3.1.1 The stable and jet phases

The MI arrival time, used here as the reference time for the stable and jet phases, has been determined by Dione et al. (2019) using a fuzzy logic method that combines an increase in the wind and a decrease in the temperature. The MI arrival time ranges from 16:00 to 20:00 UTC and is negatively correlated to the monsoon strength in the afternoon. Indeed, a strong monsoon flow favours an early MI arrival (Dione et al., 2019). When the MI arrival occurs before the establishment of the stable conditions near the surface (with negative surface sensible heat flux), the stable phase cannot be defined. This situation occurred in four IOPs out of the nine.

The stable phase is characterized by a weak monsoon flow, which persists until the MI arrival time (Fig. 3). The low layer stabilizes with an increase in the bulk Richardson number because of the decrease in the temperature near the ground. The jet phase starts when the MI reaches the site. The NLLJ usually sets in shortly thereafter (about 1 h) (Dione et al., 2019). Because the MI and NLLJ are both associated with cessation of the buoyancy-driven turbulence, due to the distance from the coast to Savè, both processes settle almost at the same time at Savè (Adler et al., 2019; Babić et al., 2019a; Dione et al., 2019). They lead to a progressive increase in the wind, which reaches 9 m s⁻¹ at the end of the jet phase. The jet core height is located at $Z_{*}=\mathrm{1}$. At the same time, a cooling and an increase in relative humidity occur up to $Z_{*}=\mathrm{4}$. An additional impact of the wind increase during the jet phase is the decrease in the bulk Richardson number below the height $Z_{*}=\mathrm{2.5}$, due to the wind shear in the layer below the jet core. Wind speed and potential temperature vertical profiles are shown schematically for stable and jet phases in Fig. 2.

3.1.2 Relevant processes leading to saturation

The contributions of temperature and specific humidity changes to the relative humidity (RH) changes have been quantified for a case study (Babić et al., 2019a) and for 11 IOPs (Adler et al., 2019) using radiosondes released at the Savè site. Figure 4 shows averaged vertical profiles, over stable and jet phases, of specific humidity and temperature contributions to the total change in RH at the Savè and Kumasi sites. The height is normalized by the cloud base when the stratus clouds form. The median values of the cloud-base at Savè and Kumasi are 227 and 137 m a.g.l., respectively (Kalthoff et al., 2018). At Savè, the cooling causes at least 80 % of the RH increase. Adler et al. (2019), who analysed the moistening for separated stable and jet phases at Savè, pointed out a weak moistening during the stable phase, which is almost compensated by a drying occurring during the jet phase, leading to a 20 % contribution of the moisture to the increase in RH at the end of the two phases (Fig. 4a). Using the frequent radiosondes released at Kumasi, the temperature and moisture contributions to the RH change have been estimated with the same method (Fig. 4b). The results for the two sites are similar, and they confirm that the cooling is mainly responsible for the saturation of the lower layer of the atmosphere.

Figure 4Specific humidity (q) and temperature (T) mean contributions to the mean total change in relative humidity (RH) at (a) Savè and (b) Kumasi averaged over all available IOPs. The shading indicates the standard deviation. The horizontal dashed line ($Z_{*}=\mathrm{1}$) indicates the CBH when the stratus clouds form.

Download

In order to quantify the processes responsible for the cooling, Babić et al. (2019a) and Adler et al. (2019) used the budget equation for the mean potential temperature (θ) (Stull, 1988):

where u, v, and w are the wind components; ρ is the mean air density; Q is the net radiation flux; c_p is the specific heat capacity of the air at constant pressure; H is the sensible heat flux; L_v is the latent heat of water vaporization; and E is the evaporation rate. The contributions to the local tendency of the potential temperature (TOT) are the horizontal advection (HADV) and vertical advection (VADV) of potential temperature, the divergence of the net radiation flux (RAD) and of the sensible heat flux (TURB), and the phase change (SQ). However, SQ can be neglected because the heat budget analysis is applied before the cloud formation. As the vertical component of the wind (w) is very difficult to retrieve from the observations, VADV is not estimated, but it is expected to be small. The two remaining terms, TOT and HADV, are then estimated using the measurements at the Kumasi and Savè supersites. TOT is deduced from the radiosonde profiles launched at the supersite. HADV is calculated by combining radiosondes launched at the supersite and at three coastal stations (Abidjan, Accra, and Cotonou) (see Adler et al. (2019), for details of the method). At last, the Santa Barbara DISORT Atmospheric Radiative Transfer (SBDART) model (Ricchiazzi et al., 1998) is applied to estimate RAD (see Babić et al., 2019a for details of the method) at the Savè site only. This method could not be applied to the Kumasi dataset because too few radiosoundings were launched during the jet phase. Additionally, applying the SBDART model to atmospheric soundings with clouds requires some cloud characteristics (cloud top, liquid water content) which are not available at the Kumasi site. Finally, the residual term (RES) includes TURB at the Savè site and both TURB and RAD at the Kumasi site. The vertical profiles for all these terms, averaged over stable and jet phases, are presented in Fig. 5. Although such an analysis was performed separately for the jet phase at the Savè site (Adler et al., 2019) (not shown), it is, however, not feasible at the Kumasi site because, among other reasons, the MI and NLLJ arrival times were not established, so the start of the jet phase is unknown.

Figure 5a shows a local total cooling that decreases with height from −0.80 K h⁻¹ at the surface to −0.15 K h⁻¹ at $Z_{*}=\mathrm{4}$. HADV is also height dependent, with a vertical profile closely linked to the horizontal wind profile of the NLLJ (Adler et al., 2019; Dione et al., 2019). HADV is −0.20 K h⁻¹ at the surface and at $Z_{*}=\mathrm{2.5}$, with a maximum cooling of −0.40 K h⁻¹ at the jet core height. RAD slightly increases with height from −0.15 K h⁻¹ at surface to −0.10 K h⁻¹ at $Z_{*}=\mathrm{4}$. This means that RES, which is mainly the sensible heat flux divergence term at Savè, is the main contribution to the cooling close to the ground, with a decreasing effect with height. Above $Z_{*}=\mathrm{1.5}$, the sum of RAD and HADV is nearly equal to TOT. These three processes and their respective role during the stable and jet phases are indicated in Fig. 2. Adler et al. (2019) conclude, when integrating the vertical profiles with height for Savè, that the advection is the main cooling process, responsible for 50 % of the cooling of the atmosphere during the stable and jet phases. Each of the radiation and heat flux divergences contributes around 22 % to the local total cooling.

Very similar results are obtained with the Kumasi dataset, though a slightly lower local total cooling is observed at Kumasi. Close to the surface, it is 25 % lower than the one observed at Savè, with a cooling of −0.60 K h⁻¹ at Kumasi instead of −0.80 K h⁻¹ at Savè. It remains 16 % lower at $Z_{*}=\mathrm{1}$. Nevertheless, as for Savè, HADV contributes about 50 % to the local total cooling at around $Z_{*}=\mathrm{1}$ and above, while cooling near the surface is mainly caused by the sensible and radiation flux divergence.

Figure 5(a) Vertical profiles of local total cooling rate (TOT) and estimates of contribution of horizontal advection (HADV) and net radiation flux divergence (RAD) terms for the Savè site during the stable and jet phases. (b) Vertical profiles of TOT and HADV for the Kumasi site during the stable and jet phases. The residual term includes the turbulent flux divergence at Savè (RES_T) and both the net radiation and turbulent flux divergences at Kumasi (RES_T+R) . The shading indicates the standard deviation. The horizontal dashed line ($Z_{*}=\mathrm{1}$) indicates the CBH when the stratus clouds form.

Download

3.2.1 The stratus and convective phases

According to the right column in Fig. 3 presenting the stratus and convective phases, the highest values of the horizontal wind speed, defining the NLLJ core, are observed between $Z_{*}=\mathrm{1.0}$ and $Z_{*}=\mathrm{1.5}$ at the beginning of the stratus phase. The correspondence of the LLSC base height with the NLLJ core can be explained by the high cooling due to horizontal advection at this level during the stable and stratus phases (Fig. 5). During the stratus phase, the wind speed decreases and becomes more constant with height between $Z_{*}=\mathrm{0.5}$ and $Z_{*}=\mathrm{2.0}$ (Fig. 3). The modification of the wind vertical profile is likely due to the turbulent mixing within the cloudy layer as induced by long-wave cooling at the cloud top. Although reduced, the maximum values of the wind speed are found towards the top of the cloudy layer during the stratus phase (Adler et al., 2019; Babić et al., 2019a; Dione et al., 2019) as schematically represented in Fig. 2. The potential temperature in the sub-cloud layer is often well mixed and two processes could contribute to this: the shear-driven turbulence below the NLLJ core and the turbulence induced by the LLSC due to the radiative cooling at the cloud top. When the convective phase starts, the wind speed is almost back to monsoon flow daytime conditions, i.e. a weak wind of about 4 m s⁻¹. The relative humidity decreases with time close to the surface and above $Z_{*}=\mathrm{2}$, which indicates a thinning of the LLSC from both the base and the top. The cloudy layer rises as the CBL develops, with the typical features of a convective mixed layer: a warming of the near-neutral stratified sub-cloud layer and a decrease in the bulk Richardson number.

3.2.2 Relevant processes linking the LLSC to the surface

As shown in Fig. 3, the LLSCs and the NLLJ could interact in two ways. First, the LLSCs could reduce the NLLJ strength because of the turbulent mixing in the cloudy layer. Such an effect of the LLSC turbulent mixing on a mesoscale phenomenon like the NLLJ is possible, because the LLSCs extend over more than 800 000 km² from the Guinean coast up to 10^∘ N latitude. Secondly, the turbulence below the NLLJ core modifies the conditions in the sub-cloud layer. As already observed during the AMMA experiment (African Monsoon Multidisciplinary Analysis; Redelsperger et al., 2006) by Lothon et al. (2008), the NLLJ can induce dynamical turbulence down to the surface, reflected by an increase in night-time turbulent kinetic energy up to 0.6 and 0.3 m² s⁻², on average, at Kumasi and Savè, respectively (Kalthoff et al., 2018). If the shear-driven and the cloud-top-driven turbulence are strong enough, the sub-cloud layer is well mixed and the lifting condensation level (LCL) (Romps, 2017) will correspond to the LLSC base. The LLSC is considered coupled to the surface (Adler et al., 2019) (indicated by LCL_coupled in Fig. 2). On the contrary, a sub-cloud layer with low turbulence will be less mixed. The cloud is then decoupled from the surface, with its base higher than the LCL (indicated by LCL_decoupled in Fig. 2). This is shown in Figure 6, where the bulk Richardson number in the sub-cloud layer is plotted versus the difference between the CBH and the LCL estimated from the radiosondes launched during the stratus phase at Savè and at Kumasi. For a bulk Richardson number below 0.1, the CBH is less than 75 m below or above the LCL. For larger values of the bulk Richardson number, the CBH is at least 150 m above the LCL. One can note only four radiosoundings for which the bulk Richardson number is above 0.1 at Kumasi, indicating larger turbulence in the sub-cloud layer at Kumasi than at Savè.

Figure 6Bulk Richardson number (Ri_b) in the sub-cloud layer versus height difference between cloud base height (CBH) and lifting condensation level (LCL) estimated from the radiosondes launched during the stratus phase at Savè and at Kumasi. Colours stand for the different IOPs.

Download

3.2.3 LLSC breakup and boundary layer evolution

The LLSC breakup occurs during the convective phase, when the CBL develops vertically due to heating of the surface. The frequent radiosondes released during the convective phase are unfortunately not suitable for determining the CBL height, because the potential temperature gradients were very weak. An estimate of the sub-cloud layer height is given by the LCL. This LCL-based estimate is motivated by the very good accordance between this level and the base height of the cumulus cloud forming in the afternoon. In addition, this method has the advantage to continuously provide the sub-cloud layer height, and it can easily be applied to the Kumasi and Ile-Ife sites where the radiosondes are less frequent.

The way in which the LLSC and the surface are coupled (or not), as discussed in the previous section, plays a determining role in the breakup. Based on CBH evolution, relative to LCL and CBH standard deviation along the stratus and convective phases, three scenarios have been defined at Savè, as illustrated in Fig. 7. For each scenario, Fig. 7 presents, in the top panel (Fig. 7a), the temporal evolution, from 00:00 to 16:00 UTC, of the CBH and cloud top (measured by the cloud radar; see Adler et al., 2019, for the method) and LCL. The half-hourly standard deviation of the cloud base, the cloud fraction (percentage of cloud base below 1000 m a.g.l. over 30 min), and the difference between cloud base and LCL are indicated in the bottom panel (Fig. 7e). A cloud fraction larger than 95 % is chosen as a criterion to determine the presence of LLSC above the supersite (Adler et al., 2019). Examples of visible and infrared images from the cloud cameras at two distinct times illustrate the cloud coverage, and they allow one to make the link between the quantified parameters presented previously and the state of the sky as recorded by the cameras.

• Scenario 1 (Fig. 7a and b) is LLSC coupled to the surface. It is illustrated by the 8 July case (IOP 8) (Babić et al., 2019a). During the night, the CBH is very close to the LCL (LCL_coupled in Fig. 2), and the CBH standard deviation is very small, which indicates a very regular and constant LLSC base. The visible and infrared images indicate a homogeneous stratus layer at 08:14 UTC. After 08:00 UTC, the LCL steadily increases to reach 1000 m a.g.l. at 16:00 UTC. Until 12:00 UTC, the increases of the LCL and CBH are simultaneous and identical. After that time, the cloud base fraction decrease indicates the LLSC breakup. The low values of the standard deviation of the CBH during daytime and the visible and infrared images at 10:28 UTC indicate a LLSC evolution towards a thinner cloud layer and stratocumulus clouds. Among the eight IOPs for which the LLSC breakup has been analysed, three follow this scenario. This breakup scenario is schematically presented in Fig. 2 with an LCL close to the CBH during the convective phase (Scenario 1).

  When the LLSC is decoupled from the surface during the night, meaning that the LLSC base is above the LCL (LCL_decoupled in Fig. 2), two scenarios occur.

  • Scenario 2 (Fig. 7c and d) is progressive coupling of the LLSC to the surface. In this scenario, the LCL rises up to the CBH, as illustrated with the 27 July case (IOP 14) and schematized in Fig. 2 (Scenario 2). After 08:00 UTC, once the LLSC is coupled to the surface, the CBH rises with the sub-cloud layer height as in Scenario 1. In IOP 14, the LLSC breakup occurs at 12:00 UTC, shown by the decrease in the cloud base fraction. In that case, the high standard deviation of the CBH indicates cumulus cloud development. The ceilometer is not only sensitive to the base of clouds, but also to the edges when the clouds pass over the instrument. In the case of cumulus clouds, the ceilometer indicates scattered heights. Visible and infrared images illustrate the passage from the LLSC at 07:18 UTC to cumulus clouds at 12:14 UTC. Two IOPs among eight follow Scenario 2.

  • Scenario 3 (Fig. 7e and f) is cumulus cloud formation below the LLSC. This scenario is illustrated by IOP 11. The LLSC is decoupled from the surface with a CBH base 150 m above the LCL during the night, increasing to 300 m at 07:00 UTC. The infrared image is a mix of orange and yellow colours, which indicates colder bases than the ones observed in IOP 8 and IOP 14 early in the morning. After 07:15 UTC, the ceilometer measures a quite scattered cloud base but coupled to the surface. This indicates some cumulus cloud formation below the LLSC. Two cloudy layers coexist until 08:30 UTC, as indicated by the 100 % cloud coverage. A thin stratocumulus layer persists after 09:00 UTC with a very low standard deviation of the base, which is at about 400 m a.g.l. When the low stratocumulus layer is broken enough, higher cloud bases at around 1500 m a.g.l. are detected by the ceilometer between 10:00 and 11:00 UTC. Visible and infrared images at 10:36 UTC illustrate this higher stratocumulus layer. Three IOPs among eight follow Scenario 3. This breakup scenario is schematically presented in Figure 2 with an LCL below the CBH for the entire convective phase and with some boundary layer cumulus formation below the LLSC (Scenario 3).

Figure 7(a) 8 July (IOP 8), (c) 27 July (IOP 14), and (e) 18 July (IOP 11) illustrate breakup scenarios 1, 2, and 3, respectively. The temporal evolution of the cloud base height (red dots), cloud top (grey dots), and LCL with its uncertainty (black line with grey shading) are presented in the top panels. The LCL uncertainty is based on the uncertainties of the temperature and humidity used for the LCL estimation. The temporal evolution of the cloud fraction (blue line), the standard deviation of the cloud base over 30 min (black line), and the difference between LCL and CBH (dashed black line) are presented in the bottom panels. The vertical dashed green lines indicate the LLSC onset and breakup times. The vertical red lines indicate the times of the visible and infrared camera pictures presented in (b), (d), and (f) for IOP 8, IOP 14 , and IOP 11, respectively. The times are 08:14 and 10:28 UTC on 8 July, 07:18 and 12:14 UTC on 27 July, and 07:02 and 10:36 UTC on 18 July. The colour scale for the infrared images (indicated at the bottom of the images) ranges from blue, for colder brightness temperatures, to white for the warmer brightness temperatures. The black rectangle on the visible images indicates the area corresponding to the infrared images. The white dot in the infrared images, at 12:14 UTC on 27 July and 10:36 UTC on 18 July, is due to the sun and is located at the centre of the solar disc.

The stratus clouds modify the surface energy balance (SEB), because they reduce the net short-wave radiation twice as much as they increase the net infrared radiation (Chen et al., 2000). In order to investigate the effect of the LLSC on the SEB during DACCIWA campaign, the temporally averaged flux from 06:00 to 16:00 UTC (indicated by 〈〉) of the net radiation (Rn), the latent heat flux (Le), and the sensible heat flux (H) are calculated for 21 d for Savè and Kumasi and 20 d for Ile-Ife.

Figure 8 shows a negative correlation between 〈Rn〉, 〈Le〉, and 〈H〉 and the LLSC breakup time, with correlation coefficients below −0.64 for Savè and Kumasi. The variability is certainly due to the day-to-day variation of soil moisture, which is more important at the beginning of the campaign than at the end, when frequent rain events maintained an almost constant soil moisture. 〈Rn〉 is reduced by 25 % and 50 % when LLSC breakup occurs after 13:00 UTC at Savè and at Kumasi, respectively. This difference may be due to LLSC macrophysical properties, like deeper clouds or larger liquid water path, but also to higher cloudy layers which also impact the net radiation. Unfortunately, the same plot cannot be provided for Ile-Ife, where the breakup time of the LLSC could not be determined. From these results, one can deduce for the first time the error related to the SEB if the LLSC breakup time is inaccurately simulated by numerical models.

The impact of the SEB on the CBL vertical development is shown in Fig. 9a. The CBL development is represented by the LCL at 16:00 UTC at the three sites and plotted versus 〈H〉. As expected, the lower 〈H〉, the lower LCL, which ranges from 300 to 1000 m a.g.l. over the three sites. Finally, the link between the CBL development and the LLSC breakup time is shown, for the Savè and Kumasi sites, in Fig. 9b. Later LLSC breakup implies lower net radiation at the surface (Fig. 8), and therefore weaker surface sensible heat flux (Fig. 9a), which leads to a lower vertical development of the CBL. The LCL is half, when LLSC breakup occurs after 11:00 UTC, compared to an LCL associated with early-morning LLSC breakup. The impacts of this on the moist convection during the afternoon need detailed investigations.

Figure 8Averaged net radiation (〈Rn〉), latent heat flux (〈Le〉), and sensible heat flux (〈H〉) at the surface from 06:00 to 16:00 UTC versus LLSC breakup time at (a) Savè and (b) Kumasi. Linear regressions are plotted and correlation coefficients are indicated.

Download

Figure 9(a) LCL at 1600 UTC versus 〈H〉 at the three sites. (b) LCL at 16:00 UTC versus LLSC breakup time at Savè and at Kumasi. The LLSC breakup time cannot be estimated at Ile-Ife.

Download