The role of droplet sedimentation in the evolution of low level clouds over southern West Africa

Large eddy simulations are performed to investigate the influence of cloud microphysics on the evolution of low level clouds that form over southern West Africa during the monsoon season. We find that, even in clouds that are not precipitating, the size of cloud droplets has a non-negligible effect on liquid water path. This is explained through the effects of droplet sedimentation, which acts to remove liquid water from the entrainment zone close to cloud top, increasing liquid water path. Sedimentation also produces a more heterogeneous cloud structure and lowers cloud base height. Our results imply that an 5 appropriate parameterization of the effects of sedimentation is required to improve the representation of the diurnal cycle of the atmospheric boundary layer over southern West Africa in large-scale models. Copyright statement. TEXT


Introduction
During the months of June to September, the climate of southern West Africa (SWA) is dominated by the southwesterly flow of the West African Monsoon (WAM), which is principally driven by a north-south pressure gradient associated with the Saharan heat low and brings seasonal rains to the region (e.g. Sultan and Janicot, 2000;LeBarbé et al., 2002). Clouds, through their diabatic effects, are known to exert an influence on the WAM circulation. For example, a number of studies have explored the role of moist convection in the Sahel (e.g. Garcia-Carreras et al., 2013;Marsham et al., 2013;Birch et al., 2014), revealing that the diurnal cy-cle of latent heating and cloud radiative forcing affects the north-south pressure gradient and hence the northward advection of moisture.
Low-level clouds (LLCs) over SWA, with bases only a few hundred metres above ground level (a.g.l.), are also a common occurrence during the WAM season (e.g. Schrage and Fink, 2012;van der Linden et al., 2015), yet it is only recently that their role has been considered in detail. LLCs typically form near the Guinea coast sometime after sunset following the initiation of the southwesterly nocturnal low-level jet. The jet is linked to the low-level pressure gradient, supplying moisture to the Sahel region where it is mixed as a result of convection during the day (Parker et al., 2005;Lothon et al., 2008;Abdou et al., 2010;Bain et al., 2010). The clouds then spread northwards inland during the night (Schuster et al., 2013;van der Linden et al., 2015;Kalthoff et al., 2018), and typically persist until the late morning, after which they transition to broken cumulus and dissipate. Through their impact on surface solar irradiance, the LLCs play an important role in the evolution of the atmospheric boundary layer (BL) (Gounou et al., 2012) and the regional climate of West Africa (e.g. Knippertz et al., 2011;Hannak et al., 2017).
The most comprehensive observational study of the atmospheric BL over SWA was conducted recently by Kalthoff et al. (2018) during the DACCIWA field campaign (Knippertz et al., 2015aFlamant et al., 2017). Between 14 June and 30 July 2016, intensive observations were made at three ground sites -Savé (Benin), Kumasi (Ghana), and Ile-Ife (Nigeria) -using a variety of instrumentation including infrared cloud cameras, radiosondes and wind profil- Figure 1. Image from the 0.6 µ visible channel of the Meteosat-10 geostationary satellite, revealing the cloud structure over southern West Africa at 10:12 UTC on 5 July 2016. Borders and coastlines are highlighted, along with the locations of Savé and Lomé, labelled "A" and "B" respectively. Image obtained from http://catalogue.ceda.ac.uk/uuid/5fa2529b973e47ae38ab3557f2018ef4 (link valid as of July 2018). ers (Derrien et al., 2016), radars and ceilometers (Handwerker et al., 2016), and microwave radiometers (Wieser et al., 2016). These ground-based observations were complemented by in situ measurements of aerosol and cloud properties from three European aircraft, which together conducted 50 research flights between 27 June and 16 July. The results presented in Kalthoff et al. (2018) reveal significant variability in the onset and dissolution of LLCs over SWA from day to day and from site to site. However the governing processes and mechanisms responsible are not fully understood. Furthermore, large-scale models struggle to represent these LLCs and their variability accurately. Hannak et al. (2017) found that many current general circulation models (GCMs) suffer a common bias in the form of insufficient low cloud cover over SWA, abundant solar radiation, and thus too large a diurnal cycle in temperature and relative humidity. They concluded that targeted model sensitivity experiments are needed to test possible feedback mechanisms among low clouds, radiation, BL dynamics, precipitation, and the WAM circulation.
Several studies have proposed specific mechanisms relevant for the formation and break-up of the cloud decks (e.g. Schrage and Fink, 2012;Schuster et al., 2013;Adler et al., 2017). Specifically, LLCs are believed to be sensitive to temperature and moisture advection from the south (controlled by the strength of the low-level jet), vertical mixing of heat and moisture arising due to shear-generated turbulence, radiative cooling at cloud top, condensational heating, sub-cloud evaporation, orographic lifting, and lifting induced by gravity wave propagation. In addition to each of these processes, it is important to also consider the role of microphysics in the evolution of LLCs and the potential modification of the cloud properties via the interaction with aerosols (Knippertz et al., 2015a, b). The combination of ground-based and in situ measurements from DAC-CIWA offers a unique opportunity to explore the links among aerosols, microphysics, and the bulk cloud properties in SWA and to inform an appropriate level of parameterization for the representation of LLCs in regional and global models.
One particular microphysical process of interest is the role of droplet sedimentation -the gravitational settling of liquid droplets suspended within the cloud layer. Previous studies of non-drizzling marine stratocumulus have demonstrated that droplet sedimentation has a non-negligible impact on the evolution of liquid water path (LWP) (e.g. Ackerman et al., 2004;Bretherton et al., 2007). However the role of sedimentation in relation to low-level clouds over SWA has not yet been investigated in detail, despite the potential for changes in aerosol properties in the SWA region to modify the size distribution of cloud droplets (and in turn their sedimentation velocity). Hence the purpose of the present study is to perform large-eddy simulations of a selected DACCIWA case study to isolate the effects of droplet sedimentation and quantify its impact on the ability of the model to reproduce the observations. In doing so, the aim is to identify an optimum configuration for the parameterization of BL clouds over SWA. The rest of this paper is organised as follows. Section 2 presents details of the case study to be simulated; Sect. 3 describes the numerical model used to perform the simulations, along with details regarding model configuration and initialisation, and the results are presented in Sect. 4. Implications of the findings are discussed in Sect. 5, before the main conclusions are summarised in Sect. 6.

Case study
For the purpose of this study, we focus on a particular case on 4-5 July 2016, the seventh intensive observation period (IOP) from the DACCIWA field campaign . The satellite image in Fig. 1 reveals the extent of the cloud coverage over SWA at 10:12 UTC on 5 July 2016. The conditions were fairly typical of the campaign as a whole, with observations collected at the ground site at Savé (labelled "A" in Fig. 1) revealing the onset of the low-level jet around 18:00 UTC on 4 July, followed by the formation of a low-level stratocumulus deck during the night. Cloud at the Savé ground site was first observed at 03:00 UTC on 5 July, which persisted until around mid-day local time, after which it began to break up into patchy cumulus (see Flamant et al., 2017, their Fig. 6). No precipitation was recorded at the Savé ground site for this case, consistent with the majority of days sampled during the campaign period. For an overview of the diurnal cycle of the atmospheric BL at Savé during DAC-CIWA, the reader is referred to Kalthoff et al. (2018).
The radiosonde data from IOP 7 provide more information on the structure and evolution of the BL on this day. Profiles of potential temperature and relative humidity from the 03:30 UTC sonde, launched approximately half an hour after the cloud was first detected at Savé, are shown in Fig. 2a, b. The relative humidity profile reveals a cloud layer approximately 200 m thick, with a cloud top height of 550 m capped by a temperature inversion of 1.5 K. The horizontal wind components, shown separately in Fig. 2c, d, reveal a low-level jet with a wind speed maximum at a height of 550 m a.g.l., and the cloud layer located directly beneath. Later sondes from 05:00, 06:28, 08:00, and 09:28 UTC (not shown) reveal that the cloud layer persisted throughout the morning, with the relative humidity occasionally peaking just below water saturation, suggesting the presence of some breaks in the cloud cover. This is consistent with images from the infrared camera at Savé (see Fig. 3), further analysis of which is presented in Dione et al. (2018) over the whole Figure 3. Images from the infrared cloud camera at Savé on 5 July 2016 (Derrien et al., 2016) for the times indicated. The images from the camera are coded in RGB colours (red, green, and blue), providing a qualitative estimate of cloud cover during the day and night. The image colour is dependent on the emissivity of the sky and consequently on the brightness temperature, such that red indicates warm and blue cold. campaign period. The low-level jet persisted until around 11:00 UTC (Fig. 4a), by which time the depth of the BL had increased to 1 km due to solar heating of the surface, resulting in lifting of the cloud deck (as shown in Fig. 2a, b). Three research aircraft were also deployed in sequence on this day, taking in situ measurements along the transect between Lomé (labelled "B" in Fig. 1) and Savé from 08:00 UTC through to 18:00 UTC in order to sample the microphysical evolution during the cloud life cycle .

Model description
To fulfill the needs of this study we utilise the Met Office NERC Cloud model (MONC; Brown et al., 2015). MONC is a rewrite of the original Met Office Large Eddy Model (LEM), which has been used extensively over the past 20 years to study cloud processes in a variety of regimes (e.g. Brown, 1999;Brown et al., 2002;Clark et al., 2005;Connolly et al., 2006Connolly et al., , 2013Marsham et al., 2006;Young et al., 2017). MONC offers several key advantages over the original LEM, including code optimisations, bug fixes, and a new solver that enables simulations to be performed with relatively large domain sizes without having to compromise on the model resolution.
Radiation is represented in MONC by the Suite of Community RAdiative Transfer codes based on Edwards and Slingo (SOCRATES; Edwards and Slingo, 1996), the same as that used in the Met Office Unified Model, specifically the Global Atmosphere Model 6.0 (Walters et al., 2017). SOCRATES is called on a 3 min time step, allowing the effects of longwave cloud top cooling and shortwave absorption within the cloud layer to be captured in the model.
Regarding the treatment of cloud processes, MONC is coupled to the CASIM (Cloud AeroSol Interactive Microphysics) module, a newly developed user-configurable multi-moment scheme that represents five hydrometeor species (cloud, rain, ice, snow, and graupel) and multi-mode aerosols. CASIM has already been used within the Met Office Unified Model to study aerosol-cloud interactions in different meteorological contexts, e.g. Grosvenor et al. (2017), Miltenberger et al. (2018), and Stevens et al. (2018). Further details on the specific configuration of CASIM used in the present study are given in Sect. 3.2.

Model initialisation and configuration
MONC is initialised using profiles of potential temperature, total water mass mixing ratio, and horizontal wind components, which are obtained from radiosondes launched from the Savé ground site. For IOP 7, we initialise the model using data from the 03:30 UTC radiosonde as shown in Fig. 2, interpolating the data onto the model grid with a vertical resolution of 10 m. Where the initial relative humidity profile is at water saturation (i.e. between 350 and 550 m in Fig. 2b), the cloud liquid water mass mixing ratio profile is calculated assuming an adiabatic cloud parcel ascent from cloud base to cloud top. The profile of total water mass mixing ratio is then calculated as the sum of the cloud liquid water and water vapour mass mixing ratios at each model level. During the first model time step, this supersaturated profile results in the immediate production of a cloud layer via condensation and at an early enough stage in its life cycle to study its subsequent evolution over a period of 7.5 h. The choice of the 03:30 UTC sonde for initialisation is justified since the aim of the present study is to focus on the role of microphysical factors that control the subsequent evolution of the LLC, rather than the meteorological factors that govern the onset of cloud formation.
Regarding the forcing of the wind field, the winds from 03:30 UTC are relaxed towards the u and v wind components from the 11:00 UTC radiosonde (as shown in Fig. 2c, d) over a period of 7.5 h. This allows the model to maintain the low-level jet throughout the simulation period. No forcing increments are applied to either the potential temperature field or the moisture fields; however a constant largescale divergence of 5 × 10 −6 s −1 is imposed throughout the model domain to produce a constant large-scale subsidence. According to the ERA-Interim reanalysis dataset (Dee et al., 2011), this value lies within the variability range over SWA during the time period of the DACCIWA field campaign.
Importantly, MONC is not coupled to an interactive land surface scheme in the present study and so to represent the effects of the surface, time-varying fluxes of sensible and latent heat are prescribed using surface measurements from the Savé ground site (Kohler et al., 2016). Fluxes from 5 July 2016 used to force the model are plotted in Fig. 4b for the simulation period indicated.
All the simulations presented in this paper use a domain size of 7.5 km × 7.5 km in the horizontal with a 30 m grid spacing and a vertical extent of 2 km with a 10 m spacing between vertical levels up to 1.5 km, increasing to a 20 m spacing between 1.5 km and 2 km. The top 500 m is a damping layer to prevent unwanted gravity waves from reflecting off the rigid model lid. The first 90 min of each simulation 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22  (between 03:30 and 05:00 UTC) are discarded to allow the model to spin up from the initial conditions, and periodic boundary conditions are used in all cases.

Details of model experiments
Here we introduce and describe two initial experiments, the results from which are analysed in the next section. The first MONC experiment with CASIM is configured for dual moment cloud and rain, while cold processes were not considered or required. Autoconversion and accretion are represented using the scheme of Khairoutdinov and Kogan (2000) and sedimentation of cloud droplets and rain is included. A saturation adjustment scheme is employed for condensation and evaporation of cloud droplets, while rain evaporation is based on the scheme used in the LEM (Gray et al., 2001). CASIM includes various options for aerosol activation and in this work we employ the scheme of Abdul-Razzak et al. (1998), with the aerosol specified as a single accumulation mode log-normal size distribution following the analysis of regional aerosol properties in Haslett et al. (in preparation, 2018). The aerosol mass and number fields are completely passive in this experiment (i.e. not influenced by cloud and rain processes) and are used only to determine the number of droplets activated. This experiment is henceforth referred to as CASIM_NO_PROC.
The second MONC experiment is identical to CASIM_NO_PROC, the only difference being that droplet sedimentation is turned off following the 90 min spin-up period to allow turbulence to develop within the BL. We refer to this experiment as CASIM_NO_SED. The rationale of this second experiment is to explore whether a simulation with sedimentation disabled is able to reproduce the observations for this case and therefore to reveal the extent to which sedimentation impacts the simulation.

Results
We begin with an initial inspection of results from the CASIM_NO_PROC experiment. Figure 5a shows a timeheight plot of the domain-average cloud mass mixing ratio for the period 05:00-11:00 UTC. The presence of a cloud layer is revealed with an initial mean cloud base around 350 m and a cloud top of 600 m. Following sunrise at 05:37 UTC, the surface fluxes of sensible and latent heat increase sharply from around 07:00 UTC as shown in Fig. 4b, resulting in a deeper BL and lifting of the cloud layer from around 08:00 UTC. The general trend in the time series of cloud base height is well captured by the model, as seen in the comparison against the ceilometer measurements from the Savé ground site (Fig. 5b). Cloud top longwave radiative cooling was found to be crucial for the development and maintenance of the cloud, through the generation of an overturning circulation within the cloud layer. Indeed, without any longwave cooling, the model was unable to sustain the cloud layer, resulting in complete dissipation by the end of the spin-up period. Figure 6 provides further information about the evolution of the mixing state of the simulated BL, in terms of pro-files of liquid water potential temperature, liquid water mixing ratio, and total water mixing ratio following the diagnostic analysis of Jones et al. (2011). Domain-average profiles from 05:30 UTC (Fig. 6a) reveal a predominantly wellmixed cloud-topped BL capped by a temperature inversion at 600 m a.g.l. A stable layer exists from the surface up to 150 m a.g.l., consistent with longwave cooling of the surface during the night, with a thin fog layer which dissipates by 06:30 UTC. By 11:00 UTC (Fig. 6b), the increase in surface fluxes produces a deeper, convective BL with an unstable layer at the surface. A well-mixed layer exists between 50 and 400 m a.g.l. in the sub-cloud region, with a hint of a second shallower well-mixed layer directly below the top of the BL, where the values of liquid water mixing ratio are largest. Thet a_liq (K) Tot al_wat er (g kg ) Liquid wat er (x10 g kg ) -1 -1 -1 -1 Figure 6. Vertical profiles of liquid water potential temperature (K; black lines), total water mass mixing ratio (g kg −1 ; blue lines), and liquid water mass mixing ratio (× 10 g kg −1 ; green lines) diagnosed at (a) 05:30 UTC and (b) 11:00 UTC from the CASIM_NO_PROC experiment. Equivalent plots for CASIM_NO_SED are shown in (c) and (d) respectively.
These layers are separated by a transition region between 400 and 900 m a.g.l., where the liquid water potential temperature gradually increases with height. Figure 7a and b show that the model captures the general deepening of the BL as seen in the observations, with a simulated BL height of 1.1 km by 11:00 UTC compared with 1 km in the corresponding radiosonde profile. The vertical structure of both potential temperature and relative humidity is also reasonably well captured by the model. Figure 8a compares the time series of simulated LWP from CASIM_NO_PROC with observations from the vertically pointing ground-based microwave radiometer at Savé (Wieser et al., 2016). Because the radiometer measurements represent the time evolution at a single location, care must be taken when evaluating the model against this dataset to account for the difference in spatial sampling. Hence in Fig. 8 we plot both the simulated LWP time series taken from the centre of the model domain diagnosed at 1 min intervals, together with the variability in LWP across the whole domain. The model simulates the evolution of LWP in a manner that is broadly consistent with the measurements, with the observations for the most part lying within ±2 standard deviations of the simulated LWP values. Peak local values of LWP also occur at approximately the correct time in the model, i.e. after 08:00 UTC when the surface fluxes have started to rise sharply. No precipitation was produced by the model during the simulation period, consistent with the measurements at Savé.
Having validated the ability of CASIM_NO_PROC to capture the key features of the observations, we now consider the impact of disabling sedimentation by analysing results from the CASIM_NO_SED experiment. Figure 8b shows that CASIM_NO_SED underestimates the variability in LWP before 07:30 UTC compared to both the observations and CASIM_NO_PROC. Maps comparing the spatial distribution of LWP within the model domain for both simulations ( Fig. 9) confirm that the cloud is much more spatially homogeneous in CASIM_NO_SED initially, resembling a largely featureless sheet of stratus as opposed to the more lumpy stratocumulus seen in CASIM_NO_PROC. A comparison of the time series of mean LWP in the domain (solid lines in Fig. 10) reveals that, although both simulations show a similar rise and fall pattern with a peak around mid-morning, there are still some notable differences despite neither simulation producing any precipitation. For instance, following completion of the spin-up phase at 05:00 UTC, the rate of LWP growth slows in CASIM_NO_SED relative to CASIM_NO_PROC such that by 07:00 UTC, CASIM_NO_PROC has the higher LWP. The peak LWP in CASIM_NO_SED occurs around the same time but persists for longer, before decreasing sharply around 10:00 UTC. The other difference between the two simulations is in the evolution of the domain mean cloud base height. Figure 5b shows that CASIM_NO_SED maintains an elevated cloud base height compared to CASIM_NO_PROC throughout the simulation period. Between 05:30 and 08:00 UTC, the mean cloud base height is 60 m higher in CASIM_NO_SED, increasing to an average of 140 m higher between 08:00 and 11:00 UTC. The link between droplet sedimentation and LWP has been explored previously by Bretherton et al. (2007) in the context of nocturnal non-drizzling marine stratocumulus layers in the subtropics. Sedimentation was found to ultimately increase LWP, caused by the removal of liquid water from the entrainment zone near cloud top. In turn this reduces the magnitudes of evaporative cooling and longwave radiative cooling, two processes which control the sinking of relatively dry air from the free troposphere into the cloud layer. Conversely, higher cloud condensation nuclei concentrations decrease the mean droplet size and fall speed, reducing sedimentation rates and thus making the cloud more susceptible to the effects of entrainment at the top of the BL. This results in a reduced LWP and a thinner cloud layer for more polluted conditions. We now conduct further analysis of the two MONC experiments to explore whether the results of the present study are consistent with the findings of Bretherton et al. (2007).
Returning to Fig. 10, following completion of the spinup phase at 05:00 UTC, both simulations have the same value of LWP. As mentioned earlier, the initial development of the cloud layer during the spin-up phase is strongly dependent on the mechanism of longwave radiative cooling. By 05:30 UTC, the lack of droplet sedimentation in CASIM_NO_SED means that this experiment is able to maintain a slightly higher liquid water content at cloud top relative to CASIM_NO_PROC, with a more sharply defined peak value (see Fig. 6c compared to Fig. 6a). Over the following 1.5 h, the larger liquid water content within the entrainment zone in CASIM_NO_SED promotes stronger evaporative cooling and longwave radiative cooling relative to CASIM_NO_PROC (see Fig. 11). This increases the downward heat flux at cloud top, reduces moisture fluxes, and reduces the circulation strength in the BL (Fig. 12). The result is a slower rate of LWP growth with time relative to CASIM_NO_PROC, such that by 07:00 UTC, CASIM_NO_PROC has the higher LWP. Thus in CASIM_NO_PROC, the removal of liquid water mass from cloud top due to droplet sedimentation effectively acts to shield the cloud layer to some extent from the effects of entrainment, allowing LWP to grow faster with time.
A closer inspection of Fig. 11 reveals more information about the relative roles of radiative cooling and evaporative cooling in the evolution of the cloud layer. In both simula- tions, it is clear that radiative cooling is the dominant process, with peak rates that are typically an order of magnitude larger than those produced by evaporation near cloud top. The absence of sedimentation in CASIM_NO_SED results in larger cooling rates associated with both processes. However, the increase in longwave cooling rates is relatively modestaround 37 % by 07:00 UTC -whereas evaporative cooling rates increase by a factor of 2 within the same time period. Thus in relative terms, the effect of sedimentation appears to have the largest impact on rates of evaporative cooling.
It is important to remember that the present study is over land and the simulation period extends into the daytime, as opposed to the nocturnal marine BL simulated by Bretherton et al. (2007). Thus it is no surprise that after 08:00 UTC, when the fluxes of sensible and latent heat dominate and the surface layer becomes unstable, the effect of sedimen-tation on LWP starts to break down. The convective vertical mixing associated with the prescribed sensible and latent heat fluxes coincides with the lifting of the cloud layer and a decrease in LWP, with a more rapid depletion evident in CASIM_NO_PROC (Fig. 10). This is consistent with stronger evaporative cooling during mixing associated with the higher LWP around 07:30 UTC.
It is interesting to consider what would happen to the evolution of LWP in the absence of surface-driven mixing. This is important because, although the mean LLC onset time at Savé is 03:00 UTC (around 3 h before sunrise), it is notably earlier at other ground sites (e.g. 00:00 UTC in Kumasi, and 21:00 UTC at Ile-Ife; Kalthoff et al., 2018). Assuming the sedimentation-entrainment feedback holds true, an earlier LLC onset would allow more time for the effects of sedimentation to impact LWP before sunrise. To  Condensation heating rate (K h ) -1 (d) Figure 11. Domain-average vertical profiles of (a) longwave radiative heating rate (K h −1 ) and (b) condensation heating rate (K h −1 ) calculated as temporal means between 05:00 and 05:30 UTC for CASIM_NO_PROC (blue) and CASIM_NO_SED (orange). Longwave and condensation heating rates for the period 06:30-07:00 UTC are shown in (c) and (d) respectively. maintained, CASIM_NO_PROC exhibits a higher LWP by around 33 % relative to CASIM_NO_SED by the end of the simulation period. Based on this analysis, we conclude that the response of the model is consistent with the reasoning of Bretherton et al. (2007).

Discussion
The numerical experiments performed in this study have shown that droplet sedimentation helps to promote a more heterogeneous cloud layer, with localised regions of both enhanced LWP and reduced LWP within the model domain relative to simulations without droplet sedimentation, whilst also lowering cloud base height. Whilst surface fluxes remain relatively small, in this case prior to 07:00 UTC, sedimentation also acts to increase the rate of mean LWP growth within the domain by buffering the cloud layer from the effects of cloud top entrainment driven by evaporative cooling and longwave radiative cooling.
Since droplet sedimentation rates are inversely proportional to number concentration, one would expect the effects of sedimentation on both LWP and cloud base height to become more prominent as cloud droplet number concentration (CDNC) reduces. In the case of CASIM_NO_PROC, predicted number concentrations lie in the range of 400-700 cm −3 at STP, which agrees well with in situ measurements with median values of around 500 cm −3 at STP (Jonathan Taylor, personal communication, 2018). In this section we perform some new experiments to explore the sensitivity to reducing CDNC. We introduce results from a new experiment, CASIM_200, which prescribes the initial CDNC to be 200 cm −3 . This new simulation produces excessive variability in the LWP field and cloud bases that are too low (Figs. 13 and 14 respectively), confirming our hypothesis. This was found to be the case even with autoconversion switched off. The depth of the BL in CASIM_200 is also too shallow by the end of the simulation period, by virtue of the effect of increased droplet size and excessive  sedimentation velocity on entrainment. However, mean LWP is slightly lower compared to CASIM_NO_PROC; this is because, around 08:30 UTC, cloud base becomes so low it touches the surface and liquid water is removed from the domain. At 200 cm −3 , this removal of liquid is predominantly due to gravitational settling of cloud droplets as opposed  to significant warm rain production. Further reductions in CDNC, down to 100 and 50 cm −3 , respectively, deplete the LWP even more as a result of an increase in autoconversion. These results, as summarised in Table 1, suggest that the effects of droplet size on cloud top entrainment rates should not be ignored when considering the diurnal cycle of LLCs in the region.
In light of this result, it is pertinent to consider the potential implications of changes in CDNC in terms of cloud radiative effects. Any elevation of CDNC within urban plumes will increase cloud optical depth in a manner that is proportional to CDNC 1/3 and LWP 2/3 for shallow clouds. However, the reduced sedimentation associated with the increased CDNC would increase cloud top entrainment and therefore reduce LWP. Hence any effect of increased optical thickness arising from enhanced aerosol concentrations will to some extent be offset by the sedimentation-entrainment feedback and is likely to lessen any first-order indirect effect.
Our findings also have implications for the diagnosis of aerosol-cloud interactions from satellite data. An adiabatic cloud profile is typically assumed when estimating cloud properties from satellites, but a relevant issue here is the extent to which the adiabatic assumption holds in these lowlevel clouds (Merk et al., 2016). Since satellites view cloud top, it is conceivable that the sedimentation-entrainment effect may well bias retrievals significantly.
An important caveat in our results is the prescription of surface fluxes in our simulations; there is no feedback among changes in cloud cover, LWP, and the land surface radiation budget. What happens after sunrise in reality is likely to be dependent on such feedbacks, which the present model configuration is not able to capture due to the lack of an interactive land surface scheme. Coupling of MONC to an interactive land surface scheme is needed to be able to comment fully on the impacts of droplet sedimentation and cloud optical depth on the diurnal cycle of these low-level clouds.
In this study, large-eddy simulations of low-level clouds over southern West Africa have been performed with a focus on establishing the sensitivity of the cloud evolution to the treatment of droplet sedimentation. The simulations are constrained and validated using the unprecedented suite of measurements collected during the DACCIWA field campaign in 2016.
Our results reveal that, even for non-precipitating clouds, the evolution of low-level clouds over southern West Africa is sensitive to the effects of droplet sedimentation, suggesting that this mechanism should not be neglected when performing large-scale simulations of the region. Sedimentation of droplets acts to remove liquid water from the entrainment zone near cloud top, reducing the magnitude of evaporative cooling and longwave radiative cooling during entrainment mixing. This increases the rate of growth of liquid water path during the night-time and early morning period. For the conditions of prescribed subsidence and surface fluxes, the simulation best able to reproduce the observations was the one that came closest to matching the observed droplet number concentrations. Ignoring droplet sedimentation completely reduced variability in liquid water path by around a factor of 2 during the early morning and also elevated the mean cloud base height by an additional 200 m by the end of the simulation period. Conversely, overestimating sedimentation rates, by virtue of reducing the droplet number concentration by a factor of 2 or more relative to observed values, caused cloud base to lower to the surface by 08:30 UTC and liquid water path variability to increase by around a factor of 2. Both these changes degraded the realism of the model simulation with respect to the available observations. In all cases, cloud top longwave radiative cooling during the night was found to be crucial for the formation and maintenance of the clouds.
The link between sedimentation and liquid water path has been noted previously in relation to nocturnal non-drizzling marine BL clouds. But the clouds considered in the present study form over land and persist into the daytime, which means that the effect of sedimentation can potentially play an important role in regulating the surface radiation budget, with consequences for the diurnal cycle of the BL in southern West Africa and possibly the circulation of the West African Monsoon. The results of our study suggest the possibility of a complex feedback chain involving aerosols, sedimentation, entrainment, liquid water path, and surface energy fluxes. We recommend as part of future work that the experiments performed in this study be repeated using an interactive land surface scheme to determine the extent to which the sensitivities shown are modified due to feedbacks between cloud cover and the surface heat flux budget.
Code and data availability. The observational data used in this paper can be accessed upon request at http://baobab.sedoo.fr/ Author contributions. CD was the primary author of the paper, and led the design of the numerical experiments and the subsequent analysis. AH provided essential development and technical support for the MONC model, and assisted in the experimental design. AH, HCoe and TC all contributed to the analysis of the model results and the writing of the paper.