1 Introduction

Central equatorial Africa (CEA) (10o S–10o N; 10o–30o E) hosts the Congo basin (CB) whose important climatic role extends well beyond the African continent. The Basin is known as one of the three hot spots of major convective activity in the global tropics (Webster 1983), experiencing the highest lightning strike frequency of anywhere on the planet (Jackson et al. 2009) and receiving around 1500–2000 mm of rainfall per year (Dezfuli 2017). In addition, its rainforest stores incredible amounts of carbon, preventing it from being emitted into our atmosphere and fueling climate change. At the local scale, through evaporation, tropical forests and woodlands exchange vast amounts of water and energy with the atmosphere, controlling the seasonality of rainfall in the region (Crowhurst et al. 2020).

Despite its importance in the local and global climate systems, CEA remains a very understudied region compared to other parts of Africa (Washington et al. 2013; Creese and Washington 2016). This is mainly due to a lack of observational data (Washington et al. 2013; Creese and Washington 2018). Although much remains to be done to cover the gap, several studies (Cook 1999; Nicholson and Grist 2003; Pokam et al. 2012, 2014; Nicholson and Dezfuli 2013; Dezfuli and Nicholson 2013; Cook and Vizy 2016; Nicholson 2018; Longandjo and Rouault 2020; Kuete et al 2019) have used reanalysis data to highlight, describe and investigate the drivers of some important features of the atmospheric circulation in the region. Among them, elements of the low-level circulation play a crucial role because they are active year-round and their changes throughout the year strongly influence the climate in the region (Nicholson and Grist 2003; Pokam et al. 2014; Longandjo and Rouault 2020).

1.1 Key elements in the low-level circulation

The lower tropospheric circulation in CEA is mainly characterised by two components which are both established throughout the year: Low-level westerlies (LLWs: Nicholson and Grist 2003) and the Congo basin cell (CB cell: Longandjo and Rouault 2020).

LLWs are associated with the southeasterly trades on the northeastern flank of Saint Helena (South Atlantic) high. Due to the Coriolis force, the southeasterlies recurve and become westerlies when crossing the equator (Pokam et al. 2014). Nicholson and Grist (2003) suggested that the strength of equatorial westerlies is related to the sea level pressure associated with the South Atlantic High (SAH). However, in investigating the drivers of LLWs, Pokam et al. (2014) split the zonal wind into its divergent and rotational component and found that north of 6o N in CEA, LLWs are primarily a rotational flow forming part of the cyclonic circulation driven primarily by the heat low of the West African monsoon system. This northern arm of the LLW is well-developed from June to August. It weakens during September–November, disappears in the December–February season and originates in the March–May season. South of 6o N, the circulation is dominated by the divergent component and the seasonal variability of the LLW is controlled by the zonal land-sea thermal contrast near the equator. By advecting moisture from the Atlantic ocean to the CEA region, the strength of LLWs is related to rainfall variability in the region where wet years exhibit a distinct westerly wind during both wet seasons (Dezfuli and Nicholson 2013; Nicholson and Dezfuli 2013). This makes LLWs an essential circulation feature in CEA.

Many studies (Pokam et al. 2014; Cook and Vizy 2016; Neupane 2016) have suggested the presence of a zonal shallow overturning cell over central Africa. The cell was finally highlighted by Longandjo and Rouault (2020) and they denoted it the Congo basin cell. It is a closed, counterclockwise and shallow zonal overturning cell that is confined at the lower troposphere (between the surface and 800 hPa) and is active throughout the year. LLWs form the lower base of the cell and similar to LLWs, the Congo Basin cell intensity and width are driven by the near-surface temperature warming on both the central African landmass and the eastern equatorial Atlantic. The cell’s maximum (minimum) intensity and width are registered in August/September (May). As shown by Longandjo and Rouault (2020) the eastern edge of the cell is associated with the Congo Air Boundary, a convergence zone where the low-level jets from the equatorial Atlantic, having crossed the central African landmass, meet the Indian monsoon system easterlies to form the ascending branch of the cell. It is a zone of maximum convection and precipitation in the region. The zonal rainfall maximum position in the region is then modulated by the width of the Congo basin cell.

Therefore, because LLWs contribute to modulate the amount of rainfall in CEA (Nicholson and Dezfuli 2013) while the CB cell plays a crucial role in rainfall redistribution over central Africa via the zonal rainfall maximum position (Longandjo and Rouault 2020), the ability of climate models to simulate their strength and drivers is important for advancing knowledge of the CEA climate.

1.2 State of evaluation

Climate models are tools used to investigate the response of climate to various forcings and for making climate projections (Creese and Washington 2016; Taguela et al. 2020). However, to assess confidence in projected outputs, models are evaluated on their ability to represent the past or present climate. In this line, the process-based evaluation method has been advocated (James et al. 2015, 2018; Rowell et al. 2015; Baumberger et al. 2017), because a better understanding of how models behave is fundamental to help determine how to improve them. In addition, it is also an important way to assess their adequacy for future projection. Although there is not much such study in CEA, there are growing efforts in that direction (Creese and Washington 2018; Tamoffo et al. 2019; Crowhurst et al. 2020; Tamoffo et al. 2021a, b).

For instance, regarding the LLWs in CEA, Tamoffo et al. (2021a, b) showed that rainfall in the regional climate model RCA4-v4 has been improved in CEA compared to its previous version (RCA4-v1) due to the stronger LLWs in the new version that advect sufficient moisture into the region and contribute to reduce the dry bias in the model. However, to the best of our knowledge, Creese and Washington 2018 is the only study that has investigated the LLWs in the hindcast model output from the Coupled Model Intercomparison Project 5 (CMIP5) in CEA. This study assessed their impact on the simulated rainfall biases in the region during the SON (September–November) season. They found that wetter models in the east of the region exhibit stronger westerly flow across the tropical eastern Atlantic, advecting a surplus of moisture in the region that contributes to an increase of rainfall. Since LLWs were not the focus of the study, they did not go deeply in their evaluation by investigating the origin of their biases. In addition, with the release of model datasets from the Coupled Model Intercomparison Project 6 (CMIP6), it will be interesting to assess how they represent LLWs and if there is an improvement compared to CMIP5 models. As far as the CB cell is concerned, it was recently highlighted and described by Longandjo and Rouault (2020) and has never been evaluated in any coupled climate models.

1.3 Aims

In light of the importance of these low-level circulation features in CEA (Sect. 1.1) and given the gap in the state of their evaluation in coupled climate models (Sect. 1.2), this paper aims to answer the following questions:

  1. a)

    How are the main features (LLWs and the CB cell) of low-level circulation in CEA represented in CMIP5 and CMIP6 models?

  2. b)

    Is there an improvement in CMIP6 compared to CMIP5 models?

  3. c)

    Do their drivers explain changes in models’ behaviour?

The paper outline is as follows: Sect. 2 presents the data and methods used in this study. The LLWs and the CB cell are evaluated in CMIP5 and CMIP6 models in Sects. 3 and 4, respectively, while Sect. 5 summarizes our results and highlights the main conclusions.

2 Data and methods

Data used in this study are output from coupled general circulation models’ (CGCMs) simulations taking part in the Coupled Model Intercomparison Project Phase 5 (CMIP5; Taylor et al. 2012) and 6 (CMIP6; Eyring et al. 2016). Only one ensemble member for each model is used: the r1i1p1 integration for CMIP5 and r1i1p1f1 for CMIP6 models. We analyse the monthly output of 16 climate models, 8 from CMIP5 and 8 from CMIP6 (see Table 1 for more details). To track any improvement, each CMIP6 model used is the new version of a corresponding CMIP5 model. For this study, model datasets cover a period from 1980–2005 to 1980–2010 for the CMIP5 and CMIP6 output respectively.

Table 1 Details of CMIP5 and CMIP6 models used in the study
Full size table

Three reanalysis datasets at monthly time scale from 1980 to 2010 are used to assess models output: MERRA2 (Gelaro et al. 2017) the updated version of MERRA, is a reanalysis dataset from the National Aeronautics and Space Administration (NASA). It is available on 72 sigma levels at a horizontal resolution of 0.5° × 0.625°; ERA-Interim (Dee et al. 2011) a reanalysis dataset produced with the ECMWF Integrated Forecast System (IFS Cycle 31r2) with a horizontal resolution of 0.75° × 0.75° on 60 levels; ERA5 (Hersbach et al. 2020), the reanalysis dataset with the highest spatial resolution used in this study, is also produced by the European Centre for Medium-Range Weather Forecast (ECMWF) Integrated Forecast System (IFS Cycle 41r2) to replace their ERA-Interim product. It has a spatial resolution of 0.25° × 0.25° with 137 levels. Some variables used in our investigations are the horizontal wind vector (both zonal and meridional components) and the near surface temperature, pressure and geopotential height.

To evaluate the ability of models to represent LLWs over Central Equatorial Africa, the horizontal zonal wind is partitioned into divergent and rotational (non-divergent) components using the Helmholtz theorem as applied in Pokam et al. (2014). This is to assess the contribution of each component to the total zonal wind. Afterwards, we explore the processes that control each component to investigate mechanisms of the LLWs in CMIP5 and CMIP6 models. For bias calculation, datasets used in this study have been interpolated to a common grid of 1° × 1° to easily compare models and reanalyses. Regarding the Congo basin cell, we use the zonal mass-weighted stream function, as computed in Longandjo and Rouault (2020) for its exploration. It satisfies the meridional mean continuity equation in spherical coordinates and is calculated at each pressure and longitude as a downward integrated zonal wind function. The zonal wind is first averaged between 5o N and 5o S before the zonal mass-weighted stream function is computed.

3 Low-level westerlies (LLWs)

The representation of the simulated LLWs together with the contribution of the divergent and rotational circulations to the total zonal wind are assessed in this section. The attention is also put on LLWs’ drivers to understand their biases.

3.1 Mean seasonal climatology of the total circulation

With the core speed of the LLWs in CEA located along the coastal region between 10° and 15o E year-round (Pokam et al. 2014), the long term mean of the zonal wind is averaged in that longitude band in all the reanalyses, CMIP5 and CMIP6 models and represented in Figs. 1 and 2 for March–May (MAM) and September–November (SON) seasons, respectively. The focus is on the MAM and SON seasons because they are wet seasons that encompass the majority of mechanisms driving the region's climate system (Tamoffo et al. 2021b). It appears in reanalyses that, although the westerlies are weaker in MAM (Fig. 1a–c) than in SON (Fig. 2a–c) season, there is an agreement (even though not perfect) with the upper boundary of LLWs found around 850 hPa and the core speed located around 925 hPa in both seasons. Furthermore, in reanalyses, the MAM season depicts a distinct arm of westerly winds located north of 6o N which is strengthened (not shown here) in June–August (JJA) and weakened significantly in SON, to disappear (not shown here) in December–February (DJF) as highlighted in Pokam et al. (2014). Although they bear some biases, the above basic features of LLWs are relatively well represented in models. In MAM, HadGEM2-CM3, BCC-CSM1.1-m and MIROC5 (GFDL-CM3, GISS-E2-R and MRI-ESM1) are CMIP5 models overestimating (underestimating) the upper boundary of LLWs found at around 800 hPa (900 hPa). However, this is slightly improved in CMIP6 models except in BCC-CSM2-MR and MIROC6. In addition to the overestimation of the upper boundary of LLWs in HadGEM2-CM3, the modelled westerlies are also extended further south in the SON season, with once more an improvement in its CMIP6 version. In the MAM season, most of the CMIP5 models fail to represent the arm of westerly winds located north of 6o N. There is no improvement observed in that feature in CMIP6 models. This could be due to the coarse resolution of both CMIP5 and CMIP6 models. However, the study will focus on the arm of westerly winds located south of 6o N because that arm is present year-round (Pokam et al. 2014). In both seasons, CMIP5 and CMIP6 models depict well the vertical location of the core speed of LLWs found at around 925 hPa as in reanalyses. Models’ ability to represent the intensity and the spatial pattern of the core speed of LLWs at this level (925 hPa) is discussed next.

Fig. 1
figure 1

March–May long term seasonal mean of zonal wind (m/s) averaged between 10° and 15° E for a–c reanalyses, d–k CMIP5 and l–s CMIP6 models. Positive values correspond to westerly winds and negative values to easterly winds

Full size image
Fig. 2
figure 2

Same as in Fig. 1, but for the September–November season

Full size image

Figures 3 and 4 show for MAM and SON season respectively, the 925 hPa seasonal mean climatology of total circulation (vectors) and zonal wind (shading) speed from all the reanalyses (Figs. 3a–c and 4a–c). Biases relative to ERA5 are shown from CMIP5 and CMIP6 models (Figs. 3d–s and 4d–s). In both seasons, reanalyses depict an inflow of westerly winds in the study region through the western boundary with the inflow stronger in SON than in MAM season. The simulated biases of CMIP5 and CMIP6 models are similar in their spatial pattern but differ in intensity. In both seasons, the positive (negative) bias along the coastal region in HadGEM2-CM3, HadGEM2-ES, BCC-CSM1.1-m and MIROC5 (GFDL-CM3, GISS-E2-R and MRI-ESM1) denotes an excess (a deficient) in the intensity of westerly winds, much more pronounced in BCC-CSM1.1-m (GFDL-CM3). Although the sign of the biases is the same, an improvement in reducing the biases depicted in CMIP5 models is observed in their corresponding CMIP6 models in both seasons. However, in the MAM season, the CNRM-CM5 model and its corresponding CMIP6 version (CNRM-CM6-1) have an opposite bias in the simulated zonal wind along the coast. CNRM-CM5 overestimates while CNRM-CM6-1 underestimates.

Fig. 3
figure 3

Long term seasonal mean of March–May 925 hPa total wind (vectors: m/s) and zonal wind (shading: m/s) for a–c reanalyses, d–k CMIP5 models biases with respect to ERA5 and l–s CMIP6 models biases with respect to ERA5. The black box is the study region, and the interest is on the inflow at the region’s western boundary

Full size image
Fig. 4
figure 4

Same as Fig. 3, but for the September–November season

Full size image

3.2 Contribution from the divergent and rotational circulations

To focus on the westerly winds at the western boundary of the study region, the 925 hPa zonal wind and its divergent and rotational components are averaged between 10° S–5° N and 10°–15° E. Their annual cycles are then displayed in Fig. 5 for reanalyses, CMIP5 and CMIP6 models (Fig. 5a–f). This is done to assess the contribution of each component to the total zonal wind and to determine to what extent they contribute to the models’ biases. Reanalyses agree well in the representation of the total zonal wind and its components. Throughout the year, the divergent circulation (Fig. 5b and e) is a westerly wind (positive values) while the rotational circulation (Fig. 5c and f) is an easterly wind (negative values) in most months. The total zonal wind being also a westerly wind (positive values: Fig. 5a and d), it follows that it is mainly divergent in its kinematic character. This is in line with the findings from Pokam et al. (2014). In addition, the reanalyses (Fig. 5a–f) show that, the total and divergent zonal wind peaks’ are found in February and August while for the rotational circulation, the peaks are found in May and November since it is principally an easterly wind as highlighted above. Although the seasonality of the total zonal circulation and its components is reasonably captured by the CMIP5 and CMIP6 models, considerable differences in terms of intensity are depicted (Fig. 5a–f). Most of the CMIP5 models overestimate the strength of the total zonal circulation at the beginning and around the end of the year (Fig. 5a). Moreover, in addition to what emerges from Figs. 3 and 4, the BCC-CSM1.1-m model overestimates the total circulation not only in MAM and SON but over the whole year, in contrast with GFDL-CM3 which underestimates the westerly flow in most months (from February to September). Examining the wind components (Fig. 5b and c), most models overestimate the westerly component of the rotational circulation throughout the year (Fig. 5c), making it a westerly flow that increases its contribution to the total wind. However, from one model to another, the positive or negative bias in the total zonal circulation (Fig. 5a) is generally attributed to the combined biases in both the divergent (Fig. 5b) and rotational (Fig. 5c) circulation. This is observed in the BCC-CSM1.1-m model that overestimates the total zonal wind as the result of the overestimation of both divergent and rotational components. On the other hand, the overestimated total zonal wind in the HadGEM2-CM3 model is only attributed to the rotational circulation in MAM and SON while from May to August, the deficit of the total wind in the GFDL-CM3 model is mainly due to its underestimated divergent component.

Fig. 5
figure 5

Seasonal cycle of 925 hPa total, divergent and rotational zonal wind averaged between 10° S–5° N and 10°–15° E for a–c reanalyses and CIMP5 models, d–f reanalyses and CMIP6 models. g–i Uncertainty ranges in total, divergent and rotational zonal wind from reanalyses (red), CMIP5 (blue) and CMIP6 (green) models. Note that westerly winds are positive while easterly are negative

Full size image

Compared with reanalyses, the rotational circulation (Fig. 5f) in the CMIP6 models has led to a stronger underestimated total wind in GFDL-CM4 and MRI-ESM2-0 during June and May (Fig. 5d). However, in comparison with the CMIP5 models, smaller biases are generally observed in their corresponding CMIP6 models, highlighting the improvement in the total wind (Fig. 5d) which is the result of the improvement in both components (Fig. 5e and f). This is well observed in Fig. 5g–i displaying the annual cycle of the uncertainty ranges in total, divergent and rotational zonal wind from reanalyses, CMIP5 and CMIP6 models. The uncertainty range here is the spread or the dispersion in a set of data. It is represented with a band and the smaller the bandwidth, the smaller the uncertainty and vice-versa. It appears that, compared to reanalyses, in most months, the spread of the total wind in CMIP6 models is slightly smaller than the one from CMIP5 models due to the improvement in both divergent (Fig. 5h) and rotational (Fig. 5i) circulation. However, in general, in both CMIP5 and CMIP6 models, the spread of the rotational component (Fig. 5i) is larger than the spread of the divergent (Fig. 5h) component. This highlights the fact that, in most months, the bias in the rotational circulation contributes most to the bias in the total zonal circulation.

3.3 Control mechanisms of LLWs

This section aims to understand simulated LLWs biases by exploring the processes that control the variability of the LLWs over CEA. As seen above, when decomposed into their two components, the LLWs biases are attributed to either the divergent or the rotational component or both. Therefore, investigating their drivers means looking at those of their two components. In addition, investigating whether drivers agree with the improvement in the simulation of LLWs from CMIP5 to CMIP6 models will add more confidence in the improvement and to the CMIP6 outputs.

Pokam et al. (2014) have shown that the divergent component of the LLWs in CEA is driven by the near-surface temperature difference between the Atlantic ocean and the Congo basin landmass. However, the area of the Atlantic ocean over which temperature variability is strongly associated with the coastal circulation variability is not clearly defined and has always been taken arbitrarily (Cook and Vizy 2016; Neupane 2016; Longandjo and Rouault 2020). To identify that area, the near-surface temperature difference is calculated between western Central Africa (averaged in the black box in Fig. 6a) and each grid point over the Atlantic Ocean. The difference found for each grid point in the Atlantic ocean is correlated with the 925 hPa divergent wind averaged over the coastal area (between 10° S–5° N and 10°–15° E) and the correlation values obtained at each grid point are then displayed in Fig. 6 for reanalyses (Fig. 6a–c), CMIP5 (Fig. 6d–k) and CMIP6 (Fig. 6i–s) models. Let us note that, the computed correlation is based on the inter-annual variability. Except in ERA5, other reanalyses such as ERAINT and MERRA2 depict a positive correlation all over the Atlantic ocean (Fig. 6a–c). This means a high (low) temperature difference is associated with a strong (weak) divergent wind. The blue box (Fig. 6a) located at the east of the Atlantic ocean (between 10° S–0° and 0°–10° E) shows the region with the highest positive correlation based on all three reanalyses. All reanalyses agree relatively well with the location of that region (Fig. 6a–c) although the strength of the correlation is less in ERA5 (Fig. 6b). Both CMIP5 and CMIP6 models succeed in reproducing the positive correlation over the Atlantic ocean and the region of maximum correlation is also well detected by most models. However, in CNRM-CM6-1 (Fig. 6l), GFDL-CM4 (Fig. 6m) and GISS-E2-1-G (Fig. 6n), the positive correlation is less strong compared to that depicted by their corresponding previous CMIP5 models (Fig. 6d–f). This feature suggests that the strength of the divergent wind is less related to the temperature difference between the Atlantic ocean and the Congo basin landmass in those models. In addition, in HadGEM2-CM3 (Fig. 6g) and BCC-CSM2-MR (Fig. 6r), the region of maximum correlation is shifted a little southwestward in the Atlantic ocean.

Fig. 6
figure 6

Correlations of 925 hPa temperature difference between Central Africa (averaged in the black box) and each grid point temperature over the Atlantic Ocean, against the divergent wind, averaged between 10° S–5° N and 10°–15° E (red box) for a–c reanalyses, d–k CMIP5 and l–s CMIP6 models. Long term annual mean data are used here and the blue box (a) shows the region of highest correlation in reanalyses

Full size image

Figure 7 is used to help further understand the divergent wind biases. It shows the annual cycle of temperature and pressure differences between the Central Africa landmass (black box in Fig. 6a) and the eastern Atlantic Ocean (blue box in Fig. 6a) from reanalyses, CMIP5 and CMIP6 models. It emerges that, throughout the year, the temperature (pressure) difference is positive (negative) because of higher (lower) temperatures (pressure) over the land than ocean. Reanalyses show two peaks found in February–March and July–August respectively. The two peaks are also observed in the seasonal cycle of the pressure difference and they coincide with the divergent wind component peaks (Fig. 5b). The second and stronger peak appears during the eastern equatorial Atlantic cold tongue (Cook and Vizy 2016; Neupane 2016). The later enhances the temperature difference between the cool SSTs and the warm continent and strengthen the divergent wind flow.

Fig. 7
figure 7

Annual cycle of 925 hPa land–ocean thermal difference and the zonal pressure difference between the Central African landmass (10° S–5° N; 15°–30° E) and the coastal Atlantic Ocean (10° S–0°; 0°–10° E) for a, b reanalyses and CMIP5, c, d reanalyses and CMIP6 models. e, f Uncertainty ranges in land–ocean thermal and pressure differences from reanalyses (red), CMIP5 (blue) and CMIP6 (green) models

Full size image

CMIP5 and CMIP6 models reasonably capture the main characteristic (the two peaks) of the annual cycle of both temperature and pressure differences, but they bear some biases. Most CMIP5 models underestimate the July–August peak of the temperature difference annual cycle (Fig. 7a). The underestimation is more pronounced in HadGEM and is associated with an underestimation of the simulated pressure difference (Fig. 7b) which can explain why its simulated divergence wind is also underestimated (Fig. 5b). In BCC-CSM the overestimated divergent wind (Fig. 5b) is exclusively associated with the overestimation of the simulated pressure difference (Fig. 7b) as the temperature difference is within the range of reanalyses (Fig. 7a). Examining the corresponding CMIP6 models, GFDL-CM and MRI-ESM deteriorate in representing the annual cycle of both temperature and pressure difference as compared with their corresponding CMIP5 models, consistent with their simulated wind divergence. However, most CMIP6 models such as HadGEM and BCC-CSM do better (Fig. 7c and d) than their corresponding CMIP5 models (Fig. 7a and b). Figure 7e and f illustrate the improvement by showing the uncertainty ranges in thermal and pressure differences from reanalyses, CMIP5 and CMIP6 models. These results reveal that, the range of CMIP6 models remains larger than that of the reanalyses range but compared to CMIP5 the bandwidth of CMIP6 models is relatively small except in a few months such as June-July in the temperature difference annual cycle. However, from one month to another, the median CMIP6 model is generally closer to the median reanalysis. Overall, this shows that the improvement in the simulated divergent wind is associated with an improvement in the simulated drivers.

Figure 8 shows the spatial pattern of the long term annual mean 850 hPa relative vorticity (shading) and geopotential height (contours) for reanalyses, CMIP5 and CMIP6 models. Here, the focus on these variables (relative vorticity and geopotential height) at the 850 hPa pressure level is because at 925 hPa (where the core of the LLWs is found), many areas are blank due to high orography, making difficult any interpretation. In the reanalyses (Fig. 8a–c), at the western boundary of the study area (red box: Fig. 8a), the negative values of the relative vorticity associated with low-pressure values reflect a clockwise circulation corresponding to the rotational component of LLWs. These negative values of the relative vorticity are strengthened in most CMIP5 models in line with the overestimation of their simulated LLWs rotational component all-year-round (Fig. 5c). The reinforcement of the negative values of relative vorticity is generally associated with a reinforcement of the simulated low pressure and this is particularly observed in HadGEM2-CM3 (Fig. 8g). By contrast, MRI-ESM1 (Fig. 8h) underestimates the strength of the negative vorticity found at the western boundary of the study region. This is consistent with the underestimated LLWs rotational component in that model (Fig. 5c). The CMIP6 models depict a similar pattern as their corresponding CMIP5 models with the negative values of the relative vorticity remaining pronounced with respect with reanalyses. However, although the improvement in CMIP6 with respect to CMIP5 is not very obvious for all the models, an improvement is observed for some of them such as HadGEM3-GC31-LL and BCC-CSM2-MR. Furthermore, the improvement in the relative vorticity in HadGEM3-GC31-LL is associated with an improvement in the simulated pressure consistent with a better representation of the rotational circulation (Fig. 5f).

Fig. 8
figure 8

Long term annual mean of 850 hPa relative vorticity (shading: 10–5 s−1) and 850 hPa Geopotential height (contours: Pa) for a–c reanalyses, d–k CMIP5 and l–s CMIP6 models. The red box is the study region, and the interest is on the value of the relative vorticity at the region’s western boundary

Full size image

4 The Congo basin cell (CB cell)

In this section, we assess models' ability to represent the CB cell with the focus on the cell’s key characteristics (intensity, width, western and eastern edge positions).

4.1 Mean seasonal climatology and link with LLWs

To investigate the Congo basin cell, Figs. 9 and 10 display the seasonal mean climatology of the mass-weighted stream-function (contours) for reanalyses, CMIP5 and CMIP6 models in MAM and SON seasons respectively. The zonal wind is also represented (shaded) in Figs. 9 and 10 to highlight the link between LLWs and the Congo basin cell. With the negative values of the mass-weighted stream-function depicting the Congo basin cell, it emerges from the figures that the three reanalyses agree relatively well in the intensity and the width of the cell in both seasons (Figs. 9a–c and 10a–c) although in MAM the cell looks wider in MERRA2 and ERA5 than ERAINT and its height is located around 800 hPa in MAM (Fig. 9a–c) but slightly higher in SON (Fig. 10a–c) season. In the MAM season (Fig. 9), CNRM-CM5, HadGEM2-CM3 and BCC-CSM1.1-m (GISS-E2-R and MRI-ESM1) are CMIP5 models that simulate a stronger and larger (weaker and narrower) cell compared to the reanalyses with the height of the cell exceeding 500 hPa in HadGEM2-CM3. To a certain extent, with this overestimated height, this could be assimilated to a large-scale circulation since there is an ongoing debate on whether the CB cell is distinct or not from the larger scale Central Africa Walker cell (Longandjo and Rouault 2020; Nicholson 2022). During the SON season, the cell remains stronger in HadGEM2-CM3 and BCC-CSM1.1-m but the width of the cell is less large in HadGEM2-CM3. Looking at the corresponding CMIP6 models, in terms of intensity, an improvement is particularly observed in CNRM-CM6-1 and HadGEM3-GC31-LL during the MAM season. Although BCC-CSM2-MR does not show a real improvement during the SON season, HadGEM3-GC31-LL shows an improvement in the cell’s intensity, width and height while GFDL-CM4 depicts a stronger cell compared to its corresponding previous CMIP5 model (GFDL-CM3). However, we observe that compared to CMIP5 models, most of the corresponding CMIP6 models better represent the cell in both seasons.

Fig. 9
figure 9

March–May mean seasonal climatology of the zonal mass-weighted stream function (contours: 1011 kg s−1) and the total zonal wind (shading: m/s) averaged between 5° S–5° N for a–c reanalyses, d–k CMIP5 and l–s CMIP6 models. Solid black, green and dashed black contours represent positive, zero and negative values of mass-weighted stream functions, respectively. Contour intervals are 10 between positive contours and 4 between negative contours

Full size image
Fig. 10
figure 10

Same as Fig. 9 but for the September–November season

Full size image

It is also observed that the bias in the representation of the CB cell is generally associated with the bias in the strength of the simulated LLWs. In both seasons (Figs. 9 and 10) the overestimated (underestimated) Congo basin cell in models is associated with stronger (weaker) simulated LLWs. This highlights the fact that LLWs and the CB cell have the same drivers, consistent with findings from Longandjo and Rouault (2020). Therefore, in MAM and SON seasons, the improvement in the representation of the cell from CMIP5 to CMIP6 models is associated with the improvement in the representation of the simulated LLWs. Does the improvement extend throughout the year? We answer this question in the following section.

4.2 Annual cycle of the cell’s key characteristics

Examining at the annual cycle of the cell in more detail, the models’ ability to represent the seasonal evolution of the cell's key characteristics (intensity, width, western and eastern edge positions) is assessed in this section. Located between 925 and 850 hPa (Longandjo and Rouault 2020) the intensity of the cell is found by vertically averaging the mass-weighted stream function between 925 and 850 hPa and the minimum value found in the longitudinal band of the cell represents its intensity. The smaller that value, the stronger the cell because negative values of the mass-weighted stream functions are those depicting the cell. The western and eastern edges are located at the longitudes where the vertically (between 1000 and 850 hPa) averaged mass-weighted stream function is equal to zero or changes sign and the difference between the two edge longitudes indicates the width or zonal extent of the cell.

The seasonal cycle of these cell’s key characteristics is represented in Fig. 11 for reanalyses, CMIP5 and CMIP6 models. The figure reveals that the seasonal evolution of the Congo basin cell is consistent in all three reanalyses with the maximum intensities found in February and August/September (Fig. 11a and e). Moreover, in agreement with findings from Longandjo and Rouault (2020), the second maximum is stronger than the first and throughout the year, the peaks of this bimodal signal are associated with the westernmost positions of the western edge (Fig. 11c and g) and the maximum width (Fig. 11b and f) of the cell. Although ERAINT does not agree with other reanalyses (MERRA2 and ERA5), the eastern edge position of the cell (Fig. 11d and h) has a low annual variability compared to the western edge position. CMIP5 models moderately capture the main features of the cell’s key characteristics. However, most fail to simulate the bimodal signal of the evolution of the Congo basin cell’s key characteristics, especially during the first half of the year. Although the second peak is reasonably well detected, most CMIP5 models fail to detect the first one, simulating it in April instead of February (Fig. 11a). In addition, especially in CNRM-CM and HadGEM (Fig. 11a), the simulated first peak found in April is stronger than the second (Fig. 11a) and is associated with a simulated larger cell (Fig. 11b) and a further west position of the simulated western edge (Fig. 11c). The simulated western edge longitude of the cell suggests that water vapour is transported from further afield in the Atlantic ocean toward central Africa inland. However, although the CMIP6 version of BCC-CSM still overestimates the strength (Fig. 11e), the width (Fig. 11f) and the western edge position (Fig. 11g) of the cell throughout the year, CMIP6 models generally perform better in representing the seasonal cycle of the cell’s key characteristics (Fig. 11e–h). This is well illustrated with the spread in CMIP6 models being smaller than that of CMIP5 models and their ensemble mean generally closer to reanalyses (Fig. 11i–k). The bandwidth of CMIP5 models is more important, showing a large range of disparity or dispersion compared to CMIP6 models.

Fig. 11
figure 11

Seasonal cycle on the Congo Basin cell intensity, width, western and eastern edge positions as depicted in a–d reanalyses and CMIP5 models, e–h reanalyses and CMIP6 models. i–l Uncertainty ranges in CB cell intensity, width, western and eastern edge positions from reanalyses (red), CMIP5 (blue) and CMIP6 (green) models

Full size image

5 Summary and conclusion

A recent study (Creese and Washington 2018) has highlighted the fact that rainfall bias in CMIP5 models over Central Equatorial Africa (CEA) is partly attributed to the misrepresentation of the important features of the atmospheric low-level circulation in the region. However, the study did not go deeply in the understanding of the origin of the misrepresentation. Acting year-round, the key features of the lower circulation in CEA are the low-level westerlies (LLWs) and the Congo basin cell (CB) cell. With the release of CMIP6 models, in this paper, we compare the ability of eight CMIP6 climate models and the corresponding eight previous CMIP5 models to simulate the long term mean climatology and the seasonal cycle of these features. In addition, we explore their drivers to understand how these features are represented in models and to explain changes in models’ behaviour.

Although both CMIP5 and CMIP6 models perform well in simulating the main characteristics of LLWs such as the vertical location of the core speed found at around 925 hPa, and the two peaks in the seasonal cycle, most CMIP5 models overestimate the strength of LLWs by up to around 2 m/s in BCC-CSM1.1-m model. The overestimation is generally attributed to both divergent and rotational components of the total wind with the rotational component contributing the most in the overestimation. In CMIP6, although most models still overestimate LLWs strength, the intensity is slightly reduced compared to CMIP5 models. That improvement is the result of a better performance in both divergent and rotational circulation. The improvement in the simulated divergent component is associated with a better representation of the near-surface pressure and/or temperature difference between the Central Africa landmass and the coastal Atlantic Ocean. Regarding the rotational circulation, in a model such as HadGEM3-GC31-LL and BCC-CSM2-MR, a simulated higher 850 hPa pressure is associated with less pronounced negative vorticity and a better representation of the rotational circulation (Fig. 5f).

For the CB cell, our results suggest that, its simulated biases are related to those of LLWs. In most CMIP5 models, the cell intensity and width are overestimated with the western edge generally found further west. These biases in the cell’s key characteristics are associated with the overestimation of the simulated LLWs in CMIP5 models. However, although in CMIP6 models the cell key characteristics are still overestimated, the strength is reduced compared to CMIP5 models. This shows that CMIP6 models perform better in representing the cell and therefore, the improvement in the representation of the cell in CMIP6 models is associated with improvement in the representation of the simulated LLWs. We note that a small improvement in LLWs translates to a larger relative improvement in the CB cell.

Given the results we have highlighted in this study, compared to CMIP5 models, more confidence can be put in CMIP6 models although the spread is still relatively large. However, to help better represent low level circulation features in the next generation of the coupled models, the land and ocean surface schemes could be advanced to resolve much better the surface characteristics which in turn can lead to an improvement of the simulated low-level circulation drivers (land–ocean thermal and pressure differences). In the meanwhile, the future work will investigate whether the improvement in the simulation of these low-level circulation features in CEA from CMIP5 to CMIP6 models has an impact like an improvement in the representation of rainfall in the region.