For pollen and spore extraction, 22 samples of 0.5 cm³ samples at intervals between 5 and 10 cm were prepared following a standard chemical protocol (Faegri and Iversen, 1975; Daniau et al., 2019). Pollen grains and fern spores were counted using a light microscope (Zeiss) at ×400 magnification, and a minimum sum of 300 terrestrial pollen grains were counted for each sample. The references consulted for identification of pollen and spores included several atlases (Straka and Freidrich, 1989; Gosling et al., 2013; Schüler and Hemp, 2016; Rasoloarijao et al., 2019), the online African Pollen Database (06/2021²) and the reference collections of University of Göttingen (06/2021³) and ISEM (06/2021⁴). Percentages of pollen and fern spores were computed on the terrestrial pollen sum and the pollen record was subdivided into significant zones by applying a constrained cluster analysis by sum of squares performed with terrestrial pollen (CONISS, Grimm, 1987). Fungal spores were counted on pollen slides at an interval varying between 10 and 30 cm. A minimum fungal spore sum of 100-300 was counted for each sample, except for the depth 96 cm where only 73 spores were found. Fungal spore and other NPP identification are based on NPP Image Database (Shumilovskikh L. S. et al., 2021). Statistical analyses were performed using the RStudio software (Version 1.2.1335) with the package “Rioja” (Version 0.9-21) (Juggins, 2015).

For macro-charcoal counting a total of 220 contiguous sediment samples of 0.5 cm³ were retrieved. The extraction of sedimentary macro-charcoal particles followed the standard protocol from Stevenson and Haberle (2005). Macro-charcoal particles were counted using a digital microscope at 5.0 magnification (Zeiss) and wooden and grass-based charcoal were distinguished following morphological description from Enache and Cumming (2006). Macro-charcoal concentration (particles cm^–3) and macro-charcoal influx (particles cm^–2 year^–1) were computed for each sample.

To reconstruct the ecological shift occurring at the transition between the two sediment cores with high precision, Poaceae pollen and macro-charcoal particle were counted on 15 contiguous additional samples from core AMP16-1 that encompasses this transition between the depth 110 and 96 cm. These detailed analyses performed every centimeter (for pollen and macro-charcoals) allow to trace environmental changes with high precision, without any potential disturbance from the switch between the two sediment cores in the composite master sequence AMP16 (Figure 2 and Supplementary Figure 1). For these contiguous samples, the same previously mentioned chemical treatment for pollen and macro-charcoal particle extraction were used. To obtain Poaceae percentages, a minimum of 100 terrestrial pollen grains were counted among which only Poaceae was identified.

Diatom analyses were performed every 6-10 cm. Sample preparation and concentration assessment were conducted following standard procedures (Kalbe and Werner, 1974; Battarbee and Bennion, 2011). If possible, a minimum of 400 valves was identified in each sample. In samples with very low diatom concentration, analyses were stopped when 1000 microspheres had been counted. A simple preservation index was tested by counting at least 100 specimens of Aulacoseira granulata and differentiate between intact and broken valves (ratio I/B). Permanent slides for light microscopy were prepared using Naphrax^® as a mounting medium. Slides analyses was performed using a Leica DM 5000 B light microscope with differential interference contrast, equipped with a ProgRes^® CT5 camera, under oil immersion at ×1000 magnification. Identification is mainly based on (Krammer and Lange-Bertalot, 1986, 1988, 1991a,b), Houk and Klee (2007); Houk et al. (2010), Lange-Bertalot et al. (2017) and other relevant taxonomic publications (e.g., Klee et al., 2000). Stratigraphic diatom zones (DZ) were defined by hierarchical cluster analysis of taxa occurring with > 1% using PAST ver. 4.04 (Hammer et al., 2001). Down-core changes in diatom results were visualized using C2 software version 1.7.7. (Juggins, 2007).

The composite magnetic susceptibility pattern shows values between 86 and 7035 × 10^–6 SI (Figure 3) with noticeable distinct narrow peaks matching the light grayish and reddish ERDs (Supplementary Figure 1). These peaks also represent the maximum values in the record. In unit I (2940 to 1300 cal BP) values are generally on a low level of around 600 × 10^–6 SI. Values drop down from around 1800 × 10^–6 SI in unit II (1300 to 1000 cal BP) to ca. 1000 × 10^–6 SI at 1000 cal BP in unit III (1000 to 820 cal BP). They distinctly increase at 890 cal BP to 3400 × 10^–6 SI and decline steadily thereafter until the top of unit III (820 cal BP) reflecting the uppermost ERD. Unit IV (820 cal BP - today) shows further declining magnetic susceptibility values with some internal variation to 800 × 10^–6 SI at 500 cal BP. After another broad peak centered at ca. 430 cal BP showing values of ca. 1600 × 10^–6 SI, the remaining record reveals fairly stable values ranging around 1100 × 10^–6 SI. However, the distinct minimum during the youngest part of the record (after 10 cal BP) is likely attributed to measurement edge effects.

Grain size distributions in all samples are characterized as bi- to polymodal and poorly sorted (Figure 3 and Supplementary Figure 3). Median grain size ranges between 4.9 and 16.4 μm. Unit I (2940 to 1300 cal BP) displays the occurrence of coarser particles between 6.6 and 16.4 μm, with a general fining trend. Particle size median (D50) is distinctively smaller 4.9 and 9.4 μm after 1300 cal BP and shows no grading trend and only minor variability. Particle sorting (σ), however generally described as poor, is quite variable over the entire covered time span. In unit I, deposits are very poorly sorted. From 1300 to 820 cal BP (unit II and III) sorting tends to be better, but highly variable and remains quite stable afterward in unit IV (820 cal BP - today). Skewness (S_k) in the distribution of all samples is positive, but comparable to the sorting showing internal variability. In the older deposits (unit I, 2940 to 1300 cal BP) S_k reveals lower values with some variability and an increasing trend. In unit II and III, from 1300 to 820 cal BP, S_k is on an intermediate level with high internal variability and remains on a high positive level toward the top of the record (unit IV).

T(O)C, TN as well as the C/N ratio follow the units of the AMP-16 record, but unit II and III (1300 to 820 cal BP) are merged (Figure 3). In general, TC varies between 2.1 and 13.4%, and TN shows contents from 0.2 to 1%. Stable, high amounts of TC and TN are observed within unit I (2940 to 1300 cal BP). In unit II (1300 to 1100 cal BP), TC and TN drop distinctly and remain low and relatively stable, particularly during unit IV (after 820 cal BP). The C/N ratio drops from relatively high values, partly exceeding 16 in unit II (at 1200 cal BP), to values below 14 in the units above. δ¹³C reveals the most negative values of around -30.3 ‰ in unit I (2940-1300 cal BP), a shift to more positive values (>-27 ‰) in unit II and III (until 820 cal BP), and values stabilize around -28 ‰ in unit IV (after 820 cal BP) (Figure 3). Stable values for δ¹⁵N of around 1.8 ‰ occur in unit I (2940 to 1300 cal BP) as well as in untis II and IV (after 1000 cal BP), but reveals a distinct enrichment of up to 4.3 ‰ in unit II (Figure 3).

Biogenic silica (BiSi) contents range between 2.6 and 23.7% (Figure 3). The amount shows a decreasing trend during unit I and unit II until 1100 cal BP, when it reaches its minimum. Contents are dramatically increasing to the maximum thereafter and are again decreasing at ca. 700 cal BP, however remaining on a constant intermediate level of ca. 12% toward the top.

n-Alkanes occur in all investigated samples, and the chromatograms show the odd over even predominance typical for plant-derived leaf-waxes (e.g., Schäfer et al., 2016). Because of the low concentrations of the even n-alkanes and insufficient peak separation from other aliphatic compounds, we quantified only the more abundant odd n-alkanes. The sum of the odd n-alkanes (nC₂₁ to nC₃₅) reaches 20 μg g^–1 dry sediment in unit I (2940 to 1300 cal BP), drops below 10 μg g^–1 in unit II and III (1300 to 820 cal BP), and increases again over the last few centuries (Figure 4). The Paq is low at ca. 0.1 in unit I, increases to values > 0.2 in unit II and III, and exceeds values > 0.4 in unit IV. The ACL mirrors this trend, starting with values around 30.5, dropping in unit II and III, and reaching values < 29.5 in unit IV. Excluding C₂₇, or even C₂₇ and C₂₉, for the calculation of the ACL′ and ACL″, respectively, of course increases the values, but it also reduces and even removes the trend toward shorter chain length. The remaining feature of the ACL″ is a pronounced maximum in unit II.

A total of 93 different pollen and spore taxa and 89 NPP types have been identified. Based on the cluster analysis (CONISS) the pollen diagram is subdivided into three significant pollen zones (PZ, Figure 5).

PZ-1 (165-100 cm/2940-1300 cal BP/unit I): Pollen assemblages mainly consist of arboreal pollen (AP 80-90%) and show a high tree diversity. The most abundant AP correspond to Moraceae/Urticaceae (20-30%), Celtis (ca. 20%), Macaranga/Mallotus (ca. 10%) and Elaeocarpaceae (ca. 10%). Other AP with low values (< 5%) are also characteristic for this zone such as Podocarpus, Weinmannia, Ilex, Meliaceae/Sapotaceae, Anacardiaceae, Noronhia, Proteaceae and Ilex. Non-Arboreal Pollen (NAP) corresponding to herbs and shrubs show low values (5-15%) and are mainly characterized by a steady occurrence of Poaceae (5-10%), Acalypha (ca. 2%) and Begoniaceae (ca. 2%). Aquatic pollen taxa do not exceed 10% and fern spores are absent. Lignicolous fungal spore assemblage is diverse and presented by Xylariaceae, Potamomyces, Brachysporium, Dictyosporium, Canalisporium pallidum. Plant pathogens are presented by Ustulina deusta and Lasioplodia. Saprotrophs/coprophilous spore assemblage is poor. Among unknown NPPs, HdV-1033 and HdV-1036 are dominant.

PZ-2 (100-27 cm/1300-600 cal BP/units II, III and IV): AP diversity decreases as well as AP percentages that drop abruptly to 30%, then increase progressively up to 40% at the end of the PZ. Among AP, only percentages of Trema increase slightly (up to 5-10%). Moraceae/Urticaceae (ca. 15-30%) and Macaranga/Mallotus (ca. 5%) still remain among the dominant AP. Poaceae pollen increase abruptly up to 60% then decrease progressively and reach 50% at the end of the PZ showing an opposite trend to AP. Other NAPs remain with low values. Ericaceae percentages show a slight increase (up to ca. 5%) and slight percentage decreases are recorded for Acalypha (<2%) and Begoniaceae (<2%). Percentages of aquatic pollen are characterized by an important increase in Potamogeton, a slight increase in Cyperaceae and Lemna is not recorded anymore. A steady occurrence of fern spores is recorded but with low values (< 3%) and the algae Botryococcus shows an increase at ca. 20% from ca. 1000 cal BP. Considerable changes in fungal spores are indicated by a sharp increase in Coniochaeta ligniaria associated with Delitschia, Cercophora and Sordaria. Spores of pyrophilous fungi Gelasinospora and Neurospora occur, while diversity of lignicolous assemblage decreases. Plant pathogens are presented by Lasioplodia, Tetraploa aristata, Dictyoarthrinium cf. sacchari and rusts uredospores.

PZ-3 (27-0 cm/from 600 cal BP/unit IV): The progressive increase in AP which started in PZ-2, continues in this zone with AP reaching 70%. AP are mainly dominated by Moraceae-Urticaceae (20-30%), Macaranga/Mallotus (5-10%) and Trema (20%). This zone is also characterized by the occurrence of new AP taxa such as Mimosa (up to 5%) or Myrtaceae (up to 3%) while several taxa characteristic of the first zone such as Podocarpus, Weinmannia, Ilex or Meliaceae/Sapotaceae are almost not recorded anymore. Among the NAP, only Poaceae pollen indicate significant changes characterized by a progressive decreasing trend from 45% to 25%. At ca. 10 cm depth an increase in the aquatic pollen Cyperaceae (up to 8%) is recorded. The other aquatic pollen, fern spores and Botryococcus do not show marked changes. PZ-3 is characterized by a further increase in diversity of coprophilous assemblages by Arnium, Apiosordaria and Podospora anserina-type. Chlamydospores Glomus-type occur constantly. Assemblage of lignicolous fungi change to Curvularia, Podospora curvispora, Canalisporium pulchrum, Megalohypha aqua-dulces, Caryospora, Zopfiella lundquistii. Plant pathogens diversity increase by several Spegazzinia species. Diversity of other spores decreases.

Prior to 1300 cal BP, macro-charcoal particles occur in every sample, but concentrations and influx remain very low (Figures 6A,B). Following morphological descriptions from Enache and Cumming (2006), most macro-charcoal particles are wood-based. At ca. 1300 cal BP (unit II, 1300 to 1000 cal BP), a marked increase of concentration and influx is recorded with peaks exceeding 300 particles cm^–3 and 40 particles cm^–2 year^–1, respectively. Grass-based charcoal particles increase at the same time with percentages varying around 30% and peaks exceeding 40%. Subsequently, percentages decrease after 1000 cal BP and remain generally below 30%. From 1000 cal BP (unit III and IV, 1000 cal BP to today), macro-charcoal concentration and influx also tend to decrease. After a last peak above 300 particle cm^–3, between 820 and 750 cal BP, concentrations fluctuate around 100 particles cm^–3. Macro-charcoal influx indicates a progressive decreasing trend from 1000 cal BP and values stabilize from ca. 750 cal BP varying around 5 particles cm^–2 year^–1. This decrease in influx is partly the result of the decrease in SAR (Figure 2).

Between 110 and 100 cm (ca. 1640-1300 cal BP) low percentages of Poaceae (<10%) and macro-charcoal (max. 16 particles cm^–3) are recorded (Figure 6C). Above 100 cm (<1300 cal BP), Poaceae show a continuous increase and reach 40% at 97 cm (ca. 1200 cal BP). Macro-charcoal particles show two important increases above 100 cm depth. The first increase corresponds to a peak at 99 cm and maximum concentration (368 particles cm^–3, ca.1300 cal BP) is reached in the sample where Poaceae percentage shows the first marked increase. This macro-charcoal peak at 99 cm occurs contemporaneously to a peak in magnetic susceptibility. The second increase starts at 97 cm (ca. 1200 cal BP) and macro-charcoal concentration reaches ca. 580 particles cm^–3 at 96 cm.

A total of 30 diatom species have been identified in the sequence. Overall, two planktonic taxa, Discostella cf. mascarenica (Klee, Houk and Bielsa) Houk & Klee and Aulacoseira granulata (Ehrenberg) Simonsen agg., clearly dominate the low diversity assemblages (Figure 7). Most samples yield high diatom concentrations [up to 60 × 10⁸ valves g^–1 dry weight (DW)]. However, between 1310-1040 cal BP, diatom concentration was low and valve preservation poor (ratio I/B: 0.6-0.7). Hierarchical cluster analysis identified three major diatom zones (DZ-1 - DZ-3) and three subzones (DZ3a-c).

In the oldest part of the core (DZ-1, 1630-1310 cal BP), diatom concentrations are high with values between 11 × 10⁸ to 60 × 10⁸ valves g^–1 DW. Planktonic species Discostella cf. mascarenica is by far the most abundant species (> 96%) and almost no benthic taxa are observed. The following zone (DZ-2, 1310-1130 cal BP) is distinguished by almost sterile samples with < 0.025 × 10⁸ valves g^–1 DW. First and frequent occurrences of Aulacoseira granulata together with a minor proportion (6.8%) of benthic and epiphytic taxa such as Navicula Bory de Saint-Vincent s.l., Nitzschia Hassall spp. and Gomphonema Ehrenberg spp. characterize DZ-3a (1130-1080 cal BP). Discostella cf. mascarenica is barely present anymore. Although diatom concentration is still low in the lowermost part of zone DZ-3b (1080-570 cal BP), numbers quickly increase up to 49 × 10⁸ valves g^–1 DW. Assemblages almost exclusively consist of Aulacoseira granulata agg. (>96%). The uppermost part of the sequence (DZ-3c, from 570 cal BP) is still overall dominated by Aulacoseira granulata agg. (91-96%) but benthic and tychoplanktonic taxa slightly increase in abundances (up to 7.3%) compared to the previous subzone DZ-3b.

Between 2940 and 1300 cal BP, sedimentological and vegetation data do not show major changes (Figures 3, 5A). The high dominance of tree taxa suggests a large area of forest. Most tree taxa that are described as typical of rain forest in modern and fossil pollen samples from Madagascar are well represented in this period (e.g., Moraceae/Urticaceae, Celtis, Elaeocarpaceae, Macaranga/Mallotus and Podocarpus; Burney, 1988; Gasse and Van Campo, 2001). This suggests that the dense and humid rain forest was the dominant vegetation prior to 1300 cal BP on Nosy Be. The low fire activity during the same time confirms the occurrence of rain forest developed under humid environmental conditions and low disturbance regime (Figure 6A). Furthermore, the presence of a wide variety of lignicolous fungi indicate the availability of dead wood substrate in the lake catchment as well (Figure 5B).

Sedimentological and geochemical results are consistent with the dominance of rain forest prior 1300 cal BP. Relatively low SAR supports the assumption of dense vegetation stabilizing the topsoils in the catchment (Figure 2). The comparably coarse particle median (Figure 3) might either be indicative for lacking grasses and thus a missing “filter” for larger minerogenic particles flushing into the lake by inflowing water, or coarser particles might point to comparably moist conditions with higher amounts of precipitation and thus higher transport energy of the inflows (Conroy et al., 2008). High organic carbon contents, high C/N-ratios, and high amounts of leaf-wax n-alkanes during this phase further suggest high productivity within the catchment and input of terrestrial organic matter (Meyers and Ishiwatari, 1993; Figures 3, 4). The depleted δ¹³C values of the bulk organic matter point to predominantly C₃ vegetation (Figure 3; Farquhar et al., 1989; Meyers, 1994; Diefendorf et al., 2010).

The n-alkane patterns show a dominance of nC₂₉, nC₃₁, and nC₃₃, and the Paq is accordingly very low. This indicates a dominant input from higher terrestrial plants over input from aquatic sources (Eglinton and Hamilton, 1967; Zech et al., 2010). Distinguishing between input from grasses and herbs versus deciduous trees, as has been done in other studies (e.g., Zech et al., 2013b; Schäfer et al., 2016), needs to take into account that common plants of rain forests synthesize high amounts of nC₃₁ and nC₃₃ (Gnecco et al., 1989; Abas and Simoneit, 1998; Li et al., 2016; Teunissen van Manen et al., 2019). The high ACL values in comparison to northern mid-latitude locations (e.g., Europe) thus reflect the local rain forest signal, and should not be mis-interpreted to indicate dominant input from grasses and herbs.

From 1300 cal BP, most tree taxa decrease and some of them almost entirely disappear (e.g., Podocarpus, Weinmannia, Ilex, Meliaceae/Sapotaceae, Figure 5A). Inversely, an increase of herbs (mainly Poaceae) is observed. Together with an enrichment in δ¹³C of the bulk organic fraction further suggesting a shift from C₃ to more C₄ vegetation (Farquhar et al., 1989; Meyers, 1994; Diefendorf et al., 2010), this points to the development of a mosaic of grassland and forest (Figure 3). Combined with the increase in macro-charcoal mainly associated with increased grass-based charcoal particles (Figure 6A), our results suggest an abrupt development of grasses dominated and fire disturbed landscape and confirm the alteration of the rain forest. High-resolution paleoecological analyses during that critical transition (see Figure 6C) reveal that grassland developed within few decades and occurs simultaneously with increases of fires. The co-occurrence between the first peak of macro-charcoal particles and the peak of magnetic susceptibility indicate that fires were local to the catchment and led to the observed forest alteration and initiated the increase of soil erosion. Similar observations were made at Laguna Potrok Aike, Patagonia where increased frequency dependent magnetic susceptibility and increased charcoal accumulation rates indicated the destruction of the vegetation cover by fire leading to enhanced soil erosion (Haberzettl et al., 2006).

Vegetation changes are also evidenced by the considerable development of saprophilous/coprophilous fungal spores (Figure 5B), that generally develop on decaying plant remains (e.g., litter) and/or herbivore livestock dung (e.g., Krug et al., 2004; Shumilovskikh L. et al., 2021; van Asperen et al., 2021). Plant parasitic fungi such as Tetraploa aristata, Spegazzinia sp., Dictyoarthrinium cf. sacchari indicate wide spread of herbaceous monocots and dicots as well while lignicolous fungi such as Xylariaceae decrease after 1100 cal BP (Shumilovskikh L. et al., 2021). Therefore, this complete rearrangement of lignicolous and plant parasites fungal spores confirms an important forest dieback at the expense of grassland. In addition to vegetation changes, enhanced soil erosion is also indicated by Glomus-type (Figure 5B), representing chlamydospores of the arbuscular mycorrhizal fungi growing symbiotically on plant roots (van Geel et al., 1989).

Sediment physical and organic characteristics are altered at this transition and SAR is increasing by almost one order of magnitude, supporting the pronounced decrease in tree cover associated with increased vulnerability of the soils to erosion. This is also visible in the higher frequency of ERDs occurring within the deposits (Figure 2). Following a short-term coarsening of the particles between 1300 and 1200 cal BP, the fining of the minerogenic particles accompanied by slightly better sorting and a more positive skewness of the distributions is observed at ca. 1150 cal BP (Figure 3). These changes suggest a selective deposition based on alteration of transport processes, i.e., lower amount of transport energy or shielding (McLaren, 1981; McLaren and Bowles, 1985). In this context, the development of Potamogeton from 1150 cal BP onward as an emergent aquatic plant, which likely covered the shallow (near shore/littoral) water areas, might have acted as sediment trap for larger particles. A lake level increase could have provided shallow water areas for macrophytes growth, since this would inundate the northwestern area of the catchment leading to the formation of a larger shallow water area, with emerging aquatic vegetation acting consequently as source of aquatic biomass and as sink for coarse minerogenic particles. A comparable process has been observed recently using aerial imagery (Supplementary Figure 4). The lake level increase might also be associated with enhanced nutrient supply, which, in combination, most likely explain the rise of aquatic vegetation and algae which is also supported by a distinct shift to rather low C/N-ratios (Meyers and Ishiwatari, 1993). The exceptional enrichment in δ¹⁵N during this phase might support the enhanced nutrient supply through the input of isotopically enriched N with either topsoil (Ochiai et al., 2015), or distinctive input from anthropogenic/livestock sources (sewage) (Enters et al., 2010). Alternatively, denitrification of the system due to tenacious anoxic conditions at the lake floor might be an explanation for the high δ¹⁵N (Enters et al., 2010).

This increase of soil erosion impacted lake ecosystem dynamics by increasing nutrient discharges which probably contributed to the development of the aquatic plant Potamogeton followed by the development of the green algae Botryococcus. A change in lake ecosystem dynamic is further supported by the diatom record characterized by minimum concentration values between 1300 and 1040 cal BP (concentration minimum peaking at 1160 cal BP) and minimum valve preservation at ca. 1100 cal BP (Figure 7). Low concentration values are likely caused by enhanced catchment erosion and input of minerogenic material into the lake with associated dilution of autochthonous material. Additionally, the low valve preservation is suggested to be linked to both, mechanical breakage in a high dynamic (energy) environment, and potentially to an increase in pH (Flower and Ryves, 2009). A distinct shift in the dominating planktonic diatom species from Discostella cf. mascarenica to Aulacoseira granulata agg. further proofs a change in the lake ecosystem following the transitional phase. Discostella mascarenica is a rare species and only little is known on its distribution and ecological preferences. It was described from a small, alkaline, eutrophic freshwater pond on Réunion island, situated 800 km to the east of Madagascar (Klee et al., 2000). Aulacoseira granulata occurs worldwide in a wide range of trophic conditions but is often associated with more eutrophic waters. Moreover, it is frequent under turbulent conditions, capable to survive in higher turbid waters, has high growth demands for silica and has also been connected to phases of increased erosion (Kilham et al., 1986; Costa-Böddeker et al., 2012; Bicudo et al., 2016; Nardelli et al., 2016). Shifts among Discostella and Aulacoseira have been associated with physical restructuring of the water column, whereby small cyclotelloid species like Discostella benefit from a strongly stratified water column and larger-celled Aulacoseira taxa are favored by strong vertical mixing (Rühland et al., 2015). Thus, the change from Discotella cf. mascarenica to Aulacoseria granulata agg. in Lake Amaprihibe supports the assumption of a catchment alteration leading to increased turbulence and turbidity.

The distinctive decrease in leaf-wax n-alkane amount, the shift to shorter chain n-alkane predominance and the marked increase in Paq further supports the emergence of aquatic vegetation (Ficken et al., 2000; Aichner et al., 2010; Castañeda and Schouten, 2011) as the result of a change in nutrient supply to the lake. Decreasing ACL values document the input of shorter homologs. In order to evaluate and reduce a potential bias of aquatic and emergent plants (likely also producing C₂₇ and C₂₉ to some degree), the modified ACL′ and ACL″ is applied. The increase at 1300 cal BP in these modified indices point to enhanced input of long-chain C₃₃ likely originating from grasses. This would also support the increase in δ¹³C of the bulk organic fraction, which suggest a shift to higher amounts of C4 vegetation.

After 1000-900 cal BP, a slow and progressive forest recovery is evidenced by slight decrease of Poaceae pollen percentages and slight reduction of fire activity, while pioneer tree taxa (e.g., Macaranga/Mallotus) tend to increase (Figures 5A, 6). The coeval pronounced increase in biogenic silica (BiSi) until 820 cal BP (Figure 3), coinciding with high diatom concentrations, especially of the heavily silicified species Aulacoseira granulata, however indicates an enhanced nutrient input (Ferris and Lehman, 2007). The further lowering in C/N as well as the high Paq afterward also suggest a predominance of aquatic (macrophyte and algae) biomass production (Meyers, 1994, 2003; Figures 3, 4). Forest recovery and the decrease in fire occurrences continues after 600 cal BP, evidenced by still decreasing Poaceae pollen percentages, moderate high charcoal concentration and a slight depletion trend in the δ¹³C record toward C₃ vegetation endmember. However, pollen assemblages reveal a different tree composition than in rain forest prior to 1300 cal BP. Pioneer tree taxa characterized by fast recovery after disturbance (Trema, Macaranga/Mallotus) are more abundant, while several tree taxa do not recover (Podocarpus, Weinmannia, Ilex, Meliaceae/Sapotaceae or Schefflera) and new ones appear (Myrtaceae and Mimosa). This new tree assemblage points to the development toward modern vegetation conditions which are characterized by a secondary complex vegetation including rain forest, wooded savanna and grasslands with crops mosaics (Mayaux et al., 2000). The result of this substantial change of the landscape is also evidenced by a stable intermediate SAR and the deposition of constantly small minerogenic particles with a sorting and a distribution skewness on a stable but very different level, compared to the initial phase (unit I, Figures 2, 3). Magnetic susceptibility on a quite stable intermediate level suggests a constant input of soil material, which is likely originating from farming activities in the catchment of the lake.

The previous study conducted on Lake Amparihibe by Burney et al. (2003) sets the date for the first occurrence of livestock and initial human impact to 1000 ⁺¹⁸⁰/_–90 cal BP (calibrated age from original publication), by moderating the occurrence of Sporormiella and an increase in sedimentary macro-charcoal particles. Our results suggest a major transformation of rain forest vegetation, to an open, fire disturbed, landscape, but starting somewhat earlier at 1306 ± 70 cal BP (Figure 8A). The rapid development of grasses (within several decades), coupled with peaks of macro-charcoal particles and saprotrophs/coprophilous fungal spores shows the first detectable human impact in northwestern Madagascar. These results illustrate the development of agriculture likely related to fire for pasture. Human impact has altered the environment and led to a major impact on the sediment accumulation increasing by one order of magnitude (0.13 to 1.3 mm a^–1) within less than 100 years. The lake system reacted with changes in the trophic status on this sudden release of nutrients transferred with the eroded soils.

Surprisingly, we were not able to confirm the predicted increase of Sporormiella in our record. After 1300 cal BP, a spore of Sporormiella has been counted only once on the top core while Burney et al. (2003) show that Sporormiella appear at the same time that macro-charcoal particles increase. Raper and Bush (2009) demonstrated that Sporormiella concentration declines from the shoreline across the first 100 m of a lake surface, highlighting the importance of the lake size and its changes in the interpretation of paleoecological records of Sporormiella abundance. Although both coring locations are in the central part of the lake, slight differences of distances from the shoreline might be an explanation since Sporormiella occur in low numbers (<2%) in Burney et al. (2003). However, evaluation of this effect in Nosy Be requires more research and other factors might also contribute to the different results (e.g., chemical treatments, spore identifications). Despite this difference, the entire assemblage of the saprotrophic/coprophilous fungi in our record strongly suggests presence of grazing around the lake (Shumilovskikh and van Geel, 2020; van Asperen et al., 2021), confirming previous results (Burney et al., 2003). By reaching the same conclusion, but with some difference in the results, we demonstrate that this is crucial to not only focus on one taxon, but to focus on the whole assemblages of fungal spores.

Comparison with archeological data reported by Douglass et al. (2019) reveals that this early human impact documented by our results on Nosy Be fits well with the major increase of radiocarbon dates in Madagascar and the maximum pace of megafauna extinction (Figure 8C). Demography history based on genomic data also revealed the same pattern with a major population expansion starting between 1250 and 750 cal BP (Figure 8C; Pierron et al., 2017). On Nosy Be, archeological surveys have shown first villages and hamlets on the western coast and overlooking Lake Amparihibe and Lake Anjavibe (Dewar and Wright, 1993; Wright and Radimilahy, 2005). Settlements from the coast had evidence of fishing activities while the interior settlements must have concerned solely with swidden cultivation of rice and the herding of cattle, goats and sheep (Wright and Radimilahy, 2005). Consequently, the establishment of such village or hamlet at 1300 cal BP, might be the best fit to explain our findings. Unfortunately, from these archeological surveys, no archeological remains (fish, turtle bones or mammal bones, carbonized coconut shells and local ceramics) have been dated on Nosy Be. However, ceramics are indistinguishable from those found during in Mahilaka, which corresponds to the oldest town described on Madagascar (Dewar and Wright, 1993; Radimilahy, 1997; Figure 1A). Mahilaka was an Islamic port located at 40 km south of Nosy Be in Ampasindava Bay and it formed, together with the Comoro islands, an important hub in the Indian Ocean trade network with several thousand inhabitants (Radimilahy, 1997; Crowther et al., 2016). Study of Crowther et al. (2016) dating rice remains in the area revealed that Comoro islands and Mahilaka were occupied at least from ca. 1250 cal BP (8th century) and from ca. 1000 cal BP (10th century), respectively. In this context, significant human impact at 1300 cal BP on Nosy Be island off the mouth off Ampasindava Bay certainly marks initial settlement of these populations in the region. After ca. 600 cal BP, human occupation declines and Mahilaka and surrounding villages in Ampasindava Bay including Nosy Be were abandoned (Dewar and Wright, 1993; Radimilahy, 1997). The reasons for this abandonment remain unclear (Wright and Radimilahy, 2005). Although recovery of forest and decrease of fires evidenced by our study may reflect this population decline, the continuous occurrences of macro-charcoal particles show that Nosy Be has certainly not been entirely deserted.

To better understand the impact of these populations on natural ecosystems it is crucial to consider the potential contribution of natural factors such as climatic changes and provide robust comparisons with paleoclimate and paleoenvironmental records available in the region. Specifically, several speleothems from northwestern Madagascar help to address past variability of precipitation in this region (Figures 1A, 8B). These records show that the mid- to late Holocene was characterized by a general drying trend punctuated by decadal-to-multidecadal megadroughts or prominent droughts (Wang et al., 2019; Li et al., 2020). A common feature revealed by these speleothems during the past 3000 years is an important megadrought centered around 900 cal BP (Scroxton et al., 2017; Voarintsoa et al., 2017; Wang et al., 2019; Li et al., 2020). Human impact and associated main environmental changes observed on Nosy Be at ca. 1300 cal BP occur earlier than this megadrought (Figure 8). Consequently, this drought cannot be considered as the main trigger of the ecosystem shift during the past millennium. A similar pattern is also recorded by δ¹³C from the same speleothems showing the replacement of C₃ woodland habitat with C₄ grasslands preceding the megadrought. To explain this transition, Godfrey et al. (2019) proposed the “subsistence shift hypothesis” which suggests a shift from hunting/foraging to herding/farming between 1250 and 1050 cal BP. During this interval, settlement of new immigrant groups, concomitant expansion of human population on Madagascar combined with spread of pastoralism altered natural ecosystems and accelerated the pace of megafaunal decline on Madagascar (Godfrey et al., 2019; Li et al., 2020; Hixon et al., 2021b). Our data support this assumption on Nosy Be, that development of herding/farming activities from 1300 to 1000 cal BP fragmented the rain forest. At ca. 900 cal BP, our results indicate the largest ERD recorded during the past 3000 years and the maximum value of δ¹³C of the bulk organic fraction (Figure 2 and Supplementary Figure 1). This 8.7 cm thick layer might be related to a lake level decrease during the megadrought and it could explain the observed reworked material and the outlier radiocarbon date (Table 1). However, one must remain careful with the interpretation of this layer because humans were already settled in the vicinity of the lake and distinguishing the effect of human impact and climate change on ecosystems is challenging in this context. Additional study sites or precipitation-related proxy are therefore necessary to confirm this hypothesis. After ca. 500 cal BP, more humid conditions are evidenced by speleothems (Scroxton et al., 2017) and may have also contributed to the observed forest recovery on Nosy Be (Figure 5). However, except for the well constrained megadrought at 900 cal BP, speleothem records are partly showing contradicting precipitation trends, despite most of them originate from the same cave (Supplementary Figure 3 in Supplementary Material from Li et al., 2020), which makes them rather debatable.

In northwestern Madagascar, other study provided paleoecological data (Figure 8B). Lake Mitsinjo sediments (Figure 1A) recorded significant grassland development and an increase in charcoal particles occurred from ca. 1000-900 cal BP, that was preceded by a hiatus in sediment deposition (Matsumoto and Burney, 1994; Burney, 1999). In Benavony (Figure 1A), a marshy meadow, the first significant peak of macro-charcoal occurred even later at ca. 800 cal BP (Burney, 1999). Age uncertainties may be responsible for these discrepancies but small-scale spatial differences in the onset and intensity of human activities may also have played a role. Furthermore, influence of climatic changes cannot entirely be excluded, since vegetation changes in northwestern Madagascar occurred in the context of a multimillennial drying trend and several megadroughts occurred during the Holocene (Li et al., 2020). Resilience of every vegetation type in this region (dry deciduous forest, semi-deciduous forest, rain forest or montane vegetation) may have been altered differently to these climatic changes and to the increase of human activities. In this context, without additional and high-resolution paleoecological records it will further remain a matter of discussion to which extent megadroughts affected human impact on natural ecosystem alteration during the past millennium.