Palaeo-trajectories of forest savannization in the southern Congo

Abstract

Tropical savannah and forest are thought to represent alternative stable states in ecosystem structure in some climates. The implication is that biomes are maintained by positive feedbacks, e.g. with fire, and that historical distributions could play a role in determining modern ones. In this context, climate alone does not govern transitions between biomes, and understanding the causes and pathways of such transitions becomes crucial. Here, we use a multi-proxy analysis of a 2000-year core to evaluate modes of transition in vegetation structure and fire regimes. We demonstrate a first transition ca 1540 BP, when a cyclic fire regime entered a forested landscape, eventually resulting, by ca 1060 BP, in a transition to a more open savannah-like or mosaicked structure. This pattern may parallel currently accelerating fire regimes in tropical forests suggesting that fires can savannize forests, but perhaps more slowly than feared. Finally, ca 540 BP, a drought combined with anthropogenic influences resulted in a conclusive transition to savannah, probably resembling the modern landscape in the region. We show here that fire interacted with drought to transition forest to savannah, suggesting that disturbance by fire can be a major driver of biome change.

1. Introduction

The Afrotropics comprise 15% of tropical forests and 70% of tropical savannahs [1], representing a key component of the global carbon cycle. Recent work suggests that transitions between savannah and forest may result not just from climate, but also from changes in disturbance regimes, which have been implicated in expanding savannah extent globally. However, our understanding of the pathways of transition between savannah and forest is limited by their slow pace (even ‘abrupt’ transitions take decades for trees to grow). Palaeo-ecological data, i.e. the sub-fossil and geochemical remains preserved in sediment sequences, could provide insights at relevant timescales but are underrepresented in the region.

Forest and savannah biomes are differentiated by discontinuities in vegetation structure and disturbance regimes. Forests feature a closed canopy, suppressing grass accumulation and preventing frequent fire spread [2], whereas savannahs feature open (although varying) canopies, with a continuous and often fire-promoting grass layer [3–5]. In Africa, both savannah and forest are widely distributed in regions with rainfall between 1000 and 2500 mm yr⁻¹ [3,6], and fire experiments show that fire can maintain savannahs in regions where forests represent the climax biome [3,7,8]. Models of biome distribution have simplified these findings to argue that feedbacks between an open canopy and frequent fires (and between a closed canopy and fire suppression) result in alternative stable states in ecosystem structure. These models imply that past distributions could be crucial in determining the distribution of savannah and forest today.

However, the past distribution of savannahs and forests in Africa are still hotly debated [9]. The possibility for alternative stable states in ecosystem structure exacerbates this debate—past changes in distributions, whether driven by climate or by human impacts, are likely to be preserved over a long timescale via feedbacks with fire. The palaeohistory of climate events (e.g. large droughts [10]) may play a crucial role in shaping the distribution of savannah and forest today, as could past human impacts. As such, palaeo-ecological insights become even more critical for determining the processes and drivers that have resulted in the distribution of savannah and forest today.

What we do know is that forests in the Congo Basin have repeatedly expanded and contracted during the past 20 000 years, driven by climatic oscillations [11–13] and, more recently, by anthropogenic forcings [14,15]. Climatically, drying has been associated with forest disturbance and contraction. Most notably, around 4000 BP, the end of the African Humid Period resulted in a decrease in rainfall [16] and savannization of sites along the forest margin [17,18]. Later, during the ‘third millennium crisis' (approx. 2800–1300 BP), an increase in seasonality [13,19] resulted in widespread forest disturbances deep in the Congo interior [20]. Meanwhile, estimates of the timing of intensification of human activity in the Basin also vary widely [15,21–23], ranging from 3500 to 2000 BP depending on location [24,25]. In general, the arrival of Bantu-speaking peoples probably began near the border between Cameroon and Nigeria, eventually reaching the southern Congo [24,26]. Some authors argue that this sequential anthropogenic intensification has been responsible for a wave of landscape change [15,27].

Disentangling the mechanisms and pathways of ecological transitions among savannah and forest presents a challenge. Well-dated, multi-proxy long-term data, especially fire records, are necessary for studying how these systems have responded to climatic and anthropogenic change, but appropriate records are lacking. To remedy this problem, we present a multi-proxy analysis of a lacustrine sedimentary sequence dating to 2000 cal. BP in the southern Congo.

2. Methods

We extracted sediments from Lake Ngofouo (figure 1) in August 2015, dated the core (8 yr cm⁻¹ resolution), and evaluated multiple independent proxies (see electronic supplementary material and [29]). We used continuous charcoal analysis to estimate fire activity and fuel type [30], and carbonate and organic matter concentrations as indicators of local hydrological changes and thus climate. Phytoliths were analysed to estimate vegetation composition and tree cover [31–34]. We paired these with a review of archaeological sites to estimate local and regional human activities [25,35]. Regime-shift analyses [36,37] were performed on charcoal, carbonates and organic matter records (electronic supplementary material, figures S2–S5).

Figure 1.

The location of Lake Ngofouo in Africa (a), the age-depth model at the site (b), and regional [28] (a) and local (c) tree cover surrounding the site. In (a), other palaeosites are represented for comparison, with local and regional archaeological sites [24,25]. In (b), ¹⁴C ages are shown in blue and ²¹⁰Pb dates in green. In (c), an NDVI image from 2018 Landsat shows the lake's surroundings, with savannah, patches of gallery forest and a fire scar. (Online version in colour.)

3. Results and discussion

(a). A closed ecosystem with no fire before ca 1540 BP

Before ca 1540 BP, the landscape was forested, without fire activity. Phytolith influx was low, D:[D + P] was high (greater than 0.6, figure 2h) [33,38], and woody-dicot morphotypes dominated assemblages (electronic supplementary material, figure S1), indicating a forested landscape [32–34,38–40]. Meanwhile, charcoal accumulation rates (CHAR) were low (figure 2a) and charcoal W : L was less than 0.6 and fluctuating (figure 2c,d). Thus, fire appears to have been minimal in the landscape, with the few fires in the region (aeolian transportation from at least 10 km [30]) likely fueled by a mix of herbaceous and arboreal types [30]. Overall, sediments featured high organic matter and low carbonate concentrations (figure 2e,f; electronic supplementary material, figures S4 and S5), suggesting a productive system with minimal erosion, i.e. a forest. Human densities around the lake remained low regionally and locally (figure 2g), despite the arrivals of Bantu-speakers elsewhere in the southern Congo (electronic supplementary material, figure S6, [24,25,35]).

Figure 2.

Multi-proxy analysis of the core NGO-A. Charcoal accumulation rates (CHAR, (a), the spectral analysis of the CHAR (b), black lines surround area of significant periodicity), the width-to-length ratio of the charcoal particles (c), the standard deviation of W/L (d), the content in organic matter (e) and the carbonates concentration of the sediments (f), the density of archaeological dating around the lake locally (crosses) and regionally (black lines and 95% CI in grey; (g)), the phytolith influxes of woody dicots, lobate, cross, rondel and saddle (h) and the phytolith tree cover index (i). (Online version in colour.)

(b). Increased fire activity and gradual landscape savannization, ca 1540–1060 BP

The end of this period, in ca 1540 BP, was marked by an increase in fire activity. However, forest apparently persisted locally, only gradually transitioning towards savannah. Ecological changes were concurrent with the appearance of humans in the region, suggesting that transitions were at least partially anthropogenic.

Biomass burning increased sharply ca 1540, with CHAR reaching 10 particles cm⁻² yr⁻¹. Meanwhile, fuel type stabilized, with more elongated and less variable particles (W : L ∼ 0.4 and low standard deviation, figure 2c,d), indicating a herbaceous fuel type [30]. High frequency of herbaceous fires suggests that, at the regional level, forests had begun to transition to savannah. Note that we interpret these fire proxies as regional in scale, because charcoal accumulates over a large depositional basin and probably does not represent a strictly local signal [30,41].

By contrast with charcoal proxies, phytoliths indicate that forests persisted in the immediate vicinity of Lake Ngofouo. Phytolith influx remained low until ca 1060 BP (figure 2h) and the phytolith tree cover index remained high (greater than 0.65, figure 2i; electronic supplementary material, figure S1), reflecting the dominance of woody dicots in the assemblage, consistent with forest surrounding the lake [32,38]. However, sediment organic matter content dropped and carbonate concentration increased (figure 2e,f), indicating either local opening of the forest landscape and/or an increase in rainfall seasonality. Moreover, throughout this period, the tree cover index decreased and lobate phytoliths (C₄ grasses [34]) increased continuously (figure 2h; electronic supplementary material, figure S1), again indicating savannization [34].

Notable during this period is a mismatch between an immediate increase in fire activity ca 1540 BP, and gradual opening of the landscape recorded in vegetation proxies between ca 1540 and 1060 BP. We suggest two possibilities. First, this pattern may reflect heterogeneous intensification of human activities throughout the region. In support of this idea, a progressive shift towards an open, fire-prone environment was also recorded at Ngamakala beginning ca 1540 BP [42], whereas other sites in the region savannized much earlier (4200 BP at Sinnda [34,43] and 2500 BP at Kitina [44]; figure 1a). Patterns of heterogeneous transition across the region, associated with the presence of archaeological sites, suggest that humans may have played a key role in transitioning forests to savannahs [45,46]. In the modern record, fire encroachment into forests and therefore forest savannization is a strongly spatially structured process [9,47], which can make locally abrupt transitions more gradual at the landscape scale [48]—a pattern consistent with this hypothesis.

However, humans arrived near Lake Ngofouo towards the beginning of this transition period, (ca 1470 BP at the latest), and their influence on vegetation should perhaps have been felt earlier. A second possibility is that the gradual change in vegetation proxies from ca 1540 to 1060 BP towards a more open, savannah-like configuration instead reflects forest resilience to intensifying human impacts. In support of this idea, fire activity during this period was strongly cyclic, with a period less than 60 years (figure 2b), perhaps implying forest successional processes [49,50]. Certainly, modern ecologists hotly debate how forests transition into savannahs and, specifically, whether and how quickly runaway fire feedbacks can lead to forest savannization [50,51]. If this is the case, then forests show a surprising degree of resilience to fire encroachment—with approximately 500 years of frequent fire needed for savannization, at least outside the context of anthropogenic climate change.

(c). A fire-prone mosaic, ca 1060–500 BP

Not until 1060 BP did vegetation shift more conclusively towards a more open savannah. Phytolith influx increased, and the phytolith tree cover index decreased to approximately 0.55 (figure 2i), reflecting a possible savannah–forest mosaic [38]. Phytolith types diversified, with increases especially in lobate phytoliths abundance (figure 2h), indicating that C₄ grasses expanded into the immediate surroundings of Ngofouo (paralleling transitions at Sinnda [34]).

Curiously, savannahs persisted in the region throughout this period, despite evidence elsewhere of increasing regional rainfall [13]. This supports the notion that savannah and forest may represent alternative stable states—and that repeated fires in savannah, once established, can prevent forest invasion, even when environmental conditions revert to favouring forests (i.e. hysteresis) [6,52].

Fire continued to burn, probably primarily herbaceous fuels, evident in substantial biomass burning (figure 2c), but at lower rates overall and with a longer period of variation (300–400 years; figure 2b). Meanwhile, human densities increased throughout this period (figure 2g; electronic supplementary material, figure S6). In the savannah context, these results are perplexing, since higher rainfall in some savannahs have been shown to increase fire frequency, via increases in fuel loads [53]. Increasing agriculturalization provides one possible explanation, if croplands decreased potential burned area [54]. However, today, this landscape burns often (2+ times per year) despite extensive cropping, so this seems unlikely. Another possibility is that the area was so wet that climate, not fuel, primarily limited fire frequency [55]. Complications in interpreting biomass burning as fire frequency or burned area are another possibility [41,56].

(d). Combined effects of people and climate, ca 540 BP, conclusively establish savannahs?

Around ca 540 BP, biomass burning abruptly increased and started cycling more quickly (figure 2b), and sediment carbonates concentrations decreased, either due to increased woody cover or decreased rainfall seasonality. A brief drought occurred in 610 BP at Sinnda [43] (figure 1a), suggesting a local drought linked to this regional event [57]. Simultaneously, archaeological-site increased to its highest level, both locally and regionally (figure 2g), suggesting an increased population density. Together, these events transitioned the system rapidly towards a true savannah, away from a savannah-forest mosaic (tree cover index approx. 0.3 [32,38]). Since then, during the last 100 years, biomass burning increased despite a steady decrease in human activity. The landscape configuration was likely similar to today's (figure 1c).

4. Conclusion

This multi-proxy analysis of a lacustrine core in southern Congo provides evidence of a transition from a fire-suppressing forest to an open savannah with an active fire regime. Although the initial shift in fire regime was abrupt, forests were resilient to change for at least 500 years despite human arrivals; rapid cycling in biomass burning suggests some process acted to erode forest extent. A second shift definitively resulted in the establishment of a dominant savannah landscape, due to the combined effect of human activities and drought.

Overall, the sequence provides a resolution on a number of controversial ideas about what determines biome distributions. First of all, while the climate was clearly implicated in biome transitions, fires also changed vegetation structure and, crucially, prevented savannahs from re-foresting, suggesting that fires probably play a key role in shaping biome distributions and transitions. Notwithstanding, forests were partially resilient to human pressures and to changes in climate, and forest savannization was slow, confirming the need for palaeo-ecological perspectives on the ecological processes involved. However, anthropogenic global change is likely to erode the resilience of tropical biomes [58], and further insight will depend on more intensive work in the region, focused on disentangling the impacts of climate from human activity. In a system where climate, vegetation structure, and fire regime are not deterministically linked [6,9,59], multiple independent proxies will be essential to avoid climate reconstructions that rely erroneously on changes in vegetation structure. This work shows that multi-proxy approaches can provide crucial—and currently missing—context for ecosystem change in the Afrotropics.

Data accessibility

All data are available from the Dryad Digital Repository at: https://doi.org/10.5061/dryad.jq1f218 [29] and charcoal data are also available from the Global Charcoal Database.

Authors' contributions

J.C.A. and A.C.S. designed the study; J.C.A., O.B., A.C.S., H.E., V.K., G.It. and G.Is. performed fieldwork; J.P. and J.C.A. analysed the charcoal and phytolith data; J.C.A. performed the numerical analyses; J.C.A., O.B. and A.C.S. wrote the manuscript with feedbacks from all authors. All authors agree to be held accountable for the content therein and approve the final version of the manuscript.

Competing interests

We declare we have no competing interests.

Funding

Support for J.C.A. was provided by Graduate Women In Science (Nell I. Mondy fellowship) and the Brown Postdoctoral Fellowship at Yale University, for A.C.S. by the National Science Foundation (DMS-1615531), and for O.B. and J.P. by an NSERC discovery grant no. (RGPIN-2017-05062).