSs for fog forecasting under different HGRs.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
2) Impact of HGR on the simulation of surface meteorological factors
To further clarify the reasons for the temporal variations in fog forecasting capability (TSs) at different HGRs, Fig. 6 presents the hourly statistical validation results of RMSE and IOA for 2-m air temperature (T2), 2-m relative humidity (RH2), and 10-m wind speed (WS10) from AWS observations and simulations at different HGRs during this fog event. Systematic changes in wind direction can indicate the transition of weather systems, but it can be easily affected by local-scale elements under weak-wind conditions. Therefore, Hangu station, located far from urban areas with flat surroundings, is selected to indicate the influence of weather systems (Fig. 6d). Additionally, Table 5 presents the evaluation statistics (RMSE and IOA) on meteorological factors simulated at different HGRs for each fog day.
Time series of (a) T2, (b) RH2, and (c) WS10 from AWS observations (black) and simulations at different HGRs (15 km: red; 5 km: green; 2.5 km: blue; 500 m: yellow) averaged over all AWS observations within the analysis domain as shown in Fig. 1 and (d) 10-m wind direction at Hangu station during this fog event. The corresponding validation results are also shown in the figure.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
Evaluation statistics (RMSE/IOA) on meteorological factors simulated using different HGRs for each fog day.
(i) Impacts of HGR on the simulation of near-surface wind field
As shown in Fig. 6, it can be observed that model HGR has a more pronounced impact on the simulation of near-surface wind speed (Fig. 6c), compared to temperature (Fig. 6a) and humidity (Fig. 6b). Regarding the 10-m wind field (Figs. 6c,d), simulations using different HGRs generally show good agreement with the observed trends of weak southerly winds in the early period and strengthened northerly winds in the later period. However, the model generally shows a systematic overestimation of near-surface wind speeds over the Tianjin area, no matter which HGR is used. Overestimation of near-surface winds by the WRF Model has also been documented in previous studies, and this systematic bias is assumed to be due to an inadequate representation of surface roughness (Cheng and Steenburgh 2005; Madala et al. 2015; Sathyanadh et al. 2017). Intercomparisons of averaged simulated wind speed with different HGRs reveal that the 15-km simulation generally exhibited the most pronounced overestimation of WS10, as well as the largest deviation from observed wind directions. Differences in wind speed and direction between 5- and 2.5-km simulations are relatively smaller, and the predicted timing of changes in wind direction at these two HGRs almost coincides with the observations. The finest HGR (500 m) appears to show the best performance in forecasting wind, with the lowest RMSE and highest IOA values, compared to coarser-HGR simulations (Table 5). The systematic overestimation of near-surface wind is generally reduced to some degree under 500-m HGR simulation, and the timing of changes in wind direction almost coincides with the 5- and 2.5-km HGR simulations. Regarding different fog days, it could be found by combining Figs. 5 and 6 that, during the first fog day when the domain-averaged WS10 is rather small (hence categorized as type-I fog day) and the differences among simulations are small as well, TSs among different HGR simulations are in close agreement. However, during the second fog day, a clear rising tendency occurs in WS10 (hence categorized as type-II fog day) and the deviations of WS10 among simulations become more evident (see Fig. 6 and Table 5), as well as the TSs (see Fig. 5). To investigate the reason for the difference in simulated WS10 and fog forecasting capabilities among different HGR simulations during the two different types of fog days, Figs. 7 and 8 exhibit the sea level pressure field, site observations of near-surface wind speed and direction, and simulated near-surface wind speed under different HGRs at 0200 LST 26 November and 2300 LST 26 November, respectively.
(a) The sea level pressure field and (b) site observations of 10-m wind speed and direction and near-surface wind speed simulations at different HGRs [(c) 15 km, (d) 5 km, (e) 2.5 km, and (f) 500 m] at 0200 LST 26 Nov.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
As in Fig. 7, but for 2300 LST 26 Nov.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
As shown in Fig. 7a, at 0200 LST 26 November, the Tianjin area was mainly under the control of marine high pressure system with weak background synoptic wind (a typical characteristic of the type-I fog day). The observed near-surface wind speeds in central and northern Tianjin are generally lower than 1.2 m s−1. A comparison between the site-observed wind field (Fig. 7b) and simulations at different HGRs (Figs. 7c–f) reveals that simulations using the 15- and 5-km HGRs would generally provide coarser wind fields and hence exhibit a weaker ability to distinguish the influence of small-scale topography (the reservoir areas, the eastern sea areas, the northern hilly areas, and the western urban areas) within the study domain on local wind fields, compared to the simulations using finer HGRs (2.5 km and 500 m). This may finally lead to the overestimation in the coverage of the low-wind areas in central Tianjin. Meanwhile, the coarse-HGR simulations overestimate the near-surface wind speeds in other areas (e.g., northern and western Tianjin) and hence result in a more pronounced overestimation in terms of domain-averaged wind speeds (as shown in Fig. 6c). As the model HGR enhanced to 2.5 km and finer (500 m), simulation results could more precisely capture the nonuniform distribution features of near-surface wind field. More specifically, the simulated minimum wind speed zones in central and northern Tianjin are in close agreement with the site observations, and the coverage of low-wind regions in the south is significantly diminished (especially for regions around the reservoir areas where the near-surface wind speed exceeds 3 m s−1). Although not all weak-wind regions lead to fog occurrence, radiation fog over land only occurs when the prevailing wind is weak. As shown in Figs. 4b(1)–b(4), all of the simulated fog areas are located within the weak-wind regions. Moreover, as the coverage of simulated low-wind area is observed to diminish when the model HGR is enhanced from 15 km to 500 m (Figs. 7c–f), the simulated fog coverage decreases correspondingly [Figs. 4b(1)–b(4)]. In summary, when the influence of large-scale background wind is weak (i.e., the overall wind speed within the study domain is low), the discrepancies in the domain average of simulated wind speeds among different HGRs are relatively stable (Fig. 6c). Nonetheless, fine-HGR simulations can better represent the nonuniform distribution characteristics of near-surface wind fields. The 15-km simulation provides not only a relatively higher value in terms of domain-averaged wind speeds, but also a larger coverage of the low-wind speed areas in central Tianjin, compared to finer-HGR simulations.
At 2300 LST 26 November (Fig. 8), influenced by the southeast penetration of the northeast cold vortex, the Tianjin area was located in the bottom of low pressure systems where the near-surface wind speeds began to rise (a typical characteristic of the type-II fog day). Site observations also show a clear rising tendency of near-surface wind speed from east to west (>3 m s−1) and the wind direction turns to northwest, which indicates the influence of synoptic-scale cold vortex flow began to enter into the study domain. Based on the statistical results of WS10 observations during fog processes from 2016 to 2019 (Fig. 9), there appears to be a threshold relationship between the near-surface wind speed and fog processes. That is, under foggy conditions, the 95th percentile of WS10 is approximately 2.2 m s−1, which means fog can hardly exist when WS10 exceeds 2.2 m s−1. Comparisons of the simulated near-surface wind field under different HGRs at 2300 LST 26 November (Figs. 8c–f) show that, when the influence of synoptic winds enters into the research domain, deviations among different HGR simulations expand. This is mainly because the coarse grids tend to simulate a faster entry (or advection) speed of synoptic wind flow, resulting in more rapid growth (namely, more pronounced overestimation) of overall wind speed across the domain. Consequently, there are almost no low-wind areas in most parts of the domain (Fig. 8c), leading to an earlier dissipation of fog processes at 15-km HGR (Fig. 5), whereas in finer-HGR simulations (5 km–500 m, as shown in Figs. 8d–f), several low-wind areas could still be captured in the central and northeast of the domain, resulting in the maintenance of simulated fog areas in these areas. As mentioned above, the predicted low-wind areas are very important for the formation and maintenance of fog processes and radiation fog over land only occurs around low-wind areas. Although the comparison of domain-averaged WS10 observations and simulations (Fig. 6c) still shows a systematic overestimation of the modeled WS10 at this time (2300 LST 26 November), the key role of the predicted low-wind areas in fog simulation would consequently result in deviations in the fog forecasting skills using different HGRs. Besides, the simulated wind field at hectometric-scale (500-m) HGR shows some kind of stripe-shaped distribution patterns. Several previous studies have also found the stripe-shaped distribution characteristics of horizontal wind speed or liquid water content in fog simulation at fine resolutions (Nakanishi 2000; Bergot 2013; Doubrawa and Muñoz-Esparza 2020). Due to the lack of correspondingly high-spatial-resolution (e.g., spatial interval of hundreds of meters, matching with the model grid intervals) observation networks, it is difficult to validate this feature and is beyond the scope of this study.
Boxplot of the percentage of near-surface wind speed with the presence of fog for the four fog cases from 2016 to 2019.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
(ii) Impacts of HGR on the simulation of near-surface temperature and humidity field
Regarding the time evolution of near-surface temperature (Fig. 6a), during the period of weak southerly winds (i.e., weak warm advections) in the first fog day (type-I fog day), there is minimal difference in temperature simulations among different HGRs, with simulated temperatures generally showing a higher bias. Evaluation statistics of simulated T2 also show minimal difference among different HGRs in type-I fog day (see Table 5), while during the second fog day (type-II fog day), because of the strengthened cold advection associated with the invasion of northerly winds (after 2200 LST 26 November), the simulated temperatures decreased more rapidly and lower than observed values at all HGRs (compared to observations) due to the overestimation of wind speeds and the consequent stronger cold air advection (Fig. 6c). In terms of the comparison of simulated temperatures among different HGRs, it can be observed that simulation at coarser HGR tends to produce a faster invasion of the cold air and hence results in a larger cold bias of simulated T2 in the second fog day. As shown in the evaluation statistics (Table 5), for the second fog day, deviations in T2 simulation using different HGRs are much larger than those for the first fog day and the 15-km simulation shows the largest bias (with the largest RMSE and lowest IOA), whereas the 500-m simulation gives the least bias. Besides, at the timing when deviations in the simulated T2 among different HGRs get enlarged in the second fog day, deviations in the simulated RH2 enlarge simultaneously (Fig. 6b). Corresponding to the colder bias at coarser HGR in T2 simulation, there appears to be a drier bias in RH2 simulation at finer HGR; namely, the 15-km simulation produces the highest RH2, while the 500-m simulation produces the lowest RH2. This further illustrates that, under the influence of cold advection, the discrepancies in the simulated saturation state at different model HGRs mainly arise from the different speeds of cold advections entering into the study domain. Additionally, when the speed of northerly winds stabilized at greater than 3 m s−1 (after 0800 LST 27 November), the simulated near-surface temperatures at different HGRs converged again (similar to the condition under weak south winds). This indicates that after the winds intensified, the cold advection passed through the whole Tianjin area and the simulated temperatures would no longer be sensitive to grid resolutions.
To examine the impact of different model HGRs on the simulated temperature and humidity distributions, Fig. 10 presents the spatial distributions of the observed (Figs. 10a,b) and simulated near-surface temperature (the left panels) and humidity (the right panels) with different HGRs at 0200 LST 26 November. As shown in Figs. 10c and 10d, the horizontal variability in the 15-km simulation results is too small to adequately represent the differences in meteorological element distributions caused by the nonuniform underlying surface. In comparison, simulations at finer HGRs (5 km, 2.5 km, and 500 m) could not only better capture the relatively warm and humid centers formed in winter nights over the two large water bodies (i.e., the Yuqiao Reservoir in the north and the Beidagang Reservoir in the south of Tianjin, respectively) but also better distinguish the small-scale humid centers along the southeastern coast.
Spatial distributions of (first row) observed and (left) simulated near-surface temperature and (right) specific humidity using (second row) 15 km, (third row) 5 km, (fourth row) 2.5 km, and (fifth row) 500 m at 0200 LST 26 Nov.
Citation: Weather and Forecasting 40, 4; 10.1175/WAF-D-24-0091.1
In summary, with the enhancement of model HGR, simulations using finer HGR could provide a more realistic representation of temperature and humidity distributions around the two reservoir areas. That is to say, the nonuniform features of simulated dynamical and thermal fields under finer HGRs would become more evident and realistic, which can better reflect the effect of small-scale underlying surface features on local meteorological fields. The impact of different HGRs on fog simulation mainly arises from discrepancies in the response speed of wind advection. Overall, the model tends to systematically overestimate the near-surface wind field in the Tianjin (plain) area. When the prevailing wind within the study domain is weak (i.e., the influence of large-scale background wind is weak), the discrepancies in the simulated wind speed among different HGRs are relatively small. Nonetheless, when the influences of synoptic winds enter into the study domain, deviations among different HGR simulations become more evident. With larger grid spacing, the influence of synoptic wind flow would enter the study domain more quickly and earlier, resulting in a more pronounced overestimation of simulated wind speed. Additionally, land fog exhibits high sensitivity to local wind intensity, hence leading to discrepancies in fog simulation under different HGRs. Although high-resolution simulations could better represent the fine-scale distribution features of meteorological fields (e.g., temperature, humidity, and wind speed), the skill scores of fog forecasting do not consistently improve with the enhancement of model HGR. As also indicated in Wilson and Fovell (2018), in some cases the NWP models yield the right fog simulations (in terms of fog coverage, i.e., higher TS values) but for the wrong representation of meteorological fields.
4. Conclusions
This study utilizes the mesoscale atmospheric model WRF Model to analyze the impact of different HGRs, ranging from ten-kilometric to hectometric scales (15, 5, and 2.5 km and 500 m), on the statistical scores of forecasting radiation fog processes in the NCP region during the period from 2016 to 2019. Using the fog event in November 2018 as a typical case, this study further investigates the influences of different HGRs on the simulated time evolution of spatial fog coverage and corresponding hourly TSs as well as assesses the model performances with different HGRs on predicting the temporal and spatial variations of near-surface meteorological elements under radiation fog conditions. Results indicated that
-
Changes in model HGR from 15 km to 500 m would have a notable impact on fog forecasting skills. Enhanced model HGR could generally allow for better simulation of spatial distributions of meteorological fields and better reflect the effect of small-scale underlying surface features on local meteorological fields. However, enhanced model HGR does not necessarily lead to improvements in the forecasting skills (i.e., TSs) of radiation fog events. Fog forecasting at different HGRs behaves differently for the two types of fog days which are primarily differentiated by whether or not the dissipation of fog is substantially influenced by the background synoptic wind flow. Generally speaking, among the four HGRs examined in this study, the 5-km HGR provides an overall better performance in forecasting the two different types of fog, followed by the 2.5-km HGR. Additionally, simulation using the finest hectometric-scale (500-m) HGR shows no superiority (or somewhat degradation) in fog forecasting skills for both of the two types of fog days.
-
The influence of model HGR on simulating the spatial distribution of fog is more pronounced during the formation and dissipation stage. During the formation stage of fog, the coarser-HGR simulation tends to produce much more widespread fog coverage than the finer-HGR simulation. During the dissipation stage of fog, there is little difference among different HGR simulations for the type-I fog day, while for the type-II fog day, the simulated fog area at different HGRs shows marked discrepancies with the simulated fog area at coarser HGR (15 km) tending to dissipate too early. Besides, simulations with finer HGRs could provide more details of small-scale fog distribution patterns; e.g., the 5-, 2.5-, and 0.5-km HGR simulations successfully captured the dense fog center in the northeast of Tianjin, making them in better agreement with the observations during fog dissipation stage when only a small patchy fog area exists. During the developing and mature stage of fog, the influence of HGR on the overall fog extent is rather limited.
-
The impact of different HGRs on fog simulation mainly arises from discrepancies in the response speed of wind advection. Overall, the model tends to systematically overestimate the near-surface wind field in the Tianjin (plain) area. When the prevailing wind within the study domain is weak (i.e., the influence of large-scale background wind is weak), the discrepancies in the simulated wind speed among different HGRs are relatively small, so are the TSs of fog forecasting among different HGRs. Nonetheless, during the periods when the influences of synoptic winds enter into the study domain, deviations among different HGR simulations become more evident due to the different response speeds of wind flow advection. With larger grid spacing, the influence of synoptic wind flow would enter the study domain more quickly and earlier, resulting in a more pronounced overestimation of simulated wind speed. Additionally, land fog exhibits high sensitivity to local wind intensity, hence leading to discrepancies in fog simulation under different HGRs.
This study investigates the impact of model HGR, from ten-kilometric to hectometric scales, on the forecasting of radiation fog processes. Although the 5-km HGR simulation generally shows better performance in fog forecasting, considerable errors still exist in the simulation of both the fog process and individual meteorological factors, and the superiority in fog forecasting at 5 km may be, to some extent, a result of compensating errors. With the continuous development and optimization of numerical models in the future, there is no doubt that the systematic errors in the simulation of meteorological elements will be inevitably diminished (or even disappeared); hence, under this condition, the forecasting of mesoscale or microscale fog must be more accurate using high-resolution models. Nonetheless, at the current stage when considerable systematic errors still exist in model physics, it is very essential to seek the optimal model configurations for fog forecasting and investigate the possible error sources, so as to better improve the current fog forecasting accuracies. This is also the primary goal of the present study. Aside from this, previous studies indicated that the vertical grid configurations (including the vertical grid spacing and the height of the lowest model level) would also play an important role in the prediction of fog events (Gultepe and Mibrandt 2007; Maronga and Bosveld 2017). An enhanced vertical grid resolution near the surface was reported to improve the simulation of the fog formation phase (Tardif 2007; Philip et al. 2016). Accordingly, we would like to investigate the impact of vertical resolution as well as the synchronous change in both the horizontal and vertical grid spacings on the simulation of fog events in future studies. Furthermore, it should also be noted that, as indicated by Boutle et al. (2014), turbulence is not the only factor that needs to adapt to model grid resolution, and other physical parameterizations (e.g., microphysics) are also necessary to possess the scale-adaptive capabilities for gray-zone simulations. We would like to examine the performance of other physical parameterizations at different model HGRs and try different approaches to improve the performance of physical parameterizations across gray-zone resolutions.
Acknowledgments.
This study was supported by the National Natural Science Foundation of China (Grants 42205092 and 42205009), the Applied Fundamental Research Project of Tianjin (No. 22JCQNJC00370), the Open Grants of the State Key Laboratory of Severe Weather (2024LASW-B23), the Bohai Rim Meteorological Science Collaborative Innovation Fund (QYXM202112 and QYXM202202), the Open Project of Tianjin Key Laboratory of Oceanic Meteorology (2022TKLOM04), and the Beijing Open Project (BJG202404).
Data availability statement.
All ground observational data used during this study are available from the Meteorological Big Data and Cloud Platform (MCP) database from China Meteorological Administration (https://10.226.64.50:8088/cmadaas/).
REFERENCES
Bergot, T., 2013: Small-scale structure of radiation fog: A large-eddy simulation study. Quart. J. Roy. Meteor. Soc., 139, 1099–1112, https://doi.org/10.1002/qj.2051.
Bergot, T., and D. Koracin, 2021: Observation, simulation and predictability of fog: Review and perspectives. Atmosphere, 12, 235, https://doi.org/10.3390/atmos12020235.
Boutle, I. A., J. E. J. Eyre, and A. P. Lock, 2014: Seamless stratocumulus simulation across the turbulent gray zone. Mon. Wea. Rev., 142, 1655–1668, https://doi.org/10.1175/MWR-D-13-00229.1.
Boutle, I. A., A. Finnenkoetter, A. P. Lock, and H. Wells, 2016: The London Model: Forecasting fog at 333 m resolution. Quart. J. Roy. Meteor. Soc., 142, 360–371, https://doi.org/10.1002/qj.2656.
Cheng, W. Y. Y., and W. J. Steenburgh, 2005: Evaluation of surface sensible weather forecasts by the WRF and the Eta Models over the western United States. Wea. Forecasting, 20, 812–821, https://doi.org/10.1175/WAF885.1.
da Rocha, R. P., F. L. T. Goncalves, and B. Segalin, 2015: Fog events and local atmospheric features simulated by regional climate model for the metropolitan area of Sao Paulo, Brazil. Atmos. Res., 151, 176–188, https://doi.org/10.1016/j.atmosres.2014.06.010.
Doubrawa, P., and D. Muñoz-Esparza, 2020: Simulating real atmospheric boundary layers at gray-zone resolutions: How do currently available turbulence parameterizations perform? Atmosphere, 11, 345, https://doi.org/10.3390/atmos11040345.
Dyer, J., C. Zarzar, P. Amburn, R. Dumais, J. Raby, and J. A. Smith, 2016: Defining the influence of horizontal grid spacing on ensemble uncertainty within a regional modeling framework. Wea. Forecasting, 31, 1997–2017, https://doi.org/10.1175/WAF-D-16-0030.1.
Ferrari, F., F. Cassola, P. E. Tuju, and A. Mazzino, 2021: RANS and LES face to face for forecasting extreme precipitation events in the Liguria region. Atmos. Res., 259, 105654, https://doi.org/10.1016/j.atmosres.2021.105654.
Gultepe, I., and J. A. Mibrandt, 2007: Microphysical observations and mesoscale model simulation of a warm fog case during FRAM Project. Pure Appl. Geophys., 164, 1161–1178, https://doi.org/10.1007/s00024-007-0212-9.
Gultepe, I., M. D. Müller, and Z. Boybeyi, 2006: A new visibility parameterization for warm-fog applications in numerical weather prediction models. J. Appl. Meteor. Climatol., 45, 1469–1480, https://doi.org/10.1175/JAM2423.1.
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.
Han, S. Q., Z. Cai, Y. Zhang, J. Wang, Q. Yao, P. Li, and X. Li, 2015: Long-term trends in fog and boundary layer characteristics in Tianjin, China. Particuology, 20, 61–68, https://doi.org/10.1016/j.partic.2014.02.008.
Hao, T., and Coauthors, 2017: The role of fog in haze episode in Tianjin, China: A case study for November 2015. Atmos. Res., 194, 235–244, https://doi.org/10.1016/j.atmosres.2017.04.020.
Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF Single-Moment 6-class microphysics scheme (WSM 6). J. Korean Meteor. Soc., 42, 129–151.
Honnert, R., V. Masson, and F. Couvreux, 2011: A diagnostic for evaluating the representation of turbulence in atmospheric models at the kilometric scale. J. Atmos. Sci., 68, 3112–3131, https://doi.org/10.1175/JAS-D-11-061.1.
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.
Ito, J., H. Niino, M. Nakanishi, and C.-H. Moeng, 2015: An extension of the Mellor–Yamada model to the terra incognita zone for dry convective mixed layers in the free convection regime. Bound.-Layer Meteor., 157, 23–43, https://doi.org/10.1007/s10546-015-0045-5.
Jia, W., and X. Zhang, 2020: The role of the planetary boundary layer parameterization schemes on the meteorological and aerosol pollution simulations: A review. Atmos. Res., 239, 104890, https://doi.org/10.1016/j.atmosres.2020.104890.
Kim, C. K., S. S. Yum, H.-G. Kim, and Y.-H. Kang, 2019: A WRF modeling study on the effects of land use changes on fog off the west coast of the Korean Peninsula. Pure Appl. Geophys., 176, 4623–4640, https://doi.org/10.1007/s00024-019-02242-z.
Lakra, K., and K. Avishek, 2022: A review on factors influencing fog formation, classification, forecasting, detection and impacts. Rend. Lincei Sci. Fis. Nat., 33, 319–353, https://doi.org/10.1007/s12210-022-01060-1.
Lawson, J. R., C. K. Potvin, P. S. Skinner, and A. E. Reinhart, 2021: The vice and virtue of increased horizontal resolution in ensemble forecasts of tornadic thunderstorms in low-CAPE, high-shear environments. Mon. Wea. Rev., 149, 921–944, https://doi.org/10.1175/MWR-D-20-0281.1.
Li, X., and Z. Pu, 2022: Turbulence effects on the formation of cold fog over complex terrain with large-eddy simulation. Geophys. Res. Lett., 49, e2022GL098792, https://doi.org/10.1029/2022GL098792.
Li, X., and Z. Pu, 2024: Effects of surface moisture flux on the formation and evolution of cold fog over complex terrain with large-eddy simulation. Quart. J. Roy. Meteor. Soc., 150, 3013–3027, https://doi.org/10.1002/qj.4748.
Li, Z., D. Liu, W. Yan, H. Wang, C. Zhu, Y. Zhu, and F. Zu, 2019: Dense fog burst reinforcement over eastern China: A review. Atmos. Res., 230, 104639, https://doi.org/10.1016/j.atmosres.2019.104639.
Lynn, B. H., S. Cohen, L. Druyan, A. S. Phillips, D. Shea, H.-Z. Krugliak, and A. P. Khain, 2020: An examination of the impact of grid spacing on WRF simulations of wintertime precipitation in the mid-Atlantic United States. Wea. Forecasting, 35, 2317–2343, https://doi.org/10.1175/WAF-D-19-0154.1.
Madala, S., A. N. V. Satyanarayana, C. V. Srinivas, and M. Kumar, 2015: Mesoscale atmospheric flow-field simulations for air quality modeling over complex terrain region of Ranchi in eastern India using WRF. Atmos. Environ., 107, 315–328, https://doi.org/10.1016/j.atmosenv.2015.02.059.
Maronga, B., and F. C. Bosveld, 2017: Key parameters for the life cycle of nocturnal radiation fog: A comprehensive large-eddy simulation study. Quart. J. Roy. Meteor. Soc., 143, 2463–2480, https://doi.org/10.1002/qj.3100.
Mohan, M., and M. Gupta, 2018: Sensitivity of PBL parameterizations on PM10 and ozone simulation using chemical transport model WRF-Chem over a sub-tropical urban airshed in India. Atmos. Environ., 185, 53–63, https://doi.org/10.1016/j.atmosenv.2018.04.054.
Nakanishi, M., 2000: Large-eddy simulation of radiation fog. Bound.-Layer Meteor., 94, 461–493, https://doi.org/10.1023/A:1002490423389.
Nakanishi, M., 2001: Improvement of the Mellor–Yamada turbulence closure model based on large-eddy simulation data. Bound.-Layer Meteor., 99, 349–378, https://doi.org/10.1023/A:1018915827400.
Nakanishi, M., and H. Niino, 2006: An improved Mellor–Yamada level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397–407, https://doi.org/10.1007/s10546-005-9030-8.
Philip, A., T. Bergot, Y. Bouteloup, and F. Bouyssel, 2016: The impact of vertical resolution on fog forecasting in the kilometric-scale model AROME: A case study and statistics. Wea. Forecasting, 31, 1655–1671, https://doi.org/10.1175/WAF-D-16-0074.1.
Pu, Z., C. N. Chachere, S. W. Hoch, E. Pardyjak, and I. Gultepe, 2016: Numerical prediction of cold season fog events over complex terrain: The performance of the WRF model during MATERHORN-Fog and early evaluation. Pure Appl. Geophys., 173, 3165–3186, https://doi.org/10.1007/s00024-016-1375-z.
Pu, Z., and Coauthors, 2023: Cold fog amongst complex terrain. Bull. Amer. Meteor. Soc., 104, E2030–E2052, https://doi.org/10.1175/BAMS-D-22-0030.1.
Román-Cascón, C., C. Yagüe, M. Sastre, G. Maqueda, F. Salamanca, and S. Viana, 2012: Observations and WRF simulations of fog events at the Spanish Northern Plateau. Adv. Sci. Res., 8, 11–18, https://doi.org/10.5194/asr-8-11-2012.
Sathyanadh, A., T. V. Prabha, B. Balaji, E. A. Resmi, and A. Karipot, 2017: Evaluation of WRF PBL parameterization schemes against direct observations during a dry event over the Ganges valley. Atmos. Res., 193, 125–141, https://doi.org/10.1016/j.atmosres.2017.02.016.
Schwartz, C. S., and R. A. Sobash, 2019: Revisiting sensitivity to horizontal grid spacing in convection-allowing models over the central and eastern United States. Mon. Wea. Rev., 147, 4411–4435, https://doi.org/10.1175/MWR-D-19-0115.1.
Shi, C., J. Yang, M. Qiu, H. Zhang, S. Zhang, and Z. Li, 2010: Analysis of an extremely dense regional fog event in Eastern China using a mesoscale model. Atmos. Res., 95, 428–440, https://doi.org/10.1016/j.atmosres.2009.11.006.
Shi, H.-Y., H.-F. Wang, L.-L. Qi, and J. Bai, 2005: Numerical simulation of radiation fog event in Yangtze River (in Chinese). J. PLA Univ. Sci. Technol., 6, 404–408.
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, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., https://doi.org/10.5065/D68S4MVH.
Smith, D. K. E., I. A. Renfrew, S. R. Dorling, J. D. Price, and I. A. Boutle, 2021: Sub-km scale numerical weather prediction model simulations of radiation fog. Quart. J. Roy. Meteor. Soc., 147, 746–763, https://doi.org/10.1002/qj.3943.
Steeneveld, G.-J., and M. de Bode, 2018: Unravelling the relative roles of physical processes in modelling the life cycle of a warm radiation fog. Quart. J. Roy. Meteor. Soc., 144, 1539–1554, https://doi.org/10.1002/qj.3300.
Steeneveld, G.-J., R. J. Ronda, and A. A. M. Holtslag, 2015: The challenge of forecasting the onset and development of radiation fog using mesoscale atmospheric models. Bound.-Layer Meteor., 154, 265–289, https://doi.org/10.1007/s10546-014-9973-8.
Sun, Y., Z. Ma, T. Niu, R. Y. Fu, and J. F. Hu, 2013: Characteristics of climate change with respect to fog days and haze days in China in the past 40 years (in Chinese). Climatic Environ. Res., 18, 397–406.
Tardif, R., 2007: The impact of vertical resolution in the explicit numerical forecasting of radiation fog: A case study. Pure Appl. Geophys., 164, 1221–1240, https://doi.org/10.1007/s00024-007-0216-5.
Tian, M., B. Wu, H. Huang, H. Zhang, W. Zhang, and Z. Wang, 2019: Impact of water vapor transfer on a Circum-Bohai-Sea heavy fog: Observation and numerical simulation. Atmos. Res., 229, 1–22, https://doi.org/10.1016/j.atmosres.2019.06.008.
Wang, R., Z. Qiang, Y. Ping, and H. Qian, 2020: Summary and prospects of numerical simulation research of the atmospheric boundary layer (in Chinese). Adv. Earth Sci., 35, 331–349.
Wei, W., X. Peng, Y. Lin, G. Zhang, Y. Yang, and J. Long , 2022: Extension and evaluation of University of Washington moist turbulence scheme to gray-zone scales. J. Adv. Model. Earth Syst., 14, e2021MS002978, https://doi.org/10.1029/2021MS002978.
Weisman, M. L., K. W. Manning, R. A. Sobash, and C. S. Schwartz, 2023: Simulations of severe convective systems using 1- versus 3-km grid spacing. Wea. Forecasting, 38, 401–423, https://doi.org/10.1175/WAF-D-22-0112.1.
Willmott, C. J., 1982: Some comments on the evaluation of model performance. Bull. Amer. Meteor. Soc., 63, 1309–1313, https://doi.org/10.1175/1520-0477(1982)063<1309:SCOTEO>2.0.CO;2.
Wilson, T. H., and R. G. Fovell, 2018: Modeling the evolution and life cycle of radiative cold pools and fog. Wea. Forecasting, 33, 203–220, https://doi.org/10.1175/WAF-D-17-0109.1.
WMO, 2008: Overview of tools and methods for meteorological and air pollution mesoscale model evaluation and user training. WMO Joint Rep. of COST Action 728 and GURME, 116 pp.
Xu, H., and Y. Wang, 2021: Sensitivity of fine-scale structure in tropical cyclone boundary layer to model horizontal resolution at sub-kilometer grid spacing. Front. Earth Sci., 9, 707274, https://doi.org/10.3389/feart.2021.707274.
Ye, X., B. Wu, and H. Zhang, 2015: The turbulent structure and transport in fog layers observed over the Tianjin area. Atmos. Res., 153, 217–234, https://doi.org/10.1016/j.atmosres.2014.08.003.
Zhou, B., J. Du, I. Gultepe, and G. Dimego, 2012: Forecast of low visibility and fog from NCEP: Current status and efforts. Pure Appl. Geophys., 169, 895–909, https://doi.org/10.1007/s00024-011-0327-x.
Zhou, B. W., J. S. Simon, and F. K. Chow, 2014: The convective boundary layer in the terra incognita. J. Atmos. Sci., 71, 2545–2563, https://doi.org/10.1175/JAS-D-13-0356.1.
