Sensitivity of low-level clouds and precipitation to anthropogenic aerosol emission in southern West Africa: a DACCIWA case study

. During the West African summer monsoon, pollutants emitted in urbanized coastal areas modify cloud cover and precipitation patterns. The Dynamics-Aerosol-Chemistry-Cloud-Interactions in West Africa (DACCIWA) ﬁeld campaign provided numerous aircraft-based and ground-based observations, which are used here to evaluate two experiments made with the coupled WRF-CHIMERE model, integrating both the direct and indirect aerosol effect on meteorology. During one well-documented week (1-7 July 2016), the impacts of anthropogenic aerosols on the diurnal cycle of low-level clouds and pre- 5 cipitation are analyzed in detail using high and moderate intensity of anthropogenic emissions in the experiments. Over the continent and close to major anthropogenic emission sources, the breakup time of low-level clouds is delayed by one hour, and the daily precipitation rate decreased by 7.5 % with the enhanced anthropogenic emission experiment (with high aerosol load). Despite the small modiﬁcations on daily average of low-level cloud cover (+2.6 %) with high aerosol load compared to moderate, there is an increase by more than 20 % from 14:00 to 22:00 UTC on hourly average. Moreover, modiﬁcations of the 10 modeled low-level cloud and precipitation rate occur far from the major anthropogenic emission sources, to the south over the ocean and to the north up to 11 ◦ N. The present study adds evidence to recent ﬁndings that enhanced pollution levels in West Africa may reduce precipitation. reaching 13.2%, and for precipitation rate reaching -10.0%, which shows the importance of anthropogenic aerosol transported over the ocean.


Introduction
In southern West Africa (SWA), the population is rapidly increasing, driving up anthropogenic emissions (AE) (Liousse et al., 2014). During the West African monsoon (WAM) period (June-September), in addition to local pollution emissions, aerosols from two remote sources are being transported toward the Guinean coast, namely mineral dust from the north and biomass burning aerosol (BBA) from Central Africa (e.g. Flamant et al., 2018a). These different sources contribute to the high aerosol 5 load over the SWA and have deleterious human health impacts (Bauer et al., 2019).
Depending on their optical and chemical properties, aerosols influence the local meteorology through the direct (and semidirect) effect via the absorption and scattering of radiation, and through the indirect effect via hygroscopic aerosol enhancing cloud droplet number concentrations, thereby decreasing droplet sizes (e.g. Haywood and Boucher, 2000;Carslaw et al., 2010).
These effects account for modifications in the meteorology especially in the boundary layer, which in turn affect pollution 10 transport and washout (Baklanov et al., 2017). An adequate representation of these effects is critical for General Circulation Models (Fan et al., 2016;Seinfeld et al., 2016), especially over West Africa, where clouds in the lowermost troposphere are not well reproduced during the WAM season (Hannak et al., 2017).
The latest generation of Numerical Weather Prediction models integrates the aerosol effects on meteorology (Baklanov et al., 2014). Over Africa, Menut et al. (2019) have shown -with the online coupled WRF-CHIMERE model -that a decrease 15 of anthropogenic emissions along the Guinean coast imposes a northward shift of the monsoonal precipitation associated with an increase of the surface wind speed over arid areas in the Sahel, inducing more mineral dust emissions.
The aerosol transport has complex pathways during the WAM season (Menut et al., 2018). The large amount of biomass burning aerosols in the accumulation mode (i.e. cloud condensation nuclei) transported from Central Africa over the ocean toward the coast of the Guinea Gulf together with the humidity in the monsoon flow impacts cloud microphysics, reducing the 20 droplet effective radius already before reaching the coast Taylor et al., 2019). Upon reaching the highly urbanized regions, these air masses are further loaded with additional aerosols from anthropogenic emissions Denjean et al., 2020). Therefore, the diurnal cycle of winds, Low-Level Clouds (LLC) and precipitations over SWA are impacted by both biomass burning and anthropogenic aerosols .
Over continental SWA and during the monsoon period, LLC are formed during the night and break-up in the afternoon, 25 however the variability of the LLC cover is only partly understood, which is mainly due to the lack of measurements in this region (Knippertz et al., 2015a). One of the goals of the recent DACCIWA project (Dynamics-Aerosol-Chemistry-Cloud Interactions in West Africa) was to address this issue thereby using the important database collected during June and July 2016 (Knippertz et al., 2015b. The interactions between the nocturnal low-level jet and the LLC have been studied with an observational campaign deployed in SWA, including aircraft (Flamant et al., 2018b), radiosondes and three super-sites that had 30 been instrumented to monitor the LLC diurnal evolution (Kalthoff et al., 2018).
During the months of June and July, the maritime air from the Atlantic reaches Savè (Benin), one of the DACCIWA supersites about 185 km north of the coast, between 16:00 and 21:00 UTC associated with the apparition of the nocturnal low-level jet and the LLC Zouzoua et al., 2020). Between 22:00 UTC and 06:00 UTC, the LLC layer continues it 2 https://doi.org/10.5194/acp-2020-1306 Preprint. Discussion started: 15 April 2021 c Author(s) 2021. CC BY 4.0 License. formation, due to an increase of relative humidity (RH) caused by the advection of cold air Babić et al., 2019;Lohou et al., 2020). The persistence of LLC has a considerable impact on the energy balance at the Earth's surface .
In general, three phases of the wind in the lowest troposphere in SWA can be distinguished according to Adler et al. (2017) and Deetz et al. (2018a, b): (i) a phase of low wind speed from 09:00 to 15:00 UTC, referred as "daytime drying"; (ii) 5 an increase of meridional wind when convection decreases in the boundary layer from 16:00 to 02:00 UTC, referred to as "maritime inflow"; and (iii) a subsequent decrease of meridional wind from 03:00 to 08:00 UTC referred as "moist morning".
Moreover,  have shown that pollutants are accumulated along the coast during the "daytime drying" phase and transported inland during the "maritime Inflow" phase, which suggests that LLC and precipitation could be especially modified by anthropogenic aerosol during a specific phase.

10
This article aims at understanding the influence of the anthropogenic aerosol emissions on the diurnal cycle of LLC and precipitation in SWA. The DACCIWA campaign provides a unique chance to analyze and validate online coupled models.
This study presents numerical modeling experiments conducted with WRF-CHIMERE in combination with aerosol and cloud observational datasets from the DACCIWA campaign, which are described in Section 2. An analysis of the location and intensity of the modeled anthropogenic plumes is given in Section 3. The mean state of humidity and wind is presented in 15 Section 4. In Section 5, we analyze the modifications of LLC and precipitation by the anthropogenic aerosol emissions from the Gulf of Guinea coast to a few hundred of kilometers north, followed by their consequences at the regional scale in Section 6. Conclusions appear in Section 7.

Modelling set-up and evaluation datasets
In this section, we briefly present the coupling of the WRF (Weather Research and Forecast) model and the CHIMERE 20 chemistry-transport model (Sect. 2.1), and the two experiments conducted using two different anthropogenic emission scenarios (Sect. 2.2). We then detail the observational datasets acquired during the DACCIWA campaign used to evaluate the two experiments and analyze LLC and precipitation. (Sect. 2.3).

CHIMERE model and its coupling with WRF
The coupled WRF-CHIMERE model integrates the direct (and semi-direct) and indirect effects between CHIMERE and WRF 25 through exchanges via an external coupling software developed primarily for use in the climate community, namely OASIS3-MCT (Craig et al., 2017), at a 10 minutes time step. Three fields are sent to the radiative scheme of WRF to represent the direct effect: (i) aerosol optical depth, (ii) single scattering albedo, and (iii) asymmetry parameter (Briant et al., 2017). The indirect effect is taken into account thereby transferring four fields of CHIMERE to the microphysics scheme of WRF: (i) aerosol size distribution, (ii) bulk hygroscopicity of internally mixed aerosols, (iii) ice nuclei, and (iv) deliquesced aerosols (Tuccella 30 et al., 2019). Over West Africa, the high aerosol load makes the integration of the aerosol effects an important steps toward the understanding of feedbacks on the meteorology during the monsoon .
The WRF model is widely used for various meteorological studies at the regional scale (Powers et al., 2017). We use the model in its 3.7.1 version (Skamarock and Klemp, 2008), allowing non-hydrostatic motion to be calculated (Janjic, 2003). The surface layer scheme is the Carlson-Boland viscous sub-layer with the surface physics calculated by the 'Noah' land surface model (Ek et al., 2003). The planetary boundary layer (PBL) physics are calculated by the Yonsei University scheme (Hong et al., 2006).

5
Meteorological initial and boundary conditions are provided by the operational analyses produced by the US National Center for Environmental Prediction (operational analyses: ds083.3 dataset, DOI: https://10.5065/D65Q4T4Z). These fields are used to nudge hourly fields of pressure, temperature, humidity and wind in the WRF simulations, with spectral nudging, which has been evaluated for regional models by von Storch et al. (2000). In order to enable the PBL variability to be resolved by WRF, low frequency spectral nudging is used only above 850 hPa. This set-up has already been used over SWA by Deroubaix et al. For the coupled WRF-CHIMERE model, two schemes are mandatory in order to reproduce: (i) the direct effects, which are taken into account through exchanges between the radiative scheme and CHIMERE, the Rapid Radiative Transfer Model for General Circulation Models with the Monte-Carlo Independent Column Approximation method of random cloud overlap from Mlawer et al. (1997); and (ii) the indirect effects, which are accounted for via the aerosol-aware microphysics scheme 15 proposed by Thompson and Eidhammer (2014) taking into account the modifications in lifetime and reflectivity of clouds due to aerosols. CHIMERE (version 2017) is a regional chemistry-transport model, which is fully described in Menut et al. (2013) and Mailler et al. (2016) for its offline mode. Bessagnet et al. (2004) describe the calculation of gaseous species in the MELCHIOR-2 (reduced) scheme and the aerosol scheme, which takes into account species such as sulphate, nitrate, ammonium, primary 20 organic matter, black carbon, secondary organic aerosols, sea salt, dust and water. All aerosols are represented using ten bins, from 40 nm to 40 µm in diameter. Menut et al. (2016) have validated and analyzed aerosol speciation and size distribution as well as the chemical mechanisms used in the CHIMERE model over Europe and North Africa. Boundary conditions at a 6-hour time step are taken from the Copernicus Atmosphere Monitoring Service made by the European Centre for Medium-Range Weather Forecasts (Inness et al., 2019) for both two experiments, which contain the biomass burning aerosols coming from 25 Central Africa (and mineral dust aerosols from the Sahel-Sahara), thanks to assimilation of aerosol optical depth retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) in this dataset (Fig. A1).

Anthropogenic emission experiments
The simulated period runs from 1 to 7 July 2016 (with a spin-up period starting on 23 June), which is entirely included in the post-onset phase of the 2016 WAM defined from 22 June to 20 July 2016 by Knippertz et al. (2017). The 32 vertical levels of  The Hemispheric Transport of Air Pollution version 2 (HTAPv2) inventory provides the anthropogenic emission flux (Janssens-Maenhout et al., 2015). This dataset is relevant for the year 2010 and given that the population is rapidly increasing in this region, it should underestimate the 2016 emissions.
Therefore, two experiments are proposed (Table 1): (i) a simulation with HTAPv2 designed to underestimate anthropogenic emission (AE), which is called AE1, and (ii) a simulation designed to overestimate the AE obtained by multiplying the emis-5 sions of HTAPv2 by a factor of ten, called AE10. This factor may seem huge but is not unrealistic, as anthropogenic emissions are estimated to increase very rapidly as an "explosive growth" (Liousse et al., 2014).

Measurements in the framework of the DACCIWA project
We use three types of measurements of the DACCIWA campaign: (i) radiosonde measurements, (ii) aircraft measurements, Visible and Infra-Red Imager (SEVIRI) images provided by EuMetSat and NAScube, are presented.
(i) We have selected six locations of the radiosonde campaign: Lamto and Abidjan in Ivory Coast, Accra in Ghana, Cotonou, Savè and Parakou in Benin ( Fig. 1). There were four releases per day at around 00:00 UTC, 06:00 UTC, 12:00 UTC and 18:00 15 UTC, and in Savè additional radiosondes were launched every 1.5 to 3 hours during intensive observation periods (Kalthoff et al., 2018).  (iii) At Savè, one of the three super-sites of the DACCIWA project, meteorological measurements were performed. Savè is a village located downstream several large coastal cities, due to the prevailing low-level southwesterlies during the WAM season.
For Savè, priority is given to the CHM15k ceilometer data providing a robust value of the cloud base height Dione et al., 2019;Lohou et al., 2020).

Evaluation of modeled pollution plumes 5
In this section, we evaluate the spatial variability of the modeled anthropogenic aerosol concentration in the lowermost troposphere, with a particular focus given to the location and the magnitude of the modeled pollution plumes. We investigate four aerosol types (BC, OC, ammonium, nitrate), which are of prime importance due to their optical and hygroscopic properties (Carslaw et al., 2010). We first analyze the modeled aerosol concentration distributions with observations acquired below 2 km amsl (above mean sea level) using all available measurements (Table 2). Secondly, we study in detail a specific flight of the 10 ATR-42 conducted on 6 July (Fig. 2).  We compare the distribution of the aerosol concentrations (BC, OC, ammonium, nitrate) of the two experiments (AE1 and AE10) with measurements of the three aircraft below 2 km (and above 300 m amsl to avoid perturbation close to the airport areas). The number of 3-min averaged observations is about 200 during the studied period, which corresponds to about 10 hours of sampling. The modeled concentrations are interpolated along the flight positions with a triple interpolation (bi-linear horizontally, linear vertically, and linear between two time steps).

5
The observations show that the OC concentration is by far the highest (Q1[OC] > Q3[others]), followed by ammonium, BC, and then nitrate ( Table 2). The model reproduces the observed order of the aerosol concentration distribution (OC > ammonium > BC > nitrate). For all aerosol types, the observations are between AE1 and AE10. With AE10, NOx are multiplied by ten, but the nitrate and ammonium concentrations are multiplied by hundred from AE1 to AE10. This shows the non-linearity of NOx to nitrate conversion and also a possible misrepresentation in the CHIMERE model, as it has already been raised by 10 Petetin et al. (2016). The underestimation of BC and OC by the two experiments could be explained by low emission factors in anthropogenic emission inventories (such as HTAPv2), as shown using in-situ measurement made during the DACCIWA campaign (Keita et al., 2018). We conclude that the range of the modeled aerosol concentration is realistic, and that the AE10 experiments is closer to the observations than AE1.
We select one specific flight in order to evaluate the location of the anthropogenic sources. This flight was performed by the 15 ATR-42 on 6 July from Lomé in Togo (take-off at 07:09 UTC) to Abidjan in Ivory Coast (landing at 10:45 UTC), during the time of day when the boundary layer height over the continent starts to increase, while being typically below 1 km amsl in the morning (Kalthoff et al., 2018). This flight is particularly interesting because it was performed mostly below 1 km amsl, over coastal, forest and urban areas, and allowed measurements in the vicinity of three major population centers (Lomé, Accra and Abidjan), with all instruments operating normally (Fig. 2). 20 The modeled aerosol concentrations are compared to aircraft measurements together with modeled and observed concen- the aerosols concentration measurements in three parts, which correspond to: (i) advecting air mass (close to position P6), (ii) Abidjan plume (between position P5 and P6), and (ii) continental (between position P1 and P4). On average, the continental aerosol concentration was 0.33 µg.m −3 for BC, 2.91 µg.m −3 for OC, 0.21 µg.m −3 for ammonium, and 0.17 µg.m −3 for nitrate.
A clear enhancement was measured in the Abidjan plume compared to continental aerosol concentrations, while advecting air mass and continental aerosol concentrations were similar (except for OC higher over the continent due to biogenic sources).   Overall, the ATR-42 flight on 6 July shows that the location of the urban plumes is well modeled, thus we can infer that the main aerosol sources are also well located. The magnitude of the anthropogenic emission sources is underestimated for the AE1 experiment, and slightly overestimated for the AE10 experiment. In conclusion, the urban plumes of Accra and Abidjan are well reproduced by the model. The range of modeled aerosol concentrations is realistic for both simulations, although the simulation AE10 is closer to the observations. The two experiments allow investigating the effect of changes in anthropogenic aerosol emission change on meteorology, as discussed in the following.
4 Sensitivity of humidity and wind speed to an anthropogenic aerosol increase 5 We now want to evaluate the modeled meteorology and quantify the sensitivity to the aerosol load described by the two experiments. We select two major variables influencing the aerosol optical properties and transport, namely relative humidity (RH) and wind speed (WS). We evaluate firstly the averaged vertical profiles using radiosonde measurements, and secondly the variability of these two variables using aircraft measurements.
In the modeled domain, there are six radiosonde sites where balloons were frequently launched (c.f. Sect. 2.3), which 10 corresponds to 28 balloon launches in Abidjan, 28 in Accra, 23 in Cotonou, 43 in Savè, 7 in Lamto, and 28 in Parakou. The averages of the observed vertical profiles are presented for each site as well as their mean and standard deviation. We analyze the vertical biases (mod-obs) of the two experiments (Fig. 3). The observed mean vertical profiles are composed of three vertical parts: (i) from the surface to 1 km amsl, close to saturation (> 90 %), which is the monsoon layer, (ii) from 1-2 km amsl, RH close to 90 % and low wind speed, which is a directional from the coast (Lamto, Savè and Parakou). Thus, it is consistent to quantify the bias of the model for the three ranges of altitude 10 described above (Table 3). There is a clear dry bias in both experiments between 1.5 and 2.5 amsl, which is also present in the data of the Integrated Forecasting System of the European Centre for Medium-range Weather Forecasts van der Linden et al. (2020).
The bias is slightly reduced for RH with AE10 compared to AE1 in the three vertical layers, by 0.21 % between 0 and 1 km amsl, by 0.24 % between 1 and 2 km amsl and 0.05 % between 2 and 4 km amsl. This suggests an underestimation of vertical 15 mixing with a PBL that is too moist and a free troposphere above that is too dry. The bias is lower from the surface to 1 km amsl than in the above layers. For wind speed, the biases remain similar between AE1 and AE10. Nevertheless, we note that the bias on WS is reduced in Accra and Cotonou in the first kilometer when comparing AE10 to AE1 (Fig. 3).
Now, we want to analyze in detail the modeled and observed variability in the lowermost troposphere (< 4 km). We compare 3-min averages of aircraft measurements with the model interpolated to the flight trajectory (with the same methodology as in 20 Sect. 3), using scatter plots (Fig. 4) together with reduced major axis regression fits (Ayers, 2001;Smith, 2009).
While the observed RH may exceed 100 % at times, the modeled values are necessarily lower than 100 % (Fig. 4)   In conclusion, the vertical mean profiles and the range of variability of relative humidity and wind speed are well represented below 4 km amsl in the two experiments. There is no clear improvement from AE1 to AE10 for relative humidity and wind speed because differences in the correlation coefficients are less than 0.1 between the two experiments. The mean state of the lowermost troposphere is only marginally modified despite the important difference in the aerosol load between the two experiments. We did not expect an important change in the relative humidity and wind speed by the introduction of the aerosol 5 effects in the coupled model. However, we expect LLC and precipitation to be more influenced by the large difference in anthropogenic aerosol load between the two experiments.

Sensitivity of low-level clouds to an anthropogenic aerosol increase
This section aims at evaluating the sensitivity of the modeled LLC diurnal cycle to the anthropogenic aerosol load differences between the two experiments. We first focus on 5 July 2016 of the DACCIWA campaign, because the three aircraft were In the morning between Lomé and Savè, the ATR-42 measured an homogenous LLC cover between 0.5 and 2 km amsl. The LWMR measured in clouds ranges from 10 −5 to 10 −3 (kg/kg). LLC are vertically located near the modeled PBL top, ranging 20 between 700 and 1000 m amsl. The model predicts a higher cloud cover for AE10 than AE1, which seems to match better with the observations. The modeled LWMR ranges from 10 −5 to 5x10 −4 (kg/kg), thus its maximum is underestimated compared to observations. Nevertheless, the spatial variability as well as the altitude of LLC are in good agreement with the observations for both experiments.
At noon, as measured by the Falcon-20 instrumentation (Voigt et al., 2010), the modeled PBL top is higher (ranging between 25 700 and 1700 m amsl) than for the ATR-42 flight and fewer clouds are observed, which is reproduced by the model. The model underestimates the LLC cover, which seems to be lower over the continent for the AE10 experiment. When the Falcon-20 flew over the ocean towards the coast (green bar on Fig. 6 corresponding to the picture of the cockpit from a camera, c.f. Fig. A2), a clear front of LLC is facing the aircraft a few hundred meters behind the shore. The aircraft sampled the lower part of the modeled front of LLC (about 5x10 −5 kg/kg) with its cloud instrumentation (Voigt et al., 2011;Kleine et al., 2018). The model  difference between the two emission scenarios (AE10 and AE1). In conclusion, the variability of the LWMR is close for the two experiments, and in good agreement with the aircraft measurements.
The spatial analysis of the 5 July flights is completed by a temporal analysis of the LLC altitude over the studied week at the site of Savè. We use the data of Cloud Base Height (CBH) estimated by the Ceilometer based on a manufactured software (Handwerker et al., 2016) to compute an hourly average, and we compare it to the modeled vertical distribution of LWMR 5 interpolated at the Savè location for the two experiments (Fig. 7).
In Savè, there is a clear diurnal cycle of LLC because they form at low altitude (few hundred meters) during the night, get elevated by the increase of the PBL height until noon, and break up between 12:00 and 18:00 UTC Zouzoua et al., 2020). We note that in the morning, the clouds are always present, mostly with CBH below 500 m amsl. From 06:00 and 12:00 UTC, the CBH rises up to 1 km amsl. From 12:00 to 18:00 UTC, the variability (shown by the standard deviation,  vertical bars in Fig. 7) of CBH increases, because LLC break up and dissipate around this time Zouzoua et al., 2020). From 18:00 to 00:00 UTC, less LLC are observed.
We compare the CBH to the modeled LWMR to analyze the LLC formation, elevation and breakup time. The modeled LWMR for the two experiments reproduces the LLC diurnal cycle depicted above, since we can see the bottom of the modeled clouds, which correspond to LWMR ranging from 10 −5 to 5x10 −4 kg/kg, moving vertically from below 500 m to 1000 m amsl 5 (in blue in Fig. 7). The observed CBH during the morning is in good agreement with the height of the PBL where the modeled cloud base is located. The duration of the LLC in the afternoon lasts longer in AE10 compared to AE1. Every day from 12:00 and 00:00 UTC, we note that the LWMR is slightly higher for the AE10 than AE1.  Both experiments reproduce a realistic temporal variability of LWMR in the studied region. The main differences between the two experiments occur between 12:00 and 24:00 UTC during the LLC breakup time. In the AE10 experiment, LLC lasts longer. However, these differences are small. In order to understand the differences between the two experiments, we need to analyze carefully the modeled features of the LLC diurnal cycle in terms of timing, intensity and altitude, as well as the implications for precipitation.

5
In the following, we carry out an analysis of an "Inland" area going from 7 • N (not including the coastline) to 9 • N, and from 1 • E to 4 • E (Savé is in the middle of this area), which is a flat terrain with little elevation without any major city, and thus few anthropogenic emission sources. It is an important region for agriculture (mostly composed of tree crop, cereal and root vegetables according to the farming system maps produced by the Food and Agriculture Organization of the United Nations), potentially strongly affected by modifications of LLC and precipitation. Using spatial averages of the modeled fields over this 10 area, we analyze the effect of the anthropogenic aerosol load increase on LWMR, liquid water path and precipitation rate.
We compute the modeled liquid water path (LWP) expressed in mass of water per square meter (kg.m −2 ) in the lowermost troposphere (below 2 km amsl) from the LWMR expressed in mass of water divided by mass of air (kg/kg), using the modeled air density (M air in kg.m −3 ) and height of the vertical levels (H level in m) with the following formula: The hourly evolution of the LWMR, LWP and precipitation are averaged over the "Inland" area (red square on Fig. 1). We analyze each day and the averaged diurnal cycles (with hourly mean and standard deviation) in order to analyze which features are seen every day, and to understand the main differences between AE1 and AE10 (Fig. 8, Fig. 9 and Fig. 10). The model simulates an increase of the LWP from 00:00 to 07:00 UTC, and a decrease from 07:00 to 00:00 UTC for the two experiments ( Fig. 8-b). This diurnal cycle is observed every day ( Fig. 8-a). We note again that, from 07:00 to 13:00, the LWP 5 in AE10 is slightly lower than in AE1, then from 13:00 to 22:00 UTC, the LWP in AE10 is higher than in AE1. The LWP in AE1 and AE10 differ mostly from 13:00 to 22:00 UTC (shadings corresponding to mean±std are less overlapping in Fig. 8-b).
In order to understand how anthropogenic emissions affect the vertical distribution of LWMR, the diurnal evolution is 10 analyzed using two iso-contours relevant for high cloudiness (two iso-contours equal to 2.5×10 −5 and to 5×10 −5 kg/kg) in the lowermost troposphere (Fig. 9). We note that for both experiments, (i) the two iso-contours of LWMR appear during the night at the same time for both experiments (at 00:00 and 02:00 UTC, respectively), (ii) the LLC altitude (i.e. low and high parts of the iso-contours) rises from 500 m amsl at the beginning of the night to 1400 m at the end of the afternoon, (iii) the main differences between the two simulations occur from 14:00 to 21:00 UTC because the presence of the two iso-contours is 15 delayed by one hour for AE10 compared to AE1.   Finally, we analyze the modeled diurnal cycle of precipitation for the two simulations (Fig. 10). The precipitation rates also show a clear cycle every day, occurring from 09:00 to 22:00 UTC in the "Inland" area, which is in agreement with the observations as shown by Knippertz et al. (2017). The more intense hourly precipitation rates are modeled for AE1 with about 0.10 mm occurring at 16:00 UTC, and for AE10 with about 0.08 mm occurring at 17:00 UTC ( Fig. 10-b). In the averaged diurnal cycle, the relative change in hourly precipitation from AE1 to AE10 reaches -28 % at 16:00 UTC and -23 % at 17:00.
However, the differences concern mostly precipitation rates greater than 0.1 mm/h that occurred only on 3 and 6 June 2016 ( Fig. 10-a).

5
In summary, AE1 produces more precipitation and an earlier maximum, while AE10 starts later and produces less precipitation, as it contains smaller drops and therefore sustains cloud cover through the afternoon. Considering daily averages during the studied week (over the "Inland" area), the AE10 experiment simulates LWMR increased by 2.6 % and precipitation rate reduced by 7.5 % compared to AE1. These low relative changes of LWMR hide contrasted periods between the morning and the afternoon. Thus, we conclude that there is a moderate influence of anthropogenic aerosol emission on LLC and precipita-10 tion over this area. Nevertheless, these aerosols modify liquid water patterns in the boundary layer, changing precipitation in the North of the "Inland" area, and potentially also over the ocean, as anthropogenic aerosols are eventually transported over the ocean. transporting polluted air masses towards the ocean, which have been observed with the aircraft measurements and analyzed by Flamant et al. (2018a). This could strongly modify the LLC and precipitation over the ocean, especially because the aerosol concentration is lower than over the continent. 20 In this section, we extend the analysis of the temporal variability of LLC and precipitation to a larger region. We study three areas, namely from north to south: Area-9-11N, from 9 • N to 11 • N, Area-7-9N, from 7 • N to 9 • N (the area studied in Sect. 5), and Area-5-7N, from 5 • N to 7 • N, all three extend from 1 • E to 4 • E. Note that the Area-5-7N contains a portion of the ocean from 5 • N to about 6 • N.
In a first step, we compare the two experiments with TRMM estimates averaged over the three areas at a daily scale ( Fig.   25 11). In a second step, we compare averages of the relative change in LWP and precipitation rate between the two experiments in average for the three areas. In a third step, we analyze the differences between the AE1 and AE10 experiments in terms of LWP and precipitation rate at an hourly time step using time-latitude Hovmoller diagrams ( Fig. 12 and 13).
The TRMM estimates show a large day-to-day variability of precipitation rate in the three areas, ranging for: Area-9-11N) from 0.1 to 13.2 mm/day, Area-7-9N) from 0 to 2.5 mm/day, and Area-5-7N) from 0.2 to 7.7 mm/day (Fig. 11). Satellite 30 estimates reveal a larger day-to-day variability than the modeled precipitation rates, which vary from 0.1 to about 2 mm/day.
The maximum modeled precipitation rates are in: Area-9-11N) 1.7 mm/day for AE1 and 1.7 mm/day for AE10, Area-7-9N) Figure 11. Daily precipitation rates measured by the TRMM instrument (black line) and modeled by the AE1 (blue line) and AE10 (red line) experiments for the period 1-7 July 2016, averaged over three areas extending from 1 • E to 4 • E, and a) from 9 • N to 11 • N, b) from 7 • N to 9 • N, and c) from 5 • N to 7 • N.
2.1 mm/day for AE1 and 1.8 mm/day for AE10, and Area-5-7N) 2.6 mm/day for AE1 and 2.0 for AE10. The model reproduces days of light precipitation but days of heavy precipitation are not reproduced.
Even if both state-of-the-art models and satellite products have difficulties in retrieving the precipitation in SWA, this comparison suggests that the model predicts too many days of low precipitation rate, as it has already been shown for the WRF model .
Liquid water path Precipitation rate In the area of Savè (Area-7-9N), LWP is increased by 2.6% in AE10 compared to AE1 on average over the studied week, while the precipitation rate is decreased by -7.5% (Table 4). North of 9 • N (Area-9-11N), LWP is increased by 2.1%, while the precipitation rate is decreased by -2.8%. The relative difference between the two experiments is greater over the ocean in the Area-5-7N for LWP reaching 13.2%, and for precipitation rate reaching -10.0%, which shows the importance of anthropogenic aerosol transported over the ocean.   To understand the meridional evolution of LWP and precipitation, we investigate the modeled differences with Hovmoller diagrams (time-latitude) from 5 • N to 11 • N and averaged over the same longitudes from 1 • E to 4 • E. Higher LWP is modeled every day in the Area-7-9N from 03:00 to 09:00 UTC for AE10 and AE1 (Fig. 12), which confirms the homogeneity of the modeled LLC cover over the Savè area.
The modeled spatial variability of the LLC cover is in good agreement with SEVIRI satellite images at 08:00 UTC and 5 16:00 UTC ( Fig. A3 and Fig. A4). In both experiments, LLC appear first over the ocean in the Area-5-7N at around 18:00 In the Area-9-11N, the LLC period is shorter, appearing at 03:00 UTC, and disappearing at 12:00 UTC.
Precipitation rates occur in the Area-5-7N over the ocean (from 5 • N to 6 • N) every day from midnight to the following morning with slightly higher rates for AE1 compared to AE10 (Fig. 13). We can see that, at the latitude of the coast (between 6 • N and 6.5 • N), precipitation rates are decreased in the morning every day. Precipitation rates are highest in the Area-7-9N, 10 occurring during the afternoon over this entire area at the same time.
The high precipitation rates occur at a different time over the ocean between 00:00 and 12:00 UTC, and over the continent between 12:00 and 00:00 UTC. In both simulations, precipitation rates are higher for AE1 compared to AE10. We note that the high precipitation occurring on 6 July during the night in the area-9-11N is stronger for AE10 than AE1 (but precipitation rates that are not in agreement with TRMM measurements on this day). 15 To conclude, aerosols emitted from anthropogenic activities have a regional scale influence on the LLC and precipitation, both inland and over the ocean. The differences for LLC between the two simulations are weak on average, however there are contrasted periods during the day.

Conclusions
This study deals with low-level clouds and precipitation over southern West Africa, and focuses on their interactions with 20 anthropogenic aerosols, which are mostly emitted by activities located close to the coast of the Gulf of Guinea. We use the coupled WRF-CHIMERE model with two experiments, which differ solely by the intensity of their anthropogenic emissions. of black carbon, organic carbon, ammonium and nitrate concentrations, which reinforce one of the main conclusion of the DACCIWA project, highlighting the importance of using specific anthropogenic emission inventory for southern West Africa Keita et al., 2020).
Taking the aerosol effect on meteorology into account, does not lead to important modifications of humidity and wind speed variability in the lowermost troposphere. The two experiments are realistic and similar for humidity and wind speed. Over an 30 "Inland" area (extending from 7 • N to 9 • N and from 1 • E to 4 • E), which corresponds to land widely used for agriculture and without any major city, the anthropogenic aerosol increase leads to an increase of hourly low-level cloud cover of more than 20 % in the afternoon, in agreement with Taylor et al. (2019). The breakup time of low-level clouds is delayed by one hour, associated with a decrease of precipitation of up to 16% . The influence of anthropogenic aerosols on low-level clouds is most significant from 16:00 to 21:00 UTC every day, which corresponds to the period of the transport of anthropogenic aerosol from the coastal urbanized area toward the north (Adler et al., 2017;Deetz et al., 2018b;. In contrast, from 09:00 to 15:00 UTC, the anthropogenic aerosols are accumulated along the coastline , where the modeled low-level cloud cover seems to be sensitive to the anthropogenic aerosol difference between the two experiments.

5
We conclude that there is a moderate effect of anthropogenic aerosol emissions on low-level clouds and precipitation in SWA.
The long-range transport of aerosol reaching the Gulf of Guinea was high during the DACCIWA campaign period, mostly because of the biomass burning emission in Central Africa Taylor et al., 2019;Denjean et al., 2020), which is taken into account in the two experiments. However, our results suggest that local anthropogenic aerosol emissions have an effect on low-level clouds and precipitation over the entire southern West African region, even over the ocean, which 10 highlights the importance of mid-level circulation from the continent to the ocean (Flamant et al., 2018a). The modifications in low-level clouds and precipitation affect in turn the agricultural activities in this region, and could also have implications for the transition from stratiform to convective clouds . Furthermore, the reduction of precipitation in southern West Africa due to the anthropogenic aerosol emitted along the coast, changes the liquid water transported towards the Sahel, and modifies the winds and soil humidity, and as a consequence dust emissions, as shown by Menut et al. (2019). The present 15 study adds evidence to the emerging hypothesis that, during the West African monsoon, increasing anthropogenic aerosol pollution in southern West Africa has already caused a precipitation reduction (Pante et al., 2020).
Acknowledgements. The research leading to these results has received funding from the European Union 7th Framework Programme