2.1 Wind profiling of the lower troposphere

An ultra-high-frequency (UHF) wind profiler operated by UPS was devoted to the study of the vertical structure of the atmospheric dynamics in the lower and middle troposphere. This 1274 MHz Doppler radar works with five beams. The three components of the wind are retrieved from these beams. This instrument was previously used for various studies of the planetary boundary layer: turbulence retrieval (Jacoby-Koaly al., 2002), African easterly jet analysis (Kalapureddy et al., 2010), the WAM diurnal cycle (Lothon et al., 2008), the NLLJ (Madougou et al., 2012), and offshore winds in high precipitation events (Said et al., 2016).

The radar ran continuously from 19 June to 30 July 2016 at Savè during the DACCIWA field experiment and provided vertical profiles of the wind every 2 min. We block averaged the data at 15 min time intervals for our wind analysis for consistency with other instruments. For our study, we considered the acquisition mode that corresponded to the highest radial resolution of 75 m, which enabled good documentation of the low troposphere from 150 m up to 3 km in height. The UHF profiler data are used here to characterize the monsoon flow, the NLLJ, and the MI.

2.2 Lower troposphere temperature profiling

A scanning microwave radiometer (humidity and temperature profiler HATPRO-G4 manufactured by RPG – Radiometer Physics GmbH) from the KIT was used to analyze the temperature profiles. The radiometer measured brightness temperature from which integrated water vapor, liquid water path, temperature profiles, and humidity profiles could be retrieved using the retrieval algorithm provided by the University of Cologne (Löhnert and Crewell, 2003; Löhnert et al., 2009). Every 15 min, temperature profiles with enhanced accuracy were obtained using low-elevation scans (Crewell and Löhnert, 2007), which are used here to detect the MI with the 302 K isotherm (Deetz et al., 2018) and the fuzzy logic algorithm described below. A systematic comparison of the radiosounding temperature profiles with the HATPRO temperature profiles (not shown) revealed a systematic cold bias of 0.2 K below 550 m, 0.5 K in the 550–1000 m layer, and 2 K in the 1000–2000 m layer. This finding is consistent with the accuracies noted by Crewell and Löhnert (2007) (<1 K below 1000 m). Only microwave radiometer measurements below 550 m are used in this study. The available dataset of this instrument covers the 30 June to 30 July time period.

2.3 Observation of surface layer conditions

A 7.77 m scaffold was mounted at Savè by UPS to study energy balance and biogenic fluxes. High-frequency measurements of air temperature, specific humidity, and three components of the wind were obtained (at 0.1 s time interval). The energy flux (latent heat, sensible heat, momentum flux) were calculated for the samples at a time resolution of 30 min. Additionally, upward and downward components of shortwave and longwave radiation, air pressure, soil temperature, and soil moisture were measured every minute.

This station operated continuously from 13 June to 30 July 2016. In this study, we use the sensible heat flux estimates to characterize the surface layer stability.

2.4 Cloud monitoring

Three co-located devices deployed during the DACCIWA field experiment are used to monitor low clouds: a ceilometer, cloud radar, and an infrared (IR) camera.

A CHM15k ceilometer, which is a 1064 nm wavelength lidar with a 5–7 kHz pulse rate, was installed by the KIT for the continuous monitoring of cloud base height (CBH). The ceilometer was operated from 3 June to 30 July 2016 at a time resolution of 1 min and a vertical resolution of 15 m. Manufacturer software automatically provided three estimates of CBH, which allowed us to detect multiple cloud layers. Here, we only use the lowest CBH, which corresponds to the low clouds under focus. We define “low-level clouds” as clouds with a base height below 1500 m a.g.l. Adler et al. (2019) used a lower altitude threshold, 600 m a.g.l., which is well adapted to the nocturnal stratus clouds. However, a 1500 m height limit allows us to extend our detection of LLSCs during daytime when the stratus cloud base height rises due to the growing convective boundary layer.

More information on cloud characteristics was provided by the KIT 35.5 GHz cloud radar (i.e., cloud top height (CTH) and cloud microphysics (rain, drizzle)). The cloud radar operated continuously from 14 June to 30 July 2016. It was run with vertical pointing every 5 min and horizontal scans every 30 min. Here, we use the observations of the vertical profiles for the CTH evaluation. After the despiking process, we averaged the reflectivity profiles of hydrometeors over 5 min and applied a threshold of −35 dBz to capture the CTH (Adler et al., 2019). Values below this threshold are considered to be related to clear air above the cloud. Additional details about the CTH retrieval technique can be found in Babić et al. (2019a). This algorithm enables a good estimation of the CTH, particularly when the clouds are uniform, which is true in the case of stratus clouds. However, CTHs are difficult to capture for scattered clouds or rain (e.g., cumulus clouds during the daytime). This fact explains some missing CTH estimates during the daytime in our later analysis. Finally, UPS installed a MOBOTIX S15 cloud camera that monitored the cloud cover all day and obtained visible and IR pictures every 2 min. The visible image was a full sky image: the aperture angles for the IR channel were $\mathrm{43}{}^{\circ }\times \mathrm{32}{}^{\circ }$ (which corresponds to 158 m × 114 m in area at a height of 200 m). In this study, we used 5 MP IR images, coded in red, green, and blue (RGB) components over 256 colors. Here, [R,G,B] denotes the relative contributions of red, green, and blue of a given pixel, all defined between 0 and 1 (with accuracy of 1/256). The color of a pixel depends on the emissivity of the corresponding sky area and consequently its brightness temperature (uncalibrated). Typically, a low cloud base is seen as red and a clear sky is seen as blue. Therefore, a homogeneous low cloud deck will create a homogeneous red color image; a fragmented stratocumulus will render an image with colors ranging from red to blue. This instrument is used here to study the horizontal homogeneity of the cloud deck and to define the onset and breakup times of the stratus deck, with a newly designed method. As far as we know, it is the first time that such methodology is used for the study of stratus cloud deck formation and breaking.

3.1 Monsoon flow detection

The monsoon flow can be detected according to the wind direction (Kalapureddy et al., 2010). In this study, it is defined as the lower layer with a 135^∘ (SE)–270^∘ (W) horizontal wind sector. The choice to use a quite large wind direction sector, including some westerly winds, is motivated by weak monsoon flow (below 1 m s⁻¹) observed during the daytime for which the direction is ill defined and can be influenced by local effects. This criterion on wind direction allows to exclude atmospheric conditions associated with vortex circulations, deep convection and Harmattan flow. The top of the monsoon flow is the level above which the horizontal wind direction is out of the 135–270^∘ sector for more than 225 m (i.e., third UHF wind profiler gate). Very often, a large wind-sheared layer between the monsoon flow and the easterlies (African easterly jet) above is observed, which makes it difficult to determine the monsoon flow depth. The strength of the monsoon flow is defined as the mean wind speed within this depth.

3.2 Monsoon flow characteristics

Applying the previous criteria to the UHF wind profiler dataset, the temporal evolutions of the monsoon flow depth and strength from 15:00 UTC on day D-1 to 15:00 UTC on day D are calculated for day D from 20 June to 30 July 2016. The results are presented in Fig. 2. Except for 13, 16 and 17 July, dates for which the UHF wind profiler data are missing, the short periods shown in white in Fig. 2, are for a wind direction not falling within the 135–270^∘ sector. As discussed in the previous section, the strong variability of the monsoon flow depth observed on some days, like 27–28 June, is associated with a large wind-sheared layer and a low wind. However, the intra-seasonal variability of the monsoon depth (from a few hundred meters to 4 km) and strength (from almost zero wind to 10 m s⁻¹) can often be linked to synoptic conditions (Couvreux et al., 2010). Several days in a row, such as 20–23 June and 10–12 July, exhibited consistent monsoon flow characteristics over time with a particularly large monsoon flow depth and strong wind speed. The period 10–12 July is included in the vortex phase (9–16 July). During this period, an unusual cyclonic circulation developed and slowly propagated from eastern Mali to Cabo Verde along with an anticyclonic vortex in the west–northwesterly direction along the Guinean coast (Knippertz et al., 2017). The wet westerly regime (21–27 July) was characterized by a particularly low monsoon flow strength associated with important oscillations of the monsoon depth due to rain events that shifted back to the coast (Knippertz et al., 2017).

Figure 2Diurnal evolution of the monsoon flow (a) thickness and (b) strength at Savè from 15:00 UTC on day D-1 to 15:00 UTC on day D during the course of the 20 June through 30 July 2016 time period. Solid black vertical lines delimit the different phases of the monsoon described by Knippertz et al. (2017). Dashed vertical lines indicate the vortex phase included in the post-onset phase. White windows indicate periods for which the monsoon flow was not defined (i.e., a wind direction beyond the 135–270^∘ sector). Grey windows indicate missing data. Red squares indicate rainy conditions.

Download

Many convective activities occurred at the Savè site during the campaign, which is typical in SWA (Laing and Fritsch, 1993). These phenomena were either detected by rain measurements (Fig. 2) or density current effects at the surface. Table 1 lists the days for which rain and/or density currents are detected after 15:00 UTC; such events are likely to have disturbed or interacted with the formation of LLSCs. The impact of density currents at the surface, in terms of temperature, relative humidity, specific humidity, wind speed, and direction depends on the intensity of the convective cell and its location. A sudden decrease in temperature and an increase in wind speed are typically observed (Table 2). A shift in wind direction is not a reliable criterion because it depends on the convective cell location with regard to the Savè supersite. As listed in Table 2, density currents are largely associated with drier air, as already discussed by Schwendike et al. (2010). All of the outflow cases listed in Table 1 are associated with a convective cell identified with the rain radar in the supersite surroundings. All of the rainy and density current cases are excluded (16 out of 41 d) from the statistics from which we estimate the primary characteristics of the monsoon flow.

Table 1Characteristics (onset and breakup time) of the NLLJ and LLSCs and the arrival time of the MI at Savé during the 20 June–30 July 2016 period. All times are indicated in UTC. For times related to the LLSCs, blanks indicate that LLSCs were not detectable (because of rain or clear sky). For times related to MI or NLLJ, blanks indicate missing data or undetectable features. The FLF_mean was used to estimate the MI onset. DC indicates density current.

Download Print Version | Download XLSX

A composite 24 h evolution of the monsoon flow depth, strength, and direction from 15:00 UTC on day D-1 to 15:00 UTC the day after during the 20 June to 30 July 2016 period enables a discussion of the diurnal evolution of the monsoon flow characteristics (Fig. 3). The median of the monsoon depth shows a weak diurnal evolution from a minimum value of 1200 m a.g.l. during the night to 2000 m a.g.l. during convective conditions (Fig. 3a), with a day-to-day variability. This finding is consistent with what was observed during the AMMA experiment during the full monsoon season (Kalapureddy et al., 2010). Unlike the monsoon flow depth, the strength and direction of the monsoon flow indicate a clear diurnal cycle (Fig. 3b and c). The median strength of the monsoon flow is roughly 3.5 m s⁻¹ between noon and 17:00 UTC with a 210^∘ direction. The median strength regularly increases between 17:00 and 01:00 UTC up to 5.5 m s⁻¹ with a simultaneous slight shift in the median wind direction (amplitude and standard deviation around 32.62 and 9.46^∘, respectively). These same changes are observed in wind surface measurements (Kalthoff et al., 2018).

Figure 3Temporal evolution of the distribution of the monsoon flow (a) depth, (b) strength and (c) direction at Savè for the period between 20 June and 30 July. The black lines in panels (a), (b) and (c) indicate the median monsoon depth, strength, and direction, respectively.

Download

4.1 Detection of the maritime inflow and nocturnal low-level jet

A low-level jet is characterized by a maximum wind speed a few hundred meters above the surface of the Earth and a clear minimum wind speed above. This situation implies significant shear below and above the jet core. Because of the occurrence of the low-level jets in various environments and forcing, different criteria can be applied to define and detect them based on vertical wind profiles. One of the first proposed definitions of the NLLJ based on observations was given by Bonner (1968). This author defined three types of NLLJs based on three different threshold values for the maximum wind speed (12, 16, and 20 m s⁻¹, respectively), and the existence of a minimum wind speed above the maximum and below 3 km a.g.l. Stull (1988) and Andreas et al. (2000) defined the NLLJ within the first 1500 m as having a maximum wind speed at least 2 m s⁻¹ faster than the minimum wind above it. Baas et al. (2009) defined NLLJs over the Netherlands as having a maximum wind speed at least 2 m s⁻¹ below 500 m a.g.l. and a wind speed 25 % faster than the local minimum wind speed above.

The detection of the NLLJ is based, in this study, on the use of dynamical and surface stability criteria: (i) the wind direction in the lowest atmosphere below 1500 m is between the southeast and west–northwest with (ii) a maximum wind speed of at least 5 m s⁻¹ and at least 2 m s⁻¹ larger than the minimum above and (iii) a surface sensible heat flux lower than 10 W m⁻². This last criterion ensures stable to neutral conditions at the surface. The onset of the NLLJ is defined when these criteria are satisfied for at least 2 h and the height of the maximum wind speed is below 500 m. The breakup time is defined when one of the three criteria mentioned above has not been satisfied for at least 1 h. The use of the surface sensible heat flux as a diagnostic of the stability may be a limitation to this method because this measurement is very local and may not represent atmospheric stability on large spatial scales. The NLLJ arrival and breakup times are determined with a 30 min temporal resolution, which corresponds to the sample duration for the sensible heat flux estimation.

The DACCIWA project focused on a region to the south of the AMMA study area that is affected by coastal phenomena, such as sea breeze. Unfortunately, no measurements provided evidence for MI formation and propagation inland. But based on simulations conducted by Adler et al. (2017) and Deetz et al. (2018), Adler et al. (2019) hypothesized that the MI penetrated 50–130 km inland. When convection and mechanical turbulence vanish at the end of the afternoon, two phenomena occur simultaneously: the monsoon flow increases and the MI, characterized by a higher wind speed than the monsoon flow further north, can propagate further inland, possibly up to Savè. Deetz et al. (2018), using COSMO model simulations for the night of 2–3 July, characterized the MI arrival by the 302 K potential temperature at a height of 250 m. This criterion was applied to the temperature measured locally by the microwave radiometer at the Savè site in order to detect the arrival of the MI. Since the 302 K criterion relies on one simulated case and because no clear justification for this 302 K value exists, another criterion based on MI characteristics observed at the surface is proposed in this study. MI arrival at Savè should be detected at the surface by a combination of both an increase in horizontal wind and a decrease in temperature. Thus, a second method was used in this study based on the combination of both an increase in horizontal wind speed and a decrease in temperature. These changes should be the signature of the MI arrival time. This method, based on the fuzzy logic method, was used by Coceal et al. (2018) to detect the sea-breeze front around London, in the south of England. Our method for MI detection at Savè follows three steps: (1) the rate of changes of temperature (T) and horizontal wind speed (ws) (termed r_T and r_ws, respectively) are calculated using 30 min averaged T and ws in the 200–550 and 150–525 m layers, respectively; (2) fuzzy logic functions for T and ws, termed FLF_T and FLF_ws, respectively, are computed using Eq. (1); and (3) the mean fuzzy logic function, FLF_mean, combining both the changes in T and ws, is the mean of FLF_ws and FLF_T.

The fuzzy logic function FLF_x for the variable x can be written as

where r_x is the rate of change of the variable x, $r_{x_{\mathrm{1}}}$ (respectively, $r_{x_{\mathrm{2}}}$) is a constant value below (above) which FLF_x is equal to y₁ (y₂). r_T is multiplied by −1 to obtain positive changes for decreasing temperature. As in Coceal et al. (2018), y₁ and y₂ are set to 0 and 1, respectively, and $r_{x_{\mathrm{1}}}$ is set to 0 (i.e., no increase in wind speed or no decrease in temperature). Instead of using the maximum value of r_x divided by two for $r_{x_{\mathrm{2}}}$ (Coceal et al., 2018), for each day, we use the value corresponding to the 99th percentile of r_x divided by two to avoid outliers. Considering the values used in this study for $r_{x_{\mathrm{1}}}$, y₁, and y₂ in Eq. (1) can be simplified as follows:

In this study, the mean fuzzy logic function (FLF_mean) is computed using equal weights for FLF_ws and FLF_T, and the same threshold of 1 is used to detect combined changes in the dynamic and thermodynamic conditions. The fuzzy logic method only makes sense if the temperature and wind speed changes are combined. However, to better understand how temperature and wind speed changes distinctly impact FLF_mean, FLF_ws and FLF_T are also discussed. FLF_ws would be a full weight applied to wind speed (i.e., considering that MI is only characterized by a dynamical change), and FLF_T would be a full weight applied to temperature in the fuzzy logic method (i.e., considering that MI arrival implies only a thermodynamical change). The MI arrival time can then be determined by noting the first time at which the fuzzy logic functions of wind speed and temperature are equal to 1. There are, in this method, important constants and thresholds to fix, and several sensitivity tests have been performed. The results for individual days can differ, but the average and the conclusions remain the same.

The use of the MI and NLLJ criteria is illustrated for three very different days in Fig. 4. Each panel presents the 24 h time–height section of wind as measured by the UHF wind profiler and the three fuzzy logic functions (FLF_ws, FLF_T, and FLF_mean). The NLLJ onset times and the two estimates of the MI arrival time (302 K potential temperature and the fuzzy logic method) at Savè are also indicated.

Figure 4Time–height sections of (color) wind speed and (arrows) direction from the UHF wind profiler on the nights of (a) 2–3 July, (b) 7–8 July, and (c) 9–10 July 2016. A leftward horizontal arrow indicates an easterly wind; an arrow from bottom to top indicates a southerly wind. The black open circles indicate the jet core height detected with a maximum wind speed of at least 5 m s⁻¹, the magenta rectangles indicate the height of the minimum wind speed above the jet core, and red open circles indicate the monsoon flow depth. The black, blue, and red lines in the lower box indicate the three fuzzy logic functions of the wind speed, temperature, and their mean, respectively. The red and black dashed vertical lines indicate the MI arrival time estimated using two different criteria: (1) the FLF_mean>1 criterion and (2) the 302 K isentrope criterion, respectively. The black vertical line indicates the NLLJ onset.

Download

A brief and weak NLLJ settled in the night of 2–3 July between 21:00 and 03:00 UTC in a very weak monsoon flow (Fig. 4a) and with a strong easterly wind above 2 km. The first simultaneous wind increase and cooling are observed at 20:00 UTC (FLF_ws, FLF_T both reach 1), which indicates a clear MI arrival time. FLF_mean values are larger than 0.5 until midnight, which demonstrates that the cooling lasts for at least 4.5 h. For that day, the 302 K potential temperature criterion is in very good agreement with the fuzzy logic method with a detection of MI arrival time at 20:15 UTC. The maximum value of the wind vertical profile reaches the threshold of 5 m s⁻¹ at 21:00 UTC, which is the onset time of the NLLJ. The second example (Fig. 4b) is for the night of 7–8 July, which is analyzed in detail by Babić et al. (2019a). The fuzzy logic functions are very similar to those of 2–3 July. The estimated arrival times of the MI are 20:00 UTC and 20:45 UTC using the fuzzy logic method and the 302 K criterion, respectively. Like the night of 2–3 July, the maximum value of the jet reaches the threshold of 5 m s⁻¹ at 21:00 UTC. The particularity of that day is that the NLLJ vanishes near the surface after sunrise on 8 July. However, the jet shape of the wind persists above unstable conditions; the height of the wind speed maximum increases up to 1000 m at 12:00 UTC which leads a strong rotation of the wind between the southwesterly in the monsoon flow and easterly above. This example illustrates the necessity of the atmospheric stability criterion for NLLJ breakup time determination. The third and last example, the night of 9–10 July (Fig. 4c), illustrates a NLLJ that was hardly detected because it develops within a strong monsoon flow observed all day long and weaker easterly wind above compared to 2–3 and 7–8 July. Six other days presenting a similar strong monsoon flow are observed during the period between 20 June and 30 July. These days correspond to synoptic anticyclonic atmospheric conditions over the Gulf of Guinea that favor a strong penetration of the monsoon flow inland (Knippertz et al., 2017). In such conditions, the NLLJ onset is not easy to determine. Figure 4c shows that the maximum wind is detected above 500 m before 19:30 UTC, which prevents us from considering this strong monsoon flow as a NLLJ until that time (criterion (i) of the NLLJ detection). The MI reaches Savè at 18:00 UTC according to the fuzzy logic method, 2 h earlier than the two previous examples. This result appears in accordance with a strong monsoon flow inhibiting the turbulent mixing and pushing the MI front faster and farther inland. On the other days, the cooling lasts until midnight. The detection of MI arrival time according to the 302 K criterion yields a later MI arrival time (19:30 UTC).

Based on these three examples, one can note the large variability that can be observed from one day to the other, which makes it challenging to define solid common criteria for MI and NLLJ detection.

4.2 MI and NLLJ characteristics

Characteristics of the MI and NLLJ are deduced from a set of days excluding rainy days and density current cases. For the NLLJ, we use the same days that we used for monsoon flow characteristics. For the MI, the 30 June through 30 July period is used (imposed by radiometer data availability), and 12 d are excluded (listed in Table 1).

Table 2Variations in surface (7.77 m height) temperature (ΔT), relative humidity (ΔRH), specific humidity (Δq), wind speed, and wind direction outflows that can disturb the MI arrival time and NLLJ onset and participate in the cooling of the atmosphere detected after 15:00 UTC. For the wind changes, the wind speed and wind direction averaged within the layer (150–525 m) before and after the event are considered.

Download Print Version | Download XLSX

The temporal occurrence of fuzzy logic functions equal to 1 for the 30 June–30 July period is shown in Fig. 5. One can note interesting results:

• A large increase in wind speed (FLF_ws=1) may be observed at any time during the day; however, the largest occurrence (≥5) is between 17:00 and 20:00 UTC. This variability is due to the day-to-day variability of the monsoon strength and the arrival time of the NLLJ during this period.

• As expected, cooling may occur between 18:00 and 00:30 UTC the following day. Contrary to the wind speed, whose fuzzy logic function reaches 1 but rarely remains at that value for several hours during the night, the temperature fuzzy logic function reaches this value many times during the night. This trend implies continuous cooling (Fig. 4). This result is in accordance with the continuous decrease in temperature within the MI from north to the south discussed by Adler et al. (2019). A large abrupt change in temperature at the front passage with constant temperature behind would have implied a very different temporal occurrence of FLF_T=1. In addition, the continuous cooling can also be explained by the turbulent mixing under the NLLJ core, which mixes the upper layer with lower layers cooled by radiative and sensible flux divergence (Adler et al., 2019; Babić et al., 2019a).

• The FLF_mean with values equal to 1 combines the wind increase and the cooling, both of which occur simultaneously during the entire night; the largest number of cases is observed between 18:00 and 21:00 UTC.

Figure 5Temporal occurrence of fuzzy logic functions equal to 1 for temperature (FLF_T), wind speed (FLF_ws), and mean fuzzy logic function (FLF_mean) for the period between 1 and 30 July.

Download

The MI arrival times are shown in Fig. 6a. Four estimates of the MI arrival times are displayed, one using the 302 K potential temperature criterion and three corresponding to the first time when the three fuzzy logic functions, FLF_ws, FLF_T and FLF_mean, attain a value of 1. Most of the observations of the MI arrival time at Savè considering only the wind speed increase fell between 16:00 and 18:00 UTC, while they fell between 16:00 and 21:00 UTC when we considered only the cooling. The arrival time of the MI deduced from FLF_mean, which couples with an equal weight cooling and wind speed increase, is observed between 16:00 and 00:00 UTC, with a median arrival time at 19:00 UTC. As noted above, the different tests performed to select the constants and thresholds for the fuzzy logic method yield different MI detection times for each day but quite similar distributions for the period of study. These results suggest that the MI arrival time is difficult to detect with local measurements. However, the MI arrival time as detected by the fuzzy logic function FLF_mean is clearly linked to the mean monsoon flow in the afternoon (Fig. 7): the stronger the monsoon flow strength in the afternoon between 12:00 and 15:00 UTC, the earlier the MI arrival time. The absolute value of the correlation coefficient between both is 0.61. The two exceptionally early arrivals at 16:00 and 16:30 UTC shown in Fig. 7 and put in Table 1 are associated with unusually strong monsoon flow all day long (e.g., the nights of 10–11 and 11–12 July).

Figure 6(a) Distribution of MI arrival times determined using the 302 K potential temperature criterion and the fuzzy logic function method (FLF_T, FLF_ws, and FLF_mean) for the period between 1 and 30 July 2016. (b) Distribution of NLLJ onset and breakup times at Savè during the period from 20 June to 30 July 2016. The MI arrival times detected with FLF_mean, already presented in panel (a), are also indicated.

Download

The NLLJ onset and breakup cannot be detected prior to 17:00 UTC and after 08:00 UTC, respectively, because of the stability criterion. The most frequent onset and breakup times of the NLLJ are 17:30 and 07:00 UTC, respectively (Fig. 6b). The MI arrival time detected with FLF_mean=1, is also indicated in Fig. 6b. One can note that, with the exception of the two exceptionally early MI arrival times, the NLLJ onset and MI arrival time distributions are quite similar to one another. Table 1 summarizes, for each day of the 20 June through 30 July period, the MI arrival time using FLF_mean, NLLJ onset and breakup times, in addition to the rainy and density current cases noted before.

Figure 8 shows a 15 min distribution of the jet core height, strength, and direction for the 20 June through 30 July period. The median jet core lowest height (350 m) is observed when the NLLJ settles. The jet core rises after 22:00 UTC to reach about 500 m at NLLJ breakup. The median strength of the jet varies between 7 and 11 m s⁻¹ (Fig. 8b), and the median direction shifts slightly from 200^∘ at NLLJ onset to 230^∘ at 03:00 UTC (Fig. 8c). The NLLJ core in Niger and Benin observed during the AMMA campaign (Lothon et al., 2008) was roughly at the same height, but the wind speed of the jet core is in contrast more pronounced in Niger, from 10 to 20 m s⁻¹.

5.1 LLSC detection

West African LLSCs have not been extensively studied in the past, generally due to a lack of observations. However, the systematic visual observations by local meteorologists at every meteorological station made over a time period of years constitute a unique dataset (Schrage and Fink, 2012). Besides this historical approach, several instruments have been used in the context of international projects to estimate LLSC characteristics from a vertical point of view: radiosoundings, cloud radar (Bouniol et al., 2012), and ceilometer (Schrage and Fink, 2012). Satellite data have also been used to analyze the horizontal variability LLSCs. Van Der Linden et al. (2015) and Bouniol et al. (2012), for example, made a climatology of the occurrence of the LLSCs in this region.

Here, we use a ceilometer to estimate the CBH (considering only CBHs below 1500 m a.g.l.), cloud radar to estimate the CTH and the IR sky camera to define the start and end of the LLSCs (Sect. 2.4). With the IR sky camera, we have the advantage of retrieving information about the occurrence of LLSCs, as well as their horizontal structure on a local scale.

Figure 7Scatter plot of the average monsoon strength between 12:00 and 15:00 UTC and the MI arrival time using FLF_mean=1, for the time period between 1 and 30 July 2016.

Download

Figure 8Distribution of the NLLJ (a) core height, (b) strength, and (c) direction at the Savè supersite. The color bar indicates the percentage of days (computed using the number of nights with NLLJ onset during the 20 June through 30 July period). The black lines in panels (a), (b), and (c) indicate the median jet core height, its strength, and its direction, respectively.

Download

To estimate the occurrence of stratus clouds at Savè, we consider the average color $[\stackrel{\mathrm{‾}}{\mathrm{R}},\stackrel{\mathrm{‾}}{\mathrm{G}},\stackrel{\mathrm{‾}}{\mathrm{B}}]$ defined by the color vector obtained when averaging all [R,G,B] pixels of the IR camera image, as well as the standard deviation σ_RGB. σ_RGB quantifies the color variability within each image, which represents the spatial variability of the lowest CBH within the $\mathrm{43}{}^{\circ }\times \mathrm{32}{}^{\circ }$ aperture:

where σ_R, σ_G, and σ_B are the standard deviations of the $\stackrel{\mathrm{‾}}{\mathrm{R}}$, $\stackrel{\mathrm{‾}}{\mathrm{G}}$, and $\stackrel{\mathrm{‾}}{\mathrm{B}}$ components, respectively, within the image.

A large standard deviation implies a large variability of temperature (i.e., a mix of clear sky and clouds) or spatial variability of the CBHs; a small standard deviation corresponds to either a homogeneous stratus deck or a fully clear sky.

Here, we set the occurrence of the low stratus clouds when (i) $\stackrel{\mathrm{‾}}{\mathrm{R}}>\mathrm{0.5}$, $\stackrel{\mathrm{‾}}{\mathrm{R}}>\stackrel{\mathrm{‾}}{\mathrm{G}}$, and $\stackrel{\mathrm{‾}}{\mathrm{R}}>\stackrel{\mathrm{‾}}{\mathrm{B}}$, (ii) ${\mathit{\sigma }}_{\mathrm{RGB}}<=\mathrm{0.15}$, for more than 2 h without any “break” longer than 30 min. Furthermore, the first time when these criteria are satisfied should occur before 06:00 UTC the following morning in order to detect only nocturnal LLSCs. The first criterion (i) corresponds to the search of a low base, with an average color close to red. The second criterion (ii) corresponds to the search for a relatively homogeneous cloud deck. The added persistence condition helps with defining a consistent nocturnal stratus cloud life cycle and avoids spurious variability. These criteria allow one to estimate, for each night, the onset and breakup times of LLSCs and therefore deduce their lifetimes. Note that during rain events, droplets are retained on the dome of the infrared camera and impact the color of the image as if there was a cloud. Therefore, LLSCs cannot be detected during rain events which are thus excluded. As far as we know, it is the first time that such methodology is used for the study of the stratus cloud deck formation and breaking.

Figure 9 shows two typical examples to illustrate this method using information provided by the ceilometer and the cloud radar. The evolution of the average color and the standard deviation during two nights (the nights of 7–8 July and 6–7 July are shown in Fig. 9a and b, respectively) are shown with the CBH estimated from the ceilometer and the CTH estimated from the cloud radar. Onset and breakup times deduced when applying the detection method are also indicated.

Figure 9Time series of the LLSC macrophysical characteristics observed on the nights of (a) 7–8 July 2016 and (b) 6–7 July at Savè. The colored band indicates the time series of the average image obtained every 2 min with the IR cloud camera. This time series is plotted arbitrarily at the 0 m abscissa for the sake of clarity. The grey line indicates σ_RGB (with its scale on the right axis) and the dashed grey line indicates for the threshold (0.15) above which the LLSC deck is not considered to be homogeneous. Black dots indicate the CBHs of the LLSCs estimated from the ceilometer. Blue dots indicate the CTHs of the LLSCs estimated from the cloud radar. The dashed black and red vertical lines indicate from left to right the NLLJ and LLSC onset and breakup times.

Download

For both examples, one can see how the smallest steady standard deviation of the IR image color consistently corresponds more to a steady low cloud base as seen by the ceilometer (see, for example, the data from 8 July from 03:00 to 08:00 UTC or the data from 7 July from 06:00 to 09:00 UTC). A larger standard deviation corresponds to more variability of the cloud base (e.g., on 8 July around 11:00 UTC or 7 July around 03:00 UTC or after 12:30 UTC).

For the night of 7–8 July (Fig. 9a), the criteria for LLSC detection are satisfied from midnight until 10:30 UTC. This period defines the LLSC lifetime for this day (Table 1). Before midnight, the sky is clear for a large part of the time (blue color with the IR camera); some clouds passed between 10:00 and 15:00 m a.g.l. The cloud base heights are variable according to the ceilometer; the IR color and σ_RGB do not satisfy the LLSC detection criteria. After 10:30 UTC, the cloud base rises and the fractionation of the cloud base (less steady red–pink color; large σ_RGB for a long duration of time) increases with the development of the convective boundary layer. It defines the end of the stratus. We consider the preceding lifting of the stratus deck from 07:30 to 10:30 UTC as being part of the LLSC lifetime (i.e., they meet our criteria for LLSC detection), which is the start of the so-called “convective phase” according to Babić et al. (2019a), who focused on the night of 7–8 July 2016 in their case study, and Lohou et al. (2019), who suggested a conceptual model of LLSCs' life cycle.

For the night of 6–7 July, the criteria for stratus LLSC are satisfied from slightly after midnight until 12:30 UTC (Table 1). The variability of cloud base found between 02:00 and 04:00 UTC, and revealed by higher σ_RGB and smaller $\stackrel{\mathrm{‾}}{\mathrm{R}}$ values, does not persist long enough to interrupt the LLSC definition.

Adler et al. (2019) used the ceilometer to estimate nocturnal LLSC characteristics (CBH, onset, and breakup times). These authors differentiated between LLSC stratus deck and preceding “stratus–fractus” phase using two lower-bound thresholds of 90 % and 50 %, respectively, for the cloud fraction computed within a 60 min time interval. An example of those two phases can be seen on the night of 6–7 July (Fig. 9b). One can see that low clouds occur between 21:00 UTC and midnight; they have a slightly higher base than later in the night, and there are long periods without clouds. This phase corresponds to the stratus–fractus phase defined by Adler et al. (2019), which precedes the “stratus phase”. Due to our criteria related to the persistence of LLSCs, this phase is excluded from our detection method. We instead detect the “stratus phase” that corresponds to full-deck LLSCs. Note that a very short stratus–fractus phase can also be seen in the previous example of the night of 7–8 July (Fig. 9a), from 23:30 to 00:10 UTC, which is also consistent with the work of Adler et al. (2019) and Babić et al. (2019a).

Clouds after 12:00 UTC for both cases correspond to convective shallow cumulus (within the “convective phase” noted by Babić et al., 2019a), as revealed by both a larger IR image standard deviation (σ_RGB) and the very large scatter in the ceilometer-based CBH between 1000 and 3000 m a.g.l. (Fig. 9). The formation of cumulus clouds below the LLSCs is one of the three LLSC breakup scenarios described in Lohou et al. (2019), but the shallow cumulus and fair weather cumulus are not detected by our criteria and indeed not under scope here.

5.2 LLSC lifetime statistics

Based on the detection criteria explained above, we noted a significant occurrence of LLSCs during the DACCIWA campaign at Savè: 65 % of the nights exhibit the development of a nocturnal stratus deck. This finding is consistent with the work of Kalthoff et al. (2018).

This prevalent occurrence is revealed in Fig. 10, which shows, from 15:00 UTC one day (D-1) to 18:00 UTC the following day, the MI arrival time and onset and breakup times of LLSC and NLLJ.

Figure 10The time series of the onset (solid lines) and breakup times (dashed lines) of NLLJ and LLSCs are shown in black and blue, respectively. The red colored circles indicate the arrival time of the MI at Savè using the mean fuzzy logical function. Solid black vertical lines delimit the different phases of the monsoon described by Knippertz et al. (2017). Dashed vertical lines indicate the vortex phase included in the post-onset phase.

Download

Figure 10 shows some intra-seasonal variability. The pronounced occurrence of LLSCs from 26 June through 11 July corresponds to the post-onset phase of the WAM. It is interrupted from 12 to 17 July by the drier vortex period identified by Knippertz et al. (2017) and noted above. From 23 to 26 July, the westerly moist regime observed (not shown here but seen in the UHF wind profiler data of Kalthoff et al., 2018) is associated with more rain (nearly on a nightly basis). This situation prevents LLSCs from being detected using the sky camera.

The shortest lifetime of the LLSCs was observed on 30 July. On this day, the onset LLSC is observed very late. For some cases, the breakup time is observed after noon, as on 3 July. A large variability in LLSC characteristics is observed: some nights exhibit steady LLSCs, and other nights have lots of intermittent LLSCs (figure not shown) (e.g.,  11, 21, 29 July). On the days that LLSCs form, they appear usually more than 3 h after the onset of the NLLJ and clear up after the NLLJ breakup time (Fig. 10).

Figure 11 gives more quantification of those time characteristics. It displays the distribution of the LLSC onset time and breakup time over the same set of 25 selected days as the monsoon flow statistics discussed above. Large variability in onset and breakup times of LLSCs is indeed observed. Overall, 26 % of LLSC onsets are observed before 00:00 UTC, and 78 % are observed before 03:00 UTC. All breakup times are observed after sunrise: 33 % of LLSC breakup times occurred before 10:00 UTC, and 74 % occurred before 12:00 UTC. We found that LLSCs always clear up after the NLLJ breakup time.

Figure 11Distribution of (blue) onset and (brown) breakup times of LLSCs at Savè from the period of 20 June through 30 July 2016.

Download