Temperature change in subtropical southeastern Africa during the past 790,000 yr

Hydrological conditions have been shown to be highly variable during the late Pleistocene in tropical and subtropical southeastern Africa (e.g., Schefuß et al., 2011; Johnson et al., 2016; Caley et al., 2018). In contrast, relatively little is known about past temperature variability. While some studies have shown that regional temperatures at the end of the last glacial period were 3.5–5.2 °C colder than present day (Kulongoski and Hilton, 2004; Chevalier and Chase, 2015; Johnson et al., 2016), the drivers of temperature changes over time remain unknown. Yet, accounting for temperature is crucial to interpret paleohydrological proxy records in terms of rainfall changes and climate dynamics, as temperature can modulate the net effect of precipitation through evapotranspiration (Chevalier and Chase, 2016; Scheff et al., 2017). Because many proxies used to infer past hydrological change (e.g., lake levels or δ¹³C from leaf-wax n-alkanes in plants) principally reflect environmental water availability and not rainfall amount per se, the absence of long and detailed temperature records restricts our ability to fully characterize past precipitation amount as a distinct variable.

Unfortunately, few proxies have the capacity to record temperature variability directly (Nicholson et al., 2013), and the trends, amplitude, and drivers of past temperature changes in southern Africa remain elusive. While some fossil pollen taxa are temperature-sensitive, methodological constraints generally limit interpretations to qualitative assessments (e.g., Dupont et al., 2011; Valsecchi et al., 2013). The development of the Bayesian climate reconstruction method CREST (Climate REconstruction SofTware) has enabled the reconstruction of quantified records of past climate change from fossil pollen data (Chevalier et al., 2014), with recent results highlighting a strong connection between terrestrial temperatures and sea-surface temperatures (SSTs) from the Mozambique Channel (Truc et al., 2013; Chevalier and Chase, 2015). However, these reconstructions are too short (20 k.y. and 45 k.y., respectively) to determine responses to a more complete suite of Pleistocene conditions, particularly periods of high eccentricity, when variations in direct insolation were substantially greater than during the last 50 k.y. (Laskar et al., 2004).

Here, we used a pollen record from marine core MD96–2048 (south of the mouth of the Limpopo River; Dupont et al., 2019) as the basis for a quantified reconstruction of temperature variability for the Limpopo River catchment during the last 790 k.y. Pollen data show a strong response of vegetation to glacial-interglacial cycles, with glacial landscapes being dominated by Ericaceae and other montane shrubs, and interglacials dominated by increased forest development (Dupont et al., 2019). These large changes allowed us to determine temperature variability at what is today the southeastern edge of the African monsoon domain (Wang and Ding, 2008) and investigate the relationship with the patterns of climate change that have characterized the late Pleistocene.

The CREST method is a Bayesian approach that combines presence-only plant occurrence data with modern climatological data to estimate the conditional responses of pollen taxa to specific climate variables, such as mean annual temperature (MAT) (Chevalier et al., 2014; Chevalier, 2019). Defined as probability density functions (PDFs), the taxon-climate response curves are fitted in two steps to incorporate the taxonomic resolution of pollen taxa (most commonly identified at the genus or family level; Fig. 1). First, Gaussian PDFs are fitted for all the species that produce the same type of pollen grains from all the climate values corresponding to the presence records (Figs. 1A and 1B). The influence of the heterogeneously distributed modern climate space is removed by weighting each climate value by the inverse of its abundance in the study area (Bray et al., 2006). Species PDFs are then combined to create a higher-order PDF that represents the climate response of the pollen type as a unit (Fig. 1C).

Past MAT values are estimated from the weighted multiplication of the pollen PDFs. Here, the raw percentages of each individual taxon were normalized by the average of the percentages observed in the fossil pollen sequence when the taxon was present to account for taphonomic processes and different pollen production rates. A value higher (lower) than one suggests that the climate at the time of deposition was more (less) favorable than the average climate in which the taxon has been observed during the studied period (Chevalier et al., 2014). The multiplication of the PDFs results in a posterior distribution of probabilities, from which climate estimates and uncertainties are derived (Fig. 1D).

Located south of the mouth of the Limpopo River (34.0167°E, 26.1667°S, 660 m water depth), marine core MD96–2048 consists of sediments from the catchments of the Limpopo, Maputo, and Incomati Rivers (>460,000 km²; Fig. 2; Dupont et al., 2011, 2019). The record spans the past 790 k.y., recording over eight complete glacial-interglacial cycles since marine isotope stage (MIS) 20. The resolution of the record varies with a mean resolution of ∼3.0 k.y. over the past 342 k.y., and ∼6.7 k.y. for the older section. As such, this record is unlikely to have recorded cyclicities with a period shorter than at least 15–20 k.y. Today, the MAT values encompassed by the catchments are fairly homogeneous (mean modern temperature ∼20.9 °C, σ = 2.5 °C; Fig. 2A). The large elevational range across the catchment (up to 2000 m in the Drakensberg region in the south, 500–1000 m on the African plateau to the west, and 0–200 m in the east) fosters a range vegetation types (four distinct biomes; Fig. 2B). We refer the reader to Dupont et al. (2011) for a more detailed description of the catchment.

To enable the reconstruction of potentially large vegetation and climatic changes, the calibration data set was designed to include a much broader area than the current catchment size (Fig. 2C), covering a temperature gradient of more than 24 °C and encompassing seven different biomes. Central Africa was not included because the distributions of plant presence records are too fragmentary in the area to estimate reliable PDFs (Chevalier, 2019). Of the 231 terrestrial and aquatic pollen taxa that have been observed, we used the climate response of 166 terrestrial taxa to reconstruct MAT (see Appendix S1 in the Supplemental Material¹).

Reconstructing climate from marine pollen records can be problematic, because marine sediments can integrate large source areas and may be subject to a range of transport vectors that can vary through time. In the case of MD96–2048, the region is dominated by easterly onshore winds year-round (Tyson and Preston-White, 2000), which is not favorable for eolian pollen transport to the coring location, and studies indicate that most of the terrestrial sediment flux is fluvial in origin (Dupont et al., 2011; Castañeda et al., 2016). The MAT value reconstructed for the most recent pollen sample (19.8 °C, σ = 2.3 °C; Fig. 3) is within the modern value calculated for the catchment as a whole (20.9 °C, σ = 2.5 °C), highlighting the fact that the pollen and associated MAT reconstructions reflect the catchment’s average temperature.

The MD96–2048 MAT reconstruction is strikingly similar to glacial-interglacial signatures of global temperature (Pearson’s correlation index ρ = 0.76 [last 342 k.y.]/0.72 [entire record]; Jouzel et al., 2007), atmospheric CO₂ concentration (ρ = 0.77/0.74; Lüthi et al., 2008; Bereiter et al., 2015), and ice volume (ρ = –0.79/−0.66; Lisiecki and Raymo, 2005) throughout the past 790 k.y. (Fig. 4; Appendix S2). These high correlations suggest that temperature variability in southeast Africa is more tightly linked to global temperature changes via glacial and CO₂ feedbacks than to local summer insolation (cf. Herbert et al., 2010).

Maximum interglacial temperatures averaged 21.0 ± 0.7 °C in the last 342 k.y. (20.6 ± 1.1 °C for the entire record), while minimum glacial temperatures averaged 17.0 ± 0.3 °C (16.9 ± 0.2 °C for the entire record). While the temporal resolution of the record masks part of the effective climate variability, an amplitude of change of ∼4 °C between full glacials and interglacials in the southeast African lowlands is consistent with reconstructions from the nearby Mozambique Channel (3–4 °C; Sonzogni et al., 1998; Caley et al., 2011) and South Africa (3.5 °C; Chevalier and Chase, 2015), and it has been shown to be strong enough to counterbalance a deficit of ∼120 mm of rain and maintain similar humidity levels (Chevalier and Chase, 2016).

Previous regional temperature reconstructions from the region also indicated a strong glacial-interglacial link between continental temperature variability and Indian Ocean SST variability (Chevalier and Chase, 2015; Castañeda et al., 2016). Our findings confirm this coeval variability, as highlighted by a correlation with a SST record from the same marine core (ρ = 0.71 [last 342 k.y.]/0.72 [entire record]; Fig. 4; Caley et al., 2018). The two records differ principally by the decline in glacial SSTs around 350 k.y. ago that is not evident in the pollen-based reconstruction. This shift in the glacial temperatures is also absent from a temperature-index record (ratio of leaf wax long-chain n-alkanes, C₃₁/[C₂₉ + C₃₁]) derived from the same core by Castañeda et al. (2016) (ρ = 0.67/0.71, after detrending for a long-term aridification trend). This SST shift is only observed in records derived from foraminifera, while other SST proxies do not record a similar trend (Caley et al., 2011). This difference may reflect different taphonomy or, more likely, different sensitivities (e.g., seasonal vs. annual signals) of the marine and terrestrial microfossils and proxies obtained from the sediment core.

The only other long quantified African terrestrial temperature record spanning several glacial-interglacial cycles comes from Lake Malawi (Fig. 2; Johnson et al., 2016). While the Malawi temperature record (partially tuned to the global benthic δ¹⁸O stack; Lisiecki and Raymo, 2005) exhibits clear glacial-interglacial signals, correlation with our reconstruction (ρ = 0.51/0.30) and other temperature records is generally more modest (Appendix S2), possibly highlighting the fact that different drivers govern temperature in these two regions of Africa (mainly tropical influences at Lake Malawi vs. subtropical and extratropical influences across our study area). It is also possible that Lake Malawi temperature estimates are partially influenced by the extreme lake-level variations that have been reconstructed over the past 800 k.y. (>10 lake-level reductions of 500 m or more; Lyons et al., 2015). These were corrected for, but this variability may have also influenced the archaea producing the tetraether index of lipids with 86 carbons (TEX₈₆) data used to derive temperature estimates, thus changing the representativity of the record over time (Johnson et al., 2016).

The MD96–2048 MAT reconstruction enabled us to analyze the ways in which pollen diversity varied under different temperature conditions and associated atmospheric CO₂ levels across glacial-interglacial cycles. Comparing measured pollen diversity (e.g., number of taxa and Margalef’s index [species diversity index; Margalef, 1957] in Figure 4) with temperature, we could not isolate systematic links (Appendix S3). While acknowledging that differences exist between pollen diversity and plant species diversity, especially when the record encompasses several biomes (Goring et al., 2013), this absence of correlation suggests that at the spatiotemporal resolution of this study, glacial and interglacial periods supported equally diverse floras. We did observe a significant pattern coherent with the ∼400 k.y. cycle of orbital eccentricity, with the low pollen diversity corresponding with high eccentricity (Appendix S3). A negative correlation between pollen diversity and the ∼100 k.y. eccentricity cycle also appears possible, particularly during high phases of the 400 k.y. cycle. While a better-resolved record is required before any clear conclusion can be drawn, we infer that the high-amplitude (low-amplitude) variability in seasonal insolation forcing under higher (lower) eccentricity may result in greater (lesser) levels of ecosystem disruption and displacement, and perhaps the establishment of more diverse vegetations under more stable climatic conditions.

Our reconstruction of MAT from fossil pollen data demonstrated the potential of CREST to reconstruct climate on long time scales and reveal past climate variability from low latitudes. Here, we have shown the existence of a strong link between southeast African and global temperatures that indicates that temperature variability in southeast Africa is primarily paced by glacial-interglacial cycles. Temperature changes of 4 °C between glacial and interglacial periods are likely to have had a significant impact on water availability in the region, and these findings reinforce the need to consider temperature as an important and independent driver of tropical climate while interpreting moisture-related proxy records.

Chevalier was supported by the Swiss National Science Foundation (SNF/FNS) ‘HORNET’ Project (#200021_169598). This research has been supported by the German Federal Ministry of Education and Research (BMBF) as a Research for Sustainability initiative (FONA; https://www.fona.de/en, last access: 25 September 2020) through the PalMod Phase II project (grant no. FKZ: 01LP1926D). Funding for this work was received from the European Research Council (ERC) under the European Union’s Seventh Framework Programme (FP7/2007–2013)/ERC Starting Grant “HYRAX,” grant agreement no. 258657. We gratefully acknowledge three reviewers for their constructive comments.