Influence of Environmental Conditions on Forecasting of an Advection-Radiation Fog : A Case Study from the Casablanca Region , Morocco

In this study, a dense advection-radiation fog that occurred over the Grand Casablanca region, Morocco, during the night of 23–24 December 2013, is investigated. The adverse visibility induced by the fog patch led to a series of collisions and loss of life on a highway of the region. This fog event is simulated by the Meso-NH research model. Conventional observations from two synoptic stations, satellite imagery, and the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis are used to analyze the physical processes during the whole life cycle of the event. Some hypotheses on the influence of environmental conditions (topography, land-sea heterogeneity, urbanization) on the numerical fog prediction are presented. Observational analysis together with numerical results show that the horizontal moisture transport, linked to a northerly wind during the afternoon, drove the onset of this fog event. The formation stage was governed by weak turbulence and nocturnal radiative cooling at both stations (costal station, GMMC, and inland station, GMMN). Horizontal moisture flux convergence was observed at the top of the fog layer, over the coastal station, during the mature phase, in addition to the radiative-turbulent interactions. The latter was the main mechanism leading to fog thickening at the inland station. The dissipation phase was governed by a zone of horizontal moisture flux divergence linked to a dry wind from the south. Sensitivity experiments show that numerical coastal fog prediction is influenced by local topography, urbanization and aerosol types, but less by land cover.

Different fog types are classified on the basis of how they form or where they occur (George, 1951).Over continental stations, the main types of fog are radiation fog and cloud base lowering fog (Roquelaure et al., 2009).However, not all types of fog can occur everywhere.The classification proposed by George (1951) takes an advectionradiation fog type into account.This fog type is common in some coastal regions such as the Los Angeles Basin (Baars et al., 2003).It results from nighttime radiative cooling of moist air that has been advected over land from the ocean or from any large water body during the previous daylight hours (Ryznar, 1977).Bari et al. (2016) have performed a climatological study in the Grand Casablanca region, Morocco.They show that advection-radiation fog is the most common fog type over this region.However, numerical modeling is still needed for a deeper understanding of the interaction among physical processes during the whole life cycle of this fog type over this region.
A review article by Niu et al. (2010) focuses on fog over China.They mention that advection-radiation fog occurrence is not only due to the effect of land and sea breezes.It could also be associated with both advection and radiation processes.The studies mentioned above lack sufficient details on the dissipation phase of such a fog type, which occurs during the night, and on how environmental conditions (e.g., topography, urbanization and land-sea heterogeneity) influence its lifetime.
Successful numerical fog forecasting depends on multiple parameters.Thus, in order to improve numerical weather prediction (NWP) of fog, many efforts have been made to study the effects of initial conditions (e.g., Bari et al., 2015), and the use of high vertical (e.g., Tardif, 2007;Philip et al., 2016) and horizontal (e.g., Boutle et al., 2016) resolutions.
Regarding the impact of surface heterogeneities on fog forecasting, Roman-Cason et al. (2016) used a statistical method to investigate the prediction of radiation fog over two contrasting sites.However, no numerical study has been performed to elucidate the influence of surface heterogeneities on different fog types over a given region.Investigations of the 3D simulations that consider land-cover characteristics and surface heterogeneities on advection-radiation fog in real synoptic conditions are still rare.
This article aims to contribute to a better understanding of the different physical processes occurring during the life cycle of advection-radiation fog and to assess the influence of some environmental conditions on its whole life cycle over the Grand Casablanca (GCB) region.This region, located on the north-western seaboard of Morocco, is one of the fog-prone regions of the Atlantic coast of North Africa that has received little attention in the literature.It is characterized by a complex landscape (Fig. 1 Section 2, devoted to the methodology, describes the observational dataset and the 3D Meso-NH model used in this study.In Section 3, we analyze the case of the advectionradiation event studied.This analysis is based on conventional observations from two synoptic stations, an observed sounding at the coastal station, reanalysis gridded data and satellite imagery.The numerical simulation and sensitivity study results are then discussed in this section.Conclusions and perspectives are presented in the last section.

Observational Data
The fog event in question occurred over the GCB region from 2010 UTC on 23 December 2013 to 0510 UTC on 24 December.European Center for Medium Weather Forecast (ECMWF) reanalyses (Dee et al., 2011) of mean sea level pressure (MSLP) and sea surface temperature (SST), extracted every 6 h at a horizontal resolution of 0.125° × 0.125°, are used to characterize the synoptic weather conditions and land-sea thermal contrast before and during the lifetime of this event.Conventional observations (temperature, visibility, cloud base height and wind) from the two synoptic stations of the region (the costal station, GMMC, and the inland station, GMMN, Fig. 1) are used to investigate the local thermodynamic conditions.According to WMO recommendations, the wind speed value is rounded to the nearest integer, and thus has an accuracy of 1 m s -1 .In addition, a sounding at the coastal station, GMMC, at 0000 UTC on 24 December 2013 was also used to describe the evolution of the lower levels of the ABL at the station.Meteosat Second Generation (MSG) satellite imagery, at 3 km horizontal 'sampling distance' at the Sub-Satellite Point, is used to describe the spatial extent of the fog patch and its evolution during the life-time of the event.During nighttime, it is difficult to distinguish fog or low clouds from other cloud types by using a single infrared channel.Therefore, EUMETSAT (www.eumetsat.int)recommends a dual channel difference method based on the Brightness Temperature Difference (BTD) between the infrared channels at 3.9 µm (IR3.9) and 10.8 µm (IR10.8)(Cermak et al., 2007).This method is based on the fact that lowlevel water clouds observed in IR3.9 have lower emissivity than the same clouds observed in IR10.8.

Numerical Modeling: Model and Configuration
The Meso-NH model (Lafore et al., 1998) was used to perform the numerical simulations.It is a non-hydrostatic, fully compressible model, which can be used in a large variety of configurations (from Large-Eddy Simulations (LES) to synoptic scales).Mixing ratios of six types of hydrometeors (cloud, rain, snow, graupel, water vapor, and ice) are included as prognostic variables.It has been successfully used for LES studies of Neutral Boundary Layer (e.g., Cuxart and Jiménez, 2007) and LES studies of fog (e.g., Bergot, 2013).Meso-NH has also been used for high-resolution meso-scale studies (e.g., Martínez et al., 2010), in particular for fog simulation (e.g., Bari et al., 2015) and low-level cloud studies (e.g., Sandu et al., 2012).
The simulation was run in a one-way nesting mode between 1200 UTC on 23 December and 1200 UTC on 24 December with a coarse 5-km domain and a nested fine 1-km grid domain.The outer domain (MNH5) covered the northwestern part of Morocco while the inner domain (MNH1) covered the GCB region (Fig. 2(a)).The two domains had horizontal resolutions of 5 km and 1 km respectively and 60 vertical levels, stretched monotonically from the surface to 6390 m (the top of the model domain).In fact, fine vertical resolution is necessary to simulate fog events (Tardif, 2007).Thus, the vertical resolution comprised 4 levels below 10 m, 18 levels below 100 m and 39 levels below 1 km with the first level at 2 m.This kind of vertical resolution is currently used in operational mesoscale models (e.g. for the French AROME model, the first level is at 5 m).The data used to define Meso-NH's initial and lateral conditions came from ERA-Interim re-analysis of ECMWF.The lateral conditions used 6-hourly data with a horizontal resolution of 0.1° × 0.1° and 70 vertical hybrid levels.The SST reanalysis varied in time and space during the model integration.Identical physical options were applied to both domains.The characteristics of the configuration used for this case study are summarized in Table 1.

Observational Analysis of the Case Studied Synoptic Weather Patterns
Before fog onset, a high-pressure system over the northern Atlantic and a weak low trough extending northeasterly to the GCB region set up a northerly marine inflow to the region at 1200UTC on 23 December 2013 (Fig. 2(a)).This favored transport of moist air from the Atlantic Ocean during the afternoon.At the same time, a strong low-pressure system was located on the western seaboard of Europe.At 1800 UTC on 23 December, the trough weakened, causing a shallow pressure distribution area (Fig. 2(b)) to be set up between the two low-pressure systems in the north and the south of the GCB region and the high-pressure system over the Atlantic Ocean.This configuration induced a weak pressure gradient on the northwestern part of Morocco and favored the occurrence of light wind conditions over the region.This pattern is particularly conducive to radiation fog formation.As the systems moved east, the synoptic flow changed, becoming from the south at 0600 UTC on 24 December (Fig. 2(c)) during the dissipation phase of the fog event.

Local Observations and Reanalysis
Based on the conventional meteorological observations (c) at 0600UTC on 24 December.The gray square refers to the GCB region.The empty square refers to a domain that will be used in the numerical simulation.at the two synoptic stations, GMMC and GMMN (Fig. 3), the atmospheric conditions were partially cloudy (cloud base height above 6000 m).The mist (visibility between 1 km and 5 km) formed first at 1840 UTC on 23 December at the coastal station, GMMC, and occurred at the inland station, GMMN, one hour later (at 1940 UTC).Then, this mist quickly transformed into fog, at 2010 UTC at GMMN, and later at 2150 UTC at GMMC.The observations indicate that this fog event lasted about 9 hours over the GCB region.Before fog onset, the land-sea thermal contrast (T2m-SST) during the afternoon was positive and reached 2°C (at around 1400 UTC) before sunset (1730 UTC, Fig. 3).This configuration could induce a cross shore circulation from sea to land.The SST reanalysis, at the grid point nearest the coastal station, GMMC (33.6N-7.7W),at 1400 UTC on 23 December was 16.5°C while the observed 2 m temperatures at the two synoptic stations at the same moment were 16.9°C at GMMC and 18.4°C at GMMN.In addition, Fig. 3 shows that 10-m wind was from the northwest to north at GMMN and GMMC during the afternoon with speeds above 2 m s -1 .Thus, this atmospheric configuration indicates that moist air originating from the sea, during the daytime, was advected to GMMN and GMMC, yielding a distinct increase in the specific humidity near the coast.This moist advection remained trapped within the stable ABL during the night.Under such synoptic conditions, with light wind, the cooling reached 0.4 °C h -1 and 0.3 °C h -1 respectively at GMMC and GMMN one hour before the fog onset (Fig. 3).
This fog event was very dense at the two stations, with minimum of horizontal visibility of 100 m at GMMC and 200 m at GMMN.During the night, the wind direction changed to the south, and thus favored the advection of dry continental air.Therefore, the fog event dissipated first at GMMN (at 0227 UTC) and then at GMMC (at 0505 UTC).
Based on the observational analysis, the studied fog event had the characteristics of an advection-radiation fog type according to Ryznar's (1977) definition and also according to the fog event classification methodology applied by Bari et al. (2016) over the Casablanca region.The moist northerly wind in the lower atmosphere seems to have been the main moisture source for fog formation, while the southerly wind seems to have been one factor among others for its dissipation.

Fog Extension as Seen by Satellite Imagery
Satellite imagery can provide valuable information on the spatial extension of fog/low clouds.The Meteosat Second Generation (MSG) images derived following Cermak and Bendix's (2007) approach are displayed in Fig. 4.This figure shows that patchy fog/low clouds covered the northwestern seaboard of Morocco.The fog/low cloud layer was located on-shore and close to the coast during the formation phase (2100 UTC in Fig. 4(a)).During the mature phase (0000 UTC in Fig. 4(b)), the fog/low clouds layer was detected on-and off-shore but still close to coast.The fog/low cloud patch spread and moved over the sea during the dissipation phase (0600 UTC in Fig. 4(c)).
To study in detail the primary mechanisms leading to this fog formation and influencing its evolution, this analysis based on observations is complemented by a numerical modeling study.

Numerical Simulation of Physical Processes
To cover the complex physical processes involved in fog formation, the control experiment of the studied fog event was run over 1-km (MNH1) and 5-km (MNH5) resolution domains.Fig. 5 shows the simulated liquid water path To distinguish the simulated fog layer, the 2m-LWC from MNH1 is also plotted in Fig. 5 (onset stage, Fig. 5(c); dissipation phase, Fig. 5(f)).Comparing the LWP area over MNH1 and MNH5, it is seen clearly that the surfaces covered by the simulated fog/low cloud patch, over the inner domain, are quite similar except during the dissipation stage, where some slight discrepancies are observed near the coastal station, GMMC.In comparison with satellite imagery (Fig. 4), Fig. 5 shows the presence of the fog/low clouds patch at a location similar to that observed in MNH1.This analysis points out that processes leading to fog formation were already resolved at 5 km, and reveals that the mesoscale processes predominated during the life cycle of this fog event.To describe the primary mechanism leading to this fog onset, we focus on the results from MNH1.This result was also found for the coastal fog, simulated by Bari et al. (2015), that occurred during the night of 21-22 January 2008 over the GCB region.

Model Evaluation at the Two Meteorological Stations
An evaluation of the numerical simulations was conducted to examine the model result at the two synoptic stations (GMMC and GMMN).Fig. 6(a) shows time series of simulated 2-m temperature, 2-m relative humidity, and 10-m wind speed and direction at GMMC and GMMN.A comparison of Fig. 3 and Fig. 6(a) shows that the simulated thermodynamic parameters (temperature, relative humidity and wind) agree with the observed ones and the temporal evolution of these simulated parameters follows that of the observed ones well.However, some discrepancies can be seen in wind speed and relative humidity at the coastal station, GMMC.In fact, stronger wind than actually observed is simulated during the second half of the night and the model underestimates the relative humidity during the afternoon.layer, in both soundings, topped by an inversion at a height of about 60 m a.g.l.This inversion is strong, exhibiting temperature increases of 3.73°C in 40 meters for the simulated sounding versus 2°C in 57 meters for the observed one.The atmospheric air is dry above the inversion.A shallow moist zone exists within the neutral atmospheric boundary layer below the inversion.This zone is linked to the fog layer during its mature phase.Thus, it can be seen from Fig. 6(b) that the model reproduced the vertical atmospheric stratification structure over GMMC reasonably well, except that the simulated surface temperature was a little colder than the observed one.
In a general sense, Meso-NH's simulations were quite close to reality during the life cycle of this fog event, except for the coastal station GMMC, where the model simulated the fog onset 1 hour earlier than it was observed and dissipated it 2 hours too early.

Simulated Back Trajectories Analysis
The origin of the moisture leading to fog formation, was analyzed using the Lagrangian analysis tool of Gheusi and Stein (2002) based on Eulerian on-line passive tracers.In Gheusi and Stein's approach, the transport processes, comprising advection and transport by subgrid turbulent and convective motions, are calculated during the model integration at every time step and at the model grid points.Thus, the motions captured by the Eulerian tracers correspond exactly to the model dynamics.This analysis was performed for different heights in the low levels of the Atmospheric Boundary Layer (2 m, 10 m and 50 m) at the two synoptic stations GMMC and GMMN as end points with an ending time of 2100 UTC on 23 December.The 9h back-trajectories of air parcels (from 1200UTC to 2100UTC) ending at the aforementioned heights in GMMC and GMMN are plotted in Fig. 7(a).The air masses arriving at GMMN and GMMC were all from the Atlantic Ocean, confirming that moist air was transported into the fog region, during the afternoon of 23 December 2013.
To better understand how atmospheric properties are modified while being advected along the back-trajectories, the time series of the simulated air temperature, and the vapor mixing ratio are illustrated in Fig. 7(b) for 2 m as end point at GMMC and GMMN.The period covers the phase before fog onset, from 1200 UTC to 2100 UTC on 23 December 2013.
At the coastal station GMMC, Fig. 7(b) shows that the parcel was traveling over the sea during the afternoon until 1800 UTC, when it reached the continental part of the region.The increase in vapor mixing ratio (r v ) of the parcel during the afternoon (9.18 g kg -1 at 1200 UTC versus 9.56 g kg -1 at 1700 UTC) highlights that it gained moisture during its passage over the sea.In addition, the temperature of the parcel decreased progressively early in the night until the fog layer formed.At the inland station, GMMN (Fig. 7(b)), the backtrajectory analysis shows that the parcel was located at 120 m a.s.l over the sea at the beginning of the simulation.During its motion, it was fed with moisture, resulting in an increase of its vapor mixing ratio (Fig. 7(b)) by 0.23 g kg -1 in 4 hours (9.04 g kg -1 at 1200 UTC versus 9.27 g kg -1 at 1600 UTC).At 1700 UTC, the parcel reached the continental part of the coastal region at a height of 180 m a.s.l.Then the parcel reached the GMMN station, remaining humid but decreasing its temperature due to radiative processes until it became saturated at 2000 UTC (Fig. 7(b)).The trajectory of the parcel points out that it underwent upward movement during its displacement.This seems to have been due to the effect of the surrounding mountain slopes.

Physical Processes Interaction as Seen by Numerical Simulation
To show the interaction between the physical mechanisms, particularly advection, radiation and turbulence, during the whole fog life cycle, the time-height evolution of radiative cooling ((∂θ/∂t) rad , temporal rate of change of temperature due to radiation), horizontal moisture flux divergence ( .(qVh ) which combines the advection and convergence contributions), and turbulent kinetic energy (TKE) are displayed in Fig. 8.
At the inland station, GMMN,Fig. 6(a) shows that simulated 2 m temperature was warmer than the simulated sea surface temperature during the afternoon and that the wind blew from the northwest to north.This induced the horizontal moisture transport (Fig. 8(d)) during the daylight hours.At 1800 UTC, the atmospheric conditions were humid (80%, Fig. 6(a)).The nocturnal cooling reached 1.09 °C h -1 one hour before fog onset at 2000 UTC (Fig. 6(a)).This radiative cooling (Fig. 8(e)) led to air saturation and then condensation.Thus, the fog formed at GMMN had advectionradiation characteristics, as deduced from observations.Its formation stage was governed by the advection process before onset during the daytime and the combination of nocturnal radiative cooling and weak turbulence (Fig. 8(f)) at fog onset.During the mature phase, the wind changed direction and began to blow from the south (Fig. 6(a)) due to the evolving synoptic situation (Fig. 2) and then the fog patch moved to the sea (Fig. 11).Classically, radiative cooling at the ground ceased and the cooling spread upward to the fog-layer top (Fig. 8(e)).During the night, the coupling between radiative cooling at the fog top and vertical turbulence (Fig. 8(f)) allowed the radiating layer to propagate condensation upward.During the dissipation phase, a zone of horizontal moisture flux divergence is clearly visible at the top of the fog layer from 0200 UTC (Fig. 8(d)).In fact, the southerly flow (Fig. 6(a)), governed the fog dissipation by advecting dryer air from the south.
At the coastal station, GMMC, the horizontal moisture transport during the afternoon (Fig. 8(a)) associated with the northerly wind, induced moist atmospheric conditions at 1900 UTC (90% in Fig. 6(a)).The fog formed under the influence of light turbulence (Fig. 8(c slightly after sunset (1730 UTC) and rapidly during the hour leading to fog onset, when cooling reached 3.06 °C h -1 (Fig. 6(a)).During the mature phase, the fog layer was maintained by a zone of moisture flux convergence (Fig. 8(a)) located near its top.During the dissipation phase, a zone of moisture divergence (Fig. 8(a)) can be seen in the bottom of the fog layer, in association with dry air advected by the strong southerly simulated wind (Fig. 6(a)).
The numerical results analysis of the fog event revealed that it had the characteristics of an advection-radiation type at the two synoptic stations, as was deduced from observations.It showed that the moist northerly winds in the lower atmosphere were the main mechanism of fog formation, while the dry southerly wind was the main reason for its dissipation.
In addition, environmental conditions may influence the fog life cycle over heterogeneous areas (Gultepe et al., 2007;Tardif and Rasmussen, 2008).Numerical results showed that the simulated fog event extended spatially along the coast and did not enter far onshore in the eastern part of the region.Thus, a possible influence of the orography on the spatial extent of the fog layer is hypothesized.Furthermore, urbanization is noteworthy in the region, particularly in the northern part.This suggests that urbanization and land cover may influence the fog life cycle and could be the main cause of the discrepancies noted during the fog formation and dissipation phases at the coastal station in the north.On the other hand, the northern part of the Casablanca region has been subject to considerable activities of industrialization and more pollutants are being discharged in association with urbanization.Since we are studying a coastal region, the continental aerosols could move over the sea and the maritime aerosols could move over the land during the evolving synoptic and mesoscale circulations.This implies high uncertainty in the aerosol type near the coast, which affected the fog microstructure.All these hypotheses will be discussed through sensitivity experiments in Section 3.3 in order to better understand the influence of environmental conditions on this fog event structure.

Numerical Experiments on Sensitivity to Environmental Conditions
To test the sensitivity of the numerical coastal fog prediction to heterogeneities of the surface and to the type of aerosols in the microphysical parameterization scheme, a series of sensitivity experiments were performed.Table 2 lists the different numerical experiments.With this approach, we aim to provide more information for a better understanding of the factors affecting coastal fog forecasting with numerical weather models.
During the analysis of sensitivity experiments results, attention was paid to the liquid water content (LWC, g m -3 ) as a qualitative and quantitative parameter for the detection of fog characteristics.Fig. 9 (for the inland station, GMMN) and Fig. 10 (for the coastal station, GMMC) illustrate the change of LWC with time and height from 1200 UTC on 23 December 2013 to 0600 UTC on 24 December 2013, for the different experiments.Fig. 11 represents the spatial extent of the simulated fog layer during the mature phase (0000 UTC), provided by the sensitivity experiments.

Sensitivity to Surface Parameters
The topographical and physiographic features of the GCB region are shown in Fig. 1(b).This figure points out significant variations in land surface characteristics (urban, rural, orography, etc.).Urbanization is important in the region.Thus, three sensitivity experiments were performed to study the influence of the surface heterogeneities on the forecasting of the studied fog event.In the URBAN and BARESOIL experiments, the land cover was taken to be uniform over the inner domain MNH1.In the NORELIEF experiment, the orography was set to zero.
The experiments assessing sensitivity to heterogeneities of the surface show that URBAN simulates the fog formation around 3 hours later than REFEXP (Table 3), while it simulates the dissipation at exactly the same time as in REFEXP at the inland station and one hour later at the coastal station.This confirms that the physical processes governing the dissipation phase are not local but linked to dry air advection from the south.URBAN also simulates a fog resulting from stratus base lowering (Figs. 9 and 10) in contrast to the advection-radiation fog in REFEXP and observations.In fact, the simulated atmospheric conditions near the surface were slightly warmer than in REFEXP (not shown) at the beginning of the first part of the night.This delayed the triggering of the mechanisms responsible for the formation of fog.On the other hand, the vertical extent of the simulated event was diminished by 20 m at the inland station, GMMN, while it is increased by 70 m at the coastal station.Furthermore, the urban conditions had an inverse influence on the fog's LWC, since the fog's maximum LWC (0.4 g m -3 ) expanded much more as a function of time and height in the control run than in URBAN experiments at both stations.
Fig. 1(a) shows that GMMN and GMMC were located within environments dominated by grassland and shrubland respectively.Comparing the time-height distribution of LWC between REFEXP and BARESOIL, in which the  ), from REFEXP and sensitivity experiments (NORELIEF, BARESOIL, URBAN, KHKOM and KHKOC) at the coastal station GMMC.
land cover is assumed to be uniform bare soil, we note that BARESOIL simulates an earlier dissipation time than in REFEXP (Table 3) while it simulates the fog onset 1 hour later at the inland station.This discrepancy is due to the difference in heat capacity of the land cover between the two experiments.In fact, the bare soil had low heat capacity and heated up fast.This induced a temperature increase and consequently a decrease in relative humidity.
During the mature phase, the vertical extent of the simulated fog patch was diminished by about 30 m at both stations.
Regarding the orography effect on the fog spatial evolution, it was found that the absence of orography favored the extension of the fog layer inland far from the sea (Fig. 11).This result confirms the findings of Bari et al. (2015) on the impact of local topography in another case study of a coastal fog event over the same region (GCB).Thus, the simulated fog event lasts longer than in REFEXP and dissipates too late (Figs. 9 and 10).In NORELIEF, in contrast to the URBAN experiment, the fog's maximum LWC (0.45 g m -3 ) expands much more as a function of time and height than in the control run (maximum LWC of about 0.35 g m -3 ).

Sensitivity to Aerosol Properties
Fog characteristics, particularly supersaturation, depend strongly on aerosols.For instance, fog is more likely to appear in an environment with large concentrations of aerosols characterized by a low activation supersaturation (Gultepe et al., 2007).To assess the influence of the type of aerosols on the life cycle of the fog event under study, two sensitivity experiments were performed using the 2-moment warm microphysical parameterization KHKO (Khairoutdinov and Kogan, 2000), which considers droplet concentration N c and mixing ratio r c as prognostic variables for the fog.The activation spectrum coefficients were those used by Cohard et al. (2000) where the activation spectrum of cloud condensation nuclei (CCN) is prescribed: where S (%) is supersaturation and F(a, b, c, x) is the hypergeometric function (Press et al., 1992).C, k, µ and βare activation spectrum coefficients that can be tuned to represent various aerosol types.The simulations were performed using aerosol size distributions and properties typical for maritime air (KHKOM experiment, composed of large sized maritime aerosols: NaCl, C, k, µ and β were set to 2.1986.10 6cm -3 , 3.251, 2.589 and 621.689) and continental air (KHKOC experiment, composed of small sized continental aerosols: (NH 4 ) 2 S0 4 , C, k, µ and β were set to 1.735.10 3 cm -3 , 1.403, 0.834 and 25.499).
Experiments on sensitivity to the type of aerosols in the microphysical parameterization scheme showed that the highest value of the fog's LWC was lower in both KHKOC (0.15 g m -3 ) and KHKOM (0.15 g m -3 ) experiments than in REFEXP (0.45 g m -3 ) where the single-moment warm microphysical scheme was used.Comparing KHKOC and KHKOM results at the two stations, we notice that the time-height distributions of LWC at the inland station GMMN are somewhat similar while a great discrepancy is observed at the coastal station GMMC (LWC core of 0.05 g m -3 in KHKOM vs. 0.15 g m -3 in KHKOC).In addition, KHKOC simulates the fog dissipation at the coastal station, GMMC, at the same time as in REFEXP, while KHKOM simulates it one hour earlier.The maritime droplets are larger than the continental ones.This produces more settling and consequently lower LWC.In KHKOC, the small droplets produce less gravitational settling and thus higher LWC.
To conclude, Fig. 11 shows that in the sensitivity experiments the spatial extent of the simulated fog areas over the continental part of the Casablanca region (inner domain) during the mature phase (24 Dec 2013 at 0000 UTC) differs from the control experiment (REFEXP), except for the BARESOIL experiment.

CONCLUSIONS
This paper presents an observational analysis and a numerical simulation, using the Meso-NH model, of an advection-radiation fog event that occurred during winter over the coastal region of Casablanca on the northwestern seaboard of Morocco.The main goal of this case study was to better understand the different physical processes occurring in advection-radiation fog and to assess the influence of surface heterogeneities and aerosol types on its numerical prediction.To achieve this, conventional meteorological parameters from two synoptic stations (coastal station, GMMC, and inland station, GMMN) were used, together with a low resolution radio-sounding available at the coastal station, MSG satellite imagery and ECMWF ERA-Interim reanalysis.  ) from MNH5 during the mature phase (24 Dec 2013 at 0000 UTC), issued from REFEXP and sensitivity experiments (NORELIEF, BARESOIL, URBAN, KHKOM and KHKOC).Filled triangle and square indicate locations of GMMN and GMMC, respectively.Square refers to MNH1.
Results from the observational analysis show that the studied event has the characteristics of advection-radiation fog type.The moist northerly wind was the main moisture source for fog formation, while the southerly dry wind was the predominant factor for its dissipation.
Furthermore, numerical modeling using Meso-NH captured the main features of this fog event well over the continental zone of this coastal region, with slight discrepancies on onset and dissipation times at the coastal station.Meso-NH reproduced the vertical structure of the Atmospheric Boundary Layer at the coastal station, GMMC, well in comparison with the observed sounding at 0000 UTC.
Regarding the main factors leading to fog onset and dissipation, numerical results confirm all the findings of the observational analysis with more details about the interaction between the physical processes during the whole life cycle of the studied fog event.The onset stage was governed by weak turbulence and nocturnal radiative cooling at both stations while the mature phase and dissipation phase were slightly different.In fact, over the coastal station, GMMC, a zone of moisture flux convergence at the fog layer top, in addition to the coupling between the radiative cooling at the fog top and the vertical mixing, helped to maintain the fog layer.This coupling was also the main mechanism leading to the fog thickening at the inland station, GMMN.However, the dissipation phase at GMMN was governed by a zone of moisture divergence at the top of the fog layer, linked to the advection of dry air from the south, while a similar zone was located in the bottom of the fog layer over GMMC.
Sensitivity experiments to surface heterogeneities demonstrated that numerical coastal fog prediction was influenced by local topography and urbanization for this case study, but less by land cover.On the other hand, sensitivity experiments to aerosols pointed out that their impact was higher near the coast.It was found that the displacement of maritime aerosols over land had little impact on the fog patch at the inland station, while the continental aerosol potentially influenced the life time of fog near the coast.In fact, continental droplets are smaller than the maritime ones.This produces less gravitational settling and consequently higher liquid water content.
This paper was limited to the study of an interesting case of advection-radiation fog in real synoptic conditions in order to help to fill some gaps in our partial knowledge of the life cycle of advection-radiation fog over heterogeneous coastal areas.Further investigation is needed with emphasis on how fog forecasts can be improved over coastal regions, particularly over the Grand Casablanca region in Morocco.
(a)) with a nearly straight coastline and significant variations in land surface characteristics (urban, suburban, and rural areas), along with high levels of pollution and low mountains in the eastern part of the region (Fig. 1(b)).

Fig. 1 .
Fig. 1.(a) Map showing the land cover of the GCB region, with the locations of the two synoptic stations used in this study, (b) Map showing the topographical (m) and physiographic features of the region of interest.

Fig. 2 .
Fig. 2. Mean sea level pressure spatial distribution at (a) 1200UTC on 23 December, (b) at 1800UTC on 23 December, and(c) at 0600UTC on 24 December.The gray square refers to the GCB region.The empty square refers to a domain that will be used in the numerical simulation.

Fig. 3 .
Fig. 3. Time series of conventional meteorological parameters: (top to bottom) cloud height base (m a.g.l), horizontal visibility (km), 10 m wind direction (1/10°), 10 m wind speed (m s -1 ) and 2 m temperature (°C), for GMMC (dashed line) and GMMN (solid line) over the period from 0000UTC on 23 December to 1200UTC on 24 December.The vertical bars refer to the observed onset and dissipation times of the fog event at GMMC (dashed line) and GMMN (solid line).The horizontal lines in the visibility plot refer to the fog (1 km, long-dashed line) and mist (5 km, dot-dashed line) thresholds.Sunrise and sunset times are shown.

Fig. 5 .
Fig. 5.The liquid water path in the lowest kilometer (1-km LWP) from MNH5 (left) and MNH1 (middle), and liquid water content at 2 m from MNH1 (right) during: formation phase (2100UTC, top row) and dissipation phase (0600UTC, bottom row).Filled triangle and square refer to GMMN and GMMC respectively.Square in MNH5 refers to MNH1.

Fig. 6 .
Fig. 6.(a) Time series of the 1-km resolution model simulated 2-m temperature (°C), wind speed (m s -1 ), wind direction (°), and 2-m relative humidity (%) at the coastal station GMMC (circles, dashed line) and at the inland station GMMN (triangles, solid line).Simulated SST (stars) is superimposed on temperatures plot.The period covers from 1200 UTC on 23 December to 0600 UTC on 24 December.The vertical bars refer to the onset and dissipation times of the simulated fog event at GMMC (dashed line) and GMMN (solid line).(b) Simulated sounding superimposed on observed sounding at 0000 UTC on 24 December 2013 at the coastal station GMMC: temperature (circle for observation and solid line with x for simulation) and dew-point temperature (+ for observation and dashed line with * for simulation).

Fig. 7 .
Fig. 7. (a) The 9 h back trajectories of air parcels ending at GMMC and GMMN, on 23 December 2013 at 2100 UTC, at the heights above ground level of 2 m (solid line), 10 m (dashed line) and 50 m (dotted line).(b) Time series of simulated water vapor mixing ratio (g kg -1 ), air temperature (°C) superimposed on dew point temperature (°C) and height (m a.s.l) along the back-trajectory with ending point at 2 m at GMMC (solid line with star) located at 60 m a.s.l and GMMN (dashed line with circle) located at 200 m a.s.l on 23 December at 2100 UTC.The vertical line corresponds to the time when the parcel reached the continental part of the coastal region (dashed line for GMMN and solid line for GMMC).

Fig. 11 .
Fig. 11.The simulated 10 m wind superimposed on liquid water content at 2 m (g kg -1

Table 2 .
Summary of the sensitivity experiments.

Table 3 .
Onset and dissipation times from the control and sensitivity experiments.Uncertainty in hours between each sensitivity experiment and the control one is shown in parentheses.