Lomonosovfonna is one of the highest ice fields in Svalbard (Fig. 1), reaching 1250 m a.s.l. in its accumulation area (Isaksson et al., 2001). For this reason, it is less affected by summer melting and meltwater percolation than other low-elevation glacier sites in Svalbard (Gordiyenko et al., 1981; Pohjola et al., 2002), making it suitable for ice core studies. Therefore this site has already been regularly studied in the past. Two deep ice cores were retrieved in 1976 and 1982 by pioneering Soviet expeditions, mainly for stratigraphic purposes (Gordiyenko et al., 1981; Zagorodnov et al., 1984). The first extensive study that retrieved both physical and chemical records from the ice was conducted on a deep ice core drilled in 1997 by an international team (Isaksson et al., 2001).

In March 2009, using the Fast Electromechanical Lightweight Ice Coring System (FELICS) (Ginot et al., 2002), a Norwegian–Swedish–Swiss team drilled a 149.5 m long ice core on Lomonosovfonna at 1202 m a.s.l. (78∘49′24.4′′ N, 17∘25′59.2′′ E), 4.6 km south of the 1997 drilling site, which could not be accessed due to the opening of a large crevasse. Bedrock at about 200 m of depth was not reached during the drilling. Further details about the drilling and the meteorological setting can be found in Wendl et al. (2015). This ice core was then shipped frozen to the Paul Scherrer Institute (PSI) in Switzerland. In addition to the LF-09 ice core, a 7.6 m shallow firn core was retrieved in April 2011 ∼110 m to the north and ∼20 m to the west of the LF-09 site by a team from the Uppsala University and processed at the Norwegian Polar Institute (NPI) in Tromsø (Vega et al., 2015b).

The LF-09 ice core was processed in a -20 ∘C cold room at PSI following well-established procedures (Eichler et al., 2000). In total, 3997 samples were cut at 3–4 cm resolution, which corresponds to about monthly resolution at the top of the core to annual resolution at the bottom. The dating was performed using a multi-parameter approach including annual layer counting of the δ18O and Na+ signals back to 1750, 210Pb decay, a two-parameter fit, and absolute reference horizons such as volcanic eruptions (sulfate peaks) and the well-documented 1963 tritium peak (Wendl et al., 2015). The LF-09 ice core thus spans the time period from 1222 to 2009, with an average accumulation rate of 0.58±0.13 m yr-1 of water equivalent (w.e.). The dating uncertainty was assessed to be ±1 year in the vicinity of the reference horizons and ±3 years in between up to 68 m w.e. of depth, ±3 years in both cases between 68 and 80 m w.e., and ±10 years below 80 m w.e. (Wendl et al., 2015). Wendl et al. (2015) also determined the annual melt percent by calculating the relative thickness of the melt features (i.e., ice lenses) for each year. For rBC analysis, two parallel samples from the inner part of the core were combined in a 50 mL polypropylene vial. When the two samples were too long to fit in a single vial, the cutting resolution was increased, overall resulting in 4046 samples. As the first 110 samples had been melted and refrozen prior to analysis, leading to potential rBC losses (Wendl et al., 2014), they were not considered in this study. Thus, the 2005–2009 time period is not covered by the LF-09 rBC record. In addition, VA and p-HBA were analyzed in the whole LF-09 ice core by Grieman et al. (2018).

The LF-11 ice core was cut at 4 cm resolution following clean protocols, resulting in a total of 155 samples (Vega et al., 2015b). The dating was performed by counting δ18O annual cycles using the winter minimum as a reference and by matching prominent ion peaks with those from the LF-09 ice core (Vega et al., 2015b, 2016). The LF-11 core thus spans the time period from 2004 to 2011, with an average accumulation rate of 0.49 m w.e. yr-1.

The entire LF-09 core and the LF-11 core were analyzed for rBC at PSI in several campaigns between 2012 and 2016 and in April–May 2016, respectively, following the procedure established by Wendl et al. (2014) for liquid samples and further evaluated by Lim et al. (2014). Discrete rBC samples were melted at room temperature, sonicated in a ultrasonic bath for 25 min, and immediately analyzed using a SP2 (Droplet Measurement Technologies, USA) (Schwarz et al., 2006; Stephens et al., 2003) coupled with a jet nebulizer (APEX-Q, Elemental Scientific Inc., USA). External calibrations from 0.1 to 50 ng g-1 (R2>0.999) were usually performed twice a week by preparing eight fresh dilutions from a rBC standard (Aquadag^®, Acheson Industries Inc., USA) (Gysel et al., 2011; Wendl et al., 2014). Before and after every day of measurements, the APEX-Q and the upstream capillaries were rinsed for 10 min with a solution of 3 % nitric acid prepared with ultrapure water (Sartorius, ≥18.2 MΩcm, with a 0.2 µm filter). The liquid flow rate of the APEX-Q was monitored several times per day to avoid changes in the nebulizing efficiency, which can impact the rBC detection. When the flow rate was not steady (i.e., presented fluctuations larger than ±10 %), the measurements were stopped. The instrumental blank was checked between every sample by rinsing the setup with ultrapure water until the rBC signal returned to the baseline value of 0–1 particle cm-3. The limit of detection (LOD) was estimated by measuring ultrapure water blanks 10 times and by calculating their average plus 3 times their standard deviation, leading to a value of 0.051 ng g-1 of rBC. The procedure blank was controlled by analyzing frozen ultrapure water treated in the same way as ice core samples. Its value was always below the LOD (typically 0.01 ng g-1 of rBC), which confirms the adequate cleanliness level of the analytical procedure. In addition, six series of 20 replicate samples were cut from parallel ice core sticks and were analyzed (Fig. 2a), showing a high level of reproducibility between original and replicate samples (r=0.73, p<0.001, n=120, and averaged relative error for the 120 samples: 23 %).

Only the first 900 LF-09 samples were manually analyzed. Then a CETAC ASX-520 autosampler (CETAC Technologies, USA) was implemented in order to speed up the measurements and improve their reproducibility. Total rBC particle counting was kept to 10 000 as recommended (Schwarz et al., 2012), the limiting condition being a measuring time between 1 and 30 min. The autosampler probe was rinsed with ultrapure water for 45 s between each sample and the waiting time in each vial before data acquisition was set to 1 min 45 s, which turned out to be sufficient for the background signal to become stable. However, some difficulties arose from the fact that rBC concentrations tend to decrease with time due to particles sticking to the walls and agglomerating beyond the SP2 detection range, which implies that rBC samples have to be measured as fast as possible after sonication (Lim et al., 2014; Wendl et al., 2014). We therefore studied the rBC degradation with time by using 24 ice core samples from Lomonosovfonna and the Swiss Alps (Colle Gnifetti and Fiescherhorn ice cores). Each sample was measured between 5 and 14 times (depending on the concentration) over 24 h. They all showed a similar decreasing trend, largely independent of the ice core site, the rBC or dust concentrations. On average, the relative apparent rBC loss was 41±9 % after 24 h, and 10±6 % after 6 h, which was thus defined as the maximum waiting time considered as acceptable. To take this decrease into account, a linear regression was performed and a systematic correction as a function of time was implemented as follows: rBCcorrected=rBCmeasured1-0.017t, where [rBC]corrected is the original rBC concentration at the initial time (t0), i.e., the end of the sonication (ng g-1), [rBC]measured is the rBC concentration measured by the SP2 (ng g-1), and t is the time in hours (h) elapsed since the end of the sonication.

Historical BC emission inventories reconstructing past emissions and atmospheric loading are used to compare the LF rBC record with estimated trends in anthropogenic BC produced by fossil fuel and biomass combustion in order to carry out source apportionment. Here, we use the BC emission inventory from Bond et al. (2007) available at 5-year resolution, between 1850 and 2000, for countries or areas identified as potential BC source regions: Canada, the USA, OECD Europe, eastern Europe, and the former USSR. This inventory includes emissions from fossil fuel and biofuel combustion, but does not include open burning such as wildfires, which contribute to a substantial part of the Arctic BC burden in summer (Stohl, 2006).

Our approach to detect years with increased forest fire activity in the LF-09 ice core follows the methodology proposed by Fischer et al. (2015), which basically uses an outlier detection approach. From the annual averages, 31-year moving medians were created. For each year, the residues between the median and average were calculated. Median absolute deviations were obtained by averaging the residues over the whole LF-09 time period for specific fire proxies (VA, p-HBA) or proxies showing little anthropogenic influence (formate), and between 1222 and 1800 for proxies influenced by anthropogenic emissions (rBC, ammonium). Then a fire threshold was defined as the median plus 3 times the median absolute deviation, in such a way that this threshold takes background variations into account, which is of prime importance in the case of nonspecific proxies having other sources than biomass burning, such as biogenic emissions in the case of formate (Legrand and De Angelis, 1996) and ammonium (Eichler et al., 2009; Fischer et al., 2015; Kellerhals et al., 2010). Every annual average above this threshold can thus be considered a year with significant biomass burning emissions, and finally the centennial frequency of such episodes was obtained. As peaks can be smeared out over several years because of layer thinning with depth and post-depositional processes, a correction for neighboring outliers was implemented: a proxy signal exceeding the detection threshold over several consecutive years was considered to be a single episode and counted only once. As VA and p-HBA were available at lower resolution than the other species (Grieman et al., 2018), annual resolution was sometimes not achieved in the deepest part of the ice core and therefore missing year values were obtained by linear interpolation between the adjacent years.

Here we present the long-term and high-resolution rBC record from Svalbard derived from the combination of the LF-09 and LF-11 ice cores spanning the time periods 1222–2004 and 2004–2011, respectively (Fig. 2a). In the LF-09 ice core, rBC concentrations are generally low with a range between the LOD (i.e., 0.051 ng g-1) up to 39.0 ng g-1 in 1980, an average of 1.2±2.3 ng g-1 (uncertainties are given as ±1σ unless otherwise stated) and a median of 0.6 ng g-1. Nominal sub-annual resolution was achieved for the entire time period, with a monthly resolution back to about 1930. Such a high resolution enables us to disentangle seasonal variations in the rBC ice core signal. Ice core rBC concentrations in Greenland (McConnell et al., 2007) as well as atmospheric BC concentrations at the Zeppelin station (Eleftheriadis et al., 2009) show a clear seasonal cycle with higher BC values in winter–early spring due to the Arctic haze phenomenon (Shaw et al., 1995). However, despite a fine resolution and contrary to other parameters such as δ18O or Na+ concentration, a clear seasonality of rBC in the LF-09 core, i.e., with winter maxima (Arctic haze) and some summer peaks (biomass burning) (Hirdman et al., 2010a; Stohl, 2006) was not observed on a regular basis, probably due to summer snowmelt affecting the rBC signal (see Sect. 3.3).

The rBC concentrations in the LF-11 ice core are comparable to preindustrial values (before 1800) measured in the LF-09 ice core, with an average of 0.5±0.4 ng g-1, a median of 0.3 ng g-1, and a range from LOD to 2.4 ng g-1. This does not seem to be related to spatial variability as the two records show similar rBC concentrations in the overlapping year 2004 (Fig. 2b) but is related to an overall decreasing trend in BC emissions in the source regions. An analogous clear drop was also observed in the HDF EC concentrations (Ruppel et al., 2017) in a recent shallow core spanning 2005–2015 (Table 1) compared to the original HDF ice core covering 1700–2004 (Ruppel et al., 2014). The very low rBC concentrations in the LF-11 core compared to the EC concentrations around local sources of contamination in Svalbard such as settlements and mining activities with values higher than 1000 ng g-1 for some samples (Aamaas et al., 2011) underlines that rBC contribution from local anthropogenic sources to the LF drill site appears to be minimal, at least for the most recent years.

LF rBC concentrations are very similar to those observed in Greenland and Canadian Arctic ice cores obtained by SP2 analyses (Keegan et al., 2014; McConnell et al., 2007; Sigl et al., 2013; Zdanowicz et al., 2017) (Table 1). However, EC concentrations in Svalbard snow (Aamaas et al., 2011; Doherty et al., 2010; Forsström et al., 2009, 2013) as well as in the HDF and Fiescherhorn ice cores (Jenk et al., 2006; Ruppel et al., 2014, 2017) are 1 order of magnitude higher than rBC concentrations in the topmost part of the LF core. This can be mainly explained by the different analytical methods employed, which do not measure the same fraction of the carbonaceous compounds, as discussed by Ruppel et al. (2014). Whereas the SP2 does not detect rBC particles larger than 500 nm, the optical and thermal-optical methods include a filtration step in which the smallest fraction of EC particles is generally lost. Lim et al. (2014) reported significant variations in the EC / rBC ratios in snow and ice, ranging from 0.5 to 3.4 according to the sample origin. Furthermore, in the aforementioned studies, most of the EC snow samples from Svalbard were collected in winter–spring, when EC concentrations in fresh snow are higher due to the Arctic haze and have not yet experienced summer melting of the snowpack, contrary to the LF ice core samples.

Up to now, it remains unclear what happens to BC when melting occurs at the surface of the snowpack. BC can be enriched at the surface due to its low solubility in water or it can be eluted with the meltwater and percolate downward through the snowpack. When the water refreezes further down, forming an ice lens, BC is trapped. If melting is considerable, runoff can occur, leading to a net loss of BC. Doherty and al. (2013) showed that BC particles tend to be retained at the snow surface when melting occurs and that only 10–30 % of the BC is eluted with meltwater through the snowpack. However, Xu et al. (2012) observed that BC concentrations were higher not only at the surface but also in firn, at the bottom part of the snowpack, due to BC percolation and enrichment on top of superimposed ice, hindering further penetration of meltwater, while the intermediate snowpack zone was depleted in BC. Moreover, fresh snow displayed higher BC concentrations compared to snow experiencing summer melting. If BC concentrations are high, percolation can even be the dominant process (Conway et al., 1996).

In Fig. 6 a qualitative assessment of the melting impact on the LF-09 record is made. Quantitative values cannot be obtained, especially as runoff might have occurred. In addition to the annual melt percent (Fig. 6b) calculated by Wendl et al. (2015), we use the melt index (Fig. 6c) defined as log⁡Na+/Mg2+ by Iizuka et al. (2002) to quantify the impact of summer melting in Svalbard as sodium and magnesium have different washout efficiencies (Beaudon et al., 2013; Grinsted et al., 2006), a higher value being indicative of a stronger melting. The highest values for both the annual melt percent and melt index are found in the 20th century, in agreement with temperature reconstructions from the Arctic (McKay et al., 2014) and the water stable isotope records from both LF-09 (Wendl, 2014) and LF-97 cores (Divine et al., 2011) showing a slight downward trend until the second half of the 19th century followed by an abrupt rise in the 20th century. Prior to 1850, colder temperatures, a lower melt index (average 1222–1850: 0.64) and melt percent (average 1222–1850: 28 %), and simultaneous peak occurrence for species with different sensitivity to elution (e.g., NH4+ and rBC) all suggest a reduced impact of melting and a better preservation of the original concentration.

The melt index shows two periods of enhanced melting in the LF-09 ice core, from the 1910s to 1930s and in the 1980s–1990s. The first one is associated with the well-known early 20th century pan-Arctic warming (Bengtsson et al., 2004), also clearly visible in the temperature series from Svalbard Airport in Longyearbyen (Fig. 6d, Nordli et al., 2014), while the second one is explained by the current global warming trend (AMAP, 2011b; IPCC, 2013). The existence of local algae in the LF-97 core only between 1900 and 1940 (Hicks and Isaksson, 2006) underlines the fact that wet surface snow was present at that time at the drill site, another clear indication of summer melting. In contrast, the annual melt percent displays the highest values around 1905 and some local maxima around 1955 and in the 1980s. Our hypothesis is that the strong 1920s melting peak was responsible for extensive water percolation through the snowpack, leading to the formation of ice lenses and producing the melt percent peak around 1905. This would correspond to a percolation length of over 15 years, which strongly exceeds the up to 8 years postulated by Kekonen et al. (2005). A nonnegligible fraction of the rBC particles might have been eluted with the meltwater. Another substantial fraction might have been lost by runoff, which could explain the rBC minima from the 1910s to 1930s. Regarding the melt index peak in the 1980s and 1990s, as no clear increase can be seen in the annual melt percent record in the previous years, percolation must have been overwhelmed by surface runoff responsible for rBC losses. The dramatic decline of the melt percent since the 1990s also confirms the dominant contribution from runoff over percolation due to increasingly warm temperatures. Kekonen et al. (2005) also noted ion losses due to runoff since the 1990s in the LF-97 ice core. Nevertheless, we suggest that the decreasing trend in rBC since the 1970s is not only an artefact due to melting issues but is primarily driven by reductions in source emissions as confirmed by atmospheric measurements throughout the Arctic.

The case of the 1885 rBC minimum is more puzzling as there is no evidence of strong melting either in the melt index or in the annual melt percent record, which could indicate that losses happened only by runoff. Melting occurrence is supported by red layers typical of algae growing only in the presence of liquid water in LF-09 ice core sections around 1879–1881 and 1883–1885. Interestingly, Keegan et al. (2014) described a widespread melting event in Greenland associated with a prominent ice layer corresponding to the year 1889, which would lie within our dating uncertainties.

Furthermore, we cannot fully exclude that the apparent loss of rBC due to melting is an artefact of the SP2 analytical method. Losses of rBC from samples which were melted and refrozen in laboratory tests can reach 45±11 % (Lim et al., 2014) or even up to 60 % for a single thaw–freeze–thaw cycle prior to analyses (Wendl et al., 2014). One possible explanation is the agglomeration of rBC particles to larger sizes during the refreezing process, beyond the SP2 detection range (Wendl et al., 2014). It is unclear if such processes similarly occur during melting and refreezing of a snowpack, which would make the use of SP2-based methods less suited for ice cores experiencing strong summer melting.

Even if the impact of summer melting and anthropogenic emissions hampered the use of rBC as a biomass burning proxy since the beginning of the Industrial Revolution, it is still possible to reconstruct past biomass burning trends in the preindustrial times (before 1800). This part of the record has a limited melting effect due to low air temperatures and presumably no anthropogenic input. In this period every rBC peak is assumed to correspond to a biomass burning episode whose emissions were transported to and deposited at the drilling site. Natural rBC emissions (wildfires) may also have contributed to the rBC record since the beginning of the Industrial Revolution but their signature is largely masked by the anthropogenic signal. For instance, the highest rBC concentrations of the record in 1980 and 1981 could be linked with strong biomass burning seasons in Canada (4.8 and 6.1 Mha, respectively; Stocks et al., 2003) potentially related to the ammonium spikes noted in Greenland ice cores (Legrand et al., 2016). The clear rBC peak visible in the LF record in summer 1994 could reflect the high fire activity in Canada for the year 1994, when 6.1 Mha burned (Stocks et al., 2003). Dibb et al. (1996) documented the advection of a biomass burning plume from the Hudson Bay lowlands, Canada, to Greenland on 5 August 1994, suggested to be responsible for an increase in NH4+, K+, and light carboxylic acid concentrations in the snowpack. Although northern Eurasia is assumed to be the main source of rBC in the LF record, those events reveal that advection of forest fire plumes from North America could possibly reach Svalbard under favorable conditions.

To identify common variability among the chemical species in the LF ice core and isolate biomass burning proxies, we performed a principal component analysis (PCA) (Table 2). We used normalized annual averages and restricted our analysis to the preindustrial period (1222–1859) as many compounds (e.g., sulfate, rBC, ammonium, and nitrate) are influenced by anthropogenic activities. Four principal components (PCs) were retrieved. PC1 has high loadings of sodium, magnesium, potassium, calcium, nitrate, sulfate, chloride, and methanesulfonate, representing 52 % of the total variance and can be explained by mineral dust and marine sources (Wendl et al., 2015). PC2 isolates light carboxylic acids (formate, acetate, and oxalate), accounting for 19 % of the total variance. These compounds are well-known proxies for biomass burning in Greenland ice cores, especially formate (Legrand and De Angelis, 1996; Legrand et al., 2016). PC3 contains high loadings of rBC and ammonium, contributing to 15 % of the total variance and forming another group of biomass burning proxies. It is interesting to note that the preindustrial LF rBC record shows the highest correlation coefficient with ammonium independent of the resolution (significant at the 0.01 level). Most of the rBC peaks correspond to ammonium peaks (see below), reflecting a similar sensitivity to biomass burning emissions, transport, and deposition. We therefore suggest that ammonium is not only a proxy for Eurasian biogenic emissions as stated by Wendl et al. (2015) but also reflects a contribution from biomass burning, with biogenic emissions driving background variations while sharp peak events are associated with forest fires. Lastly, PC4 shows high loadings of specific organic markers of biomass burning (VA and p-HBA) and explains around 13 % of the total variance. The fact that biomass burning proxies are split into three different groups highlights their different behaviors towards the processes driving biomass burning emissions, transport, and deposition to and conservation within the snowpack, thus advocating for a multi-proxy reconstruction. These discrepancies can arise from the nature of the fire event (flaming or smoldering), the kind of vegetation burnt (grass, conifers, deciduous trees), or the sensitivity towards water percolation induced by summer melting. For instance, BC is preferentially emitted by flaming fires whereas smoldering fires are dominant in boreal regions (Legrand et al., 2016). VA is mainly produced by incomplete combustion of conifers while p-HBA is dominant for grass burning (Simoneit, 2002). Contrary to the other proxies used in this study, secondary production of formate from formaldehyde and numerous volatile organic compounds including alkenes is possible when the fire plume ages (see Fig. 1 in the review from Legrand et al., 2016). Formate, like other light carboxylic acids, can also undergo post-depositional effects such as revolatilization and diffusion in the snowpack (De Angelis and Legrand, 1995). Ammonium is almost not affected by summer melting and remains well preserved in the snowpack (Pohjola et al., 2002) while VA and p-HBA have been shown to be fully eluted by meltwater in an ice core from the Swiss Alps (Müller-Tautges et al., 2016).

These discrepancies are also reflected in the diverging long-term trends. rBC (Fig. 7a) and ammonium (Fig. 7b) display a relatively flat background over the preindustrial time period, meaning that they do not indicate any significant change in biogenic emissions and biomass burning. Only slight increases in background concentrations are visible around 1370 and 1545 for ammonium and around 1290–1340, 1470, 1545–1565, and after 1750 (possibly already influenced by anthropogenic emissions) for rBC. Conversely, formate, VA, and p-HBA (Fig. 7c to e) show more pronounced long-term variations. Elevated VA concentrations before 1400 (especially around 1250–1280 and 1360–1390) and three multi-decadal peaks in p-HBA around 1250–1280, 1520–1570, and 1610–1640 were attributed by Grieman et al. (2018) to North Atlantic Oscillation (NAO) changes, a positive mode implying a decrease in precipitation over Europe and central Asia believed responsible for enhanced biomass burning. The LF formate record displays the closest similarities with the p-HBA record. Three multi-decadal periods of elevated concentrations are also visible around 1260–1280, 1480–1560, and 1620–1650, superimposed on a general decreasing trend in background concentrations throughout the Little Ice Age, as observed for the VA and p-HBA (Grieman et al., 2018), coinciding with a decrease in temperature deduced from the LF δ18O record until 1880 (Divine et al., 2011). The timing of elevated background concentrations is in agreement with other ice-core-based studies of biomass burning from Siberia (Eichler et al., 2011; Grieman et al., 2017), thus supporting the attribution of northern Eurasia as a major source region for the LF site. Eichler et al. (2011) found prominent peaks in nitrate, potassium, and charcoal in the Belukha ice core, Siberian Altai, between 1600 and 1680 induced by the strongest regional forest fire episodes of the last 750 years following extremely dry conditions in central Asia around 1540–1600 responsible for dry dead wood accumulation. In the Akademii Nauk ice core, Siberian Arctic, Grieman et al. (2017) reported higher concentrations of VA and p-HBA for the time periods 1460–1660 and 1460–1540, respectively. Conversely, paleofire trends from the LF core notably differ from Greenland paleofire reconstructions that record low fire activity between 1600 and 1800 (Legrand et al., 2016).

The frequency counting of forest fire episodes enables to focus on episodic biomass burning plumes reaching Svalbard and can provide complementary information in addition to long-term variations. For rBC and ammonium, the centennial frequency of biomass burning episodes did not show a systematic trend throughout the Little Ice Age but showed fewer biomass burning episodes in the 13th and 15th centuries and more in the 18th century, despite colder temperatures, though for the latter a contribution of early anthropogenic emissions cannot be fully excluded. The VA, p-HBA, and formate records provide complementary information as increases in the frequency are concomitant with increases in background concentrations, meaning that fire episodes are not just stochastic events but are associated with longer-term regional climate variations. This is the case in the 17th century for all of the three species and also in the 13th century for VA. Table 3 shows the number of peaks matching (within ±1 year due to the different sampling) among the different proxies, associated with the percentage it represents in relation to the total number of peaks for each proxy. The best agreement is obtained between rBC and ammonium, with 21 peaks matching, which corresponds to 60 % (42 %) of the total number of ammonium (rBC) peaks. However, substantial dating uncertainties (Wendl et al., 2015) have to be considered in the bottom part of the LF-09 ice core, preventing us from assigning a specific biomass burning episode to every single peak.

Interestingly, severe droughts were reported over central Europe in 1540 (Wetter et al., 2014) and over northern central Europe between 1437 and 1473 (Cook et al., 2015). The case of the 1797 peak in the rBC and VA records is also remarkable as outstanding values in various biomass burning proxies were detected in several ice cores from Greenland during the last decade of the 18th century. In the NEEM ice core, rBC and levoglucosan were greatly enhanced between 1787 and 1791 (Sigl et al., 2013; Zennaro et al., 2014), while a very strong peak was visible in 1794 in the D4 ice core (McConnell et al., 2007) and in 1799 in the Summit 2010 ice core (Keegan et al., 2014). Ammonium records from the ice cores mentioned above all showed peak values in the same decade (Legrand et al., 2016), supporting the fact that this period of enhanced biomass burning could originate from the same decadal-scale climatic event. Severe drought conditions prevailed in central Asia during this decade due to South Asian monsoon failure (Cook et al., 2010). According to the dust proxy records from the Dasuopu ice core (Tibet), this decade experienced the driest conditions of the last millennium for this part of the globe (Thompson et al., 2000). In the Altai region, 9 out of 10 years between 1783 and 1792 belonged to the 10 % of the coldest years of the time period 1200–1850, while the following summers between 1793 and 1811 were clearly warmer (Büntgen et al., 2016). Those cold and dry conditions likely promoted dry dead wood accumulation, which then facilitated fire spread when temperatures rose later in the 1790s, a situation in agreement with the findings from Eichler et al. (2011). In Fig. 8 an exhaustive comparison is made between enhanced background concentration periods and peak years versus summer temperature anomalies and drought reconstructions from two regions of northern Eurasia for which datasets are available, namely northern Europe (Cook et al., 2015; Esper et al., 2014) and the Altai (Büntgen et al., 2016; Cook et al., 2010). It appears that biomass burning episodes frequently occurred in concert with decadal-scale summer temperature increases. Conversely, the link with moisture variations seems less consistent as conditions were either drier or wetter than average depending on the period, suggesting that summer temperature is the controlling factor for biomass burning activity in these regions. While dry conditions can lead to dead fuel accumulation (Eichler et al., 2011), wet conditions promote biomass productivity, especially for grasslands (Pederson et al., 2014). Both of these different mechanisms can enhance fire severity in the context of decadal-scale temperature increases.