1 Introduction

Climatology is a branch of science where experiments are very difficult, if not impossible, to perform. Therefore, we are bound to use as much information from observations as possible, to try to understand the mechanisms of climate change and its possible impacts on society and ecosystems. Although climate models and basic physical considerations unambiguously agree that an increase in atmospheric greenhouse gases must lead to generally warmer temperatures, regional climate changes are more uncertain, in particular concerning other climate variables. For instance, questions such as whether warmer climates will lead to increased or decreased precipitation over a particular region are much more difficult to answer. The climate of the past can help find an answer.

The observational record is, however, usually too short, spanning at most the last 200 hundred years and more usually only the last few decades. Many relevant questions about the present climate, such as the unprecedented character of current temperatures, climate trends, and climate extremes, are better addressed by looking beyond the period covered by observations. This goal can be partially achieved by analysing indirect climate information from the so-called proxy data—tree rings, lake sediments, etc.—that are natural archives, sensitive to past environmental conditions. The period covered by these natural archives can be vast, but the past few centuries span climate conditions that are, from the geological perspective, not very much different from the present and future climate, so that the lessons learnt there may find applications for the understanding of present climate trends.

Simulations with Earth System Models that cover the past few hundred years can, together with proxy data, provide useful insights about the relevant climate mechanisms. Here, each of these two sources of information serve as independent confirmation (or rebuttal) of the other. Both are inherently uncertain, displaying different sources of error, and their combination leads to more robust conclusions than each of them taken in isolation would be able to provide.

In the following sections in this chapter, we will review the existing literature on climate variability in southern Africa during approximately the past millennium. We start with a summary of evidence available from proxies, both at the global scale and more specifically for southern Africa. These sections are followed by selected results obtained from climate simulations. Finally, we discuss their agreements and inconsistencies and conclude with the main take-home implications for future climate changes in this region.

2 The Climate of the Past Millennium: Global Background

Our knowledge of the climate of the past millennium is derived from indirect indicators that archive information about past environmental conditions. Trees tend to form thicker annual growth rings or produce wood of higher density in years with more suitable environmental conditions, usually warmer and/or wetter. These biological characteristics can be calibrated to reconstruct past physical magnitudes such as temperature or precipitation variations by statistically comparing recent tree rings to meteorological observations. Apart from dendroclimatological data, other proxy records also contain information from past environmental conditions: carbon and oxygen stable isotopes in old wood, oxygen isotopes in stalagmites, pollen assemblages in lake sediments, historical documents, etc. In this fashion, global networks of proxy data can be translated by means of complex statistical methods to annually resolved climate patterns in past periods (Li et al. 2010). These climate reconstructions, however, rely on some general assumptions that may not be always fulfilled. For instance, dendrochronological proxy records reflect the environmental conditions during the growing season, i.e., are seasonally biased. Also, other non-climatic factors may affect the growth of trees, such as availability of nutrients, fires, etc. Other types of records suffer from other corresponding caveats, so that it is not totally surprising that discrepancies between reconstructions derived from different proxy records arise. This highlights the need to combine different sources of information to reach robust conclusions.

In addition, depending on the method applied to translate the proxy information and on the network of proxy data, the reconstructions of past climate may differ on the amplitude of past climate variations and on the specific regional details. However, most of the temperature reconstructions published so far indicate that the Earth’s climate of the past millennium can be described by relatively warm centuries around year 1000 CE (common era)—the Medieval Climate Anomaly or Medieval Warm Period, followed by colder centuries between around 1500 CE and 1850 CE—usually denoted as the Little Ice Age—which in turn were followed by a warming trend that has strongly intensified from around 1980 CE onwards until present (Crowley 2000).

These centennial climate fluctuations have been attributed to different external climate forcings (Schmidt et al. 2012). One is volcanic activity that tends to cool the global climate due to the volcanic aerosols ejected to the stratosphere, dimming the incoming solar radiation. Solar output itself is also variable through time. Land-use and forest cover can modulate the regional climate due to the implied changes in the surface reflectivity. Concerning land use, major changes have occurred in some regions such as Europe and East Asia over the past few centuries. Finally, greenhouse gases since the industrial revolution have contributed to the warming trend since the end of the Little Ice Age and are the single most important factor for the warming since the mid-twentieth century.

The variations of these external factors can be reconstructed by the analysis of polar ice cores. They archive the composition of the past atmosphere. Ice acidity records show sharp peaks due to the deposition of volcanic aerosols, allowing for an accurate dating and estimation of the strength of eruptions. The concentrations of cosmogenic isotopes such as \({ }^{10}\)Be are indicative of past solar activity. These records show that around the Medieval Climate Anomaly volcanic activity was sparse and the Sun was stronger than in the ensuing centuries. By contrast, the Little Ice Age witnessed an intense and frequent volcanic activity and a weaker Sun. The recent decades are characterised by an almost constant solar output, relatively weak volcanism, and a very strong forcing due to greenhouse gases.

In addition, internal climate variations, not caused by any particular external factor but due to the slow variations of ocean currents and the inter-play between ocean and atmosphere, could have also contributed to some of these past variations. For instance, it is unclear yet as to whether the external factors could have been solely responsible for the Medieval Climate Anomaly or whether some sort of slowly varying internal mechanism might have contributed to generally warmer temperatures. This is perhaps more relevant at regional scales, for which internal climate patterns such as the El Niño-Southern Oscillation (ENSO) may have a stronger immediate influence than global external forcings.

3 The Climate of the Past Millennium: Southern Africa

Earlier literature reviews on the paleoclimate of southern Africa during the last 2000 years (Tyson and Lindesay 1992; Hannaford and Nash 2016) have identified the warm and cold climate phases in the past millennium previously mentioned in Sect. 5.2. A long 3000-year-long stalagmite record from Cold Air Cave in the Makapansgat Valley, which displays colour banding that is correlated to local temperature, also confirms the sequence of warm–cold–warm periods over the past millennium (Holmgren et al. 2001). According to this record, the Little Ice Age would have been about 1 ∘C colder than present. This is approximately confirmed by a stalagmite-based \({ }^{18}\)O isotope record from the same site spanning the past 350 years, which has been interpreted as indicators of a sharp and well-defined cold multi-decadal period centred around 1720 CE (Sundqvist et al. 2013). The cooling may have amounted to 1.4 ∘C colder than present. The rise and fall in temperature in these different thermal phases would have been roughly homogeneous over the whole region.

The picture derived for precipitation is more nuanced. As explained in Chap. 6, southern Africa is characterised by two regions with different annual precipitation regimes: a (mostly) winter-precipitation region around Cape Town and a summer-precipitation region located further to the east and northeast (Reason 2017). During the Little Ice Age, the winter-precipitation region may have received more precipitation due to a northward displacement of the belt of westerly winds (Dunwiddie and LaMarche 1980), whereas the summer-precipitation zone of southern Africa would have faced a generally drier climate due to diminished evaporation from the Indian ocean leading to lower air humidity (Woodborne et al. 2015). The reversed precipitation pattern is found in the prior warmer centuries during the Medieval Climate Anomaly. In this period, proxy indicators of precipitation in the summer-rainfall region based on stable carbon isotopes in baobab trees show increased rainfall (Woodborne et al. 2015).

Thus, an important feature of the pattern of the variability of annual precipitation totals in southern Africa, as derived from proxy information, appears to be a see-saw pattern between the winter-precipitation zone in the southwest and the summer-precipitation zone in the northeast (Fig. 5.1). Whether or not this see-saw type of variability recurs over time no matter which is the main external climate driver is indeed an interesting question, relevant for future climate changes.

Fig. 5.1
A line graph depicts reconstructions of Southern African precipitation in delta 13 C and ring index versus the calendar year. The values of the curves highly fluctuate.

Reconstructions of southern African precipitation over the past 500 years in the winter-rainfall zone (blue, left axis, based on cedar tree-ring widths) and summer-rainfall zone (red, right axis, based on concentrations of \( {{ }^{13}C}\) in baobab trees). Figure copied from Woodborne et al. (2015) (freely available)

Full size image

The nineteenth century was relatively colder than the twentieth century. This period, with a more dense network of direct observations, offers the opportunity to test that working hypothesis suggested in the previous paragraph. Unfortunately, the available studies that provide more detailed analysis for the nineteenth century precipitation variations reach contradicting conclusions, indicating either drier (Nicholson et al. 2012) or wetter conditions (Neukom et al. 2014; Nash et al. 2016; Nash 2017) during the nineteenth century in the summer-precipitation zone. By contrast, precipitation in the winter-rainfall zone has likely remained temporally stable over the last two centuries (Nash 2017). An explanation for this discrepancy may lie in the different nature of the records analysed. Whereas Nicholson et al. analysed long instrumental and documentary records, the conclusion reached by the other studies is derived from a more comprehensive set of data, including, in addition to instrumental and documentary records and indirect proxies (dendroclimatological, corals). This highlights the difficulty of inferring past climates and the need to combine all available sources of information.

4 Paleoclimate Simulations with Earth System Models

Comprehensive climate models, very similar to those used to project the impact of greenhouse gases on future climate (Edwards 2011), have also been used to retrospectively simulate the climate of the past millennium (Fernández-Donado et al. 2013). These models, akin to weather prediction models, incorporate our knowledge of the main climate processes. They contain a representation not only of the atmosphere, but also of the ocean, of sea-ice, of soils and some of them also of the terrestrial and oceanic biosphere. All in all, they are one of the most complex software packages actually in use. Nevertheless, the climate system is very complex, with processes that typically occur over a vast range of spatial and temporal scales, from seconds to millennia and from millimetres to thousands of kilometres. Due to computing limitations, some of these processes need to be represented in a more simplified form. Typically, a climate model has a spatial resolution of about 100 km, and all smaller-scale processes are represented in an averaged fashion. Climate models, for instance, do not directly simulate the formation of clouds, but only the average effect of clouds over an area of typically 1~00 km long and wide. This leads to inaccuracies and uncertainties in the simulation of precipitation and to differences in climate projections obtained with different models. In spite of mentioned model uncertainties, climate models are indeed able to reasonably replicate the two precipitation regimes observed in southern Africa. An example is shown in Fig. 5.2, which can be compared with the corresponding figure in Chap. 6. This means that the main mechanisms behind these two precipitation regimes, namely extratropical cyclones for the winter-precipitation zone and convective precipitation within the South Indian Convergence Zone for the summer-precipitation zone (Cook 2000; Reason 2017), are reasonably well simulated by global climate models.

Fig. 5.2
A bar graph depicts millimeters in a month versus month for winter rainfall zone, and summer rainfall zone. The summer-rainfall zone occupies more values, and the winter rainfall zone occupies less values.

An example of the annual cycle of precipitation in the two southern African precipitation zones simulated by the climate model MPI-ESM-P in the pre-industrial centuries (850–1800 CE). The magnitude of precipitation depends on the location of the selected model grid cells

Full size image

To simulate the climate of the past, climate models need to be driven by the external factors that provide or modulate the energy that reaches the Earth. As mentioned in Sect. 5.2, in the past millennium, these external factors were volcanism, solar output, and atmospheric greenhouse gases (Schmidt et al. 2012). The magnitude of these factors in the past millennium can be approximately reconstructed from chemical analysis of polar ice cores, including the air bubbles trapped in them, and then used for climate simulations. There exists a relatively large set of climate models that has been used to estimate future climate change, as included in the different reports by the Intergovernmental Panel on Climate Change (IPCC). Some of these models have also been used to simulate the climate of the past centuries, more precisely the period 850–2005 CE. These simulations are part of the Coupled Model Intercomparison Project CMIP5 (Taylor et al. 2012). As indicated in the Introduction, the results of these simulations may differ from model to model, so that it is necessary to consider several models not only to identify the robust results but also to be aware of the uncertainties inherent in these simulations.

Figure 5.3 displays the near-surface air temperature in southern Africa simulated by a suite of climate models in the period 850–2005 CE. Table 5.1 lists these climate models and their spatial resolution. All simulated temperatures display clear similarities but also differences. Most models, with the only exception of HadCM3, estimate colder pre-industrial temperatures of the order of 0.5 ∘C relative to the twentieth century mean. There seems to exist little doubt, from the models’ perspective, that temperatures in the twentieth century have clearly been above the pre-industrial average level. The model HadCM3 estimates twice as cold pre-industrial temperatures with respect to the twentieth century mean.

Fig. 5.3
A line graph depicts deviations from twentieth century mean, and 20-year running means in degrees Celsius versus year for M P I-E S M = -P, C C S M 4, Had C M 3, I P S L, G I S S-E 2-E R, and B C C-C S M 1. The values of the curves highly fluctuate.

Near-surface annual mean air temperature averaged in southern Africa (land areas between 10S-40S and 10E-45W) as simulated by a suite of climate models from the Climate Model Intercomparison Project CMIP5 for the period 850–2005 CE (Taylor et al. 2012). The time series represent deviations from the twentieth century mean temperature and have been smoothed with a 20-year running-mean filter

Full size image
Table 5.1 List of global climate models and their spatial resolution (atmospheric model only) in geographical degrees used in this study
Full size table

The warming trend simulated during the twentieth century stands out compared to the trends in all other centuries, with the possible exception of the nineteenth century. The climate of the nineteenth century is, however, strongly impacted by the two first decades of intensive volcanism (see next paragraph), and by its associated strong cooling. The models that estimate the strongest impact of these eruptions (e.g., GISS-2-ER) are the ones for which the ensuing warming trend is also stronger. Even so, considering all models, the simulated twentieth century warming, about 1.2 ∘C, is larger than the bracket of nineteenth century warming of 0.5 –0.8 ∘C spanned by the majority of models.

Further back in time, superposed to colder pre-industrial mean temperature, the models produce temperature variations that can be attributed to the impact of the external forcings. For instance, the clear cooling simulated after the mid-thirteenth century is due to the very strong eruption in Samalas (Indonesia) in 1258 CE (Guillet et al. 2017). Also clear in the figure are the series of eruptions in the early nineteenth century, one of them the famous Tambora eruption in 1815, which caused in Europe “the year without summer” (Raible et al. 2016) and serious societal disruptions there in 1816 CE. After around 1800 AD, the current warming trends set in, mainly caused by an increase of the solar output during the nineteenth century and by the increase of greenhouse gases in the second half of the twentieth century.

The succession of warm–cold–warm centennial or even multi-centennial periods (MCA-LIA-present) previously identified in the proxy-based reconstructions is not so clearly recognised in the simulations. Although the centuries around 1000 CE are slightly warmer in the simulations, the LIA is barely recognisable in the simulated series as a clearly cold differentiated centennial or multi-centennial period. As a result, the simulated temperature evolution during the pre-industrial period appears rather stable, with the interruptions caused by volcanism. In this respect, all models seem to agree. The possible reasons for the discrepancies between models and reconstructions are later discussed in Sect. 5.5.

Concerning past changes in precipitation, we again need to differentiate between the winter-rainfall zone and the summer-rainfall zone (see Chap. 2, Fig. 2.2). Since the mechanisms behind precipitation in these two zones are different, their long-term evolution could also diverge. The evolutions of simulated precipitation in these two zones in the past millennium are displayed in Fig. 5.4. Precipitation in the winter-rainfall zone (around Cape Town and along southern coast in the pre-industrial period) is in all simulations generally larger than in the twentieth century. The decadal variations of the pre-industrial mean precipitation are, however, much larger than for temperature, and a few decades in the simulations are indeed as dry as some decades of the twentieth century. Generally, the models do seem to indicate a tendency towards a drier climate in this region in the recent decades, in accordance with the analysis of observational data shown in Chap. 6. Precipitation in the summer-rainfall zone (continental northeast during the pre-industrial period, Fig. 5.4b) also shows a slight tendency to be larger than during the twentieth century. However, the suite of models displays a larger spread than for the winter-rainfall zone, so that the uncertainty is here also larger. One model, again the model HadCM3, behaves rather differently from the others and shows the opposite result.

Fig. 5.4
2-line graphs depict deviations from the twentieth century, 20-year running mean in millimeters versus year for M P I-E S M-P, C C S M 4, Had C M 3, I P S L, G I S S-E 2-E R, B C C-C S M 1 in winter, and summer rainfall zone. The values of the curves highly fluctuate.

Near-surface annual precipitation in southern Africa (land areas between 10S-40S and 10E-45W) as simulated by a suite of climate models from the Climate Model Intercomparison Project CMIP5 over the period 850–2005 CE, (a) for the winter-rainfall zone and (b) summer-rainfall zone (see Chap. 6). The time series represent deviations from the twentieth century mean precipitation and are smoothed with a 20-year running-mean filter

Full size image

Are the variations of the precipitation in the summer-rainfall and winter-rainfall zones mutually related? This is an interesting question regarding future climate change. If, for instance, the see-saw pattern suggested by proxy data (Sect. 5.3) is temporally stable, it is plausible that it will also be present in future climate, with enhanced rainfall in the summer zone and weaker in the winter zone, or vice versa. The model simulations, however, do not support this result. The temporal correlations, within each simulation, between annual mean precipitation in the two zones are very small (all cases smaller than 0.2). This happens irrespective of the timescale, for instance, after decadal or multi-decadal smoothing of the time series. All in all, the simulated precipitation shows rather wide interannual (not shown) and decadal variations, compared to any centennial or multi-centennial trends. The possible reasons for the disagreement between models and reconstructions are discussed in Sect. 5.5.

Variables other than temperature or precipitation may also be important to characterise past climates and estimate the impact of climate change on ecosystems. This is the case for wind, and more particularly coastal winds. Winds flowing along the western southern Africa coast are the main drivers of coastal upwelling in the Benguela Upwelling System and, therefore, are critical for the biological productivity of that ocean region. How upwelling could change under anthropogenic climate change was the focus of the hypothesis put forward by Bakun (1990); Bakun et al. (2015). According to this hypothesis, the intensity of upwelling-favourable winds should increase in the future due to the widening temperature difference between continental and oceanic surface caused by global warming. The studies that have analysed this hypothesis using observations or climate simulations are, however, not univocally conclusive (Sydeman et al. 2014). A confirmation by possible paleoclimate data from oceanic sediment cores, which may record past ocean productivity, and paleoclimate simulations could in theory shed light on this question. However, the analysis of paleoclimate simulations over the past millennium does not indicate that the intensity of upwelling-favourable winds had in the past varied hand-in-hand with global temperatures (Tim et al. 2016). Thus, these simulations do not in principle support Bakun’s hypothesis. This can happen because the effect predicted by Bakun’s hypothesis has been in the past too small compared to the natural wind variations, or because the modelled variations of past temperatures are unrealistically too narrow (Sect. 5.5). Other possible explanation assumes that state-of-the-art global climate models cannot realistically represent the reaction of coastal wind to variations in external forcing yet, due to their too coarse spatial resolution (Small et al. 2015).

In summary, the results from the global climate simulations regarding southern Africa can be interpreted as follows. The past evolution of temperatures clearly shows the impact of volcanic eruptions and, in the twentieth century, the impact of anthropogenic greenhouse gases. The twentieth century would have been the warmest of the past millennium. By contrast, precipitation would have varied little along the whole millennium, with a tendency of the twentieth century to be somewhat drier than the previous centuries but only in the winter-rainfall zone. The simulations do not show clear, well-defined, warm, cold, wet, or dry centuries in the pre-industrial centuries. Thus, the pre-industrial period appears in the simulations as a rather stable climatic background against which the 20th temperatures, but not so much precipitation, clearly stand out.

5 Comparison Between Proxy Data and Model Simulations

The combined analysis of climate simulations and climate reconstructions aims at a better understanding of the processes of climate variability and change and helps identify robust model results but also deficiencies that need to be addressed to increase our confidence in spatially detailed future climate projections.

The simulated temperatures over the past millennium show agreements with proxy-based reconstructions but also clear discrepancies. Both clearly suggest that the twentieth century has been a very warm period against the backdrop of the pre-industrial centuries. However, looking further back into the past, several climate reconstructions using independent data do show a well-defined cold period—the Little Ice Age, around 1700 CE—differentiated from a prior warmer period during medieval times. The simulations do not show these thermally differentiated periods. Instead, the simulations yield rather stable pre-industrial temperatures, punctuated by decadal cooling episodes caused by volcanic eruptions, which do not extend over several centuries.

One reason for this discrepancy may lie in the external forcing used to drive the climate models. The forcing used in the CMIP5 model suite of past millennium simulations follows a commonly agreed protocol that incorporates the knowledge of past variations of solar output and volcanism. These estimations are to some extent uncertain. Whereas volcanic forcing is relatively short-lived, the variations of solar output display longer timescales, with cycles of several hundred years. The assumed amplitude of those solar variations of past solar output in the CMIP5 protocol represents a temporally narrower version of previous solar output reconstructions, which tended to have much wider, even tenfold, amplitude of variations (Schmidt et al. 2012). It is conceivable that this assumed reduction of the amplitude of past solar output variability turns out to be somewhat unrealistic. Stronger solar output variability in the past could have led to wider temperature variability at centennial timescales.

Concerning past precipitation, the proxy-based reconstructions in southern Africa indicate two important characteristics. One is the link between past precipitation and past temperature. Precipitation in the summer-rainfall zone varies, according to the reconstructions, in accordance with temperature, with lower precipitation during the Little Ice Age and higher precipitation during the Medieval Climate Anomaly. In the winter-rainfall zone, by contrast, the link between precipitation and temperature is opposite. Thus a second characteristic is the opposite variations in the summer-rainfall and winter-rainfall zones.

The simulations do not show this behaviour. Since the simulations do not present clearly defined thermal sub-periods, they cannot show a link between temperature and precipitation variations. Also, the simulated precipitation in the summer- and winter-rainfall zones does not appear anti-correlated in any of the model simulations. Should the see-saw character of precipitation variability be confirmed by further analysis of additional proxy records, it would raise the question as to why the models fail in this respect, and whether this deficiency is critical for more robust projections of future precipitation.

It is known that the simulation of precipitation can be a challenge for global climate models. Due to their coarse resolution (Table 5.1), the much smaller-scale formation of clouds and the condensation of water vapour to liquid water cannot be explicitly represented in the models. Instead, they use heuristic equations that translate large-scale humidity convergence and atmosphere stability to average precipitation over a whole grid cell. These heuristic, empirically derived, equations are prone to errors. One possible explanation is, therefore, that precipitation in the summer-rainfall zone, which is caused by convective systems associated to the South Indian Convergence Zone (Cook 2000), might not be perfectly simulated. However, other studies raise the question as to whether climate models realistically represent the connection between large-scale patterns of climate variability, such as ENSO, and southern African rainfall. It has been found that the CMIP5 models do not replicate the observed link between southern African rainfall and ENSO (Dieppois et al. 2015, 2019). Therefore, a targeted investigation on whether this deficiency is solved in the next generation of climate models is needed to increase the confidence of future precipitation projections in this region.

Concerning climate reconstructions based on proxy data, we have to bear in mind that the amount of those types of data in the southern African region is sparse compared to other regions, such as Europe or North America. On the other hand, there exist a few sources of historical data that can be further exploited. One example is the collection of log books from the British Navy, which contain valuable information about wind intensity and duration at daily timescales. These data can be used to reconstruct atmospheric variables in the region over the few past centuries (Hannaford et al. 2015). A more dense network of proxy data could allow to construct spatially resolved gridded climate field reconstructions (Tingley et al. 2012) covering the whole region, instead of a collection of climate reconstructions that refer to the relatively smaller subregions where the proxy records are located. A gridded reconstruction covering the whole southern African region would allow a much better comparison with the output of climate models. This also remains a research focus for the future.

6 Conclusions and Outlook

The main conclusions that can be derived from this brief review of the climate of southern Africa over the past millennium are as follows:

  • Proxy records, such as dendroclimatological data, indicate a succession of a relatively warm medieval period around 1000 CE, following by colder centuries denoted as the Little Ice Age, and a warm twentieth century.

  • Precipitation in the two relevant precipitation zones—winter-rainfall and summer-rainfall zones—was linked to the average regional temperatures, with precipitation in the summer-rainfall positively correlated to temperature. Precipitation in the winter-rainfall zone was negatively correlated to temperature.

  • Climate simulations with state-of-the-art climate models do indicate a particularly warm twentieth century but show no thermally differentiated sub-periods nor correlations between temperature and precipitation. They do show, however, that the twentieth century might have been dry in the context of the past millennium.

These conclusions already indicate a possible way forward for future research. The disagreement between climate reconstructions and model simulations should be clarified in order to buttress the future climate projections obtained with those same climate models. Thereby, a particularly important aspect is the correlation between temperature and precipitation changes.