Genesis and development of an interfluvial peatland in the central Congo Basin since the Late Pleistocene

The central Congo Basin contains the largest known peatland complex in the tropics. Here we present a detailed multi-proxy record from a peat core, CEN-17.4, from the centre of a 45 km wide inter ﬂ uvial peatland (Ekolongouma), the ﬁ rst record of its kind from the central Congo peatlands. We use pollen, charcoal, sedimentological and geochemical data to reconstruct the site's history from the late Pleistocene to the present day. Peat began accumulating at the centre of the peatland ~19,600 cal BP (~17,500 e 20,400 cal BP, 95% con ﬁ dence interval), and between ~9500 (9430 e 9535 cal BP) and 10,500 (10,310 e 10,660 cal BP) cal BP towards the margins. Pollen data from the peatland centre show that an initial grass-and sedge-dominated vegetation, which burned frequently, was replaced by a Manilkara -type dominated ﬂ ooded forest at ~12,640 cal BP, replaced in turn by a more mixed swamp forest at ~9670 cal BP. Mixed swamp forest vegetation has persisted to the present day, with variations in composition and canopy openness likely caused at least in part by changes in palaeo-precipitation. Stable isotope data ( d D n-C29-v & icecorr ) indicate a large reduction in precipitation beginning ~5000 and peaking ~2000 cal BP, associated with the near-complete mineralization of several metres of previously accumulated peat and with a transition to a drier, more heliophilic swamp forest assemblage, likely with a more open canopy. Although the peatland and associated vegetation recovered from this perturbation, the strong response to this climatic event underlines the ecosystem's sensitivity to changes in precipitation. We ﬁ nd no conclusive evidence for anthropogenic activity in our record; charcoal is abundant only in the


a b s t r a c t
The central Congo Basin contains the largest known peatland complex in the tropics.Here we present a detailed multi-proxy record from a peat core, , from the centre of a 45 km wide interfluvial peatland (Ekolongouma), the first record of its kind from the central Congo peatlands.We use pollen, charcoal, sedimentological and geochemical data to reconstruct the site's history from the late Pleistocene to the present day.Peat began accumulating at the centre of the peatland ~19,600 cal BP (~17,500 e20,400 cal BP, 95% confidence interval), and between ~9500 (9430e9535 cal BP) and 10,500 (10,310 e10,660 cal BP) cal BP towards the margins.Pollen data from the peatland centre show that an initial grass-and sedge-dominated vegetation, which burned frequently, was replaced by a Manilkara-type dominated flooded forest at ~12,640 cal BP, replaced in turn by a more mixed swamp forest at ~9670 cal BP.Mixed swamp forest vegetation has persisted to the present day, with variations in composition and canopy openness likely caused at least in part by changes in palaeo-precipitation. Stable isotope data (dD n-C29-v&icecorr ) indicate a large reduction in precipitation beginning ~5000 and peaking ~2000 cal BP, associated with the near-complete mineralization of several metres of previously accumulated peat and with a transition to a drier, more heliophilic swamp forest assemblage, likely with a more open canopy.Although the peatland and associated vegetation recovered from this perturbation, the strong response to this climatic event underlines the ecosystem's sensitivity to changes in precipitation.We find no conclusive evidence for anthropogenic activity in our record; charcoal is abundant only in the Pleistocene part of the record and may reflect natural rather than anthropogenic fires.We conclude that autogenic succession and variation in the amount and seasonality of precipitation have been the most important drivers of ecological change in this peatland since the late Pleistocene.
© 2023 The Authors.Published by Elsevier Ltd.This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Introduction
The central Congo Basin (CCB), or Cuvette Centrale, spans parts of the Republic of Congo (ROC) and the Democratic Republic of Congo (DRC) (Fig. 1).This innermost part of the Congo Basin is dominated by forested wetlands (Bwangoy et al., 2010), large areas of which have recently been shown to have accumulated substantial thicknesses of peat (Dargie et al., 2017).The peatlands are estimated to occupy ~167,600 km 2 with a mean peat thickness of ~1.7 m and total carbon storage of ~29.0 Pg C, a substantial portion of the ~105 Pg C total peatland carbon storage in the tropics (Crezee et al., 2022).Dargie et al. (2017) provided the first evidence for the age of the CCB peatland complex, reporting basal radiocarbon dates from nine locations in the northern CCB, all but one of which fell between ~10,550 and ~7140 cal BP (the outlier was a date of ~2160 cal BP).Dargie et al. proposed that a regional increase in rainfall during the African Humid Period (~14,800e5500 cal BP;deMenocal et al., 2000;Schefub et al., 2005;Shanahan et al., 2015) may have facilitated waterlogging and hence peat initiation during the most humid part of the early Holocene.Dargie et al. (2017) encountered peat consistently under three vegetation types: hardwood swamp forest (including Uapaca mole Pax, Carapa procera DC. and Xylopia rubescens Oliv.), and two types of palm-dominated swamp forest, one dominated by Raphia laurentii De Wild., and one much rarer type dominated by R. hookeri G. Mann and H. Wendl., which occurs in abandoned river channels.Previous studies elsewhere in the tropics have suggested that there are consistent relationships between vegetation type and 1) peat thickness, 2) nutrient availability, and 3) hydrological regime (i.e.degree and duration of flooding; e.g.Anderson, 1983, Anderson andMuller, 1975;Page et al., 1999;Llampazo et al., 2022).Since the build-up of peat over time at a site affects all of these factors, their interaction can lead to the development of complex spatiotemporal patterning of vegetation and other ecosystem properties over decadal to millennial timescales (e.g.Phillips et al., 1997;Roucoux et al., 2013;Kelly et al., 2020).Palaeoecological studies are central to identifying the relationships between vegetation, peat thickness, nutrient availability, and hydrology, all of which remain to be explored in the CCB.
Previous work on one of CCB peatlands, Ekolongouma (also the focus of the present study), found that it has a shallow domed surface, implying a largely rain-fed system (Dargie et al., 2017;Davenport et al., 2020).Various scenarios could account for the development of shallow peat domes in the CCB, including terrestrialisation of abandoned river channels or paludification of waterlogged floodplains followed by a later transition to ombrotrophy, but these hypotheses remain to be tested.
The peatland ecosystems of the CCB appear to be largely intact at the present day, at least in terms of hydrology and forest canopy cover, but they are vulnerable to anthropogenic pressures (Dargie et al., 2019).The communities living in and around the peatlands use various non-timber forest resources including fish, game, honey and fruit.Populations are concentrated in riverside settlements, including Bantu people who typically have a subsistence economy based on fishing and farming, and Aka and Twa autochthon (native or indigenous) communities who pursue, at least in part, a hunter-gatherer economy in CCB forests and swamps (Campbell, 2005;Olivero et al., 2016;Hewlett, 2017).Trails and in some places (but not, as far as we know, at our study site) narrow canals have been constructed to facilitate access to the swamp forest (Blake, 1993;Lewis, 2002).Expanding road networks and new pressures from oil exploration, logging, and commercial plantations could potentially damage the hydrological and ecological integrity of the peatlands and, inadvertently, lead to the release of stored peat carbon (Dargie et al., 2019).An understanding of past human impacts on the peatlands will help to contextualize the apparently low level of anthropogenic impact observed today and the potentially greater impacts that could accompany socioeconomic development, if due care is not taken.
The CCB peatlands may also be vulnerable to 21 st century climate change.The preservation and accumulation of peat, especially in ombrotrophic peatlands, requires the soil water table to be close to the surface for most of the year (e.g.Moore and Bellamy, 1974; Page et al., 2006;Rydin et al., 2013;Cobb and Harvey, 2019).
Water tables can be maintained by abundant rainfall and an absence of prolonged droughts.However, mean annual rainfall has declined in the region since 1979 and the boreal summer dry season has become more intense (Cook et al., 2020).A shift to drier conditions could lead to lowering of the water table, resulting in aerobic decomposition and mineralization of the peat (Hooijer et al., 2010;Dohong et al., 2017), releasing stored carbon to the atmosphere.Additionally, drying can leave the peat surface vulnerable to fire and to erosion via runoff, washing organic material into adjacent river systems (cf.Hemingway et al., 2017).
Studies based on modelling and palaeoecological data from tropical peatlands outside Africa have suggested that future fluctuations in the intensity and pattern of rainfall could change the height and spatial distribution of water tables and increase the likelihood of peat surface drying (Hapsari et al., 2017;Leng et al., 2019).Palaeoclimatic data indicate that rainfall patterns in Central Africa have fluctuated throughout the Holocene (Schefub et al., 2005;Garcin et al., 2018); analysing the effects of these variations on past peat accumulation and other ecosystem properties will help to assess the potential impacts of future climate change.
To address the knowledge gaps identified above, here we present a detailed palaeoenvironmental proxy record from a peat core from the centre of the most intensively-studied peatland in the CCB, referred to here as the Ekolongouma peatland (informally named after an adjacent village; Fig. 1).The core, CEN-17.4,at 5.99 m, is the thickest peat sequence yet reported from the CCB.Here we use palynological and sedimentological data to describe and interpret the entire history of environmental change at the core site, from peat initiation to the development of the present-day domed, forested peat swamp.We use charcoal as a proxy for fire activity to assess the degree of past forest disturbance, and plant wax stable hydrogen isotopes as a proxy for precipitation to assess the influence of precipitation change on the peatland and vegetation over time.To place the CEN-17.4record in context with the wider Ekolongouma peatland, we use published radiocarbon dates from four additional peat cores and peat depth measurements from three transects (Fig. 2), yielding insights into the spatio-temporal pattern and underlying processes of peat accumulation in this part of the CCB peatland complex.

Aims
1. Assess the spatio-temporal pattern of peat initiation and its subsequent accumulation and identify the likely drivers of these patterns.2. Reconstruct past environments at the centre of the peatland from the late Pleistocene to the present day, particularly in terms of vegetation composition.3. Assess the evidence for past human impacts on the peatland ecosystem.4. Assess the effect of past climatic change on ecosystem properties at the site.

Site description
The CCB is thought to be underlain by 4e9 km of sediments ranging in age from Neoproterozoic to Neogene (Crosby et al., 2010;De Wit et al., 2015;Delvaux et al., 2021).The upper 1 km of these sediments are described as comprising Lower Cretaceous to Quaternary sediments (Kadima et al., 2011), likely deposited in fluvial, lacustrine and aeolian environments (De Wit et al., 2015).Dargie The Ekolongouma peatland, the focus of the present study, is located in Likouala Department, Republic of Congo.The peatland is ~45 km wide from east to west, occupying an interfluve ~300 m above sea level between the southwards-flowing black-water Likouala-aux-Herbes and white-water Ubangui rivers.Three study transects (ITG,CEN and EKG;Fig. 1) together form a nearcontinuous transect across the interfluve from Itanga village in the west to Ekolongouma village in the east.The peat across the interfluve is underlain by clays and silts (Dargie, 2015).Peat thickness reaches a maximum of 5.99 m at the site of , the main core studied here (Dargie, 2015).The peatland surface topography, reconstructed using LiDAR data, appears to increase in height from the peatland margin to the centre by ~1.8 m, indicating that the peatland is shallowly domed (Davenport et al., 2020).
Rainfall at the nearest meteorological station (Impfondo, ~1760 mm y À1; Samba et al., 2008) is low compared to some other tropical peatland regions (e.g.>3000 mm y À1 in Peruvian Amazonia; Marengo, 1998).There are two wet seasons each year, peaking in April/May and in October/November, and the intervening dry periods are short and rarely intense.The mean annual temperature is 25.6 C with little variation throughout the year (Samba et al., 2008).
The vegetation at the core site is dominated by palms and hardwood trees (Fig. 3) including R. laurentii, Carapa procera, Uapaca mole and Xylopia aethiopica (Dunel) A. Rich.; the understorey is composed of Aframomum sp., Myristicaceae sp., Melastomataceae sp., Palisota mannii C. B. Clarke., Pandanus candelabrum P. Beauv., and ferns (Dargie, 2015).The canopy at the coring site is lower and more open than along the rest of the CEN transect (Dargie et al., 2017).Two end-member vegetation types were described, hardwood swamp forest at the western end and R. laurentii-dominated palm swamp at the eastern end of the CEN transect, with gradual variation in the relative proportions of hardwood trees and palms in between (Dargie, 2015).

Peat sampling and peat depth
'Peat' is defined here as material with loss-on-ignition (LOI) > 65%, following Dargie et al. (2017).The peat core analysed in this study was retrieved in March 2014 at 01.1835 N, 017.6399E. The core was taken with a 5 cm diameter, 50 cm long Russian-type corer in 13 overlapping sections in two parallel holes to a depth of 6.29 m, coring past the base of the peat (5.99 m) into the underlying clays which prevented further sampling.Each section was wrapped, transported under licence to the UK and refrigerated at 4 C.The entire core was sub-sampled in the laboratory at 4 cm resolution for isotopic analysis and up to 2 cm resolution (including analytical intervals of 16, 8 and 4 cm) for palaeoecological analysis.Sampling for testate amoebae and plant macrofossils was carried out but preservation was too poor to allow analysis.Where necessary, depths were corrected using linear interpolation to account for up to 5 cm of compaction during transportation from the core site.Additional peat cores from the Itanga and Ekolongouma transects were collected and studied by Dargie (2015).Peat thickness was measured at 250 m intervals along the transects by probing the peat with a pole (Dargie, 2015).

Radiocarbon dating and age model construction
Twenty-two bulk peat samples from core CEN-17.4,sieved at 150 mm to remove roots as far as possible, were dated by radiocarbon AMS analysis at the Alfred Wegener Institute and first published by Garcin et al. (2022).Dates from four other cores (ITGkm6, EKGkm9, EKGkm7-2015, EKGkm4) along the Ekolongouma and Itanga transects were previously published by Dargie et al. (2017).All dates were recalibrated for this analysis using a 50:50 mix of the IntCal20 and SHCal20 calibration curves (Hogg et al., 2020) in the IntCal package in R v. 4.0.3(R Core Team, 2021); for brevity, ages cited in the text are median calibrated ages unless otherwise stated.Radiocarbon dates are presented in full in Appendix A. The age model for CEN-17.4 was constructed using the R package rBacon v.2.5.8 (Blaauw and Christen, 2011), setting the priors to allow for low memory and a wide range of accumulation rates (acc.mean¼ 40, acc.shape ¼ 1, mem.mean ¼ 0.1, ssize ¼ 15,000, mem.strength ¼ 2).The resulting age model places little weight on three samples (at 396, 596 and 625 cm) that returned ages substantially younger than adjacent dates.This is a common occurrence, particularly in the dating of slow accumulating basal peats and can be due to a number of factors e.g.downward penetration of roots and/or contamination during coring (Quik et al., 2022).

Palynology
Pollen preparation followed conventional methods (Moore et al., 1991), substituting density separation (Nakagawa et al., 1998) for HF treatment.Two Lycopodium tablets (Lund University  Spores were excluded from the pollen sum (and calculated as a percentage outside of the pollen sum), but the sum does include unidentified morphotypes, which make up less than 5% of the assemblages.Palaeoecological interpretations are based primarily on ecological information from the African Plant Database v. 3.4.0and on vegetation census data from the Ekolongouma site, supplemented by information in the palaeoecological and ecological literatures (e.g.Elenga et al., 2000;Brncic et al., 2007;Kiahtipes et al., 2011;Moutsambot e, 2012;Ifo et al., 2016: Bocko et al., 2016); see section 5.2 for further discussion.

Palynological data analysis
Pollen assemblage zones were defined using stratigraphically constrained cluster analysis, a Bray-Curtis dissimilarity matrix and the CONISS method, including only taxa achieving >5% abundance in at least one sample (Birks and Gordon, 1985).The choice of number of zones (five) was informed by a broken stick model (Bennett, 1996) and visual examination of key vegetation transitions in the pollen data.Indicator species analysis (Dufrêne and Legendre, 1997) was performed to determine which taxa were characteristic of each zone, referred to here as indicator taxa.As an aid to discussion, additional subzones were defined using the cluster analysis as a guide; one additional subzone boundary (at 180 cm) was added to facilitate comparison with the other proxy data.To further inform our interpretation of the structure of the dataset, non-metric multidimensional scaling (NMDS) ordination (Legendre and Birks, 2012) was carried out on the same dissimilarity matrix.In order to highlight associations between individual taxa throughout the dataset, we applied a correlated topic model (CTM; Blei and Lafferty, 2007), based on taxon abundances, to the same subset of pollen data.This method, only recently applied in palaeoecology, identifies sets of taxon associations within a dataset, allowing one taxon to characterise more than one association, which allows for the overlapping species associations found in nature.Finally, rarefied richness was used to estimate palynological diversity.These analyses were carried out in R using the labdsv (v.2.0e1, Roberts, 2019), rioja (v. 0.9e26, Juggins, 2020), stm (v. 1.3.6, Roberts et al., 2019) and vegan (v. 2.6e2, Oksanen et al., 2020) packages.

Sedimentology
LOI analysis was carried out on 1 cm 3 sub-samples at 550 C for 4 h following Heiri et al. (2001).Particle size analysis was carried out on the lowermost peat and underlying sediments, below 550 cm, using a Beckman Coulter LS230 Particle Size Analyser (Goossens, 2008) on 1 cm 3 sub-samples which had been digested in 30% H 2 O 2 then 10% HCl to remove organic matter and carbonates.

Organic geochemistry
Total organic carbon (TOC), total nitrogen (TN), and d 13 C TOC were measured on bulk samples, and stable isotope ratios (D:H and 13 C: 12 C) were measured on C 29 -alkanes extracted from the peat.
dD n-C29 values (Garcin et al., 2022) reflect variations in the isotopic composition of precipitation representing changes in past precipitation regimes, i.e. the amount and seasonality of precipitation (Sachse et al., 2012;Ladd et al., 2021).dD n-C29 is negatively correlated with precipitation amount in Central Africa (Collins et al., 2013;Garcin et al., 2018).As these isotopic values can be influenced by vegetation type (Sachse et al., 2012) and global ice volume (Waelbroeck et al., 2002), we corrected for both vegetation and ice volume here (not applied in Garcin et al., 2022), i.e. dD n-C29-v&icecorr (Collins et al., 2013;Feakins, 2013;Shanahan et al., 2015;Tierney et al., 2017;Garcin et al., 2018).The organic geochemistry data were further compared with the previously acquired I-index, which is derived from Rock-Eval® analysis (Garcin et al., 2022).The I-index relates to the ratio between the thermally labile and resistant pools of organic matter (Sebag et al., 2016).Further methodological details are given in Appendix B.

Inorganic geochemistry
The concentrations of major elements were analysed by Inductively Coupled Plasma Mass Spectrometry (ICP-MS).Peat samples were dried at 65 C, milled, and digested in HNO 3 then H 2 O 2 before analysis.Further methodological details are given in Appendix B.

Microscopic and macroscopic charcoal
Microscopic charcoal fragments (>10 mm) (representing regional fire activity) were quantified at the time of pollen counting using the pollen-slide method (Clark, 1988).Macroscopic charcoal fragments (>125 mm) (representing local fire activity) were extracted from 1 cm 3 subsamples by sieving, bleaching, and disaggregation using 40 ml 15% Na 6 P 6 O 18 to disaggregate organic matter, following Hawthorne et al. (2018).Macroscopic charcoal fragments were counted and measured using digital image analysis (ImageJ software: Abr amoff et al., 2004).The macroscopic records of charcoal area vs number were compared, confirming that particle breakage was not an issue during processing.

Peat thickness, basal peat age and long-term net accumulation rate
The maximum peat thickness across all transects, 5.99 m, was recorded at the site of core CEN-17.4(Fig. 1), close to the centre of the interfluve.Peat thickness decreases towards the margins of the basin (Fig. 2).Small-scale variations in peat thickness along each transect likely reflect both surface microtopographic variation (undulations <1 m were observed in the field) and unevenness in the underlying substrate.
Our age-depth model (Fig. 4) yields central age estimates for the base of CEN-17.4 at ~20,400 cal BP (~22,000e18,050 cal BP, 95% confidence interval), with peat initiation (at 599 cm) estimated at ~19,600 cal BP (~20,400e17,500 cal BP, 95% confidence interval).These age estimates should be treated with caution because two of the dates near the base (at 596 and 625 cm) are out of stratigraphic sequence (younger than surrounding dates).It would be possible, though less parsimonious, to construct an age model that placed more weight on the dates at 596 and 625 cm, which would give a younger date for peat initiation, closer to ~17,500 cal BP; or alternatively to yield a slightly older estimate for peat initiation by placing more weight on the date at 599 cm, which calibrates to ~20,100e20,500 cal BP (95% confidence interval).
Peat initiation dates from the other cores at the site are considerably younger, dating to the early Holocene and showing a trend of decreasing age from the centre to the edge of the site, with median ages of 10,525 cal BP at EKGkm9, 10,230 cal BP at EKGkm7-2015, and 9490 cal BP at EKGkm4, to the east of ; and 9530 cal BP at ITGkm6, to the west (Fig. 2).
The long-term net peat accumulation rate at CEN-17.4 is 0.31 mm y À1 .Net long-term peat accumulation rates for cores EKGkm9, EKGkm7-2015, EKGkm4 and ITGkm6 are 0.26, 0.23, 0.16 and 0.20 mm yr À1 respectively, declining systematically from the centre towards the edge of the site.These figures, however, mask considerable stratigraphic variation evident in the shape of the agedepth models (Figs. 2 and 4).Age-depth curves may not represent the net accumulation rate that occurred at the time of peat formation because of subsequent decomposition (Young et al., 2021), and additional proxy evidence (discussed below) is needed to fully understand how peat carbon storage has changed over time.However, an important feature of  is the substantial change in the age-depth relationship between 190 and 151 cm (~7500e2000 cal BP).The depth/age gradient here is 5e8 times lower than in the rest of the sequence.This stratigraphic feature was termed the 'Ghost Interval' by Garcin et al. (2022), who interpreted it as representing a section of peat that has largely been lost due to enhanced contemporaneous and secondary decomposition during a dry interval from ~5000 to 2000 cal BP.The age-depth models for additional cores from Ekolongouma with multiple dates all show a similar feature in the mid-Holocene, though less well defined perhaps due simply, to the lower density of dates (Fig. 2).

Palaeoenvironmental proxy data from CEN-17.4
The palynological analysis of CEN-17.4 yielded 155 taxa of which 24 are unidentified morphotypes, never totalling more than 5% of the assemblages.Taxa achieving >5% abundance in at least one sample are plotted in Fig. 5. Pollen preservation was generally excellent throughout the core.The pollen assemblage zones provide a convenient stratigraphic framework within which to describe the other proxy data.The proxy data are presented in detail in Table 1.For brevity, in the text below we describe only the most important features of the palaeo-proxy record.
The 629 cm-long sequence consists of fine clayey silts below 599 cm, with a sharp transition to peats at this depth; LOI and TOC data (Fig. 8) indicate that the ash content of the peat continues to decline slowly throughout the remainder of Zones I and II.Palynologically, several taxa are present throughout the entire record, including Pycnanthus, Macaranga, Syzygium, Lophira, Raphia and Pandanus.Overlying this picture of continuity are some systematic long-term changes, notably a gradual increase in the abundance of Raphia, Pandanus and Symphonia globulifera over time, reflected in the general trend of the samples from positive to negative NMDS axis 1 scores (Fig. 6).There is also a lot of high-frequency variation: in detail, the relative abundances of these taxa rarely remain constant for long.Zones I and II are defined by distinctive palynological assemblages, but Zones III to V vary mainly in the relative abundances of a consistent set of taxa.
Zone I (20,400e12,650 cal BP) extends from the base of the sequence to its upper boundary at 12,650 cal BP.The palynology of this zone is dominated by herbs, notably Cyperaceae and Poaceae, and the aquatic Alisma-type.Zone I is divided into five subzones which reflect an oscillation between a dominance of herbs (Subzones Ia, c, e) and trees including Pycnanthus, Macaranga, Syzygium, Lophira, Raphia and Pandanus (Subzones Ib, d).Micro-and macrocharcoal are abundant throughout Zone I, covarying with Cyperaceae and Poaceae.No diatoms or other siliceous microfossils were observed in the basal sediments (below 599 cm, within Subzone Ia), which are unlaminated Al-rich clayey silts with a median particle size ranging from 4 to 6 mm; Alisma-type is also absent from these basal sediments.The organic geochemistry of the peat in Zone I (above 599 cm) is variable with generally very positive d 13 C Bulk values and low C/N, markedly different from the peats in Zones IIeV.The inorganic geochemistry (Fig. 9) of this Zone I peat is also distinct from the peat in other zones, being proportionately rich in Al, Ti, Pb, P and Cu especially in Subzone Ic where the peat LOI values show a marked dip.
Zone II (12,650e9680 cal BP) is distinguished palynologically by an increase in Manilkara-type and several other hardwood tree taxa which are only abundant in this zone (Khaya, Carapa-type    The NMDS analysis (Fig. 6) highlights an overall trend in the dataset with decreasing Axis 1 scores moving up-core.The NMDS analysis also emphasizes the overlap between assemblages in Zones IIIeV, and shows Zone IV as being moderately distinctive (positive Axis 2 scores).The CTM analysis provides a different, complementary perspective on the structure of the pollen data which may better reflect ecological realities (Fig. 7).It identifies five 'topics' (groups of covarying taxa) which may be interpretable as representing consistent species associations in the ecosystem (see discussion below).The first of these topics is characterised by Cyperaceae, Poaceae and Alisma-type, already noted as indicator Fig. 6.Non-metric multidimensional scaling (NMDS) ordination of pollen data (>5% abundance).Symbols and colours represent the different pollen assemblage zones identified via CONISS analysis.

Inorganic geochemistry:
The clayey silts below 599 cm have a characteristic signature with high concentrations of Al accompanied by a suite of rarer elements (Pb, Cs, U, Rb, Li, V, Cu, Ag, Cr).These all become less abundant above 599 cm and throughout Subzone Ib, but almost all of them increase again in the middle of Subzone Ic, coinciding with the drop in LOI to 79.2%.There are slight differences in the signatures of these events (e.g. less Li and K, more Ni and As in Subzone Ic).In the peats above 599 cm, Al and Fe are proportionally the most important elements, Fe tending to increase proportionally at the expense of Al, with Ti and Mg also consistently abundant.

Peat initiation and peatland development
Our age model indicates that peat initiation began at the core site of  sometime between 20,400 and 17,500 cal BP, close to the centre of what would at the time have been a large, shallow depression (~45 km across with 3e4 m of elevation change along our eastewest transect line; Davenport et al., 2020).The presentday landscape has a gentle northesouth slope e the Ubangui and Likouala-aux-Herbes rivers which bound the site to the east and west, both flow southwards, falling only ~15 m over ~275 km e but in the absence of peat thickness data along a northesouth transect we cannot say whether the site was a shallow valley or a closed basin before the peatland developed.At the time of peat initiation, fine-grained mineral sediment was being delivered to the site, most likely by low-energy overbank flooding by sediment-laden waters.The resulting impermeable substrate would have helped to sustain the waterlogged conditions necessary for peat initiation.High concentrations of elements including Al and Ti (Fig. 9) suggest that these sediments were derived from highly weathered soils.Around ~19,600 cal BP (599 cm) the material crossed the 65% LOI threshold that we use to define 'peat', but the fact that LOI continues to increase slowly upwards, associated with a decline in the concentration of lithophilic elements including Al and Ti, suggests that the site was repeatedly flooded (seasonally or more episodically) until at least 10,450 cal BP (top of Subzone IIa, when LOI values first exceed 95%).
Whether peat initiation at the site of CEN-17.4 was driven by climatic change or some other factor, such as a geomorphological change (e.g.drainage blockage) or an autogenic process such as paludification, is unclear from our data.Dargie et al. (2017) suggested that peat initiation in the CCB might largely have been due to a shift to a wetter climate at the start of the Holocene, as the peat initiation dates then available were almost all from the early Holocene.The dating of CEN-17.4 demonstrates that peat was in fact accumulating long before the start of the Holocene, at least in places.Palaeoclimatic records from the region and climate modelling studies (Hamilton and Taylor, 1991;Schefub et al., 2005;Weijers et al., 2007;Liu et al., 2009;He et al., 2013) indicate that the climate of equatorial Africa between ~19,000 and ~17,000 cal BP was drier and cooler than today, with the driest/coldest conditions around the time of the deposition of Heinrich 1 (H1) in the north Atlantic, ~17,000e16,000 cal BP (e.g.Weijers et al., 2007;Bayon et al., 2012, cf. Andrews andVoelker, 2018).Temperatures began to increase gradually from ~17,000 and levelled off around 8000 cal  BP (Weijers et al., 2007).Precipitation is thought to have increased from ~17,000 cal BP until between ~12,500 and 11,500 cal BP, when a marked decline is reflected in dD n-C29-icecorr (Schefub et al., 2005) and Al/K (Bayon et al., 2012) records from the Congo Fan.Outside the Congo Basin itself, evidence for low lake levels has been observed at the same time at several sites (Roberts et al., 1993;Gasse, 2000;Stager et al., 2002;Peck et al., 2004).Precipitation increased again at the start of the Holocene, with peak climatic wetness occurring in many records during the early to mid-Holocene (~10,000e6000 cal BP: Schefub et al., 2005;Bayon et al., 2012;Shanahan et al., 2015; see discussion below and Fig. 10).
The dD n-C29-v&icecorr data from CEN-17.4, the most direct line of evidence for palaeoprecipitation available for our site, is broadly consistent with these other records in indicating that the climate during the Pleistocene was generally much drier than during the earlyemid Holocene.However, the data indicate a small, shortlived increase in inferred precipitation at ~19,750 cal BP (603 cm), just before peat initiation.The clayey substrate, the continued importance of river flooding (evidenced by mineral material in the peat and sediment), and the site's topographic location at a low point on the interfluve, would all have tended to favour waterlogging at that time, so that even a slight increase in rainfall might have been sufficient to initiate peat formation.Low atmospheric CO 2 concentrations (Gasse, 2000) and low atmospheric water vapour (Bush and Philander, 1998) might be expected to have a negative effect on plant water use efficiency and hence litter production under a dry climate, but the pollen and d 13 C bulk data from Zone I indicate that C 4 plants (which photosynthesize more efficiently than C 3 plants under low atmospheric CO 2 ) were the dominant contributors of litter to the peat.
By ~12,100 cal BP (459 cm), well over a metre of peat had accumulated (some may have subsequently been lost by decomposition) at the site of CEN-17.4.The age-depth curve then steepens, coincident with warmer and wetter climatic conditions evidenced by dD n-C29-v&icecorr and dD n-C29-icecorr (Garcin et al. (2022);Schefub et al., 2005), reconstructed MAT (Weijers et al., 2007), and sediment Al/K ratios (Bayon et al., 2012) and with an increase in lignin-rich C 3 vegetation (evidenced by the palynological evidence for replacement of herbs by trees, and the decrease in and by an increase in light-demanding and pioneer taxa, as water tables likely deepened.This change in forest structure and composition (discussed further below) involved a shift to taxa tolerant of the increasingly dry conditions.The I-index from Rock-Eval pyrolysis (Fig. 8) indicates intense decomposition of peat during this interval, associated with higher TOC values, indicating that the peat is dominantly composed of highly refractory organic matter (Garcin et al., 2022).The C/N ratio reaches a substantial peak at 164 cm, indicating preferential loss of nitrogen through aerobic decomposition of peats formed just before the driest interval.A similar peak in d 13 C bulk likewise suggests preferential loss of labile organic matter.Garcin et al. (2022) estimate that up to 4 m of peat may have been lost to secondary decomposition during this interval.
Above 151 cm (~2000 cal BP) the gradient of the age-depth curve increases, suggesting that peat accumulation resumed.Hydrophilic swamp forest taxa (Pandanus and Pycnanthus) gradually recover, suggesting that soil moisture stabilized or increased, despite dD n-C29-v&icecorr values remaining high and stable until ~900 cal BP.This may be due to a decrease in the strength of the dry season, which could result in higher water tables with no detectable change in dD n-C29-v&icecorr values.
Taken together, the evidence presented here from CEN-17.4,along with the independent palaeoclimatic data from the Congo Fan, suggests that climatic change has been important in controlling peat accumulation and decomposition in the Ekolongouma peatland.Late Glacial and early Holocene patterns of peat accumulation have yet to be studied at other peatlands within the CCB.However, the age-depth models from Lokolama and Bondamba presented by Garcin et al. (2022) suggest that a 'Ghost Interval' may be a widespread, perhaps ubiquitous feature of the CCB peatland stratigraphies, reflecting a severe drying event culminating around 2000 cal BP.This conclusion is consistent with a wide body of evidence from West and Central Africa which indicates increased aridity after ~4000 cal BP (Shanahan et al., 2015;Collins et al., 2017;Garcin et al., 2018).In particular, our record is relevant to understanding the Late Holocene Rainforest Crisis (LHRC), a widespread disturbance of rainforest vegetation between ~3000 and ~2000 cal BP identified at a number of locations in central Africa (Ngomanda et al., 2009;Maley et al., 2018).The exact timing, extent and causes of this event have been debated (Vincens et al., 1999;Giresse et al., 2020;Bayon et al., 2019;Garcin et al., 2018), but the evidence presented here and by Garcin et al. (2022) points to climatic drying being an important influence on the CCB peatlands at this time.
Despite the apparent substantial loss of peat represented by the 'Ghost Interval', the long-term net peat accumulation rate at CEN-17.4 is the highest recorded in the Ekolongouma peatland, with a value of 0.31 mm y À1 .This value is nonetheless at the low end of the ranges reported from domed peatlands elsewhere in the tropics (e.g.typically 0.54 mm y À1 in Central Kalimantan, 1.77 mm y À1 in Sumatra and 1.89 mm y À1 in East Kalimantan (Dommain et al., 2011); 1.27 mm y À1 at San Jorge (Kelly et al., 2017) and 1.88 mm y À1 at Aucayacu (Swindles et al., 2018), both in Peru; and 2.98 mm y À1 at San San Pond Sak, Panama (Phillips et al., 1997).
The spatial pattern of basal peat ages suggests that the nascent peatland in the Ekolongouma interfluve expanded laterally from the centre towards the margins sometime between ~19,600 and ~9500 cal BP (central estimates).The Pleistocene peatland at CEN-17.4 may have been part of a single continuous body of peat, or alternatively, just one of many small, separate peat bodies which occupied the deepest, wettest depressions in the interfluve.This pattern of peatland development is known from lowland tropical peatlands elsewhere, e.g. in Madre de Dios (Householder et al., 2012) and Quistococha (Kelly et al., 2020), Peru, where small pockets of peat formed in swales and pools, eventually infilling and coalescing over time to form the present day peatlands.
Long-term net peat accumulation rates decrease systematically in cores towards the margins of our site (Fig. 2), where peat thicknesses of 1.2e2.9m are associated with basal peat ages of between ~9500 and ~10,500 cal BP.The lateral expansion of peat from the centre to the interfluvial margins may have been either a steady or an episodic process; the evidence currently available does not allow us to distinguish definitively between these possibilities given the large distances between dated cores, but the cluster of early Holocene dates suggests a burst of peat initiation at that time, perhaps related to wetter and warmer climates.

Interpreting the CEN-17.4 pollen record
Here we discuss our overall interpretation of this first pollen record from the CCB peatlands.The principles discussed here underpin the more detailed zone-by-zone palaeoenvironmental interpretations in Section 5.3.
Little is known specifically about the pollen source area and taphonomy of central African swamp forest taxa (Julier et al., 2021;Henga-Botsikabobe et al., 2020;Bengo et al., 2020), but many well-established principles can be assumed to apply to our record (e.g.Tauber, 1967;Prentice, 1985;Jackson and Lyford, 1999).The Zone I sediments and peats are likely to have a wider pollen source area than the peats in Zones IIeV given the geochemical and sedimentological evidence for repeated flooding of the site by sedimentladen waters, and the dominance of herbs over trees at that time (evidenced by d 13 C bulk as well as by the pollen data).The background presence of tree taxa in this zone may reflect pollen introduced in floodwaters rather than local production.As peat swamp forest became established, the pollen source area likely decreased as river flooding ceased and the forest canopy thickened, reducing wind speeds (Andersen, 1974;Whitney et al., 2019), and numerically important taxa in the pollen assemblages were likely to be growing at or very close to the sampling location.Some of our taxa are known to produce abundant pollen grains (e.g.Uapaca, Alchornea, Macaranga and Celtis; Lebamba et al., 2009;Tovar et al., 2019) and others not to (e.g.Gilbertiodendron: Tovar et al., 2019), but pollen productivity remains unknown for most taxa.
The integrity of the peat sequence as a palaeoenvironmental archive also needs to be considered critically.A key interval in  in this respect is the 'Ghost Interval' between ~2000 and 7500 cal BP (151e190 cm).Pollen preservation is good throughout the interval, and the high pollen concentrations and steady pollen influx suggest that, as the peat matrix was mineralized away, the pollen was largely left intact.However, the peat sequence was greatly reduced stratigraphically and aerobic conditions may have facilitated bioturbation, mixing and disturbance of the remaining peat.Therefore, we interpret this part of the record with caution as it may not reflect a secure stratigraphic sequence.
Palynological taxa (i.e.identifiable morphotypes) may include pollen from plant species with very broad ecological tolerances, or several species with differing ecological tolerances, or species whose pollen has not yet been described in the literature.In many cases, the ecology of the plants themselves is not yet well understood.Accordingly, while we referred to local field data and to the ecological literature (particularly the African Plant Database v.3.4.0) for information on plant ecology, range and growth form, we used two complementary statistical analyses (NMDS and CTM) to explore the pollen record, to seek insights into the structures (e.g.species associations or successional patterns) that may exist in the vegetation that it represents.
The NMDS biplot (Fig. 6) maps the compositional dissimilarity between samples.Overall, there is a trend from right to left representing the overall trajectory of vegetation change over time, though with many reversals, particularly in Zones III to V. Samples in Zones I and II stand out as distinct from one another, while samples in Zones III to V overlap.This suggests that a key vegetational change reflected in the pollen dataset is the transition from Poaceae-Cyperaceae communities (Zone I), to communities dominated by trees typical of seasonally inundated forest (Manilkaratype, Khaya, Carapa-type procera: Zone II), to a mixed community of hardwood and more abundant monocot tree taxa (Zones IIIeV) which includes many of the taxa that are abundant in our census plots at the site today.Within that community there is differentiation between samples which are rich in hardwood swamp taxa such as Uapaca and Gaertnera (high NMDS axis 2 scores, especially in Zone II), and samples that are rich in the monocotyledons Raphia and Pandanus, along with Pycnanthus and Symphonia globulifera (low NMDS axis 1 and 2 scores, Zone V).
The CTM analysis identifies five 'topics' (Figs. 5 and 7), consistent co-occurrences of pollen taxa that may represent recurring ecological communities.This analysis suggests (in agreement with the NMDS) that the assemblages in Zones I and II represent distinct communities (Topic 1 [dominated by Cyperaceae and Poaceae] in Zone I, and Topic 2 [Manilkara-type and Carapa-type procera] in Zone II).Within Zones III-V, the CTM suggests that there is some structure, but it is not as clearly defined as in Zones I and II: the topics are also not well defined, containing several overlapping taxa.Topics 3 and 4 represent a signal of hydrophilic swamp forest, oscillating between two dominant swamp forest taxa, Pandanus and Pycnanthus, with Raphia and Symphonia globulifera as accessory taxa in Topic 3; and Raphia (again), without Pandanus in Topic 4. Topic 5, dominated by the hardwood swamp trees Uapaca and Syzygium along with the light-demanding taxa Lophira and Macaranga, is most abundant in subzone IVa and b, following the 'Ghost Interval' and representing a drier and more disturbed forest association.
Taking both analyses into account, we conclude that during Zones IIIeV vegetational change at the site did not involve a succession of discrete communities but, rather, shifts in the relative abundance of taxa within a single community (i.e. a 'Gleasonian' scenario rather than a 'Clementsian' one: van der Valk, 1981;Burrows, 1990;Pickett et al., 2009).Compositional changes within Zones III and V may simply represent random interactions in the dispersal, recruitment, competition and mortality of individual taxa (Hubbell et al., 2001;Vellend, 2010).We interpret the expansion of Topic 5 in Zone IV as representing an environmentally-driven change associated with low precipitation, the changing composition reflecting an ecological response to a deeper water table and an increase in canopy openness.Once palaeo-precipitation increases (evidenced by a decline in dD n-C29-v&icecorr values) the palynological assemblages are once more dominated by Topics 3 and 4, indicating hydrophilic swamp forest.
Observations of modern vegetation support these interpretations.Vegetation types that may be good analogues for Topics 1 (flooded grass/sedge fen), 2 (Manilkara-type dominated inundated swamp forest), and 5 (Lophira-dominated periodically inundated forest) are described by Moutsambot e (2012, discussed further below).A similar suite of taxa to those dominating Topics 3 (Pandanus, Pycnanthus and Raphia) and 4 (Pycnanthus, Raphia, and Syzygium) are observed in present-day vegetation plots along the CEN and EKG transects.400e12,650

cal BP
The abundance of Cyperaceae, Poaceae and Alisma-type pollen in Zone I indicates an open, herbaceous and/or aquatic palaeoenvironment.The d 13 C Bulk record confirms that the organic matter in Zone I is mostly derived from C 4 vegetation (i.e.grasses and sedges).Cyperaceae (e.g.Cyperus difformis L., C. haspan L.) and Poaceae (e.g.Echinocloa stagnina (Retz.)P. Beauv., Vossia cuspidate (Roxb.)Griff.) are frequently recorded in Congo Basin rivers and/or lakes, often forming floating mats (Bouillenne et al., 1955;Deuse, 1960;Hughes, 1992;Laraque et al., 1998;Moutsambot e, 2012).Elenga et al. (1991) record an 8 km-wide peatland, Bois de Bilanko in the Bateke Region, ROC, with a vegetation cover of Syzygium forest (Syzygium pollen is present in low abundance in Zone I at   Alisma-type pollen is recorded in Zone I. Moutsambot e (2012) records a community dominated by Ranalisma humile (Rich.)Hutch.(Alismataceae), which produces pollen of this type (Argue, 1974;Qing-feng et al., 1997), growing in shallow, calm, humicrich muddy waters, near Mokole in the eastern DRC; a similar community is also frequently found in small tributaries of the Congo River, in the Yangambi region of the DRC (L eonard, 1996).Ranalisma is considered a pioneer aquatic genus, floating or submerged, which can colonize recently exposed substrates and shallow creeks.Alismataceae pollen was also found in sediments at Lac Tele, just north of our site (Laraque et al., 1998).
Throughout Zone I, especially in Subzones Ib and Id, numerous tree taxa are present in small amounts, including Carapa-type procera, Manilkara-type, Pycnanthus, Macaranga, Raphia, and Pandanus.High dD n-C29-v&icecorr values in Subzones Ib and Id suggest that dry conditions during these intervals may have caused temporary reductions in flooding depth and/or frequency, allowing trees and shrubs to increase in abundance.Carapa-type procera and Manilkara-type can be abundant in flooded forests (see Zone II, below).The other taxa listed above are also found in open, wet environments, so it is likely that these trees were present, either in stands or as scattered individuals, close to the core site, or that their pollen was transported to the site from adjacent flooded forests.The present-day forest bordering the seasonally-inundated savanna of the Likouala-aux-Herbes is characterised by Lophira alata and other hydrophilic taxa including Manilkara-type, Alchornea and Macaranga (Moutsambot e, 2012), which occur in low abundance in Zone I.It is not unusual to find the pollen of latesuccessional swamp forest taxa in earlier successional phases, as later vegetation communities were likely to have been recruited from nearby assemblages (Morley, 1981).However, it is possible that some of the tree pollen in this zone was transported from further afield (cf.Section 5.2, above); Macaranga in particular is known to be well-dispersed with high pollen production (Laraque et al., 1998;Vincens et al., 2000Vincens et al., , 2006)), and the bisaccate pollen of Podocarpus, a montane genus, may be accounted for by long distance dispersal.
Oscillations in the pollen and dD n-C29-v&icecorr records during Zone I highlighted by the subzonation may reflect climatic changes associated with intervening cool/dry and warm/wet periods between the base of our record (~20,400 cal BP) and 12,650 cal BP, reconstructed from Congo Fan sediments and other palaeoclimatic records (discussed in Section 5.1).However, the interaction between precipitation and evaporation (influenced by changes in temperature) and the changing influence of river flooding at the site make it difficult to predict the relationship between climate, local water tables, and peatland ecology during Zone I. Climatic changes were likely spatially variable, and notable differences exist between our CEN-17.4dD n-C29-v&icecorr record and the dD n-C29-icecorr record from core GeoB 6518e1 from the Congo Fan (Schefub et al., 2005), possibly because the latter incorporates a broader, basinwide signal.Seasonality is also potentially important e a longer and/or more intense dry season can be more important to plants than a decline in total annual precipitation e but it is difficult to estimate seasonality reliably from the proxy records available.
The dD n-C29-v&icecorr record at CEN-17.4,which we interpret as a proxy for the total amount of precipitation at the site following Garcin et al. (2022), indicates generally low precipitation in Zone I, consistent with existing reconstructions of a dry climate.Oscillations in this record indicate that the period ~16,000e14,000 cal BP (Subzone Ic) was relatively humid, compared to Subzones Ib (~18,200e17,550 cal BP) and Ie (~14,000e12,650 cal BP) which were generally drier.These oscillations do not correlate consistently with patterns in the pollen record or in other proxies such as dry bulk density or C:N ratio (Fig. 8).The final maxima in Poaceae and Cyperaceae coincide with low precipitation during the interval 14,000e11,450 cal BP (Subzones Ie and Iia e discussed further below).There is evidence for episodic fluvial inundation (e.g.fluctuating LOI in Subzone Ic) at the core site during this interval which may reflect river channel migration, potentially another important influence on site development during Zone I.
The geographically closest Late Glacial palynological record to CEN-17.4 is Ngamakala Pond (Elenga et al., 1994(Elenga et al., , 2004)), a wetland occupying a small basin on the Bateke Plateau (400 m a.s.l.), 550 km to the south.This record spans the past ~24,000 years and records a transition from "hydromorphous forest" (including Sapotaceae undiff.and Syzygium) to "hydrophytic grassland" (including Xyris and Nymphaea lotus L.) between ~24,000 and ~13,000 cal BP, attributed to an inferred decrease in precipitation and increased seasonality.The Late Glacial vegetation histories are broadly similar, with both sites being dominated by open sedge/grassland, but in detail the palynological assemblages at the two sites differ markedly, with CEN-17.4 being dominated by Cyperaceae, Poaceae and Alisma-type, versus Xyris and Nymphaea lotus at Ngamakala, possibly reflecting individual site characteristics (e.g.soil or water chemistry).
Another pollen record from the southern ROC, Bilanko, also spans this period; though poorly dated, it shows abundant montane taxa (Ilex, Olea and Podocarpus) prior to 11,000 kyr BP.Barombi Mbo (Cameroon) likewise shows abundant Olea.The migration of montane taxa to lower altitudes during the LGM has been related to an increase in trade-winds, which caused strong marine upwellings and productivity, and a fall in sea surface temperatures (SST) (Jansen et al., 1984;Mix and Morey, 1996).Lower SST may have increased the formation of low stratiform clouds, promoting the survival of montane taxa at lower altitudes than present (Maley and Elenga, 1993;Maley, 1996).Several authors (e.g.Colyn et al., 1991;Jahns, 1996;Maley, 1996) have raised the possibility that rainforest taxa may also have survived the driest conditions of the last glacial on floodplains in the CCB.At CEN-17.4, the dominance of Poaceae and Cyperaceae throughout the Pleistocene part of the record indicates that the site itself was not forested, but the persistent presence of several tree taxa discussed above indicates their presence somewhere within the pollen catchment of the site.
Charcoal influx rates are high throughout Zone I, especially in Subzones Ia, c and e, indicating that fire was an important process.Today, the Congo Basin is one of the most lightning-prone regions on Earth (Christian et al., 2003) and it is quite plausible that natural fires could take hold in dry grass-sedge vegetation during the dry season(s).Anthropogenic fire ignitions cannot be excluded, however, especially given that the open Pleistocene landscape is likely to have been easier to access than today's dense swamp forest.The fires are strongly associated with Topic 1, characterised by Cyperaceae and Poaceae, suggesting that fuel type may have been important in determining the fire regime, although fuel load and moisture typically also play a role in fire ignitions (McLauchlan et al., 2020).Fire is thought to be important in maintaining open savannah ecosystems where climatic and soil conditions would be sufficient to support forest cover, and fire may have played a role in delaying succession to swamp forest at CEN-17.4 (Veenendaal et al., 2018;Aleman et al., 2019).

Zone II: 12,650e9680 cal BP
The very top of Subzone Ie and the lower part of Subzone IIa (422e464 cm) show some evidence for reduced precipitation in the dD n-C29-v&icecorr values, which are generally higher than in the previous subzones.The dry conditions ~12,000 cal BP are not associated with peat decomposition at this point in the record, likely due to the dominance of C4 vegetation, evidenced in the d 13 C n-c29 record (Fig. 8).The peat was also shallower here than during the Ghost Interval, with likely less surface water lost via overland flow, lessening the impact of any drying (Garcin et al., 2022).Proxy records from Lake Barombi Mbo in Cameroon (Giresse et al., 1994;Maley and Brenac, 1998), Lake Bosumtwi in Ghana (Shanahan et al., 2015) and Ngamakala Pond in ROC (Jansen et al., 1995) also indicate increased aridity at this time (Weijers et al., 2007;Schefub et al., 2005).However, the CEN-17.4data show high sample-to-sample variability, particularly ~12,000 cal BP, similar to the dD n-C29-icecorr record from core GeoB6518-1 from the offshore Congo fan (Fig. 10; Schefub et al., 2005).Renssen et al. (2018), modelling global hydroclimate, suggest that the Congo Basin may have seen (at most) only small decreases in temperature (<1 C) and soil moisture in response to perturbation of the global climate system between 12,900 and 11,700 cal BP, related to deglaciation and the drainage of Lake Agassiz, among other forcings.This finding is consistent with the CEN-17.4dD n-C29-v&icecorr data, which suggest some impact of regional-or global-scale perturbations to the climate system, but substantial modulation and modification of the signal by other factors.
By contrast with the dD n-C29-v&icecorr data, the pollen data from CEN-17.4 spanning the Zone I and II boundary indicate a marked reorganisation of the ecosystem from a sedge-grassland with few trees, to a closed-canopy forest dominated by taxa typically found today in seasonally-inundated forests.The inferred decline in sedges and grasses at the site is supported by the d 13 C Bulk record which shows an initially abrupt decrease around ~12,600 cal BP indicating a shift from C 4 -to C 3 -dominated vegetation, a transition that was fully completed by the start of the Holocene.This is consistent with the d 13 C n-c29 values which show a shift from high to low values ~12,000 cal BP reflecting the shift from a grass-to treedominated environment.Indicator taxa characterising Zone II include Manilkara-type, Carapa-type procera, Khaya, Sapotaceae undiff.and Entandrophragma.This assemblage, representing a seasonally-inundated forest community, is distinctive, separated clearly in the NMDS biplot and well defined by the topic model analysis (Topic 2).Other swamp forest taxa (notably Raphia, Pycnanthus, Pandanus, and Uapaca), which are present from the beginning of the record, remain consistently present in small amounts.The rarefied richness curve (Fig. 5) suggests that the diversity of the vegetation increased.Adjanohoun et al. (1988) and Moutsambot e (2012) describe a potential analogue for the Zone II environment in north-eastern ROC, south of Souanke, where the forest is dominated by Manilkara-type, Entandrophragma, Carapatype procera, Uapaca, Macaranga, and Pycnanthus, with standing water.Campbell (2005) summarises earlier vegetation classifications (Lebrun and Gilbert, 1954;Evrard, 1968;Fay et al., 1989) indicating that all these taxa (except Carapa-type procera and Pycnanthus) are dominant in near-permanently inundated swamp forest.Therefore, the inference of a closed canopy, waterlogged, swamp forest environment for the vegetation in Zone II is well supported by the ecological literature.
It may be no coincidence that the major shift in vegetation defining the Zone I/II boundary took place during a climaticallydriven reduction in precipitation, which may have allowed trees to become established at the site, outcompeting herbaceous vegetation.On the other hand, autogenic succession related to the gradual accumulation of peat (more than a metre by this point) could achieve the same effect, and certainly once the flooded forest community was in place, it persisted without interruption well into the Holocene, through the warming of climatic conditions and a precipitation increase indicated in the CEN-17.4dD n-C29-v&icecorr record from ~11,450 cal BP.
As peat continued to accumulate throughout the time of Zone II, there was a gradual reduction in the importance of river flooding over time, evidenced by high LOI and reduced concentrations of lithophilic elements (notably Al and Ti).This decrease in allogenic input, accompanied by distinct shifts in vegetation composition (e.g.Zones IIeIII), suggests that a fenebog transition was underway, as the peatland became progressively more ombrotrophic (Hughes, 2000;Hughes and Barber, 2004;L€ ahteenoja et al., 2009;Kelly et al., 2017).At CEN-17.4 this appears to have been a gradual process.The decline in inputs of mineral material may have been driven not just by the upward accumulation of peat but also by the lateral expansion of the peatland towards the adjacent rivers, which would act to limit the ingress of floodwater and reduce its velocity and therefore sediment load (cf.Mertes, 1997).
The near-total absence of micro-and macrocharcoal in Zone II (and most of the remainder of the record) suggests that fire had ceased to be important.It is possible that, beneath the swamp forest canopy, the litter did not become seasonally dry, so natural fires could not take hold.Alternatively, whereas deliberately burning the dense grass/sedge vegetation in the late Pleistocene (Zone I) may have been useful for hunting, the anthropogenic use of fire in the Holocene flooded forest (Zone II) may have been less practical and/or advantageous; to our knowledge, burning is not carried out in the swamp forest today.

Zone III: 9680e2320 cal BP
In Zone III, mixed swamp forest replaced the Manilkara-type forest of Zone II.The CEN-17.4 dD n-C29-v&icecorr record indicates an increase in precipitation from the start of the Holocene to a peak between 8000 and 9000 cal BP.This peak is paralleled in other palaeoclimatic records from central Africa, which indicate high precipitation, high humidity and a reconstructed annual mean air temperature (MAT) in the Congo Basin of ~24e25 C (e.g.Schefub et al., 2005;Weijers et al., 2007;Bayon et al., 2012;Shanahan et al., 2015).The  pollen record indicates that the flooded forest of Zone II was gradually replaced during Zone III by a mixed swamp forest, largely lacking Manilkara-type and the other indicator taxa for Zone II, and dominated instead by taxa including Pycnanthus, Gaertnera, Syzygium, Lophira, Uapaca, Raphia and Pandanus, all of which have been recorded in census plots on the peatland today.Palynological richness is at its maximum, suggesting that the forest had become more diverse over time, which may simply represent an autogenic process driven by gradual immigration and establishment of additional taxa, rather than any environmental driver (cf.MacArthur and Wilson, 1967).
While peat appears to have been accumulating steadily in Subzone IIIa, in Subzones IIIb and IIIc the age-depth curve flattens (the 'Ghost Interval', i.e. 151e190 cm/2150e7520 cal BP, Section 5.1 above).The near-total loss of several metres of peat means that the stratigraphic integrity of this part of the record is doubtful.Although pollen concentrations are very high, especially in Subzone IIIc (suggesting that the pollen remained intact as the peat mineralized), and there is substantial sample-to-sample variation in the pollen taxa, overall, we infer that there was little compositional shift in the vegetation during the period represented by the 'Ghost Interval' until ~2000 cal BP, in Zone IV (discussed below).
The dominant taxa in the pollen assemblages in Zone III e those that characterise Topics 3 and 4 e are all abundant in census plots at the site today (indeed they become important again in Zone V, in the uppermost part of the sequence).The modern vegetation is diverse and spatially variable; some of the variation may relate to the large-scale structure of the peatland (e.g. the increasing abundance of Raphia towards the centre of the peatland), but much of the variation is not easily explainable.The pollen data suggest that, throughout the 7000 years covered by Zone III, the vegetation composition at the location of CEN-17.4 fluctuated around a mean, possibly simply as a result of random vegetation dynamics.Besides the gradual decline of Topic 2 taxa (Manilkara-type, Carapa-type procera) across the Zone II/III transition, there is little evidence for directional, successional change in Zone III, suggesting that the peatland vegetation may have achieved something close to a state of dynamic equilibrium.

Zone IV: 2320e1010 cal BP
This zone includes the end of the 'Ghost Interval', ~2000 cal BP, when there was a significant decline in several swamp forest taxa including Raphia, Pandanus and Pycnanthus, and an increase in pioneer and light-demanding taxa including Lophira, Alchornea, and Macaranga (Laraque et al., 1998;Brncic et al., 2007;Kiahtipes et al., 2011).Other indicator taxa for this zone include Syzygium and Uapaca, with Croton and Homalium in smaller proportions; Topic 5 (Uapaca, Syzygium, Lophira, Macaranga) becomes dominant, especially in Subzone IVa.Despite the increase in pioneers and light demanders, there is little expansion of grasses, sedges or herbs, which peak at just 2% at ~2000 cal BP, the point of lowest precipitation as inferred from the dD n-C29-v&icecorr record, which suggests that the forest canopy remained largely intact.Rarefied richness declines markedly throughout this zone, suggesting the forest became less diverse (Fig. 5).
The shift in Zone IV (particularly ~2000 cal BP) to more heliophilic and drought-tolerant vegetation is consistent with a welldocumented period of vegetation change in central Africa between 2500 and 2000 cal BP, the Late Holocene Rainforest Crisis (Giresse et al., 2020;Bayon et al., 2019;Maley et al., 2018;Garcin et al., 2018).Palaeoenvironmental records from Nyabessam swamp in south-western Cameroon (Ngomanda et al., 2009), Lakes Barombi Mbo (Maley and Brenac, 1998) and Ossa (Reynaud-Farrera et al., 1996) in southern Cameroon, Lakes Kitina (Elenga et al., 1996) and Nguene (Giresse et al., 2009) in Gabon, the Mopo Bai swamp in north-western ROC (Brncic et al., 2009) and Ngamakala Pond in south-western ROC, Elenga et al., 1994), all within the Greater Congo Basin, consistently show a shift to pioneer and grassland taxa and a decline in mature swamp forest taxa at this time (Maley et al., 2018), though there are variations in the nature and timing of the ecological changes.At Goualougo Lake, 200 km northwest of CEN-17.4, Brncic et al. (2007) record an increase in herbs, shrubs and light-demanding trees (especially Lophira alata) at ~2000 cal BP and independent evidence for drought from ~2800 to 2000 cal BP (an increase in dust proxies in the geochemical record).A record from marshland adjacent to the Bodingue River in the Ngotto forest of the Central African Republic, 300 km northeast of CEN-17.4 (Kiahtipes et al., 2011), shows an increase in pioneer taxa (Triplochiton) and oil palm (Elaeis) around 2000 cal BP, followed by an expansion of light-demanding taxa (Asteraceae, Cyperaceae and Pycnanthus) by ~800 cal BP; though poorly constrained, the chronology also shows a pronounced reduction in sedimentation rate after ~2000 cal BP, which Kiahtipes et al. attributed to reduced precipitation.
The cause of the Late Holocene Rainforest Crisis has been widely debated: it may have been driven by climatic drying, anthropogenic clearance, or a combination of the two (e.g.Maley et al., 2018;Garcin et al., 2018;Bayon et al., 2019;Giresse et al., 2020).The absence of compelling evidence for anthropogenic activity such as crop production or burning in the CEN-17.4record suggests at most a minor, if any role for human impacts at our study site.
Subzone IVa sees a small increase (<1%) in Elaeis guineensis (oil palm) pollen.Elaeis has long been cultivated by Bantu populations and is sometimes considered to be an anthropogenic indicator in pollen diagrams (e.g.Sowunmi, 1999;Maley, 1987;Brncic et al., 2009;Bostoen et al., 2013), but it also grows wild in peatland forests today (e.g.Maley and Chepstow-Lusty, 2001;Kahlheber et al., 2014, and authors' personal observation); other palaeoecological studies have also suggested that it is not a consistent indicator of human activity.For example, Brncic et al. (2007) noted a low abundance of Elaeis pollen from ~3300 cal BP at Goualougo Lake, but concluded that fluctuations in its abundance were more likely to have been driven by climate than by anthropogenic activity, as its abundance was correlated with dry climatic intervals.Several studies (Maley, 1999;Maley and Chepstow-Lusty, 2001;Maley et al., 2018;Maley et al., 2018Maley et al., , 2018) ) have concluded that the expansion of Elaeis following the forest fragmentation at ~2500 cal BP reflects its status as a natural pioneer, preceding the reestablishment of a mature forest, rather than related to anthropogenic activity.The limited expansion of Elaeis in the CEN-17.4record during Subzone IVa therefore does not necessarily imply any human activity.There is likewise no evidence in the charcoal record for burning at the site at this time.There is thus no strong evidence that humans had a substantial impact this far into the peatland, almost 20 km from the nearest river.
In Subzone IVb swamp forest taxa re-expand, initially led by the heliophiles Pandanus and Raphia, despite dD n-C29-v&icecorr values remaining high and stable indicating continued low precipitation and/or increased seasonality.Lophira remains a substantial component of the forest structure here, and has been seen to expand at numerous sites across the region around this time, with some variation in timing (Brncic et al., 2007(Brncic et al., , 2009;;Ngomanda et al., 2009;Maley et al., 2018;Giresse et al., 2020).A mature swamp forest eventually became re-established, with a composition similar to that prior to the 'Ghost Interval' (i.e. in Zone III) but associated with markedly lower precipitation (more positive dD n- C29-v&icecorr ).Thus, some recovery of the ecosystem appears to have occurred once the new precipitation regime stabilized.Recovery of forests following the LHRC is seen in other palaeoecological records from the region.For example, following a forest contraction at Ngamakala Pond ~3000 cal BP, the forest expanded again, persisting to the present day (Elenga et al., 1994).Small increases in rainfall (e.g.~2300e2110 cal BP and ~1880e1345 cal BP) at Mopo Bai in the north-western ROC were accompanied by a rapid expansion of hydrophilic and shade-tolerant taxa (Brncic et al., 2009), suggesting that the vegetation responded rapidly to increasing precipitation.In addition to the forest recovery, the resumption of peat accumulation at CEN-17.4 after ~2000 cal BP is also in line with other records from across the Basin (Garcin et al., 2022;Elenga, 1992;Elenga et al., 1994).For example, in the Lope National Park, Gabon, Bremond et al. (2021) record that most peaty marshes established between 2150 and 1950 cal BP.The detailed reconstruction of the 'Ghost Interval' in the CEN-17.4record thus contributes to the growing body of evidence for a substantial climatic change with widespread and profound ecological consequences in Central Africa beginning ~5000 cal BP and peaking ~2000 cal BP.In particular, our record indicates that the LHRC was not restricted to terre firme rainforests, but also had a substantial impact on peatland carbon storage and on the composition of peatland vegetation.Maley et al. (2018) proposed that the floodplains of the CCB may have provided refugia for rainforest taxa, a suggestion supported by our pollen evidence for continued presence of forest throughout the Holocene, albeit with substantial turnover in community composition.

Zone V: 1010 cal BP e present
Overall, the data for Zone V indicate a similar environment to the pre-'Ghost Interval' swamp forest of Zone III.The key pollen indicator taxa for Zone IV/Topic 5 become scarce, replaced by Topics 3 and 4 (Pycnanthus, Raphia and Pandanus).We therefore interpret this zone as indicating that the swamp forest had largely recovered from the climatic event that created the 'Ghost Interval'.In detail, however, there are some important ways in which the ecosystem continued to change during Zone V.
Subzone Va continues the recovery initiated in the previous subzone and the assemblages here become very similar to those of Zone III, with abundant (but variable) Pycnanthus.Although microcharcoal influx is usually very low throughout Zone V, at the Subzone Va/b boundary there is a moderate peak in microcharcoal.The fact that this peak occurs during a transition to wetter climatic conditions does not rule out the possibility that this is a chance, natural, local phenomenon, but equally we cannot exclude the possibility of small-scale, local anthropogenic burning (the continuing low influx of microcharcoal is evidence against widespread burning).
Towards the top of Subzone Va and into Subzone Vb both Raphia and Pandanus become more abundant, followed a little later by Symphonia globulifera.Arguably, the expansion of these taxa can be interpreted as the continuation of a longer-term trend: all three taxa show a tendency towards greater abundance moving up through the core from Zone III onwards, albeit with numerous reversals.Inspection of the NMDS biplot (Fig. 6) confirms that, ignoring the disturbance associated with the 'Ghost Interval' in Zone IVa, the pollen assemblages show a steady compositional trend towards more negative Axis 1 scores across Zones II, III and V.This suggests that there may be a weak successional trend, spanning the last ~12,000 years, underlying substantial superimposed higher-frequency variation in vegetation composition (presumably driven partly by climatic change and partly by stochastic ecological processes such as competition).It is not clear whether this temporal trend bears a functional relationship to the present-day spatial vegetation patterning at the site (Raphia being more common towards the centre than the margins of the site; cf.e.g.Anderson and Muller, 1975;Phillips et al., 1991).Further research is needed to understand the drivers of present-day vegetation gradients and to test whether a similar long-term increase in Raphia, Pandanus and Symphonia globulifera occurs in other sequences from Ekolongouma and at other sites in the CCB.
On a finer temporal scale, variations in the pollen assemblages within Zone V may be partly attributable to climate change.Values of dD n-C29-v&icecorr decrease throughout Subzone Va (indicating increasing precipitation), reach a minimum (i.e.wet conditions) in Subzone Vb between 300 and 600 cal BP, and then increase again (drying) into Subzone Vc at the top of the sequence.Several records from central Africa show similar evidence for climate variability during the last millennium, although the timing of inferred changes is not always the same, possibly reflecting the limitations of agedepth models rather than indicating regional variation in climate histories.For example, at Ngamakala Pond, Elenga et al. (1996) record a decline in heliophilic tree species such as Pandanus and Poaceae from ~490 cal BP, interpreted as indicating a return to wetter conditions.At Mopo Bai, Brncic et al. (2007) record an increase in moisture between 600 and 100 cal BP accompanied by the expansion of hydrophilic taxa such as Pycnanthus.At Lake Sinnda, Niari Valley, southern ROC (Vincens et al., 1994), Poaceae declines ~650 cal BP.Several proxies in the record from Kivesso, a coastal swamp near Diosso, southwest ROC, show a clear increase in precipitation from ~500 cal BP until ~300 cal BP (Malounguila-Nganga et al., 2017).Thus, the CEN-17.4dD n-C29-v&icecorr record adds to a regional body of data suggesting that the climate was wetter than today around 500 years ago.These climatic changes may have influenced vegetation changes e for example, peak abundance of Pandanus and Raphia coincides very approximately with the dD n- C29-v&icecorr minimum e although there is no simple relationship between the isotope record and the pollen composition.
The uppermost sample is similar in composition to samples from Zone IV (the 'Ghost Interval'), plotting in a similar position in the NMDS biplot, unlike the other samples in Zone V.The presentday vegetation surrounding the coring site includes Uapaca mole, Raphia laurentii, Carapa procera, Xylopia aethiopica and Pandanus candelabrum P. Beauv.Uapaca and Pandanus are well represented in the uppermost pollen sample, while Raphia, Carapa-type procera and Xylopia are also present but less abundant, probably reflecting variations in pollen production between species.While taphonomic effects should not be ruled out, replication in other cores and sites could shed light on the extent to which CCB peatland ecosystems have changed in recent decades.Our data suggest that, throughout its history, the ecosystem at the site of CEN-17.4 has rarely been static for long, which suggests that we should expect the presentday peatlands to be variable in time and space.

Timing and pattern of peat accumulation
This detailed multi-proxy study has explored the genesis and development of an interfluvial peatland in the central Congo Basin.Our age model indicates that peat began to accumulate some time between 20,400 and 17,500 cal BP on a clay-silt substrate, at the lowest point measured along the transect between the Ubangui and Likouala aux Herbes rivers, close to the centre of the peatland.Basal dates from four other cores at the site indicate that the peatland was much smaller during the Pleistocene than it is today, and that it expanded rapidly laterally during the early Holocene.Net long-term apparent peat accumulation rates decrease from the centre towards the peatland margins.Vertical peat accumulation at the peatland centre (CEN-17.4) was likely slow during the Pleistocene, then more rapid during the early Holocene.A drying interval, peaking ~2000 cal BP, caused the near-complete mineralization of several metres of pre-existing peat.Less well-constrained agedepth models from three other cores from Ekolongouma (and from two other sites in the CCB, Lokolama and Bondamba: Garcin et al., 2022), all show a similar age-depth pattern in the mid-Holocene.The driest period at the end of the decomposition interval identified at CEN-17.4 is coincident with an arid phase, the LHRC, seen across central Africa and often represented by a vegetation change between 3000 and 2000 cal BP.All four dated cores from Ekolongouma indicate that peat accumulation resumed after ~2000 cal BP.

Pattern of vegetation change through time
Our palaeoenvironmental record shows that the site of CEN-17.4 has changed dramatically since the beginning of peat accumulation.The early peatland was characterised by a frequently inundated floodplain environment, dominated by Cyperaceae, Poaceae, and an aquatic plant producing Alisma-type pollen, possibly Ranalisma sp.At the start of the Holocene the vegetation shifted to Manilkaratype dominated, possibly seasonally inundated forest.After 9670 cal BP, a mixed swamp forest replaced the Manilkara-type forest and has remained in place ever since, with at times substantial variations in the relative proportions of taxa.In particular, there is a long-term trend towards greater abundance of Raphia, Pandanus and Symphonia globulifera, perhaps reflecting slow successional processes; the driest climatic conditions at ~2000 cal BP are associated with a marked increase in heliophiles including Lophira (supporting existing narratives around vegetation change during the LHRC); and throughout the record there are highfrequency fluctuations in the relative abundance of pollen taxa, which may simply reflect stochastic ecological processes.

Fire history and human impact
Fire was prevalent in the grass-sedge vegetation of the early peatland.Plausibly, these fires were lightning-ignited, natural events, but we cannot exclude the possibility that these open grassdominated wetlands were burned in the dry season by humans, perhaps to facilitate hunting and fishing.Fire activity appears to have all but ceased once the peat swamp forest became established at the start of the Holocene.Moderate peaks in macroscopic charcoal in Zone V likely represent isolated fire occurrences of local extent only.There is no evidence anywhere in our record that the peat itself burned.Likewise, the brief expansion of Elaeis in the record is not an unequivocal indicator of human impact.The absence of evidence for human impact does not of course prove that humans were not present, and they may have shaped the environment in ways that are invisible to the proxy techniques employed here (e.g.hunting, digging access canals [Blake, 1993]).

Climate versus autogenic drivers of ecosystem change
The pattern of ecosystem change in the early part of our record e the shift from open vegetation, to Manilkara-type dominated flooded forest, to mineral-poor, mixed swamp forest e is consistent with the pattern of successional change frequently observed in peatlands in similar topographic settings (e.g.Phillips et al., 1997;Roucoux et al., 2013;Kelly et al., 2020).The geochemical data suggest that the degree of flooding with mineral-rich water gradually declined over time, and as the peat built up it is likely that the depth and frequency of flooding by river water declined.This autogenic process alone may have been sufficient to drive the major variation in our record (Zones I to III).However, our data also indicate that peat accumulation and forest structure and composition have been sensitive to past climatic change.Most significantly, the substantial decline in precipitation culminating at ~2000 cal BP that caused the mineralization of several metres of peat is associated with a substantial change in the composition of the forest, including an increase in the abundance of taxa indicating drier and more open conditions (notably Lophira).Peat accumulation and the composition of the swamp forest both recovered after this event.Together with the limited evidence for directional successional vegetation change since ~9670 cal BP and the absence of evidence for human impact, our data suggest that changes in precipitation (and possibly its seasonality) have been an important driver of ecological change at the CEN-17.4site at least since the mixed swamp forest was established ~9670 cal BP, if not before.
This first palaeoenvironmental record from the CCB peatlands has substantially revised our understanding of the timing and pattern of peat initiation, which is now known to be much older and more variable than thought by Dargie et al. (2017).We have shown that the ecosystem at the site today has changed substantially over time, partly following autogenic processes, but also reacting substantially to past climatic change.Further research is required to improve our understanding of the spatio-temporal patterning within the enormous peatland complex at Ekolongouma, and to test whether Ekolongouma is representative of peatlands elsewhere within the CCB.
wrote the paper, with input from all co-authors.

Appendix A. Extended Data Table
Radiocarbon dates from peat cores reported in the text and used to generate age-depth models and net peat accumulation rates.

Fig. 1 .
Fig. 1.Location of CCB peatlands (shown in grey), within the larger Congo hydrological basin (part shown, in light green) in Central Africa (panel A).Black rectangle shows location of panel B. Panel B shows the Likouala aux Herbes and Ubangui Inter-fluvial Basin, including the location of Transects (Itanga, Central and Ekolongouma), and the coring location (CEN-17.4km).Additional sites mentioned in the text are labelled where appropriate in both panels.

Fig. 2 .
Fig. 2. Transect characteristics (Itanga, Centre and Ekolongouma).Peat depths (m) along each transect (distance, km, top panel), with peat core locations shaded in black.Corresponding radiocarbon chronologies for each peat core (middle panel), with basal dates, (median ages, cal BP; the basal peat date for CEN-17.4 is a central estimate from the age model (see section 4.1 for discussion)), associated age ranges (95%) and net accumulation rates (mm yr À1 ) (bottom panel).

Fig. 5 .
Fig. 5. CEN-17.4pollen percentage diagram showing taxa >5% abundance and indicator taxa mentioned in the text (coloured by zone), plotted against depth (cm) and age (cal BP).The diagram is divided into five zones determined via CONISS analysis and fifteen subzones.Pollen concentration (grains cm À3) and influx (grains cm À2 yr À1 ) are also shown, along with rarefied palynological richness (E [T300], expected number of taxa in a count of 300 grains), monoloete spores (%) and charcoal influx rates (fragments cm À2 y À1 x10 e4 ).Correlated topic model community proportions (%) are also presented.
and more Fe) and less variable than in Zone I.These factors, along with the shape of the age-depth curve, suggest that peat accumulation rates increased during this period.Charcoal influxes drop to very low levels at the start of Zone II and remain low until Zone V.Zones III (9680e2320 cal BP) sees a decline in Manilkara-type and an expansion of several taxa which remain important for the rest of the record.Zone III is characterised by Pycnanthus and (in the upper part) Gaertnera, but Syzgium, Uapaca, Raphia, and Pandanus are also important.The inorganic geochemistry of Zone III is marked by a large increase in the relative abundance of Ca and Na at the expense of Fe.The 'Ghost interval', the stratigraphic interval coinciding with a shallower gradient in the age-depth profile, starts just before the base of Subzone IIIc and ends at the base of Zone IV (i.e. between 190 cm/7520 cal BP and 151 cm/2150 cal BP).This peat representing this interval shows a sharp increase in C/N and minor changes in several other proxies (increased TC, decreased LOI and d 13 C Bulk ); pollen concentration increases substantially, but there is little compositional shift in the pollen assemblages.Zone IV (2320e1010 cal BP) sees an increase in the gradient of the age-depth profile and an inferred recovery of peat accumulation.C/N, TC, LOI and d 13 C Bulk also revert to values similar to those in Subzone IIIa.There is a marked palynological change at the base of the zone towards assemblages dominated by Lophira and Uapaca, joined by Syzygium and Symphonia globulifera in Subzone IVb.Zone V (1010 cal BPepresent) sees a return to pollen assemblages very similar to those in Zone III, but with a greater abundance of Pandanus and, in the upper part, Symphonia globulifera and Uapaca.Isolated peaks in macrocharcoal (but not microcharcoal) occur, especially at the Subzone Va/b boundary.P and K concentrations both increase markedly towards the top of the sequence.
D. Hawthorne, I.T. Lawson, G.C. Dargie et al.Quaternary Science Reviews 305 (2023) 107992 taxa for Zone I.The second topic is dominated by Manilkara-type and Carapa-type procera, indicator taxa for Zone II, and the topic model indicates that Pandanus is consistently important alongside them.Topics 3e5 vary in abundance throughout Zones III to V: Zone III sees an early decline in Topic 2 (Manilkara-type, Carapatype procera), followed by successive increases in Topic 3 (Pandanus, Raphia, Symphonia globulifera) and Topic 4 (Pycnanthus, Raphia, Syzygium).The early part of Zone IV sees a distinctive increase in Topic 5 (Uapaca, Syzygium, Lophira) before a gradual return to a mix of Topics 3 and 4 in the later part of Zone IV and Zone V.

Fig. 7 .
Fig. 7. Correlated topic model bar charts.Beta values show the relevant contribution of the dominant taxa to each topic model community.

d
13 C bulk ), suggesting rapid litter and hence peat accumulation until 190 cm (~7520 cal BP).At this point e the lower boundary of the 'Ghost Interval' of Garcin et al. (2022) e the age-depth curve flattens markedly, until 151 cm (~2000 cal BP).Garcin et al. (2022) argued, using the palaeoenvironmental proxies discussed here and additional data from two other sites in the CCB (Lokolama and Bondamba), that climatic drying from during the second half of the Holocene resulted in falling water tables, which led to contemporaneous and secondary decomposition of peat.The drying trend, marked in CEN-17.4 by a trend towards more positive dD n-C29- v&icecorr values, began ~5000 cal BP and strengthened over time, culminating around 2000 cal BP.The trend in dD n-C29-v&icecorr is accompanied by a decline (from ~30 to <5% by 2000 cal BP) in two of the most abundant swamp forest taxa, Pandanus and Pycnanthus,
and floating mats of Poaceae and Cyperaceae.An alternative, but less likely analogue is a different kind of seasonallyinundated open savannah vegetation (including the grasses Hyparrhenia diplandra (Hack.)Stapf and Setaria restioides (Franch.)Stapf), presently found bordering the Likouala-aux-Herbes River, developing hydromorphic soils on recently deposited alluvium (Moutsambot e, 2012).These grasslands form part of a succession from savanna to seasonally-inundated forest dominated by Lophira alata Banks ex C. F. Gaertn.and Daniella pynaertii De Wild.(Moutsambot e, 2012).
Dargie et al.Quaternary Science Reviews 305 (2023)107992 batch no.201890, 11,267 ± 370 spores per tablet) were used as an exotic marker in each 1 cm 3 sample of peat.Samples were mounted in silicone oil and routinely counted and identified at Â1000 magnification using a Zeiss Axioskop microscope.Identifications were made using a pollen key derived from reference material compiled from the literature and personal observations and/or photographs from reference collections at the University of Oxford,

Table 1
Description of results.Fibrous woody peat.Age-depth curve is steady and steep.Pollen: Indicator taxa for this zone include the palm Raphia and two hardwood taxa, Pandanus and Symphonia globulifera(means 11.3, 41.3 and 7.0% respectively).Pycnanthus is abundant in the lower part of the zone and Uapaca is abundant in the uppermost sample, but most of the indicators of Zone II (Syzygium, Lophira, Homalium) are scarce.Pollen richness is lower than in Zone II.Charcoal: Low influxes throughout (<75 fragments cm À2 yr À1 ).Organic geochemistry: The mean C/N ratio is 39.7, falling to 23.4 in the uppermost sample, reflecting an increase in total nitrogen.dDn-C29-v&icecorrdecreases to a minimum of À31.7‰ at 46 cm, then increases to À17.0‰ towards the top of the sequence.Inorganic geochemistry: In Subzone Va the concentrations of most elements fall substantially, recovering again in Subzone Vb.P and K are relatively more abundant here than elsewhere in the peats, and the uppermost sample shows a sharp peak in P, K, Mg, Mn, Cd and Zn.Fibrous woody peat.Age-depth curve is steady and steep.Pollen: Indicator taxa include three abundant taxa, Syzygium, Lophira and Uapaca (means 12.3, 15.5, and 22.4% respectively), along with Macaranga, Alchornea, Croton, and Homalium in smaller proportions.All of these indicator taxa are hardwood trees.Pycnanthus, although present at up to 26%, is typically less abundant (mean 5.4%) than in the zone below.Subzone IVa is marked by peak values of Uapaca (up to 49.7%).In Subzone IVb the palynological assemblages become similar again to those in Zone III, with an increase in Pycnanthus and Pandanus at the expense of Uapaca (see NMDS biplot, Fig.6).Palynological diversity remains high, though decreasing gradually upwards through the zone.Monolete pteridophyte spores are occasionally abundant, reaching their peak value of 100% (outside the main pollen sum) at 129 cm.Palynological richness is higher than in Zone I and increases upwards through Zone II, with many tree taxa present in low abundances.Monolete pteridophyte spores are consistently less abundant than in Zone I. Charcoal: Influxes are substantially lower than in the zone below, typically less than 50 fragments cm À2 yr À1 .The mean C/N ratio is 42.6, similar to Zone 1, while d 13 C bulk is markedly more negative (mean À29.6‰).dD n-C29-v&icecorr values fluctuate around a mean of À24.7‰.Concentrations of most inorganic elements decline gradually upwards through the zone (Mg, Cd and Zn are exceptions in showing no declining trend).In proportional terms, Fe and Mg become more important at the expense of Al.Sapotaceae undiff.and Entandrophragma), accompanied by the other tree taxa found in Zone I.The chemistry of the peat is different (more negative d 13 C Bulk , higher C/N, proportionally less Al Charcoal: Low influxes through most of the zone, exceeding 120 fragments cm À2 yr À1 in only two samples (at 1710 and 1170 cal BP).Organic geochemistry: The mean C/N ratio is 47.3, similar to Zone III.dD C29 is above À20.7‰,lessnegativethan in Zone III.Inorganic geochemistry: Similar to Zone III.Stratigraphy: Fibrous woody peat.The age-depth curve is steep between 327 and 190 cm (Subzones IIIa and b) and much shallower between 190 and 154 cm (Subzone IIIc).Pollen: The Manilkara-type dominated assemblages of the zone below are rapidly replaced at the base of Zone III by a range of hardwood taxa.Pycnanthus (mean 19.2%) and Gaertnera (1.1%) are indicator taxa for this zone, although several other taxa (Syzygium, Lophira, Uapaca, Raphia, Pandanus) are also abundant.Palynological diversity is at its maximum in this zone.Monolete pteridophyte spores are abundant, especially in Subzones IIIa and IIIb.Charcoal: Low influxes through most of the zone, with the notable exception of a peak at 218 cm to ~134 fragments cm À2 y À1 .Organic geochemistry: In Subzone IIIa the mean C/N ratio is 47.8, mean d 13 C bulk is À29.8‰ and dD n-C29-v&icecorr values are in places very negative (mean À36.7‰ but reaching À48.2‰ at 274 cm).In Subzone IIIb C/N and d 13 C bulk show little change, while dD n-C29-v&icecorr values vary around a mean of À33.8‰.In Subzone IIIc there is a marked peak in C/N to an extremely high value of 82.3, accompanied by a well-marked minimum in d 13 C bulk to À31.0‰, d 13 C n-C29 values remain invariant, a peak in d 15 N bulk to 7.0‰, and a less distinct minimum in dD n-C29- v&icecorr to À40‰.Inorganic geochemistry: Ca, Sr and Na increase substantially just above the base of the zone, at the expense of Fe, and remain proportionally important throughout the rest of the sequence.Stratigraphy: Well humified, possibly detrital peat with occasional leaves or seeds below c. 375 cm, grading into fibrous woody peat above.Age-depth curve is steep.Pollen: At the base of the zone Cyperaceae and Poaceae rapidly decline, replaced by a distinctive assemblage of tree taxa; indicator taxa for this zone include the numerically-dominant Manilkara-type (mean 25.2%) and, in smaller amounts, Khaya, Carapa-type procera, Sapotaceae undiff.andEntandrophragma(means 1.6, 8.2, 2.9 and 4.2% respectively).from>60% in the underlying zone to <20% at the base of this zone, while tree taxa expand to >70%.Pycnanthus, Pandanus, and to a smaller extent Raphia and Uapaca are also important in this zone.Carapa-type procera, Manilkara-type, Pycnanthus, Macaranga, Syzygium, Lophira, Raphia and Pandanus.Monolete pteridophyte spores are abundant.Charcoal: Microscopic charcoal is very abundant throughout this zone, especially in Subzones Ia, c and e, peaking at ~1420 fragments cm À2 yr À1 at 603 cm.Organic geochemistry:The mean LOI value in the silts (below 599 cm) is 16.8%.d 15 N bulk values in the silts are at a maximum for the sequence, peaking at 8.0‰.LOI in the peats reaches 90.4% at 597 cm and increases gradually throughout the remainder of the zone, interrupted by reversals (e.g. to 79.2% at 545 cm in Subzone Ic).d 15 N bulk values are <2‰ in the peats.C/N values are low throughout the zone, with a mean of 39.3; peak values are

Table 1
(continued ) Pollen zones (Subzones) Depth (cm) Age (cal BP) Description achieved in Subzones Ib and d. d 13 C bulk values oscillate markedly around a mean of À24.8‰ with d 13 C n-C29 values around a mean of À32‰.dD n-C29-v&icecorr values fluctuate around a mean of À21.1‰.