Study sites
The three study sites are situated in the Maurienne valley that borders the Vanoise massif to the south, in western France. The three sites, namely St Michel, Aussois, and Lanslevillard, are situated at the same latitude (ca 45.25°N) along a west–east transect of ca 40 km corresponding to a gradient of humidity linked to a complex rain-shadow process. St Michel is the most western site, 21 km from Aussois, the central site, and Lanslevillard is the easternmost of the three – 17 km from Aussois (geographic coordinates: Table 1).
St Michel is characterized by south-facing slopes and acidic soils on bedrock composed of glacial thin deposits of admixed coalesced sandstones and shale rock fragments (Table 1). Aussois is also on south-facing slopes but has alkaline bedrocks mostly composed of thick glacial deposits (>150 cm): an admixture of dolomitic-origin, cargneule, and gypsum. Lanslevillard is on north-facing slopes but with alkaline bedrock resulting from a glacial admixture of marbles and calcareous shale fragments (Geologic maps 775 and 776). At St Michel and Lanslevillard, forest soils are acidic or neutral brunisol, respectively, whereas at Aussois soils are calcimorph with pH >7 (Bartoli 1967; Mourier et al. 2008). Slope degrees range between 20 and 30° (Table 1), but if St Michel and Lanslevillard show regular slopes, Aussois presents more heterogeneous terrain.
Climate features vary between sites according to slope exposure and altitude of weather stations: ca 1400 m above sea level (hereafter asl) at St Michel and Aussois, and 2000 m asl at Lanslevillard, resulting in temperature values that are not totally comparable (Table 1). However, mean annual temperatures appear similar (ca 7°C at 1400 m on south-facing slopes and ca 2.5°C at 2000 m on north-facing). The mean monthly temperatures for the three winter months (TDJF) are around 0°C at 1400 m and -4°C at 2000 m, and the three of the growing season (TJJA) are ca 14.5°C at 1400 m and 10°C at 2000 m. Snow covers the ground for 3–5 months per year at 2000 m on southern slopes and 5–7 months on north-facing slopes (Table 1). Mean annual precipitation ranges between ca 900–950 mm at St Michel and Lanslevillard, and ca 700 at Aussois. Aussois is clearly the driest site, with only ca 160 mm cumulated during the growing season, while both other sites receive 200–250 mm, i.e. +40% (Table 1).
To sum up, Aussois is the most stressed site for biological processes in terms of climate and soil, lowest precipitation, south-facing slopes (high solar radiation on soil increasing the effect of air temperature), alkaline and calcimorph deep morainic soil depleted in clay (and thus poor water storage). Lanslevillard is the coldest site with a shorter growing season, and St Michel is the site with the most acidic soil features.
Soil sampling and charcoal extraction
At St Michel, ten soil profiles were collected, eight at Aussois, and six at Lanslevillard. All sampling plots were selected in the subalpine belt (1700–2500 m asl) in Aussois and St Michel, whereas at Lanslevillard the sampling plots were selected between 1600 and 2300 m asl to take into account the slope exposure that triggers a lowering of about 200 m of vegetation belt limits between south- and north-facing slopes (less solar radiation, shorter growing season, colder air and soil temperatures, longer snow cover in spring). The mean (±SE) sampling altitudes are 2109 ±87 m at St Michel, 2051 ±80 m at Aussois, and 1970 ±79 at Lanslevillard.
Soils were collected according to the standardized pedoanthracological method (Carcaillet and Thinon 1996; Carcaillet and Talon 2001). Plots disturbed by human settlements were avoided, as well as soil located at the foot of steep or long slopes susceptible to concentrated charcoal transported by run-off (Talon et al. 2005), or eroded or hydromorphic soils poorly favorable to particulate sequestration. Soil profiles were sampled in trenches dug down to the bedrock whenever possible, or to the limits of plant roots on thick morainic deposits. Soil material was sampled by cutting soil blocks from the face of a vertical profile. Blocks were collected at different depths from the bottom to the surface to avoid particles falling from the upper horizons. Block limits were separated in several layers when the horizon thicknesses were greater than 20 cm or when the soil horizon limits were not visible. About ten to 15 liters of dry fine material were sampled per layers of ca 20 cm. Each complete dried profile weighed 30 to 100 kg depending on the soil texture, the organic content, and the horizon structures, and the depth of the profiles that determine the number of layers, from one to seven for 30 and 150 cm thick profiles, respectively.
A flotation procedure with an ascending water flow was used to extract charcoal from the soil matrix (Carcaillet and Thinon 1996). Flotation with wet sieving followed by manual sorting under a binocular microscope, allowed the final separation of charcoal fragments from other soil particles. Only fragments larger than 0.4 mm in diameter were considered. Charcoal that is smaller than 0.4 mm may be transported over long distances, while the stand origin of charcoal larger than 0.4 mm is unequivocal (Clark et al. 1998; Ohlson and Tryterud 2000; Lynch et al. 2004).
Dating and biomass burning chronology
In total, 93 AMS 14C datings were obtained from soils at St Michel (43), Aussois (26), and Lanslevillard (24). Thirty-four (34) out of these 93 AMS 14C datings were already published, 18 at St Michel and 16 at Aussois (Carcaillet 1998). Because in this preliminary study of 1988 (i) no dates covered the recent period from 2200 calendar years before the present (hereafter, cal BP) above 2000 m asl at Aussois and above 1700 m asl at St Michel, and (ii) none was obtained before 6500 cal BP (Carcaillet 1998), the strategy was to increase significantly the number of datings to fill in these gaps in the chronologies, or to strengthen the already reported pattern. The point was to question whether these chronological gaps resulted in a problem of low numbers or to see if they correspond to an actual lack of burning in some sites at certain altitudes. The additional site (Lanslevillard) allows us to test whether these reported chronologies per altitude class were strictly local/artifactual or corresponded to a general pattern. In this region of the western Alps, no accurate sedimentary charcoal series are available below 2000 m asl (Leys and Carcaillet 2016), which prevents us formulating a response to this issue of fire patterns according to altitude. We assume that fire occurrences should be more frequent at lower altitudes and on south-facing slopes compared to higher altitudes and north-facing slopes due to greater dryness (precipitation increases with elevation and south-facing slopes receive more solar radiation) (Schumacher and Bugmann 2006), and posit their greater attractiveness to prehistoric humans who altered fire patterns in response to the drivers prevailing thousands of years ago, e.g., climate, vegetation, topography, soil, etc. (Leys and Carcaillet 2016; Vannière et al. 2016).
With a few exceptions (Table S1), each dating was measured on one charcoal fragment. Prior to AMS dating, charcoal fragments were cleaned using an ultrasonic wave generator under a microscope (×40) to remove small roots, fungi hyphae, and particles that could potentially alter the dating result. Further, each fragment was processed with a solution of Na4P2O7 or NaP2O4 to extract the organic compounds adsorbed in the charcoal porosity. The solution was changed daily up to the release end of organic compounds by the charcoal. Generally, five to seven days were needed to chemically clean the charcoal fragments (Carcaillet and Thinon 1996).
The 14C measurements were processed by several laboratories: the NSF Arizona Facility Laboratory (Arizona, USA), the Centre de Datation par le Radiocarbone (France), and the LMC14 (a national facility lab, also in France). Dating codes were respectively labeled AA-xxxx, Lyon(OxA)-xxxx, and SacA-xxxx (Table S1).
The radiocarbon measurements were calibrated against dendrochronological years using the CALIB program version 8.2, based on the IntCal20 dataset (Stuiver et al. 2021), and reported as intercept with 2 sigma and a probability range of 1.0. The probability distributions of 14C datings were cumulated by site (e.g., Hajdas et al. 2007) and labeled local chronology. The probability distributions were obtained using the sum probability function of the CALIB program. Contrary to some assumptions (e.g., Payette et al. 2012; Couillard et al. 2021), such distribution cannot provide a chronology of fire events, but rather a chronology of biomass burned. Important lags between the 14C date (radiocarbon age of wood) and the age of fire can exist, notably if the charcoal fragments used for dating came from the inner part of trunks for long-lived species such as those growing in subalpine forests (Carcaillet 1998). To bypass this problem – which Gavin (2001) called the ‘inbuilt age’ – and in the absence of a fuel load/consumption model to fix it (Gavin 2001), charcoal fragments corresponding to twigs were here selected for dating, when possible to reduce the range of the so-called inbuilt age. However, a maximum of cumulated calibration does not correspond to a fire event but to a stochastic period of burned biomass characterized by a concentration of 14C dating and associated probability of calibration.
The statistical analyses of radiocarbon dating were based on their distribution revealed through violin-plots (R package, ggplot2). The violin-plot is more informative than the box-plots sometimes used to establish dating distribution (e.g., Gavin et al. 2003), because a multimodal distribution is visual and illustrated by two bulges at least, whereas a unimodal distribution should present only one. The site distributions were statistically compared to each other based on ANOVA followed by Tukey HSD post-hoc test to determine whether differences occur between pairs of sites. Calendar dates cannot be directly used because a date corresponds to a complex range of values, sometimes to several ranges associated with different probabilities within the full calibration range of the date considered. Consequently, the mode of each calendar date was extracted from CALIB and used as representative of the dating for violin-plots and ANOVA. The mode value divides the graph of the density function into two parts with equal probability, i.e. p = 0.5 (Michczyński 2007). This process is not perfect but better than to use the median (Michczyński 2007), which does not consider the probability variability within this calibration range (Telford et al. 2004). The mode (and the median) does allow for a unique value per date that is needed for the ANOVA.
Charcoal botanical identification and analysis
Extracted charcoal fragments were observed under an incident light microscope (×200, ×500, ×1000) to identify the woody taxa that burned. Identifications were realized with the help of specialist reference wood anatomy books (e.g., Jacquiot 1955; Jacquiot et al. 1973; Schweingruber 1990), and some specific studies on wood charcoal anatomy notably for Larix and Picea, for Betula, and for Empetraceae/Ericaceae. Because Larix and Picea can only be distinguished from each other in exceptional cases, they have been gathered under the taxon “type Larix-Picea” (Talon 1997). Betula trees can be distinguished from Betula shrubs (Hellberg and Carcaillet 2003). With regard to Ericaceae and Empetraceae, genera or species have been identified when possible based on the anatomy of the medulla: Arctostaphylos, Empetrum, Erica, Rhododendron, Vaccinium myrtillus, V. vitis-idaea, V. uliginosum (Talon 2004). Further, Pinus sylvestris (a mountain tree, growing up to 2000 m) and P. uncinata (a true subalpine tree growing from 1500 to 2500 m asl) cannot be distinguished based on their wood anatomy, and are thus labeled P. sylvestris/uncinata. The assemblages of charcoal are considered to correspond mainly to dead woody debris, especially that lying on the ground, and to living twigs of trees and shrubs which can easily burn due to their small diameters or their low height in the vegetation (Bégin and Marguerie 2002; Kasin et al. 2017).
Charcoal was weighed and expressed in terms of charcoal mass (= anthracomass) relative to the dry soil mass (mgCHAR.kgsoil-1 or ppm; see Carcaillet and Talon 2001). The charcoal mass was calculated per profile and per layer. The representation of past burned biomass based on anthracomass presents distortion potential due to the different consumption of species (Fréjaville et al. 2013a), burning conditions (Asgough et al. 2011), and the assumed resistance of charcoal to weathering processes in soil depending on soil type, burning conditions, and plant anatomy (Preston and Schmidt 2006). However, the representation of assemblages based on charcoal mass still remains an efficient way to figure out what the plant biomass had been at the time of the burning events.
Charcoal assemblages were expressed as a percentage of the identified charcoal mass per taxon and per layer; unidentified charcoal fragments were not considered in this calculation. Unidentified charcoals were those fragments with too few anatomical features for definitive identification, or glassy charcoals whose provenances still remain unknown despite experimentations (Théry-Parizot 2002). Taxa assemblages per layer were used for Correspondence Analysis (CA) to statistically distinguish the main organization of burned woody assemblages, and notably to test whether assemblages are distributed according to sites and to elevation. CA were computed using FactoMineR of the R package (Lê et al. 2008). For synthetic representation, charcoal concentrations per layers were averaged per taxon and per profile to calculate the composition gradient along the elevation. This averaging was possible because each charcoal layer is not stratified due to soil bioturbation (Carcaillet 2001). The layers from a given profile were thus considered as replicates for that profile. For that purpose, infrequent or poorly abundant taxa were clustered, as in “broadleaved trees” for Acer, Alnus cf. incana, Betula, Corylus, Fraxinus, cf. Populus; “Ericaceae” for Arctostaphylos, Erica, Rhododendron, Vaccinium; “other shrub” for Alnus cf. viridis, Berberis, Juniperus, cf. Ononis, Rhamnus, Rosa, Rosaceae Maloideae, and cf. Salix; Clematis alpina, the only liana found, was included with “other shrubs” because it generally spread in the shrub layer. The full list of taxa identified per profile and per layer is in the supplementary information (Table S2).