Waxing and Waning of Forests: Late Quaternary Biogeography of Southeast Africa

Received: 12 October 2017 | Revised: 5 March 2018 | Accepted: 9 March 2018 DOI: 10.1111/gcb.14150

PRIMARY RESEARCH ARTICLE

Waxing and waning of forests: Late Quaternary biogeography of southeast Africa

Sarah J. Ivory1,2 | Anne-Marie Lezine 3 | Annie Vincens4 | Andrew S. Cohen5

1Department of Anthropology, Ohio State University, Columbus, OH, USA Abstract 2Department of Geosciences, Penn State African ecosystems are at great risk. Despite their ecological and economic impor- University, State College, PA, USA tance, long-standing ideas about African forest ecology and biogeography, such as 3LOCEAN, CNRS, Paris, France the timing of changes in forest extent and the importance of disturbance, have been 4CEREGE, CNRS, Aix-en-Provence, France 5Department of Geosciences, University of unable to be tested due to a lack of sufficiently long records. Here, we present the Arizona, Tucson, AZ, USA longest continuous terrestrial record of late Quaternary vegetation from southern

Correspondence Africa collected to date from a drill core from Lake Malawi covering the last Sarah J. Ivory, Department of Geosciences, ~600,000 years. Pollen analysis permits us to investigate changes in vegetation Penn State University, State College, PA, USA. structure and composition over multiple climatic transitions. We observe nine Email: [email protected] phases of forest expansion and collapse related to regional hydroclimate change.

Funding information The development of desert, steppe and grassland vegetation during arid periods is US National Science Foundation–Earth likely dynamically linked to thresholds in regional hydrology associated with lake System History Program, Grant/Award Number: EAR-0602350; International level and moisture recycling. Species composition of these dryland ecosystems Continental Scientific Drilling Program; varied greatly and is unlike the vegetation found at Malawi today, with assemblages National Science Foundation Graduate Research Fellowship, Grant/Award Number: suggesting strong Somali-Masai affinities. Furthermore, nearly all semiarid 2009078688 assemblages contain low forest taxa abundances, suggesting that moist lowland gallery forests formed refugia along waterways during arid times. When the region was wet, forests were species-rich and very high afromontane tree abundances suggest frequent widespread lowland colonization by modern high elevation trees. Further- more, species composition varied little amongst forest phases until ~80 ka when disturbance tolerant tree taxa characteristic of the modern vegetation increased in abundance. The waxing and waning of forests has important implications for under- standing the processes that control modern tropical vegetation biogeography as well as the environments of early humans across Africa. Finally, this work highlights the resilience of montane forests during previous warm intervals, which is relevant for future climate change; however, we point to a fundamental shift in disturbance regimes which are crucial for the structure and composition of modern East African landscapes.

KEYWORDS Africa, biogeography, global change, hydrology, Lake Malawi, palaeoclimate, palaeoenvironments, tropical forests

1 | INTRODUCTION emergence of disturbance-maintained Zambezian miombo woodlands, have yet to be resolved due to the availability of sufficiently long Tropical African ecosystems are immensely important for their ecolog- terrestrial records which capture previous warm intervals. Marine ical value, as bastions of biodiversity and endemism, as well as for cores covering this period show a large expansion of afromontane essential resources for millions of people for daily subsistence (e.g. forest, suggesting that ancient forests may have been substantially Doherty, Sitch, Smith, Lewis, & Thornton, 2010; Mittermeier, Turner, different than lowland communities observed today, however, no Larsen, Brooks, & Gascon, 2011; White, 1981). However, African for- terrestrial record has ever documented their presence or extent ests and semiarid lands are already changing as a result of pressure (Dupont et al., 2011; Ning & Dupont, 1997). In addition, palaeocli- from climate and land-use change (Fugere, Kasangaki, & Chapman, mate records from across the continent show strongly antiphased 2016; Newmark & McNeally, 2018; Niang, Ruppel, & Abdrabo, 2014). signals implying that these earlier high amplitude wet–dry cycles of In turn, feedbacks associated with altered disturbance regimes, such as the late Pleistocene occurred asynchronously across the continent. wildfires, on landscapes processes could result in detrimental changes For example at Lake Malawi from ~600 to 100 ka, lake levels to important ecosystems services (e.g. Bond, Woodward, & Midgley, decreased multiple times by as much as 550 m, whereas north of 2005; Midgley & Bond, 2015; Tilman et al., 1997; Van Langevelde the equator, lakes expanded in the Kenyan rift valley (Ivory et al., et al., 2003). Therefore, it is increasingly important to be able to pre- 2016; Salzburger, Van Bocxlaer, & Cohen, 2014; Trauth et al., 2007). dict long-term species responses to climate. However, particularly in These large hydroclimatic shifts and patchiness of continental cli- Africa, modern ecological observations are sparse, and the range of mate likely resulted in massive reorganization of species as they observations is short (e.g. Feeley & Silman, 2011; Willis & Birks, 2006). tracked optimal climate. Perhaps more importantly, although we Furthermore, as tropical climate is likely moving to a state that does know that species respond individualistically to climate (e.g. Palmer not exist today, projecting species’ responses to future change relies et al., 2015; Stewart, 2009), it is yet unclear how ecosystem compo- on extrapolation (Niang et al., 2014; Williams & Jackson, 2007; Wil- sition and structure changed or how differential dispersal may have liams, Jackson, & Kutzbach, 2007). Thus, long-term (>105 year) records influenced community composition and forest succession. which observe the climatic processes that control vegetation structure Lake Malawi, southeast Africa, sits currently at the southern and composition are required (Dietl et al., 2015; Dietl & Flessa, 2017; extent of the Intertropical Convergence Zone (ITCZ) and is sensitive Lindbladh, Fraver, Edvardsson, & Felton, 2013; ). to fluctuations in regional climate (Figure 1; Cohen et al., 2007; Ivory Palaeoecological records have provided insights about long-term et al., 2016; Scholz et al., 2007). Long drill core records from this ecological processes and allowed us to look at vegetation change in lake preserve a regional climate signal over the late Quaternary with response to conditions warmer than today and over periods of rapid rapid transitions from arid to humid climates (Ivory et al., 2016). Fos- change which can be used as future analogues (e.g. Blois, Zarnetske, sil pollen analysis of sediments from this record provides the ideal Fitzpatrick, & Finnegan, 2013; Hughen, Eglinton, Xu, & Makou, opportunity to better understand ecological processes that operate 2004; Salzmann & Hoelzmann, 2005). For example in Africa, the over many climatic transitions. The unique position of the lake with prevalence of forests throughout the region when global tempera- respect to tropical circulation as well as ecologically important tures are higher, such as during interglacials, has been thought of as ecosystems within the African tropics allows us to look at changing canonical, despite some evidence to the contrary (Beuning, Zimmer- forest composition and vegetation structure through time as well man, Ivory, & Cohen, 2011; Ivory, Lezine, Vincens, & Cohen, 2012; alterations to disturbance regimes. Furthermore, this unique record Vincens, Garcin, & Buchet, 2007). Warming and wetting over the last is the first long, terrestrial record of vegetation change from East deglacial period associated with the reinforcement of regional mon- Africa covering the entirety of the late Quaternary (>500 ka) and soons resulted in an expansion of woodlands and forests as well as provides insights into large-scale reorganization of biomes over mul- the retreat of dryland ecosystems, which dominated during the Last tiple wet-dry cycles in African climate. Glacial Maximum (LGM; e.g. Izumi & Lezine, 2016; Maley, 1992). However, hydrological records reconstructed from ancient lake 1.1 | Modern setting deposits and marine cores suggest that LGM to modern climate changes in Africa were small in amplitude and occurred over thou- Lake Malawi, the southernmost lake in the East African Rift System, sands of years with respect to those which occurred earlier in the is composed of a series of alternating N-S oriented half-graben late Pleistocene (Dupont & Kuhlmann, 2017; Dupont et al., 2011; basins (Figure 1). The lake is hydrologically open and is drained by Scholz et al., 2007; Trauth et al., 2007). Thus, as most existing the Shire River to the south (sill depth = 6 m), although most water records do not extend beyond the last glacial period, it is unclear loss occurs via evaporation (Eccles, 1974). Mountain ranges in the whether the LGM-modern vegetation response is indicative of larger northern (Livingstone Mountains and Rungwe Highlands) and the alterations in climate over multiple cycles. However, the lack of long western (Nyika Plateau) watershed rise steeply from the lake shore records has made it challenging to test this assumption. (478 m asl) to nearly 3,000 m. The lake currently sits at the southern Moreover, long-standing, fundamental biogeographical questions, limit of the ITCZ (Nicholson, 1996; Figure 1). Within the watershed, particularly regarding the distribution of afromontane forest and the MAP is very heterogeneous and ranges from 800 mm/year in the IVORY ET AL. | 3

(a) (b)

MAL05-1B/1C

Desert Semi-desert Steppe/Grassland Wooded Grassland Tropical Woodland Tropical Rainforest

Mediterranean Forest/Scrub Dry miombo woodland Wet miombo woodland Coastal Forest Tropical seasonal forest Afromontane forest Afromontane Forest

FIGURE 1 (a) Map of the phytogeographic regions of Africa following White (1983) with the afromontane region highlighted in purple. The black rectangle shows the position of the lake and represents the area of the map of the watershed. (b) Map of the watershed and vegetation of Lake Malawi. Colours represent dominant vegetation types present in the legend. Black lines are the watershed boundary and principal waterways. Drill site for core MAL05-1B/1C is marked lowlands to 2,400 mm/year on the slopes of Rungwe Highlands whereas Podocarpus, Juniperus, Olea africana and various Ericaceae (DeBusk, 1997). Rainfall is highly seasonal, with a single rainy season are common in drier areas above 2,000 m asl (800–1,700 mm/year; from November to April, when prevailing surface winds are north- ~4 dry months; White, 1983). Grasslands are widespread at all eleva- easterly (NE), and rainfall reaches 250 mm/month. A long dry season tions but are concentrated at the southern end of the lake where from May to October lasts ~6 months when virtually no precipita- rainfall is lowest (White, 1983). tion occurs and strong southeasterly (SE) tradewinds prevail. A prior taphonomic and pollen transport study suggests that pol- Vegetation in the Malawi watershed is largely constrained by len deposited in the central basin is transported predominantly by rainfall and rainfall seasonality (Ivory et al., 2012; Vincens, Garcin wind (DeBusk, 1997). Today, most wind transported pollen comes et al., 2007). In the low to midaltitudes (<1,500 m asl), low-diversity from north of the lake during the austral summer, when NE trade- deciduous Zambezian miombo woodlands grow in areas of highly winds predominate and when the majority of plants flower. Although seasonal rainfall and are dominated by deciduous trees such Brachys- no large rivers drain into the central basin adjacent to the core site, tegia, Berlinia and Isoberlinia (Figure 1). These woodlands can be sub- the presence of riparian taxa, such as tropical seasonal forest, aquat- divided into a drier and wetter type found in areas of more or less ics and ferns, suggests that some pollen is delivered to the lake than 1,000 mm/year, respectively (White, 1983). The wetter miombo along streams (DeBusk, 1997). has higher, denser canopy cover and is frequently associated with other trees, such as Uapaca (White, 1983). The drier miombo dis- plays a more open canopy and higher proportion of grasses and 2 | MATERIALS AND METHODS trees from the Combretaceae family (White, 1983). Closed-canopy tropical seasonal forests are dominated by semideciduous and ever- Drill cores MAL05-1B and MAL05-1C were collected at the same green trees like Myrica, Macaranga and various Ulmaceae and site several metres offset in the central basin of Lake Malawi in Moraceae species, as well as shade-loving herbs like Urticaceae and 2005 (Figure 1; 11°180S, 34°260E; 592 m water depth). The cores ferns. These forests are not widespread but also occur in the low were collected using a hydraulic piston corer in the upper ~80 m of to midaltitudes along streams as gallery forests and in areas sediment and rotational drilling in more consolidated sediments with moister edaphic conditions (Figure 1). Flooded grasslands with below 80 mblf. Further details of the core site are provided in Lyons Cyperaceae and Typha are present at the lakeshore. et al. (2015), Scholz et al. (2011), and the age model in Ivory et al. Above 1,500 m asl, afromontane forests occur as discontinuous, (2016). This study focuses on a new dataset from a 170-m core sec- often degraded patches within extensive high elevation grasslands tion (10–180 mblf). An age model was constructed for the entirety (Figure 1). Olea capensis is typical of moister areas (1,500– of core MAL05-1B/1C based on a cubic spline interpolation using 2,000 mm/year; 0–3 dry months) from 1,500 to 2,500 m asl, 16 AMS radiocarbon dates, presence of tephra from the Toba 4 | IVORY ET AL. volcanic eruption (75 ka), 16 palaeomagnetic excursion events, 2 Ar- undifferentiated, Myrica, Juniperus-type, Faurea-type and Ilex mitis. Ar dates on tephra and seven palaeomagnetic excursion, intensity Miombo woodland is only represented by Brachystegia and Uapaca and reversal events (Ivory et al., 2016; Lane, Chorn, & Johnson, in summary figures, as these are the highest-dispersing, most indica- 2013; Table S1 and Figure S1). Core lithology is characterized by tive taxa in this vegetation type. high amplitude facies changes associated with lake level and is domi- To compare pollen assemblages from the fossil record to those nated by silty and clayey diatom oozes and homogenous silty clays in Africa today, biomization was conducted. Biomization is a tech- (Lyons et al., 2015). Lake levels were reconstructed by Ivory et al. nique which assigns pollen taxa to plant functional types which can (2016) by characterizing aquatic and terrestrial biofacies (ostracodes, be used to reconstruct biomes from fossil pollen samples based on chaobrid fragments, fish fossils, fern sori, charred particles and green well-studied modern species assemblages and affinities (Jolly et al., algae) and authigenic minerals (quartz, mica, pyrite, siderite and 1998). All fossil samples were quantitatively assigned to African ooids). Details on this dataset can be found in Ivory et al. (2016); biomes based on the biomization technique developed by e.g. Elenga however, the first principal component of this dataset explains et al. (2000), Jolly et al. (1998) and Izumi and Lezine (2016), and ref- 14.27% of the variance and was found to be strongly related to lake erences therein. The method uses sums of square-root transformed level fluctuations. abundances for each sample to create biome affinity scores. African The fossil pollen record is based on 402 samples from the two biomes include afromontane (WAMF), afroalpine (XERO), tropical adjacent cores (Figure 2). Core MAL05-1C represents the upper por- rainforest (TRFO), tropical seasonal forest (TSFO), tropical deciduous tion of the record comprising 40 samples from 10.5 to 62 mblf and forest (TDFO), wooded grassland (SAVA), steppe (STEP) and desert was presented in Beuning et al. (2011). The remaining 362 previ- (DESE). Vincens et al. (2006) conducted biomization on modern pol- ously unpublished samples from core MAL05-1B were taken from len surface samples from East Africa. Pollen samples were assigned 62 to 170 mblf with an ~30 cm resolution (Table S2). We will dis- the correct biomes for 82.6% of the full sample set. Vincens et al. cuss the composite data referred to as MAL05-1B/1C. All samples (2006) also found that incorrectly assigned samples tended to have a were processed using standard methods with the addition of Lycopo- slight bias towards drier biomes types and that degraded/mosaic dium spores (Faegri & Iversen, 1989; Vincens, Garcin et al., 2007) vegetation were the most difficult to correctly assign. and were sieved at 10 microns to remove clays. Between 300 and To look at species turnover and dissimilarity amongst the sam- 500 grains were counted per sample at 1,0009 magnification with ples, several multivariate statistical techniques were employed. First, the exception of the six uppermost samples from MAL05-1B which detrended correspondence analysis (DCA) was performed using the range from 17 to 149 grains/sample due to extremely low pollen vegan package in R on all samples and pollen taxa with a minimum concentrations. Average temporal resolution for the whole fossil pol- value of >2% (Oksanen et al., 2013; R Core Team, 2017). DCA is len record is 1,400 years, however, a higher average resolution of well-suited to ecological abundance data with potentially long uni- 620 years was obtained from 250 to 20 ka. The lake level recon- modal environmental gradients, and variance along the first axis is struction from Ivory et al. (2016) has a temporal resolution over the frequently interpreted as proportional to the amount of species turn- last 600 ka of 500 years, with a 250 years resolution in the last over among samples (ter Braak, 1985). DCA was performed on Wis- 250 ka. A total of 227 pollen taxa were identified using the African consin-transformed abundances with 26 segments for detrending Pollen Database (http://apd.sedoo.fr/apd/accueil.htm) and atlases of and rescaling of the axes with 4 iterations. Following DCA on all fos- pollen morphology (Bonnefille & Riollet, 1980; Maley, 1970). sil samples only, a further DCA was performed by passively project- Pollen percentages were calculated against a sum of all pollen ing a small subset of modern pollen samples from lake surface and spores less undeterminable grains, aquatics (Cyperaceae, Typha, sediments onto the biplot created with the fossil samples. These 54 Nymphaea, Polygonum senegalense-type, Ottelia, Laurembergia) and modern samples were analysed and presented in DeBusk (1997). undifferentiated bryophytes, which were calculated separately (Fig- Biome affinity scores produced during the biomization of the fossil ure 2). Pollen diagrams were drawn using Tilia (Grimm, 1990), and and modern samples were then used to fit environmental vectors to zonation was determined quantitatively by constrained cluster analy- the DCA biplots. This allows us to investigate relationships among sis using CONISS (Grimm, 1987). This zonation highlighted alterna- the modern biome affinities of the fossil and modern samples. Envi- tions between zones dominated by semiarid and forest pollen taxa, ronmental fit of biomes only shows significant environmental vari- which led to their classification as either forest or semiarid phases. ables determined by p-values based on 999 permutations. The abundance values presented in the results section are average Next, a large dataset of 1,001 modern pollen samples from percentages for a zone unless otherwise specified. Pollen nomencla- throughout Africa was acquired from the African Pollen Database ture follows Vincens, Lezine, Buchet, Lewden, and Le Thomas (http://apd.sedoo.fr/accueil.htm). To investigate dissimilarity between (2007), and pollen assemblages described here are based on biomes the fossil pollen assemblages and modern pollen assemblages across outlined by Vincens, Bremond, Brewer, Buchet, and Dussouillez the continent, we calculated minimum squared chord distance. This (2006) and used in previous studies in this watershed (Ivory et al., technique has been used in other palynological studies to determine 2012; Vincens, Garcin et al., 2007). Tropical seasonal forest is the whether fossil assemblages from a sediment core have modern ana- sum of Moraceae, Macaranga-type, Celtis and Trema-type orientalis, logues anywhere within a study region (Gill, Williams, Jackson, Linin- and afromontane is the sum of Podocarpus, Olea, Ericaceae ger, & Robinson, 2009; Overpeck, Webb, & Prentice, 1985). IVORY ET AL. | 5

-type

-type type

s e re 0 Poaceae AmaranthaceaeAcacia CommiphoraBrachystegia africanaUapaca Celtis MoraceaeMacaranga-Podocarpus Olea EricaeaeJuniperusIlex mitisPrunus africanaT Herbs CONISS

50 F9

S8 100 F8 S7 F7 S6 150 F6 S5 F5 200 S4 F4

250 S3 300

350

Age (kyr BP) F3 400

450

S2 500 F2

550 S1

F1 600 20 40 60 80 100 20 20 20 40 60 20 20 40 60 80100 10 20 30 40 Total sum of Dry Woodland/ Miombo Tropical Afromontane squares Wooded Grassland Woodland Seasonal Forest Forest

FIGURE 2 Pollen percentage diagram of core MAL05-1B/1C. Only dominant pollen taxa are pictured, and these are colour-coded based on vegetation type. To the right of the diagram is the CONISS cluster analysis as well as a bar representing the assignment of zones to forest (green) or semiarid (yellow) phases

Minimum squared chord distance was calculated using the analogue 3 | RESULTS package in R for each fossil sample against the full dataset of modern African pollen samples, and fossil samples were considered “no ana- Pollen preservation is good throughout this record, and concentra- logue” if minimum squared chord distance was greater than 0.3 tions are high (x = 23,480 grains/cm3), with the exception of the six (Simpson, Oksanen, & Simpson, 2016). samples from MAL05-1B (x = 2,758 grains/cm3). These samples, To estimate species richness in the fossil pollen data, we calcu- which occur from 61.9 to 63.7 mblf, are coeval with a period of lated Margalef’s Index (Dmg), which is highly correlated with species extreme aridity, as described in Cohen et al. (2007), when extremely richness in a population (Margalef, 1958). This is calculated as: low lake levels and coarser sediments resulted in pollen reworking and abrasion. Abundance of broken or reworked grains is typically Dmg ¼ðS 1Þ= ln N less than 2%, except in these six samples likely due to abrasion from where S is the total number of taxa in a sample and N is the total an increased sand fraction, when values are still never higher than number of individual grains counted. Goring, Lacourse, Pellatt, and 5.8%. Our pollen stratigraphy is divided into 17 zones (Figure 2). Mathewes (2013) observed that diversity indices calculated on large These zones highlight an alternation between dominance of arboreal regional pollen datasets do not always match diversity of the corre- and nonarboreal pollen taxa. For simplification of discussion, these sponding vegetation due to taxonomic complexity and morphological zones have been termed “forest phases” when arboreal pollen taxa specificity of pollination strategy. However, Felde, Peglar, Bjune, are dominant or “semiarid phases” when pollen from grasses and Grytnes, and Birks (2016) have observed that the comparison of nonarboreal taxa is more prevalent. diversity metrics from pollen samples from equivalent regional biome In the MAL05-1B/1C pollen stratigraphy, nine zones indicative types (i.e. forest to forest) rather than across nonequivalent types of forest are recorded (Figure 2). These are called F1–F9 in order of (forest to desert) are in fact comparable to better understand occurrence and are observed from 170 to 166 mblf (F1: 589 ka changes in diversity within a certain community type through time 4.82 kyr to 581 ka 7.34 kyr), 162–157 mblf (F2: 522 ka 11.1 kyr and can eliminate the effect of changes in overrepresented taxa. to 499 ka 11.8 kyr), 154–140.5 mblf (F3: 470 ka 9.83 kyr to Therefore, in this paper, we compare Dmg only across equivalent 274 ka 7.99 kyr), 135.5–126.5 mblf (F4: 241 ka 8.08 kyr to biome types, and use it only as a relative indicator of richness. 209 ka 5.35 kyr), 119.5–118 mblf (F5: 193 ka 6.51 kyr to 6 | IVORY ET AL.

189 ka 6.71 kyr), 110.5–94.5 mblf (F6: 165 ka 6.09 kyr to that four standard deviations are indicative of complete species turn- 142 ka 5.45 kyr), 88–81 mblf (F7: 137 ka 2.35 kyr over (Figure 3; ter Braak, 1985). An axis length of 2 in the MAL05- to 130 ka 2.28 kyr), 75.5–65.5 mblf (F8: 122 ka 4.50 kyr to 1B/1C biplot suggests turnover is large, but that 25% of taxa are 97 ka 8.16 kyr) and 34.5–10.5 mblf (F9: 80 ka 9.50 kyr to common to most samples (ter Braak, 1985). 21 ka 0.470 kyr). All forest phases are similar, with high percent- Axis 1 represents a structural gradient from more open landscapes ages of lowland forest, woodland and afromontane forest pollen. on the positive end to closed-canopy forest on the negative end of the The most characteristic forest taxon was the afromontane forest axis. This contrasts taxa frequently associated with semiarid biomes, conifer, Podocarpus. In addition to a high mean value (31%) among such as Commiphora africana-type (3.1), Amaranthaceae (1.8), Hymeno- forest samples, Podocarpus varied little once dominant (standard cardia (2.5) and Combretaceae undiff. (1.3), with those that typify the deviation <10%; minimum: 3.8%) and occasionally reached very high densest, moist forests at all elevations, such as Cliffortia nitida-type percentages (62%). Other afromontane taxa occurred in higher per- (2.2), Prunus africana-type (1.9) Anthospermum-type (2.0). In centages than in the semi-arid phases as well, such as Olea (mean = 5%; range = 0%–13%), Ericaceae (mean = 2%; range = 0%– 11%) and Myrica (mean = 1%; range = 0%–6%). Some taxa only (a) occurred during certain phases, such as Juniperus which was uncom- Miombo mon during the early phases but reached maximum values in F8 Afromontane Savanna

1.0 1.5 Steppe (mean = 1%; range = 0%–5%). Seasonal Forest The lowland arboreal assemblages, both tropical seasonal forest 0.5 + and miombo woodland, showed a much more complex pattern of +

occurrence during the forest phases. The only taxon that occurred 0.0 regularly in each phase was the miombo woodland tree Uapaca DCA2 Desert (mean = 2%; range = 0%–8%). Although tropical seasonal forest taxa, such as Moraceae (mean = 1%; range = 0%–6%), Macaranga-type Last 80 ka (mean = 1%; range = 0%–4%), Celtis (mean = 1%; range = 0%–18%) Modern sample Semi-arid phase sample and Trema-type orientalis (mean = 0.3%; range = 0%–6%), reached Forest phase sample relatively high percentages, these lowland forest pollen taxa occurred –1.5 –1.0 –0.5 –2 –1 0 1 2 only ephemerally. In contrast, the semiarid phases S1–S8 and are observed from DCA1 166 to 162 mblf (S1: 581 ka 7.34 kyr to 522 ka 11.1 kyr), + 157–154 mblf (S2: 499 ka 11.8 kyr to 470 ka 9.83 kyr), 140.5– (b) Hagenia abyssinica 3 – 135.5 mblf (S3: 274 ka 7.99 kyr to 241 ka 8.08 kyr), 126.5 Commiphora africana-type +

2 119.5 mblf (S4: 209 ka 5.35 kyr to 193 ka 6.51 kyr), Maesa-type lanceolata Urtica-type + + + 118–110.5 mblf (S5: 189 ka 6.71 kyr to 165 ka 6.09 kyr), Cliffortia nitida-type+ Syzygium-type +Artemisia Juniperus-type Moraceae +Trema-type orientalis 1 Ilex mitis + + + +Hymenocardia 94.5–88 mblf (S6: 142 ka 5.45 kyr to 137 ka 2.35 kyr), 81– Dodonaea viscosa +Macarangaacarangangaa-type-t + CombretaceaeC undiff. + CeltisC Anthospermum-type + + – + OleaO ea + BrachBBrachystegiachystegich i 75.5 mblf (S7: 130 ka 2.28 kyr to 122 ka 4.50 kyr), 65.5 0 + Prunus africana-type Uapacac DCA2 ++ Ericaceae undiff.ndnndifff. + 34.5 mblf (S8: 97 ka 8.16 kyr to 80 ka 9.50 kyr). These phases AlchornAlchornean Fernsrnsrn + were characterized by dominance of grasses and other nonarboreal + Podocarpusarpusarpuu + taxa (Figure 2). Grass accounted for on average 70% of the pollen PoaceaePoacePoacea + assemblage and never less than 29%. Other herbs occurred during Amaranthaceae = Palmae each semiarid phase, such as Amaranthaceae (mean 1%; –3 –2 –1 + range = 0%–8%). Similarly, dry woodland trees, such as Acacia (mean = 0.1%; range = 0%–1%) and Commiphora africana-type –4 –2 0 2 4 (mean = 0.1%; range = 0%–2%), were present, but did not become DCA1 frequent in occurrence until S3. FIGURE 3 (a) Detrended correspondence analysis (DCA) biplot A DCA was performed on all fossil samples, and the first two for all fossil and modern pollen samples. Dots and clouds are axes are shown in the biplot of Figure 3. Axis 1 has an eigenvalue of samples that are coloured green for forest samples, yellow for 0.1604 and an axis length of 2.0320 standard deviations. Axis 2 has semiarid samples, black for modern samples and the orange cloud an eigenvalue of 0.1025 and an axis length of 2.3553 standard devi- represents the area of the biplot occupied by samples from the last ations. However, the second axis of a DCA is strongly affected by 80 ka. Vectors represent the biome loadings of the pollen samples. (b) the same DCA biplot, except the scale is enlarged to illustrate the an arch effect created by the detrending process (ter Braak, 1985). positions of the pollen taxa with respect to the space filled by forest Because of this, we entirely focus on the inferences from DCA axis samples (green cloud) and semiarid samples (yellow cloud). Red 1. When used with ecological data, the first DCA axis provides a crosses are important pollen taxa that are shown in detail in the metric of species turnover amongst samples along that axis, such lower panel IVORY ET AL. | 7 addition, the position of the grasses (Poaceae) near the extreme posi- reduction of up to 61% of modern required to drive a 550 m tive end of axis 1 corroborates this interpretation. When biome vec- regression such as occurred from 80 to 107 ka. This decrease would tors are added to the biplot, this gradient in vegetation structure result in an average MAP of 372 mm/year in the basin instead of illustrates a transition from afromontane and lowland forest at the the modern average of 955 mm/year (Lyons et al., 2011). This value negative end, to woodland then savanna and desert biomes towards is consistent with the absence of charcoal during these intervals, the positive end of the axis. which suggests rainfall less than 400 mm/year results in low vegeta- Fossil samples show a strong clustering of forest samples near tion density, severely reducing fuel loads and decreasing wildfire negative values and semi-arid samples near positive values along axis activity (Cohen et al., 2007). 1. Forest phase samples cluster more tightly, suggesting that species However, in contrast to the hypothesis of grassland expansion in turnover amongst forest phases is low, regardless of when samples the Afrotropics during glacial periods and forest expansion during occurred in the temporal record, and thus there is little difference in interglacial periods, we observe at Lake Malawi that near modern species composition of forests through time. In contrast, the semi- lake levels and expansive forests were present during many of the arid phases are more scattered with respect to axis 1, representing a glacial periods in this record, suggesting some wet glacials (ex. mar- spread of ~2 standard deviations. This suggests that there is more ine oxygen isotope stage [MIS] 2, 4, 8; Figure 4). This implies that species turnover from sample to sample between semiarid phases the large environmental gradients observed over the late Quaternary than during the forest phases. In addition, it also suggests in the southern Africa cannot result from glacial-interglacial forcing that the majority of turnover in the entire record results from alone and must instead be linked to higher amplitude forcing. Ivory changes in species composition of the driest vegetation types et al. (2016) suggest a strong linkage between lake level responses through time. and the amplitude of insolation forcing resulted in the nonlinear response of the lake due to its half-graben morphometry, such that majority of its water volume (70%) is carried in the upper 200 m of 4 | DISCUSSION AND IMPLICATIONS the lake (Lyons et al., 2011). This results in a threshold-like response of lake levels to high amplitude changes in summer insolation when Over the last 600,000 years, we observe dramatic cycles of alternat- eccentricity is high. In contrast, when eccentricity is low and insola- ing forest and semi-arid vegetation within the Lake Malawi water- tion changes are small in amplitude, water levels remain persistently shed (Figure 2). Forest taxa dominated from 589–581 ka, 522– high, as was the case from 470 to 274 ka and has been the case for 499 ka, 470–274 ka, 241–209 ka, 193–189 ka, 165–142 ka, 137– the last 70 ka (Scholz et al., 2007). Interestingly, we observe here a 130 ka, 122–97 ka and 80–21 ka. As the watershed drains an extre- similar nonlinear response of forest extent to these nonlinear hydro- mely large area (~100,500 km2) and as anemophilous pollen may logical changes. Thus, highstand/lowstand transitions as well as for- even be sourced from outside of the watershed itself, vegetation est/semiarid transitions may occur coevally between two adjacent changes indicated by fossil pollen abundances integrate a signal from samples during our highest resolution intervals. Although other inter- a regional source area (DeBusk, 1997). Regional records such as this vals of this record have a much higher temporal resolution due to have the ability to capture landscape change representative of the slow sedimentation rates (e.g. ~400 ka) which do not allow us to highest amplitude of variability (i.e. regional biome change rather determine whether large-scale landscape transformation occurred, than stand-level change; Prentice, 1985). Thus, this suggests vegeta- the average time span between two samples over these higher reso- tion variability observed in the Lake Malawi drill core record during lution transitions is <250 years for complete transformation of the the late Quaternary in southeast Africa illustrates extremely large regional landscape. Currently, 80% of the rainfall within the water- environmental gradients through time with many transitions from shed is derived from moisture recycled off of the lake surface (Lyons dense tropical montane forests to desert or semiarid bushlands et al., 2011). It is likely that this hydrological forcing threshold is set throughout the region. by decreasing lake surface area as water level decreases, reducing Furthermore, these cycles of forest expansion are mirrored by potential moisture recycling and exacerbating decreases in regional increases in water level. This suggests a strong control of regional moisture fluxes. Thus we suggest that critical transitions in forest hydroclimate variability on vegetation structure and that desert bio- extent are related to strong feedbacks between vegetation and climatic conditions have occurred in the past at Lake Malawi during moisture recycling in the lake on a regional scale. extremely arid times (Figure 4). Lake level reconstructions on the Although vegetation structure varies between two extreme end drill cores based both on physical properties as well as bio- and members, we also observe alterations to vegetation composition minerogenic facies have pointed to repeated large regression events within a single end member through time. However, these alter- on the order of 200 to 550 m below the current water level coeval ations differ in character and timing between the forest and semiarid with semiarid phases (Figure 4; Lyons et al., 2015). Lyons, Kroll, and phases. Amongst the forest phases, the DCA suggests only minimal Scholz (2011), Lyons et al. (2015) and Stone, Westover, and Cohen compositional change until very recently (Figure 3). Afromontane (2011) also suggested that despite recorded aridity in the region at and tropical seasonal forest arboreal taxa dominate the pollen the LGM which resulted in a decline in lake level of ~100 m, these assemblages during F1–F8, suggesting widespread expansion into earlier arid periods were much more severe with precipitation the lowlands of taxa now restricted to the high elevations. Although 8 | IVORY ET AL.

PC1 (aquatic biological and mineralogical indicators) many afromontane trees, such as Podocarpus, are known to produce –0.2 large amounts of readily dispersed pollen grains, the percentages in 0 the Malawi record during forest phases are frequently greater than 0.2 40% (Figure 2). Pollen transport studies within extant afromontane 0.4 communities today have indicated that such percentages (>40%) are 0.6 Poaceae % only found within dense forests (Vincens, 1982). Although less con- 100 Podocarpus % sistently abundant, other tree taxa common in the afromontane for- 80 Not fire est, such as Olea and Juniperus, also attained abundances similar to 60 tolerant those found within modern forests (Olea:5%–10%; Juniperus:3%– 40 30%; Vincens, 1982). Thus, high abundances at Lake Malawi indicate 20 that forest phases F1–F8 from 600 to 80 ka were periods of exten- 0 sive colonization of the watershed down to the level of the lake 20 Olea % shore by montane forest. This suggests a lowering of the lower ele- 16 Fire tolerant vational limit of many of these taxa from ~1,500 to 2,000 m asl 12 today to below 500 m asl. 8 In contrast to pollen records and modelling studies which show a 4 strong influence of reduced atmospheric CO for lowering treelines 0 2 during the LGM in Africa (ex. Jolly & Haxeltine, 1997), we show that 6 Miombo % Fire tolerant these forest phases occurred during both interglacial (ex. MIS11) and 4 glacial conditions (ex. MIS6). This suggests, that beyond the LGM,

2 climatic factors other than the physiological CO2 effect had a much more powerful effect on treeline over long time scales (105 years). 0 6 Charcoal Influx Instead, we suggest that effective moisture may have been the (gr cm–2 year–1) strongest control on species’ distributions. Furthermore, this is not a 4 unique feature to the Malawi watershed. Comparison of the Lake 2 Malawi pollen record with those from marine core records covering 0 the last ~300 ka off of West and East Africa, we observe that the 8 Species 6 Richness expansion of afromontane taxa, such as Podocarpus, occurred at the 4 sub-continental scale (Figure 5; Dupont et al., 2011; Ning & Dupont, 2 1997). In fact, the timing of the most recent large-scale increase in 0 Podocarpus in all three records is coeval, lasting from about 115 to 0.6 Min. Squared 90 ka, the beginning of the last glacial period. Further, although the Chord Distance timing of previous forest dominated periods were not always con- 0.4 temporaneous or of the same duration, multiple earlier phases of 0.2 forest development overlapped significantly and occurred during 0 Marine Oxygen Isotope Stages periods of globally cooler temperatures (Figure 5). Thus, on numer- 12 3 4 56 7 8910 11 12 13 14 15 ous occasions in the past, currently elevationally restricted taxa have 0100 200300 400500 600 expanded into the lowlands and may have been regionally continu- Age (kyr BP) ous from the lake to each coast. The cause of forest expansion during glacial periods could be related to quasi-period warming events FIGURE 4 Hydrology, vegetation and ecological indicators from within the glacial, such as Dansgaard-Oeschger Events, however, drill core MAL05-1B/1C. From top to bottom, lake level reconstruction from principal components analysis (PC1) from Ivory previous studies of hydroclimate at Malawi suggest significant vari- et al. (2016) with relative water level indicated by the arrow ability of water level corresponding to changes in orbital precession (blue = high lake level; red = low lake level), fossil pollen percentages (Ivory et al., 2016). This could suggest that vegetation is responding for grass (Poaceae) and the fire intolerant tree Podocarpus, fossil pollen to similar forcing. percentages for two fire tolerant taxa (Olea and miombo woodland), Forest composition changes little throughout the record as sug- charcoal influxes, species richness calculated from Margalef’s Index gested by the clustering of forest phase samples on the DCA biplot and minimum squared chord distance (blue line is a five point smooth). The relative position of the Marine Oxygen Isotopes Stages of (Figure 3). Although some of the forest phase afromontane taxa vary Lisiecki and Raymo (2005) have been added. The green and yellow throughout the record, in particular, the increased frequency and bars represent forest or semiarid phases as indicated by the abundance of Juniperus only after 250 ka as well as co-occurrence CONISS cluster analysis, however, the last 80 ka has been of afromontane and tropical seasonal forest taxa (such as Celtis) from coloured orange to represent a critical transition in vegetation 320 to 350 ka as well as the last 100 ka, many of the main compo- composition nents (Podocarpus, Olea, Ericaceae) are consistently present and IVORY ET AL. | 9

Podocarpus % contrast, the fossil assemblages from the drill core plot almost com- pletely outside of the area occupied by the modern lake samples and Atlantic Ocean Lake Malawi Indian Ocean (GeoB-1016) (Limpopo River Delta) instead along a biome vector suggesting strong affinities to the 020406080 020406080020406080 0 afromontane forest biome (Figure 3). Furthermore, although afromontane forests do exist today within the basin, pollen samples taken from within these local communities show some dissimilarities to the fossil pollen samples. In particular, because of the degraded 100 nature of modern afromontane forest (Meadows, 1984), modern pollen assemblages typically have a larger contribution of grasses than observed in the fossil samples (DeBusk, 1997). Furthermore, how- 200 ever, the forest phases from 600 to 80 ka appear to have no modern analogue in Africa, as minimum squared chord distance is always >0.3 (Figure 4). This implies that although these ancient forests include many taxa indicative of afromontane communities today, 300 they also include a large admixture of taxa more typical of lowland environments today, such as Celtis and Uapaca. Age (kyr BP) The change in forest composition to a more miombo dominated 400 landscape suggests that 80 ka represents a critical transition in forests in the region. Beyond the change in species composition, addi- tionally, species richness, which was very high during the preceding 500 forest phases remains low for the remainder of the record despite dominance of forest taxa (Figure 4). In fact, although species richness values achieve similarly low values briefly during the semiarid phases before 250 ka, the last 80 ka represents the first time within 600 the last 600 ka when such low values were sustained for longer than FIGURE 5 Percentages of Podocarpus, a dominant afromontane ~10 kyrs, regardless of dominant vegetation type. tree, at three sites that straddle southern Africa. Records from the Modern vegetation in the Zambezian region, one of Africa’s lar- ° 0 ° 0 ° 0 Atlantic Ocean (11 46 S, 11 41 E) and Indian Ocean (26 10 S, gest phytogeographic regions (2.7 million km2; Frost, 1996), is floris- 34°010E) are based on marine cores from Ning and Dupont (1997; tically impoverished in comparison to tropical African forests such as GeoB-1016) and Dupont et al. (2011; Limpopo River Delta) in the Guineo-Congolian or afromontane regions (White, 1983). Many consider the degraded nature to be the result of long droughts abundant during forest phases (Figure 2). We suggest that the rela- and harsh conditions; however, studies have linked the presence of tive stasis of forest community composition results from the persis- and diversity within miombo woodlands to disturbance frequency, tence of small relict populations during arid periods. This mostly wildfire (Bond et al., 2005; Sankaran et al., 2005). Natural interpretation is supported by the observation of low abundances of fires depend on lightning as a sole ignition source, and thus most most forest tree taxa even during very arid periods (Figure 2). The frequently occur during the wet season. This results in lower inten- suggestion that forest taxa could have persisted as refugia in edaphi- sity fires which reduces tree mortality and promotes eventual con- cally moist areas, such as along streams, even when rainfall was version of miombo to mature forest (Frost, 1996). In contrast, quite low, means that when rainfall increased following an arid intense dry season fires halt forest regrowth by precluding tree taxa, phase, local populations existed to seed forests as they developed, such as Podocarpus, from frequently disturbed areas. Charcoal analy- rather than depending on long distance dispersal from external for- sis on the core suggests that influxes to the basin increased dramati- ests further afield. cally at this time (Figure 4). This suggests that the critical transition Despite the similarity amongst these ancient forests over much at 80 ka to lower regional diversity and a persistently miombo domi- of the last 600 ka, a comparison of the fossil pollen assemblages nated landscape is a relatively new occurrence and likely related to with modern surface samples from within the basin suggests that an increase in dry season fire. forest composition fundamentally altered after 80 ka (Figure 4). Pol- The afromontane forest archipelago has long interested biogeog- len transport studies by DeBusk (1997) have characterized typical raphers because of its remarkable compositional similarity despite pollen assemblages within the lake and watershed. Biomization of occupying disjunct populations separated by large distances (Hamil- the modern samples suggests that these assemblages have close ton, 1981; Livingstone, 1967; White, 1983). White (1983) remarked affinities with the miombo woodland vegetation which dominates that, despite their discontinuous ranges, individual populations of the watershed and thus plot along the miombo vector in the DCA forest are quite similar, with only a handful of dominant taxa, includ- biplot (Figure 3). This implies that pollen preserved in the lake sedi- ing: Podocarpus spp., Juniperus procera, Ericaceae, Prunus africana, ments are representative of watershed vegetation. However, in Olea africana and Ilex mitis. One possible explanation is that, 10 | IVORY ET AL. although barriers to dispersal exist today that isolate a single moun- Kenya (White, 1983). The presence of these assemblages outside of taintop, it is plausible that in the past, certain taxa may have occu- their modern range periodically in our record implies that they must pied continuous lowland corridors, allowing for greater dispersal. It have migrated into the region following forest collapse. This means has been suggested that this might have occurred during recent gla- that despite high turnover of vegetation during dry times, commu- cial periods, when cooler temperatures allowed cold-adapted mon- nity composition remained similar to plant communities that exist tane species to thrive in the lowlands (Hamilton, 1981; White, today, and that semi-arid vegetation re-established in the region 1981). However, this idea has been discounted as palaeoecological many times over the course of the record. studies covering the LGM observe that, although in some areas montane taxa do grow at lower elevations due to a tolerance of cooler 4.1 | Summary and implications temperatures such as around northern Lake Tanganyika; these cooler periods were also characterized by enhanced aridity around Lake The pollen record from Lake Malawi drill core MAL05-1B/1C cap- Malawi and in equatorial East Africa which resulted in an overall tures many landscape transitions over the last 600–20 ka. Vegeta- reduction of forest extent at all elevations (Bonnefille & Chalie, tion turnover from wet to arid periods is nearly complete at a 2000; Hamilton, 1981; Ivory & Russell, 2016; Livingstone, 1967; regional scale. On a first order, these phases track regional moisture Maley, 1992). Instead, we suggest that afromontane forest did once balance. During wet phases, dense, species-rich forests were com- occupy a continuous lowland corridor throughout the Malawi region, mon throughout the watershed, while severe reductions in rainfall which may account for recent mixing of populations and composi- with respect to modern resulted in a dramatic reorganization of veg- tional homogeneity. In contrast, instead of during cooler, drier peri- etation communities to semiarid assemblages composed of many ods, increases in afromontane forest pollen throughout East Africa taxa unknown in the basin today (DeBusk, 1997; White, 1983). seem to have occurred during periods of wetter conditions with little Variability in lake level and vegetation structure appear to have wildfire activity, suggesting that moisture deficit combined with dis- responded nonlinearly to local summer insolation forcing when the turbance is a strong limitation of forest extent in the lowlands. amplitude of insolation was high enough to drive large enough mois- Composition of the palynoflora of the semi-arid phases changes ture deficits to significantly lower lake level and subsequently dramatically through time (Figures 2 and 3). Thus, although forest decrease moisture recycling. This control appears much more com- composition remains stable through persistence of small relict popula- plex than a simple relationship to temperature, moisture or CO2 fluc- tions during dry times, this appears not to have been the case during tuations associated with glacial-interglacial cycles. During periods of the wet periods for the semiarid taxa. We observe that the taxa which high precipitation, dense forests with large contributions of afromon- characterize many of these phases (Commiphora africana-type, Acacia, tane trees were prevalent in the watershed, even periodically reach- Amaranthaceae) reach zero values once forest expansion was under- ing the lake shore. These forests had no modern analogue and were way (Figure 2). Biomization on these samples suggests that these much more diverse than vegetation communities around Malawi phases are represented by a wider array of biome types than the forest today. Furthermore, the stable species composition of these forests, phases (e.g. desert, steppe, savanna; Figure 3). Although the most suggests that such communities of species may persist for long peri- common are wooded grassland or savanna biomes, some samples even ods and be more resilient to change than would be expected. are classified as woodland or tropical seasonal forest because of the As modern lowland vegetation is dominated by low diversity persistence of edaphically supported tree communities. Some dryland miombo woodlands, these periodic expansions of moist species-rich vegetation, such as Commiphora spp. and Acacia spp., are insect polli- lowland and highland forest suggests very different ecological pro- nated and therefore do not readily disperse large quantities of pollen. cesses at work in these ancient forests than today. In particular, much Therefore the distinction of biome changes may be difficult if based denser heterogeneous tree cover with closed canopies indicates much on the presence or absence of these taxa alone. However, other taxa higher biomass on the landscape and dramatic alterations to distur- which help differentiate these biomes (such as Artemisia and Combre- bance regimes, particularly fires. In addition, this study implies that the taceae for steppe and savanna vs. Poaceae and Amaranthaceae for remarkable biogeographical similarity of modern afromontane popula- desert) produce much readily transported pollen, do not suffer from tions may be explained by lowland expansion at various times prior to that constraint, and thus support the interpretation that shifts in spe- 80 ka. This allowed many of these taxa to form continuous popula- cies composition occur often during these phases. This suggests that tions which decreased barriers to dispersal observed today. semiarid vegetation was extremely heterogeneous from phase to During arid periods, arboreal taxa decreased and forests collapsed. phase as well as within a single phase. Unlike the ancient forests, these ancient semiarid assemblages showed Furthermore, the semiarid phases are unique in that dissimilarity high species turnover from phase to phase. This high turnover sug- indices such as minimum squared chord distance suggest that these gests that semiarid taxa established in the watershed from outside of assemblages are compositionally similar to extant biomes despite the the region repeatedly and did not remain during wetter periods. Fur- presence of taxa which are uncommon in this area today. In fact, thermore, although these taxa are not found in the region today, semi- these assemblages have striking affinities to the Somali-Masai phyto- arid phase assemblages have modern analogues elsewhere in Africa, geographic region Acacia-Commiphora woodlands, an ecoregion com- and represent a wide variety of dryland biomes from woodland to mon today in very arid parts in the Horn of Africa in Ethiopia and desert. For example many of the taxa found during these times have IVORY ET AL. | 11 affinities with the Somali-Masai phytogeographic region found today Palaeoclimatology, Palaeoecology, 303,81–92. https://doi.org/10. in Ethiopia and Kenya (White, 1983). In addition, these semiarid com- 1016/j.palaeo.2010.01.025 Blois, J. L., Zarnetske, P. L., Fitzpatrick, M. C., & Finnegan, S. (2013). Cli- munities always include small percentages of tropical seasonal forests, mate change and the past, present, and future of biotic interactions. which likely persisted along streams. This suggests that during these Science, 341, 499–504. https://doi.org/10.1126/science.1237184 times that landscape was a complex mosaic of desert, steppe and Bond, W. J., Woodward, F. I., & Midgley, G. F. (2005). The global distri- savannah, interspersed with gallery forests that acted as refugia for bution of ecosystems in a world without fire. New Phytologist, 165, 525–538. moister taxa during times of nonoptimal climate. Bonnefille, R., & Chalie, F. (2000). Pollen-inferred precipitation time-ser- We suggest that these large-scale alterations of vegetation from ies from equatorial mountains, Africa, the last 40 kyr BP. Global and 600 to 80 ka as well as the change in structure, composition and Planetary Change, 26,25–50. https://doi.org/10.1016/S0921-8181 functioning after 80 ka may have had important implications for the (00)00032-1 Bonnefille, R., & Riollet, G. (1980). Pollens de savanes d’Afrique orientale development and dispersal of early humans. We suggest the dense (pp. 140). Paris, France: CNRS. forest was regional in scale and may even have extended from the ter Braak, C. J. (1985). Correspondence analysis of incidence and abun- Atlantic to Indian Ocean coasts numerous times during the late Qua- dance data: Properties in terms of a unimodal response model. Bio- ternary. As forest as well as extreme aridity may have acted as barri- metrics, 41, 859–873. https://doi.org/10.2307/2530959 ers to the dispersal, it is possible that hominin access to the region Cohen, A. S., Stone, J. R., Beuning, K. R., Park, L. E., Reinthal, P. N., Dett- man, D., ... Brown, E. T. (2007). Ecological consequences of early even as a corridor from southern to eastern Africa was punctuated Late Pleistocene megadroughts in tropical Africa. PNAS, 104, 16422– and limited to transitional periods. Lastly, this study points to a fun- 16427. https://doi.org/10.1073/pnas.0703873104 damental difference between the ancient forests that dominated the DeBusk, G. H. (1997). The distribution of pollen in the surface sediments region and modern arboreal assemblages. Ancient forests in the low- of Lake Malawi, Africa, and the transport of pollen in large lakes. Review of Palaeobotany and Palynology, 97, 123–153. https://doi.org/ land around the lake included many species that were long thought 10.1016/S0034-6667(96)00066-8 to be cool-adapted; however, we suggest that instead that these Dietl G. P., & Flessa K. W. (Eds.) (2017). Conservation paleobiology: Science species thrived in lowland areas until increases in wildfires following and practice. Chicago, IL: University of Chicago Press. a large arid episode at 80 ka. This change to more modern vegeta- Dietl, G. P., Kidwell, S. M., Brenner, M., Burney, D. A., Flessa, K. W., Jack- son, S. T., & Koch, P. L. (2015). Conservation paleobiology: Leverag- tion resulted in an increase in disturbance tolerant trees, such as ing knowledge of the past to inform conservation and restoration. ~ miombo woodland taxa, at 80 ka. Although the local archaeological Annual Review of Earth and Planetary Sciences, 43,79–103. https://d record at Malawi itself is sparse, making it difficult to make a direct oi.org/10.1146/annurev-earth-040610-133349 comparison to the timing of expansion at the regional scale, this tim- Doherty, R. M., Sitch, S., Smith, B., Lewis, S. L., & Thornton, P. K. (2010). Implications of future climate and atmospheric CO2 content for ing is consistent with evidence of increases in modern human popu- regional biogeochemistry, biogeography and ecosystem services lations (Thompson, Mackay, De Moor, & Gomani-Chindebvu, 2014). across East Africa. Global Change Biology, 16, 617–640. https://doi. It may be possible that fire use by early humans served as an impor- org/10.1111/j.1365-2486.2009.01997.x tant driver of landscape change after this time. Dupont, L., Caley, T., Kim, J., Castaneda, I., Malaize, B., & Giraudeau, J. (2011). Glacial-interglacial vegetation dynamics in South Eastern Africa coupled to sea surface temperature variations in the Western Indian Ocean. Climates of the Past, 7, 1209–1224. https://doi.org/10. ACKNOWLEDGEMENTS 5194/cp-7-1209-2011 We thank three anonymous reviewers and all participants in the Dupont, L., & Kuhlmann, H. (2017). Glacial-interglacial vegetation change in the Zambezi catchment. Quaternary Science Reviews, 155, 127– Lake Malawi Drilling Project, in particular Chris Scholz, Tom Johnson, 135. https://doi.org/10.1016/j.quascirev.2016.11.019 Robert Lyons and Eric Brown. Initial core processing and sampling Eccles, D. H. (1974). An outline of the physical limnology of Lake Malawi were conducted at LacCore at the University of Minnesota. Funding (Lake Nyasa). Limnology and Oceanography, 19, 730–742. https://doi. came from the US National Science Foundation–Earth Sciences org/10.4319/lo.1974.19.5.0730 Elenga, H., Peyron, O., Bonnefille, R., Jolly, D., Cheddadi, R., Guiot, J., ... (EAR-0602350), International Continental Scientific Drilling Program Hamilton, A. C. (2000). Pollen-based biome reconstruction for south- and a National Science Foundation Graduate Research Fellowship ern Europe and Africa 18,000 yr bp. Journal of Biogeography, 27, (2009078688). 621–634. https://doi.org/10.1046/j.1365-2699.2000.00430.x Faegri, K., & Iversen, J. (1989). Textbook of pollen analysis (4th edn by Faegri, K., Kaland, PE & Krzywinski, K.). Chichester, UK: John Wiley and Sons. ORCID Feeley, K. J., & Silman, M. R. (2011). The data void in modeling current and future distributions of tropical species. Global Change Biology, 17, Sarah J. Ivory http://orcid.org/0000-0003-4709-4406 626–630. https://doi.org/10.1111/j.1365-2486.2010.02239.x Felde, V. A., Peglar, S. M., Bjune, A. E., Grytnes, J., & Birks, H. J. (2016). Modern pollen–plant richness and diversity relationships exist along a REFERENCES vegetational gradient in southern Norway. The Holocene, 26, 163– 175. https://doi.org/10.1177/0959683615596843 Beuning, K. R., Zimmerman, K. A., Ivory, S. J., & Cohen, A. S. (2011). Frost, P. G. H. (1996). The ecology of miombo woodlands. In B. Cmpbell Vegetation response to glacial–interglacial climate variability near (Ed.), The miombo in transition: Woodlands and wellfare in Africa (pp. Lake Malawi in the southern African tropics. Palaeogeography, 11–57). Bogor, Indonesia: CIFOR. 12 | IVORY ET AL.

Fugere, V., Kasangaki, A., & Chapman, L. J. (2016). Land use changes in Lyons, R. P., Scholz, C. A., Cohen, A. S., King, J. W., Brown, E. T., Ivory, S. an afrotropical biodiversity hotspot affect stream alpha and beta J., ... Blome, M. W. (2015). Continuous 1.3-million-year record of diversity. Ecosphere, 7, e01355. https://doi.org/10.1002/ecs2.1355 East African hydroclimate, and implications for patterns of evolution Gill, J. L., Williams, J. W., Jackson, S. T., Lininger, K. B., & Robinson, G. S. and biodiversity. Proceedings of the National Academy of Sciences, (2009). Pleistocene megafaunal collapse, novel plant communities, 112, 15568–15573. and enhanced fire regimes in North America. Science, 326, 1100– Maley, J. (1970). Contributions a l’etude du Bassin tchadien Atlas de pol- 1103. https://doi.org/10.1126/science.1179504 lens du Tchad. Pollen et Spores, 40,29–48. Goring, S., Lacourse, T., Pellatt, M. G., & Mathewes, R. W. (2013). Pollen Maley, J. (1992). The African rain forest vegetation and palaeoenviron- assemblage richness does not reflect regional plant species richness: ments during late Quaternary. Climatic Change, 19,79–98. A cautionary tale. Journal of Ecology, 101, 1137–1145. https://doi. Margalef, D. R. (1958). Information theory in ecology. Society for General org/10.1111/1365-2745.12135 Systems Research, 3,36–71. Grimm, E. C. (1987). CONISS: A FORTRAN 77 program for stratigraphi- Meadows, M. (1984). Late Quaternary vegetation history of the Nyika cally constrained cluster analysis by the method of incremental sum Plateau, Malawi. Journal of Biogeography, 11, 209–222. https://doi. of squares. Computers & Geosciences, 13,13–35. https://doi.org/10. org/10.2307/2844640 1016/0098-3004(87)90022-7 Midgley, G. F., & Bond, W. J. (2015). Future of African terrestrial biodi- Grimm, E. (1990). TILIA 2.0. Springfield, IL: Illinois State Museum. versity and ecosystems under anthropogenic climate change. Nature Hamilton, A. C. (1981). The quaternary history of African forests: Its rele- Climate Change, 5, 823. https://doi.org/10.1038/nclimate2753 vance to conservation. African Journal of Ecology, 19,1–6. https://doi. Mittermeier, R. A., Turner, W. R., Larsen, F. W., Brooks, T. M., & Gascon, org/10.1111/j.1365-2028.1981.tb00647.x C. (2011). Global biodiversity conservation: The critical role of hot- Hughen, K. A., Eglinton, T. I., Xu, L., & Makou, M. (2004). Abrupt tropical spots. In F. E. Zachos & J. C. Habel (Eds.), Biodiversity hotspots (pp. 3– vegetation response to rapid climate changes. Science, 304, 1955– 22). Berlin, Heidelberg, Germany: Springer. https://doi.org/10.1007/ 1959. https://doi.org/10.1126/science.1092995 978-3-642-20992-5 Ivory, S. J., Blome, M. W., King, J. W., McGlue, M. M., Cole, J. E., & Newmark, W. D., & McNeally, P. B. (2018). Impact of habitat fragmenta- Cohen, A. S. (2016). Environmental change explains cichlid adaptive tion on the spatial structure of the Eastern Arc forests in East Africa: radiation at Lake Malawi over the past 1.2 million years. Proceedings Implications for biodiversity conservation. Biodiversity and Conserva- of the National Academy of Sciences, 113, 11895–11900. https://doi. tion, 27, 1387–1402. org/10.1073/pnas.1611028113 Niang, I., Ruppel, O., & Abdrabo, M. (2014). WGII, Chapter 22: Africa. Ivory, S. J., Lezine, A., Vincens, A., & Cohen, A. S. (2012). Effect of Geneva, Switzerland, IPCC, Fifth Assessment Report. aridity and rainfall seasonality on vegetation in the southern trop- Nicholson, S. E. (1996). A review of dynamics and climate variability in ics of East Africa during the Pleistocene/Holocene transition. Qua- Eastern Africa. In T. C. Johnson & E. O. Odada (Eds.), The limnology, ternary Research, 77,77–86. https://doi.org/10.1016/j.yqres.2011. climatology and paleoclimatology of the East African lakes (pp. 25–56). 11.005 Amsterdam, the Netherlands: Gordon and Breach. Ivory, S. J., & Russell, J. (2016). Climate, herbivory, and fire controls on Ning, S., & Dupont, L. M. (1997). Vegetation and climatic history of tropical African forest for the last 60ka. Quaternary Science Reviews, southwest Africa: A marine palynological record of the last 148, 101–114. https://doi.org/10.1016/j.quascirev.2016.07.015 300,000 years. Vegetation History and Archaeobotany, 6, 117–131. Izumi, K., & Lezine, A. (2016). Pollen-based biome reconstructions over https://doi.org/10.1007/BF01261959

the past 18,000 years and atmospheric CO2 impacts on vegetation in Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., equatorial mountains of Africa. Quaternary Science Reviews, 152,93– O’hara, R. B., ... Oksanen, M. J. (2013). Package ‘vegan’. Community 103. https://doi.org/10.1016/j.quascirev.2016.09.023 ecology package, version, 2, 10, 631–637.

Jolly, D., & Haxeltine, A. (1997). Effect of low glacial atmospheric CO2 Overpeck, J., Webb, T., & Prentice, I. (1985). Quantitative interpretation on tropical African montane vegetation. Science, 276, 786–788. of fossil pollen spectra: Dissimilarity coefficients and the method of https://doi.org/10.1126/science.276.5313.786 modern analogs. Quaternary Research, 23,87–108. https://doi.org/10. Jolly, D., Prentice, I. C., Bonnefille, R., Ballouche, A., Bengo, M., Brenac, 1016/0033-5894(85)90074-2 P., ... Edorh, T. (1998). Biome reconstruction from pollen and plant Palmer, G., Hill, J. K., Brereton, T. M., Brooks, D. R., Chapman, J. W., Fox, macrofossil data for Africa and the Arabian peninsula at 0 and R., ... Thomas, C. D. (2015). Individualistic sensitivities and exposure 6000 years. Journal of Biogeography, 25, 1007–1027. https://doi.org/ to climate change explain variation in species’ distribution and abun- 10.1046/j.1365-2699.1998.00238.x dance changes. Science Advances, 1, e1400220. https://doi.org/10. Lane, C. S., Chorn, B. T., & Johnson, T. C. (2013). Ash from the Toba 1126/sciadv.1400220 supereruption in Lake Malawi shows no volcanic winter in East Africa Prentice, I. C. (1985). Pollen representation, source area, and basin size: at 75 ka. Proceedings of the National Academy of Sciences of the Uni- Toward a unified theory of pollen analysis. Quaternary Research, 23, ted States of America, 110, 8025–8029. https://doi.org/10.1073/pna 76–86. https://doi.org/10.1016/0033-5894(85)90073-0 s.1301474110 Salzburger, W., Van Bocxlaer, B., & Cohen, A. S. (2014). Ecology and evo- Lindbladh, M., Fraver, S., Edvardsson, J., & Felton, A. (2013). Past forest lution of the African Great Lakes and their faunas. Annual Review of composition, structures and processes – How paleoecology can con- Ecology, Evolution, and Systematics, 45, 519–545. https://doi.org/10. tribute to forest conservation. Biological Conservation, 168, 116–127. 1146/annurev-ecolsys-120213-091804 https://doi.org/10.1016/j.biocon.2013.09.021 Salzmann, U., & Hoelzmann, P. (2005). The Dahomey Gap: An abrupt cli- Lisiecki, L. E., & Raymo, M. E. (2005). A Pliocene-Pleistocene stack of 57 matically induced rain forest fragmentation in West Africa during the globally distributed benthic d18O records. Paleoceanography, late Holocene. The Holocene, 15, 190–199. https://doi.org/10.1191/ 20(1). https://doi.org/10.1029/2004PA001071 0959683605hl799rp Livingstone, D. A. (1967). Postglacial vegetation of the Ruwenzori Moun- Sankaran, M., Hanan, N. P., Scholes, R. J., Ratnam, J., Augustine, D. J., tains in equatorial Africa. Ecological Monographs, 37,25–52. https://d Cade, B. S., ... Ardo, J. (2005). Determinants of woody cover in Afri- oi.org/10.2307/1948481 can savannas. Nature, 438, 846–849. https://doi.org/10.1038/nature Lyons, R. P., Kroll, C. N., & Scholz, C. A. (2011). An energy-balance 04070 hydrologic model for the Lake Malawi Rift Basin, East Africa. Palaeo- Scholz, C., Cohen, A., Johnson, T., King, J., Talbot, M., & Brown, E. geography, Palaeoclimatology, Palaeoecology, 75,83–97. (2011). Scientific drilling in the Great Rift Valley: The 2005 Lake IVORY ET AL. | 13

Malawi Scientific Drilling Project: An overview of the past to southern Tanzania. Review of Paleobotany and Palynology, 140, 145,000 years of climate variability in Southern Hemisphere East 187–212. https://doi.org/10.1016/j.revpalbo.2006.04.003 Africa. Palaeogeography, Palaeoclimatology, Palaeoecology, 303,3–19. Vincens, A., Garcin, Y., & Buchet, G. (2007). Influence of rainfall seasonal- https://doi.org/10.1016/j.palaeo.2010.10.030 ity on African lowland vegetation during the Late Quaternary: Pollen Scholz, C. A., Johnson, T. C., Cohen, A. S., King, J. W., Peck, J. A., Over- evidence from Lake Masoko, Tanzania. Journal of Biogeography, 34, peck, J. T., ... Lyons, R. P. (2007). East African megadroughts 1274–1288. https://doi.org/10.1111/j.1365-2699.2007.01698.x between 135 and 75 thousand years ago and bearing on early-mod- Vincens, A., Lezine, A., Buchet, G., Lewden, D., & Le Thomas, A. (2007). ern human origins. Proceedings of the National Academy of Sciences, African pollen database inventory of tree and shrub pollen types. 104, 16416–16421. https://doi.org/10.1073/pnas.0703874104 Review of Palaeobotany and Palynology, 145, 135–141. https://doi. Simpson, G. L., Oksanen, J., & Simpson, M. G. L. (2016). Package ‘ana- org/10.1016/j.revpalbo.2006.09.004 logue’. White, F. (1981). The history of the Afromontane archipelago and the Stewart, J. R. (2009). The evolutionary consequence of the individualistic scientific need for its conservation. African Journal of Ecology, 19,33– response to climate change. Journal of Evolutionary Biology, 22, 2363– 54. https://doi.org/10.1111/j.1365-2028.1981.tb00651.x 2375. https://doi.org/10.1111/j.1420-9101.2009.01859.x White, F. (1983). The vegetation of Africa. A descriptive memoir to accom- Stone, J. R., Westover, K. S., & Cohen, A. S. (2011). Late Pleistocene pany the UNESCO/AETFAT/UNSO vegetation map of Africa (pp. 356). paleohydrography and diatom paleoecology of the central basin of Paris, France: UNESCO. Lake Malawi, Africa. Palaeogeography, Palaeoclimatology, Palaeoecol- Williams, J. W., & Jackson, S. T. (2007). Novel climates, no-analog com- ogy, 303,51–70. https://doi.org/10.1016/j.palaeo.2010.01.012 munities, and ecological surprises. Frontiers in Ecology and the Environ- Team, R. C. (2017). R: A language and environment for statistical computing ment, 5, 475–482. https://doi.org/10.1890/070037 [Internet]. Vienna, Austria: R Foundation for Statistical Computing. Williams, J. W., Jackson, S. T., & Kutzbach, J. E. (2007). Projected distri- Thompson, J. C., Mackay, A., De Moor, V., & Gomani-Chindebvu, E. butions of novel and disappearing climates by 2100 AD. Proceedings (2014). Catchment survey in the Karonga district: A landscape-scale of the National Academy of Sciences, 104, 5738–5742. https://doi. analysis of provisioning and core reduction strategies during the Mid- org/10.1073/pnas.0606292104 dle Stone Age of northern Malawi. African Archaeological Review, 31, Willis, K. J., & Birks, H. J. B. (2006). What is natural? The need for a 447–478. https://doi.org/10.1007/s10437-014-9167-2 long-term perspective in biodiversity conservation. Science, 314, Tilman, D., Knops, J., Wedin, D., Reich, P., Ritchie, M., & Siemann, E. 1261–1265. https://doi.org/10.1126/science.1122667 (1997). The influence of functional diversity and composition on ecosystem processes. Science, 277, 1300–1302. https://doi.org/10. 1126/science.277.5330.1300 Trauth, M. H., Maslin, M. A., Deino, A. L., Strecker, M. R., Bergner, A. G., SUPPORTING INFORMATION & Dahnforth, M. (2007). High-and low-latitude forcing of Plio-Pleisto- Additional Supporting Information may be found online in the cene East African climate and human evolution. Journal of Human Evolution, 53, 475–486. https://doi.org/10.1016/j.jhevol.2006.12.009 supporting information tab for this article. Van Langevelde, F., Van De Vijver, C. A., Kumar, L., Van De Koppel, J., De Rid- der, N., Van Andel, J., ... Prins, H. H. (2003). Effects of fire and herbivory on the stability of savanna ecosystems. Ecology, 84, 337–350. https://doi. org/10.1890/0012-9658(2003)084[0337:EOFAHO]2.0.CO;2 How to cite this article: Ivory SJ, Lezine A-M, Vincens A, Vincens, A. (1982). Palynologie, environnements actuels et pliv-pleistocenes Cohen AS. Waxing and waning of forests: Late Quaternary al’est du lac Turkana (Kenya). PhD Dissertation, University of Aix- biogeography of southeast Africa. Glob Change Biol. Marseille. 2018;00:1–13. https://doi.org/10.1111/gcb.14150 Vincens, A., Bremond, L., Brewer, S., Buchet, G., & Dussouillez, P. (2006). Modern pollen-based biome reconstructions in East Africa expanded Graphical Abstract The contents of this page will be used as part of the graphical abstract of html only. It will not be published as part of main article.

Very little is known of African forest history to understand how forests may change in the future. This study presents 600,000 years of vegetation history from Lake Malawi showing massive reorganization of the vegetation due to changes in regional rainfall. We find that forest were widespread for much of the last 600,000 years, and that modern miombo woodland only emerged within the last 80 ka.