Impact of air–sea coupling
Comparing GA6-GOML and GOML-OBS (and GA6-OBS) cleanly identifies the impact of air–sea coupling (see Sect. 2.1, Table 2), analysis that is made possible by using the MetUM-GOML configuration. The seasonal cycle of precipitation over the LDS region (Fig. 4a) from these three simulations is shown in Fig. 6a. All three simulations show similar seasonal cycles that agree with GPCP from January-August and December, but underestimate the second wet season during September-November. The correlation matrix in Fig. 6b shows strong correlations, with coefficients greater than 0.9, between the three seasonal cycles, and statistically significant positive correlations with GPCP, with coefficients greater than 0.81. Some slight differences between GA6-OBS and GA6-GOML (Fig. 6) suggest that small SST biases in GOML-OBS influence the precipitation seasonal cycle here. Including air–sea coupling, while maintaining the same mean SST, has a minimal impact on the representation of the seasonal cycle over the LDS region and does not improve the intensity of the second wet season.
Impact of ocean mean states
Comparing GOML-GC2 to GOML-OBS isolates the effect of the ocean mean states on the seasonal cycle of precipitation over the southern part of West Africa (see Sect. 2.1, Table 2). Both GC2–GC2 and GOML-GC2 misrepresent the seasonal cycle, with one wet season per year, with the peak in rainfall occurring when the LDS should occur (Fig. 6a), hence the inclusion of coupled model ocean mean state biases leads to the incorrect seasonal cycle. Figure 6a shows that the difference between GOML-GC2 and GOML-OBS is much greater than the difference between GOML-OBS and GA6-OBS, indicating that GC2 ocean mean state biases have a bigger impact on the seasonal cycle of precipitation in the LDS region than the inclusion of air–sea coupled physics.
The seasonal cycle from GOML-ATL-N96 (Fig. 6a) shows similar patterns to GOML-GC2, with one wet season per year, peaking in July/August, during the observed LDS. GOML-ATL-N96 underestimates rainfall relative to GOML-GC2-N216, but has similar rainfall totals with GOML-GC2-N96 (Fig. 6a), suggesting this underestimate is related to horizontal resolution rather than differences in the ocean mean state.
The correlation of the mean annual rainfall cycle across the simulations (Fig. 6b) demonstrates greatest agreement between the simulations with the same ocean mean state (e.g. between GA6-OBS and GOML-OBS, and between GC2–GC2 and GOML-GC2). Conversely, agreement is much lower between simulations with the same model and resolution but different ocean states (e.g. GOML-OBS and GOML-GC2). Figure 6b indicates better agreement between GOML-ATL-N96 and either GOML-GC2-N96 or GOML-GC2-N216 (correlation coefficients of 0.974 and 0.931 respectively), than between GOML-ATL-N96 and GOML-OBS (correlation coefficient of 0.304). Thus we surmise that the incorrect representation of the seasonal cycle of rainfall over the southern part of West Africa in GOML-GC2 is related to GC2 Atlantic Ocean mean state biases.
The incorrect seasonal cycle for simulations with coupled model ocean mean state in the Atlantic (GC2–GC2, GOML-GC2 and GOML-ATL-N96, Fig. 6a) can be partitioned into a number of components: a late onset and deficient rainfall in May; excess rainfall in July-August, during the peak of the LDS; and insufficient rainfall in October (seen in all simulations). The first two factors, which are not exhibited in GA6-OBS and GOML-OBS, will be explored further.
To compare the location of the rainfall among simulations, Fig. 7 shows the mean monthly position of the TRB (Sect. 2.3), in May and August. As in Fig. 6, the GOML simulations are similar to the GA6 and GC2 simulations with the same ocean mean state; GOML-ATL-N96 is similar to GOML-GC2 and GC2–GC2. In May and August all simulations place the TRB south of the observed position in GPCP, especially in those simulations with GC2 ocean state in the Atlantic (GC2–GC2, GOML-GC2 and GOML-ATL-N96).
In May, the TRB is just south of the coastline in GPCP (Fig. 7a). The TRB in GA6-OBS and GOML-OBS is just south of the GPCP mean position (Fig. 7a), but in GC2–GC2, GOML-GC2 and GOML-ATL-N96 the TRB is further south, just north of the equator. The northern and southern limits (solid and dashed lines respectively; Fig. 7c) confirm this southward bias; the TRB is over approximately \(0^\circ \hbox {N}\)-\(10^\circ \hbox {N}\) in GPCP, GA6-OBS, and GOML-OBS, but over approximately \(5^\circ \hbox {S}\) to the coastline in GC2–GC2, GOML-GC2 and GOML-ATL-N96, consistent with the lower rainfall in May over the southern part of West Africa (Fig. 6). Previous studies suggest that the southward bias in mean TRB position is related to warm SST biases in the Gulf of Guinea (Fig. 1, Roehrig et al. 2013), which will be discussed in more detail in Sect. 4.3. Consistent results across GOML-GC2 and GOML-ATL-N96 confirm this bias is related to Atlantic Ocean mean state SST errors.
In August, the TRB is over Burkina Faso in GPCP, while GA6-OBS and GOML-OBS exhibit a southward shift, with the TRB over northern Ghana and Ivory Coast (Fig. 7b). Again, GC2–GC2 and GOML-GC2 place the TRB even further south, with GOML-ATL-N96 exhibiting an additional southward bias (Fig. 7b). The position of the northern boundary is the same in four simulations (GA6-OBS, GC2–GC2, GOML-OBS and GOML-GC2), passing through Senegal, Southern Mali and along the southern boundary of Niger (Fig. 7d). The key difference between these simulations is related to the position of the southern boundary, which leads to the differences in mean position (Fig. 7b). In GPCP, GA6-OBS, and GOML-OBS the southern part of the Ivory Coast and Ghana are outside the southern limit of the TRB in August, consistent with the low rainfall in August (Fig. 6a) and the correct representation of the LDS. In GC2–GC2, GOML-GC2 and GOML-ATL-N96 the southern limit of the TRB is south of the coastline between \(20^\circ \hbox {W}\) and \(10^\circ \hbox {E}\), consistent with the high rainfall over the southern part of West Africa and the incorrect representation of the LDS. The different positions of the southern boundary over the LDS region (\(10^\circ \hbox {W}\)–\(2^\circ \hbox {E}\)) can clearly be seen in Fig. 7d. This indicates that the incorrect representation of the LDS in simulations with GC2 SST biases is not solely related to an overall southward shift of the TRB, but may also be related to more local factors (see Sect. 4.4), including differences in regional patterns of ascent and descent. GOML-ATL-N96 exhibits a southward shift in both the northern and southern boundaries, which is related to horizontal resolution; see Supplementary Information for Fig. 7 replicated at N96 (Fig. S3).
In the next sections May and August are considered separately, and factors related to the rainfall biases in these months are presented. In Sect. 4.3 the southward bias in the TRB position in May, and associated patterns of wind and SST biases are discussed, while in Sect. 4.4 the rainfall overestimate in August is explored together with the patterns of ascent and descent along the coastline.
Southward Bias in the TRB position in May
In May, simulations using coupled model ocean mean state (GC2–GC2, GOML-GC2, and GOML-ATL-N96), which includes a warm bias over the southern tropical Atlantic, underestimate rainfall over the LDS region (Fig. 6a) as part of a wider southward bias in the position of the tropical rain belt (Fig. 7a, c).
A number of studies have identified a southward bias in the ITCZ in CGCMs, and associated this with SST biases over the tropical Atlantic (Richter and Xie 2008; Richter et al. 2012; Roehrig et al. 2013; Toniazzo and Woolnough 2014). Coupled climate models, including GC2 (Fig. 2), exhibit a large warm bias in the south east tropical Atlantic, peaking at the Angola/Namibia coastline and extending north-west towards the equator, covering much of the basin (Eichhorn and Bader 2017). Furthermore, coupled climate models fail to capture the equatorial cold tongue that forms in the eastern equatorial Atlantic during boreal summer (Fig. 1); combined with the cold bias to the west, this reverses the equatorial zonal SST gradient (Richter et al. 2012). SST sensitivity experiments have shown that improved representation of Atlantic SSTs (Eichhorn and Bader 2017), and in particular the Atlantic cold tongue, improves the onset and seasonal evolution of the West African monsoon (Steinig et al. 2018), as colder SSTs in the cold tongue enhance the land-sea temperature contrast and strengthen the monsoon flow (Okumura and Xie 2004; Chang et al. 2008).
GOML-GC2 and GOML-ATL-N96 show a cold SST bias north of the equator and warm bias south of the equator in May (Figs 1, 8), which likely contributes to the southward bias in the position of the TRB by altering the interhemispheric temperature gradient. A warmer Southern Hemisphere (and cooler Northern Hemisphere) is associated with a northward cross-equatorial atmospheric energy transport and a southward displacement of the tropical rain belt (Hwang and Frierson 2013; Hawcroft et al. 2017).
ERA-I and observed SST (Smith and Murphy 2007) exhibit a northwest-southeast temperature gradient across the tropical Atlantic, with south-easterly winds from the cooler waters off the Angola/Namibia coastline towards the warmer western equatorial Atlantic (Fig. 8a). The same pattern is found in GOML-OBS, with small biases (Fig. 8b). GOML-GC2 and GOML-ATL-N96 (Fig. 8c, d) contain a simpler north-south temperature gradient in the equatorial region, demonstrated by the warm bias in the east and cool in the west, with associated northwesterly wind anomalies between \(0^\circ \hbox {S}\) and \(5^\circ \hbox {S}\). These wind biases are also likely to be linked to the southward shift of the TRB. Although the investigation of the relationship between biases in Atlantic SST, wind and precipitation has been the focus of many studies (Okumura and Xie 2004; Richter and Xie 2008; Richter et al. 2012, 2014), establishing causal mechanisms remains a challenge, as in other basins (Shonk et al. 2018).
Richter and Xie (2008) and Richter et al. (2012) argue that the westerly bias in surface winds over the equatorial Atlantic during boreal spring, also present in atmosphere-only simulations, causes equatorial Atlantic SST biases. Weakened easterlies are associated with a deeper thermocline in the east and reduced equatorial upwelling, which inhibits equatorial cold tongue formation. Similarly, Fig. 8b shows small north-westerly wind biases in the western equatorial Atlantic in May. Voldoire et al. (2019) found that imposing the correct wind stress over the equatorial Atlantic reduces biases in SST and equatorial thermocline depth. The eastern warming and western cooling in turn induces westerly wind biases via a Bjerknes feedback mechanism (Richter and Xie 2008). Richter et al. (2012) propose that this westerly wind bias originates from excess convection over tropical Africa and reduced convection over South America, which initiates a pressure gradient that drives the westerly wind anomalies (Richter and Xie 2008). In addition, Richter et al. (2014) highlighted the role of latitudinal position of the boreal spring ITCZ on equatorial surface winds, with a southward shift of the ITCZ linked to the westerly wind bias at the surface. The same pattern of biases is seen in Fig. 8 (and Fig. 7), which may suggest that the same processes and feedbacks are active in GOML-GC2 (and GOML-ATL-N96). Additionally, other studies have noted the role of the West African monsoon winds on SST, as the cross-equatorial southerlies induce Ekman upwelling south of the equator that cools the eastern equatorial Atlantic (Okumura and Xie 2004; Hagos and Cook 2009). Reduced cross-equatorial southerlies, as seen in Fig. 8c, d, will therefore also reduce equatorial upwelling in fully coupled simulations, contributing to the warm bias.
The results here demonstrate that ocean mean state biases in the Atlantic are associated with a southward shift of the TRB in boreal spring, related to changes in the meridional temperature gradient, and equatorial wind biases, which also affect and respond to the position of the tropical rain belt. Further investigation is required to investigate the complex interplay of factors, including precipitation, wind and SST biases that develop over the Atlantic during boreal spring in coupled simulations.
Overestimation of rainfall during the August LDS
In order to understand the overestimation of rainfall during the August LDS, vertical cross sections of zonal wind and vertical velocity compare regions of ascent and descent and rainfall in the GOML simulations with ERA-I reanalysis over the LDS region (Fig. 9). For August, ERA-I and GOML-OBS show similar patterns, in agreement with Nicholson (2009), Nicholson (2013) and James et al. (2018). Two regions of ascent are identified: one centred around \(20^\circ \hbox {N}\) (shifted slightly south in GOML-OBS) and another deeper region centred around \(10^\circ \hbox {N}\). The ascent at \(20^\circ \hbox {N}\) corresponds to the surface ITCZ (Nicholson 2009), while most of the rainfall is associated with the ascent at \(10^\circ \hbox {N}\), just north of the coastline. Both ERA-I and GOML-OBS have a weaker, more southerly rainfall peak compared with GPCP (dashed black line). The southward shift of the northern region of shallow ascent in all GOML simulations when compared with ERA-I may indicate that the surface ITCZ does not propagate far enough north. This may be related to the dry bias over the Sahel in JJA seen in both GA6-OBS and GC2–GC2 (see Supplementary Information) and the southward shift of the TRB in Fig. 7.
Descent over the northern Gulf of Guinea (Fig. 9a, b), which encroaches onto the southern part of West Africa, caps the shallow ascent along the coastline, and gives lower rainfall totals here. ERA-I and GOML-OBS show reduced precipitation along the coast, consistent with the LDS; shallow ascent prevails at the coast due to upper level descent. While GOML-GC2 and GOML-ATL-N96 (Fig. 9c, d) also capture the two main regions of ascent, they do not capture the region of descent encroaching onto the coastline. The ascent at the coastline is deeper, associated with a second rainfall peak on the coast, consistent with earlier results showing rainfall along the coastline in August in GOML-GC2 and GOML-ATL-N96 (Fig. 6a). The ascent in GOML-ATL-N96 at \(10^\circ \hbox {N}\) is weaker than in GOML-GC2, but this is a consequence of resolution rather than ocean mean state biases (see Supplementary Information). All simulations show a southward shift in the position of the African Easterly Jet (AEJ) compared to ERA-I: while in ERA-I (and Nicholson 2013) the axis of the AEJ is north of the main region of ascent, the GOML simulations show the axis of the AEJ co-located with the ascent at \(10^\circ \hbox {N}\). James et al. (2018) also identified a southward shift in the AEJ in GC2. This may indicate errors in the representation of the meridional temperature gradient, as Parker and Diop-Kane (2017) report that the AEJ is in approximate thermal wind balance with the lower tropospheric temperature gradient. Convection occurs more frequently south of the AEJ than north of the AEJ (Parker and Diop-Kane 2017), hence a southward bias in the position of the AEJ is consistent with the southward shift of the TRB in Fig. 7 in all simulations. Since GOML-OBS, GOML-GC2 and GOML-ATL-N96 all contain a southward bias in AEJ position and TRB position, yet only those simulations forced with Atlantic SST bias (GOML-GC2 and GOML-ATL-N96, Fig 9c, d) fail to capture the LDS, this supports the conclusion from Fig. 7d that the LDS in August is associated with local factors. The stronger AEJ in GOML-ATL-N96 compared with GOML-GC2 and GOML-OBS is not a consequence of resolution (see Supplementary Information, Figure S4), and is driven by other factors.
Figure 9 suggests that the descent above 500hPa and limited ascent along the coastline is associated with reduced rainfall over the coastline during August in ERA-I and GOML-OBS. In GOML simulations forced by the coupled model ocean mean state over the Atlantic (GOML-GC2 and GOML-ATL-N96) the region of descent is shifted south, the ascent along the coastline is deeper, and higher rainfall is seen along the coastline. Parker and Diop-Kane (2017) state that high pressure over the Gulf of Guinea extends onto the coastline in July-August, with the associated descent inhibiting rainfall, leading to the LDS. Over the northern Gulf of Guinea, GOML-GC2 and GOML-ATL-N96 exhibit lower mean sea-level pressure in August, compared with GOML-OBS/GOML-OBS-N96 (not shown). Although it was not quantitatively shown, Odekunle and Eludoyin (2008) and Odekunle (2010) also proposed that increased static stability over the coastline limits convection and leads to the reduced rainfall associated with the LDS. They suggest that this increased static stability results from the cool SSTs along this coastline during the boreal summer, due to local upwelling and the advection of cold upwelled waters from other regions. Similarly, Parker and Diop-Kane (2017) note that the LDS is weak or absent where warm onshore waters persist, for example, to the east around the Niger delta in Nigeria and off the coast of Liberia. Upwelling between the Liberia/Ivory Coast border and Ghana is a consequence of the non-linear dynamics of the Guinea Current and its detachment from the coast, while upwelling east of Ghana is driven by local winds (Djakouré et al. 2017), hence reduced upwelling in coupled models is consistent with poor representation of the Guinea Current and the westerly wind biases present over this region from June-August (result not shown).
Figure 9 demonstrates that when GOML is constrained to the observed ocean state, with cooler SSTs in August (Fig. 1d–f), upper level descent reduces rainfall along the coastline, whereas the introduction of GC2 ocean mean state biases, including a warm bias over the northern Gulf of Guinea (Fig. 1j–l), leads to ascent along the coastline, preventing occurrence of the LDS in those simulations. Further investigation, with additional simulations, is required to elucidate specific regions of influence and mechanisms.