Holocene wildfire regimes in forested peatlands in western Siberia: interaction between peatland moisture conditions and the composition of plant functional types

10 Department of Physical Geography, Goethe University, Altenhöferallee 1, 60438 Frankfurt am Main, Germany Senckenberg Biodiversity and Climate Research Centre (BiK-F), Senckenberganlage, 25, 60325, Frankfurt am Main, Germany 3 STAR-UBB Institute, Babeş-Bolyai University, Kogălniceanu 1, 400084, Cluj-Napoca, Romania Department of Geology, Babeş-Bolyai University, Kogălniceanu 1, 400084, Cluj-Napoca, Romania 15 Department of Biogeography, Paleoecology and Nature Conservation, Faculty of Biology and Environmental Protection, University of Lodz, Banacha 1/3, Lodz, Poland School of Science, Engineering and Environment, University of Salford, Salford, Greater Manchester M5 4WT, UK Bio-Clim-Land Center of Excellence, National Research, 634050, Tomsk State University, Tomsk, Russia Laboratory of Palynology, Faculty of Biology, Sofia University St. Kliment Ohridski, Dragan Tsankov 8, 1164 Sofia, Bulgaria 20 Tuvan State University, Russia; Bio-Clim-Land" Center of Excellence, Tomsk State University, Russia


Introduction
Wildfire is the most common type of disturbance in boreal forests. It can change forest composition and accelerate climate 60 warming via carbon release into the atmosphere and alteration of the radiative balance due to changes in land surface albedo (Rogers et al., 2015;Kharuk et al., 2021). However, the impacts of wildfire on vegetation and climate strongly depend on the fire regime. High-intensity crown fires kill most trees and alter species composition for an extended period. In the short term, https://doi.org/10.5194/cp-2021-125 Preprint. Discussion started: 4 October 2021 c Author(s) 2021. CC BY 4.0 License.
In this study, we used a multi-proxy (pollen, macrocharcoal morphologies, testate amoeba) analysis on two new peat records from Tomsk Oblast to explore the patterns and drivers of the western Siberian fire regime throughout the Holocene.
Specifically, we have determined how peatland moisture has interacted with community composition (with regard in plant functional types) and fuel flammability in driving the frequency and type of fires. Combining our new data with our two other 100 published records from the region allowed a quantification of the degree of spatiotemporal synchronicity in fire activity.

Geographical location and charcoal site selection
The study areas stretch along a west-east transect in a permafrost-free area in Tomsk Oblast, Siberia (Bleuten and Filippov, 2008;Kirpotin et al., 2021). The region has a continental climate with subarctic influences. The mean annual temperature is 105 0°C, the mean temperature of the coldest/warmest season is −17°C/17°C, and the mean annual precipitation is ~ 525 mm falling predominantly in the warm season (www.climexp.knmi.nl). Typical forests in this region are characterised as light taiga type composed of Pinus sylvestris, Betula pubescens, B. pendula, with a greater proportion of dark taiga, including Pinus sibirica, Picea obovata, and Abies sibirica towards the east (Berezin et al., 2014;Rybina et al., 2014). Fires, at present primarily ignited by humans (80%), occur predominantly between spring and early autumn, but summer and autumn fires are 110 more frequently of greater severity (Kukavskaya et al., 2014(Kukavskaya et al., , 2016Kharuk et al., 2021).
We cored two sites, Rybnaya Mire (N57,27566800°, E84,48758900°) located on the river Rybnaya, a tributary of the Ob river, and Ulukh-Chayakh Mire (N57.34326197°, E88,32306189°) located on a terrace of the Chulym river, near the village of Teguldet at the border of the Krasnoyarsk Krai (Fig. 1). The local vegetation at both coring sites includes mesotrophic open sedge-Sphagnum communities with young birch trees at Rybnaya Mire and standing dead tree trunks at Ulukh-Chayakh Mire.

115
The dead trees reflect a rise in water levels across the mire due to road construction in the 1960s. Peat cores (50 cm length, 5 cm diameter) totalling 400 cm at Rybnaya Mire and 348 cm at Ulukh-Chayakh Mire were extracted with a Russian-type corer in 2017 and 2019, respectively.

120
We established the chronology of both cores based on AMS radiocarbon measurements performed at Isotoptech, Debrecen, Hungary (File S1). The 14 C AMS age estimates were converted to calendar years BP using the IntCal20 data set of Reimer et al. (2020) and we constructed the age-depth models using the smooth spline method implemented in the CLAM software (Blaauw, 2010). In the age-depth models, we assigned a surface age of -67 (coring year 2017 for Rybnaya) and -69 (coring year 2019 for Ulukh-Chayakh) to the surface samples of each sequence. We inferred changes in local-scale fire regime based on peat samples of 2 cm 3 extracted at 1 cm contiguous intervals (Whitlock and Larsen, 2001). Charcoal preparation, identification and the categorisation of charcoal morphotypes followed Feurdean et 130 al. (2020a) and Feurdean (2021). To deduce the predominant fire type burnt, we pooled the charcoal morphologies into two categories: non-woody (graminoids, forbs) and woody types (deciduous leaves, wood, and resins) and calculated their ratio.
These two categories represent the dominant fuel sources for surface and crown fires, respectively (Enache et al., 2006;Jensen et al., 2007;Courtney-Mustaphi and Pisaric, 2014;Feurdean et al., 2017;2020a;Feurdean, 2021). At Ulukh-Chayakh, we additionally measured the aspect ratio (length:width) of charcoal particles in selected samples to discriminate between the 135 relative dominance of graminoids versus leaves and wood morphologies (Feurdean, 2021;Vachula et al., 2021). We separated the charcoal fractions into two size classes (150-500 μm and >500 μm) to deduce the approximative location of fires, i.e., local versus on-site fire events. We calculated the influx of charcoal morphotypes and size classes (particles cm -2 yr -1 ) by dividing their respective concentration (particles/cm 3 ) by the sediment deposition rate determined from the age-depth models (yr/cm).
We estimated the frequency and severity of fire episodes using charcoal peaks identified from macrocharcoal particles using 140 the method of Higuera et al. (2009). This methodology involved interpolating CHAR values to constant time steps (30 yr), decomposing the record into a background and a peak component (local fire episodes) using a robust LOWESS smoother (900 yr), and evaluating charcoal peaks using the 95th percentile of the modelled noise distribution. We used the signal-to-noise index (SNI) to assess the suitability of the record for peak detection (Kelly et al. 2011;File S2).
To integrate fire history across the study area, we created composite records of CHAR and fire frequency (FF) by combining 145 the two new records with our two published profiles from Plotnikovo Mire (Feurdean et al., 2019;2020a;Fig. 1). Composite CHAR was created using the Power et al. (2008) protocol implemented in the R palaeofire package version 4.0 (Blarquez et al., 2014) with the following setup for Z score transformation: 8.5 ka to present as the base period, and LOWESS smoothing to a 25-yr mean. The composite FF was created by pooling the identified fire events from all four records and applying a 150yr smoothing. 150

Fire type identification from satellite images and forest statistics
To evaluate the ability of charcoal records to capture the occurrence and severity of fires near the two sites, we used satellite imagery of vegetation cover from LANDSAT 4-5 ТМ, LANDSAT-7 ЕТМ+, and LANDSAT 8 OLI/TIRS (https://earthexplorer.usgs.gov/). We also used observation-based fire data from the Forest Department of the Tomsk Region from 1950 to 2020 CE and data from MODIS sensors onboard the NASA Terra-1 and Aqua-1, which detect forest fires 155 (http://fires.ru/). All maps and images were processed using ESRI ArcGIS version 10.8.

Pollen-based reconstruction of vegetation dynamics
To reconstruct past vegetation dynamics, we analysed pollen and spores mainly at 4 cm intervals in both profiles following the protocol of Willis and Bennett (2001). A pollen sum exceeding 500 terrestrial pollen grains, excluding Cyperaceae, was counted for most samples and converted into percentages based on the total terrestrial pollen sum. The percentages of Cyperaceae, ferns and mosses were calculated relative to the terrestrial pollen sum and their respective individual sum. The bottom part of the Ulukh-Chayakh profile, i.e., from 300-340 cm (age >6500 cal yr BP), contained extremely poorly preserved pollen and was excluded from further analysis.
We assigned pollen-derived tree taxa to one of the fire-related functional PFT groups -resisters (Pinus sylvestris, Larix spp.), avoiders (Picea obovata, Abies sibirica, Pinus sibirica), or invaders (Betula spp.). The PFT-specific fire coping adaptations 165 relate to a taxon's strategy to complete its life cycle in the context of a given fire regime (Gill, 1981;Wirth, 2005). We also separated the tree composition into taxa associated with dark taiga (Pinus sibirica, Abies sibirica, Picea obovata) and light taiga (Pinus sylvestris, Betula spp., Larix spp.) and creating an index for forest density by calculating the ratio between the relative abundance of dark taiga and light taiga tree taxa (Higuera et al., 2014). Additionally, we pooled the non-arboreal pollen types as shrubs, herbs, ferns and mosses. To determine the regional changes in forest composition, we created composite 170 records of PFTs. This was done by averaging the Z-scores and smoothing them to 300 years using a locally weighted regression and a 95% confidence envelope on the same sites used for composite biomass burning reconstructions in the R palaeofire package.

175
To determine changes in peatland hydrology at each site, we used testate amoeba-based water table reconstructions. The testate amoebae preparation and identification were conducted at 4 cm intervals following the methods of Hendon and Charman (1997) and Charman et al. (2000). Based on the available literature (Grospietsch, 1958;Ogden and Hedley, 1980;Mazei and Tsyganov, 2006), we identified and counted a minimum of 150 testate amoebae per sample. No testate amoebae were present in core segments below 285 cm (age > 6000 ka BP) in the Ulukh-Chayakh sequence. We used the transfer function developed 180 for the pan-European region (Amesbury et al., 2016) to derive the water depth. For the composite water table reconstruction, we used the same setting as for pollen.

Numerical methods
To quantify the relationship between peatland moisture conditions, biomass burning, forest density and the relative abundances 185 of PFTs, we performed a correlation analysis. We used microcharcoal influx (CHAR particles >150 μm) as a determination of biomass burning, pollen percentages as a determination of the three PFTs (resisters, invaders, avoiders, and others) and testate amoebae-based water table depth as a determination of peatland moisture. The group "others" includes pollen and spores of shrubs, herbs, ferns and moss. Before the analysis, we interpolated all datasets fed into the model to a 100-year interval using linear interpolation. We also developed generalised linear models (GLMs) to explore the response of biomass burning (CHAR) 190 to changes in peatland moisture conditions (DTW) and to test for the existence of thresholds in the water level at which biomass burning may increase.

195
The age-depth model of the Rybnaya sequence spanned ~ the last 8400 years and showed a mean peat accumulation rate of 25 yr/cm (ranging between 6-36 yr/cm). The Ulukh-Chayakh sequence may cover ~ the last 8500 years, but the chronology of the bottom part of this site (>6000 years) relied on linear extrapolation and it is therefore highly uncertain (File S1). The subsequence covering the last 6000 years had an average temporal resolution of 21 yr/cm (ranging between 3-37 yr/cm).

Site-specific and composite record of biomass burning and fire frequency
Both records displayed a high signal-to-noise index (File S2) with almost all peaks well above 3 (a mean of 4.98 at Rybnaya and 6.45 at Ulukh-Chayakh), the theoretical minimum value for justification of peak analysis (Kelly et al., 2011). At Rybnaya, CHAR was highest between 7.5 and 6 ka and around 4.5 ka, whereas the frequency of high-severity fires inferred from the charcoal peak magnitude and abundance of woody morphologies ranged between 1 and 2 fires/900 yr (Figs. 2a,b,4; Files S2, 205 S3). At Ulukh-Chayakh, CHAR was highest >6 and 3.5 ka, and the fire frequency varied between 2 and 5 fires/900 yr. Both profiles showed a decline towards a minimum in CHAR and frequency of severe fires i.e., low peak magnitude and reduced occurrence of woody charcoal morphologies, approximately between 4 and 1.5 ka (3-1.5 ka at Ulukh-Chayakh). The CHAR increased towards the present at both sites, but the charcoal peak amplitude was considerably higher at Ulukh-Chayakh.
The composite CHAR record showed augmented regional biomass burning between 7.5 and 4 ka (positive Z-scores) with 210 peaks centred at 7.5 and 4.5 ka (Fig. 5). We found a pattern of CHAR decline towards a minimum in the composite profiles between 4 and 2 ka (negative Z-scores), followed by another increase between 2 ka BP to the present (positive Z-scores). The composite fire frequency showed an increase until ca. 5-4.5 ka and then again from 2 ka to the present.

215
Early Holocene data is only available from the Rybanya site (8.5 and 7.5 ka BP), indicating that light taiga and the fire-invader Betula were most abundant (up to 60%) during that period, whereas other light taiga fire-resisters (Pinus sylvestris) and dark taiga and fire avoiders (mainly Picea obovata), had an abundance of 10% and 5%, respectively ( Fig. 3; File S4a). The high proportion of light taiga resulted in a low forest density (dark-to-light taiga ratio of 0.1). Denser forests, with an increased dark-to-light taiga ratio (> 0.2), were identified between 7.5 and 4 ka at Rybnaya and between 5 and 2.5 ka at Ulukh-Chayakh 220 (Fig. 5;File S4a,b). The light taiga was composed of up to 70% fire-resisters (Pinus sylvestris and Larix spp.), while the dark taiga, with its fire avoiders (Pinus sibirica, Picea obovata, and Abies sibirica), made up to 25% of the forest. A decline in fireresisters (down to 50%) and forest density and increasing invader abundance (up to 60%) occurred after 4.5 ka at Rybaya and after 2.5 ka at Ulukh-Chayakh, indicating a change in forest composition. Between-site variability was highest within the avoider group. Picea obovata had higher values at Rybnaya (up to 10%) than at Ulukh-Chayakh (3%), whereas Abies sibirica 225 was more abundant at Ulukh-Chayakh (15%) than at Rybnaya (7%; Fig. 4). The composite vegetation record showed a higher proportion of fire-resisters between 7.5 and 5 ka, fire avoiders between 7.5 and 3 ka, and fire invaders over the last 4.5 ka (

Site-specific and composite records of peatland moisture conditions
At Rybnaya, the testate amoeba-based reconstruction of the water table indicated relatively high levels (around 15 cm below surface) between 8.5 and 7.5 ka, 4 and 2.5 ka, and over the last 1 ka (Fig. 4). We reconstructed low levels (20-30 cm below surface) for the time interval between 7.5 and 4.5 ka and 2.5 and 1 ka (Fig. 5). At Ulukh-Chayakh, the water table was high between 3.5 and 2.5 ka, 2 and 0.5 ka, and towards the present day. There, times of low water table occurred between 6 and 3.5 240 ka, 2.5 and 2 ka, and around 0.5 ka. The composite record of the water table showed high levels approximately between 8.5 to 7.5 ka, 3.5 and 3 ka and during the last 1.5 ka, and low levels between 7.5 and 4.5 ka, and around 2.5 ka (Fig. 5).

Numerical analysis
The correlation analysis at both sites showed that the decline in water table position (higher values representing drier 245 conditions) was positively associated with the proportion of fire avoiders, resisters, and forest density, and negatively with invaders and others (Fig. 6a, b). Although the correlation coefficients were stronger for Rybnaya than Ulukh-Chayakh, none of these were statistically significant (Appendix A1). The generalised linear model showed no significant influence of water levels above 20 cm on the probability of fire occurrence and severity at both sites (Fig. 6c,d). However, a drop in water level below 20 cm contributed statistically significantly to the increase in fire occurrence and severity at Rybnya but had no 250 statistically significant influence at Ulukh-Chayakh.

Changes in fuel, fire type and fire frequency over the Holocene
Stand-replacing fires combust substantial amounts of biomass because they burn entire trees. Such fires manifest themselves 255 in a greater abundance of charcoal particles that exceed the quantities typically produced by surface fires. This feature allows a distinct separation of charcoal peaks from charcoal background levels, which yields a higher signal-to-noise index (Higuera et al., 2005;Courtney Mustaphi and Pisaric 2014). Indeed, the comparison of sedimentary charcoal from sub-recent samples with satellite imagery of the region (Plotnikovo Mire) has confirmed that the marked increase in charcoal particles towards the present day at this location, closely agrees with the occurrence of high-severity fires (Feurdean et al., 2020a). pieces) from sub-surface samples in both cores (Fig. 2). A high-severity fire event occurred in 2015 some ca. 30 km from Ulukh-Chayakh, which is not documented in our surface charcoal record and corroborates the localised origin of charcoal found in this sedimentary profile. Samples from the Ulukh-Chayakh site show a minor increase in charcoal (2-7 pieces) ca. 265 fifty years ago, possibly related to a local fire event in 1993. The first clear charcoal peak at this site (250 pieces), with numerous wood fragments, was found ca. 200 years ago, whereas the preceding charcoal peak (400 pieces) was ca. 900 years ago (Fig. 2a,b; File S3). These two peaks therefore demonstrate the intermittent occurrence of infrequent, high-severity local fires that produce large quantities of charcoal, embedded in what is otherwise predominately a surface fire type regime generating low levels of charcoal. It also indicates that the charcoal peak extraction method applied here reliably reflects the 270 occurrence of high-severity local fires.
We reason that the prevalence of a high charcoal influx with well-defined charcoal peaks and abundant woody morphotypes found between ca. 7.5 and 4 (3) ka at both sites, and over the past 1.5 ka at Ulukh-Chayakh, is indicative of the predominance of high-severity local fires ( Fig. 2; File S3). Between 4 (3) and 1.5 ka, the observed low charcoal influx and peak magnitude were coupled with a lower abundance of woody morphologies at both sites, although the latter feature was more pronounced 275 at Ulukh-Chayakh ( Fig. 2; File S3). For this period, we inferred predominately low-temperature surface fires that mostly burned understory biomass. Experimental production of charcoal additionally provides evidence that graminoids and Sphagnum produce little charcoal during high-temperature fires (Hudspith et al., 2017;Feurdean, 2021). Measurements of aspect ratio and surface area on selected charcoal samples at Ulukh-Chayakh show that the aspect ratio only partially agrees with the predominant fuel type inferred from charcoal morphologies. A lower aspect ratio of charcoal particles, typical for 280 wood and leaf morphotypes, was what we expected at times of increased woody morphologies, and an increased aspect ratio with a rise in the relative proportion of graminoids (Feurdean, 2021;Vachula et al., 2021). This is probably the results of a mixed morphology of charcoal types in the same sample (Fig. 2). The overlapping morphologies of leaves and wood from trees and shrubs in the same sample also make it hard to distinguish between the dominant fire type as high-severity surface fires can combust both living shrubs and dead wood and leaves. Our measurements of the surface area indicate high values at 285 times of an increased abundance of leaves ( Fig. 2; File S3) with a typically high surface area (Crawford and Belcher, 2014;Feurdean, 2021), but also with the presence of larger size particles, which complicates any interpretation of a link between charcoal surface area and the increased predominance of leaves.
Our reconstructed fire return interval is ~ 450 yr (2 fires/900 yr) throughout most of the Holocene (Fig. 4). However, fire frequency (at Rybnaya) and severity (at Ulukh-Chayakh) increased over the last 1.5-2 ka BP compared to the long-term 290 averages of both records suggesting an intensification of fire activity during the most recent two millennia. The composite fire frequency derived from all four records also shows increasing biomass burning and fire frequency over the last 2-1.5 ka (Fig.   5). The reconstructed fire return interval (FRI) in the study area is in line with those from other sites located in forested peatlands (Pinus-Betula dominated, Picea obovata, Abies sibirica) in Russia, which report an FRI range between 100 and 600 yr (Barhoumi et al., 2019(Barhoumi et al., , 2020. In contrast to these studies, we have not found a gradual increase in fire frequency from the 295 early Holocene towards the present, but rather two distinct periods of enhanced activity, between 7.5 and 4 ka, and approximately over the last 2 ka (Fig. 5). A similar increase in fire frequency during the last two millennia has also been recorded at other sites in southern Siberia, such as Lake Baikal (Bourhami et al., 2021)

Drivers of vegetation and fire regime change
Climate conditions influence peatland moisture, fuel dryness, flammability and the composition of moisture-and fire-related PFTs. The relative abundances of the fire-related PFTs, in turn, influences the prevalence of specific fire regime types, e.g., low-intensity vs. high-intensity. Below, we discuss how climate change affects hydrological conditions in peatlands. We also highlight how climate and hydrology interact with forest and peatland plant composition, and fuel flammability. between 8 and 30 cm below the peat surface and was around 20 cm at both sites during large parts of the Holocene (Fig. 4).
We found that the probability of occurrence and intensification of fire severity was greater at times when the water level dropped below 20 cm (drier conditions), and lower at water levels between 10 and 20 cm below the surface (wetter conditions), 315 a pattern that is more evident at Rybnaya (Fig. 6 c,d). The stronger moisture-fire relationship at Rybnaya may be related to the higher amplitude of changes in local hydrological states and the overall drier conditions at this site (8-30 cm) compared to Ulukh-Chayakh (10-23 cm), which remained moister over the Holocene (Fig. 6). Water tables of mesotrophic mires are coregulated by the difference between precipitation (P) and evapotranspiration (Et) and external water inflow. The detrital element Ti, a possible indicator of water influx, was high in the bottom profiles that were rich in minerogenic material. Ti 320 increased between 4 and 3 ka and 1.5-0.1 ka at Ulukh-Chayakh and only slightly over the last 1 ka at Rybnaya (File S5). This evidence suggests that, except of these time intervals, the water table reconstructions primarily reflect the P-Et balance.
At a Holocene scale, the intensity of fires and/or fire size was greatest between 7.5 and 6.5 ka and 5 and 4 ka, but not fully synchronous between the sites (Figs. 4, 5). From 9 to 4.5 ka, annual temperatures in the Northern Hemisphere were up to 3.5°C warmer than at present (Fig. 5) and exhibited a more pronounced seasonality, i.e., higher summer and lower winter 325 temperatures (Kauffman et al., 2020;Bova et al., 2021). Proxy records from Siberia attest to warmer and drier-than-present climate conditions between 9 to 6 ka and a temperature and moisture optimum between 6 and 4.5 ka BP (Groisman et al., 2012). Warm summer temperatures likely enhanced evapotranspiration and consequently lowered peatland water levels, leading to drier surface conditions (Fig. 5). These conditions, in turn, increased fuel flammability, allowed a denser tree cover to establish on the peatlands and fostered enhanced fire severity (see 4.3).
Contrastingly, the interval of lowest biomass burning and fire severity approximately between 4 (3) and 1.5 ka BP coincides with one of the periods of wettest peatland conditions during the Holocene (at Rybnaya) or a highly fluctuating water table (at Ulukh-Chayakh; Fig. 4) Annual and summer temperatures declined after 4.5 ka to values below the present day (Kauffman et al., 2020;Bova et al., 2021). Regionally, this cooling trend, which started at approx. 4.5 ka (Groisman et al., 2012) (Fig.4).
We hypothesise that the climatic linkage between the cool and wet conditions that imply high fuel moisture conditions and typically low fire activity was likely disrupted by anthropogenic activity (Fig 3; File S4a, b). Historical records indicate the 340 widespread colonisation of the region by the Russians at this time that was associated with the conversion of forest to arable land and pastures, and likely increased anthropogenic ignitions (Naumov, 2006). Most recently, local fire suppression near the village of Teguldet may have hampered contemporary fire spread on the mire leading to the recent pattern of low fire activity.

Feedbacks between peatland moisture, community PFT composition and fire regime
Peatland moisture conditions affect the relative abundances of PFTs. Fire-related PFTs, in turn, create positive or negative feedbacks for fire regimes via specific fire-related traits (Feurdean et al., 2020a;Kharuk et al., 2021). We found a dominance of open light taiga species and fire invaders (Betula spp) at times of higher water table (wet conditions) between 8.5 and 7.5 ka and after 4.5 and 3.5 ka (Figs. 2, 4). However, Betula spp. also remained dominant during the periods of water table decline recorded between 2.5 and 1.5 ka. B. pubescens grows on wet peatlands, whereas B. pendula grows more frequently on drier 350 soils and peatlands with higher fire incidence (Wirth, 2005;Groisman et al., 2012). Unfortunately, pollen analysis cannot differentiate the prevalence of birch species with contrasting hydrological and fire tolerances because their pollen is too similar.
Denser forests of light taiga and fire-resisters (Pinus sylvestris) prevailed between 7.5 and 4.5 ka. Dark taiga taxa and fire avoiders, primally Pinus sibirica at Rybanya and Abies sibirica at Ulukh Chayakh (only from 5.5 ka), intersperse these forests ( Fig.3; File S4a,b). This forest composition coincides with drier peatland conditions and an increased severity in fires.

355
Although not statistically significant, our correlations suggest a positive association between a high-water level and Betula, and low water level conditions with all conifer types (Fig. 6 a,b). We deduce from this a greater tolerance of broadleaf trees to wetter conditions compared to conifers. Visual trends and correlation reveal that the biomass burning increase with forest density was probably connected to dry peat conditions (Figs. 4, 6). The greatest biomass burning was found for intermediate forest density indices from 0.15 to 0.3 (max. 0.6; Fig. 4). This finding is consistent with emerging evidence on fire-fuel 360 relationships in boreal forests, which points to a higher fire hazard in less dense than denser forests, which allow fuel to dry (Scheffer et al., 2012;Feurdean et al., 2020b).
The presence of the fire-resister, such as Pinus sylvestris, in Siberia is associated with a light surface fire regime (mean FRI of al., 2016). Fire avoiders typically experience severe stand-replacing crown fires at relatively long return intervals (mean FRI 365 150 years; range 99 and 300 years; Goldammer and Furyaev, 1996;Schulze et al., 2005;Wirth, 2005;Kharuk et al., 2021).
The post-fire regeneration pathway of conifer species involves recruitment directly after burning (P. sylvestris) or after an early successional phase initiated by Betula (Pinus sibirica, Picea obovata, and Abies sibirica). A mixed taiga forest with broadleaf trees and dark taiga conifer taxa typically develops between 60-120 years after a fire, followed by the dominance of conifers as a late-successional stage (Tautenhahn et al., 2016;Coop et al., 2020;Kharuk et al., 2021). We reason that the 370 prevailing periodicity of severe fires (every 180 to 450 years) between 7.5 and 4.5 ka provided a fire-free period that was long enough for the fire avoiders Pinus sibirica, Picea obovata, and Abies sibirica to reach reproductive age. Alternatively, burning may have been too patchy to create sufficiently large forest gaps to limit seed dispersal and inhibit post fire recovery. Pinus sibirica is typically classified as a fire avoider but behaves like a fire resister when reaching old age, which could also explain its increased presence at times of high fire activity. Fire also allows dark taiga species to compete with P. sylvestris and become 375 established beyond poor soils and boggy areas (Kharuk et al., 2021).
A different fire regime and altered vegetation feedbacks emerged over the last two millennia at Ulukh-Chayakh, the site that experienced more severe fire episodes. It should be acknowledged that each identified fire peak could have come a few consecutive fire events that we were not able separate using the charcoal records. The intensification of local fire severity at Ulukh-Chayakh paralleled a decline in the proportion of fire avoiders (mostly Abies sibirica and Picea obovata), an increase 380 in fire invaders (Betula) and the abundant occurrence of heathland shrubs (Figs. 3. 4). Shrubs burn hotter than other surface fuel types and reach temperatures that kill most conifer seedlings (Tautenhahn et al., 2016). High severity burned areas have also been found to be more prone to repeated burning and have a more negative impact on conifers (Kukavskaya et al., 2016;Whitmann et al., 2019;Coop et al., 2020). We propose that the intensification in fire severity at Ulukh-Chayakh may have eliminated mature Abies sibirica and Picea obovata, limited their seed dispersal and disrupted their successional pathways.

385
These dynamics further resulted in a shift in forest composition towards more post-fire-adapted invader communities with better dispersal capabilities, recruitment strategies for burned areas, and rapidly maturing taxa. Our results support emerging findings of increases in broadleaf trees (Betula or Populus) at the expense of evergreen conifers that are associated with contemporary warming and increasing fire severity in forests in Siberia and North America (Kelly et al., 2013;Tautenhahn et al., 2016;Mekkonen et al., 2019;Whitman et al., 2019;Kharuk et al., 2021). At Rybnaya, the modest increase in biomass 390 burning over the past two millennia coincided with wet conditions and a forest composition remaining dominated by Betula, indicating that fire has played a more limited role in forest dynamics at this site.

Conclusion
This study provides novel insights into past fire regime based on Holocene records from forested peatlands in western Siberia. greater forest density. Although the probability of fire occurrence on these peatlands increased with drier conditions and forest density, the greatest fire incidence occurred when the water level declined below 20 cm and at an intermediate forest density.

400
At a Holocene timescale, we found two scenarios of moisture-vegetation-fire interactions. In the first, an enhanced fire severity episode occurred between 7.5 and 4.5 ka and was associated with a low water level and a higher proportion of dark taiga/fire avoider taxa (mainly Pinus sibirica at Rybnaya and Abies sibirica at Ulukh Chayakh) in a light taiga and fire resister (Pinus sylvestris) dominant community. However, the second period of increased fire severity (i.e., ca. last 1.5 ka) coincided with a reduced abundance of fire avoiders (mostly Abies sibirica and Picea obovata) and an expansion of fire invaders (Betula).

405
These community changes demonstrate that frequent fires of higher severity can lead to compositional changes in forests, either because the trees could not reach their reproductive age between burning events or the fires created substantial forest gaps that hindered seed dispersal. This study also shows certain prolonged periods of synchronous fire activity across sites, suggesting that the magnitude of centennial to millennial-scale climate variability over the Holocene was marked enough to drive fire regimes on a regional scale.

410
Based on our findings from fossil records, the frequent warm and dry spells predicted in Siberia for the future by climate change scenarios will likely enhance peatland drying and may convey a competitive advantage to some conifers, especially Pinus sylvestris. But dry peatland conditions, particularly water levels below 20 cm, will also exacerbate the likelihood of fire frequency, increase their severity, and likely disrupt the successional pathway of other typical taiga conifers, notably that of fire avoiders. Such a fire regime change may accelerate a shift towards broadleaf fire invader-type trees. Future climate-415 disturbance-fire feedbacks will not only accelerate changes in boreal forest structure and composition, but also impact the carbon balance of the forested peatlands and ultimately lead to albedo-mediated feedbacks on the regional climate system.

Data availability
All data generated for this paper will be deposited in Neotoma database upon publications.    and Ulukh-Chayakh (d).