Western equatorial African forest-savanna mosaics : a legacy of late Holocene climatic change ?

Western equatorial African forest-savanna mosaics: a legacy of late Holocene climatic change? A. Ngomanda, A. Chepstow-Lusty, M. Makaya, C. Favier, P. Schevin, J. Maley, M. Fontugne, R. Oslisly, and D. Jolly Institut de Recherche en Ecologie Tropicale, IRET/CENAREST, BP. 13354, Libreville, Gabon Institut des Sciences de l’Evolution de Montpellier, UMR 5554, Université de Montpellier II, Place Eugène Bataillon, cc 61, 34095 Montpellier Cedex 5, France Université des Sciences et Techniques de Masuku, Departement de Geologie, BP. 901, Franceville, Gabon Laboratoire des Sciences du Climat et de l’Environnement, UMR CEA/CNRS 1572, Domaine du CNRS, 91198 Gif sur Yvette cedex, France IRD – Cameroun, BP. 1857, Yaoundé, Cameroun


Introduction
In western equatorial Africa, the tropical lowland rainforest of Gabon is well known for its high biodiversity (Sosef, 1994).This part of the Guineo-Congolian rainforest (White, 1983) is today well conserved because of the absence of intensive agricultural activities, coupled with low population densities.However, although the Gabonese rainforest shows a relatively homogeneous aspect, it is included by grass-dominated savannas, most notably at its periphery in the coastal area, or within the forest block (e.g. the forestsavanna mosaic in the Middle Valley of Ogooué, see Fig. 1).
The origin and persistence of these savanna patches, typical of drier environments, remains controversial, as at these equatorial latitudes the current humid conditions favour rainforest development (Aubréville, 1967).A number of ecologists have assumed that the effects of recent human impact (e.g.cultivation, forest logging or savanna fires) can be extrapolated to explain the dynamics of the forest/savanna Published by Copernicus Publications on behalf of the European Geosciences Union.ecotone during the past few millennia (Fontes, 1978;White, 1992White, , 2001;;King et al., 1997).
Indirect evidence for a climatic rather than an anthropogenic origin for these savannas is provided by regional palaeoclimatic data from Lakes Barombi-Mbo (Giresse et al., 1994;Maley and Brenac, 1998) and Ossa (Reynaud-Farrera et al., 1996;Wirrmann et al., 2001;Nguetsop et al., 2004;Giresse et al., 2005) in Cameroon, and Lakes Sinnda (Vincens et al., 1998) and Kitina (Elenga et al., 1996) in the Congo, which are situated at the northern and southern limits of the Guineo-Congolian rainforest, respectively.Unlike the savannas formed during the coolest and driest conditions of the Last Glacial Maximum (Maley and Brenac, 1998), which were subsequently recovered by rainforest expansion, palaeoclimatogical data indicate that the late-Holocene savannas correspond with a period of aridity and marked lakelevel reduction.For example, Lake Sinnda became completely dry about 4000 C 14 yr BP (Vincens et al., 1998).This aridity produced a sudden contraction of rainforests, followed by a renewed expansion of savannas associated with secondary forests.
Archaeological findings for the same period indicate that Bantu-speaking peoples arrived in Central Africa synchronously as the widespread appearance of savannas and fragmented rainforests (Schwartz, 1992).It has been suggested that human migration from the grasslands of the Nigerian-Cameroonian border into central Africa at the onset of the late-Holocene was favoured by canopy openings in the dense rainforest.The band of small coastal savannas, which today run down from Equatorial Guinea through Gabon to the Congo, may even constitute part of the old migration route (Oslisly, 2001).
Here, we present two new high-resolution pollen records from Gabon, spanning the late Holocene in order to determine the origin, timing and persistence of the forest-savanna mosaic on the coast and within the dense rainforest block inland.

Regional setting
Lakes Nguène and Maridor are shallow lakes (∼3 to 5 m depth) located in the western part of Gabon, at the same equatorial latitude, though in highly contrasting geomorphological and vegetational settings.Lake Nguène is situated on the southern slopes of the Cristal Mounts (0 • 12 S-10 • 28 E, 20 m a.s.l.), approximately 160 km from the northwest coast of Gabon (Fig. 1).This ancient fluvial depression of the Abanga River has an area of ∼6 km 2 , with large areas of the littoral zone covered by a dense swamp of Cyperaceae (Cyperus papyrus, Killinga sp., Cyperus sp., Eleocharis sp., etc.).In its northern part, the lake basin has an extensive floodplain, with abundant herbaceous plants (Asteraceae, Amaranthaceae, Polygonum sp., etc.), grasses (such as Phragmites sp. and Echinoclora sp.), as well as some flood-tolerant shrubs, particularly Nauclea pobeguinii (Rubiaceae) and Uapaca heudolotii (Euphorbiaceae).On the waterlogged soils of sheltered lake shores, dense hygrophytic vegetation occurs, mainly dominated by shrubs of Alchornea cordifolia (Euphorbiaceae).The natural riparian vegetation of Lake Nguène is dominated by trees belonging to the Caesalpiniaceae family (Anthonota macrophylla, Cynometra sp., Guibourtia sp., etc.).Away from the lake, the Nguène region, located in one of the most species-rich areas of western equatorial Africa (Sosef, 1994), supports an evergreen rainforest dominated by Aucoumea klaineana, Dacryodes büettneri (Burseraceae), Desbordesia glaucescens (Irvingiaceae), and Monopetalanthus sp.(Caesalpiniaceae) (Nicolas, 1977;Caballé and Fontes, 1978).
By contrast Lake Maridor (0 • 10 S-9 • 21 E) is located approximately 3 km from the coast of Gabon.The lake basin, maintained both by direct precipitation, small streams and probably by groundwater infiltration, has a surface area of about 0.25 km 2 .Its shallow water is partly covered by macrophytes, while a marsh dominated by monocotyledonous herbs and Cyperaceae separate the lake basin from a small Mitragyna ciliata (Rubiaceae) and Uapaca guineensis dominated swamp forest (Christy et al., 1990).This small wetland is surrounded by a forest-savanna mosaic with secondary forests rich in Aucoumea klaineana and low-diversity savannas, mostly consisting of grasses in the tribe Andropogoneae.These secondary forests have been suggested to result from human disturbance of the coastal mature evergreen rainforest, rich in Aucoumea klaineana, Saccoglottis gabonensis (Humiriaceae) and Erismadelphus exsul (Vochysiaceae) (Nicolas, 1977;Caballé and Fontes, 1978;Christy et al., 1990), during the last few centuries, notably by logging.According to Christy et al. (1990), the mature evergreen rainforest today around Lake Maridor represents only 30% of the original forest cover.
Climatically, these two lakes lie in a humid part of Gabon, with mean annual rainfall ranging from 1916 mm at Lambaréné, 100 km south of Lake Nguène, to 2834 mm at Libreville, 70 km north of Lake Maridor (period from 1953Maridor (period from -1989)).Precipitation is seasonal, with a wet season which lasts about nine months (September to May), interspersed by a "short dry season" centred on January with a distinct reduction in rainfall.Temperatures vary little throughout the year, with a range of 20-33 • C; these are lowest during the major dry season (June to September), when cloud cover is almost constant, because of lower sea-surface temperatures in the Gulf of Guinea (Leroux, 1983).Similarly, mean relative humidity also varies little over the year and does not fall below 70%

Materials and methods
Sediments were collected using a Vibracorer from the central part of Lakes Nguène and Maridor at 2 m water depth.Using a sampling strategy of 5 and 10 cm intervals, 55 samples were obtained from core NGUE1 and 58 from core MAR2, respectively.Pollen preparation followed the standard methodology (Faegri and Iversen, 1989): dissolution of the carbonates and silicate with diluted HCl (10%) and cold HF (70%), respectively; removal of colloidal silica with warm diluted HCl, and destruction of humic acids by dilution in KOH (10%) solution.The obtained residue was diluted in glycerol.Identification was based on the reference collection at the Institut des Sciences de l'Evolution de Montpellier (Université Montpellier II), as well as published pollen images for tropical Africa (Assemien et al., 1974;Ybert, 1979;Bonnefille and Riollet, 1980).Pollen nomenclature follows the African Pollen database (http://medias.obs-mip.fr/apd/).Pollen percentages are based on a sum of at least 600 pollen grains and pteridophyte spores.However, to show a clear representation of the forest component, at least 250 terrestrial pollen grains were counted at each level, excluding Gramineae.These latter, as well as the local marshy herbaceous taxa (Cyperaceae, pteridophytes, Nymphaea, etc.) are excluded from the pollen sum used to express pollen abundance of terrestrial taxa, because their high pollen abundance masks the forest signal.
The chronology of the cores NGUE1 and MAR2 is based on radiocarbon ages determined by a gas proportional counter and AMS methods using samples of bulk organic matter (Table 1).These radiocarbon dates were converted to calendar years using the INTCAL04 calibration curve (Reimer et al., 2004), and a continuous chronology based on the stratigraphically consistent series of dates from each sequence was derived by linear interpolation between the calibrated ages.

Age models and sedimentation rates
Of the 11 radiocarbon dates from core NGUE1, ten dates are stratigraphically consistent, although two minor dating inversions can be noted.If the base of core NGUE1 is considered as dating from 4780 cal yr BP, two possible chronologies can be derived from the 11 radiocarbon ages (Fig. 2).The first chronology (Fig. 2, dotted line) shows a significant change in the sedimentation rate through the gley-mud at 187 cm.However, this change is neither associated with lithology nor with marked variation in the sand content, and hence suggests a regional environmental shift.Thus, the 3310±40 14 C yr BP date, stratigraphically and sedimentologically anomalous, is probably caused by allochtonous inputs or reworked organic matter.In contrast, the second chronology (Fig. 2, solid line), based on nine radiocarbon dates, clearly shows that changes in accumulation rates are systematically synchronous either with increasing sand content or with lithological variation.For example, between 2540±40 14 C yr BP and 2429±40 14 C yr BP, the accumulation rate is very high (0.20 cm/yr) and contemporaneous with high sand content (between 2 and 6%) suggesting an active erosive phase or strong stream transport.In the light of these stratigraphical and lithological considerations, the age-model illustrated by the solid line has been retained.
Core MAR2 shows a sedimentary discontinuity around 405-400 cm with the A2 horizon of the podzolic soil missing (Fig. 3), which suggests a hiatus of several hundred years.This hiatus, however, does not appear in the MAR1 test-core as this core shows a 20 cm-thick A2 horizon in the podzolic soil, inferring that the sediment break in the MAR2 core was not caused by the drying up of Lake Maridor.Above the Table 1.Radiocarbon dates of bulk organic matter from cores MAR2 and NGUE1 of Lakes Maridor and Nguène, respectively.All radiocarbon dates were calibrated using the program INTCAL04 (Reimer et al., 2004)  4500 cal yr BP, were obtained from bulk samples of clayey sediments containing black organic-rich layers.This suggests that the organic laminae are mostly allocthonous organic matter originating from the underlying podzolic horizon.The alternative age-model (solid line), constructed using radiocarbon dates, obtained from the non-perturbed clayey sediments and showing more variation, which are either synchronous with increasing sand content or with lithological changes, can be assumed to be closer to reality.

Nguène pollen record
A total of 121 pollen taxa were identified from the 55 samples of core NGUE1, with changes in the relative pollen abundance clearly reflecting the vegetation dynamics (Fig. 4).Based on floristic changes of the forest component, three pollen zones characterizing the main successive stages have been visually defined.
Zone N2 (ca.3200-1400 cal yr BP; 355-115 cm) indicates that major changes occurred in both the marsh and rainforest pollen signal.
Between 3200 and 2400 cal yr BP (subzone N2c: 355-255 cm), arboreal pollen taxa characteristic of the mature rainforest are progressively replaced by pioneering plants, notably Alchornea cordifolia (up to 60%) and Elaeis guineensis (∼10%), which favour disturbed forest habitats (Maley and Chepstow-Lusty, 2001).In Guineo-Congolian wetlands, Alchornea cordifolia grows in well-drained lake shore soils, as well as open areas of fringing forest bordering the lower reaches of lowland rivers (Lebrun and Gilbert, 1954;Evrard, 1968;Schnell, 1976).This species, which can tolerate marked flooding regimes (Evrard, 1968), may gradually colonize swamp shorelines as the mean water-level lowers.Thus, an abrupt rise in Alchornea cordifolia pollen, following a corresponding decrease in Cyperaceae pollen, clearly indicates that dense stands of littoral vegetation progressively invaded the lake basin.The significant presence of Elaeis guineensis pollen, remaining remarkably constant (∼10%) throughout zone N2, associated with Tetrorchidium (Euphorbiaceae) pollen (∼3%) and declining mature forest pollen taxa, indicate the creation of openings in the closed canopy forest surrounding the lake basin.
From 2400 to 2000 cal yr BP (sub-zone N2b: 255-175 cm), fern spores and grass pollen markedly increase, reaching maxima of 40% and 10% of the total pollen sum, respectively.Many of the marshy pteridophyte species found today around Lake Nguène are either epiphytic ferns in the surrounding Cyperus papyrus swamp, or terrestrial ferns invading the lake shores during dry season low-stands (Ngomanda, personal observation).As fern spores were not identified beyond family level, a distinction was not established between epiphytic wetland and more terrestrial forms.Nevertheless, the rise in fern spores, associated with both a slow increase in grass pollen and declining Cyperaceae pollen abundance, suggests that lake-levels were low during this period.It is also notable that a rise in marshy herbaceous pollen taxa, following a significant decline in Alchornea cordifolia-type pollen, is accompanied by an increasing abundance of secondary forest pollen taxa [e.g., Aphania-type (∼5%), Lannea-type (Anacardiaceae) (2-15%), Anacardiaceae (∼5%), Trilepisium madagascariensis (∼5%) and Flacourtiaceae (∼3%)].However, Caesalpiniaceae and Nauclea-type pollen continue to maintain significant relative abundances (5-10%).This pollen assemblage suggests that a closed canopy rainforest persisted, but with increasing fragmentation allowing secondary forest taxa to colonize the gaps.
Zone N1 (ca.1400-20 cal yr BP; 115-5 cm) A detailed high-resolution description of this section of the Nguène pollen diagram has been published already (Ngomanda et al., 2007).Here, we sub-divide Zone 1 into two sub-zones to facilitate interpretation.

Maridor pollen record
A total of 158 pollen and spore taxa were identified in the 58 samples analysed (Fig. 5).Changes in relative pollen abundance in the Maridor record, reflecting the vegetation dynamics, suggest three major zones with a total of four subzones.
In Sub-Zone M1a (ca. 1940-290 cal yr BP;110-5 cm), pioneering tree pollen (e.g., Elaeis guineensis and Macaranga) pollen rise again.Although, this is interpreted as an expansion of pioneer forest, it was not associated with a decline of pollen from secondary forest (e.g., Aucoumea klaineana, Tetracera, etc.), swamp forest (Uapaca guineensis, Melastomataceae cf.Dissotis congolensis and Raphia), or mature rainforest (e.g., Drypetes [Euphorbiaceae], Caesalpiniaceae, Mimosaceae, Plagiostyles africana [Euphorbiaceae]) taxa.These pollen assemblages reveal a picture of mixed rainforest types similar to the current complex landscape of forestsavanna mosaic.In addition to the importance of swampy environments around Lake Maridor, the high abundance of Cyperaceae pollen and pteridophyte spores suggest a mosaic of marshy-swampy vegetation.

Discussion
Pollen data from Lakes Nguène and Maridor clearly show that a well-developed moist rainforest existed around the lakes before 4200 cal yrs BP, as Caesalpiniaceae, a major indicator of Guineo-Congolian moist evergreen rainforest (White, 1983), is abundant in the pollen assemblages.Similar regional vegetation reconstructions have been described from other palaeoecological records across western Equatorial and West tropical Africa.Notably, pollen records from Cameroon (at ∼2 • N) showed that a Biafrean-type rainforest, dominated by Caesalpiniaceae, existed around Lakes Barombi-Mbo (Maley and Brenac, 1998) and Ossa (Reynaud-Farrera et al., 1996) at that time.During the same period, a semi-evergreen rainforest surrounded the catchment of Lakes Sinnda (Vincens et al., 1998) and Kitina (Elenga et al., 1996) in southern Congo (∼ at 2 • S), as well as the catchment of Lakes Bosumtwi and Sélé in Ghana and Benin (Salzmann and Hoelzmann, 2005), respectively.
From 4000 cal yr BP, the contraction of moist evergreen rainforest suggests the onset of aridity during the late Holocene.This major vegetation change occurred progressively, first by altering the floristic composition of mature evergreen rainforest (as shown by the replacement of Caesalpiniaceae by semi-deciduous trees), followed by expansion of savannas and/or open forest formations, which reached their maxima between 2700 and 2400 cal yr BP.The opening of the forest and its substitution by savannas and pioneer formations is attested in various late Holocene palaeoecological sites across inter-tropical Africa.Central-East African pollen records clearly show that mountain forest openings occurred synchronously in the highlands of Burundi, Rwanda and Uganda at 4300, 3800 and 2500 14 C yr BP (Taylor, 1990;Jolly and Bonnefille, 1991;Taylor, 1992Taylor, , 1993;;Jolly et al., 1994Jolly et al., , 1997)).In West and Central Africa, pollen data suggest a breakdown of African lowland rainforest into two distinct phases: the first one around 4000 cal yr BP, and a second phase around 2500 cal yr BP.The first "crisis" impacted the periphery of the central African forest block and was marked by the widespread appearance of savannas.At the northern periphery, at Lac Sélé in Benin, the opening of the "Dahomey Gap", a savanna corridor interrupting the West African rainforest, is dated between 4500 and 3400 cal yr BP (Salzmann and Hoelzmann, 2005).In the south, at Lake Sinnda in southern Congo, the semi-deciduous forest was replaced by savannas after 3990±70 14 C yr BP (4530-4400 cal yr BP).This phase of savanna expansion is represented by the lower zone M2 in the Maridor pollen diagram.Sites further within the forest block, such as Lakes Barombi-Mbo and Ossa, the Nyabessan swamp (Cameroon), and Lake Kitina (Congo), do not show signs of ecosystem succession for this period, and maintain stable rainforests dominated by Caesalpiniaceae (Elenga et al., 1996;Reynaud-Farrera et al., 1996;Maley and Giresse, 1998;Ngomanda et al., 2009).At these sites, forest opening occurred between 2700 and 2400 cal yr BP, and is widely considered to have been completed within a few decades (e.g., Maley, 2002Maley, , 2004;;Ngomanda et al., 2009).This phase of forest breakdown, visible in zone Ng2 of the Nguène pollen diagram ended around 2000 cal yr BP.
The spread of savannas in the Maridor region around 4000 cal yr BP, however, occurred as the lake-level increased, as shown both by the sharp rise of pollen from marsh communities, (e.g., Cyperaceae, aquatic herbs) and by the sedimentary changes observed, with the deposition of sandyclays, typical of lake deposits (Giresse et al., 2009), following podzolic soil formation.This latter deposition is typical of lowland areas in the sedimentary sandy basin of coastal Gabon, as well as observed on the Congolese Batéké sands (Schwartz, 1988).Different morphodynamic processes, operating within the lake basin due to climatic oscillations, may explain the apparent contradiction between lake-level fluctuations and vegetation change around Lake Maridor.The sediments at the base of core MAR1, showing gleymuds and podzols, typical of marshy environments, suggest that Lake Maridor was initially a swamp system.Around 4000 cal yr BP, the base of this former swamp acted as a valley bottom; active erosion from the surrounding Plio-Pleistocene sedimentary basement, and strong stream transport led to accumulation of coarse deposits in the bottom of the swamp.This process is clearly shown by the sedimentary change observed at ∼320 cm in the MAR1 core, which indicates deposition of coarse-grained sands of alluvial origin.Decreasing water level, due to declining rainfall or more accentuated seasonality, led to subsequent emergence of these coarse alluvial sediments, allowing the development of the podzolic horizon.The accumulation of these podzolised deposits blocked the outflow around 4000 cal yr BP, facilitating the formation of permanent open water.
It has also been suggested that the major expansion of Elaeis guineensis, coupled with increasing grass cover may be due to human agricultural activity (Sowunmi, 1999), as the recovery of pollen and macro-remains of this oil palm in palaeoecological records has often been used as an anthropogenic indicator (Clist, 1995;Oslisly, 1995Oslisly, , 1998Oslisly, , 2001;;Assoko Ndong, 2002).This hypothesis is supported by the fact that in many Neolithic (4500-2500 cal yr BP) and Iron age (2400 cal BP to present) Gabonese archaeological sites, substantial amounts of palm nuts have been recovered, suggesting the importance of this resource to humans during the late Holocene (Clist, 1995;Oslisly, 2001;Assoko Ndong, 2002).However, no evidence for systematic cultivation of Elaeis guineensis in the region has been formerly demonstrated, although the Bantu are considered as farmers; simply gathering palm nuts in secondary forests, natu-rally abundant in E. guineensis, may easily explain its concentration in archaeological sites.Indeed, in other Holocene sites from western equatorial Africa, pollen data clearly show that the expansion of Elaeis guineensis always followed the widespread establishment of grassland savannas (Elenga et al., 1992(Elenga et al., , 1994) ) or temporary forest openings (Elenga et al., 1996;Reynaud-Farrera et al., 1996;Maley and Brenac, 1998;Ngomanda et al., 2005).Furthermore, it can be emphasized that the Lake Nguène vegetation record shows a significant decrease in E. guineensis pollen during the last five centuries.This is exactly the time interval when anthropogenic impact on rainforests (e.g., due to increasing human population densities) would be expected to rise.
The rapid ecosystem transition between 4200 and 4000 cal yr BP, from stable moist evergreen forest to drier vegetation formations (i.e.savannas and semi-evergreen forests) and the degradation of rainforest over the third millennium BP, suggest that the vegetation changes observed in Gabon mainly reflect the regional variability of effective moisture.This hypothesis is supported by the records of terrigenous dust deposited in lake sediments downwind from the Sahara, and continental records of past precipitation changes in Central Africa (Nguetsop et al., 2004;Weldeab et al., 2005Weldeab et al., , 2007)).These combined records show a continuous decrease of rainfall from 5200 cal yr BP and the appearance of a marked dry season during the Northern Hemisphere winter months with very low atmospheric humidity between 2400 and 2100 cal yr BP.The presence in the pollen diagrams of Nguène (e.g.zones N3b and N2) and Maridor (zone M2) of deciduous trees (e.g.Holoptelea grandis, Trilepisium madagascariensis, Celtis sp., Pycnanthus angolensis, Blighia sp., Lannea sp., Aphania senegalensis, Sterculiaceae, etc.) which lose totally or partially their leaves during the Northern Hemisphere winter dry season (Letouzey, 1968;White, 1983) reinforces the evidence for seasonality change occurring from ca. 4000 cal yr BP.
The progressive drying of central African terrestrial ecosystems during the mid-and late-Holocene, which coincided with the desiccation of the Sahara (Kröpelin et al., 2008), is widely thought to have been linked to weaker insolation forcing of the African monsoon and eastern equatorial Atlantic sea-surface temperature anomalies.Furthermore, this would have affected the southward extension of the Intertropical Convergence Zone (ITCZ) and its associated rainfall belt (Nguetsop et al., 2004;Weldeab et al., 2005Weldeab et al., , 2007;;Ngomanda et al., 2009).

Conclusions
Pollen data from Lakes Maridor and Nguène in Gabon show that hydrological changes occurring over the past 4500 yr were the major driving forces controlling rainforest dynamics in this part of western equatorial Africa.They clearly demonstrate that the current forest-savanna mosaics of the coastal region of Gabon began about 4000 cal yr BP, following a rapid climatic-induced deterioration of the evergreen rainforest which covered that region during the mid-Holocene.

Fig. 1 .
Fig. 1.Simplified vegetation map of Gabon in central Africa and location of Lakes Maridor and Nguène.

Fig. 2 .Fig. 3 .
Fig. 2. Age-depth plot and sand content of core NGUE1.The age-model illustrated by the solid line has been used to derive the chronology of the core NGU1.Sediment accumulation rates are based on linear interpolation of calibrated radiocarbon dates.

Fig. 4 .
Fig.4.Summary pollen diagram from Lake Nguène.The pollen sum of terrestrial taxa excludes pteridophyte spores, aquatic herbs, Cyperaceae and Poaceae pollen.The 14 C ages in parentheses are omitted from the chronology of the core NGUE1 because they may be derived from allochtonous inputs or reworked organic matter.
Summary pollen diagram from Lake Maridor.The pollen sum of terrestrial taxa excludes pteridophyte spores, aquatic herbs, Cyperaceae and Poaceae pollen.The 14 C ages in parentheses are omitted from the chronology of core MAR2 because they may be derived from allochtonous inputs or reworked organic matter. .