Twenty-first century emergence of alpine fire in Central African mountains

2026-05-21Mason,AndreaL.,Pereboom,EleanorM.B.,Ivory,SarahJ.,Vachula,RichardS.,Kelly,MeredithA.,Nakileza,Bob,Russell,JamesM.

Nature 2026年8115期

关键词：公式

Abstract

Wildfires are an escalating global hazard that threaten ecosystems, air quality and societies 1 , 2 . Although tall tropical mountains have generally been considered too cool and moist to burn 3 , the recent occurrence of high-elevation wildfires on Africa’s highest mountains suggests the emergence of a new and potentially transformative threat. A lack of historical records of fire in these environments limits our understanding of fire activity, its environmental impacts and the resilience of Afromontane ecosystems. Here we show that a twenty-first century fire was the first to affect Afroalpine elevations (>3,800 m above sea level (a.s.l.)) over the past 12 thousand years in the Rwenzori Mountains, Central Africa. At mid-elevations (2,990 m a.s.l.), fire increased abruptly around 2 thousand years ago, coincident with regional evidence for changing human activity 4 , and this was followed by an increase in bamboo-dominated ecosystems. Our results highlight the role of humans as an important driver of Afromontane fire activity and show that high-elevation tropical Afroalpine fire is a new twenty-first century disturbance that could transform high-elevation tropical ecosystems.

Main

Fire is an ecological process that shapes plant community composition 5 , 6 and carbon cycling 6 , 7 , 8 . It is critical for the maintenance of many terrestrial ecosystems but can also transform vegetation community structure, particularly in the context of rapid climate change 5 . Climate is a primary driver of fire activity through its influence on fuel availability and flammability; however, human activity has played a critical role in fire dynamics by both promoting and suppressing fire 6 , 8 . The tropics have recently experienced substantial changes in fire regime or the frequency, extent and intensity of fires over time. Although burned area has decreased substantially in tropical savannahs as a result of agricultural expansion and fragmentation of fire-prone savannah vegetation 6 , fire has increased in other ecosystems owing to increasingly fire-prone weather conditions and intensifying human activity 1 , 2 . This includes lowland tropical forests in Central Africa, where human activity, coupled with climate warming and drying, have increased fire frequency and severity 9 .

Whereas fire is well studied in lowland African ecosystems, far less is known about fire activity in Africa’s high tropical mountains 10 , where persistently cold (mean annual temperature <6 °C) and moist (>2,000 mm year − 1 precipitation) conditions are thought to suppress fire 3 . Tropical African mountains constitute a unique biodiversity hotspot, hosting rich and distinctive endemic flora (Afromontane and Afroalpine communities) on a series of isolated ‘sky islands’, making them uniquely threatened by their limited range for migration in a warming climate 1 . In 2012 CE, Africa’s third tallest mountain range, the Rwenzori Mountains, Uganda–Democratic Republic of Congo, experienced a widespread and destructive fire that burned approximately 42 km 2 of fragile, high-elevation Afroalpine habitat 11 . Similar fire outbreaks on Mount Kilimanjaro and Mount Kenya in the past decade further suggest that wildfire is becoming a pressing environmental concern in tropical African mountains 12 , 13 . However, with a limited understanding of post-fire recovery on tropical African mountains, the long-term resilience of Afromontane forests to sustained changes in fire activity and the impacts of fire on higher-elevation Afroalpine ecosystems are largely unknown. Recent wildfires across Africa’s highest iconic mountains mean that we require new data to understand historical fire activity and the interactions between climate change, fire activity and plant communities.

Rwenzori Afromontane fire

We investigated the historical context of recent Afroalpine wildfire, the role of climate change in Afromontane fire activity and the response of Afromontane ecosystems to fire. We collected lake sediment cores from the Rwenzori Mountains, Uganda–Democratic Republic of Congo (Fig. 1 and Extended Data Fig. 1) spanning the past approximately 12 thousand years (kyr) from Lake Mahoma, situated at middle elevations (2,990 m a.s.l.), and Lake Kopello, situated in the Afroalpine zone (4,017 m a.s.l.) 14 . Our records include sediments spanning the 2012 CE Rwenzori fire that occurred in the Lake Kopello watershed (Extended Data Fig. 1), allowing validation of fire proxies.

**Fig. 1: The location of the Rwenzori Mountains and Africa’s other high-elevation mountains.**

Steep climatic and ecological gradients result in distinct ecosystems at different elevations in the Rwenzori. In the lowlands (<1,800 m a.s.l.), vegetation is dominated by bushland and savannah grasslands. Afromontane forests inhabit 1,800–2,400 m a.s.l., above which lies a zone dominated by dense mountain bamboo. Ericaceous forest extends from about 2,700 to 3,900 m a.s.l., above which are Afroalpine communities and herbaceous cover 10 , 15 . Lake Kopello is in the Afroalpine zone, whereas Lake Mahoma is in the bamboo zone. We generated new high-resolution records of fire using sedimentary charcoal and compared these with new and existing records of precipitation 14 , temperature 15 and vegetation 10 , 14 , 16 .

The sediment record from Lake Kopello reveals very low rates of charcoal accumulation throughout the Holocene, ranging from 0 to 1.3 particles per cm 2 per year (Fig. 2c). By contrast, charcoal accumulation is much greater in sediments spanning the past century, ranging from 0 to 87 particles per cm 2 per year (Fig. 2d). Charcoal accumulation was enhanced from 1920 to 1960 CE and gradually declined until the twenty-first century. The largest peak in charcoal accumulation occurs in sediments dated between 2012 (the year of the recent fire) and 2014 CE, when it reached between 62 and 87 particles per cm 2 per year. In Lake Mahoma, charcoal accumulation rates are higher, ranging from 0 to 2 particles per cm 2 per year during most of the Holocene (Fig. 3d). Accumulation rapidly increased to about 6.5 particles per cm 2 per year and remained high over the past approximately 2 kyr. The differences in charcoal accumulation trends between the two sites reveal different fire histories at these elevations.

**Fig. 2: Records of Afroalpine fire in the Rwenzori during the twenty-first century and Holocene.**

**Fig. 3: Records of montane fire in the Holocene in the context of climate, vegetation and human activity.**

Modern emergence of alpine fire

Lake Kopello charcoal accumulation is an order of magnitude lower in the Holocene compared with the the past century and nearly 100 times lower than the highest accumulation rates observed during the 2012 fire (Fig. 2c). Although charcoal accumulation rates were increased over the past roughly 100 years compared with the previous approximately 12 kyr, no local fires were observed during scientific exploration and tourism over the twentieth and twenty-first centuries in the Lake Kopello catchment before the fire in 2012 (refs. 16 , 17 ). Charcoal accumulating in Lake Kopello during the twentieth and early twenty-first centuries therefore must derive from the long-distance transport of charcoal to Lake Kopello from remote fires 18 , 19 . Moderate Resolution Imaging Spectroradiometer (MODIS) data from the past 25 years confirm that several low-elevation fires within a 50-km radius of the Rwenzori produced injection heights sufficient to deposit charcoal at elevations reaching as high as Lake Kopello (Extended Data Fig. 2). Also, charcoal morphology, measured by length:width (L:W) ratios, has been shown to record fire fuel source in the Afrotropics 20 and corroborates that much of the twentieth century charcoal is derived from grassland fires (Fig. 2e), such as those in the lowland savannah (grass fuel charcoal L:W ratio >3.5) 21 . By contrast, charcoal L:W ratios for the 2012 fire event were 3.4, consistent with more woody fuel sources such as those found in the Lake Kopello watershed today. Burning of Afroalpine sedges such as Carex runssoroensis, which produce a high charcoal L:W ratio 22 and surround Lake Kopello, could explain why charcoal recording the fire event has L:W ratios between the range for grasses and shrubs 21 , 22 . Despite possible contributions of elongated sedge charcoal, a significant difference between the L:W ratios before and after the 2012 fire (Cliff’s δ = 1, 95% confidence interval (CI) [0.71, 1.00]; Wilcoxon rank-sum test, n = 14, W = 24, P = 0.021) is consistent with a switch in fire fuel source between these intervals driven by a change from lowland savannah to local fire.

Trends in twentieth century charcoal accumulation (Fig. 2d) align with socioeconomic changes that affected lowland Uganda, further supporting a low-elevation fire source. Increased and variable charcoal accumulation rates characterize much of the early twentieth century, and charcoal accumulation significantly decreases (pre-1962 versus post-1962 (1880–2009 CE), Cliff’s δ = 0.483, 95% CI [0.11, 0.73]; Wilcoxon rank-sum test, n = 41, W = 310, P = 0.008) in the mid-twentieth century, coincident with the end of British colonial rule and Ugandan independence in 1962 CE (Extended Data Fig. 3). Similar mid-century charcoal declines are also observed at Lake Katinda (1,024 m a.s.l.) 18 , located in the lowlands about 50 km from the Rwenzori (Supplementary Fig. 1). These records may reflect a regional decline in lowland fire activity that could be associated with a decrease in pastoralism and the expansion of sedentary agricultural practices 18 , 23 . The British colonial government enacted measures to discourage pastoralism, such as imposing boundaries to restrict movement, creation of protected game and forest reserves and subsidies for those who implemented modern methods of cattle production. These strategies were continued by the post-colonial government following independence 23 and may have altered fire activity.

Lake Kopello charcoal thus tracks both local and regional fire, suggesting that large-scale high-elevation fires, had they occurred, would be captured in the sedimentary record. There may have been a small-scale fire in the Afroalpine zone of the Rwenzori in the late 1990s (CE); however, it is thought to have occurred largely in the Democratic Republic of Congo region of the mountains during a period of local conflict when Rwenzori Mountains National Park was closed and, thus, is not well documented. The large increase in charcoal accumulation in Lake Kopello during the only observed high-elevation fire demonstrates that the 2012 fire is unprecedented in the past approximately 12 kyr in the Afroalpine zone of the Rwenzori Mountains.

Because our Lake Kopello record comprises, to our knowledge, the first complete Holocene record of fire from the alpine zone in tropical Africa, it is difficult to assess whether such low fire frequencies also characterize other Afroalpine environments. Fire has been a recurrent ecosystem disturbance over the Holocene in the high elevations of the Bale Mountains, Ethiopia; however, the Bale Mountains are considerably drier and differ in their Afroalpine vegetation community structure. The Bale ericaceous belt extends to a higher elevation in the Afroalpine zone (3,000–4,300 m a.s.l.) 24 and because Erica is highly flammable, it can create a positive fire feedback in drier ericaceous shrublands 24 , 25 . However, mature Erica forests such as those on Rwenzori and Kilimanjaro persist in wetter regions with low fire activity. Recent fires on Kilimanjaro have destroyed large tracts of old-growth Erica forest, which were subsequently replaced with low-stature Erica shrubs 7 . Although we lack records from the ericaceous zone in the Rwenzori, changes in Erica abundance over the Holocene are not coupled with fire activity (Extended Data Fig. 4), indicating differences in Erica-fire feedbacks between moist tropical and drier subtropical mountain ecosystems.

Despite considerable Holocene climate change, including moist conditions during the African Humid Period from 11 to 5 thousand years ago (ka) 26 and late Holocene century-scale droughts that influenced fire activity in lowland western Uganda 18 , fire remained absent in the Rwenzori Afroalpine zone. Fire was also suppressed during a prolonged approximately 3 °C warming in the mid-Holocene 14 (Fig. 3a), a magnitude similar to the warming predicted for the coming decades that will probably result in the complete deglaciation of the Rwenzori Mountain range 15 , 27 , 28 . These patterns of temperature and rainfall change (Fig. 3a,b) are consistent with records from across much of tropical East Africa 4 , 14 , 15 and led to the reorganization of Rwenzori plant community composition 4 , 10 , 16 . For instance, palynological records from the Rwenzori Afroalpine zone document gradual upslope and downslope migration of the tree line during the mid-Holocene associated with warming and cooling 10 , 16 (Fig. 3a–c). However, our charcoal record shows that substantial Holocene climate changes and plant community restructuring were insufficient to alter fire activity in the Rwenzori Afroalpine zone 29 , 30 , 31 .

Montane fire changed vegetation

Our record of fire at Lake Mahoma reveals the impacts of past changes in fire activity on mid-elevation Afromontane ecosystems. Despite substantial changes in climate (Fig. 3a,b), the mid-elevations experienced low fire activity through most of the Holocene (Fig. 3d), similar to high elevations (Fig. 2c and Extended Data Fig. 4). Modest increases in charcoal (<2 particles per cm 2 per year) associated with the Younger Dryas (about 12 ka), an interval of high-latitude cooling and regional drying in East Africa, and after the end of the African Humid Period (about 10–6 ka) probably reflect fire by-products derived from lowland savannah sources 32 (Extended Data Fig. 5). These time intervals include the coldest and warmest intervals in our record, respectively, suggesting that temperature alone had a limited impact on fire. However, they are both periods of low precipitation, highlighting the sensitivity of lowland savannah fire to changes in moisture 18 .

The fourfold increase in charcoal accumulation in the late Holocene (about 2 ka) is notable and is similar to broader patterns of fire activity in East Africa 4 , 18 , 24 , 33 . Although late Holocene accumulation rates at Lake Mahoma are far lower than what is observed in modern Lake Kopello during the 2012 fire (Methods), changepoint analyses (R ‘changepoint’ v2.3 (ref. 34 )) of Lake Mahoma charcoal and Poaceae (grass) pollen time series data reveal that the sudden increase in charcoal accumulation (1.73 ka changepoint; Fig. 3d) immediately preceded an abrupt increase in both Poaceae pollen percentage (1.67 ka changepoint; Fig. 3c) and pollen influx (Extended Data Fig. 4). This near-simultaneous increase in charcoal and grass pollen indicates a local disturbance that transformed late Holocene montane ecosystems (Fig. 3c,d).

Today, Lake Mahoma is surrounded by dense stands of native bamboo (Sinarundinaria alpina), a C 3 fire-tolerant mountain grass 10 (Extended Data Fig. 6). Mountain bamboo in East Africa is considered a successional stage of montane forest following disturbance, and taxa associated with lower montane forest regeneration have even been found to co-occur within bamboo ecosystems 35 . Late Holocene vegetation succession at Lake Mahoma is marked by the gradual decline of lowland montane forest such as deciduous tree species Celtis africana and the gradual expansion of African yellowwood (Podocarpus) as climate cooled and dried, followed by the abrupt increase in Poaceae (bamboo) 4 , 10 when fire increased (Fig. 3c). The rapid establishment of a mountain bamboo-dominated ecosystem shortly after an abrupt increase in fire at about 2 ka demonstrates the consequence of fire disturbance in Afromontane ecosystems.

Although natural climate variability can cause ecosystem disturbance and catalyse changes in fire activity, the abrupt increase in fire at about 2 ka was not associated with substantial changes in temperature or precipitation at Lake Mahoma (Fig. 3a,b). Nearby evidence for late Holocene hydroclimate fluctuations comes from Lake Edward (912 m a.s.l.), located south of the Rwenzori, in which drying initiated around 5.2 ka and was followed by a succession of century-scale droughts, including drought-induced lake low stands at 4.2 and 2 ka (ref. 36 ). Lake George, which is connected to Lake Edward by the Kazinga Channel, was desiccated around 4 ka (ref. 36 ). This rainfall history is similar to but also differs from that of the central Congo, in which an interval of drying and reduced rates of peat accumulation from about 5 to about 2 ka was followed by a return to much wetter conditions, similar to the African Humid Period 37 . Similarly to Lake Mahoma, both Lake Edward 4 and Lake Katinda 18 (1,024 m a.s.l.) show an increase in fire activity around 2 ka (Supplementary Fig. 2). Dry conditions and persistent intervals of drought following the end of the African Humid Period may have contributed to the increase in fire activity at Lake Mahoma and regionally between 3 and 2 ka, but this does not explain the sudden shift in fire activity at Lake Mahoma.

We instead attribute the abrupt increase in fire to an intensification of human activity in Central Africa between 2 and 3 ka (refs. 4 , 38 , 39 ). Evidence for human occupation in the lowlands surrounding the Rwenzori extends into the Pleistocene 40 , 41 , 42 , 43 , 44 , and existing records are insufficient to estimate changes in human populations during the Holocene. However, changing technological and subsistence practices, such as the emergence of agricultural, pastoralist and iron-smelting practices 35 , 45 , intensified human impacts on the landscape over the past 2 kyr, similar to changes in Western Africa 39 and the Congo Basin 46 .

Human activity has been invoked to explain environmental changes in Uganda during the past two millennia 4 , 43 , 47 , and although archaeological records in the eastern Congo are sparse, evidence for human activity in western Uganda includes dated archaeological materials such as pottery and metallurgical objects, as well as evidence for changes in land cover, fire and surface runoff 18 , 35 , 39 , 43 , 45 , 46 . We synthesized 14 C dates from four archaeological sites in southwestern Uganda (Fig. 3e, Supplementary Figs. 3 and 4 and Supplementary Table 3) into a summed probability distribution (SPD) (Supplementary Information). Although there are limitations to SPDs, including sampling and preservation bias, SPDs can provide useful information about past human activity. Their summed probability shows a high density of dated archaeological materials at about 2 ka (Fig. 3e), coincident with the increase in fire and change in fire activity at Lake Mahoma (Fig. 3d). Schmidt et al. 43 attribute burning events at the base of the Rwenzori Mountains in the Ndali Crater Lakes Region (Supplementary Fig. 1) at about this time to forest clearance by Later Stone Age peoples. Although iron smelting is often used to explain changes in human activity in Central Africa at about 2 ka (refs. 4 , 39 ), archaeological evidence suggests that inhabitants of southwestern Uganda about 2,000 years ago relied on lithic technology, with iron smelting arriving later 41 , 43 .

There is no direct archaeological or palynological (for example, crop taxa) evidence for human occupation in the Rwenzori Mountains; however, evidence for human occupation has been documented as high as about 4,200 m a.s.l. in Africa during the Pleistocene in the Bale Mountains 48 , and humans who periodically occupied caves on Mount Sodicho, Ethiopia affected surrounding Afromontane vegetation 49 . Furthermore, a loss of forest taxa in the lower montane forests (about 2,000 m a.s.l.) of the Rukiga Highlands in southwestern Uganda at around 2.3–2.2 ka has been attributed to forest clearance by humans 47 . We cannot be certain of local human activity within the Rwenzori owing to a lack of archaeological exploration, and our work highlights the need for further archaeological studies in Afromontane environments. It is possible that fires in the lowlands moved upslope, potentially abetted by dry late Holocene conditions, affecting elevations as high as Lake Mahoma. Although the exact relationships between late Holocene human demographics, agricultural practices, metallurgy and fire use in western Uganda remain uncertain, our data show that an abrupt increase in fire activity across the region at around 2 ka (refs. 4 , 18 , 33 , 43 ) (Supplementary Fig. 2) occurred alongside an intensification of human activity and led to a fire-mediated vegetation community transformation of Rwenzori Afromontane ecosystems.

Fire impacts and management

Previous work has suggested that tropical Afroalpine fire is rare; therefore, fire management plans have not been considered essential 25 . However, human activities, including poaching and a growing tourism industry in the mountains, can lead to increased ignitions. Our work indicates that tropical Afroalpine fire outside the context of human ignition is rare, but climate change (that is, surface warming and drying) can promote favourable conditions to spread fires 1 , 9 ignited by anthropogenic activity. We show that the Afroalpine ecosystem of the Rwenzori has no natural fire activity, which implies that modern fires could have an enormous impact on local plant communities, such as established species becoming outcompeted in a new disturbance regime 50 . Recent fires have resulted in the loss of a third of Kilimanjaro’s forest cover over the past 70 years and have pushed the upper limit of the closed forest downslope by 400 m (ref. 7 ). On Mount Elgon, Uganda–Kenya, experimental burns revealed that Afroalpine ericaceous species were slow to recover post-fire, whereas Afroalpine grasses experienced fast recovery. This difference in post-fire recovery rate led to the replacement of ericaceous heathers with grassland 25 —similar to the replacement of montane forest with bamboo at Lake Mahoma approximately 2 ka. Afroalpine plant communities are already projected to experience extraordinary range reductions as their climatic range moves to higher elevations with less available land area as temperatures warm, with the potential for the total elimination of their climatic range in Africa in the future. Fire, especially under changing climate conditions, has the potential to greatly accelerate this ecosystem transformation, posing a potential threat to Afromontane ecosystems already jeopardized by climate change.

Recent montane fires have amplified natural disasters affecting downstream communities, most notably through deadly flooding and landslide events. The 2012 Rwenzori fire burned 18% of the catchment area, which reduced water retention capability and, soon after, extreme rainfall led to massive floods, resulting in numerous fatalities and substantial infrastructure damage 11 . Thus, the creation of management plans that focus on preventing human ignitions and preparedness for suppression are critical to protect tropical Afromontane habitats, tourism infrastructures and local communities in the twenty-first century, especially in the context of future climate extremes 1 .

Methods

Site description

The Rwenzori Mountains are the third tallest mountain range in Africa (5,109 m a.s.l.) and straddle the Uganda–Democratic Republic of Congo border 29 . They are one of three glaciated mountain ranges in Africa at present, alongside Mount Kilimanjaro and Mount Kenya, but the Rwenzori contains the largest number of glaciers and most expansive Afroalpine ecosystem in Africa 16 . The altitudinal range of the Rwenzori hosts rich floral biodiversity across five main vegetation zones: the savannah (<1,800 m a.s.l.), the montane forest zone (about 1,800–2,400 m a.s.l.), the bamboo zone (about 2,400–3,000 m a.s.l.), the ericaceous belt (about 2,700–3,900 m a.s.l.) and the Afroalpine belt (>3,900 m a.s.l.) 10 , 15 , 16 . Lake Mahoma is located on a moraine ridge in the bamboo zone, and Lake Kopello is located in a glacially carved valley in the Afroalpine zone. The fauna of the Rwenzori Mountains today consists of eastern chimpanzees (Pan troglodytes) along with other primates, such as blue monkeys (Cercopithecus mitis), as well as populations of red duiker (Cephalophus natalensis) and rock hyrax (Procavia capensis). Today, no large herbivores reside in the mountains with the exception of a very small population of African forest elephants (Loxodonta cyclotis), estimated to be around seven individuals as of 2021 (ref. 51 ).

Sediment cores and age model

Two lake sediment cores were recovered from Lake Mahoma (2,990 m a.s.l): a Bolivia core recovered in 2008 (MAHO08: coordinates 0.34° N, 29.96° E, water depth 11.8 m) 15 and a Livingstone core recovered in 2019 (MAHO19: coordinates 0.34° N, 29.96° E, water depth 11.5 m) 15 . Similarly, two sediment cores were recovered from Lake Kopello (4,017 m a.s.l.,): a Livingstone core recovered in 2019 (LKOP19: coordinates 0.31° N, 29.89° E, water depth 11 m) 14 and a gravity core recovered in 2024 (LKOP24, coordinates 0.31° N, 29.89° E, water depth 11.4 m) 14 . A previously published composite stratigraphy from Garelick et al. 15 was made for the 2008 and 2019 Lake Mahoma sediment cores by cross-correlating visual features such as a shared tephra layer and magnetic susceptibility 14 (Supplementary Table 2). The Lake Mahoma core composite has an average sedimentation rate of 37.04 years per cm. A similar approach was used by Mason et al. 14 to create a composite stratigraphy for the 2019 Lake Kopello sediment cores 14 . In this study, we build on the Lake Kopello core composite by including the 2024 Lake Kopello gravity core, which adds a sediment sequence from 0–52 cm to the total composite. We failed to recover a total of 18 cm of sediment (52–70 cm composite), which is missing in the final Kopello core composite (Supplementary Table 1). We denote this missing sediment sequence in Fig. 2c with hash marks on the x axis. The average sedimentation rate is 11.44 years per cm in the deeper Livingstone core sediments, and the 2024 gravity core, which captured the sediment water interface, has an average sedimentation rate of 2.63 years per cm. The Lake Kopello composite is a total of 13 sections and the Lake Mahoma composite is a total of six sections (Supplementary Information).

We used a recently published age model for Lake Mahoma based on radiocarbon ages 14 and constructed a new age model for Lake Kopello that incorporates the recently recovered Lake Kopello 2024 gravity core. The new Lake Kopello age model was constructed using previously published radiocarbon ages 14 calibrated using IntCal20 (ref. 52 ) from the 2019 Livingstone cores and 12 new 210 Pb dates from the 2024 gravity core analysed by Flett Research Ltd. The Lake Kopello 2024 gravity core was also analysed for 137 Cs to constrain the 1963/1964 nuclear bomb testing; however, no 137 Cs was detected. In the recent gravity core samples, an age–depth relationship was calculated using the constant rate of supply model (CRS) for 210 Pb (ref. 53 ) (Extended Data Fig. 7). We made a final age–depth model for Lake Kopello using the ‘plum’ function in the R package Bacon (v3.5.2) 54 , which incorporated the previously published radiocarbon dates 14 and the new CRS 210 Pb results (Extended Data Fig. 7).

Charcoal analysis

The Lake Kopello 2024 gravity core was subsampled continuously at 1-cm resolution and the 2019 Lake Kopello Livingstone cores were subsampled continuously at 2-cm resolution. The Lake Mahoma 2019 Livingstone cores and 2008 gravity core were subsampled continuously at 1-cm resolution owing to slower sedimentation rates than at Lake Kopello. Sediment was volumetrically sampled (2 cm 3 ) for Lake Mahoma and Lake Kopello. All sediment samples were treated with 1 M sodium pyrophosphate (Na 4 P 2 O 7 ) and 2.5% sodium hypochlorite (NaClO), then left to soak for 12 h. After soaking, sediments were wet sieved at >250 µm to isolate larger charcoal particles because this is most likely to identify relatively proximal fires 19 , 55 , 56 . Sieved samples were transferred to a plastic dish suspended in water and counted manually using a binocular dissection microscope at 6.4× magnification. Charcoal L:W ratio data from the Lake Kopello gravity core was quantified by capturing photos of the particles with a Nikon DSFi3 microscope camera and measuring images with CharTool (v3.2) 57 . A total of 475 charcoal samples were counted from Lake Kopello and a total of 329 from Lake Mahoma. Charcoal counts were converted to charcoal accumulation rate using the following calculation:

Charcoal sources to Lake Mahoma and Lake Kopello

We attribute the differences in charcoal accumulation rate at Kopello and Mahoma to the very different fluvial and geomorphic settings of the two lakes that affect charcoal transport. Charcoal accumulation rates for local fire events differ between Lake Kopello and Lake Mahoma, probably because of differences in charcoal transport mechanisms. Charcoal particle size influences dispersal distance, but other factors such as particle shape, density and clumping, as well as fire size, severity and intensity and vegetation type, are also important in determining charcoal dispersal distance 58 , 59 , 60 . Basin and watershed size as well as the type of vegetation burned are also known to affect charcoal accumulation in lakes 61 . Lake Kopello and Lake Mahoma differ in their geological settings, which probably affected the variation in charcoal accumulation at each site. Lake Kopello is located in a valley and receives riverine fluxes 29 . Lake Mahoma is located on a moraine ridge 15 and has no river inputs; thus, charcoal transport is largely aeolian. Lake Mahoma background charcoal levels are higher than for Lake Kopello over the Holocene, which is probably because of Mahoma’s relative proximity to lowland ecosystems and shorter transport distance.

To reconstruct a record of local fire and to reduce the influence of long-distance transport of regionally sourced charcoal, we focused our analysis on the >250 µm charcoal fraction. Although we analysed a relatively large size fraction, both lakes still received charcoal, especially grass-fire-derived charcoal, from lower elevations. This is confirmed by similar trends in charcoal accumulation at both lakes over the Holocene, such as enhanced charcoal accumulation over the past roughly 2 kyr and the end of the African Humid Period (Extended Data Fig. 4). The similarity between these two records suggests that both lakes receive at least some charcoal input from regional fires in the form of what has been termed in the literature ‘background charcoal’ 62 .

In Lake Kopello, the long-distance origin of this charcoal is confirmed by the exceptionally low fluxes of charcoal to sediments before 2012 (Fig. 2c) and L:W ratios (as well as width:length (W:L) ratios; Extended Data Fig. 8 and Extended Data Table 1), which indicate that the charcoal in pre-2012 sediments is derived from grass and differ from the L:W ratio of the 2012 fire, which burned the Lake Kopello catchment (personal observation; Extended Data Fig. 1). We find that pre-2012 mean L:W values are in the range of grass fuel sources and differ significantly from post-2012 values (Wilcoxon rank-sum test, n = 14, W = 24, P = 0.021). We postulate that this significant difference between before and after the 2012 fire is caused by the change from a regional background input of charcoal to a localized fire signal. On the basis of these data, we conclude that most of the Lake Kopello gravity core charcoal is derived from distant lowland savannah fires.

In the case of Lake Mahoma, charcoal derived from long-distance transport does comprise some of the record (Extended Data Fig. 8 and Extended Data Table 2). However, we interpret the increased accumulation around 2 ka to reflect local fire because of the magnitude of the increase in charcoal particle flux. Furthermore, this increase occurs simultaneously with an abrupt ecological shift documented in fossil pollen that would be possible only with a localized disturbance.

Further evidence for regional charcoal input is evident during the twentieth and twenty-first centuries at Lake Kopello, when charcoal accumulation is observed during times when there are no documented or observed fires in the catchment or the surrounding Rwenzori Mountains National Park (Fig. 2d). This charcoal therefore cannot be derived from local fires but rather must be derived from regional fires. Modern satellite-derived fire data provide further evidence for regional charcoal deposition. We examined NASA Fire Information for Resource Management System (FIRMS) MODIS fire point data spanning 2002 CE to the present (2025 CE) using QGIS (v3.24.0). These data provide evidence for numerous fires that occurred within a 50-km radius (a reasonable estimate for charcoal sourcing 58 ) over the past few decades. Furthermore, many of these fires produced injection heights great enough to reach the elevations of both Lake Kopello and Lake Mahoma (Extended Data Fig. 3). By extracting the MODIS-derived fire radiative power for each of these fires (within 50 km) and comparing these data with the USGS GMTED2010 digital elevation model, we calculated the smoke fallout horizon for each fire in the dataset (that is, particle injection height) using the equations inherent to most charcoal dispersal models (originally presented by Clark 63 but then also used by Higuera et al. 62 and Vachula and Richter 64 ). We then added these injection heights to the elevation of the fire (as derived from intersecting the MODIS fire points with the digital elevation model) to calculate the absolute particle injection height for each fire. We then identified those fires that had fire radiative powers sufficient to inject charcoal particles into the atmosphere to a height equivalent to the elevations of Mahoma and Kopello (Extended Data Fig. 2).

Lake Mahoma pollen

We present a pollen record from Lake Mahoma to understand the impact of local fire on vegetation communities. Pollen samples were processed using standard methods 65 . Lycopodium tablets were added to calculate pollen concentrations and samples were sieved with 10-μm screens to remove clays. Pollen grains were identified using a pollen reference collection in the Ivory lab at Penn State University and pollen atlases 66 , 67 . A minimum of 300 grains were counted in each sample, and taxon names follow the standard schema of the African Pollen Database 68 , 69 . Pollen abundances are based on the sum of terrestrial pollen types excluding aquatic plants (calculated separately). Pollen counts from Lake Mahoma are a combination of new data from this study, published pollen counts from ref. 10 and published pollen counts from ref. 16 . In this study, we add eight pollen counts to the previous 18 pollen samples from ref. 10 , which were also analysed on cores MAHO19 and MAHO08. We also include 37 previously published counts from ref. 16 , which were counted on a different previously recovered core. To create a new comprehensive Mahoma pollen composite, we aligned the 37 Livingstone counts to the 26 counts from the MAHO08 and MAHO19 cores. The pollen alignment between these two cores was based on radiocarbon ages, tephra location and matching vegetation trends. We present a total of 63 pollen samples with an average pollen resolution at Lake Mahoma of 0.184 kyr and higher-resolution counts over the late Holocene fire interval (past approximately 2 kyr) averaging 0.086 kyr resolution. Across all samples, a total of 141 taxa were identified, but we chose to highlight Celtis africana, Podocarpus and Poaceae to represent the lower-elevation forests, upper montane forest and bamboo zone, respectively. Pollen concentrations are not available from the Livingstone samples, so we present pollen percentage data, which shows a sharp increase in Poaceae in late Holocene sediment (Fig. 3c). Pollen concentration data confirm that the increase in fire at roughly 1.7 ka is followed by a sharp spike in Poaceae pollen inputs (Extended Data Fig. 4).

Mahoma charcoal and pollen changepoint analysis

We used changepoint analysis to identify the relationship between fire and ecosystem transformation in our Lake Mahoma Holocene time series. We used the R package ‘changepoint’ (v2.3) 34 and the pruned exact linear time method to identify several changes in both mean and variance throughout the time series. We identified a total of four changepoints in charcoal accumulation (Fig. 3d) and two changepoints in Poaceae pollen (Fig. 3c). The charcoal changepoint at 11.630 ka occurs around the end of the arid Younger Dryas, and the changepoint at 6.192 ka marks the end of the wet African Humid Period. We infer that these two changepoints are the result of changes in regional precipitation driving changes in lowland fire activity. By contrast, the changepoints at 1.730 ka and 0.612 ka reflect changes in fire that are not associated with changes in aridity and coincide with a period of cooling. These changepoints denote the beginning and end of enhanced late Holocene fire. The initiation of fire at 1.730 ka shortly precedes a changepoint in Poaceae at 1.665 ka. Mountain bamboo is a pioneer species, which is able to expand quickly following disturbance 35 , so we interpret the near simultaneous rise in charcoal and Poaceae to reflect a local disturbance. Another Poaceae changepoint in the older section at 10.019 ka probably reflects a decrease in grass as precipitation increases at the start of the African Humid Period. We focus our discussion on the charcoal accumulation changepoint at 1.730 ka and the Poaceae pollen changepoint at 1.665 ka. We suggest that the changepoint in charcoal shortly before a changepoint in Poaceae is evidence for a fire-mediated ecosystem transition.

Lake Mahoma bamboo expansion

We conclude that fire at about 1.7 ka catalysed the abrupt shift to a grassy ecosystem at Lake Mahoma around 1.6 ka. Grass abundances reached >30% of total pollen counts at times in the late Holocene, and, although it is not possible to distinguish what type of grass increased in abundance using pollen analysis, today, Lake Mahoma exists in the bamboo zone 10 , 14 , 15 , dominated by Sinarundinaria alpina, a tropical C 3 grass (Extended Data Fig. 6). A C 31 leaf wax alkane δ 13 C record from Lake Mahoma 14 shows that values across the vegetation assemblage varied during the late Holocene (Extended Data Fig. 6) but did not fall within ±1 standard deviation of the median δ 13 C value for C 31 leaf wax n-alkanes for tropical C 4 grasses (22.1 ± 2.6‰, n = 12) 70 , 71 or the C 31 alkane African C 4 endmember (20.1 ± 2.5‰) 72 described in ref. 72 . It is clear that the grass expansion during the late Holocene was not a C 4 grass and was more likely to be a C 3 grass given the δ 13 C isotope values over this interval. Sinarundinaria alpina is a C 3 grass, and other grasses are uncommon in the Lake Mahoma catchment today (personal observation; Extended Data Fig. 6). We conclude that fire in the late Holocene catalysed a transition to the modern-day C 3 bamboo forest around 2 ka.

Rwenzori Mountains climate records

We include existing Holocene records of precipitation and temperature from the Rwenzori Mountains (Fig. 3b). We used leaf wax C 31 δ 2 H precipitation records from ref. 14 from Lake Kopello and Lake Mahoma, which used the same sediment cores as this study. Mason et al. 14 include an extra precipitation record from Lake Africa (3,895 m a.s.l.), which is located about 2 km downstream from Lake Kopello. Trends in the Rwenzori δ 2 H precipitation records over the Holocene match regional trends in Africa: drying (δ 2 H enrichment) in the Younger Dryas (about 12 ka), a wetter African Humid Period (10–5 ka, δ 2 H depletion) and return to drier conditions (δ 2 H enrichment) in the late Holocene (about 2 ka) 14 .

Furthermore, we include a mean annual air temperature record from ref. 15 , which reconstructed temperature using branched glycerol dialkyl glycerol tetraethers in the same Lake Mahoma sediment cores as this study (Fig. 3a). Temperatures were calculated using the East African calibration of MBT′ 5ME which is based on the methylation of branched tetraethers of 5-methyl isomers 15 . At Lake Mahoma, temperatures were coolest in the Younger Dryas (about 12 ka) and late Holocene, whereas temperatures were warmest during the mid-Holocene warm period (about 7–5 ka).

登录APP查看全文