South-western Africa vegetation responses to atmospheric and oceanic changes during the last climatic cycle

Terrestrial and marine climatic tracers from marine core MD96-2098 collected in the southwestern African margin and spanning from 194 to 24 (thousand years before present) documented three pronounced expansions of Nama-Karoo and ﬁne-leaved savanna during the last interglacial (Marine Isotopic Stage 5 – MIS 5). Nama-Karoo 5 and ﬁne-leaved savanna expansions were linked to increased aridity during the three warmest substadials of MIS 5. Enhanced aridity potentially resulted from a combination of reduced Benguela Upwelling System (BUS), expanded subtropical high-pressure cells, and reduced austral-summer precipitation due to a northward shift of the In-tertropical Convergence Zone (ITCZ). Decreased austral-winter precipitation was likely 10 linked to a southern displacement of the westerlies. In contrast, during glacial isotopic stages MIS 6, 4 and 3, Fynbos expanded at the expense of Nama-Karoo and ﬁne-leaved savanna indicating a relative increase in precipitation probably concentrated during the austral winter months. Our record also suggested that warm-cold or cold-warm transitions between isotopic stages and substages were punctuated by short 15 increases in humidity. Increased aridity during MIS 5e, 5c and 5a warm substages coincided with minima in both precessional index and global ice volume. On the other hand, austral-winter precipitation increases were associated with precession maxima at the time of well-developed northern-hemisphere ice caps.

northward migration. These three climatic factors combined would result in a slight increase in humidity in northeastern South Africa during interglacials. On the other hand, increased aridity during the last interglacial in southern Africa has been suggested based on ratios of aeolian dust and fluvial mud in marine sediments (Stuut and Lamy, 2004). 15 Vegetation-based climate reconstructions for southern Africa have been less straight forward given the paucity of records (Dupont, 2011) and fragmentary nature of some terrestrial sequences (Scott et al., 2012;Meadows et al., 2010). On one hand, some records point to expansions of the Fynbos biome (Shi et al., 2001) and the winterrainfall zone during glacial periods (Chase and Meadows, 2007), and to a contracting and savanna biome reductions, instead of expansions, during the last interglacial precession minima. Whether the last interglacial was characterised by orbitally-driven increased aridity or increased precipitation may have significant implications for resource availability and climate in the region today, hence the importance of further investigating the glacial-interglacial climate and vegetation dynamics of southern Africa. 5 In this study we aim to disentangle the contrasting hypotheses of orbital-scale climate change in southern Africa by combining terrestrial and marine tracers from the marine sequence MD96-2098. We use pollen and charcoal as terrestrial tracers, and δ 18 O from benthic foraminifera as a marine tracer. Vegetation reconstructions from marine records have contributed to our understanding of ocean-land interactions in 10 many regions of the world, including the Iberian Peninsula (Sánchez Goñi et al., 2000), the eastern subtropical Pacific (Lyle et al., 2012), and the tropical Atlantic (González and Dupont, 2009). Studies from the African margin (e.g. Dupont, 2011;Dupont and Behling, 2006;Hooghiemstra et al., 1992;Leroy and Dupont, 1994) have demonstrated that pollen records from marine sequences are reliable and useful tools to reconstruct 15 changes in the regional vegetation of adjacent landmasses and the climate dynamics at orbital and suborbital timescales. In arid environments, marine sequences are particularly essential in providing continuous records of vegetation change at the regional scale. The pollen sequence from MD96-2098 presented here covers the period between Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by different seasonal precipitation (Rutherford, 1997). The Nama-Karoo is influenced primarily by austral summer precipitation, while the distribution of the Succulent-Karoo coincides with the austral-winter rainfall region (Chase and Meadows, 2007). To the northwest, the Nama-Karoo biome transitions into the Desert, where mean annual precipitation can be as low as 20 mm yr −1 (Jurgens et al., 1997). The Desert reaches 5 300 km inland and its extent is linked to the intensity of BUS (Lutjeharms and Meeuwis, 1987). High precipitation seasonality (i.e. difference between dry-season and rainy-season precipitation) and high austral-summer rainfall characterize the savanna. The savanna biome represents a mosaic that includes shrublands, dry forests, lightly-wooded grass-10 lands, and deciduous woodlands (Scholes, 1997). At the landscape scale however, the savanna can be subdivided into the fine-and broad-leaved savannas based on moisture conditions and soils (Scholes, 1997). The fine-leaved savanna (Fn-LSav) is found in dry and fertile environments (between 400 and 800 mm yr −1 ), and the broadleaved savanna (Bd-LSav) is found in nutrient-poor and moist environments (up to 15 1500 mm yr −1 ) (Scholes, 1997). Additionally, in the Fn-LSav fuel load and fire frequency are very low, while Bd-LSav has high fuel load and fire frequency (Scholes, 1997;Archibald et al., 2010). The Fn-LSav is found to the northeast of the Nama-Karoo biome ( Fig. 1), known as the Kalahari Highveld transition zone (Cowling and Hilton-Taylor, 2009). Due to the transitional character of the Fn-LSav, some of its outer 20 parts have been classified as grassland or Nama-Karoo (White, 1983). The composition of the Fn-LSav can be similar to that of the Nama-Karoo, with dominance of C4 grasses (Poaceae) and succulent plants, but it differs in having scattered trees ( Fig. 2b) (Cowling et al., 1994). The Bd-LSav is characterized by broad-leaved trees from the Caesalpinaceae and Combretaceae families and an understory dominated by grasses 25 (Scholes, 1997). The grassland biome is dominated by C4 grasses and non-grassy forbs as Anthospermum sp., Lycium sp., Solanum sp. and Pentzia sp. (O'Connor and Bredenkamp, 1997). At the high elevations the biome is dominated by C3 grasses. In the grass- lands, precipitation is highly seasonal with mean annual rainfall ranging between 750 and over 1200 mm, falling primarily during the austral-summer months (O'Connor and Bredenkamp, 1997) (Fig. 1). The southernmost part of Africa is characterised by the Fynbos biome, a fire-prone vegetation dominated by Ericaceous and Asteraceae shrubs, diverse Protea shrubs 5 and trees, and Restionaceae herbs (Cowling et al., 1997a). The Fynbos biome receives relatively high annual precipitation (1200 mm year −1 ) concentrated during the austral-winter months (Rutherford, 1997). The coastal forest biome is found along the eastern coast of the subcontinent and often occurs in small patches with high abundance of Podocarpus (Rutherford, 1997). Podocarpus patches can also be found in the 10 southeastern part of the Fynbos.

Marine core description and pollen analysis
Pollen analysis was conducted on marine core MD96-2098 (25 • 36 S, 12 • 38 E). This giant CALYPSO core was collected during the IMAGES II-NAUSICAA cruise at 15 a 2910 m water depth from the Lüderitz slope in the Walvis Basin, approximately 500 km northwest of the Orange River mouth (Fig. 1). The sediments of this 32 m long core were composed of calcium carbonates (foraminifera), biogenic silica (nannofossil mud), clays and organic matter (Bertrand et al., 1996). The core was sampled every 10 cm between 450 and 1940 cm (uncorrected depth) for pollen analysis. The uncor-20 rected depth did not take into account artificial gaps created during piston extraction (Bertrand et al., 1996).
Sample volumes were estimated by water displacement. Pollen concentrations per unit volume were calculated based on a known spike of exotic Lycopodium spores added to each sample. Pollen extraction techniques included treatment with hydroflu-Introduction allowed eliminating small non-palynomorph particles and concentrating pollen grains and spores. An independent test of this protocol showed that the use of a 10 µm sieve had no effect on the pollen composition of marine samples, i.e. comparison of filtered and unfiltered sample counts showed that taxa were not selectively filtered out during pollen preparation and concentration (see http://www.ephe-paleoclimat.com/ephe/ Lab%20Facilities.htm for a detailed pollen preparation protocol). We used the pollen spectra from 31 surface samples collected along a transect ( Fig. 1) from Cape Town (South Africa) to Lüderitz (Namibia) and designed to cover the four major biomes of southwestern Africa (Table S1). The transect included samples from the Desert, Fynbos, Nama-and Succulent-Karoo. The samples were treated 10 with standard acetolysis (Faegri and Iversen, 1989) and scanned under the microscope until completing pollen sums were greater than 300 grains (Supplement). We also used previously-published pollen spectra from 150 additional surface samples collected between 22 • and 35 • latitude south (APD, Gajewski et al., 2002). These pollen spectra were used to assess the distribution of Poaceae pollen percentages and other potential 15 indicators of large biomes in southern Africa. ArcGIS 10 was used to draw iso-lines of pollen percentages by interpolating values from a total of 178 surface samples through the natural neighbour method. Additionally, we analysed two marine pollen samples from the upper part of core MD96-2098 (at 5 and 10 cm depth). We compared the pollen spectra from these core top samples with the pollen signal of the modern vege-20 tation to evaluate how well marine sediments represent the vegetation of the adjacent landmasses, and to aid interpretation of the pollen record. Pollen identification was aided by the pollen reference collection of the Department of Plant Sciences at University of the Free State (Bloemfontein, South Africa), the African Pollen Database (APD) (http://medias3.mediasfrance.org/pollen), the Univer- 25 sal Pollen Collection (http://www.palyno.org/Pollen), and pollen descriptions published by Scott (1982). Pollen grains from the Asteraceae family were grouped into three pollen taxa: Artemisia-type, Stoebe-type and other morphotypes were classified into Asteraceae-other. Some morphotypes were grouped into family types: Acanthaceae, To summarize changes in the fossil pollen record, detrended correspondence analysis (DCA) was used in tandem with non-metric multidimensional scaling (NMDS) as parametric and non-parametric alternatives (McCune and Grace, 2002). Results from 5 the DCA ordination were preferred when NMDS was unable to reach a stable solution after several random starts, and when stress levels were too high to allow a meaningful interpretation (McCune and Grace, 2002). These ordinations were performed filtering out pollen morphotypes that only occurred in one sample to avoid overweighting rare taxa.

Marine core chronology
Two sediment gaps between 693 and 709 cm and 759 and 908 cm were described in the core log. These gaps were considered artificial and linked to piston extraction (Bertrand et al., 1996), thus the record could be assumed continuous. Depths were corrected to take into account these artificial sediment gaps. An age model was 15 established for the record based on 16 marine isotope events (MIE) from the Cibicidoides wuellerstorfi δ 18 O benthic record of MD96-2098 (Bertrand et al., 2002) and 14 Accelerator Mass Spectrometer radiocarbon ages (AMS 14 C) from mixed planktonic foraminifera extracted from MD96-2098 (Table S1). The 14 AMS 14 C dates were produced at the Laboratoire de Mesure du Carbone 14. One single 14 C date showed an 20 age reversal and was therefore excluded from the chronology on the principle of parsimony. AMS 14 C ages were calibrated using the marine09.14c curve (Hughen et al., 2004) from CALIB REV5.0 (Reimer et al., 2013). We applied a 400 year global reservoir correction factor and a weighted mean Delta R of 157 years derived from 9 regional reservoir error values from the Marine Reservoir Correction Dataset (Dewar et al., 25 2012; Southon et al., 2002). MIE ages were derived from LR04 global stack (Lisiecki and Raymo, 2005) and additional sources ( (Bertrand et al., 2003). The low net primary productivity that characterizes southwestern Africa (Imhoff et al., 2004) could explain relatively low pollen concentrations in the continental margin ( Fig. S2).
Pollen grains are part of the fine sediment fraction and can be transported by two 15 main vectors: aeolian or fluvial (Hooghiemstra et al., 1986;Heusser and Balsam, 1977). Dupont and Wyputta (2003) modelled present-day wind trajectories for marine core locations between 6 and 30 • S along the coastline of southern Africa. They suggest aeolian pollen input to the Walvis area (23 • S) via the south-east trade winds during austral summer, and dominant east-to-west wind directions during the austral fall and 20 winter. These winds transport pollen and other terrestrial particles from the Namib Desert, southern Namibia and western South Africa. The direction of the winds indicate that the Namib Desert, Nama-Karoo and Succulent-Karoo are the most likely sources of pollen in the Walvis area (Dupont and Wyputta, 2003 Dupont and Wyputta (2003) to be dominated by terrestrial aeolian input. However, given that this threshold was established using only two marine sites located 6 • apart at 23 • 26 (GeoB1710-3) and 29 • 27 (GeoB1722-1), it is difficult to conclude that MD96-2098 only receives wind-transported pollen. MD96-2098 likely receives fine sediments from the Orange River plume. Sedimen-5 tological analyses of the Orange River delta and plume indicate that fine muds are transported both northwards and southwards (Rogers and Rau, 2006). Additionally, an analysis of the imprint of terrigenous input in Atlantic surface sediments found relatively high Fe/K values along the Namibian and South African margin that could reflect the input of Orange River material (Govin et al., 2012). Pollen grains are hence likely to 10 reach the coring site from the Orange River catchment area. Scott et al. (2004) argue that pollen in marine sediments can be the result of longdistance transport by ocean currents, suggesting that pollen assemblages in marine sediments do not reflect accurately past changes in vegetation and climate. However, the highest pollen influx in marine sediments along this margin is near the coast and 15 the vegetation source (Dupont et al., 2007), not along the paths of oceanic currents (i.e. Benguela Current). Additionally, analyses of pollen transport vs. source in northwestern Africa show that pollen grains can sink rapidly in the water column (Hooghiemstra et al., 1986) before they can be carried away by ocean currents. As a result, influence of oceanic currents on the composition of pollen assemblages is probably negligible. 20 Overall, the marine site MD96-2098 might receive both aeolian and fluvial pollen input from the vegetation located east and southeast to the site.
The pollen composition of the two core-top samples from core MD96-2098 are dominated by Poaceae (30 and 40 %), Cyperaceae (20 %) and Chenopodiaceae-Amaranthaceae (20 and 30 %) (Fig. S2). This composition corresponds well with the 25 pollen spectra from the three major biomes occupying today the adjacent landmasses ( Fig. 1a): Desert, Nama-Karoo and Fn-LSav (Figs. S3 and 5). Pollen percentages from Fynbos taxa are less than 10 %, Podocarpus is weakly represented, and taxa specifically found in the broad-leaved savanna (e.g. Caesalpinaceae, Combretaceae) are not 355 grass species in virtually all southern African biomes (Cowling et al., 1997b). Altogether, the spatial distribution of Poaceae pollen percentages appears to be essential information to distinguish the pollen signal from major biomes, and therefore climatic zones, in this region. In the eastern and northeastern part of southern Africa, the highest percentages of Poaceae pollen (up to 90 %) are found in the pollen rain of the 10 Bd-LSav and grasslands. In the western half of southern Africa, Poaceae pollen percentages in terrestrial surface samples are up to 60 % in the Nama-Karoo and its transition with the Fn-LSav (Fig. 3). This suggests an overrepresentation of Poaceae in the pollen rain of the Nama-Karoo biome where grasses can be abundant but are not necessarily dominant. 15 Poaceae is likely to be well represented in other parts of the Fn-LSav, but the paucity of surface samples from this biome hinder drawing further conclusions. In the Namib Desert where abundance of grasses in the vegetation is low, Poaceae pollen percentages are as high as 25 % in surface samples, comparable to 20 % reported from hyrax dung (Scott et al., 2004). 20 In marine surface samples along the southwestern African coast, Poaceae pollen percentages are as low as 10 % in samples collected in front of the Bd-LSav at around 15 • S (Fig. 3). Nama-Karoo and Fn-LSav biomes. As Poaceae pollen percentages in Desert surface samples are less than 25 %, high percentages of grass pollen from marine sediments in the southwestern Africa margin should be interpreted as indicator of the Nama-Karoo and the Fn-LSav, where Poaceae is as high as 70 % in terrestrial surface samples. Our field observations also support this view as we observed large grass-dominated 5 vegetation in the Nama-Karoo and Fn-LSav. (Fig. 2).

Southwestern Africa vegetation and climatic changes from MIS 6 to 2
The pollen record presented here spans from 24.7 to 190 ka. A log transformation of concentration values in MD96-2098 results in a curve remarkably similar to that of δ 18 O benthic values (Fig. 4) and may be linked to changes in pollen input at the coring 10 site. Relative increases in pollen concentration could indicate an increase in pollen arrival during low sea-level stands when the vegetation source was closest (i.e. during glacial stages). However, this is unlikely because of the width of the Walvis continental Shelf (i.e. rapid depth change in a few kilometres). An increase of pollen concentration might indicate instead an increase in pollen input during glacials, and/or 15 an increase in pollen preservation linked to upwelling enhancement as suggested by Pichevin et al. (2005). Glacial-interglacial pollen concentration variations have no effect on the interpretation of the pollen record which is based on relative frequencies, but they do indicate the influence of the obliquity signal in the pollen record from MD96-2098. 20 The axis scores on DCA1 reveal changes in the composition of pollen assemblages that also resemble variations in the δ 18 O benthic record (Fig. 4). DCA1 axis scores from MIS 5 and 3 are overall positive in value, while scores from MIS 6 and 4 are negative. A series of large-magnitude changes in DCA1 axis scores (i.e. when adjacent sample scores switch from positive to negative values) are also visible and increase in ampli-25 tude after ca. 100 ka. Such changes in DCA Axis1 scores are also observed during MIS 6 but are of lesser magnitude. Changes in DCA axis scores suggest significant changes in vegetation composition from one time step to the next. 357 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Nama-Karoo and Fn-LSav pollen percentages are up to 60 % during MIS 5 and display three percentage peaks that correspond with δ 18 O benthic and precession minima (Fig. 4a). These percentage peaks are centred at 125, 107 and 83 ka. The composition of pollen spectra from warm marine substages MIS 5e, 5c and 5a is comparable to the core-top samples (Fig. S2) and corresponds well with the modern pollen spectra 5 from Nama-Karoo and Fn-LSav (Fig. S3). At the same time, Nama-Karoo and Fn-LSav pollen percentages in the core-top samples are relatively low compared to their maximum during MIS 5e (Fig. 4b). During

20
The increases of Nama-Karoo and Fn-LSav during MIS 5e, 5c and 5a suggest an increase in aridity in southwestern Africa that likely resulted from expansions in three directions (Fig. 5). The Nama-Karoo and Fn-LSav probably expanded to the northwest into the present-day area of the coastal Namib Desert as the intensity of BUS weakened during MIS 5 warm substages. This weakening has been documented through 25 alkenone-based SST from marine core GeoB1711-3 (Kirst et al., 1999) (Fig. 4c), foraminifera-assemblage based SST (Chen et al., 2002) and grain-size end-member modelling (Stuut et al., 2002). Stuut and Lamy (2004)  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | glacials, resulting in a reduction of the wind-driven upwelling. A weakened BUS and the associated relative increase in humidity likely led to a colonization of Desert areas by Nama-Karoo or Fn-LSav (Fig. 5a). Comparable contractions of the Namib Desert linked to increased SSTs and weakening of BUS during the present interglacial are documented by Shi et al. (2000). 5 To the south, the Nama-Karoo and Fn-LSav likely expanded at the expense of the Succulent-Karoo and Fynbos. Warm Antarctica temperatures recorded during warm substages of MIS 5 (EPICA, 2006) would drive the southern westerlies polewards (Ruddiman, 2006), contributing to the ventilation of deep CO 2 -rich waters in the southern Ocean (Toggweiler and Russell, 2008). This mechanism would explain the parallel-10 ing trends observed between MIS 5 Nama-Karoo and Fn-LSav expansions in southern Africa and the atmospheric CO 2 record (Petit et al., 1999;Bereiter et al., 2012) (Fig. 5). A poleward migration of the westerlies during the present interglacial, relative to their position during the previous glacial, has been suggested by Weldeab et al. (2013), and correlated with nssCa 2+ from Antarctica (Röthlisberger et al., 2008). Such poleward 15 migration of the westerlies during the present interglacial is equivalent to the westerlies migration we propose for warmest periods of the last interglacial. The increased Agulhas leakage documented in the Cape basin record during the last interglacial (Peeters et al., 2004) (Fig. 4c) has been linked to a southward migration of the subtropical front and the westerlies, reducing austral-winter precipitation over southern Africa. Such an 20 atmospheric configuration would in turn favour the development of the Nama-Karoo at the expense of Succulent-Karoo and Fynbos biomes (Fig. 5a).
To the northeast, Nama-Karoo and Fn-LSav likely pushed the limit of Bd-LSav equatorward as austral-summer precipitation decreased (Fig. 5a). Austral-summer precipitation reductions in Southern Africa have been linked to reduced austral-summer in-25 solation in the Pretoria saltpan (Partridge et al., 1997) and to reductions of grassfuelled fires during precession minima reconstructed from MD96-2098(Daniau et al., 2013 (Fig. 4b) south Atlantic and the Indian Oceans would expand (Fig. 5a) (Ruddiman, 2006). Such changes in the tropical and subtropical pressure systems would allow the expansion of the Nama-Karoo and Fn-LSav to the northeast. In contrast with the results presented here, previous studies report poleward interglacial expansions of savannas based on pollen records from marine sediments along 5 the southwestern African coast (Dupont, 2011). However, these studies univocally interpret the Poaceae pollen percentage increases as the result of savanna expansions. Such an interpreation is potentially plausible in marine records collecting pollen from broad-leaved savanna vegetation, e.g. the Limpopo basin (Dupont et al., 2011). However, our study shows that Poaceae pollen percentage increases in sequences located 10 off the southwestern African coast can alternatively indicate the expansion of fineleaved savanna and Nama-Karoo vegetation. Previous studies do not differentiate between the broad-leaved and fine-leaved savannas, despite the significant climatic and structural differences between these two types of vegetation. The Bd-LSav is influenced by fire and receives a considerable amount of precipitation during the austral 15 summer (Scholes, 1997). The Fn-LSav is structurally and climatically more similar to the Nama-Karoo biome, as it receives very low austral-summer precipitation and does not burn (Archibald et al., 2010) despite being under a regime of significant precipitation seasonality (Scholes, 1997). If high Poaceae pollen percentages during MIS 5 warm substages in our record were related with expansions of the Bd-LSav and increased 20 summer precipitation, the fire activity should also increase during these substages. Instead, an independent charcoal record from the same marine sequence MD96-2098 (Fig. 4b) documents reductions of grass-fuelled fires and a decrease in austral-summer precipitation during MIS 5 precession minima (Daniau et al., 2013). An atmospheric configuration with reduced austral summer precipitation in Southern Africa and the 25 ITCZ shifted northward during the warmest periods of MIS 5 is also consistent with documented strengthening of Asian monsoon and weakening of the South American monsoon during the last-interglacial precession minima (Wang et al., 2004).

360
Our results suggest that the Bd-LSav retreated equatorwards during MIS 5 precession minima, while Nama-Karoo and Fn-LSav expanded. Nama-Karoo and Fn-LSav probably covered a surface area larger than at present during MIS 5 warm substages, as indicated by up to 70 % pollen from this biome during MIS 5 compared to 35 % in the core-top samples. This is despite the difference in precession parameters between 5 the last millennium and MIS 5 warm substages. Recent model experiments on the impact of precession changes on South African vegetation indicate that high precession is linked to reductions of the broad-leaved savannas (Woillez et al., 2014). Altogether these vegetation changes point to increased aridity in southwestern Africa during the warmest periods of the last interglacial.

10
During glacial isotopic stages, contractions of the Nama-Karoo and Fn-LSav would result from a different atmospheric configuration (Fig. 5b): a southward migration of the ITCZ and the associated South African monsoon (Daniau et al., 2013;Partridge et al., 1997) increasing austral-summer rainfall over southern Africa; an intensification of BUS and decreased SST off the Namibian coast (Stuut and Lamy, 2004;Kirst et al., 1999) 15 leading to aridification of coastal areas; and lastly, an equatorward migration of the westerlies increasing austral-winter precipitation and allowing a northward expansion of the winter-rain zone in Southern Africa (Chase and Meadows, 2007). The proposed glacial precipitation changes are consistent with recent estimates of Last Glacial Maximum palaeoprecipitation based on glacier reconstruction and mass-balance modelling 20 (Mills et al., 2012), with leaf-wax reconstructions of hydroclimate (Collins et al., 2014), and with simulated glacial climatic fluctuations in southern Africa (Huntley et al., 2014).
The pollen record from MD96-2098 also suggested glacial expansions of Fynbos (Fig. 4b), as pollen percentages of Artemisia-type, Stoebe-type, Passerina and Ericaceae were higher during MIS 6, 4 and 3 than in the core-top samples (Fig. S2). These 25 results were consistent with glacial northward expansions of Fynbos documented in other pollen records from southern Africa (Shi et al., 2000;Dupont et al., 2007). Our record also documented a large peak in Fynbos indicators (Fig. 4b) that coincided with a fast decrease in Nama-Karoo and fine-leaved savanna pollen percentages at the 361 pansion of the winter-rain zone of southern Africa. Transitions MIS 6/5 and 4/3 were characterized by small but rapid increases in Podocarpus, potentially linked to a short increase in annual precipitation. Such increases in Podocarpus have also been documented in other records from southern Africa (Dupont et al., 2011;Dupont, 2011).
Finally, the amplitude of millennial-scale vegetation changes increased between 10 ca. 100 and 25 ka, and was highlighted by switches from negative to positive DCA1 scores (Fig. 4b) and increased variability of Restionaceae pollen percentages. These results could indicate either enhanced trade-wind variability or Fynbos vegetation expansions. Other Fynbos indicators did not display such trend (Fig. 4), suggesting that Restionaceae variability between 100 and 24 ka were more likely the result of en-15 hanced variability of southeast trade winds. Restionaceae pollen percentage data from a record 2 • North of our marine site also showed comparable increases in the amplitude of millennial-scale changes (Shi et al., 2001). Grain-size wind strength tracers from the Walvis Ridge displayed enhanced millennial-scale variability, although only after ca. 80 ka (Stuut et al., 2002). An analysis of BUS dynamics over the past 190 ka found 20 increased millennial-scale variability of wind strength after ca. 100 ka and the windiest conditions in this zone during MIS 4 and 3 (Stuut et al., 2002). Such millennial-scale atmospheric reorganisations are recorded in the pollen-based DCA analysis as rapid biome shifts in southwestern Africa.

25
Terrestrial and marine markers from the marine core MD96-2098 documented expansions of the Nama-Karoo and fine-leaved savanna during MIS 5e, 5c and 5a warm 362 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | substages. Northwestern expansions of the Nama-Karoo and Fn-LSav are potentially linked to the reduction of BUS and a local increase in humidity in the desert area, while aridification increased at a regional scale. Towards the east, Nama-Karoo and Fn-LSav expansions probably resulted from increased subtropical high pressure, a northward shift of the ITCZ, and reduced austral-summer precipitation. Nama-Karoo and Fn-LSav 5 expansions to the southern boundary are possibly associated with southern displacement of the westerlies and the subtropical front, decreasing austral-winter precipitation. During glacial isotopic stages MIS 6, 4 and 3, Fynbos biome expansions are probably linked to the increased influence of the southern westerlies and austral-winter precipitation in southwestern Africa. Our pollen record also suggested that warm-cold 10 or cold-warm transitions between isotopic stages and substages were punctuated by short increases in humidity. Increased variability of vegetation changes at millennial timescales ca. 100 ka was also documented and could be associated with previouslyidentified enhanced variability of the southeastern trade winds.
Interglacial-glacial southern Africa biome dynamics were linked to atmospheric and 15 oceanic dynamics resulting from changes in global ice volume and precession at orbital timescales. Atmospheric configurations with westerly winds shifted southwards relative to today have been suggested for other interglacials (Peeters et al., 2004) and are projected for the end of 21st-century under current global warming (Beal et al., 2011). This is likely to reduce austral-winter precipitation over southern Africa and 20 favour expansions of the Nama-Karoo at the expense of the winter-rain fed Fynbos and Succulent-Karoo biomes. However, taking the current orbital configuration alone, the Nama-Karoo and Fn-LSav in southern Africa might naturally remain relatively reduced for several millennial ahead.   Scholes (1997), location of the Orange River and major tributaries, oceanic currents (grey arrows) and southeastern trade winds (black arrows). Temperature seasonality (b) and annual precipitation in mm yr −1 (c) extracted from the WorldClim dataset (Hijmans et al., 2005). Black dots indicate location of marine core MD96-2098 and numbers indicate the location of surface sample collection points described in Table S1 Scholes, 1997;Mucina et al., 2007). The broad-leaved savanna distribution includes the Mopane and mixed savannas described by Scholes (1997). Iso-lines are plotted based on pollen percentage data from surface samples analysed in this study (red dots) and pollen spectra from other samples previously published and extracted from the African Pollen Database (black dots) (Gajewski et al., 2002). Poaceae pollen percentage in the marine domain are redrawn from Dupont and Wyputta (2003) (Laskar et al., 2004). (b) Stable oxygen profile of benthic foraminifera Cibicidoides wuellerstorfi (Bertrand et al., 2002), log-transformed total pollen concentration, detrended correspondence analysis Axis1 scores, pollen percentages of indicator taxa for Nama-Karoo and fine-leaved savanna (Acanthaceae, Aizoaceae, Crassulaceae, Euphoria, Poaceae, and Tribulus,) and Fynbos (Artemisia-type, Ericaceae, Passerina, Protea, and Stoebe-type), charcoal concentrations in number of particles per gram (nb g −1 ) from marine core MD96-2098 plotted against age in ka (thousands of calibrated/calendar years before present). Stars on the left correspond to percentage of pollen taxa in two top-core samples dating 530 and 1060 calibrated years before present. (c) Independent climatic records discussed in the text. Gray curve: low-resolution CO 2 record from Vostok (Petit et al., 1999); black curve: high-resolution CO 2 record from Bereiter et al. (2012). Stage boundary ages for 3/2, 4/3, and 5/4 from (Sanchez Goñi and Harrison, 2010) and 6/5 from (Henderson and Slowey, 2000).