Wildfire history of the boreal forest of southwestern Yakutia (Siberia) over the last two millennia documented by a lake- sedimentary charcoal record

Wildfires, as a key disturbance in forest ecosystems, are shaping the world’s boreal landscapes. Changes in fire regimes are closely linked to a wide array of environmental factors, such as vegetation composition, climate change, and human activity. Arctic and boreal regions and, in particular, Siberian boreal forests are experiencing 15 rising air and ground temperatures with the subsequent degradation of permafrost soils, leading to shifts in tree cover and species composition. Compared to the boreal zones of North America or Europe, little is known about how such environmental changes might influence long-term fire regimes in Russia. The larch-dominated eastern Siberian deciduous boreal forests differ markedly from the composition of other boreal forests, yet data about past fire regimes remain sparse. Here, we present a high-resolution macroscopic charcoal record from lacustrine 20 sediments of Lake Khamra (SW Yakutia, Siberia) spanning the last c. 2200 years, including information about charcoal particle sizes and morphotypes. Our results reveal a phase of increased charcoal accumulation between 600 and 900 CE, indicative of relatively high amounts of burnt biomass and high fire frequencies. This is followed by an almost 900-year-long period of low charcoal accumulation without significant peaks, likely corresponding to cooler climate conditions. After 1750 CE fire frequencies and the relative amount of biomass burnt start to 25 increase again, coinciding with a warming climate and increased anthropogenic land development after Russian colonisation. In the 20 century, total charcoal accumulation decreases again to very low levels, despite higher fire frequency, potentially reflecting a change in fire management strategies and/or a shift of the fire regime towards more frequent, but smaller fires. A similar pattern for different charcoal morphotypes and comparison to a pollen and non-pollen palynomorph record from the same sediment core indicate that broad-scale changes in vegetation 30

lists only a few continuously sampled macroscopic charcoal records across the Siberian boreal forest. Only recently 60 have charcoal records of sufficient temporal resolution allowed the assessment of fire return intervals in western Siberian evergreen forests and they reveal close fire-vegetation relationships (Barhoumi et al., 2019;Feurdean et al., 2020). More studies have been conducted in North America (e.g. Frégeau et al., 2015;Hély et al., 2010;Hoecker et al., 2020;Waito et al., 2018) and boreal Europe (e.g. Aakala et al., 2018;Feurdean et al., 2017;Molinari et al., 2020;Wallenius, 2011). However, comparisons across boreal study sites are complicated by the differing 65 predominant fire regimes in North America (high-intensity crown fires) and Eurasia (lower-intensity surface fires) (de Groot et al., 2013).
The main fire regimes in the European and western Siberian evergreen boreal forest also differ markedly from those of its larch-dominated, deciduous counterpart in eastern Siberia. Many prevalent evergreen conifers (Pinus sibirica, Picea obovata, Abies sibirica) are commonly seen as fire avoiders and are more susceptible to crown fires 70 (Dietze et al., 2020;Isaev et al., 2010;Rogers et al., 2015). The predominant eastern Siberian larches (Larix gmelinii, L. cajanderi, L. sibirica), on the other hand, can resist fires with an insulating bark protecting the cambium from heat, while their deciduous and self-pruning nature restricts fires from reaching the crown (Wirth, 2005). Moreover, larches are thought to benefit from occasional surface fires, leading to more saplings by clearing the lower vegetation layers of plant litter and mosses, although this might become a risk for young trees if fire 75 frequencies increase above a certain threshold (Sofronov and Volokitina, 2010). Surface fires in larch forest might also play a role in the long-term preservation of permafrost (Dietze et al., 2020;Herzschuh et al., 2016). These different fire strategies within the Siberian boreal forest reinforce the need for fire reconstructions towards the eastern part to evaluate changes in fire regimes depending on the prevalent tree species and to obtain a biome-specific overview of fire regimes throughout time. 80 The main goal of this study is to start filling a pronounced gap in the global distribution of macroscopic charcoal records by providing the first continuously sampled, high-resolution macroscopic charcoal record from eastern Siberia, using charcoal size classes and morphotypes. We specifically aim to answer the following research questions: (I) How did the fire regime in south-west Yakutia change throughout the last two millennia? (II) How might reconstructed fire history relate to common drivers behind changes in fire regimes (climate, vegetation, 85 humans)?
https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. of c. 5000 inhabitants. Traces of logging activity are visible on satellite imagery c. 10 km from the lake, while in its direct vicinity only winter forest tracks are kept open. We discovered traces of recent wildfire disturbance at vegetation plots around the lake, i.e. trees with fire scars on their trunk or burnt bark. These traces likely correspond to recent fires within the catchment in 2007 and 2014 as captured by the remotely sensed forest loss data of Hansen 120 et al. (2013) (Fig. 1b, c).

Fieldwork and subsampling
Fieldwork at Lake Khamra was conducted in August 2018. The 242-cm-long sediment core EN18232-3 was 130 obtained using a hammer-modified UWITEC gravity corer at the deepest part of the lake (22.3 m), based on water depth measurements using a surveying rope and a hand-held HONDEX PS-7 LCD digital sounder. From the same location, a parallel short sediment core (EN18232-2, 39 cm length) was retrieved and subsampled in the field at increments of 0.5 cm (upper 20 cm) and 1 cm (lower 19 cm) for lead-210 and caesium-137 (Pb/Cs) age dating.
The sediment core, stored in a plastic tube, and all sediment samples, kept in Whirl-Pak bags, were shipped to the 135 Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research (AWI) in Potsdam and placed in storage at 4°C. In October 2018, sediment core EN18232-3 was opened and cut in half in a cooled room under sterile conditions. While one half was archived, the other half was subsampled for proxy analysis and radiocarbon ( 14 C) dating. The sampling scheme included one 2 mL sample from the midpoint of every 2 cm increment (n = 120), used for the combined palynological analysis and charcoal extraction, and two c. 1 mL samples between 140 each pair of larger samples (n = 234), used specifically for charcoal extraction to ensure a continuous record of charcoal concentration. Bulk sediment samples for 14 C age dating were extracted every 20 cm (n = 12). At 85.5 cm core depth, a c. 2 cm long piece of wood was removed from the sediment core for 14 C dating. Due to a lack of any other larger organic structures, more macrofossil samples were picked while wet-sieving bulk sediment samples (n = 15). Except for one case, a clear determination of their origin was not possible because of the small 145 size of these samples.

Lithology and age dating
Sediment core EN18232-3 was visually described before sampling. Water content and bulk density were determined in subsamples from 120 sampling increments every 2 cm. 150 To establish a chronology, all bulk and macrofossil samples were sent to AWI Bremerhaven for 14 C age dating at the MICADAS (Mini Carbon Dating System) laboratory. Subsamples of the parallel short core EN18232-2 (n = 19) were sent to the University of Liverpool Environmental Radioactivity Laboratory for Pb/Cs age dating, analysing 210 Pb, 226 Ra, 137 Cs, and 241 Am by direct gamma assay with Ortec HPGe GWL series well-type coaxial low background intrinsic germanium detectors (Appleby et al., 1986). After careful evaluation of age dating 155 results, an age-depth model was computed using Bacon v.2.4.3 (Blaauw and Christen, 2011;package "rbacon";https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. Blaauw et al., 2020), combining Pb/Cs and adjusted 14 C ages of all bulk samples calibrated with the IntCal20 14 C calibration curve (Reimer et al., 2020).

Macroscopic charcoal
We developed a sample preparation protocol that allows for the extraction of both macroscopic charcoal and the 160 smaller pollen fraction including non-pollen palynomorphs (NPP) from the same sediment sample. Lycopodium tablets (Department of Geology, Lund University), used as marker grains in the palynological analysis, were dissolved in 10% HCl and added to the sediment samples. These were subsequently wet-sieved at 150 µm mesh width for separation of the macroscopic charcoal from smaller fractions (e.g. Conedera et al., 2009;Dietze et al., 2019;Hawthorne et al., 2018). The suspension with the <150 µm-fraction was collected in a bowl below the sieve. 165 This "pollen subsample" was then iteratively added to a falcon tube, centrifuged, and decanted prior to further preparation for palynological analysis. The >150 µm fraction in the sieve was rinsed together under a gentle stream of tap water before being transferred into another falcon tube. This macroscopic "charcoal subsample" was then left to soak overnight in c. 15 mL of bleach (<5% NaClO) to minimise the potential counting error from darker, non-charcoal organic particles (Halsall et al., 2018;Hawthorne et al., 2018). 170 Counting of 304 charcoal samples was done under a reflected-light stereomicroscope at 10-40x magnification. All particles that appeared opaque and mostly jet-black with charred structures were counted in every given sample (see Brunelle and Anderson, 2003;Hawthorne et al., 2018). In addition, counted particles were grouped into three size classes (150-300 μm, 300-500 μm, and >500 μm measured along a particle's longest axis; Dietze et al., 2019) and after similarities in shape (charcoal morphotypes; Enache and Cumming, 2007). For size reference, preparatory 175 needles with known diameters of 300 and 500 μm were used that could be placed next to a charcoal particle. These needles also allowed the evaluation of the flexibility of particles of uncertain origin, since charcoal fragments are described as fragile and non-bendable (Whitlock and Larsen, 2001).
Grouping of particles after their shape was based on the morphotype classification scheme by Enache and Cumming (2007) and extended by three additional types to represent the variety of charcoal particles found at the 180 study site. The original scheme differentiates between irregular (types M, P), angular (types S, B, C), and elongated (types D, F) shapes and further divides those depending on whether they show a visible structure or ramifications.
The three added types appear as highly irregular particles (type X), elongated, fibrous particles (type E), and slightly charred, partially transparent particles (type Z), the latter of which are not included in the total charcoal sum. For correlations and visualisations, the morphotypes were grouped in their respective main categories 185 (irregular, angular, or elongated). Within the topmost c. 50 cm of the sediment core, relative morphotype https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. distributions were retrospectively derived from counts of 67 subsamples. At that time, 11 samples had already been used for other purposes and thus lack information on morphotype classification.
Eight randomly selected samples were counted a second time to obtain an estimate of counting uncertainty and ambiguity in charcoal identification. 190

Palynological samples
Established protocols following Andreev et al. (2012) were applied to the "pollen subsample" (n = 35). Of these samples, 11 were chosen specifically from intervals with high charcoal concentrations, whereas the others were spread across the sediment core. Samples were treated with boiling potassium hydroxide for 10 min, sieved and left to soak in 18 mL hydrofluoric acid (40% HF) overnight. After two additional treatments with hot HF (1.5 h 195 each), acetolysis was performed using acetic acid and in a second step a mixture of acetic anhydrite and sulfuric acid. After being fine-sieved in an ultrasonic bath, the samples were suspended in glycerol.
Pollen and NPPs were counted together with the added Lycopodium spores on pollen slides under a transmitting light microscope, with a minimum count sum of 300 particles. For subsequent analyses, pollen and spore amounts relative to the respective total counts were used. 200

Statistical methods
Statistical analysis was carried out in R v.4.0.2 (R Core Team, 2020). To asses fire history, two different approaches were applied that decompose the charcoal record into a background component, representing longterm variations in charcoal accumulation and particle taphonomy, and a peak component, representing predominantly local charcoal accumulation during fire episodes (Higuera et al., 2007(Higuera et al., , 2009Kelly et al., 2011). 205 First, we used the well-established "CharAnalysis" method (Higuera et al., 2009; R script by Dietze et al., 2019), referred to as "classic CHAR". We interpolated the charcoal record to equally spaced time intervals according to its median resolution and calculated the charcoal accumulation rate (CHAR, particles cm -² yr -1 ; package "paleofire" v.1.2.4, function "pretreatment"; Blarquez et al., 2014). A background component was determined by computing a locally estimated scatterplot smoothing (LOESS) at a window width of 25% of the total record length (package 210 "locfit" v.1.5-9.4, function "locfit"; Loader et al., 2020). This window width was found to result in an efficient distribution of a signal-to-noise index (SNI) >3 after Kelly et al. (2011), which indicates a high degree of separation between signal and noise (Barhoumi et al., 2019;Kelly et al., 2011). A peak component was created by subtracting the background component from the timeseries. By fitting two Gaussian distributions into the histogram of the peak component (package "mixtools" v.1.2.0, function "normalmixEM"; Benaglia et al., 2009), a global threshold 215 was defined at the 99th percentile of the noise distribution (Whitlock and Anderson, 2003). All peak component values exceeding the threshold were subsequently identified as signal (representing fire episodes) and marked when they overlapped with periods of SNI >3. Longer window widths in the statistical analysis of classic CHAR would have resulted in more fire episodes (up to 100%) above the proposed cutoff-value of SNI = 3 (Kelly et al., 2011), but also in a strong averaging of the record. With all of the few fire episodes at SNI <3 lying very close to 220 the cut-off value, this was seen as a reason to include them in the estimation of fire return intervals (FRIs).
FRIs were determined as the temporal difference between subsequent fire episodes. An illustration of fire frequency was derived by counting all identified fire episodes within a moving window spanning 200 yrs before applying a LOESS of the same window width to provide a clearer visualisation.
To account for accumulated uncertainties from both the chronology and the counting procedure, we used a Monte 225 Carlo (MC) based approach (for detailed description and R script see Dietze et al., 2019), referred to as "robust CHAR". In short, it describes both the age and proxy values of each sample as Gaussian distributions, creating a pool from which random values are sampled. As inputs we used the 2σ range of Pb/Cs ages and 1σ range of calibrated and adjusted 14 C ages to scale the general magnitudes of uncertainty between the two dating methods to comparable dimensions. For proxy uncertainty, the average deviation between the repeatedly counted samples was 230 used (c. 20%). Five thousand MC runs were performed and output data resolution set at three times the record's median temporal resolution (18 yrs). Robust CHAR was then divided into background and peak components, similarly to classic CHAR, but by computing 1000 randomly sampled LOESS fits at varying window widths ranging from 5-25% of the record length, thereby not relying on an individual user-input value (Blarquez et al., 2013). 235 The statistical approach outlined above (classic and robust CHAR including the determination of SNI, FRIs, and fire frequency) was also applied to the charcoal size classes and morphotype categories, respectively (see Supplement).
A principal component analysis (PCA; package "stats"; R Core Team, 2020) was used to assess relationships among the various centred log-ratio (clr) transformed (package "compositions"; van den Boogaart et al., 2020) 240 relative distributions of charcoal particle classes. To evaluate potential associations between charcoal accumulation and vegetation we applied correlation tests using Kendall's τ (package "psych"; Revelle, 2020) to clr-transformed relative distributions of pollen types and charcoal classes following Dietze et al. (2020).

Lithological sediment properties and chronology 245
Sediment core EN18232-3 shows no lamination or visible changes in its brown colour or the texture of the sediment matrix. Its homogenous appearance is underlined by both uniform mean dry bulk density (184 ± 4 mg cm -³, mean ± 1σ) and water content (83 ± 4%).
Only 9 of the 15 macrofossil samples were large enough in size to be used for 14 C dating, although mostly only slightly above the minimum amount of carbon (Table 1a). As Pb/Cs dating from the parallel short core EN18232-250 2 reveals no disturbances in its uniform sedimentation rate, the ages are expected to be applicable to the main core EN18232-3 from the same location within the lake and thus can be compared to the respective 14 C ages. Noticeably, the topmost 14 C bulk sample dates to 1415 ± 27 14 C yrs BP (before present, i.e. before 1950 CE), whereas the Pb/Cs method from the parallel core confirms the expected recent surface age (Table 1b). This 14 C age offset can have a variety of causes, such as sediment mixing processes (Biskaborn et al., 2012), the presence of old organic 255 carbon (Colman et al., 1996;Vyse et al., 2020), or dissolved carbonate rock (hard-water effect; Keaveney and Reimer, 2012;Philippsen, 2013).
In the case of Lake Khamra, located in a zone of discontinuous permafrost and bedrock containing early Palaeozoic carbonates, the observed 14 C age offset is a likely consequence of input of both old organic and inorganic carbon through the south-western inflow stream. The magnitude of this offset is documented by the difference between 260 the 14 C bulk surface sample and the corresponding Pb/Cs age, which shows a recent surface age and is not affected by "old carbon" input. Furthermore, multiple indicators support the assumption that accumulation of old carbon not only happened recently, but rather constitutes an ongoing process in this lake system: (I) Treating the surface 14 C age as an outlier without also adjusting the other 14 C dates would necessarily lead to a strong shift in sedimentation rates, which is neither reflected by the homogenous appearance and density, nor the uniform 265 sedimentation rate as implied by the Pb/Cs method. (II) Macrofossil 14 C ages provide direct evidence for the influence of old carbon on bulk samples at various depths (e.g. macrofossil age of 9902 ± 97 14 C yrs BP in a sediment matrix that dates back to only c. 100 yrs BP according to the parallel core's unaffected Pb/Cs method at depth 21-22 cm).
For these reasons, the documented age offset (mean ± 1σ) in the topmost 14 C sample was used to adjust all other 270 14 C bulk samples (Colman et al., 1996;Vyse et al., 2020). Although macrofossil samples are usually thought to be superior in precision to bulk sediment for age dating purposes (Hajdas et al., 1995;Wohlfarth et al., 1998), within https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. the present environment we cannot exclude a potential permafrost origin as well as measurement uncertainties due to their small size. For this reason, they were not used for constructing the chronology.
The age-depth model created according to these observations (Fig. 2) shows a smooth transition from Pb/Cs dates 275 towards adjusted bulk 14 C ages. Its uniform sedimentation rate mirrors the sediment core's homogenous composition and supports the underlying assumption of a rather constant magnitude of old carbon influence on bulk 14 C ages. Based on this chronology, the sediment core spans c. 2350 years across its 242 cm length. The continuously sampled charcoal record spans c. 2160 years and thus reaches back from 2018 CE until c. 140 BCE.

Reconstructed fire regime
The median temporal resolution of the charcoal record is 6 yrs and its samples contain 8.1 ± 5.1 (mean ± 1σ) charcoal particles per cm³ of sediment (min: 0, max: 38.8). Interpreting the classic peak component as occurrences 295 of fires, 50 fire episodes within the continuously sampled core segment, spanning the last c. 2200 yrs, were identified. This results in a record-wide mean FRI of 43 yrs (min: 6, max: 594).
Classic and robust CHAR analysis distinguished four distinct phases, representing different states of the fire regime (see Fig. 3). Phase 1 (c. 200 BCE to 600 CE) is characterised by relatively high CHAR of 0.83 ± 0.43 particles cm -² yr -1 and numerous (n = 23) peaks (i.e. fire episodes, Fig. 3c), with a SNI that is slightly above 3 for the most 300 part but slowly decreasing towards the following phase 2 (Fig. 3b). The mean FRI is 31 yrs (min: 6, max: 144).
Robust CHAR shows a steadily increasing background component (Fig. 3d), whereas both its peak component https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. (Fig. 3e) and sum (Fig. 3f) remain at low levels. More fire episodes (n = 16) occur in the following, shorter phase 2 (600-900 CE), with a lower mean FRI of c. 14 yrs. It incorporates some of the highest peaks of the whole record with both the classic and robust approaches, resulting in a high CHAR of 1.3 ± 0.54 particles cm -² yr -1 (including 305 the maximum of 2.8 particles cm -² yr -1 ) while SNI falls below 3. In the transition to phase 3 at around 900 CE a pronounced decrease in charcoal input leads up to a long period of few to no fire episodes (n = 6) and longer mean FRI of >60 yrs (min: 6, max: 594). Hence, CHAR of phase 3 (900-1750 CE) is comparably low with 0.53 ± 0.29 particles cm -² yr -1 , while robust CHAR background remains below average. The SNI decreases to a record-wide minimum due to the lack of any CHAR peaks above the global threshold in the second half of phase 3. Finally, 310 phase 4 (1750-2018 CE) has higher CHAR (0.61 ± 0.47 particles cm -² yr -1 ) and more frequent occurrences of fire episodes (n = 5), with mean FRI sharply decreasing to 40 yrs (min: 6, max: 78). An outstanding peak around 1880 CE leads to a maximum SNI >6 during this phase. Although phase 4 sees increasing CHAR and fire frequency after the low CHAR of phase 3, charcoal input decreases to a minimum within the last century (mean CHAR of the last 100 yrs before core extraction = 0.56 ± 0.29 particles cm -² yr -1 ). In contrast, the robust CHAR background 315 and sum show increases within phase 4 (Fig. 3d, e), and the robust peak component shows two maxima around the early 1800s and 1950 CE (Fig. 3e). In general, the older half of the record (c. 200 BCE to 1000 CE) has higher mean CHAR and a higher variability (0.97 ± 0.5 particles cm -² yr -1 ) compared to the younger half (1000-2018 CE; CHAR = 0.51 ± 0.32 particles cm -² yr -1 ). Even with added uncertainties from counting and the chronology, maxima of robust CHAR mostly replicate periods of increased classic CHAR. 320 Over the entire record, 43.7% of charcoal particles belong to the smallest size class (150-300 µm), while 28.8% and 27.5% are part of the medium (300-500 µm) and large size classes (>500 µm), respectively. When assessed individually, more fire episodes are identified for smaller particles than for larger particles ( Table 2).
The most prevalent morphotypes present in the sediment are types F (elongated, 31.7%), M (irregular, 28.4%), S (angular, 20.6%), and B (angular, 7.2%), with all others (X, C, D, E, P) ranking at or below 3% each. Noticeably, 325 two of the highest peaks of the record at c. 650 and 1880 CE are composed primarily of large (>500 µm), elongated (type F) particles. The total relative amount of type F particles seems to correlate with the largest particle size class, whereas type M is more closely associated with smaller particles (see Fig. 4 and Appendix B). Furthermore, the PCA indicates that there are rather weak grouping patterns of morphotype or size-class distributions in samples of increasing core depth. The charcoal morphotypes show a similar temporal pattern for their background and peak 330 components (Fig. 5c, d), mostly mirroring the decreased variability in the second half of the record as described for the sum of all particles. https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License.

Fire regime history of the last two millennia at Lake Khamra
The macroscopic charcoal record at Lake Khamra (Fig. 3) reveals gradually increasing, but relatively stable fire activity from 200 BCE to 600 CE (phase 1). A period of high fire activity takes place between 600-900 CE, 360 expressed as higher CHAR and shorter FRIs (phase 2). It then transitions into an almost 600-year period without any identified fire episodes and low CHAR (phase 3). From around 1750 CE the modern fire regime begins to take shape (phase 4), with regularly identified fire episodes marking increasing fire frequency. However, the most recent levels of CHAR are still lower than those of the maximum in phase 2 and reach a minimum in the 20th century, meaning that the amount of modern charcoal accumulation is not unprecedented within the last c. 2200 365 yrs.
Robust CHAR (Fig. 3d-f), incorporating uncertainties from the age-depth model and charcoal counting, necessarily loses the original charcoal record's short-term variability. It needs to be noted that any uncertainty potentially arising from the assumed constant rate of old carbon input to the lake, underlying the sediment core's chronology, is not included here, as any changes in magnitude of this reservoir-like effect are impossible to 370 quantify. This issue is common in studies in permafrost regions that use 14 C age dating (Biskaborn et al., 2012;Colman et al., 1996;Nazarova et al., 2013;Vyse et al., 2020). In such instances, applying Pb/Cs age dating adds valuable non-carbon-related estimates of sediment accumulation rates for the upper part of a sediment core (Whitlock and Larsen, 2001).
The general trend of fire regime changes in SW Yakutia over the last c. 2200 yrs captured by the CHAR 375 background component (Fig. 5c) and described above, shares many similarities with a charcoal record from the evergreen forest in the Tomsk region (c. 1500 km west of Lake Khamra; Feurdean et al., 2020). There, one period of exceptionally high CHAR was observed around 700 CE and then again starting around 1700 CE towards the present, in parallel with Lake Khamra's fire history. The onset of increased biomass burning around 1700 CE was also reconstructed using aromatic acids from an ice core on the Severnaya Zemlya archipelago (c. 2200 km north-380 west of Lake Khamra), and it indicates a sudden decrease at the beginning of the 20th century (Grieman et al., 2017). However, the same study finds another maximum of biomass burning around 1500 CE, in contrast to a clear period without fire episodes between c. 1250-1750 CE at Lake Khamra. Although comparisons across such long distances and between different climate zones, archives, and fire proxies are likely to show differing results, some of the described trends seem to be recorded at several sites worldwide. A global charcoal record synthesis 385 by Marlon et al. (2013) indicates decreasing biomass burning from c. 0 CE towards the industrial era, where, after https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. a maximum around 1850 CE, it decreases with the onset of the 20th century. Charcoal records from Siberia are underrepresented in such synthesis studies (Marlon et al., 2016), which, together with a lack of comparable records closer to SW Yakutia, underline the issue of sparse data in the region.
The reconstructed record-wide mean FRI of 43 yrs, incorporating both the exceptionally short FRI of phase 2 (14 390 yrs) and the long FRI of phase 3 (60 yrs, excluding the c. 600-yr-long period without identified fires), lies within the range of the few comparable studies in western Siberia. Barhoumi et al. (2019) found the shortest FRIs of the Holocene ranging from 40-100 yrs between 1500 CE and present day in macroscopic charcoal records from the northern Ural region. A mean FRI of 45 yrs during recent centuries was inferred by Feurdean et al. (2020). Other studies using tree-stand ages and fire scars in tree-ring chronologies suggest mean FRIs of 80-90 yrs for mixed 395 larch forests between the Yenisei and Tunguska rivers c. 1000 km north-west of Lake Khamra since c. 1800 CE, although the FRI of individual study sites could be as short as c. 50 yrs (Kharuk et al., 2008;Sofronov et al., 1998;Vaganov and Arbatskaya, 1996). A mean FRI of 52 yrs was reported for the northern Irkutsk region c. 300 km west of Lake Khamra in the 18th century  and 50-80 yrs for some sites in the north-eastern larch-dominated forests (Kharuk et al., 2011;Schepaschenko et al., 2008). In general, FRIs increase with latitude 400 due to lower incoming solar radiation, shorter fire seasons, and lower flammability of moist biomass (Kharuk, 2016;Kharuk et al., 2011), which likely contributes to a relatively short mean FRI at Lake Khamra. Additionally, studies of tree-ring chronologies or stand ages usually convey direct fire impact and are therefore more locally constrained, whereas a charcoal record incorporates fires from a larger source area (Remy et al., 2018).
Smaller charcoal particles record shorter FRIs than larger particles when analysing size classes individually (see 405   Table 2 and Supplement), which could be because smaller particles have a larger source area, thus incorporating more fire events into the signal. However, the wide-spread assertion that larger charcoal particles generally originate from fires within a few hundred metres of the lake archive (Clark et al., 1998;Higuera et al., 2007;Ohlson and Tryterud, 2000) has been challenged (Peters and Higuera, 2007;Pisaric, 2002;Tinner et al., 2006;Woodward and Haines, 2020). As wildfires in the Siberian boreal forest are predominantly considered low-410 intensity surface fires (de Groot et al., 2013), the potential of the resulting convection to transport large charcoal particles is probably limited compared to high-intensity crown fires. We therefore assume a charcoal source area of a few hundred metres around the lake for low-intensity fires (Conedera et al., 2009) and up to several kilometres for more intense fires producing stronger convection.
At Lake Khamra with a catchment to lake-area ratio of 23:1, the charcoal record predominantly consists of fragile 415 morphotypes (types F, M, and S alone make up >80%). Enache and Cumming (2007) explain how a large catchment to lake-area ratio might favour secondary deposition of compact morphotypes, while fragile https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. morphotypes are more prone to fragmentation during surface runoff and thus rather represent primary input through the air. However, the present large catchment to lake-area ratio at Lake Khamra might indicate that morphotype distribution is controlled by the type of biomass burning rather than secondary charcoal transport. 420 This is also implied by the gentle and densely vegetated slopes that do not favour surface erosion.
Experimental charring studies have shown how different types of vegetation produce varied charcoal appearances (Jensen et al., 2007;Mustaphi and Pisaric, 2014;Pereboom et al., 2020). Pereboom et al. (2020) found elongated charcoal particles after experimentally burning tundra graminoids, but this result is likely not directly applicable to a study site in the boreal forest. Noticeably, type F particles at Lake Khamra closely match the appearance of 425 charred Picea needles reported by Mustaphi and Pisaric (2014). Together with the potential of more intense fires producing larger charcoal particles (Ward and Hardy, 1991), this could mean the two previously noted peaks of CHAR (c. 1880 CE at 19.5-20.5 cm depth and 650 CE at 124.5-125 cm; both dominated by high shares of type F particles >500 µm) are evidence of higher-intensity fires burning conifer trees more severely and within few kilometres from the lake shore. Charring experiments with local vegetation and a regionally adapted morphotype 430 classification scheme would potentially benefit future studies by providing clear ground-truthing for links between morphotypes and vegetation.

Vegetation
The overall stable vegetation composition during the time covered by the charcoal record, as implied by the pollen 435 and NPP record (see Appendix A), indicates that vegetation changes were unlikely to be the main driver behind changes in the fire regime, and/or that changes in fire regime did not lead to large-scale shifts in vegetation composition. Similarly, no prominent shift in charcoal morphotype composition, and hence in the type of biomass burned over time, can be inferred (Fig. 5c, d). Although some studies draw a similar conclusion (e.g. Carcaillet et al., 2001 in eastern Canada), this result contrasts with many other studies from the Eurasian and North American 440 boreal zones, where vegetation changes were found to be closely connected to changes in fire regimes (Barhoumi et al., 2019(Barhoumi et al., , 2020Feurdean et al., 2020;Gavin et al., 2007;Kelly et al., 2013). However, on a shorter, multidecadal timescale, phases with more Cyperaceae pollen (sedges) in the Lake Khamra record and a higher ratio of evergreen to deciduous arboreal pollen types coincide with periods of high fire activity in phases 2 and 4 (see Fig.   5b). This could be due to either the ability of sedges to quickly settle on freshly disturbed and cleared out forest 445 areas, and/or sedges growing in wetter areas, possibly right at the lake shore, which are spared by fires (Angelstam and Kuuluvainen, 2004;Isaev et al., 2010;Ivanova et al., 2014). Increased numbers of evergreen trees might enable https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. more intense crown fires. Indirectly, dry periods could lead to receding lake levels and thus an increase in both shoreland sedges and fire ignitions. However, such clear links are difficult to infer without hydrological data. In addition to differences in pollen source area and taphonomy compared to that of macroscopic charcoal, a variety 450 of factors likely obscures traces of potential fire impacts: surface fires in the deciduous forests in central Yakutia mostly result in the elimination of only a share of a tree population depending on fire intensity (Matveev and Usoltzev, 1996), while herbs or shrubs may recover too quickly for changes to be detected in our record with a median temporal sampling resolution of 6 yrs and potential mixing processes and residence time of pollen grains before settling in the lake sediment (Campbell, 1999;Faegri et al., 1989). These factors, together with a remaining 455 ambiguity in morphotype classification, likely explain the rather weak correlations of pollen and charcoal records.

Climate
The low fire activity in the latter half of phase 3 (900-1750 CE)  Although it has been demonstrated that the timing and extent of these climatic phases are heterogeneous (Guiot et al., 2010), evidence for their occurrence in Siberia is seen in other proxy studies (Churakova Sidorova et al., 2020;Kharuk et al., 2010;Osborn and Briffa, 2006), albeit less pronounced when it comes to vegetation response in the West Siberian Lowland (Philben et al., 2014). Neukom et al. (2019) show how these climatic periods arising from 465 averaged reconstructions at many individual study sites are not spatially or temporally coherent on the global scale, and conclude that environmental reconstructions "should not be forced to fit into global narratives or epochs". This might be especially true for studies using chronologies that have 14 C reservoir effects. Due to a lack of regional studies, the PAGES Arctic 2k temperature reconstruction (McKay and Kaufman, 2014) was used to provide a comparison between the reconstructed fire activity and large-scale changes in Arctic climate north of 60°N (Fig.  470   5a). This synthesis of circumpolar temperature reconstructions incorporates many records, although data from Siberia are sparse and thus it is underrepresented compared to Greenland, North America, and Europe. However, climate at Lake Khamra is likely to be strongly influenced by the conditions further north, as Arctic temperatures affect the strength of the AO, which has been indirectly linked to fire activity further south (Balzter et al., 2005;Kim et al., 2020). The reconstructed PAGES Arctic 2k temperature provides evidence for a colder climate around 475 c. 1600 CE, coinciding with the LIA and low fire activity at Lake Khamra. The following onset of increased fire frequency in phase 4 is concurrent with a gradual increase in Arctic temperatures during the last two centuries https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. (Fig. 5a, c), although not exceeding the maximum in fire frequency of phase 2. The older half of the charcoal record (before c. 1000 CE) does not match the temperature reconstruction as clearly. This might be due to a more regionally constrained climate, or a consequence of the lack of comparable data about humidity, which also affects 480 fire activity (Brown and Giesecke, 2014;Power et al., 2008). A larch tree-ring based, c. 1500-yr-long reconstruction of summer vapour pressure deficit in north-eastern Yakutia (c. 2000 km from Lake Khamra) indicates that the modern, increasing level of drought stress is not yet unprecedented, being surpassed by a high vapour pressure deficit during the MCA (Churakova Sidorova et al., 2020). This is similar to trends observed in the fire reconstruction at Lake Khamra. Another possibility for a less clear relationship between Arctic temperature 485 and fire regime changes in the older half of the record is that the assumed constant impact of old carbon on the 14 C age dating might be less pronounced in this older millennium. Yet, an impact of colder Arctic temperatures on the reconstructed low fire activity in phase 2 at Lake Khamra seems likely and calls for more palaeoclimatic reconstructions and high-resolution charcoal records within the vicinity of the lake to better assess the role of climate on shifts in the regional fire regime. 490

Human activity
The discrepancy of last century's low CHAR just after a phase of increasing fire frequency and rising Arctic temperature could potentially be a sign of direct and indirect consequences of human activity around Lake Khamra.
The anthropogenic influence on fire regimes may be the most difficult to quantify due to missing information about the kind and extent of human fire use and management throughout time, and the complex disentanglement 495 of other drivers like climate and vegetation (Marlon et al., 2013). Although Lake Khamra is located in a sparsely populated region, humans have historically been shaping the surrounding landscapes by building forest winter tracks and roads (c. 0-30 km distance), logging (c. 10 km distance), and building villages and towns (ca. 30 and 40 km distance, respectively). Yakutia has been populated by humans since at least c. 28,000 BCE (Pitulko et al., 2004), although the population first noticeably increased at the end of the LGM around 17,000 BCE (Fiedel and 500 Kuzmin, 2007), eventually forming the indigenous hunting and reindeer herding tribes of Evens, Evenks, and Yukaghirs (Keyser et al., 2015;Pakendorf et al., 2006). Between 1100-1300 CE (in the first half of phase 3 in the charcoal record) the Sakha people moved in from the south, pressured by an expanding Mongol empire (Fedorova et al., 2013). They established a new and distinct form of semi-nomadic livelihood based on horse and cattle breeding (Pakendorf et al., 1999). Fire was mainly used at hearths to provide warmth and light, but also to sustain 505 grasslands for grazing (Kisilyakov, 2009;Pyne, 1996). Population density likely started to increase rapidly after Russian colonisation in the early 17th century (Crubézy et al., 2010), and even more drastically with the onset of https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License.
new industries like large-scale logging and mining in the 20th century (Pyne, 1996). When compared to the pastoralist societies that existed up to that point, anthropogenic influence on the fire regime likely increased at the end of phase 3 (c. 1700 CE onwards) and throughout phase 4, after colonisation and industrialisation (Drobyshev 510 et al., 2004). It has been shown that human livelihoods and the mentality towards fire use can often better explain shifts in fire regimes than population density alone (Bowman et al., 2011;Dietze et al., 2018). For example, a formerly smaller population could have relied on practices like slash-and-burn agriculture until there was a transition towards more industrialised, urban livelihoods and a new focus on active fire suppression to protect forestry resources despite an increasing population (Dietze et al., 2019;Marlon et al., 2013). This is also thought 515 to explain a pronounced decrease in boreal fire activity within the last century in tree ring studies from central Siberia (300 km west of Lake Khamra) and Fennoscandia (Wallenius, 2011;Wallenius et al., 2011). Forest roads and clearings could have acted as fire breaks, while the emergence of fire suppression in Russia was conceived as early as 1893 CE and later led to the founding of the first aerial firefighting unit in 1931 CE (Pyne, 1996). Adding to this, slash-and-burn agriculture was officially banned at the end of the 18th century, but likely still practiced 520 frequently up until the early 20th century (Drobyshev et al., 2004;Konakov, 1999;Kozubov and Taskaev, 1999).
Higher fire frequency after c. 1750 CE, marking the onset of phase 4, therefore coincides with both a rapid increase in anthropogenic activity and land development, as well as a warming Arctic climate. Low CHAR within the last century on the other hand might be a consequence of a cultural shift towards seeing fire as a hazard to ban, control, and suppress. Whereas high fire frequency in the past corresponded with high amounts of biomass burned, the 525 recent century also sees increasing fire frequencies with decreasing total biomass burned (Fig. 5c, d). This might indicate that the current fire regime differs from that experienced by indigenous and Sakha people a few hundred years ago, now potentially consisting of more frequent, but smaller fires. A better understanding of fire use and management of the various Yakutian societies throughout history is needed to judge the extent of the human influence on fire regimes in relatively remote regions, especially since it has the potential to obscure the effects of 530 recent global warming in fire reconstructions.

Conclusions 540
With its continuous sampling scheme and high median temporal resolution, the macroscopic charcoal record at Lake Khamra provides first insights into changes of the boreal fire regime of the last c. 2200 yrs in eastern Siberia, where comparable data are still lacking. Contrary to other studies, current levels of charcoal accumulation at Lake Khamra are not unprecedented within the last two millennia. The reconstructed fire regime changes do not coincide with large-scale shifts in vegetation composition, although short-term increases of evergreen trees and sedges 545 broadly coincide with periods of increased biomass burning around 700 and 1850 CE (phases 2 and 4). Also, low fire activity from c. 900 to 1750 CE (phase 3), expressed as long FRIs and low charcoal accumulation, corresponds to a colder Arctic climate during the LIA. Despite the generally low population density, increased anthropogenic https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License.
forcing after the colonisation of Yakutia by the Russians in the early 17th century might have contributed to an increase in fire frequency, together with rising temperatures. Although northern regions have been warming rapidly 550 in recent decades, charcoal input to the lake has been minimal during the last century, coinciding with new fire management strategies and a ban on fire-related agricultural practices. The mean FRI of 43 yrs is at the lower end of published literature for the wider region and incorporates a range of individual values of up to almost 600 yrs.
Overall charcoal accumulation (classic CHAR background component) and the frequency of identified fire episodes seem to be directly related to each other for the majority of the record. 555 Although this new charcoal record improves data availability from eastern Siberia, more reconstructions, especially from distinctly deciduous regions, are needed to form a detailed analysis of past fire regimes in the Siberian boreal forest. An improved understanding of both fire activity and its drivers throughout history will eventually enable a meaningful assessment of the presence and future of Siberian wildfires and their consequences.

Code availability
The R script used to analyse the charcoal record presented in this study will be made permanently available on Zenodo and linked here. Review access to the script via GitHub: https://github.com/rglueckler/CharcoalAnalysisR

See separate Supplement file
Author contribution 575 ED and UH conceived and designed the study. LP, UH and SK organised the expedition to Yakutia. SV and BB collected the samples and conducted field measurements. RG and SV sampled the sediment core and performed the lab analysis. RG conducted all the charcoal proxy analysis, supported by ED and SK. AA conducted the pollen https://doi.org/10.5194/bg-2020-415 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. and non-pollen palynomorph analysis. BW and ED analysed remote sensing data and created maps of the study location. RG wrote the paper with inputs from ED. All the authors reviewed the final manuscript.