Interannual variability of African rainfall impacts local and global communities, but its past behavior and response in future climate projections are poorly understood. This is primarily due to short instrumental records and a lack of long high-resolution palaeoclimate proxy records. Here we present an annually resolved 91,000 year Early Pleistocene record of hydroclimate from the early hominin-bearing Makapansgat Valley, South Africa. Changes in speleothem annual band thickness are dominated by precession over four consecutive orbital cycles with strong millennial-scale periodicity. The frequency of interannual variability (2.0–6.5 yr oscillations) does not change systematically, yet its amplitude is modulated by the orbital forcing. These long-term characteristics of interannual variability are reproduced with transient climate model simulations of water balance for South Africa from the Late Pleistocene to Recent. Based on these results, we suggest that the frequency of interannual variations in southern African rainfall is likely to be stable under anthropogenic warming, but that the size of year-to-year variations may increase. We see an orbitally forced increase in the amplitude of interannual climate variability between 1.8 Ma and 1.7 Ma coincident with the first evidence for the Acheulean stone tool technology.


Year-to-year variations in rainfall patterns in eastern and southeastern Africa are influenced by dominant modes of climate variability (Christensen et al., 2013) such as the El Niño/Southern Oscillation (ENSO; Cai et al., 2015) and the Indian Ocean Dipole (IOD; Saji et al., 1999). Substantial effort has been devoted to understanding the past (Tudhope et al., 2001; Emile-Geay et al., 2016) and future (Cai et al., 2015) behavior of these climate modes. However, there are additional long-term influences on African interannual rainfall variability, including sea-surface temperature (Nicholson, 2000; Wolff et al., 2011) and orbital precession (Clement et al., 2004). Instrumental records are too short to provide insights into modes of interannual climate variability, whose properties often vary on centennial time scales (Wittenberg, 2009). This leads to a reliance on proxy records (typically of less than 1000 yr) of past interannual variability from tree rings (Fowler et al., 2012), corals (Tudhope et al., 2001; Cobb et al., 2013; Emile-Geay et al., 2016), varves (Moy et al., 2002) and annually layered speleothems (Rasbury and Aharon, 2006). Speleothem time series of annual band thickness of up to 1000 yr duration have been obtained from Holocene African stalagmites (Brook et al., 1999; Baker et al., 2010). The length of these records has been limited by the sporadic preservation of annual laminae and reliance on manual band counting and measuring.


Here we present a replicated record of band thickness measurements from an Early Pleistocene flowstone from Buffalo Cave, South Africa (Fig. 1A). The flowstone is 2.4 m thick, horizontally laminated, and exposed at the base of a mined paleo-cave sequence (Hopley et al., 2007). Fluorescent bands were imaged using confocal microscopy (Fig. 1B; see the GSA Data Repository1). Two parallel grayscale lines from the same interval (Fig. 1B, Line 1 with 91,015 laminae and Line 2 with 91,502 laminae) were extracted from the photomosaic image and automated methods were used to count and measure the bands (see the Data Repository; Fig. DR1). We performed time-series analysis using smoothed periodograms derived using the Lomb-Scargle method applied to both the speleothem lamina data and the CCSM3 (Community Climate System Model) Transient Climate Evolution model (TraCE) simulation (Liu et al., 2014). Significant spectral peaks were distinguished from the power-law background exceeding the 1% false-alarm level (see the Data Repository for further details). Radiometric and magnetostratigraphic dating techniques are not sufficiently precise for the Early Pleistocene to resolve the annual nature of the laminae (see Figs. DR5–DR7). However, assuming that the laminae are annual and complete, the time interval covered by the laminae is the same as that derived independently from half-precession tie-points from a co-located oxygen isotope data set (Hopley et al., 2007; Fig. DR2). In addition, similar laminae from an Early Holocene flowstone from the same cave are demonstrably annual (Figs. DR3 and DR4). The absolute age of the Early Pleistocene flowstone interval is inferred to be between 1.81 Ma and 1.72 Ma from orbital tuning of the oxygen isotopes (Hopley et al., 2007).

The processes that determine the thickness of each individual lamina cannot be determined through direct observation, as the flowstone is no longer active, nor is the mined paleo-cave available for cave-monitoring studies. However, the active speleothems in Cold Air Cave (1 km southwest of Buffalo Cave; Fig. 1A) are sensitive to seasonal changes in rainfall (Finch et al., 2001) and to regional climatic variability (Sundqvist et al., 2013). Present-day variations in annual speleothem growth rates in this region depend primarily on water availability (Green et al., 2015). We consider variations in the Buffalo Cave lamina thickness to reflect the changing net water budget (precipitation minus potential evapotranspiration), primarily during austral summer (see the Data Repository for a discussion).

The two Early Pleistocene laminae thickness time series have power spectra with gently sloping backgrounds (i.e., following power laws; Figs. 2A and 2B), consistent with the properties of long records of measured precipitation (Fleming, 2013). Identification of regular cyclicity was based on the presence of spectral peaks, at the same frequency for both Lines 1 and 2, which exceed the 1% “false-alarm level” (see the Data Repository). Spectral peaks denoting cycles of periods of ∼23 k.y. (Figs. 2A and 2B) are consistent with orbital-precession forcing, as demonstrated previously for mean climate at Buffalo Cave (Hopley et al., 2007). Low-pass–filtered laminae thickness data for Lines 1 and 2 show 23 k.y. oscillations that are approximately in phase and have relative amplitude variations similar to the orbital precession (Figs. 3B and 3C).


Significant spectral peaks denoting cyclicity of ∼4.3 and ∼1.3 k.y. are recorded for Lines 1 and 2 (Figs. 2A and 2B). The periods of these peaks are similar to the 4.0–2.5 k.y. and 1.25– 0.75 k.y. cycles observed in the Late Pleistocene to Holocene δ18O record of the nearby Cold Air Cave (see Fig. 1) speleothems (Holmgren et al., 2003). Millennial-scale climate cycles with periods of 1.5 k.y. in the North Atlantic (Dansgaard-Oeschger events) have been demonstrated in the Late Pleistocene and Holocene (Bond et al., 1997). Similar climate cycles have been documented for many terrestrial temperature and hydrological (wet-dry) indicators (Wang et al., 2005). Such rapid climate changes are connected to variations in the thermohaline circulation controlled by meltwater pulses, and are known to modulate the strength of interannual variability (Liu et al., 2014; Turney et al., 2004).

Unlike ENSO temperature-related records from the Pacific (Tudhope et al., 2001; Rasbury and Aharon, 2006; Cobb et al., 2013; Emile-Geay et al., 2016), here we see no robust concentrations of variance in the 2–7 yr band (Figs. 2A and 2B) in the Early Pleistocene. Generation of low-resolution power spectra using the Bartlett method, using subsections 1000 yr long, demonstrates that the lack of concentrated power at these time scales is not an artifact of the exceptionally high-frequency resolution of the spectra in Figures 2A and 2B (see the Data Repository for methods; Fig. DR8). Similarly, there are no such concentrations of variance at the 2–7 yr scale in instrumental records of South African rainfall (Kane, 2009), Late Holocene corals from Madagascar (Zinke et al., 2004), the Early Holocene laminae record from Buffalo Cave (Fig. 2C), nor within the net water budget in South Africa for the past 22 k.y. from TraCE (Liu et al., 2014) (Fig. 2D).

The replicate time series were high-pass filtered with a cut off at 6.5 yr to extract interannual variability (Fig. 3), as well as band-pass filtered to extract decadal variability with periods of between 10 and 28.5 yr (Fig. DR9). The average period of both the interannual and decadal variability in 100 yr intervals appears to change randomly around a mean; there is no indication of orbital forcing (Fig. 3; Fig. DR9). In contrast, there are large long-term (>1000 yr) changes of the amplitude in both frequency bands, which become more prominent in the later portion of the record (Figs. 3D and 3F). The TraCE simulation (Liu et al., 2014) for the past 22 k.y. of summer water balance in South Africa shows similar long-term constancy of the period of interannual variations, but long-term changes in amplitude (Fig. 3).

The apparent discrepancy between the early 21st Century slowdown of rising tropical surface temperatures and climate model projections has been attributed to multi-decadal variability being undersampled by models (Kosaka and Xie, 2013). The Buffalo Cave record suggests that the amplitude of decadal climate variability is underestimated in the TraCE simulation (Figs. DR10 and DR11), as implied by the instrumental record (Laepple and Huybers, 2014). Very long, annually resolved paleoclimate records, such as presented here, provide further insight into multidecadal variability and, in turn, better representation in climate models.

Previous paleoclimate (Tudhope et al., 2001; Moy et al., 2002; Conroy et al., 2008; Koutavas and Joanides, 2012; Carré et al., 2014) and modeling (Clement et al., 2000; Timmermann et al., 2007; Zheng et al., 2008) studies have found a reduction in the frequency and/or amplitude of interannual variability during the mid-Holocene boreal autumn insolation maximum, although recent studies (Cobb et al., 2013; Emile-Geay et al., 2016) have suggested that the link between interannual variability and precession can be masked by large millennial-scale changes. In contrast to these discontinuous proxies, the Buffalo Cave record clearly shows both orbital and millennial-scale changes in the amplitude of interannual rainfall variability. However, an ensemble of climate model simulations of Last Glacial Maximum (LGM) to recent rainfall anomalies suggest that the present-day teleconnection between ENSO and southern African rainfall is not robust to changes in orbital forcing (Fig. DR12), explaining the lack of variance in the 2–7 yr band seen in the region (Fig. 2). Therefore, the Buffalo Cave record of rainfall variability should not be viewed as a proxy for ENSO amplitude, but rather represents a local hydroclimate response.

There is an increased amplitude of interannual variations at ca. 1.775 Ma (Fig. 3F) that is temporally coincident with the emergence of the Acheulean lithic technology (Fig. 3A; see the Data Repository for a discussion). The increased amplitude (Fig. 3B) may be a consequence of the shift to larger precession changes (Clement et al., 2004), but could also be plausibly linked to an enhancement of the tropical circulation at ca. 1.8 Ma (Ravelo et al., 2004; Hopley et al., 2007).


It is still an open question as to what effect future anthropogenic climate change will have on tropical climate variability. Model simulations of the 21st Century do not show robust changes in either the frequency or dynamics of interannual variability, such as ENSO (Collins et al., 2010; Stevenson, 2012); however, there are robust amplitude increases in their hydroclimatic response (Fig. DR 13; Cai et al., 2015). Along with increasing extreme floods and drought, this is a simple response to the higher saturation vapor pressure of warmer air (Christensen et al., 2013). Orbital variations are known to change the circulation, temperature, and hydroclimate of subtropical Africa (Joussaume et al., 1999). This new record demonstrates that past interannual hydroclimate variability was controlled, at least in part, by local temperatures (associated with increased local solar insolation). It provides robust observational confirmation of the expected long-term increase in the amplitude of South African rainfall variability during anthropogenic warming (Christensen et al., 2013).

It has been suggested that long-term climate changes influenced morphological and technological change in hominin evolution (Antón et al., 2014). The development of plausible mechanisms for this influence has been hampered by a lack of knowledge about how those long-term climate changes were manifested during an individual’s lifespan. This record (and others developed using the novel method described here) could provide essential data for exploring the role, if any, that climate variability and resource uncertainty played in hominin evolution.


We thank Mark Maslin for reading the manuscript and for supporting this project from its initiation. Thanks to Steve Noble for technical support at NERC Isotope Geosciences Laboratories and to Ming Tan and two anonymous reviewers for their constructive comments. Funding was provided to Hopley by the Natural Environment Research Council (NERC, UK), awards NE/J00443X/1 and IP/1065/1108. Weedon was supported by the Joint UK DECC/Defra Met Office Hadley Climate Centre Programme (GA01101). Herries was supported by Australian Research Council Future Fellowship FT120100399. Brierley is grateful to the substantial effort of the TraCE modeling and the Paleoclimate Modelling Intercomparison Project (PMIP) modeling groups, who have kindly donated all the simulation output. TraCE-21ka was made possible by the U.S. Department of Energy (DOE) INCITE computing program, and supported by the National Center for Atmospheric Research (NCAR), the U.S. National Science Foundation P2C2 program, and the DOE Abrupt Change and EaSM programs.

1GSA Data Repository item 2018267, additional information on chronology, climate controls on annual band thicknesses, model/data comparison, and the links between climate variability and hominin evolution, is available online at www.geosociety.org/pubs/ft2018.htm, or on request from editing@geosociety.org. The annual band thickness datasets are archived at the British Geological Survey (BGS) National Geoscience Data Centre (NGDC): http://dx.doi.org/10.5285/59750e40-ac38-4321-ad74-f001a03ebd16.
Gold Open Access: This paper is published under the terms of the CC-BY license.