3.1 Climatology over Southern Africa
This section discusses how well MPAS reproduces the climatology over southern Africa with a focus on the spatial distribution of temperature (Figs. 2a, 2c and 2e), rainfall (Figs. 2b, 2d and 2f) and 500 hPa geopotential height (Fig. 3). Figure 2 shows that MPAS simulates well the spatial distribution of temperature over southern Africa as compared to the observation and ERA5 reanalysis. The MPAS model has a strong spatial correlation (r > 0.9) with the observation, indicating that it reproduces the dominating features in the observed temperature field. It captures the temperature gradient between the tropics and midlatitudes and the relatively cold surface temperatures associated with the Benguela current near the western regions of the continent and the warm surface temperatures associated with the Agulhas Current in the eastern parts of the continent. However, the model simulation shows a cold bias of about 2°C over much of the Indian Ocean, which may be due to the coarser resolution in the model data (~240km) as compared to the observation (~100km). This bias suggests a weaker response of the MPAS atmosphere to SSTs over the Indian Ocean. The model shows a cold bias over the land, more especially over Botswana, Angola, Zambia and western parts of Tanzania. This bias, which also features in ERA5 results, may be due to how the MPAS resolves the topography over these high elevation areas (see Fig. 1b).
The MPAS simulated rainfall field also has common features with GPCP observation (Figs. 2b, 2d and 2f). GPCP shows an area of minimum precipitation over much of the Atlantic Ocean and a band of maximum rainfall (associated with South Indian Convergence Zone (SICZ)) over the Indian Ocean. The SICZ associated rainfall band shows a local maximum over the northern parts of the Indian Ocean (5-15°S; 60-80°E) and Madagascar (about 12 mm day-1). MPAS reproduces the tropical rainfall band but with a wet bias of about ±1 mm day-1 and fails to capture the local maximum rainfall over Madagascar in comparison to GPCP. The failure of the model to capture the local maximum over Madagascar may be attributed to the overestimation of deep convection over the Mozambique Channel, as the strong convection over the channel can suppress convection over the Malagasy Island, leading to a lack of local maximum rainfall. Additionally, the deep convection over the Mozambique Channel may be due to the dry bias in MPAS simulation over Mozambique and southern Tanzania. Despite that, MPAS shows good agreement with GPCP (r=0.89) over the southern African continent, where all datasets show more rainfall in the tropics (6-8mm/day) and lesser rainfall near the subtropics (about 2-3 mm/day).
MPAS generally agrees with the two reanalysis datasets (ERA5 and 20C) on the spatial pattern of the 500hPa geopotential height (Fig. 3). All the datasets show areas of high geopotential height over the tropics and low geopotential height poleward of 30°S. However, the MPAS model gives higher values of the 500hPa geopotential height over much of the study area compared to the ERA5 reanalysis. The MPAS bias is more than 10m for much of the area, but it is comparable with the difference between the two reanalyses (i.e. reanalysis uncertainty). There are also inconsistencies between the model simulation and the reanalysis over parts of Sub-Saharan Africa and over the Indian Ocean. For instance, the model shows a high-pressure centre (which resembles the Botswana High) farther east as compared to the reanalysis, which shows the high-pressure centre between Namibia and Botswana’s border. Furthermore, MPAS struggles to capture the overall pattern of the 500hPa geopotential height over the Indian Ocean as compared to the reanalysis. These inconsistencies may be due to the low 240km resolution in the model simulation. Using a higher resolution may improve the model results. Regardless of that, MPAS features a high spatial correlation (r=0.98) with respect to the ERA5 reanalysis.
3.2 Spatio-temporal distribution of the Botswana High
There is a good agreement between MPAS and reanalysis data on the variability of the pressure gradient over southern Africa as well as the pattern and location of the Botswana High, although with some discrepancies (Fig. 4). In each of the datasets, EOF analysis indicates that mode 1 (EOF1) has a pattern reminiscent of the Botswana High, which is centred between 10°S and 25°S with an area of low pressure south of the ridge. Compared to the actual summer position of the Botswana High, EOF1 is displaced slightly equatorward as in Reason (2018). Despite that, EOF1 in the MPAS model shows an area of high pressure centred between the borders of Botswana, Namibia, Angola and Zambia and accounts for approximately 80% in the variance of the 500 hPa geopotential height in the region, which is similar to ERA5 (85%) and 20C (83%). The 80% variance has also been found by Reason (2018). However, the MPAS high shows a broader centre located farther north in comparison to the reanalysis whilst 20C shows a weaker and narrower centre which is located more eastward compared with the ERA5 and MPAS. Additionally, the combination of the model data and reanalysis EOF analysis (Fig. 4d) shows a similar pattern to Figs. 4a-4c, but more comparable with the ERA5 high and explains 82% variance in the 500 hPa geopotential height over the region.
The time series associated with EOF1 shows a strong agreement among the three datasets (MPAS, ERA5 and 20C reanalysis) in depicting the interannual variability of the Botswana High. However, the 20C reanalysis exhibits a weaker Botswana High in comparison to ERA5 and the MPAS results (Fig. 5). In addition, even with the combination of the datasets (as in Fig. 4d), the time-series results (figure not shown) are comparable to Fig. 5, indicating that EOF1 shows the same feature in both the model and reanalysis. In agreement with the previous studies by Reason (2016) and Reason (2018), Fig. 5 shows a strong correlation between ENSO and the Botswana High Index during the study period. All three datasets indicate that the Botswana High was strongest in the years 1983, 1998 and 2010, which is consistent with Reason (2016). In addition, all three cases of the strongest Botswana High occurred during the mature phase of El Nino summers (1983, 1998 and 2010), while the majority of the weakest Botswana Highs occurred during La Nina summers except for 1994, which occurred in a neutral ENSO year. However, MPAS simulation indicates that the weak Botswana Highs occurred in 1985, 1986, 1989, 1999 and 2008, while ERA5 shows the weak high in 1985, 1986, 1989, 1996 and 2000. On the other hand, the 20C showed the highest number of weak highs (1985, 1986, 1989, 1994, 1996, 2000, 2001 and 2008), possibly due to the generally weaker high in the reanalysis. However, the time-series shows that the MPAS results strongly correlated with that of ERA5 (r=0.83, p<0.0001), meaning that the MPAS model has a strong skill in simulating the interannual variability of the Botswana High.
A wavelet analysis of EOF1 in the reanalysis and model simulation indicates that the Botswana High is dominated by a strong 4-5-year cycle band before the early 1990s and a weak 2-3-year cycle band afterwards (Fig. 6). All datasets show similarity in the 4-5-year cycle between 1981-1989; however, the power spectrum is stronger in the ERA5 than in 20C and MPAS. There are also some discrepancies between the MPAS and reanalysis on the 2-3-year cycle. For example, before the 1990s, the 2-3-year periodicity was weaker in MPAS than the reanalysis but stronger between 1995-2001 and 2007-2010. The dominance of the 4-5-year and 2-3-year periodicities of EOF1 compares well with the ENSO wavelet power spectrum (Fig. 6d) and suggests that the Botswana High variability may be related to forcing from ENSO that also has leading periodicities in these two frequency bands.
3.4 The influence of SSTs on the Botswana High
The correlations between the JFM Botswana High index and SST anomalies in different seasons (OND, JFM, and AMJ) indicates a good agreement between MPAS, ERA5 and 20C (Fig. 7). All the datasets feature a high correlation between the Botswana High index and global SST anomalies. During early summer (OND), the most significant correlations occur over the North Atlantic as well as parts of the South-West and North-East Pacific. However, during concurrent summer (JFM), the correlation becomes stronger over the tropical Pacific, tropical Atlantic, South-West Pacific, tropical Indian Ocean, Enderby Plain (60°S, 40°E) as well as over the Kerguelen Plateau (43°S, 72°E). In the Autumn (AMJ), the correlation decreases over the tropical Pacific as well as over the South-West Pacific and parts of the southern Indian Ocean. The decrease in correlation over the Pacific indicates that the Botswana High is highly correlated with the early development phase of ENSO and progressively becomes weaker during the mature phase.
To quantify the link between the Botswana High and ENSO, Table 1 shows the coefficient of correlation between the Botswana High and Nino 3.4 seasons. All datasets indicate a strong correlation between Botswana High and ENSO in JFM; however, the correlation becomes rapidly weaker when negative lags are introduced, indicating a very weak relationship between the Botswana High and decaying phase of ENSO. On the other hand, introducing positive month lags, the correlation becomes progressively weak but remains strong (0.61 - 0.69) at 95% significance up to a lag of 2, which is related to the ENSO development phase. At lag 3, which is indicative of the mature phase of ENSO, the correlation becomes weaker heading to the decaying phase of ENSO in lag 4 and 5. Despite the strong correlation between the Botswana High and the tropical Indian Ocean SST (as shown Fig. 7), there seems to be no link between the high and the Indian Ocean dipole index (Table 1).
To further assess the relationship between the Botswana High and ENSO, Fig. 8 shows composite anomalies of JFM 200hPa velocity potential and stream function (contours) during strong Botswana High Years (1983, 1998 and 2010) and weak Botswana High Years (1989, 1994 and 2008). The results show that +ve Botswana High years are characterised by anomalous upper tropospheric divergence (negative velocity potential) west of the prime meridian (i.e. 0° longitude) and anomalous upper tropospheric convergence (positive velocity potential) east of the prime meridian and vice-versa during -ve years. This upper- level convergence-divergence pattern during +ve(-ve) Botswana High years is reminiscent of the weakening (strengthening) of the Walker Circulation and typically forms during El Nino (La Nina) years. Furthermore, the increase (decrease) in upper-level convergence over southern Africa may lead to increased (decreased) subsidence over the region and the strengthening(weakening) of the Botswana High. The 200hPa eddy stream function during +ve(-ve) Botswana High years are characterised with anticyclonic(cyclonic) anomalies in the upper troposphere over the central and eastern pacific. This pattern is characteristic of a Gill-Matsuno type response, which typically occurs during El Nino (La Nina) Years due to warm(cold) SST anomalies in the tropical Pacific Ocean (Gore et al. 2019). This response indicates a weakening (strengthening) of the upper-level cyclonic flow in the eastern Pacific and the weakening (strengthening) of the Walker Circulation. During +ve (-ve) years, the upper-level stream function shows anomalous cyclonic (anticyclonic) flow over the Indian Ocean, Asia and southern Africa, which corresponds to the increase (decrease) in convergence over the area leading to lower-level subsidence (convection).
Figures 9 and 10 show the anomalous 500hpa geopotential height and vertical cross-section of the geopotential height during +ve and -ve Botswana High years. In general, the position of the Botswana High varies between the datasets during +ve years (Figs. 9a, 9c and 9e). The MPAS model shows a Botswana High located over northwestern South Africa and southeastern Botswana (Fig. 9e), while ERA5 shows the high over much of Namibia (Fig. 9a). However, 20C reanalysis fails to adequately capture the net sinking motion (ω>0) associated with the Botswana High over Namibia, Botswana or South Africa (Fig. 9c). The reason for this is that the 20C Botswana High is weaker (Fig. 10c) as compared with ERA5 (Fig. 10a) and MPAS (Fig. 10e). This may also explain why the 20C EOF 1 (Fig 4b) was weaker as compared to ERA5 and MPAS (Figs. 4a and 4c). Nevertheless, all the datasets show sinking motion associated with the Botswana High, which is characterised by rising air to the west of the High, which sinks and manifests as upper-level convergence in the vicinity of the high, leading to increased subsidence in the mid-levels. Figures 9b, 9d and 9f also show variability in the position of the Botswana High during -ve years, however, the position of the Botswana High is more similar as compared to +ve years. Despite that, MPAS agrees well with ERA5 reanalysis on the depth of the -ve Botswana High mode (Figs. 10b, 10d and 10f) and shows a -ve Botswana High that is characterised by lower divergence which might lead to increased convection.
3.5 Impact of Botswana High on climate variables and droughts
3.5.1 Positive phase of the Botswana High
Figures 11a, 11c, 11e, and 12a, 12c, 12e show the anomalous 500hPa geopotential height and anomalous surface temperatures, respectively, during +ve Botswana High years. The composite of standardised anomalies for 500 hPa geopotential height during +ve Botswana High years shows a good agreement between the MPAS simulation and reanalysis (20C, ERA5), in a sense that all datasets indicate enhanced geopotential height over the tropics and little to no change poleward of 30°S. Higher geopotential heights are typically associated with warm air masses and increased subsidence and may lead to an increase in temperature over the tropics. When an air mass descends, the pressure on the air mass increases. Because of the increase in pressure, the volume of the air mass decreases, increasing its internal energy, which manifests itself in the increase in temperature of that mass of air. This increase in the geopotential height agrees with Fig. 12, which shows that +ve Botswana High years are generally characterised by warmer temperatures over the land surface as well as over the tropical Atlantic and the tropical Indian Ocean. During +ve Botswana High years, the observational data shows above normal temperatures mostly over the eastern parts of the continent (Zimbabwe, Zambia, and Mozambique) with slightly warmer conditions over the northeast regions of South Africa (Figs. 12a, 12c and 12e). The reanalysis shows a similar pattern; however, the warmer conditions extend throughout the eastern half of South Africa. On the other hand, the model simulation shows above-normal temperatures for much of the eastern parts of the continent as well as over South Africa. In addition to that, the positive anomaly extends along the coastline of Namibia and Angola as compared to the CRU CFSR and ERA5. The above-normal temperatures in the MPAS simulation may be attributed to the anomalous local maximum in the geopotential height over the South African Development Community (Fig. 11e), which is stronger in MPAS as compared to the reanalysis (Figs. 11a and 11b).
MPAS and reanalysis datasets indicate that +ve Botswana High years are typically characterised by neutral to positive OLR anomalies (Figs. 13a, 13c, and 13e), lesser convection (Figs. 9a, 9b, and 9c) and lesser rainfall (Figs. 14a, 14b, and 14c). Positive OLR anomalies indicate suppressed convection, lesser cloud coverage and more radiation emitted back into space, which is consistent with the increased subsidence in the area. However, there are notable discrepancies between 20C, ERA5, and the model simulation of OLR, especially over the ocean. The differences in OLR may be due to the Tiedtke convective parameterisation scheme used in the model, which removes convective instability before resolved-scale motions can respond to it. Another reason could be that the Tiedtke parameterisation is sensitive to entrained air from the free troposphere, and thus convection can be reduced by dry, free troposphere (Ali et al. 2014).
The discrepancies in the position of the Botswana High sinking motion may impact rainfall patterns in the datasets over southern Africa (Fig. 14). In actuality, there is poor agreement between the datasets on the +ve phase rainfall anomalies over land. GPCP shows below-average rainfall for much of the southern African continent and mildly wet conditions over South Africa, northern Angola, and Tanzania. The reanalysis shows an agreement with GPCP over southern Africa, Angola, and Tanzania but shows disagreement with the observation over Madagascar, Zambia and the Democratic Republic of Congo (DRC). The MPAS model also shows disagreements with GPCP and ERA5 over Botswana and northern Namibia, where it simulates above average rainfall, whereas the other datasets show below-average rainfall. The MPAS model’s inability to adequately capture the precipitation anomalies may be due to the low resolution (240km) used in the simulation. A higher resolution simulation could allow for better representation of orography and surface fields vital for the initiation of convection in complex terrains (Hohenegger et al. 2008).
MPAS shows a broad agreement with reanalysis on drought patterns during +ve Botswana High years, although with some discrepancies. Foremost, all datasets show a warm bias in PET over much of the subcontinent (Figs. 15a, 15c and 15e); however, the model shows higher anomalies than reanalysis. Higher PET anomalies indicate enhanced evaporation, leading to more severe drought conditions over the region in the absence of precipitation. Despite the good agreement among the dataset on the PET pattern, the disagree on the meteorological drought patterns (i.e. SPI and SPEI) during +ve Botswana High years (Figs. 16a, 16c 16e, 17a, 17c and 17e). For example, both MPAS and 20C show normal to wet conditions over Namibia, while ERA5 shows parched conditions. On the other hand, both reanalysis datasets show a relatively dry bias over Botswana, Zimbabwe and the northeastern regions of South Africa, while the model shows a wet bias. Despite that, some areas show agreement between the datasets, such as Angola, parts of Tanzania, and central parts of South Africa. However, MPAS shows better agreement with 20C in SPEI anomalies than with ERA5. For instance, both MPAS and 20C show +ve SPEI anomalies over central Namibia, Angola and northeastern South Africa, while ERA5 shows severe drought over much of the region.
3.5.1 Negative phase of the Botswana High
MPAS and reanalysis data show that -ve Botswana High years are characterized by a decrease in the geopotential height (Figs. 11b, 11d, and 11f) which is typically associated with cooler air masses and is consistent with the decrease in the surface temperatures over land (Figs. 12b, 12d and 12f). During -ve years, the most obvious disagreement between the reanalysis and MPAS is the local minimum geopotential heights simulated over the Atlantic and the Indian Ocean as compared to the reanalysis. In addition to that, the reanalysis shows a slight increase in the geopotential height over the subtropics whilst the MPAS simulation indicates a decrease in the geopotential height (Figs. 11b, 11d, and 11f). In connection to the decrease in geopotential height over the region, both CRU CFSR and ERA5 show below normal temperatures over much of the continent, however, the model simulation shows not much change in temperature except over Zambia, the DRC, Tanzania and northern Mozambique. Despite this, the model shows a good comparison with observation and reanalysis regarding surface temperatures over the ocean. The datasets show two warm tongues over the Atlantic and a local minimum surface temperature south of Madagascar.
There is a better agreement in the OLR, omega, and rainfall patterns between the datasets during -ve Botswana High years as compared with +ve years. However, there are notable discrepancies between the model simulation and ERA5 reanalysis, especially over the ocean. In general, the datasets show negative OLR anomalies during -ve years are indicative of enhanced convection, which may lead to more cloud coverage and rainfall (Figs. 13b, 13d and 13f). More cloud coverage may increase the albedo effect leading to colder temperatures over the study areas as shown by Figs. 12b, 12d, and 12f, however, the effect of cloud cover on temperature is not singular as the role of horizontal advection is also important. Additionally, more convection implies higher and colder cloud tops that emit less radiation back into space. 500hPa Omega anomalies during -ve Botswana High years (Figs. 9b, 9d and 9f) indicate areas of convection in the mid-levels (~500 hPa) in Namibia, southern Angola, Zambia, Botswana and parts of the interior of South Africa. The anomalous convection in those areas is consistent with the increased rainfall over the region (Figs. 14b, 14d, and 14f). Both GPCP and ERA5 show wetter conditions for much of the southern continent except for southern Mozambique, Zimbabwe, and parts of Angola. Similarly, MPAS simulates wetter conditions over much of the southern African continent except over Namibia, however, these are weaker as compared to the observation and reanalysis. This result is consistent with the result of Fig. 13f, which shows weaker negative OLR anomalies for MPAS as compared to Figs. 13b and 13d.
The agreement between MPAS and reanalysis on drought patterns is better during the -ve Botswana High years than during +ve years. As expected, the datasets generally show a cold bias in PET anomalies over the region, which is indicative of cool (Figs. 12b, 12d and 12f), cloudy (Figs. 13b, 13d and 13f) and rainy conditions (Fig.14b, 14d and 14f). However, the MPAS model shows near-normal PET anomalies for South Africa, Botswana and Namibia, which disagrees with the reanalysis results. This discrepancy may be due to the near-normal temperatures simulated by the MPAS model in Fig. 12f, which the PET calculation is dependent on. The SPI also shows a good agreement between the datasets especially over South Africa, Botswana, Angola, Zambia, southern DRC, northern Mozambique and much of Tanzania (Figs. 16b, 16d and 16f). However, the model fails to capture the wet (dry) spells near Namibia (Zimbabwe and southern Mozambique) as in the reanalysis. Nonetheless, SPEI also shows a better agreement between the MPAS and reanalysis data during -ve Botswana High Years. Again, the datasets indicate +ve anomalies over much of the region except over Angola, and the southern coastline of South Africa (Figs. 17b, 17d and 17f).