3.1 Relationship between the Botswana High and ENSO
Figure 3a shows a strong association in the interannual variability of the Botswana High and ENSO. The MPAS, MPAS5 and ERA5 show a high correlation between Botswana Highs and ENSO (r = 0.84, r = 0.81 and r = 0.89, respectively). In addition, the Botswana High in MPAS shows a higher correlation with one in ERA5 (r = 0.89) as compared to MPAS5 Botswana High (r = 0.87). These results show that the MPAS model improves the simulation of the interannual variability of the Botswana High, although by a small margin. Both models and reanalysis indicate the strongest Botswana Highs during the strong El Niño years (1983, 1998 and 2010), while the weakest Botswana Highs occurred in La Niña years (1985, 1989, and 2000) (Fig. 3a). This is in agreement with previous studies by Driver and Reason (2017). However, the relationship is not strictly linear as there are cases where the intensity of ENSO is not proportional to the intensity of the Botswana High (e.g. El Niño 1983 vs El Niño 1998). Kao and Yu (2009) showed that the nonlinear relationship could be explained by the internal variability of ENSO SST anomalies within the Pacific Ocean, creating different ENSO patterns. Furthermore, Johnson (2013) and Gore et al. (2019) identified eight ENSO SST patterns over the Pacific Ocean (four El Niño patterns and four La Niña patterns) which might impact the strength of the Botswana High differently.
Figure 3b shows a scatter plot between the Botswana High and ENSO. As mentioned above, there seems to exist a quasi-linear relationship between the Botswana High and ENSO. At a 95% confidence limit, the adjusted R2 values indicate that ENSO explains about 70%, 65%, and 79% of Botswana High variability in the MPAS, MPAS5 and ERA5 datasets respectively. Again, the MPAS model improves the relationship between Botswana High and ENSO (70%) as compared to MPAS5 (65%). Given the strong association between the Botswana High and ENSO, the question then arises as to whether the Botswana High is a result of forcing from ENSO. To address this, the response of the Botswana High to the removal of ENSO forcing is discussed in Sect. 3.3.
3.2 Composite Anomalies over Southern Africa
We evaluate the MPAS simulation over southern Africa by comparing it with GPCP observation, ERA5 reanalysis and MPAS5 model data (Figs. 4 and 5). The validation focuses on the spatial distribution of rainfall (Fig. 4) and the 500 hPa geopotential height (Fig. 5) in JFM of El Niño years (1983, 1992, 1998; and 2010) and La Niña Years (1985, 1989, 2000, and 2008). Figure 4 shows that MPAS simulates well spatial distribution of rainfall over southern Africa, except that the simulation generally has a better agreement with GPCP during El Niño years composite (r = 0.85; RMSE = 1.67) than during La Niña years composite (r = 0.80; RMSE = 2.12). During the El Niño years, the model captures well the band of maximum rainfall over the tropics (associated with the South Indian Convergence Zone, SICZ) with a wet bias of about ± 1 mm day− 1. It also reproduces well the rainfall over the eastern parts of the continent and areas of minimum rainfall over the central and western parts of South Africa and Namibia. Over Madagascar, the model fails to capture the local maximum rainfall as in GPCP. This lack of maximum rainfall over Madagascar may be attributed to the overestimation of deep convection over the Mozambique Channel (MC) area (Figs. 4b, 4f, 4j, 4n and 4r), which may suppress convection over adjacent areas leading to the inability of MPAS to stimulate the local maximum rainfall over Madagascar. Despite that, the MPAS composite (r = 0.85; RMSE = 1.67) shows that the MPAS variable-resolution generally improves the simulation of rainfall over the region as compared to the uniform resolution MPAS5 composite (r = 0.83; RMSE = 1.78) during El Niño years.
During the La Niña years, the MPAS model underestimates the summer rainfall over the eastern and central parts of Botswana, Mozambique and South Africa (by about 3 mm day− 1). This may be attributed to the overestimated deep convection extending farther south of the MC (Figs. 4d, 4h, 4l, 4p, and 4t), leading to the suppression of convective rainfall and a dry bias over the eastern and central parts of the southern African continent. Furthermore, the overestimation of deep convection over the MC is greater in the MPAS variable-grid than MPAS5 uniform-grid during La Niña. This reduces the accuracy of the MPAS in simulating the regional rainfall as compared to MPAS5. This result is consistent with Maoyi et al. (2017) and Driver et al. (2018), who noted a similar wet bias over the MC using a 50 km stretched-grid GCM (called CAM-EULAG). The wet bias could be due to the convective parameterisation in the model, which might be too sensitive to the warm boundary layer over the region. Another cause for this wet bias could be the horizontal resolution sensitivity of moist physics as highlighted by Williamson (2008) or the parameterisation time step in the model simulation as indicated by Williamson and Olson (2003).
MPAS shows good agreement with ERA5 on the spatial pattern of the 500 hPa geopotential height during El Niño years composite (r = 0.96; RMSE = 10.08) and La Niña years composite (r = 0.96 RMSE = 7.63). In general, both the MPAS and reanalysis show higher values of the 500 hPa geopotential height during El Niño years and Lower values during La Niña years, suggesting that El Niño (La Niña) summers may be the driving force in the strengthening (weakening) of the 500 hPa geopotential height over southern Africa, which may lead to the strengthening (weakening) of the Botswana High. This result is in agreement with previous studies by Reason (2016) and Driver and Reason (2017), who also found that the high is always stronger (weaker) during El Niño (La Niña) summers. Furthermore, the strong spatial correlations (r > 0.9) indicate that the model has adequate skill in reproducing the dominating features of the 500 hPa geopotential height over southern Africa but with some discrepancies. For instance, the model shows enhanced geopotential heights in the subtropics and lower anomalies poleward of 30°S as in ERA5. In addition to that, the model captures the local maximum geopotential height pattern over southern Africa (which is reminiscent of the Botswana High) as in ERA5 (e.g. Figures 5a, 5e, 5i, 5m, 5q). Despite the higher resolution of MPAS variable-grid simulation (48 km) than the MPAS5 uniform-grid simulation (240 km) over southern Africa, MPAS results do not seem to always improve on that of MPAS5 as shown by the correlation and RMSE. While MPAS simulation outperforms the MPAS5 simulation in some years (i.e. 1992, 2010, 1989 and 2008), the opposite is true in other years (1983, 1998, 1985 and 2000). Several factors could make the higher resolution MPAS simulation not always outperform the coarser resolution MPAS5 simulation. For example, both models use different land-use and topography data, which may impact results over high lying regions over southern Africa. Another reason could be the difference in the physics scheme between MPAS (WRF v4.0.3 physics) and MPAS5 (WRF v3.8.1 physics), which might have their strengths and weaknesses over southern Africa.
3.3. Sensitivity Experiments
3.3.1 The response of the Botswana High to the removal of ENSO forcing
Figure 6 shows the interannual variability of the Botswana High in CTRL and NoENSO for the EOF analysis (Fig. 6a) and spatial average analysis (Fig. 6b). The time series associated with the EOF Botswana High index (Fig. 6a) shows good similarity with the spatial average Botswana High index. This result indicates that the EOF Botswana High is physically related to the temporal variability of the 500 hPa geopotential height over the Botswana High core region. In light of this, the impact of ENSO on the temporal variability of the Botswana High is evident. In general, NoENSO shows that with the removal of the ENSO forcing (El Niño or La Niña), the amplitude of variability in Botswana High reduces; however, the signal of the Botswana High variability still remains. This indicates that the Botswana High modes are independent of ENSO forcing while the magnitude of the variability is enhanced during ENSO events. Figure 6 also suggests that there are cases when ENSO forcing may alter the signs of the Botswana High mode, either from + ve phase to -ve phase (as in 1992) or from + ve phase to -ve phase (as in 2008). This may be due to the internal variability of ENSO SST anomalies within the Pacific Ocean, creating different ENSO flavours that may impact the Botswana High distinctively.
The 500 hPa geopotential height composites show good agreement with the Botswana High’s interannual variability during El Niño Years (Figs. 7a and 8a). At regional scale (over southern Africa), there is an overall decrease in the 500 hPa geopotential height over southern Africa (Figs. 7a, 7c, 7e, 7g and 7i) which is consistent with the decrease in the interannual variability of the Botswana High index in 1983, 1992, 1998 and 2010; however, the largest decrease is over the northeastern parts of the southern African region, which is typically an area of high precipitation associated with the SICZ. Basically, the removal of El Niño influence leads to increased convection over the SICZ area leading to increased precipitation over northern Madagascar, Mozambique Channel, and much of the northeastern parts of the subcontinent. At global scale (Figs. 8a, 8c, 8e, 8g and 8i), all El Niño years (1983, 1992, 1998, 2010) show that with the removal El Niño event, there is a decrease in the 500 hPa geopotential height over the tropics and an increase in the subtropics. This response represents a weakening of the Hadley circulation (Cook 2000), and it is generally known from modelling and observational studies (Oort and Yienger 1996; Roechner et al. 1996). In addition to that, this decrease in the 500 hPa geopotential height over the tropics leads to a reduction in subsidence in the 500 hPa pressure level over southern Africa (Figs. 7a, 7c, 7e, 7g and 7i), which in turn reduces the strength of the Botswana High as seen in Fig. 6.
La Niña composites (Fig. 7b and 8b) show an opposite pattern in the 500 hPa geopotential heights compared to El Niño Composites. The regional circulation anomalies show an increase in subsidence over the region, especially over the Mozambique Channel, northern Mozambique and northern Madagascar (Figs. 7b, 7d, 7f, 7h and 7j). This anomalous increase in subsidence over the SICZ area may inhibit convection, increase temperatures, weaken the SICZ and lead to decreased precipitation. Reason (2016) attributed the formation of the Botswana High to heat released by tropical regions of high precipitation near the SICZ area. In light of this, the anomalous increase in subsidence and temperature over the SICZ may lead to the formation and strengthening of the Botswana High as Reason (2016) stated. However, a model experiment similar to Lenters and Cook (1997) is required to prove the veracity of the results, which is beyond the scope of this work.
Globally, there is an overall increase in the 500 hPa geopotential height (NoENSO minus CTRL) over the tropics and a decrease over parts of the subtropics (Figs. 8b, 8d, 8f, 8h and 8j). This pattern represents the strengthening of the Hadley circulation (Cook 2000). This increase in the 500 hPa geopotential height leads to an increase in subsidence over southern Africa, strengthening the Botswana High. The years 1985 and 1989 indicate a large increase in the 500 hPa geopotential height in the northeastern regions of southern Africa, which are the same years where there was the highest increase in the Botswana High strength due to the absence of La Niña (see Fig. 6). This result implies that the Botswana High strength is sensitive to subsidence over this area during La Niña summers.
3.3.2 Atmospheric dynamics associated with El Niño
The difference in the 200 hPa velocity potential between NoENSO and CTRL years show an upper tropospheric divergence (negative velocity potential) over the Indian and Western Pacific Ocean (Figs. 9a, 9c, 9e, 9g and 9i). This upper-level divergence is more pronounced over the northwest pacific and maritime continent, which coincides with deep convection from the warmer SSTs in the region. In contrast, there is a strong upper tropospheric convergence (positive velocity potential) over the eastern Pacific and South America, which coincides with cooler SSTs over that region. This convergence-divergence pattern across the Pacific is a reverse pattern of normal El Niño years and similar to a typical La Niña pattern. In normal El Niño years, the upper-level divergence is located over the eastern Pacific and South America, whilst the upper-level convergence is typically located over the western Pacific and maritime continent, including Australia. This normal pattern indicates a weakening of the Walker Circulation, which is typical during El Niño years (Reason and Jagadheesha 2005, Ashok et al. 2007, Yuan et al. 2014, Gore et al. 2019). Therefore, the reverse pattern, which is the difference between NoENSO and CTRL, indicates that the Walker Circulation will strengthen in the absence of El Niño. 1983 and 1998 show the most strengthening of the Walker Circulation as they were the strongest El Niño years in the study period. The southern African continent generally shows divergent flow, which may strengthen the SICZ and aid in the formation of tropical temperate troughs (Tyson and Preston-Whyte 2000) while weakening the strength of the Botswana High.
Figures 10a, 10c, 10e, 10g, and 10i show the difference between NoENSO and CTRL in the 200 hPa eddy stream function for each El Niño year, highlighting the large-scale stationary wave response. Taking the difference filters out the zonal mean response and only shows the difference in the wave response. All El Niño years show a strengthening of the cyclonic flow in the upper troposphere over the central and eastern Pacific Ocean, associated with the intensity of cooling of SSTs in the region. The strong ENSO event of 1998 generated the strongest anomalies, which indicate that the SST cooling was more intense during that year while the cyclonic flow extended as far as the western Pacific as in 2010. On the contrary, the South Atlantic and South Indian ocean show an increase in the upper tropospheric anticyclonic flow, indicating the warming of SSTs in the region. This warming of the two ocean basins may increase convergence over the southern African continent, leading to an overall weakening of subsidence in the 500 hPa level as in Fig. 7, thus reducing the Botswana High’s strength.
3.3.3 Atmospheric dynamics associated with La Niña
Figures 9b, 9d, 9f, 9h, and 9j show changes in the upper tropospheric divergence flow between NoENSO and CTRL. 1985, 1989 and 2008 show upper tropospheric divergence over much of the central and eastern Pacific and over much of the Atlantic Ocean, which coincides with deep convection from warmer SSTs. Contrary to that, there is an upper-level convergence (positive velocity potential) over much of the Indian Ocean, including Australia and the maritime continent, which corresponds to colder SSTs. This upper-level convergence over the Indian Ocean may result in a north-eastward shift of the SICZ (Cook 2001) and tropical temperate troughs (Tyson and Preston-Whyte 2000), strengthening drier conditions over southern Africa. This convergence-divergence pattern is reminiscent of the weakening of the Walker Circulation (Gore et al. 2019) and typically takes place during El Niño events. In general, the weakening of La Niña events will lead to an El Niño type of response in the atmosphere and vice-versa. The year 2000 showed a similar pattern with other La Niña years, except for a weak convergence over the Pacific. 1985 and 1989 had the highest velocity potential over southern Africa’s northeast regions, which corresponds to the increase in subsidence over the region due to cooler SSTs. This corresponds well to the increase in the 500 hPa geopotential height as in Fig. 8d and 8f and the larger strengthening of the Botswana High (1985, 1989) as in Fig. 6 as compared to other La Niña Botswana High years (2000, 2008).
The difference in the eddy stream function at 200 hPa (NoENSO minus CTRL) shows anticyclones over the eastern and western Pacific Ocean (Figs. 10b, 10d, 10f, 10h, and 10j). These anticyclones are characteristic of a Gill-Matsuno type response which is imminent due to the warming of Pacific Ocean SSTs (Cook 2001, Wilson et al. 2014, Gore et al. 2019). This increase in the strength of anticyclones leads to a reduction in the cyclonic flow over the Pacific Ocean and the weakening of the Walker Circulation. All La Niña years indicate an increase in the cyclonic flow over the southern parts of South America, Africa, the Atlantic Ocean and the Indian Ocean. This anomalous cyclonic flow in the upper troposphere indicates cooling SSTs over the region and may increase subsidence over southern Africa, thus strengthening the Botswana High.