The drivers of late Quaternary climate variability in eastern South Africa 1 2

1 2 Charlotte Miller1*, Jemma Finch2, Trevor Hill2, Francien Peterse3, Marc Humphries4, Matthias Zabel1, 3 Enno Schefuß1 4 5 1MARUM Center for Marine Environmental Sciences, University of Bremen, Bremen, Germany 6 2School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, 7 Pietermaritzburg, South Africa 8 3Department of Earth Sciences, Utrecht University, Netherlands 9 4Molecular Sciences Institute, School of Chemistry, University of the Witwatersrand, Johannesburg, 10 South Africa 11 12 *Correspondence email: lottiemiller2@gmail.com 13 14 Abstract 15 The scarcity of continuous, terrestrial, palaeoenvironmental records in eastern South Africa 16 leaves the evolution of late Quaternary climate and its driving mechanisms uncertain. Here we use a 17 ~7-m long core from Mfabeni peatland (KwaZulu-Natal, South Africa) to reconstruct climate variability 18 for the last 32 thousand years (ka BP). We infer past vegetation and hydrological variability using stable 19 carbon (δCwax) and hydrogen isotopes (δDwax) of plant-wax n-alkanes and use Paq to reconstruct water 20 table changes. Our results indicate that late Quaternary climate in eastern South Africa did not respond 21 directly to orbital forcing nor to changes in sea surface temperatures (SSTs) in the western Indian 22 Ocean. The arid conditions evidenced at Mfabeni during the Last Glacial Maximum (LGM) are a 23 consequence of both low SSTs and an equatorward displacement of the southern hemisphere 24 westerlies due to increased Antarctic sea ice extent. The increased humidity at Mfabeni between 19– 25 14 ka BP likely resulted from decreased Antarctic sea ice which led to a southward retreat of the 26 westerlies and increased the influence of the moisture-bearing tropical easterlies. Between 14–5 ka 27 BP, when the westerlies were in their southernmost position, local insolation became the dominant 28 control, leading to stronger atmospheric convection and an enhanced tropical easterly monsoon. 29 Generally drier conditions persisted during the past c. 5 kyrs, but were overlain by high amplitude, 30 millennial-scale environmental variability, probably resulting from an equatorward return of the 31 southern hemisphere westerlies and heightened ENSO activity. Our findings stress the influence of the 32 southern hemisphere westerlies in driving climatological and environmental changes in eastern South 33 Africa. 34 35


Introduction
Last glacial (c.21 ka BP) to present-day changes in vegetation, precipitation and temperature in eastern South Africa are poorly constrained.Whether eastern South Africa was characterized by aridity or increased humidity during the last glacial period remains unclear.Proxy data show spatial complexity (e.g.Baker et al., 2016;Chase et al., 2017;Chevalier and Chase, 2015& 2016, Dupont et al., 2011;Schefuß et al., 2011;Scott et al., 2012;Scott, 2016;Schmidt et al., 2014;Simon et al., 2015), and modelled LGM precipitation for the region are highly variable and often do not even agree on the sign of precipitation change.For example, the PMIP3 model ensemble mean suggests increased last glacial precipitation in the east of South Africa with dry conditions towards the north and south (compared to the present day; Braconnot et al., 2007).Conversely, the NCAR CCSM3 model indicates drier than present conditions in the centre of South Africa and along the eastern coast (Otto-Bliesner et al., 2006).These contrasting simulations for the last glacial period highlight the difficulty in simulating past precipitation in South Africa with a lack of proper understanding of relevant processes (Stone, 2014).
The mechanisms driving Quaternary climate variability in South Africa are complex and spatially heterogeneous.For example, hydroclimate may be paced by austral summer insolation fluctuations, resulting from changes in the Earth's orbital precession on 23-19 ka timescales.Strong summer insolation (during precession maxima) causes stronger atmospheric convection and an increase in the land/ocean temperature contrast, which results in higher moisture transport by the tropical easterlies and higher precipitation in eastern South Africa (e.g.Simon et al., 2015).Climate may also be influenced by high-latitude forcing related to changes in the Earth's orbital obliquity and eccentricity on longer, i.e. glacial-interglacial timescales, which may result in latitudinal contraction and expansion of the climatic belts (e.g.Dupont, 2011).The model of Nicholson and Flohn (1980) suggests an equatorward displacement of the intertropical convergence zone (ITCZ; Fig. 1) during the last glacial period, although proxy data from South Africa provide no conclusive support for this scenario.In addition, during glacial periods, the Walker Circulation may have been weaker with its ascending limb further to the east, within the Indian Ocean.This possibly resulted in an eastward displacement of the coastal cloud band and thus a drier summer rainfall zone (SRZ) and a wetter winter rainfall zone (WRZ; Tyson, 1986).Furthermore, changes in the latitudinal position of the southern hemisphere westerlies (as a response to fluctuations in Antarctic sea ice extent) have been invoked to influence climate in South Africa (Chase and Meadows, 2007;Chevalier and Chase, 2015;Chase et al., 2017).The western South African region has received most focus regarding the southern hemisphere westerly influence (Grundling et al., 2015).Furthermore, ET is higher in the swamp forest than in the sedge and reed fen, therefore a change in vegetation composition at Mfabeni has the potential to impact ET rates.The depositional setting of the Mfabeni peatland provides a unique opportunity to reconstruct past eastern South African climate variability at centennial-scale resolution from the Late Pleistocene to the present day.

Methodological background
To reconstruct past vegetation and hydroclimate changes we use the distribution, and the carbon and hydrogen isotopic composition, of long chain n-alkanes derived from plant-waxes.

Distributions of plant-waxes
To obtain information on water table variations, we quantify the relative contribution of plant-waxes derived from submerged and floating macrophytes relative to that of emergent and terrestrial plants (Paq).Odd-numbered n-alkanes (C25-C35) are derived from the epicuticular wax coating of terrestrial higher plants (Eglinton and Hamilton, 1967).Conversely, aquatic plant-waxes (of submerged macrophyte origin) are dominated by mid-chain n-alkanes (typically C23 and C25; e.g.Baker et al., 2016;Ficken et al., 2002).Thus we quantify Paq using Equation 1 (Ficken et al., 2000).
Eq. 1 with Cx the amount of each homologue.
To assess n-alkane degradation we used the carbon preference index (CPI; Bray and Evans, 1961).The CPI reflects the molecular distribution of odd-to-even n-alkanes, within a certain carbon number range (here, n-C26 to n-C34; Equation 2).High CPI values indicate a higher contribution of odd-numbered nalkanes (relative to even), indicating the n-alkanes are derived from higher terrestrial plants.Low CPI values indicate either low contribution from terrestrial higher plants or high organic matter degradation (Eglinton and Hamilton, 1967).

Carbon and hydrogen isotopes of terrestrial plant-waxes
To reconstruct vegetation changes, we use the carbon isotopic composition of terrestrial plantwaxes ( 13 Cwax).On late Quaternary timescales the primary factor determining the amplitude of respectively.
The δD compositions of long-chain n-alkanes were measured using a Thermo Trace GC coupled via a pyrolysis reactor (operated at 1420 °C) to a Thermo Fisher MAT 253 isotope ratio mass spectrometer (GC/IR-MS).The GC column and temperature program was similar to the δ 13 C analysis.
The δD values were calibrated against external H2 reference gas and are reported in ‰ VSMOW.The H 3+ factor was monitored daily and fluctuated around 5.2 ppm nA -1 during analyses.An n-alkane standard of 16 externally calibrated alkanes was measured every sixth measurement.The long-term precision and accuracy of the external n-alkane standard was 2.7 and 2‰, respectively.Samples were analysed in duplicate when n-alkane concentrations were adequate for multiple runs.The internal standard (squalane, δD= -180‰; ±2), yielded an accuracy of 0.9‰ and a precision of 1.9‰ (n=36).For δD the average precision in replicates was 1‰ for both n-C29 and n-C31 alkanes (n=52).
Last glacial Mfabani Dwax values were corrected to account for the effect of changes in global ice volume (Collins et al., 2013;Schefuß et al., 2005).For this the benthic foraminifera-based oxygen isotope curve (Waelbroeck et al., 2002) was interpolated to each sample age and then converted to D values using the global meteoric water line (Craig, 1961). 5.

Results
This study focusses on the last 32 ka BP (c.590 cm).The average temporal resolution between the 62 samples analysed for  13 C and D is c. 500 years.From 590 cm (32 ka BP) to 70 cm (c. 2 ka BP) the core is very dark brown in colour containing peat with humus, fine detritus and silt.From 70 cm to core top, the sediments are similar in colour to the peat below and contain fibrous peat with humus and herbaceous fine detritus (Humphries et al., 2017).Between 457 and 358 cm (c.23-14 ka BP; comprising the LGM) mean grain sizes average at 110 μm, with smaller diameters averaging at 50 μm between 298 and core top (c.11 ka-present, Holocene; Fig. 4g).The lithology of core MF4-12 does not exactly match with that observed from core SL6 (Baker et al., 2014;2016;2017), although sandy peat is observed during the LGM at both locations.This result is not surprising, multiple cores taken in transects across the bog indicate peat heterogeneity (Grundling et al., 2013).
Long chain n-alkane CPI values are generally around 6 (ranging from 2-13), indicating good nalkane preservation.The two samples with CPI values of 2, potentially containing more degraded nalkanes, are highlighted in red (Fig. 4b & c; Fig. 5b & c; Fig. 6).However, the in-or exclusion of these samples does not affect the observed pattern of changes and we thus consider the record to be suitable for palaeoclimate reconstruction.The samples contain n-alkanes with carbon chain lengths ranging from C17-C35, with C29 and C31 having the highest abundance.The high abundances of C29 and C31 enabled reliable isotopic analyses.The relationship between the δD and δ 13 C of the C29 and C31 nalkanes is strong, with R 2 values of 0.8 and 0.9, respectively.Consequently, for the δ 13 Cwax and δDwax, we use the amount-weighted mean of the C29 and C31 n-alkanes.
Between 14 and 5 ka BP,  13 Cwax values are relatively stable and average at -28‰ (Fig. 4b).Dwax values become gradually lower during this period reaching -173‰ at 7.5 ka BP.At 5 ka BP, Dwax values shift towards more positive values by 16‰ (Fig. 4c).Relatively high Paq values occur between 14-5 ka BP (Fig. 4f).After c. 5 ka BP several high amplitude millennial-scale fluctuations in both  13 Cwax and Dwax values are evident.These fluctuations interrupt a trend where the isotope values of both  13 Cwax and Dwax gradually increase towards present day.A pronounced shift to higher  13 Cwax and Dwax values occurs at 2.8 ka BP.From c. 900 yr BP,  13 Cwax and Dwax values become higher reaching core top values Good preservation of n-alkanes in Mfabeni peat was also observed in nearby core SL6, but this was based on a CPI calculated using n-C21-31 (Baker et al., 2016).No relationship exists between the CPI and Paq (R 2 = 0.11), this suggests that CPI variations at the location of core MF4-12 are not related to changes in organic matter preservation due to water table level variations.
The main source of carbon for terrestrial higher plants (the source of the C29 and C31 n-alkanes) is atmospheric CO2, whereas aquatics also assimilate dissolved carbon, complicating the interpretation of their carbon isotope signal.We thus focus solely on C29 and C31 n-alkanes that are predominantly derived from terrestrial plants (Eglinton and Hamilton, 1967).The majority of the samples (67 %) have dominant n-alkane chain lengths of C29 and C31.For the remaining 33 % of the samples, concentrated between 6 and 1.1 ka BP, the dominant chain length switched to n-C25, indicating a higher n-alkane input from submerged macrophytes (Ficken et al., 2000).The n-C25 are unlikely to be sourced from mosses, as mosses are rare in subtropical peatland environments (Baker et al., 2016).Instead, the C25 is likely mainly derived from aquatic plants, which produce mid-chain n-alkanes as dominant homologues (C20-C25; Ficken et al., 2000).This increase of n-alkanes sourced from aquatic plants c. 6-1.1 ka BP is unlikely to have had any impact on the isotopic composition of the long-chain n-alkanes (C29 and C31) as these are minor components in aquatic plants (e.g.Aichner et al., 2010).Therefore, we interpret the  13 Cwax as changes in the C3/C4 ratio of terrestrial higher plants.
C4 grasses are dominant within the SRZ, with C3 grasses more prevalent in the WRZ at higher altitudes (Vogel et al., 1978).Previous palynological studies indicate that the dominant components of the pollen assemblage at Mfabeni are Poaceae and Cyperaceae (Finch and Hill, 2008).Although Cyperaceae species can be either C3 or C4, most Cyperaceae in eastern South Africa (67 %) are of the C4-type (Stock et al., 2004) suggests that the sea level effect was minor.Indeed, the organic geochemical proxies agree with palynological data indicating regional grassland dominance (high Poaceae, Cyperaceae and Asteraceae) with low amounts of arboreal taxa (Fig. 6; Finch and Hill, 2008).Regional aridity and/or stronger wind strength are also indicated by increased grain size of the lithogenic sediment fraction at Mfabeni (Fig. 4g; Humphries et al., 2017).The dry conditions at Mfabeni during the LGM appear to be part of a wider eastern South African pattern, since they are consistent with regional precipitation and the aridity stacks (Fig. 4d & e).
The shift to more negative  13 Cwax values following the LGM, at c. 19 ka BP, indicating that the vegetation at Mfabeni changed to more C3-type plants (Fig. 4b), is also evident in Mfabeni core SL6 (Fig. 4h; Baker et al., 2017).This change is thus representative of C3/C4 changes across the peat bog.
However, the palynological record indicates no shift towards arboreal taxa at this time but instead a continuation of grasslands (Fig. 6; Finch and Hill, 2008).Baker et al. (2017) suggested that this carbon isotope shift resulted from the increase in temperature across the glacial-interglacial transition.
However, higher growing-season temperatures would favour C4 over C3 grasses (Ehleringer, 1997) and we therefore suggest that the carbon isotope decrease represents a shift from C4 grasses to C3 grasses ka BP corresponds well with a decrease in aeolian dust (Humphries et al., 2017) and the  13 Cwax record from Mfabeni core SL6 (Baker et al., 2017;Fig. 4b & 4h).There are, however, some minor differences between the two  13 Cwax records.We attribute these to small-scale variations in vegetation across the peatbog, the lower sampling resolution of core SL6 and to dating uncertainties in both records.The shift to lower  13 Cwax values at 19 ka BP occurs at the same time as a rise in the water table as documented by an increase in Paq values (Fig. 4f).The gradual shift to lower Dwax values around 19 ka BP occurs during decreasing local summer insolation, suggesting that this moisture shift was unlikely to be a result of increased precipitation, but more likely resulting from lower ET rates due to decreasing wind strength.
At the same time, gradually decreasing Dwax values indicate increasing humidity.Pollen data from Mfabeni provide evidence for an expansion of arboreal type vegetation at c. 12 ka BP (Fig. 6; Finch and Hill, 2008).The pollen data thus suggest the establishment of swamp forest vegetation during the early Holocene, indicative of a moist climate (Fig. 6).Mfabeni aeolian sediment grain size are low and stable throughout this period, also suggesting a moist climate and low wind strength (Fig. 4g).The moist climate likely resulted in vegetated dunes, reducing the amount of material available for aeolian transport.The relatively high Paq values between 14-5 ka BP indicate a high and stable water table at this time (Fig. 4f).Elevated total organic carbon percentages within Mfabeni core SL6 during the Holocene, also suggest increased water levels (Baker et al., 2017).
Between c. 5-0 ka BP several high-amplitude millennial-scale C3/C4 vegetation changes are evident superimposed on an overall shift from predominantly C3 to more C4-type vegetation towards the present-day (Fig. 5b).This variability contrasts with the more gradual C4/C3 vegetation transition from the Glacial to Holocene.The  13 Cwax values from Mfabeni core SL6 between c. 6-1 ka BP also indicate a period of predominantly C4-type vegetation, implying arid conditions during this time (Baker et al., 2017;Fig. 4h).A similar pattern of a long-term trend with superimposed short-term variability is visible in the in Dwax record.The general enrichment in D reflects gradual drying, punctuated by millennial-scale pulses of aridity, with the most pronounced arid event at c. 2.8 ka BP (Fig. 5c).
Counterintuitively, the high abundance of n-C25 alkanes and high but variable Paq values between c. 5-0 ka BP indicate a generally high water table, interrupted by brief periods of a lower water table (Fig. 5d).After 2.3 ka BP, both  13 Cwax and Dwax values become higher and Paq values lower (Fig. 5b-d).This suggests increased C4-type vegetation cover, decreased summer precipitation and/or higher ET amount and low water table levels.The increased variability between 5-0 ka BP could be an artefact of the high temporal resolution of our record during this interval (~220 vs ~700 years per sample for the remainder of the record).Nevertheless, other data from the region (e.g.Baker et al., 2017, Humphries et al., 2017;2016, Finch and Hill, 2008, Neumann et al., 2010)  instability and pulses of arid climatic conditions during the last c. 5 ka BP, suggesting that the observed variability is real (Fig. 5e).The long-term drying trend is unlikely to be caused by decreased summer precipitation because local summer insolation and Mozambique Channel SSTs are high (Fig. 5a & Fig. 4j).Instead, the general drying trend is more likely a result of heightened ET during the late Holocene.

Climate driving mechanisms
Modern observations suggest that warm SSTs within the Mozambique Channel and Agulhas Current induce increased evaporation (e.g.Walker, 1990), resulting in increased onshore airflow and advection of moist air and higher rainfall in the SRZ (Tyson, 1999).Variations in local SST are thus thought to be an important driver of hydroclimate in eastern South Africa.This mechanism may also play a key role on longer time scales.Indeed, Chevalier and Chase (2015) mention this hypothesis, invoking SSTs as the dominant driver of precipitation variability during the LGM, and Baker et al. (2017) argue for the existence of a short arid event at 7.1 ka BP that corresponds to a decrease in Mozambique Channel SSTs.However, Mfabeni vegetation and hydrology reconstructions over the last 32 ka BP do not show a close relationship with changes in southwest Indian Ocean SSTs (Fig. 4j, Wang et al., 2013).
This suggests that SST variability is unlikely to be the sole driver of the changes in hydroclimate at Mfabeni over the 32 ka.Thus, we suggest an additional role, namely the southern hemisphere westerlies.
We attribute the arid climate and the associated expansion of drought tolerant C4 plants and a low water table at Mfabeni during the LGM, in part, to a northward displacement of the southern hemisphere westerly winds, shifting the hydroclimate to a more evaporative regime, where ET exceeds precipitation.In addition, lower SSTs (Fig. 4j) in the Mozambique Channel at this time likely reduced moisture availability.It is possible that the combination of a northward displacement of the southern hemisphere westerlies and lower SSTs shifted the fine balance between precipitation and ET at Mfabeni towards higher ET rates during the LGM.
Numerous palaeoenvironmental studies (e.g.Lamy et al., 2001;Lamy et al., 2010;Stuut and Lamy, 2004;Chase et al., 2017) and climate model simulations (e.g.Cockcroft et al., 1987;Rojas et al., 2009;Toggweiler et al., 2006), indicate an intensification and equatorward migration of the southern hemisphere westerlies in response to the increased extent of Antarctic sea ice during the LGM.Such changes may have expanded the limit of the WRZ in South Africa northward, to around 25°S in the west and 30°S in the east (Cockcroft et al., 1987).This would have put Mfabeni (at 28°S) within the range of the southern westerlies.Regions on the east coast, such as Mfabeni, then experienced stronger winter winds, causing heightened ET (Humphries et al., 2017).With more northerly westerlies, the duration of the dry season at Mfabeni may also have been extended diminishing the influence of the easterlies.This shortened the rain season and heightened ET rates.The northward shift in the westerlies during the LGM is also visible in records from the present WRZ (e.g.Chase et al., 2017), which show increased winter rainfall and moist conditions (Fig. 4i).
Although during the LGM the southern hemisphere westerlies were in a more northerly position and had the potential to provide rainfall, we do not see any evidence that the source of precipitation changed.Today the moisture at Mfabeni is mainly provided by the tropical easterlies (Kruger et al., 2010;Tyson, 1999), and thus we see enriched D values on the east coast and more depleted values further inland (Dprecip data from GNIP (IAEA, 2018) and groundwater D data from West et al., 2014).Precipitation sourced from the southern hemisphere westerlies would be strongly depleted in D because of the large continental transport distance from the Atlantic Ocean.However, there is no evidence of lower Dwax values during the LGM at Mfabeni, indicating the westerlies did not bring moisture to Mfabeni during the LGM but that increased wind strength led to increased ET and a more arid climate.
We suggest that the shift to more humid conditions at c. 19 ka BP was related to the retreat of the southern hemisphere westerlies from this area as Antarctic sea ice began to retreat poleward at this time (Fig. 4k), leading to less ET and allowing an increased influence of the moist tropical easterlies.This shift was unlikely driven by a change in local summer insolation (i.e.Chevalier and Chase, 2015) because insolation was decreasing at this time, which would have caused reduced, instead of enhanced, summer precipitation.We suggest that the abrupt shift to more C3 vegetation was a non-linear response to increasing moisture availability in the peatbog (Fig. 4c).Precipitation amount may have reached a critical threshold at c. 19 ka BP for the establishment of C3 type vegetation, resulting in the observed abrupt vegetation shift (Fig. 4b).We propose that at c. 19 ka BP, the position of the southern westerlies had a greater climatic influence than the local insolation forcing.Further south, within the present WRZ, the retreat of the westerlies at this time resulted in a shift to more arid conditions (Fig. 4i; Chase et al., 2017).The timing of increased humidity at Mfabeni at c. 19 ka BP corresponds well to a reduction in Antarctic sea ice extent, which is thought to be the main driver of the latitudinal position of the westerlies (Fig. 4k; Fischer et al., 2007).
Between 14-5 ka, low Antarctic sea ice (Fig. 4k & 5g), resulted in a more poleward position of the westerlies.The diminished effect of the westerlies in eastern South Africa at this time permitted the tropical easterlies, and thus local summer insolation, to dominate the climatic regime at Mfabeni.Indeed, increasing humidity at Mfabeni at this time, corresponds with increasing southern hemisphere summer insolation (Fig. 4a).The importance of insolation for South African climate variability during the late Quaternary has been suggested before (e.g.Partridge et al., 1997;Simon et al., 2015;Schefuß et al., 2011;Chevalier and Chase, 2015).However, we suggest that direct local insolation forcing is only dominant when the westerlies are located far south.Meadows and Baxter, 2001), Cecilia Cave (Fig. 1; Baxter, 1989) and Eilandvlei (Wündsch et al., 2018) document increased moisture supply to the WRZ, implying a recurring more northerly location of the westerly storm tracks at this time.Chevalier and Chase et al. (2015) propose that increased precipitation in the WRZ during the late Holocene was due to both the warmer interglacial climate and the northward expansion of the westerly storm tracks.Although no indication for an increase in sea ice is evident from EPICA salt concentration data (Fig. 4k), diatom data (Fragilariopsis curta and F. cylindrus) from PS2090/ODP1094 in the southern South Atlantic document an increase in sea ice during the late Holocene (Fig. 5g), which may have pushed the southern westerlies equatorward.In addition, climate modelling results imply a northward shift of the southern hemisphere westerlies at this time (Hudson and Hewitson, 2001).Consequently, in a comparable way to the LGM, the increased sea ice during the late Holocene (Fig. 5g), may have displaced the westerlies equatorward, increasing winter wind strength and the length of the dry season at Mfabeni, leading to a decreased influence of the moisture bearing tropical easterlies (Mejía et al., 2014;Toggweiler et al., 2006;Williams and Bryan, 2006).Furthermore, although the westerlies may have had a more northerly position during this time, simultaneous high local summer insolation and warm SSTs (causing strong convective rainfall during summer; Fig. 5a) may have been the cause of the relatively high water table (Fig. 5d) and transitory peaks in precipitation and C3-type vegetation expansion (Fig. 5b and c).Interestingly, the hydrological variability at Mfabeni (Fig. 5c) during the last c. 5 ka BP, is not present in the central and eastern South African precipitation stack (Fig. 4d).We attribute this to the highly sensitive balance between ET and precipitation at Mfabeni (Grundling et al., 2015), and the fact that the precipitation stack smoothes local hydrological variability.
It is possible that anthropogenic influences also played a role in shaping the environment at Mfabeni during, at least, during the late Holocene.However, unequivocal agricultural and exotic pollen indicators are absent from the pollen record and although pollen data indicate that deforestation occurred during the late Holocene, it is unclear whether this was related to human influence or regional climatic change (Finch and Hill, 2008).The deforestation could have affected the water table and increased the relative amount of C4-type vegetation.The appearance of Morella and Acacia in the late Holocene may indicate the development of open vegetation or secondary forest due to fire disturbance (Finch and Hill, 2008).Human activities or climate change may be responsible for changes in fire regime.With no palaeo-charcoal data available for Mfabeni, no direct evidence for increased fire activity during the late Holocene exists.In addition, the palaeoenvironmental evidence available suggests that the arid conditions during the late Holocene were regional in nature (Scott, 1999;2003;Humphries et al., 2016, Neumann et al., 2010).Thus, any human activity was unlikely the primary cause of the late Holocene regional aridity and the large magnitude of environmental changes observed at Mfabeni.
Today ENSO activity is one of the most important driving mechanisms for inter-annual climatic variability in South Africa (Tudhope et al., 2001).Southern Africa's seasonal rainfall is linked to ENSO, with dry (wet) conditions associated with El Niño (La Niña) events (Archer et al., 2017;Mason and Jury, 1997).According to the model of Tyson (1986), during El Niño events, high Indian Ocean SSTs produce a weaker landward pressure gradient over the southeast African coast, diminishing low-level confluence over the continent.This results in a longitudinal migration of the ascending limb of the Walker circulation eastwards, leading to lower rainfall on the continent (i.e at Mfabeni) but higher rainfall towards the east (i.e Madagascar; Mason and Jury, 1997).Furthermore, during El Niño events, a northward shift of the westerlies may occur, which could increase rainfall over western South Africa but lead to aridity in the east (i.e. at Mfabeni ;Lindesay, 1988).Although after c. 5 ka BP the Mfabeni sampling resolution is higher, we document some evidence for heightened climatic variability (in comparison with the rest of the record) and generally drier conditions at Mfabeni over the last c. 5 ka BP.We speculate that this variability may have been the result of amplified ENSO activity (e.g.Humphries et al., 2017).Palaeoenvironmental studies in the Pacific Basin and South America indicate that during the early Holocene El Niño events were smaller and occurred less frequently, with a shift to stronger ENSO activity after c. 5 ka BP (Fig. 5h, Moy et al., 2002;Huffman, 2010;Rodbell et al., 1999;Sandweiss et al., 1996;Tudhope et al., 2001).
We therefore invoke a combination of both the northerly-displaced southern hemisphere westerlies and the impact of high ENSO variability as climatic drivers during the last c. 5 ka BP.The high-amplitude, millennial-scale vegetation and hydrological instability documented at Mfabeni during the last c. 5 ka BP contrasts with the relatively stable conditions during the LGM and early Holocene.
This increased environmental variability during the late Holocene could be the result of increased and strongly fluctuating sea ice extent during this period, overlain by strong ENSO activity (Moy et al., 2002).

Conclusions
Compound specific carbon and hydrogen isotope data and n-alkane distributions (Paq) from  Red/orange = summer rainfall zone (SRZ).Green = year-round rainfall zone (YRZ).Blue is winter rainfall zone (WRZ).The white arrows are atmospheric circulation and the blue arrows are oceanic circulation.
Map is courtesy of B. Gijsbertsen, UKZN Cartography Unit.Central and eastern South African regional precipitation stack (red line; Chevalier and Chase, 2015).e) Southern African regional aridity stack (Chevalier and Chase, 2016).f) Paq at Mfabeni, indicating the amount of aquatic vs. terrestrial n-alkanes (high/low water table).g) Mean grain size data of the lithogenic sediment fraction from Mfabeni, with increased grain size indicating increased wind strength (Humphries et al., 2017).h) Mfabeni core SL6 stable carbon isotope composition (weighted mean) of C29-C31 n-alkanes (Baker et al., 2017).i) Combined nitrogen isotope data from Seweweekspoort rock hyrax middens, reflecting changes in humidity (Chase et al., 2017).j) U K′ 37 derived SSTs from core GIK16160-3 in the Mozambique Channel (Wang et al., 2013).k) Sea salt sodium concentrations from the EPICA DML ice core in Antarctica, reflecting changes in sea ice coverage (Fischer et al., 2007).).Blue dashed lines highlight trends.e) Mfabeni calcium/scandium ratio, indicating changes in water table (Humphries et al., 2017).f) Bulk carbon isotope data from Verlorenvlei (Carr et al., 2015).g) Extent of Antarctic sea ice.Estimation is based on the abundance of

Clim.
Past Discuss., https://doi.org/10.5194/cp-2019-4Manuscript under review for journal Clim.Past Discussion started: 21 January 2019 c Author(s) 2019.CC BY 4.0 License. of -21 and -128‰, respectively (Fig. 4b and c).Generally high, but variable and rapidly fluctuating Paq values are evident between c. 5-0 ka BP.Paq values decrease substantially after 1.3 ka BP from 0.6 to a core top value of c. 0 (Fig. 4f).proxy signals The relatively high CPI27-33 values indicate that the long-chain n-alkanes within the peat are derived from terrestrial higher plants and are relatively non-degraded.The long-chain n-alkanes are likely sourced directly from the local vegetation surrounding the coring location.It is possible that during times of stronger wind strength (i.e. during the LGM; Humphries et al., 2017) aeolian transport resulted in a higher biomarker contribution from more distal sources (i.e. the surrounding dune vegetation).
. The C4 vegetation at Mfabeni is thus mostly Poaceae or Cyperaceae from the sedge and reed fen.The C3 vegetation at Mfabeni is comprised of arboreal taxa from the swamp forest (e.g.Myrtaceae and Ficus) and locally distributed Podocarpus (Finch and Hill, 2008; Venter, Clim.Past Discuss., https://doi.org/10.5194/cp-2019-4Manuscript under review for journal Clim.Past Discussion started: 21 January 2019 c Author(s) 2019.CC BY 4.0 License.
Mfabeni peatbog are used to reconstruct climatic conditions, over the last 32 ka BP, in eastern South Africa.The LGM at Mfabeni was characterized by a high contribution of C4 grasses, high ET rates and a low water table.During the LGM increased Antarctic sea ice extent led to an equatorward displacement of the southern hemisphere westerly winds, which extended the length and increased the intensity of the dry season at Mfabeni.Between c. 19-5 ka BP an expansion of C3-type vegetation occurred, with more rainfall and a higher water table at Mfabeni.At c. 19 ka BP, Antarctic sea ice 16 Clim.Past Discuss., https://doi.org/10.5194/cp-2019-4Manuscript under review for journal Clim.Past Discussion started: 21 January 2019 c Author(s) 2019.CC BY 4.0 License.decreased, which resulted in a southward retreat of the southern hemisphere westerlies.This southward retreat of the westerlies after c. 19 ka BP and an increase in local summer insolation after c. 12 ka BP resulted in more precipitation and increased wet season length at Mfabeni.When the westerlies were in their southernmost position (c.14-5 ka BP), local insolation became the dominant control on Mfabeni climate, leading to stronger convection and enhanced monsoonal precipitation from the tropical easterlies.The late Holocene (c.<5 ka BP) was characterized by increased environmental instability and increasingly arid conditions.We attribute these trends to concurring high local summer insolation and the recurring influence of the southern westerlies and/or heighted ENSO activity.The Mfabeni record indicates that climate and environmental variability in eastern South Africa over the last 32 ka BP is driven by a combination of i) enhanced/reduced moisture transport by the tropical easterlies, driven by variations in southern hemisphere summer insolation, and ii) latitudinal displacements of the southern hemisphere westerlies.With the expansion and retreat of Antarctic sea ice responsible for the displacement of the westerlies, we invoke high-latitude climate forcing as an important driver of climate in eastern South Africa.

Figure captions Figure 1
Figure captions Figure 1.a) Rainfall seasonality map for southern Africa showing the major oceanic and atmospheric currents, the position of the intertropical convergence zone (ITCZ) and the Congo Air Boundary (CAB).

Figure 3 .
Figure 3. Depth-age model of core MF4-12 produced using Bacon, based on 24 14 C AMS dates.Blue symbols are AMS dates and grey shading indicates 95% confidence interval on the mean age (red line).

Figure 4 .
Figure 4. Climate and environmental change at Mfabeni compared with regional records and orbital insolation.a) December-January-February (DJF) insolation for 28°S (blue line; Laskar, 2011).b) Stable carbon isotope composition (weighted mean) of C29-C31 n-alkanes from Mfabeni, reflecting changes in C3/C4 vegetation type.c) Hydrogen isotope composition (weighted mean) of C29-C31 n-alkanes from Mfabeni, reflecting changes in summer precipitation and ET amount.Red is the Dwax corrected for ice volume changes.Error bars on isotope data reflect analytical uncertainty of duplicate analyses.d) The two Mfabeni samples with CPI values of c. 2 are highlighted in red (4b & c).SHW = southern hemisphere westerlies.Blue shading = Mfabeni wet, orange = Mfabeni arid.

Figure 5 .
Figure 5.Comparison of Mfabeni data with other records of environmental variability over the last 15 ka BP. a) DJF insolation for 28°S (black line; Laskar, 2011).b) Carbon isotope composition (weighted mean) of C29-C31 n-alkanes from Mfabeni, reflecting changes in C3/C4 vegetation type.c) Hydrogen isotope composition (weighted mean) of C29-C31 n-alkanes from Mfabeni, reflecting changes in summer precipitation and ET amount.d) Paq at Mfabeni, indicating the amount of aquatic vs. terrestrial n-alkanes (high/low water table).Blue dashed lines highlight trends.e) Mfabeni calcium/scandium

Fragilariopsis curta and
Fragilariopsis cylindrus at site PS2090/ODP1094 (SW of Cape Town; Bianchi and Gersonde, 2004).h) Red colour intensity time-series from Laguna Pallcacocha.High values are light coloured inorganic clastic laminae, which were deposited during ENSO-driven episodes (Moy et al., 2002).The Mfabeni sample with a CPI value of c. 2 is highlighted in red (5b & c).

Figure 6 .
Figure 6.Summary figure highlighting the main climate phases and driving mechanisms at Mfabeni.From left: hydrogen isotope composition (weighted mean) of C29-C31 n-alkanes from Mfabeni.Red is the Dwax corrected for ice volume changes.Stable carbon isotopic composition (weighted mean) of C29-C31 n-alkanes from Mfabeni.The two Mfabeni samples with CPI values of c. 2 are highlighted in red.Summary of the palynological data from Finch and Hill (2008) and possible climate driving mechanisms at Mfabeni during the last 32 ka BP.