For the investigation of atmospheric dynamics, analysis and reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) are used. Most of the analyses are based on the ERA-Interim (hereafter ERA-I) reanalysis at about 0.7∘ grid spacing (Dee et al., 2011), which allows the computation of background climatologies back to 1979. For investigations focusing on the campaign period in 2016 alone, the higher-resolution operational analyses (native resolution of ∼ 9 km; model version Cy41r2; see www.ecmwf.int) are employed. As there was no change to the operational system during the study period, these data can be regarded as homogeneous, in contrast to longer time spans of operational data. The majority of radiosondes launched during the DACCIWA field campaign were distributed through the Global Telecommunication System and were assimilated at ECMWF. For the analysis of ocean influences on West Africa, the 0.25∘ daily Reynolds Optimum Interpolation SST data are used. The dataset combines observations from different platforms (satellites, ships, buoys) on a regular global grid. A spatially complete SST map is produced by interpolating to fill in gaps (Reynolds et al., 2007). Data have been retrieved from the NOAA NCDC (National Oceanic and Atmospheric Administration – National Climatic Data Center) FTP site (http://www.ncdc.noaa.gov). Monthly anomalies for June–July 2016 and daily anomalies are based on the 1981–2016 climatology.

As a precipitation estimate, the standard Tropical Rainfall Measuring Mission (TRMM) product 3B42 (v7) with 0.25∘ grid spacing is used (Huffman et al., 2007). This product combines information from space-borne radar and microwave and infrared channels, subject to monthly calibration with surface rain gauges if available. Since September 2014, the real-time calibration of microwave radiances using the precipitation radar has ceased due to the decommissioning of the TRMM satellite and was replaced by using climatological adjustments. Although this caused a discontinuity, the TRMM 3B42 product was prioritised over the Global Precipitation Measurement (GPM) Integrated Multi-satellite Retrievals for GPM (IMERG) successor product due to the longer availability (1998–2016), which allowed for the calculation of anomalies. The temporal resolution of this product is once every 3 h, but here daily accumulations (22:30–22:30 UTC) are used for most investigations. In addition, outgoing longwave radiation (OLR) data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on the geostationary Meteosat Second Generation (MSG) satellites with a spatial and temporal resolution of ∼ 3 km at nadir and 15 min, respectively, are used as a proxy for convective activity (Schmetz et al., 2002). In particular, channel 9 (i.e. approximately 9.80–11.80 µm) of the thermal infrared band is taken to retrieve cloud-top temperatures and to ensure day and night coverage. Different types of clouds are analysed using information on cloud-top characteristics (CTX) and the cloud mask (CMA) from the Satellite Application Facility on Climate Monitoring (CM SAF). Both the CTX and CMA subsets are part of the CLAAS-2 (Finkensieper et al., 2016) dataset, which is derived from information provided by SEVIRI (Stengel et al., 2013). Therefore, CLAAS data have the same temporal and spatial resolution as the SEVIRI dataset.

As this paper is meant to give a broad overview of meteorological and chemical conditions only, a detailed analysis of DACCIWA field campaign data is left to follow-up studies. The only exception is radiosonde data from Abidjan (for location, see Fig. 1) used to illustrate a period of unusual dryness during July 2016. Relative humidity was derived from (00:00, 06:00, 12:00 and 18:00 UTC) soundings four times daily using the high-resolution vertical profiles obtained from the MODEM radiosonde system. The analysis will concentrate on the main DACCIWA study region (8∘ W–8∘ E, 5–10∘ N, see Fig. 1), but influences on that region from a much wider area will be considered.

In order to better understand and characterise atmospheric variability during the DACCIWA campaign, a number of features important for SWA were objectively or subjectively identified: 1.

Equatorial waves: the presence of CCEWs is identified using the wave filtering method in specific wave-number–frequency domains as described in Wheeler and Kiladis (1999). In addition to the CCEWs (Kelvin waves, MJO, mixed Rossby-gravity and equatorial Rossby waves), tropical depression-like disturbances (TD) are filtered following the method by Roundy and Frank (2004). These often correspond to AEWs over West Africa. The filtering is applied to the 3-hourly TRMM 3B42 (v7) precipitation dataset (see Sect. 2.1) within the northern equatorial band 5–15∘ N, which contains the bulk of the convective precipitation during the campaign period but excludes some heavier oceanic rainfalls (see Fig. 5b).

In order to guide the discussion of the DACCIWA field measurements, the study period is divided into distinctive phases and the most significant weather systems are labelled for better reference in other papers. The division into phases is mainly based on the north–south precipitation difference (NSPD hereafter) between the coastal zone (0–7.5∘ N) and the Sudanian–Sahelian zone (7.5–15∘ N), both averaged across the longitude range 8∘ W–8∘ E (see Fig. 6 for orientation). Figure 5a shows daily values of the NSPD based on TRMM precipitation estimates for June–July 2016. Figure 5b shows the corresponding zonally averaged rainfall values against latitude. Four distinct phases are recognisable from this analysis.

Phase 1 lasts from 1 to 21 June 2016 and is characterised by a rainfall maximum near the coast. However, it shows large fluctuations with periods around 5 days (Fig. 5a). Particularly the middle part of Phase 1 is very wet, while the earlier and later parts are characterised by more isolated rainfall peaks near 4∘ N (Fig. 5b). This modulation is consistent with the QBZD index, showing a significant minimum around 14  2016 (see http://misva.sedoo.fr). Rainfall during this period is unusually intense offshore of the Niger Delta area stretching across the Gulf of Guinea towards Cape Palmas (Fig. 6a; see also Fig. S1a for anomalies). A second rainfall maximum is located over the tropical Atlantic to the west of West Africa in Fig. 6a, where SSTs are climatologically much warmer (not shown). Precipitation does already stretch far inland into the Sahel but amounts are relatively low with the exception of the Cameroon line highland region along the border of Nigeria and Cameroon. The moderate changes from drier to wetter and back to drier conditions in the Sahel during Phase 1 are reflected in weak but hardly significant undulations of the intra-seasonal Sahelian index reaching a minimum on 12 June 2016 (see http://misva.sedoo.fr). The pre-monsoonal conditions are also reflected in fields of zonally averaged total column water vapour (TCWV) with values above 45 mm mostly restricted to south of 12∘ N (Fig. 7). The ITD (identified by the 14 ∘C isoline of 2 m dew point) fluctuates around 16∘ N (Fig. 7).

Phase 1 corresponds closely to the period of anomalously strong SHL that fluctuates from east to west discussed in Sect. 3 (see Fig. 4), while correspondence to SST behaviour in the Gulf of Guinea (Fig. 3) is less clear. It is interesting to note that despite a strong SHL and an established ACT, rainfall remains strongest along the coast, indicating that monsoon onset has not yet occurred. This aspect will be discussed further in Sect. 4.3. In Fig. 5, significant synoptic disturbances are labelled with the capital letters A–J. These were subjectively identified from 850 hPa streamline plots at 00:00 UTC each day and often show a noticeable correspondence to the precipitation and TCWV behaviour. A summary of their most important characteristics is given in Table 1 and individual tracks are shown in Figs. 11, 13, 14 and 16. The locations of the feature labels in Figs. 5 and 7 correspond to the times when they cross the Greenwich meridian (i.e. centre of the DACCIWA focus region) and their latitudinal position (Figs. 5b and 7 only).

Phase 2 lasts from 22 June to 20 July 2016 and is characterised by a rainfall maximum inland with smaller and less regular fluctuations of the NSPD (Fig. 5a) and only occasional and weaker convective systems around 4∘ N (Fig. 5b). This indicates a fully developed WAM with a deeper penetration of rainfalls and TCWV into the continent, and a northward-shifted ITD, while marine precipitation is restricted to the Bight of Bonny and the waters along the West African west coast (Figs. 6b and 7). Large parts of the inland DACCIWA region were virtually dry during this period, much drier than in other years (Fig. S1b), despite relatively high TCWV values (Fig. 7). The transition from Phase 1 to Phase 2 is marked by strikingly dry conditions across most of the area of interest (Fig. 5b), much reduced TCWV (Fig. 7), strong fluctuations of the ITD (Fig. 7) and an abrupt breakdown of the SHL (Fig. 4a). During Phase 2 the SHL then gradually intensifies and shifts westward (Fig. 4). There is also a gradual increase in coastal upwelling during this period (Fig. 3), which is consistent with more stable, near-surface monsoonal winds. The behaviour of the ACT, which is relatively weak and shifted to the south during most of Phase 2, does not seem to be closely related to the precipitation shift.

During 21–26 July 2016 (Phase 3), the rainfall maximum shifts back to the coastal zone (Fig. 5a), accompanied by wet conditions spanning large parts of the latitude band from 1 to 22∘ N (Fig. 5b), where TCWV is enhanced and the ITD reaches its northernmost extension (Fig. 7). A horizontal distribution of rainfall during this period (Fig. 6c) shows unusually intense convection across the entire Gulf of Guinea, widespread rain across the entire Sudanian zone and more patchy local maxima stretching into the Sahel and even southern Sahara (see anomalies in Fig. S1c). Even larger amounts are found along the coast of Guinea and Sierra Leone. This wet phase is preceded and accompanied by a second breakdown of the SHL as well as a marked westward shift in its centre (Fig. 4). Coastal upwelling is increased during this period, while no major change in the ACT is seen (Fig. 3).

During the last 5 days of July 2016 (Phase 4), the WAM system returns to a more typical behaviour for this time of the year with a precipitation maximum in the Sahel, similar to Phase 2 (Fig. 5). As in Phase 2, the southern parts of the DACCIWA region are rather dry and coastal rainfalls are restricted to the Niger Delta region (Fig. 6d). Rainfall along the coast of Guinea, however, is even more abnormal than in Phase 3 (Fig. S1d). Overall, conditions are somewhat wetter than during Phase 2. This is accompanied by a partial recovery of the SHL (Fig. 4) and weakening of coastal upwelling (Fig. 3).

In the remainder of this section, the four phases outlined above as well as the transition between Phases 1 and 2 (the monsoon onset) will be analysed in detail, focusing on the synoptic-scale features labelled in Fig. 5. To aid the characterisation of these features, the following additional diagrams will be considered (see Sect. 2.2 for more details): (a) objective analyses of AEJ position and speed (Fig. 8), (b) Hovmöller plots of 850 hPa vorticity and meridional wind for the 4–18∘ N latitude band (Fig. 9), and (c) Hovmöller plots of equatorial wave disturbances in the 0–15∘ N band based on TRMM rainfall as well as tracks of long-lived MCSs (Fig. 10).

As stated in Sect. 4.1, the pre-onset period is characterised by a coastal rainfall maximum (Fig. 5), a strong eastward-shifted SHL (Fig. 4) and a weak ACT (Fig. 3). The AEJ is still located close to the coast during most of this phase (Fig. 8a; see also Fig. S2a). The first week (1–6 June) is relatively quiet with overall little rainfall across the region (Fig. 5b). The AEJ is anomalously far south (Fig. 8a) with a below-normal intensity (Fig. 8b). No significant coherent features are detected during this period, neither in 850 hPa vorticity and meridional winds (Fig. 9) nor in terms of filtered equatorial waves (Fig. 10). The enhanced vorticity feature starting on 5 June 2016 (Fig. 9) is related to a northern area of high horizontal wind shear (not shown) and is thus not associated with coherent meridional wind signals. The activity of long-lived MCSs is also relatively weak (black lines in Fig. 10; see also Fig. S3).

Between 7 and 15 June 2016 the AEJ begins shifting northward, showing two distinct mean speed maxima of more than 14 ms-1 (Fig. 8). On 7 June the jet maximum is located over southern Chad (not shown). The enhanced shear associated with this feature appears to have supported the formation of a large and long-lived MCS (Fig. 10) that brings substantial rainfall to southern areas (Fig. 5b) and thus creates a minimum in NSPD (Fig. 5a). In the following days, three relatively weak cyclonic disturbances cross the region (Fig. 11). As already mentioned, Table 1 provides a summary of the main characteristics of these disturbances and all subsequent ones. The first disturbance (labelled A in Fig. 11) propagates quickly westward from eastern Chad to northern Côte d'Ivoire between 9 and 11 June 2016, in accordance with the relatively strong AEJ during this period (Fig. 8b). When it passes the DACCIWA region, the increase in southerly flow seen in Fig. 9 (solid black lines) is associated with an increase in rainfall inland, while coastal rainfall is also still active, leading to an NSPD near zero (Fig. 5a). The vorticity signature of Feature A is relatively weak (Fig. 9), but there are several long-lived MCSs embedded in this system and the latter stages are identified as a TD (Fig. 10).

The immediately following second disturbance (labelled B in Fig. 11) propagates a little slower and on a more southern track from the border of Nigeria and Cameroon parallel to the coast out to the Atlantic past 20∘ W. When the centre of the vortex passes the DACCIWA region on 12 and 13 June, a strong increase in rainfall over the ocean is observed, creating a sharp minimum in NSPD (Fig. 5a). The slower propagation of this feature is consistent with the larger distance to the strong AEJ core near 9∘ N (Fig. 8). Feature B shows a more coherent signature in vorticity and meridional wind (Fig. 9), as well as TCWV (Fig. 7), and is identified as a TD with two very long-lived and intense MCSs embedded over the DACCIWA region (Fig. 10).

From 15 to 18 June 2016 the AEJ is weak and shifts northward (Fig. 8), while a third cyclonic feature becomes evident in the 850 hPa streamlines (labelled C in Fig. 11). It propagates relatively slowly from eastern Nigeria across the DACCIWA region, reaching northwestern Côte d'Ivoire by 00:00 UTC on 18 June. It is associated with a moderate increase in rainfall inland, while the coast is conspicuously dry, leading to a slightly positive NSPD (Fig. 5a). Interestingly, Feature C appears to be related to a longer-lived, somewhat patchy vorticity and meridional wind feature (white line in Fig. 9), which is also identified as a TD (Fig. 10). The vortex identified from the streamlines appears to move somewhat slower than this disturbance and is only found during the period of strongest southerlies immediately over the DACCIWA region (not shown). The vorticity feature moves at a similar speed as the embedded MCS (Fig. 10). Interestingly, during the middle part of Phase 1, the rainfall in the 5–15∘ N latitude band appears to be modulated by two Kelvin waves propagating across the DACCIWA region (green lines in Fig. 10), which superpose with the TD signals. There is also some indication for equatorial Rossby wave activity in the western part of the domain, but this signal is harder to see in the unfiltered TRMM data (Fig. 10).

The monsoon onset is often defined as a more permanent shift in the rainfall maximum into the continent (e.g. Fitzpatrick et al., 2015). According to the NSPD (Fig. 5a), this occurred on 21–22 June in 2016, the transition from Phase 1 to Phase 2. As this date is of such a large importance for the WAM, a dedicated discussion of the 5 days before and after this date is presented here. Overall this 10-day period has relatively low rainfall in the DACCIWA region, the two noteworthy exceptions being the enhanced coastal rainfall around 19 and 20 June and a Sahelian maximum on 24–25 June (Fig. 5b). Consistently, the activity of equatorial waves and long-lived MCSs is strongly suppressed (Fig. 10).

The synoptic development during the onset is characterised by substantial extratropical influences disturbing the circulation over northern Africa with high-amplitude waves and wave breaking in the subtropical upper troposphere. On 15 June 2016, the polar and subtropical jets merge over the Mediterranean Sea and a high-amplitude ridge is located upstream over the central North Atlantic (not shown). This period is characterised by a substantial drop in the inertial stability index defined by Cook (2015) but negative values are only reached for short periods (not shown). While the Atlantic ridge continues to propagate eastward, the downstream trough and ridge amplify strongly until 17 June. On this day, the trough stretches all the way to the Mauritanian coast and leads to a strong southwesterly flow at 600 hPa across the western Sahara (Fig. 12). It is conceivable that subsidence associated with the ridge stretching from eastern Europe into northeastern Africa (Fig. 12) contributed to the suppression of rainfall evident from Fig. 5b. The inflow of cool maritime air from the Atlantic Ocean leads to an abrupt ventilation of the SHL, causing a weakening of its intensity and rapid eastward shift in the centre (Fig. 4) with some resemblance to the situation Todd et al. (2013) refer to as the “maritime phase”. According to http://misva.sedoo.fr, the intra-seasonal SHL index reached a distinct maximum on 17 and 18 June 2016. The extreme nature of this cold surge is visible in 20∘ W–0∘ E averaged temperature anomalies at 850 hPa, showing a very distinct and unusual cooling during this period, with anomalies below -6 K north of 30∘ N and below -4 K down to almost 20∘ N (Fig. S4a). Over the next few days, the whole wave slowly drifts eastwards, allowing the northerlies associated with the trough to penetrate into the eastern parts of the Sahara as well. The associated cooling, which is overall less dramatic in the east (Fig. S4b), finally leads to the conspicuous collapse of the SHL between 20 and 25 June 2016 shown in Fig. 4a. This development is also reflected in an abrupt northward jump of the AEJ core accompanied by a significant weakening around 21 June 2016 (Fig. 8).

Within the large area of reduced surface pressure to the southeast of the trough that stretches unusually far south (Fig. 12), three cyclonic vortices form at 850 hPa (Fig. 13). The first one (labelled D1) first appears over the Hoggar Mountains in southern Algeria on 15 June 2016 and then remains rather stationary over northern Niger between 16 and 18 June 2016. The second (labelled D2) forms farther to the east and slowly moves along the border between Chad and Niger between 17 and 19 June 2016 (see also Fig. 12). The last one (labelled D3) is only discernable in 850 hPa streamlines on 19 and 20 June 2016 along the border of northern Chad and Sudan. On these two days, the three centres form a zonally elongated area of cyclonic rotation with marked northerlies to the west and southerlies to the east. Between 19 and 21 June 2016, D1 and D2 rotate cyclonically around each other while beginning to propagate westward in a fashion similar to an AEW (Fig. 13). Both cyclonic centres slow down and weaken between 22 and 25 June 2016 near the West African west coast. To the northwest of D1 and D2, northerly flow reaches values of 15–25 ms-1 between 18 and 21 June 2016 (not shown), which leads to a marked southward push of TCWV (Fig. 7).

Figure 9 shows how this unusual development is reflected in 850 hPa vorticity and meridional wind (labelled D). On 18 June 2016, D1 and D2 are still located to the north of 18∘ N and thus signals in the Hovmöller plot are weak. On 19 June 2016, the strong northerlies begin to penetrate into the DACCIWA region, helping to suppress rainfall (see Fig. 5b). This is followed by unusually large vorticity values on 20 June 2016, when D1 moves south of 18∘ N (Fig. 13). On 20 and into 21 June 2016 a wide area of very strong southerlies spreads across the DACCIWA region. These bring moisture far into the continent, shifting the ITD northward (Fig. 7), and thus create the conditions for an inland rainfall maximum between 23 and 25 June 2016, indicating that the onset has in fact occurred. After the turbulent transition phase, the WAM system becomes relatively quiet and the AEJ slowly gets re-established near its climatological latitudinal position until 26 June 2016 (Fig. 8) with the SHL also beginning to re-intensify (Fig. 4).

The analysis above strongly suggests that in 2016 the monsoon onset was triggered by very strong interactions with the midlatitudes that supported a suppression of rainfall over West Africa. Low-rainfall conditions around the onset have been documented for other years as well (Sultan and Janicot, 2003).

Phase 2 comprises a period of relatively undisturbed monsoon conditions. The entire DACCIWA aircraft campaign fell into this period (see Flamant et al., 2017). The NSPD is positive through most of this phase and is modulated by the significant weather systems E–I, with centres between 12 and 16∘ N and thus farther north than the Phase 1 features A–C (see Fig. 5). This period was anomalously dry along the Guinea Coast (Fig. S1b). Features E–I also modulate the speed and latitude of the AEJ (Fig. 8).

At the very beginning of this phase, between 23 and 26 June 2016, while the monsoon is still being established, a relatively weak (and therefore unlabelled) cyclonic feature crosses the southern part of SWA, creating some moderate rainfalls in the Sahel around 24 June 2016 (Fig. 5b). This is the first time during the DACCIWA campaign that the precipitation maximum has fully shifted inland. After this system, the SHL starts intensifying and shifts northward (Fig. 4a), the AEJ accelerates (Fig. 8b) and a deep southwesterly monsoon flow gets established (not shown).

Between 27 June and 8 July 2016, three AEWs (Fig. 14a) associated with moderate fluctuations in TCWV (Fig. 7) develop. The first two (Features E and F) form on a relatively well organised AEJ (Fig. 8b) and show rather classical propagation characteristics with coherent signals in meridionally averaged vorticity and meridional wind at 850 hPa (Fig. 9). Both have two cyclonic centres at 850 hPa straddling the jet to the north and south (Fig. 14a) and are also objectively identified as TDs (Fig. 10). Feature E, which forms near the Greenwich meridian, appears to be associated with the slight rainfall enhancement on 26 and 27 June 2016, while Feature F forming near 12∘ E creates a peak in rain in the northern box on 29 June 2016 (Fig. 5b). Finally, Feature G forms during a period when the AEJ weakens and becomes more fragmented, which makes its latitudinal position vary strongly (Fig. 8). This leads to less clear and slower propagation behaviour (Fig. 14a). The northern centre propagates from central Niger to eastern Mauritania between 3 and 6 July 2016 and then drifts northwestward towards the border with Western Sahara. This behaviour is accompanied by a rapid shift in the SHL to the west (Fig. 4b). The southern centre shows a less coherent propagation. This is consistent with the relatively patchy signals in wind and vorticity shown in Fig. 9, apart maybe from the final stages over the open Atlantic Ocean. Feature G is also not matched with a TD like the previous features are (Fig. 10). Nevertheless, a marked increase in rainfall is observed when this feature crosses the DACCIWA region on 4 July 2016 (Fig. 5b).

After that, between 9 and 16 July 2016, a fundamentally different and quite unusual development occurs. While in the north, a cyclonic feature slowly tracks from eastern Mali to Cabo Verde between 8 and 13 July 2016 and then out to the Atlantic (Feature H1 in Fig. 14b), there is no clear corresponding southern vortex. Instead, an anticyclonic system (H2) slowly propagates from Gabon on 11 July across the tropical eastern Atlantic, reaching the coast of Sierra Leone on 14 July 2016, after which it begins to weaken over the ocean to the west. As this system moves a little faster than its cyclonic counterpart to the north, the two centres approach each other, creating an area of marked low- to mid-level southwesterly winds in between them, particularly on 12–14 July 2016 along the western border of the DACCIWA region (arrows in Fig. 14b). This behaviour is associated with a weakening and northward shift in the AEJ (Fig. 8). It is conceivable that these westerly wind anomalies also helped to intensify coastal upwelling as shown in Fig. 3. Given the zonal distance between the two centres, both positive and negative vorticity signals are apparent in the Hovmöller plot shown in Fig. 9, although the negative one is only strong past 10∘ W. Propagation of these two features is relatively slow, with about 7 ms-1. While the signal in the northerlies at 850 hPa is somewhat patchy, the signal in the southerlies, created by the positive superposition of the wind disturbances associated with the staggered northern and southern vortices, is coherent and strong, particularly to the west of 10∘ W (Fig. 9), as also reflected in TCWV (Fig. 7). This has likely supported a deeper inland penetration and slight intensification of rainfall (Fig. 5b). Given the somewhat unusual behaviour of this system, it is no surprise that there is no matching between the TDs and long-lived MCSs objectively identified during this period (Fig. 10). This propagating cyclonic–anticyclonic vortex couplet appears unrelated to any of the classical equatorial waves, but the slow propagation speed and the opposing circulation centres are consistent with the 6–9-day wave regime described by Diedhiou et al. (1999). To the best of our knowledge, the dynamical origin of such features is still somewhat unclear. In particular, the southern origin of the anticyclonic centre and its faster propagation seem unusual.

An interesting effect on the coastal region is that the southern anticyclonic vortex (Feature H2 in Fig. 14b) appears to have brought with it dry air from the area of subsidence in the equatorial zone or even SH. To illustrate this, Fig. 15 shows a time series of radiosoundings made four times daily from Abidjan for the period 7–16 July 2016. While most days show a well-developed monsoon layer with high relative humidity and winds from westerly directions, very dry air suddenly intrudes into the 850–700 hPa layer at midday on 11 July (drop from ∼ 85 to under 20 %) persisting until the morning of 14  July 2016. Winds blow from southwesterly to westerly directions during this period. Indications of aged aerosol particles were found when the DACCIWA research aircraft penetrated this layer over several days. These aged aerosol particles were likely from fires in the SH (Flamant et al., 2017). Interestingly the usually easterly 600–925 hPa shear vector backs to northerly just before the dry event – signalling the changes in the 600 hPa circulation due to the vortex couplet.

Phase 3 is characterized by wet conditions stretching from the tropical Atlantic far into the Sahel and even southern Sahara, particularly on 23–25 July 2016 (Fig. 5). Rainfalls are most abundant over the ocean, particularly off the coast of Nigeria and stretching west to Côte d'Ivoire as well as off the coasts of Liberia and Guinea (Fig. 6), creating large positive anomalies (Fig. S1c). There is also a marked Sahelian band that dips south into Ghana. The relatively large rainfall over the ocean coincides with an area of enhanced meridional SST gradients (Fig. 3) that influence surface wind convergence. Within the DACCIWA region (8∘ W–8∘ E), rainfalls to the south of 7.5∘ N dominate, leading to a sharp drop in NSPD to negative values during this period (Fig. 5a). Similar to the monsoon onset period during the transition from Phase 1 to Phase 2, Phase 3 is characterised by a breakdown of the SHL (Fig. 4a) and low inertial stability according to the index defined by Cook (2015) (not shown), but this time the responsible cold intrusion from the midlatitudes occurs over northeastern Africa (Fig. S4b), leading to a marked westward excursion of the SHL (Fig. 4b). Triggered by an upper-wave in the subtropics, an intensification of the low-level northeasterlies is observed, particularly on 23 and 24 July 2016 over Egypt and Sudan. Such a situation has been referred to as a cold surge by Vizy and Cook (2009), and this type appears to be more frequent than the event in the west before the onset (see Sect. 4.3).

An unusual and interesting synoptic development leads up to this event. On 17 July 2016, an anticyclonic centre appears over the northern Central African Republic and swiftly propagates westward, reaching the border between Nigeria and Benin on 19 July 2016 (propagation speed ∼ 12 ms-1; Feature I2 in Fig. 16). Strong southerlies ahead of this system lead to an increase in TCWV (Fig. 7). After this day, the vortex slows down substantially (average propagation speed ∼ 4 ms-1) and shifts to a more southern track just off the Guinea coast and then out to the open Atlantic, reaching 25∘ W on 27 July 2016. During this period the vortex centre is not always clearly identifiable. Somewhat similar to Feature H (see Fig. 14b), the anticyclonic vortex is accompanied by a cyclonic centre to the north, which is first evident in streamlines at 850 hPa on 19 July 2016 close to the border between northeastern Nigeria and Chad (Feature I1 in Fig. 16). During the slow propagation phase from 21 to 26 July 2016, when the two centres are almost aligned latitudinally with a distance of about 10∘, a strong westerly jet develops between them with a maximum near 10∘ N, which also propagates westward. The occurrence of this jet (see arrows in Fig. 16), which brings large amounts of moisture into SWA from the west, where much warmer SSTs prevail, exactly marks the beginning and end of Phase 3. Nicholson (2009) and Nicholson and Webster (2007) have shown that summers with a strong westerly flow at 850 hPa are on average associated with particularly wet conditions. The situation discussed here is therefore one possible synoptic-scale manifestation of this climatological result. Finally, southerlies to the east of I1 lead to a northward extension of the moist zone and an unusual ITD position at 24∘ N (Fig. 7).

As with Feature H, the increasing westerly flow between the cyclonic and anticyclonic centres leads to a weakening and north- and eastward shift in the AEJ (Fig. 8, Fig. S2c) as well as an increase in coastal upwelling (Fig. 3). The weakening of the AEJ may also be related to the weakened SHL discussed above. However, in contrast to Feature H, the latitudinal alignment of the vortices leads to a cancellation of signals in the Hovmöller diagram, making Feature I barely detectable in Fig. 9. So unusual is this situation that there is also very little in terms of objectively identified wave features during this period (Fig. 10). However, some modulation of rainfall by an eastward-propagating Kelvin wave is evident between 18 and 22 July 2016. Mounier et al. (2008) also discuss enhanced westerly inflow, moist conditions and a Kelvin wave influence in connection with the QBZD but the match with their concept is hard to establish for a single case. Feature I is too slow to match the 6–9-day wave regime described by Diedhiou et al. (1999).

On 26 July 2006 the widespread rainfall characterising Phase 3 ceases and the precipitation maximum shifts back to the Sahel for the rest of the DACCIWA campaign period, as indicated by a positive NSPD (Figs. 5 and 6d). There is also evidence for a return of the SHL and the ITD to more climatological positions and intensity (Figs. 4 and 7). Thus, overall this phase marks the return to more undisturbed monsoonal conditions similar to Phase 2. A last significant cyclonic feature occurs during this period (Fig. 16, labelled J). This feature is first detected over South Sudan on 23 and 24 July 2016. Until 27 July it swiftly crosses the DACCIWA region, reaching the Guinea highlands. After that, it slows down over the Atlantic on 28–30 July 2016. Figure 8 shows that the period of fast propagation is concomitant with an enhanced and southward-shifted AEJ. Feature J can be identified well in the north–south averaged 850 hPa vorticity in the eastern and western parts of the study region but is somewhat diffuse around 10∘ W (Fig. 9). Nevertheless, the fast-propagation phase is evident from the vorticity as well. Meridional wind signals associated with Feature J, however, are rather weak and only the western parts are concomitant with an objectively identified TD, which in turn appears to be related to a long-lived MCS (Fig. 10). Feature J also creates some mild fluctuations in TCWV (Fig. 7).