14,000 years of climatic and anthropogenic change in the Afromontane forest of São Tomé Island, Gulf of Guinea

Quaternary Science Reviews 322 (2023) 108381 Available online 11 November 2023 0277-3791/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/bync/4.0/). 14,000 years of climatic and anthropogenic change in the Afromontane forest of S˜ ao Tom´ e Island, Gulf of Guinea Alvaro Castilla-Beltr´ an a , b , c , * , Ricardo Faustino de Lima d , e , Laura Benitez Bosco e , f , g , Rosa Delia Castillo Armas c , Nichola Strandberg h , Tariq St´ evart i , Lea de Nascimento c , Jos´ e María Fern´ andez-Palacios c , Sandra Nogu´ e j , k a Department of Geography and History, University of La Laguna, Spain b Archaeological Micromorphology and Biomarker Research Lab, Instituto Universitario de Bio-Org´ anica Antonio Gonz´ alez, Universidad de La Laguna, Spain c Island Ecology and Biogeography Group, Instituto Universitario de Enfermedades Tropicales y Salud Pública de Canarias, University of La Laguna, Spain d cE3c – Center for Ecology, Evolution and Environmental Changes, CHANGE – Global Change and Sustainability Institute & Departamento de Biologia Animal, Faculdade de Ciˆ encias da Universidade de Lisboa, Edifício C2, 5◦Piso, Sala 2.5.15b Campo Grande, 1749-016, Lisboa, Portugal e GGBC – Gulf of Guinea Biodiversity Center, S˜ ao Tom´ e, S˜ ao Tom´ e and Príncipe f Fauna & Flora International, Cambridge, UK g Fundaç˜ ao Príncipe, Santo Ant´ onio, S˜ ao Tom´ e and Príncipe h School of Geography and Environmental Science, University of Southampton, UK i Missouri Botanical Garden, Africa & Madagascar Department, Madagascar j Universitat Aut` onoma de Barcelona, Bellaterra (Cerdanyola del Vall` es), Catalonia, Spain k CREAF, Bellaterra (Cerdanyola del Vall` es), Catalonia, Spain ARTICLE INFO Handling editor: Donatella Magri Keywords: Afromontane forest Climate change Fossil pollen Gulf of Guinea islands Holocene Human impacts Islands Late pleistocene Multi-proxy Paleoecology ABSTRACT S˜ ao Tom´ e (Gulf of Guinea, Central Africa) is a 854 km 2 tropical island that had a pivotal role in early European colonial expansion through the Atlantic between the 15th and 16th centuries. Historical sources suggest that native vegetation has been heavily impacted since human arrival (1470 CE) due to monoculture economies and the introduction of mammals and plants, some of which now have established wild populations. The Afromontane forest of S˜ ao Tom´ e, located above 800 m.a.sl., is particularly rich in endemic plant species and has remained relatively unaffected by direct human impacts. Here, we explore how environmental change influenced this forest through the study of a sedimentary sequence from the volcanic crater of Lagoa Am´ elia (1340 m a.s.l.), a palustrine system located at the boundary between submontane (800–1400 m a.s.l.) and mist forest (above 1400 m a.s.l.). We used fossil pollen, non-pollen palynomorphs, sedimentology and charcoal to determine forest dynamics from the Late Pleistocene to the present. From 14,000 to 12,500 cal yr BP the forest was dominated by taxa from higher altitudes, adapted to cooler and drier climates (e.g. Afrocarpus mannii trees and Psychotria nubicola). After 12,500 cal yr BP, a potential uphill migration was identified by an increase in taxa like the trees Symphonia globulifera and Craterispermum cerinanthum. From 11,200 cal yr BP through the rest of the Holocene taxa from lower altitudes became dominant (e.g. Prunus africana, Polyscias, and Sabicea), except at c. 8500 cal yr BP when rapid cooling led to forest opening. Charcoal showed that fires were frequent during the Late Pleistocene (14,000 to 11,200 cal yr BP), becoming rare during the Holocene until anthropogenic fires started at c. 220 cal yr BP. Other recent anthropogenic impacts detected in Lagoa Am´ elia included the appearance of pollen of introduced plant species (e.g., Cestrum), and the increase in pollen of economically important species (Elaeis guineensis, Zea mays) and in fungal spores related to introduced herbivores. Our results reveal that climate changed the altitudinal distribution of the Afromontane forest in S˜ ao Tom´ e during the Late Pleistocene, as observed on the African continent, and that this ecosystem was also strongly impacted by human arrival, through fire, farming, and introduced species. * Corresponding author. Department of Geography and History, University of La Laguna, Spain. E-mail address: [email protected] (A. Castilla-Beltr´ an). Contents lists available at ScienceDirect Quaternary Science Reviews journal homepage: www.elsevier.com/locate/quascirev https://doi.org/10.1016/j.quascirev.2023.108381 Received 4 September 2023; Received in revised form 21 October 2023; Accepted 24 October 2023 Quaternary Science Reviews 322 (2023) 108381 2 1. Introduction Since the Pleistocene, the climate of Central Africa has been characterised by periods of cool and arid conditions (e.g. Last Glacial Period) alternated with humid phases (e.g. African Humid Period, c. 14,800 to c. 5500 BP) (deMenocal et al., 2000; Weldeab et al., 2005). These climate changes are known to have impacted biodiversity and human societies, causing significant ecosystem changes, such as the fragmentation of lowland tropical forests (Maley, 1989) and elevational shifts in vegetation belts (L´ ezine et al., 2019, 2023). Afromontane forests are often distributed in altitudinal bands between 800 and 3500 m a.s.l., and can be identified by characteristic taxa, such as conifer trees from the Podocarpaceae family. Palaeoenvironmental research has shown how Afromontane forests responded to past climate changes and shifting fire regimes. For instance, sediments from Lake Bambili (Cameroon), showed how the upper Afromontane forest treeline currently located at c. 2800 m a.s.l., migrated downslope during glacial cooling and upslope during postglacial warming (L´ ezine et al., 2019). In addition, a record in Mount Kenya (Kenya) showed that during the Late Pleistocene-Holocene transition, Afromontane forests taxa, growing at c. 2000 m a.s.l., shifted to mixed montane forests due to increased forest fires (Rucina et al., 2009). These examples from continental Afromontane forests suggest diverse responses to climate change since the Late Pleistocene. However, it remains unclear if forests on African oceanic islands, such as S˜ ao Tom´ e, in the Gulf of Guinea, exhibited similar dynamics or if, alternatively, they remained stable due to their insularity and the ocean acting Fig. 1. Location of S˜ ao Tom´ e Island and Lagoa Am´ elia (white star). Vegetation map modified after Soares (2017). Lagoa Am´ elia satellite image from Google Earth. Photos of the study site and sedimentary sequence with the detail of the waterlogged wood at 260 cm were taken by the team. Sediment description into first-order size classes after Blott and Pye (2012). 4 cm gap between core sections corresponds to sediment recovered in the percussion corer tip. A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 3 as a buffer to climatic shifts (Fern´ andez-Palacios et al., 2011). The oceanic islands of the Gulf of Guinea include Príncipe, S˜ ao Tom´ e (together constitute the Democratic Republic of S˜ ao Tom´ e and Príncipe), and Annob´ on (belonging to the Republic of Equatorial Guinea). These are part of the Cameroon volcanic line, which has a NE-SW disposition and extends to the continental island of Bioko (also belonging to Equatorial Guinea) and to Mount Cameroon in the African mainland, which is a 4040 m a.s.l. high active volcano, 50 km NE of Bioko. These islands are a biodiversity hotspot and have been called the “African Gal´ apagos". Among them, S˜ ao Tom´ e (854 km 2 ) is home to a remarkably high number of endemics, including c. 150 plant taxa and 40 land vertebrates (e.g., the shrew Crocidura thomensis and the bat Chaerephon tomensis) (Ceríaco et al., 2022a,b; Garcia et al., 2022; St´ evart et al., 2022). Knowledge of the threatened flora and fauna of S˜ ao Tom´ e has been growing in recent years (e.g., https://cepf-stp-threat-flora.netlify. app/). The Afromontane forest of S˜ ao Tom´ e, distributed above 800 m a.s.l., is composed of the montane forest and the mist forest. These forest types (Fig. 1, represented in the vegetation map as ‘native forests’) have been threatened by a rapidly expanding human population (Heleno et al., 2022). S˜ ao Tom´ e Island remained uninhabited until Portuguese settlement in the late 15th century (Garfield, 2015; Mu˜ noz-Torrent et al., 2022). At the crossroads of many Portuguese colonial routes, and having fertile soil and diverse microclimates (Ceríaco et al., 2022a,b) S˜ ao Tom´ e became an experimental ground for agriculture, receiving crops from around the tropics (Mensing et al., 2020). Our limited knowledge of historical land use changes suggests that native forests were extensively cleared, especially in the lowlands (de Lima et al., 2022). In 2006, around one-third of the island area was classified as Obˆ o Natural Park (Direcç˜ ao Geral do Ambiente, 2006), a protected area that covers most of the best-preserved Afromontane forest, where Lagoa Am´ elia, our study system, is located. In this paper, we will assess whether the Afromontane forest of S˜ ao Tom´ e responded to past environmental changes, as described in continental Afromontane forests (e.g. L´ ezine et al., 2013, 2019, 2023, Rucina et al., 2009), and how these have been affected by human colonization of the island. To this end, we used a battery of palaeoenvironmental methods and analysed a 14,000-year sedimentary sequence from Lagoa Am´ elia, a palustrine system crater at 1340 m a.s.l. At the boundary between montane and mist forests, this palustrine wetland represents a promising ecosystem to study the effects of different drivers of change through time. 2. Material and methods 2.1. Site location and fieldwork In June 2021, we conducted fieldwork in Lagoa Am´ elia (1340 m a.s. l., 0◦16 ′ 44.8"N, 6◦36 ′ 30.3"E), a volcanic caldera that contains a palustrine wetland. Attempts to core the unconsolidated floating bog at the centre of Lagoa Am´ elia were unsuccessful. Alternatively, we extracted a 4.58 m deep sediment core from the edge of the lagoon, using a percussion corer (5 cm diameter). The sedimentary sequence was composed of 14 sections or drivers and it is called Lagoa Am´ elia edge (LAE) (Fig. 1). The flora in the study site included the shrubs Myrsine melanophloeos and the endemics Leea tinctoria and Heteradelphia paulowilhelmia, but was dominated by herbs, such as grasses (e.g., Panicum hochstetteri and Oplismenus hirtellus), ferns (e.g., Nephrolepis cordifolia var. pumicicola and Lepisorus excavates), lycophytes (e.g., Lycopodiella cernua), begonias (e. g., the endemics Begonia crateris and B. baccata), and orchids (e.g., Solenangis clavata, and the endemics Rhipidoglossum brevifolium and Dinklageella scandens). Other flora included the climbers (e.g., Anthocleista scandens), the tree fern Alsophila manniana, and the tall native montane forest on the inner slopes of the crater. Lagoa Am´ elia is known as a biodiversity hotspot for ferns (Figueiredo et al., 2011). 2.2. The afromontane forests of S˜ ao Tom´ e The Afromontane forests include the montane (800–1400 m a.s.l.) and mist forests (1400–1800 m a.s.l.) (Exell, 1944; Figueiredo et al., 2011; Dauby et al., 2022). The montane forest is characterized by tree species such as Prunus africana, Symphonia globulifera, Syzygium guineensis, and the endemics Croton stellulifer, Discoclaoxylon occidentale, Homalium henriquesii, Polyscias quintasii, Tabernaemontana stenosiphon, and Trichilia grandifolia, as well as understory species, such as Oxyanthus speciosus, Piper umbellatum, and the endemics Alsophila welwitschii, Begonia subalpestris, Craterispermum cerinanthum, and Erythrococca molleri (Oliveira, 1993; Figueiredo et al., 2011). The mist forest is characterized by the endemic conifer tree Afrocarpus mannii, and other trees such as Melchiora mannii, Olea capensis, Peddiea thomensis, Psychotria guerkeana, and P. nubicola, with an understory dominated by Begonia spp., Pilea spp., and Tristema spp., and the climber Anthocleista scandens. Above these vegetation belts (over 1800 m a.s.l.) there is a low forest, which includes scrublands (with the endemic shrub Erica thomensis), and grasslands (with herbs such as the endemic Lobelia barnsii) (Dauby et al., 2022). In contrast to the more humid montane forests and mist forests, this low forest has shallow soils and, at the peak of the dry season, can become dry enough to be vulnerable to fire (Monod, 1960). 2.3. Laboratory methods Once transported to the Ecology Laboratory at the University of La Laguna (Canary Islands), we photographed the sediment core sections, described its colour and macroscopic characteristics, and carried out a sampling of 1 cm 3 samples, isolating finds of interest, such as rock fragments and macro-remains (e.g., waterlogged woody material, Fig. 1). We carried out multiple analyses in a set of 49 samples: 1) fossil pollen and fern spores to track changes in regional vegetation; 2) nonpollen palynomorphs (NPPs), namely fungal spores to characterise changes in decomposers communities and coprophilous fungi that can identify the presence of herbivores (e.g., bovids, goats, etc.) (Cugny et al., 2010); 3) charcoal particles to study the occurrence of local (macro-charcoal particles over 150 μ m) and regional fires (micro-- charcoal particles under 150 μ m). Finally, in another set of 45 samples we measured granulometry, sediment elemental composition, and organic content through loss on ignition (LOI), to study the dynamics of sediment deposition, erosion, and organic matter accumulation into the crater basin, respectively. 2.3.1. Palynomorph identification and counting and statistical analysis For palynomorph sample preparation, we spiked the samples with exotic Lycopodium spores (batch 1031), sieved (using 150 μ m and 10 μ m meshes), floated with sodium polytungstate at 1.9 gravity, acetolysed, and mounted in silicon oil (Erdtman, 2011). Slides were inspected at ×400 magnification, counting microfossils until a sum of at least 300 pollen grains was reached, or over 150 in samples with low pollen concentration and/or poor preservation. For NPPs quantification, at least 300 fungal and fern spores were counted. To carry out fossil pollen identifications, we used a pollen reference collection assembled from plant samples stored at the S˜ ao Tom´ e and Príncipe National Herbarium (STPH) and the Coimbra Herbarium (COI). The S˜ ao Tom´ e pollen reference collection, currently located at the University of La Laguna’s Ecology Laboratory, includes over 60 taxa, most distributed in Afromontane forests, and other taxa native to the island (see Table S4). We checked the taxonomy using the African Plant Database (APD, v3.4.0) and MBG Tropicos (v3.3.2). To complement the information from the pollen reference collection, we used regional literature from tropical Africa, such as pollen atlases (Gosling et al., 2013; Schueler and Hemp, 2016; Rasoloarijao et al., 2019) and NPPs research literature (e.g. Gelorini et al., 2011). We established the correspondence between pollen types and the species occurring in S˜ ao Tom´ e (Table S2). For analysis of pollen assemblage, all pollen grains were included in A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 4 the pollen sum, including potential wetland taxa since in some cases these could not be distinguished from terrestrial species (e.g. some pollen assigned to Cyperaceae). To divide the record into zones, we performed a CONISS stratigraphically constrained analysis using the packages Rioja and Vegan in RStudio (Oksanen et al., 2013). We calculated NPPs abundance as percentages over the pollen sum (Cugny et al., 2010). We also calculated microfossil concentrations relative to the known amount of exotic Lycopodium spores. We used the package Vegan to calculate detrended correspondence analysis (DCA) and used axis 1 as an indicator of pollen assemblage turnover through time (e.g. Bush et al., 2004; Blarquez et al., 2014; Castilla-Beltr´ an et al., 2021; Nogu´ e et al., 2021). In addition, we carried out a rarefaction analysis based on pollen counts to calculate palynological richness using the package Hotelling and plotted the results with a linear regression using RStudio (Curran, 2018). 2.3.2. Macroand micro-charcoal analyses For macro-charcoal analysis, we used a set of 49 1-cm 3 sediment samples sieved using a 150 μ m mesh and counted charcoal fragments under a low magnification microscope. Samples rich in organic matter were bleached using a low concentration of H 2 O 2 , to distinguish charcoal from other dark organic particles. For micro-charcoal analysis, we counted opaque, black angular pieces between 5 and 100 μ m, alongside fern spores and Lycopodium spores in the pollen microscope slides (Finsinger and Tinner, 2005). We aimed for >100 elements between Lycopodium spores and micro-charcoal or stopped counting after reaching >300 Lycopodium spores in samples with minimal charcoal content. Micro-charcoal concentrations were estimated relative to the known concentration of exotic Lycopodium spores. Macroand micro-charcoal results are displayed as particles per cm 3 (Fig. 2). 2.3.3. Sedimentological analyses We dried and homogenized a set of 45 samples (1 every 10 cm) for elemental composition, granulometric analysis, and loss on ignition (LOI). For elemental composition analysis, we carried out 160-s measurements using a hand-held Fluorescence X-Ray using a Niton XL3T GOLDD mounted on a base. The semi-quantitative results are expressed as detected elements in percentages and ppm, presented as ratios for palaeoenvironmental interpretation, such as the Ti/Ca ratio, which can be interpreted as an indicator of detrital inputs into lake basins (Croudace and Rothwell, 2015). The remaining material of this non-destructive technique was used for granulometric analysis, using a Malvern Mastersizer-Hydro at the School of Geography and Environmental Science, University of Southampton, with soil analysis settings adjusted to non-spherical particles. 20-s measurements were averaged if the standard deviation was below 5%, which was achieved by adjusting the obscuration parameter. Results are expressed per fraction (Dx10, Dx15, Dx90) and as percentages of clay, silt and sand, including subcategories such as fine and coarse sand. To estimate organic matter content, we ashed the dried material of the same set of samples at 550 ◦C for 4 h and calculated weight LOI using a high precision balance by weighing samples before and after (Heiri et al., 2001). 2.3.4. Radiocarbon dating and chronological modelling To develop a chronological model, we dated 11 samples using AMS radiocarbon dating at three different laboratories: Queen’s University Belfast, Scottish Universities Environmental Research Centre (SUERC), and BETA analytics (Table 1). We selected and sieved the sediment Fig. 2. Multi-proxy summary diagram showing sedimentological data (LOI, ratio Ti/Ca, granulometry), charcoal (microand macro-charcoal), fossil pollen and NPPs concentrations, DCA axis-1 results, rarefaction analysis, and pollen zones for Lagoa Am´ elia (LAE) core. A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 5 samples to isolate macrofossils that could yield precise dates. In the absence of seeds, we aim for wood cellulose and twigs. Due to bad macrofossil preservation, we dated bulk sediment samples for section 40-230 cm (Table 1). We used the package RBacon in RStudio to develop an age-depth model based on Bayesian statistics (Blaauw and Christen, 2011), using nine radiocarbon dates. Bulk dates from levels 140 and 60 cm (SUERC-104996 and SUERC-104997), were excluded from the age-depth model since the resulting 14 C dates were showing inversions likely due to the input of old or minerogenic carbon from the volcanic caldera walls in this section of the record (Bj¨ orck and Wohlfarth, 2001). The RBacon age-depth model results were used to calculate the sedimentation rate and the mean age (cal yr BP) of each sample. 3. Results 3.1. Paleoenvironmental description The resulting age-depth model suggests that the Lagoa Am´ elia sedimentary sequence was deposited during the last 14,000 cal yr BP. Based on stratigraphically constrained CONISS analysis, the record was divided into 8 different zones (LAE-1 to LAE-8), including 2 zones with a single sample each (LAE-3 and LAE-5). 3.1.1. LAE-1, 450-415 cm (14,000–13,500 cal yr BP) Zone LAE-1 showed medium levels of organic matter content (avg. 18% LOI), and low values of the Ti/Ca elemental ratio. Stones were very rare, median grain size increased throughout the zone from 57 to 159 μ m, and coarse sand was absent or present in very low percentages (maximum of 0.75%) (Fig. 2). Sedimentation was the slowest in the record (of c. 10 yr per cm 3 ) (Fig. S1). Pollen was well preserved, with high concentration values (up to 240,000 grains/cm 3 ) except for the 420 cm deep sample that showed low pollen concentration (c. 2600 grains/cm 3 ). The assemblage displayed high percentages of Afrocarpus mannii (avg. 10%), Olea capensis (avg. 9%), Craterispermum cerinanthum (avg. 5%), Celtis spp. (avg. 6%), Trema orientalis (avg. 3%), Moraceae/Urticaceae (avg. 3%), as well as high values of Asteraceae (avg. 8%) and Poaceae (avg. 3.5%). Alchornea, Begoniaceae, Commelinaceae, Homalium, Macaranga, Piperaceae, Uapaca, and Utricularia pollen was also present, and pollen of Elaeis guineensis was found at level 440 cm (Fig. 3). Rarefied species richness values were some of the lowest in the record compared to other pollen zones (avg. 28), and DCA axis-1 pollen scores oscillate thought the zone at values around −0.5. The fern spore assemblage was dominated by Cyatheaceae tree-fern spores (Alsophila, up to 230% relative to the pollen sum), as well as morphotype monolete psilate (up to 430% relative to the pollen sum), with variable levels of other fern spores (Woodsiaceae, Polypodiaceae, and Pteridaceae). There was a high diversity of fungal spores, including spores of coprophilous fungi, such as Cercophora, Chaetomium, Coniochaeta, Delitzchia, Podospora, Sordaria, Sporormiella and, the most abundant being Podospora (up to 6%/pollen sum). We recorded a high percentage of the saprophytic/parasitic fungi Rosellinia (avg. 10%/pollen sum) (Fig. 4). There was a peak in macro-charcoal particle concentration at 450 cm depth (88 fragments/cm 3 ). Micro-charcoal concentration was consistently high, varying between 3037 particles/cm 3 at level 420 cm to 32,307 particles/cm 3 at level 440 cm (Fig. 2). 3.1.2. LAE-2, 415-325 cm (13,500–11,350 cal yr BP) Zone LAE-2 held medium levels of organic content (avg. 15% LOI), with few rocks and gravel, despite an increasing trend of median grain size (ranging from 82 to 207 μ m), and peaks up to 6% of coarse sand at levels 370, 340, and 330 cm. The pollen assemblage showed continuity from LAE-1, yet held lower pollen concentration values (avg. 50,000 grains/cm 3 ). DCA axis-1 scores increased from −0.64 to −0.36. We recorded a marked increase in Symphonia globulifera (with peaks of up to 13 and 16%), an increase in Craterispermum cerinanthum, and decreases in Olea capensis, Celtis and Afrocarpus mannii (down to 6%, 3% and 1% respectively). From level 380 cm–370 cm there was an increase in Sabicea spp. (avg. 6%), Clematis hirsuta (avg. 4%), and Psychotria-type (up to 15%). There was also a moderate increase in Commelinaceae, and a moderate decrease in Poaceae (both taxa up to 3%). At the end of the zone (360-330 cm), we recorded punctual increases in Apocynaceae and Vigna-type, as well as of Plantago. Ferns showed a similar assemblage to LAE-1, with peaks in Alsophila manniana and monolete psilate fern spores in sample 360 cm (370% and 670% over the pollen sum, respectively). Overall, this section records some of the highest rarefied species richness values of the record (e.g., 36 at level 380 cm). Coprophilous fungi percentages also increased around levels 360340 cm, and there were small peaks in Tetraploa fungal spores (up to 3%/pollen sum), as well as sustained decreases in Rosellinia and Kretzschmaria fungal spores, with peaks around 360-340 cm. We detected an overall high level of macroand micro-charcoal concentrations, with peaks in macro-charcoal at levels 390 cm and 350 cm reaching 92 and 86 particles/cm 3 , respectively, and in microcharcoal at level 360 cm (59,000 particles/cm 3 ). 3.1.3. LAE-3, 320 cm (11,100–10,850 cal yr BP) Zone LAE-3 (sample 320 cm) contained a high percentage of organic material (58% LOI), and no coarse sand, gravel, or rocks. Pollen concentration was high (c. 214,000 grains/cm 3 ), and DCA axis-1 value was −1.06, the lowest value of the record. The pollen assemblage was strongly dominated by Trema orientalis (44%), while the rest of the assemblage remained stable, even though it was probably underrepresented due to the hyperabundance of Trema. We also recorded Lobelia and Vigna-type taxa, as well as a decrease in the proportion of fern spores. Coprophilous fungi were present, showing an abundance of Sordariaceae spores, as well as high levels of Rosellinia (10%/pollen sum). Charcoal concentrations were low, 14 macro-charcoal particles/cm 3 , and 2500 micro-charcoal particles/cm 3 . 3.1.4. LAE-4, 315-255 cm (10,850–9200 cal yr BP) Zone LAE-4 was characterized by medium levels of organic content (avg. 13%), except for a pronounced peak at level 260 cm (82% LOI), Table 1 Radiocarbon dating of Lagoa Am´ elia core (LAE). Data in bold was used to develop the age-depth model. Code Material Depth (cm) Radiocarbon age Error Calibrated yr BP (% 95.4) UCIAMS264635 Twig 20 264 35 450–151 Beta651538 Wood cellulose 30 130 30 275–8 SUERC104996 Bulk 60 7110 37 8014–7850 SUERC104997 Bulk 140 7847 39 8930–8540 SUERC105002 Bulk 200 3702 36 4151–3925 SUERC104991 Wood cellulose 258 8188 40 9276–9018 SUERC104992 Wood cellulose 292 9134 39 10481–10224 SUERC104993 Wood cellulose 320 9769 42 11255–11113 SUERC104994 Wood cellulose 385 8289 40 9428–9132 Beta651539 Wood cellulose 411 11690 40 13607–13458 UBA46606 Wood cellulose 450 11752 50 13752–13500 A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 6 with very variable values of median grain size (ranging from 113 to 202 μ m), sand content (maximum 3% at level 300 cm) and Ti/Ca (ranging from 0.2 to 8.2%). The sedimentation rate gradually slowed down from 30 yr/cm 3 to c. 50 yr/cm 3 . This zone showed a transition of DCA axis-1 pollen scores, from −0.25 to −0.05. There was a decline (or absence) in dominant pollen types from zones LAE-1 and LAE-2 (e.g., Afrocarpus, Olea, Craterispermum, Symphonia), while Psychotria-type (up to 15%), Hymenodyction (up to 6%), Psydrax (up to 4%), Sabicea (up to 6%), and Clematis hirsuta (up to 8%) showed increases. Asteraceae, Cyperaceae, Commelinaceae, and Poaceae also showed increases (up to 18%, 11% and 4% respectively). Trema orientalis pollen decreased until it disappeared from the assemblage. There were punctual increases in Discoclaoxylon occidentale (6% at level 260 cm). The fern spore assemblage was dominated by monolete psilate spores. There was a decrease in fungal spores, except for sample 260 cm, where coprophilous types were present, as well as abundant Rosellinia spores. There were generally lower concentrations of macro-charcoal particles (below 15 particles/cm 3 ) and micro-charcoal (c. 2000 particles/ cm 3 around 310-300 cm, decreasing in 290–270 cm, and then increasing to over 3000 at 260 cm). 3.1.5. LAE-5250 cm (c. 8500 cal yr BP) Zone LAE-5 (sample 250 cm) showed medium levels of organic content (21% LOI), low median grain size (89 μ m), and no rocks or coarse sand. Fig. 3. Pollen diagram showing the percentage of taxa categorized as canopy trees, small trees, climbers, shrubs, and herbs, over 14,000 cal yr BP for Lagoa Am´ elia (LAE) core. The taxonomic selection is based on habit category (see Table S2), and the asterisk symbols denote taxa that can be assigned to two or more categories (e. g. Asteraceae, includes shrubs and herbs). For curves of rare taxa (<5%), we used exaggeration ×5 (lighter colour). Taxa that did not reach a sum of 10 pollen grains in the record were excluded from the diagram. A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 7 Pollen concentration was high (112,000 grains/cm 3 ), its DCA axis-1 score −0.49, and the pollen assemblage was very distinct, with increases in Lobelia-type (12%) and Rosaceae1-type (5%), and abundant Ficus (9%) and Moraceae/Urticaceae (5%). Percentage values of Poaceae were low (0.5%), and Lannea-type and Apocynaceae pollen were present. We recorded the lowest proportions of fern spores relative to pollen, with a dominance of monolete psilate types (80%/pollen sum). The highest abundance of coprophilous (e.g., Cercophora at 14%) and Rosellinia (54%) and Kretzschmaria (5%) fungal spores so far in the record were detected in this zone, with high concentration value (175,000 spores/cm 3 ). There is an abundance of micro-charcoal particles (estimated 62,100 particles/cm 3 ), and moderate macro-charcoal concentrations (28 particles/cm 3 ). 3.1.6. LAE-6, 245-115 cm (8500-~2750 cal yr BP) Zone LAE-6 had relatively low organic content (avg. 13%), various peaks in median grain size (highest in level 130, at 205 μ m), sand content (also highest in level 130 cm, at 78%) and Ti/Ca ratio (highest in section 140-120 cm). Coarse sediments, such as rocks and gravel, were also found. Sedimentation in this zone varied from c. 80 yr to c. 30 yr per cm 3 . Pollen preservation was variable, with poor preservation at levels 140-120 cm (pollen sums <150 grains), and rarefied species richness generally showing low values (avg. 29). Pollen and NPPs concentrations were stable, except high concentrations in sample 190 cm (207,000 and 2,033,000 grains/cm 3 respectively). Pollen assemblage resembled LAE4, from which it could be distinguished by the small increase in Olea capensis (200-190 cm), the increase in Lamiaceae, and the sustained presence of Prunus africana, Lannea-type, Impatiens, and Vitaceae. The fern spore assemblage was dominated by monolete psilate types, with an increasing percentage of Alsophila manniana and Polypodiaceae spores. Fungal spores were mostly absent. Micro-charcoal estimates showed a peak at level 230 cm (c. 5000 particles/cm 3 ), and then a decreasing trend, below 1000 particles/cm 3 , while macro-charcoal particles were always very rare or absent. 3.1.7. LAE-7, 115-35 cm (2500-470 cal yr BP) Zone LAE-7 showed stable low organic content (avg. 14%), but peaks in median grain size (up to 161 μ m), coarse sand content (up to 4%) and presence of rocks. It showed the highest values of Ti/Ca (up to 17), as well as the presence of gravel and rocks, especially between 75 and 45 cm. Sedimentation rate was stable at c. 35 yr/cm 3 . Pollen preservation varied from a high abundance of degraded pollen at level 110 cm to excellent preservation at level 90 cm. The pollen assemblage is similar to zones LAE-4 and LAE-6. Key differences included Clematis and Sabicea reaching maximum values (up to 23% and 15%, respectively), a peak in Symphonia globulifera (8% at level 60 cm), and the sustained presence of Uapaca-type, Trichilia, Campylospermum, Impatiens and Oxyanthus speciosus pollen. Poaceae and Cyperaceae pollen decreased, yet Asteraceae remained stable. Fern spores were still dominated by monolete psilate types, despite a marked peak of other monolete types (cf. Woodsiaceae) at level 90 cm and of Alsophila manniana at level 70 cm, and an overall increase in Lycopodiaceae spores. However, there were very few fungal spores, limited to isolated occurrences of saprophytes such as Coniochaeta and Rosellinia. Macro-charcoal concentration varied between zero and 4 particles/ cm 3 , while micro-charcoal stayed below 1000/cm 3 . 3.1.8. LAE-8, 35-0 cm (250 cal yr BP -present) Zone LAE-8 showed increasing values of organic content (ranging from 17% to 90% in the surface sample), while grain size values, sand content percentage, and Ti/Ca ratio showed a steady decrease, and rocks became rare. This zone sees an acceleration of the sedimentation rate from c. 30 yr–10 yr per cm 3 (Fig. S1), increasing pollen concentration values (up to 287,000 grains/cm 3 ), DCA axis-1 values (from 0.65 to 1.24), and palynological richness values (up to 37, highest of the record). In contrast to the previous zone, Clematis, Hymenodyction, Prunus africana, Trichilia, and Psychotria-type showed an overall reduction in abundance, while Poaceae and Moraceae/Urticaceae increased (up to 13 and 12%, respectively). There were moderate increases in Nuxia-type, Tabernaemontana stenosiphon, Sterculia, and Homalium henriquesii. Fig. 4. Non-pollen palynomorphs (NPPs), including tree ferns (pink), other ferns (grey) and fungal spores (green), plotted over 14,000 cal yr BP for Lagoa Am´ elia (LAE) core. NPPs percentages were calculated over the pollen sum. A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 8 Scheffera-type, Rosaceae (cf. Rubus) and Commelina-2 showed a sustained presence, and Elaeis guineensis and Zea mays were also present. This section shows the highest values of rarefied species richness. As pollen concentration rose, the relative percentage of fern spores decreased. The fungal spore assemblage was diverse, showing increases of Rosellinia (up to 54%/pollen sum), and abundant coprophilous fungi (e. g., Cercophora peaked at 11% at level 10 cm). There were also increases in the presence of Sporochisma (7%), Meliola (28%), Sporidesmium, and Isthmospora spinosa (up to 19%). Micro-charcoal particles showed a moderate increase, up to 7000/ cm 3 at level 10 cm, where macro-charcoal concentrations also peaked, at 28 particles/cm 3 . 4. Discussion 4.1. Sedimentation process, chronological framework, and bog formation in Lagoa Am´ elia The Lagoa Am´ elia Edge (LAE) sediment core studied here was deposited during the last 14,000 cal yr BP, following a complex process that led to clearly differentiated facies. Based on mineral, organic and sediment elemental composition (e.g., granulometry, XRF, LOI data), the sequence can be divided into sections dominated by fine and organic sediments, coarse sediment, and with presence of gravel (Figs. 1 and 2). Sedimentation rates indicate slower deposition on the bottom section of the record (458-250 cm), but changes in grain size and composition towards coarser material in the upper section of the record (250-30 cm) strongly suggest that sedimentation rates were shifting depending on the stability of the crater slopes (Fig. 2). This is in accordance with the Ti/Ca elemental ratio), which shows higher values in the upper section of the core, suggesting increased detrital inputs from the slopes that could have been caused by increased aridity potentially linked to more pronounced seasonality after c. 6000 cal yr BP. The record also indicates local bog formation at five brief intervals: 14,000–13,900 cal yr BP, 12,700–12,600 cal yr BP, 11,200–10,800 cal yr BP, 9400–8400 cal yr BP, and 200 cal yr BP-present day. During these periods there was abundant preservation of organic matter (up to 90%), with high levels of pollen concentration and well-preserved wood fragments (Figs. 1 and 2), and an increased percentage of Poaceae and Cyperaceae pollen (Fig. 3), spores of saprophyte soil fungi (e.g. Rosellinia), and herb parasite fungi (e.g. Tetraploa) (Fig. 4). These plant and fungal communities are representative of locally abundant herbaceous taxa, suggesting that peat bogs were developing in the surface and edges of the crater lake. This situation is analogue to the present-day state of the lake, in which the floating bog is mostly constituted by herbaceous species (Fig. 1). Being a sequence formed by a complex sedimentation process, the LAE age-depth model also deserves discussion. We excluded two radiocarbon dates (SUERC-104996 and SUERC-1049979) representing inversions. We interpret the inversions as old carbon from the crater walls being deposited in rapid erosive events (Bj¨ orck and Wohlfarth, 2001; Grimm et al., 2009) (Fig. S1). This explanation aligns well with the acceleration of sedimentation detected at 140 and 60 cm. In addition, fast-paced deposition of mineral material is also the most likely cause of the ‘outlier’ radiocarbon dating results at level 200 cm (SUERC-105002, Fig. S1) that contrasts with an otherwise linear model. Finally, the date at 385 cm (SUERC-104994) yielded a younger age than expected, of which the most likely explanation is the intrusion of roots from vegetation growing at the edge of the lake. 4.2. Late Pleistocene to holocene transition: from mist to montane forest Has the montane forest of S˜ ao Tom´ e been stable during the global climatic changes occurring during the past 14,000 years, or has it shifted like continental Afromontane forests (Table 2)? The Lagoa Am´ elia Table 2 Summary table of African mainland palaeoecological records cited in the text that show Afromontane responses to climate change in the Late Pleistocene and Holocene periods, including four records from Cameroon and four records of central and eastern Africa (see Table S3 for full details, including marine records). Record Reference Pleistocene-Holocene change Bambili twin crater system (NW Cameroon) ~2300 m a.s.l. L´ ezine et al. (2019) Increases in Podocarpus, Olea, Celtis, Alchornea, and Clematis pollen represent an uphill treeline expansion during the PleistoceneHolocene transition. During the Heinrich 1 event (c. 16,000 yr BP) there was a severe drought, and the expansion paused, even reversing. The upslope movement resumed around 15,000 years ago, reaching the modern-day tree line elevation by the onset of the Holocene around 11,000 years ago. Mbi crater lake (Cameroon) 2015 m a.s.l. L´ ezine et al. (2023) Early stages of tree expansion at Mbi started around 14,500 cal yr BP. Another phase of forest development occurred between 13,500 and 12,500 cal yr BP. However, this expansion was interrupted by a dry event (Younger Dryas). The transitions between the B¨ olling/Allerød and Younger Dryas events were more clearly observed at Mbi compared to Lake Bambili. During the onset of the Holocene (11,500–10,800 cal yr BP), there was a decrease in Poaceae and an increase in forest pollen, including Schefflera, Alchornea, Olea, Nuxia, and Podocarpus. Lake Mbalang (Cameroon) m a.s.l. (1110) Vincens et al., (2010) The transition from a forested environment to an open savannah took place due to long-term aridification following the African Humid Period. The transition began around 6100 BP, when the percentages of pollen from Podocarpus, Olea capensis, and Alchornea started to decrease. The modern savannah environment (increase in Poaceae pollen) became established around 3000 BP. There was a brief phase of forest regeneration between 5200 and 4200 cal yr BP. Lake Barombi Mbo (Cameroon) 300 m a.s.l. Maley and Brenac (1998) The representation of Olea capensis pollen indicates a relatively cool climate until approximately 13,000 yr BP, c. 10,000–2800 yr BP, the climate became very wet, and the representation of Poaceae pollen sharply decreased, while forest trees reached their maximum extension between 9500 and 3000 yr BP. Cores KH3 and KH4, Lake Tanganyika (Tanzania, DR Congo), 655 m a.s.l. Ivory and Russell (2016) Between 15,000 and 11,000 years ago, there was a significant decline in previously abundant Afromontane trees such as Podocarpus and Juniperus procera, while Olea reached its maximum value. This period also saw an increase in Poaceae and Artemisia pollen, marking the beginning of the transition from Afromontane forest to the dominance of herbs. From 11,000 to 7200 years ago, Olea decreased while most lowland tree taxa reached their maximum (continued on next page) A. Castilla-Beltr´ an et al. Quaternary Science Reviews 322 (2023) 108381 9 palaeoenvironmental record (LAE) reveals that the montane forest of S˜ ao Tom´ e has shifted during the Late Pleistocene and the onset of the Holocene (c. 13,000 and 11,200 cal yr BP), both in altitudinal distribution and in species composition, in response to distinct environmental stressors. Between 14,000 and c. 12,500 cal yr BP the forest around Lagoa Am´ elia was dominated by Afrocarpus mannii, Olea capensis, Psychotria nubicola, and Celtis (Figs. 3 and 5). These trees are currently found in montane forest, A. mannii and P. nubicola being restricted to higher elevations, namely to mist forest on ridges (Dauby et al., 2022). High concentrations of macroand micro-charcoal during the Late Pleistocene (Figs. 2 and 5) must evidence frequent fires around Lagoa Am´ elia and cannot be explained by volcanism, since the last eruptions in S˜ ao Tom´ e occurred c. 1.5 Ma (Lopes, 2020). A first shift in forest composition 13,000 cal yr BP, consisted of an increase in Symphonia globulifera, a tree currently associated with montane forest (800–1400 m a.s.l.) even though it can also be found on ridges at lower altitudes (Dauby et al., 2022). Another species that also showed higher frequencies at this time was the small tree Craterispermum cerinanthum, another typical montane species (Figueiredo et al., 2011) that can also be found at lower altitudes (Fig. 3). Both species became abundant at Lagoa Am´ elia when palynological richness (values up to 36), and vegetation turnover were high, and fires were still frequent, suggesting fast species replacement. From 13,000 onwards, the regional climate warmed moderately and became relatively dry (Weldeab et al., 2005; Collins et al., 2017). LAE sedimentological data showed increases in the Ti/Ca ratio c. 12,500 cal yr BP, indicating detrital input to the crater lake, which could evidence the synchronicity with regional hydrological changes during this dry period (Weldeab et al., 2005). In addition, the record also shows local bog formation occurring on the edges of the palustrine area, namely c. 14,000, 12,600, and 11,200 cal yr BP, which could be related to dry conditions due to the potential intermittent desiccation or at least to low water levels. The documented changes during the Late Pleistocene in S˜ ao Tom´ e Afromontane forests are best understood in comparison to mainland African records. In western Cameroon, montane forest taxa such as Olea capensis, currently inhabiting high elevations in areas with relatively low temperatures, were especially abundant at lower altitudes during the cool and dry glacial periods between 28,000 and 13,000 cal yr BP (Maley and Brenac, 1998). Similarly, the abundance of Olea capensis and Afrocarpus mannii recorded at LAE suggests they were more abundant around the crater during the Late Pleistocene. The global deglaciation brought changes in the distribution and composition of Afromontane forests in tropical Africa: between 15,000 and 11,000 yr BP the Afromontane treelines in Northwest Cameroon migrated upslope to reach modern-day elevations around Lake Bambili (L´ ezine et al., 2019), and forests expanded in the area of Mbi crater lake after 14,500 cal yr BP (L´ ezine et al., 2023). However, the process of forest migration (expansion and contraction) was interrupted during periods of rapid climatic change, such as the Younger Dryas (L´ ezine et al., 2013). There is also evidence that fire regimes were changing, for instance, in Mount Kenya (study site at c. 2000 m a.s.l.), increased forest fires during the Late Pleistocene-Holocene took place along with forest shifts from montane to mixed montane forests (Rucina et al., 2009). Our data supports a similar process, in which Late Pleistocene dry conditions led to more frequent fires near Lagoa Am´ elia, likely due to lightning strikes during the dry season (Ceríaco et al., 2022a,b). In sum, these shreds of evidence suggest that the dynamics of island Afromontane forests during the Late Pleistocene deglaciation might have been analogue to those in the continent since the Afromontane forests of S˜ ao Tom´ e were shifting according to regional climate conditions during this period. Between 14,000 and 11,200 cal yr BP mist forest was likely distributed at lower elevations than in the present (1400 m a.s.l.) due to cooler and dryer climatic conditions. Forest fires held a role in these shifts and are linked to dryer regional conditions. Although we cannot rule out other types of processes, such as compositional changes not related to altitudinal shifts, our results suggest that dry tolerant taxa of the Afromontane forests were distributed at lower elevations during the Late Pleistocene, and gradually moved upwards as the climate warmed and fires frequency increased during the deglaciation period (Table 2). 4.3. S˜ ao Tom´ e’s afromontane forest during the holocene The end of the Younger Dryas brought increased precipitation in Central Africa (Collins et al., 2017). Afterwards, the Holocene (c. 11,700 BP) was characterised by warmer and more humid environmental conditions (Weldeab et al., 2005), which is considered a main driver of the current distribution of Afromontane forests (White, 1993). In addition, the temperature during the Early Holocene also shifted between cooler and warmer conditions in the Eastern Equatorial Atlantic, where S˜ ao Tom´ e is located, with sporadic cooling episodes, such as the one that took place between 9000 and 8000 cal yr BP (Weldeab et al., 2005). During the onset of the Holocene (11,700 cal yr BP) the environmental conditions in Lagoa Am´ elia, were characterised by infrequent fires, likely due to high humidity, the consolidation of the cloud belt, and increased rainfall (Fig. 5). There is a period c. 11,200 cal yr BP (LAE 3, one sample), when the forest showed a sharp peak in the abundance of Trema orientalis trees, currently considered a pioneer species associated with disturbance (Fig. 3). After this peak, the forest changed through an increased abundance in canopy trees (e.g., Prunus africana and Polyscias quintasii), and diverse climbers (Sabicea spp., Clematis hirsuta). Importantly, most of these taxa are currently more abundant in submontane forest (800–1400 m a.s.l.), suggesting an uphill migration of this forest type after 11,200 cal yr BP. At the same time, the abundance of mist Table 2 (continued) Record Reference Pleistocene-Holocene change percentages, including Alchornea, Bridelia, Macaranga, Myrianthus holstii, and Uapaca, signifying the final decline of all Afromontane taxa. Virunga volcanoes (Rwanda) 3474 m a.s.l. McGlynn et al. (2013) Around 5000 cal yr BP, there was a decline in ericaceous vegetation and an expansion of Afroalpine vegetation, followed by an increase in taxa associated with lower montane forest, particularly Podocarpus. These changes in vegetation reflect increasing aridity during the Midto Late Holocene. Lake Rukwa (Tanzania) 800 m a.s.l. Vincens et al. (2010) During the deglaciation period (16,500 to 12,300 yr BP), the climate transitioned to warmer conditions, leading to a retreat of Afromontane taxa at higher altitudes on the plateau. This was accompanied by the expansion of wooded formations in the region. Around 12,100 yr BP, the local woodland and bushland experienced maximum development and diversity, indicating an increase in rainfall Mount Kenya (Kenya) 2154 m a.s.l. Rucina et al. (2009) The record shows reduced levels of Hagenia, Juniperus, Olea, and Podocarpus around the time of the Younger Dryas. During the Late Pleistocene to Holocene transition, a variable fire regime drove the reorganization of the ecosystem composition in the montane forest. This shift reflects the onset of a warmer and moister climate from the onset of the Holocene, as mixed montane forests became more established. A. Castilla-Beltr´ an et al.