West African rainforests are amongst the world’s premier biodiversity hotspots. Over time, the size and distribution of these rainforests have changed significantly due to fluctuations in climate. However, compared to other biodiversity hotspots, our understanding of potential West African rainforest refugia remains relatively limited. Studies from different disciplines have provided valuable insights into refugia location, age and evolutionary role. Fossil pollen data and comparative biogeography studies have revealed cycles of rainforest contraction and expansion linked to aridification and have used these patterns to identify large-scale refugia. Phylogeographic studies mostly corroborated the existence and locality of large-scale refugia, but also unearthed the putative presence of microrefugia; smaller isolated regions that jointly form larger refuge zones (macrorefugia). Moreover, some refugia seem to harbour deep evolutionary lineages, potentially reflecting long-term stability, whereas others may have been stable during more recent aridification cycles. Phylogeographic splits are congruent across species, but asynchronous and frequently align with both climatically unstable regions and landscape features, such as rivers. These temporal and biogeographic aspects have been further explored using demographic and Species Distribution Modelling (SDM). Across various rainforest taxa, these methods show that populations and suitable habitat retracted and expanded, promoting macro-evolutionary change. Climatic fluctuations thus appear to have played an important role in shaping West African biodiversity. Overall, research shifted from identifying refugia to quantifying their role in driving evolutionary change. However, the scarcity of studies linking detailed changes in demography with SDM limits our ability to make general statements regarding refugia dynamics, predict species response to climate change and prioritise future climate refugia.

biogeography, climate refugia, conservation, diversifi­cation, genomics, palynology, phylogeography, refugia dynamics, species distribution modelling

Tropical rainforests only cover around 10% of the Earth´s surface, but are global epicentres of biodiversity and home to over two-thirds of all known species (Groombridge and Jenkins 2002; Lewis 2005). High rates of habitat loss and overexploitation are posing a serious threat to these biodiversity hotspots (Bradshaw et al. 2009) and the rising temperature and shifting precipitation patterns caused by climate change are adding further pressure (James et al. 2013). Predicting the effects of climate change on complex ecosystems such as rainforests remains difficult (Zhou et al. 2013; Feng et al. 2018). However, it is not the first time that the Planet has undergone drastic changes in climate and much can be learnt from the past (McGuire et al. 2023). For example, Pleistocene glaciations also led to radical changes in temperature and precipitation (Bennett et al. 1991; Chiang 2009). Many ecosystems and even entire biomes have been impacted by these fluctuations. Some species experienced range shifts or extinctions (Lorenzen et al. 2011) while, at the same time, these environmental changes also promoted species diversification (Svenning et al. 2015). In the Tropics, the climate became more arid and cooler during glacial times, which presumably led to the fragmentation and contraction of rainforests into so-called climate refugia (Haffer 1969; Carnaval et al. 2009; Hardy et al. 2013). Here, we define rainforest refugia as regions of climatic stability, where rainforests and their biota persisted during cold and arid periods and from which they may have expanded once conditions improved. Understanding how these large-scale environmental changes impacted species helps us to predict their resilience and adaptability against future climate change and provide information for predictions about their distribution (Bellard et al. 2012; Dietl et al. 2015; McGuire et al. 2023). Refugia are, therefore, key conservation targets and can enhance management strategies due to their capacity to function as a buffer, mitigating the adverse impacts of climate change on species and ecosystems (Keppel and Wardell-Johnson 2012; Keppel et al. 2012; Morelli et al. 2020).

Studies on climate refugia have overwhelmingly focused on the Northern Hemisphere (Keppel et al. 2012), but the existence of refugia in the Tropics has also been intensely discussed. Extensive research on potential refugia in Amazonia suggests that changes in vegetation distribution did happen in response to climatic fluctuations (Cowling et al. 2008). However, this likely did not lead to massive forest dieback forcing all species to retract uniformly (Bush 1994; Cowling et al. 2008). Furthermore, the impact of Pleistocene glacial periods was not consistent across the entire Amazon Basin. For example, while climate in eastern Amazonia seems to have drastically oscillated between dry and wet conditions, conditions in western Amazonia seem to have been more stable (Cheng et al. 2013). These observations underscore the need to investigate the consequences of past climatic fluctuations in ecologically diverse taxa and at regional scales to capture the unique climatic and ecological histories of different areas. African rainforests have received far less research attention, despite pollen data suggesting that the effects of aridity and cooling were more intense in Africa, compared to other global rainforests (Maley 1996; Corlett and Primack 2011). A few excellent reviews with a continental-scale perspective have discussed the effects of fragmentation on the evolution of African rainforests and their biota (Plana 2004; Hardy et al. 2013; Couvreur et al. 2021). However, no review has yet focused on refugia at a regional to local scale, which is most relevant for conservation (Balantic et al. 2021). Given that the main African rainforest blocks exhibit distinct ecological, climatic and geological characteristics, gaining insights into the individual histories of each would be highly valuable (Billick and Price 2011; Willis et al. 2013). With the growing body of literature specifically focusing on organismal groups inhabiting the West, Central or East African rainforest blocks, we think it is timely to synthesise our understanding of the role of past climate in shaping African rainforest biodiversity at a regional scale.

In this review, we focus on West African rainforests. They encompass a global biodiversity hotspot – the ‘Guinean Forests of West Africa’ (Myers et al. 2000) – and are widely regarded as a distinct biogeographic unit, harbouring high levels of taxonomic uniqueness (De Klerk et al. 2002a; Penner et al. 2011; Linder et al. 2012; Droissart et al. 2018). Habitat alteration, conversion and fragmentation have resulted in an overall loss of more than 80% of the former forest cover (Carr et al. 2015; Aleman et al. 2018; Rödel et al. 2021). Contemporary climate change is already impacting these rainforests by leading to a general reduction in average rainfall (Wittig et al. 2007) and, therefore, it is of utmost importance to come up with effective conservation measures. Here, we discuss the role of West African rainforest refugia in supporting species persistence and diversification. We focus primarily on climatic fluctuations during the Pleistocene, but also discuss patterns shaped at deeper time-scales. We first define our study area and provide context regarding the link between rainforests, geography and climate in this region. Then, we discuss how different fields of study have contributed to our collective understanding of macrorefugia, microrefugia and the role these regions have played in shaping biodiversity. Finally, we discuss how more detailed genomic studies at a broader taxonomic scale can yield important insights into refugia dynamics across time and space. Combined with SDM, such efforts can enable us to better understand how species have responded to climatic fluctuations in the past and how evolutionary refugia across West Africa may mitigate the effects of contemporary climate change (Gavin et al. 2014).

While the distinction between rainforest flora and fauna in West and Central Africa is widely acknowledged (De Klerk et al. 2002a; Penner et al. 2011; Linder et al. 2012; Droissart et al. 2018), the precise geographic position of the border between these two rainforest blocks is a matter of debate. Several landscape features, such as the Volta River in Ghana (Fig. 1a), the Dahomey Gap from central Ghana to southern Benin (Fig. 1c), the Niger or Cross Rivers in Nigeria (Fig. 1a) or the Sanaga River in Cameroon, have been suggested as potential biogeographic boundaries. Which of these ‘borders’ is favoured appears to vary depending on the taxonomic group under consideration (Booth 1958; White 1979; Penner et al. 2011; Linder et al. 2012; Droissart et al. 2018). Most phylogenetic studies show that Lower Guinean populations west of the Cross River (Figs 1a, 3a) have a closer phylogenetic affinity to their West, rather than Central African sister taxa (see phylogeography section). Therefore, we herein define West African rainforests as extending from Sierra Leone and Guinea in the west to the Niger Delta and the Cross River in the east (Fig. 1a). The reason behind the existence of a distinct biogeographic boundary in the Cross River is a topic of discussion. Compared to other African regions, rivers in West Africa have remained relatively stable over time, due to longer periods of tectonic stability (John 1986; Penner et al. 2011). Rivers have been found to act as long-standing barriers for the dispersal of many mammal, amphibian and understorey bird species (Nicolas et al. 2008a; Voelker et al. 2013; Penner et al. 2019). Additional reasons underlying the presence of a biogeographic boundary in this region could be recurrent rainforest fragmentation during periods of aridity or marine incursion events, that isolated West Africa from the rest of the continent between 94–50 million years ago (Mya) (Maley 1996; Heine and Brune 2014; Rödel et al. 2021; Scotese 2021).

West African rainforests comprise several forest sub-types or -regions with distinct species assemblages (ecoregions; Olson et al. (2001)). These include: Western Guinean lowland forests, Eastern Guinean forests, Guinean montane forests, Nigerian lowland forests, Niger Delta swamp forests and Cross-Niger transition forests (Fig. 1c). The arrangement of forest types (deciduous versus semi-deciduous) is influenced by the movement of the Inter-Tropical Convergence Zone (ITCZ) in a north-south direction, which determines the rainfall pattern (Poorter et al. 2004). As a result, different latitudinal belts of forests are formed. Along the Atlantic coastline (where annual rainfall exceeds 1,500 mm; Fig. 1b), evergreen rainforests thrive, gradually transitioning into semi-deciduous forests several hundred kilometres inland (annual rainfall ranging from 1,250 to 1,750 mm; Fig. 1b). In the drier interior, rainforests give way to forest-savannah mosaics, eventually becoming savannah and semi-desert vegetation and ultimately desert in the far North. The West African rainforest belt stretches continuously along the Atlantic coast (Fig. 1b, c). It is only naturally interrupted by dry forest and savannah between southern-central Ghana and Benin, the so-called Dahomey Gap (Fig. 1c). This region receives unusually low annual rainfall levels (~ 700 mm; Fig. 1b). The northern savannah also extends comparatively far south into the central Ivorian rainforest belt, east of the Bandama River (Fig. 1a), forming the V-Baoulé (Fig. 1c). This savannah intrusion is slowly disappearing and has been shown to be a legacy from the last dry palaeoclimatic fluctuation (Peltre 1977; Nicolas et al. 2008a). Finally, the pattern of parallel west-eastern biome-bands is altered by altitude, with higher mountains receiving more precipitation than lowlands of the same latitude (Fig. 1a, b). Mountains, ranges and/or plateaus such as those forming the Guinean highlands receive exceptionally high rainfall allowing rainforests to thrive further north than elsewhere in West Africa (e.g. the Fouta Djallon located in central-northern Guinea receives 1,600–2,000 mm annual rainfall despite being located around 10–11°N; Fig. 1b, c). Changes in past climate have been related to north-south oscillations of the latitudinal vegetation bands. Currently, the rainforest zone extends approximately 300 km from the Atlantic coast into the north, but it is believed to have extended up to 400–500 km further north of their current position during humid periods (Malhi et al. 2013). On the contrary, during cold and dry phases, the ITCZ shifted southwards, below the Congo fan (Leroux 1993; Kim et al. 2010), leading to a southwards shift of the rainfall and vegetation gradient and presumably allowing rainforests to persist only in the southern tips of the Gulf of Guinea. Further north, it has been hypothesised that rainforests could have persisted in upland areas capturing humidity of stratiform clouds and receiving orographic rainfall (Maley 1996; Plana 2004).

Phylogeographic, biogeographic and pollen data played an important role in uncovering the location and extent of both macro- and potentially microrefugia. However, they are less suited to quantify how refugia have shaped population dynamics across space and time. By employing demographic and SDM, we can investigate fluctuations in population size, changes in gene flow and shifts in suitable habitat over time. This provides valuable insights into refugia connectivity, their capacity to support viable populations based on population size trajectories and their role in driving diversification. Understanding these dynamics is key to assessing the micro- and macro-evolutionary consequences of refugia and predicting how species may respond to future changes in climate.

Demographic inference

Early studies investigating demographic signatures of refugia used 16S rRNA and/or cytB mitochondrial sequences (Hillers 2008; Nicolas et al. 2008a). However, our ability to accurately infer divergence times and to reconstruct demographic history based on such a small number of genes is very limited (Felsenstein 2006; Gutenkunst et al. 2009). With the development of new genetic markers and the introduction of high-throughput sequencing, dozens and even thousands of loci can now be utilised to more accurately characterise refugia dynamics at the temporal and spatial scale (Lexer et al. 2013; Horníková et al. 2021; Mestre et al. 2022). In West Africa, demographic modelling has, to date, been largely based on microsatellites (Duminil et al. 2013; Demenou et al. 2016, 2018; Migliore et al. 2019), but also on genome-reduction approaches, such as ddRADseq (Portik et al. 2017; Allen et al. 2021). These studies employ model-based demographic inference by comparing the observed genetic data with simulated data run under competing demographic models. Demographic models can be based on single populations and model exclusively changes in population size (e.g. bottlenecks, expansions, contractions) or they can be based on multiple populations and model divergence in terms of population size and/or migration rates between daughter populations (e.g. divergence in strict isolation, secondary contact and divergence with migration; Nkonmeneck et al. (2022); De Jode et al. (2023); Fig. 5g).

The importance of climate refugia in generating the rich biodiversity found in rainforests has been widely debated, but they are not the only processes that can lead to lineage diversification (Moritz et al. 2000; Rocha and Kaefer 2019). One major objective of demographic studies is to weigh competing mechanisms leading to population differentiation. The three most frequently tested mechanisms of diversification are: the a) refugia, b) landscape barrier and c) ecotone model and each makes particular demographic predictions (Fig. 5g). In the landscape barrier model (Wallace 1854; Ayres and Clutton-Brock 1992; Moritz et al. 2000), daughter populations diverge in strict isolation from each other, for instance on opposing river banks. In the refugia model (Haffer 1969; Diamond and Hamilton 1980; Haffer 1997; Moritz et al. 2000), daughter populations diverge in isolation followed by population growth and secondary contact. In the ecotone model, daughter populations diverge with continuous gene flow, which may decrease over time (Endler 1977; Rice and Hostert 1993; Smith et al. 1997; Moritz et al. 2000). Studies on rainforest reptiles and amphibians in West and Central Africa predominantly support the role of refugia in driving divergence, with populations initially diverging in isolation, followed by expansion and secondary contact (Portik et al. 2017; Charles et al. 2018; Leaché et al. 2019; Jaynes et al. 2022; Nkonmeneck et al. 2022). However, the refugia model has also been rejected in favour of the landscape barrier (Allen et al. 2021) or the ecotone model (Freedman et al. 2023). Ecological traits strongly influence species responses to climate-driven rainforest fragmentation. For instance, the arboreal snake species investigated by Allen et al. (2021) occur in rainforests, but also woodlands, including the dry forest-savannah mosaic in the DG. They are, therefore, likely not as susceptible to rainforest fragmentation. Nonetheless, taken together, these findings underscore the importance of refugia in driving diversification at least for species that are strongly tied to rainforests.

Figure 5. 

SDM and demographic modelling. a-f: climatic stability maps inferred by modelling the distribution of suitable habitat over the last 120 Kya.(a) Leptopelis macrotis Schiøtz, 1967 and L. millsoni (Boulenger, 1895) (Jaynes et al. 2022);(b) Chiromantis rufescens (Günther, 1869) (Leaché et al. 2019);(c) Scotobleps gabonicus Boulenger, 1900 (Portik et al. 2017);(d) Pan troglodytes (Blumenbach, 1775) (Barratt et al. 2021);(e) Toxycodryas pulverulenta (Fischer, 1856) (Allen et al. 2021);(f) Toxicodryas blandingii (Hallowell, 1844) (Allen et al. 2021); (a–c): frogs; (d): chimpanzee, (e, f): snakes. Areas with the highest probability of stability are shown in red;(g) schematic representation of the competing models employed to test for the role of landscape barriers, refugia or ecotones in driving the divergence.

Demographic modelling has also been used to explore the role of refugia at a more local scale, focusing on how specific landscape features influence population dynamics. Demenou et al. (2016, 2018) focused their research on the DG to reconstruct the origin of isolated rainforest tree populations in this naturally dry savannah corridor. Two co-distributed tree species presented very similar spatial genetic structure, with distinct clusters in Upper Guinea, Lower Guinea and isolated forest fragments in the DG. For all species, the DG populations seem to have experienced bottlenecks and demographic models suggested that these populations resulted from a mixture of Upper Guinean and Lower Guinean populations. However, the timing of the split between Upper and Lower Guinean populations, as well as the establishment of populations in the DG differed between species. Demenou et al. (2016) found an earlier split between Upper and Lower Guinean populations (ca. 500 Kya) and a post-LGM colonisation of the DG. In contrast, Demenou et al. (2018) found a later split between Lower and Upper Guinean populations (ca. 73 Kya) and a pre-LGM colonisation of the DG. Again, these differences may be explained by variations in the ecology of each species. The species investigated by Demenou et al. (2016, 2018) share similarities as pioneering, light-demanding trees that are dispersed by wind and likely pollinated by insects. However, the study species of Demenou et al. (2018) prefers semi-deciduous forests over secondary evergreen forests. Although both tree species were affected by rainforest fragmentation during arid periods, their distinct ecological traits led to different responses across multiple aridification cycles. This underscores the importance of considering species’ ecological differences to predict the response of rainforest biota to climatic changes.

Despite being overshadowed by research in Central or East Africa, studies on refugia in West Africa have yielded significant insights into their location, age and evolutionary role. Palynological and biogeographical studies have shown that rainforests fragmented and contracted during arid periods (Maley 1996). These approaches pointed to the existence of one to four large refugia (Fig. 2): Mount Nimba, Cape Three Points, Cape Palmas and the Niger Delta region (Fig. 1b). Intraspecific phylogeographic patterns support the positions of these refugia and have identified additional large-scale refugia in regions like: the Fouta Djallon highlands, Sierra Leone, western Liberia and south-eastern Côte d’Ivoire (Fig. 4). Furthermore, they revealed the presence of putative microrefugia coalescing into macrorefugia likely due to more consistent historical connections (Hillers 2008; Fig. 4a). For example, three to four microrefugia were postulated within macrorefugia located in Sierra Leone, western Liberia, south-eastern Guinea, south-eastern Côte d’Ivoire and Ghana (Hillers 2008; Fig. 4a). Molecular analyses on leaf-litter frogs also showed that regions like Fouta Djallon and Ghana harboured lineages since the Miocene, while other areas remained stable since the Pleistocene (Hillers 2008; Fig. 4a). This potentially reflects the varying degrees to which refugia are decoupled from surrounding climatic conditions (Stralberg et al. 2020; Fig. 4a). Phylogeographic splits, while generally congruent across species, are often asynchronous and match both climatically unstable regions and landscape features like rivers which complicates the determination of underlying drivers (Hillers 2008; Nicolas et al. 2008a; Fig. 3). The variation in the timing of divergence, often, but not always matching the timing of aridification events (Bell et al. 2017; Hassanin et al. 2020), indicates that the same spatial phylogeographic patterns across species may have distinct underlying causes (Donoghue and Moore 2003). Alternatively, the observed asynchrony between studies may also be caused by differences in study design, dataset type and inference approaches and calls for a more standardised framework. Climate-driven rainforest fragmentation is probably an important, but not the sole diversification mechanism operating in West Africa. Demographic analyses can provide a more mechanistic understanding of how species have been impacted by climatic fluctuations. Several studies found signs of population contraction and expansion, as well as secondary contact which have been interpreted as signatures of refugia dynamics (Hillers 2008; Nicolas et al. 2008a; Portik et al. 2017). Additionally, there are several SDM studies which indicate that species have experienced substantial range fluctuations through history (Portik et al. 2017; Leaché et al. 2019; Barratt et al. 2021; Jaynes et al. 2022; Fig. 5a–d). Similar range shifts have been detected in rainforest taxa in Central (Bell et al. 2017; Helmstetter et al. 2020) and East Africa (Demos et al. 2014; Barratt et al. 2018). However, the response of species to climate-driven rainforest fragmentation likely varies across taxonomic groups depending on their biological properties, such as degree of rainforest dependency (Bell et al. 2017; Helmstetter et al. 2020). So far, studies exploring demographic signatures of refugia in West Africa are limited to few species and many regions, such as the Niger Delta and Fouta Djallon highlands, remain understudied. This limits our full understanding of West Africa’s biogeographic history.

Rainforest refugia in West Africa have been mostly treated as clearly outlined regions to which rainforests and their associated biota retracted during periods of aridity. While this has provided a useful framework, the real nature of refugia is likely more complex. West African rainforests encompass a diverse range of forest habitats: evergreen, semi-deciduous, deciduous, lowland and montane forests (Fig. 1c), each of which probably has been impacted in different ways by palaeoclimatic oscillations. Rainforest-dependent species in other African regions tend to exhibit diverse responses to climatic changes, depending on their biological properties (Bell et al. 2017; Helmstetter et al. 2020). Furthermore, refugia differ in spatial scale (micro- versus macrorefugia) and possibly also in temporal persistence (Miocene versus Pleistocene refugia) (Hillers 2008; Fig. 4a). Globally, the focus is shifting from simply mapping refugia to obtaining a process-based understanding of population persistence (Bennett and Provan 2008) and the environmental and physical geographic properties producing refugial habitats (Keppel et al. 2012; Stralberg et al. 2020; Balantic et al. 2021). West Africa has a diverse landscape, which includes hilly regions, isolated mountains, mountain ranges and topographically heterogeneous areas with deep valleys. Persisting under these different types of landscape settings likely has different demographic implications for populations, with regards to abundance, range and/or connectivity amongst refugia. A more comprehensive understanding of refugia dynamics requires examining a wider range of taxa with finer spatial resolution and genomic data (Lexer et al. 2013; McGaughran et al. 2022). Future research would benefit from exploring also lesser-known regions within West Africa (e.g. the Guinean Fouta Djallon or the Niger Delta region) and adopting landscape demography approaches for a comprehensive understanding of species persistence in varying environments (Gurevitch et al. 2016).

Technological and methodological advances are rapidly expanding the potential of population genomics research across a wider range of taxa, including non-model organisms (Lexer et al. 2013; Ellegren 2014). Furthermore, the development of new sequencing approaches, such as long-read sequencing, enables the use of haplotype information in demographic models. This enhances the resolution for more recent timescales, such as the Pleistocene (Bofill and Blom 2024). Alongside these advances, bioinformatic tools have been developed to support integrative and comparative approaches, facilitating the joint analysis of multiple species or independent data sources. For example, tools integrating genomic data with SDM help explore spatiotemporal refugia dynamics combining multiple lines of evidence (He et al. 2017), while comparative phylogeographic methods facilitate the cross-species comparison of demographic histories (Xue and Hickerson 2017). Although beyond the scope of the current study, future meta-analyses would benefit from using comparative methods capable of accounting for methodological variation and uncertainties across studies, such as those proposed by Hickerson and Meyer (2008). When genetic markers are comparable and both sequence as well as metadata are publicly available, aggregation tools can repurpose these datasets and test phylogeographic hypotheses across taxa (Pelletier et al. 2022). By harnessing these methodological developments, future studies will be better equipped to identify the environmental and biological factors shaping species responses to climatic fluctuations. Such insights are crucial for the development of effective, long-term conservation strategies in the face of ongoing climate change.

The authors would like to express their gratitude to all who contributed valuable insights and discussions that helped shape this paper, particularly the members of the ‘Blom lab’, the ‘Rödel lab’ and the ‘Mayer lab’ at the Museum für Naturkunde Berlin. Special thanks to Sofia Hayden, Guillaume Demare and Ninda Baptista for their support during the writing process, Jack Ullrich-Lüter for organising the Academic Writing Bootcamp and Agustin Elias-Costa for help with Fig. 4. ME was supported by the Elsa-Neumann scholarship from the state of Berlin, Germany. Finally, we thank Thomas LP Couvreur and Bryan Carstens for carefully reading this manuscript and giving constructive feedback in their reviews.

All authors contributed to the conception and the design of this study. ME compiled and analysed the literature and designed the figures. The manuscript was written by ME with input from MPKB and MOR. All authors contributed to data interpretation and writing and approved the final manuscript.

The authors declare no conflict of interests.