Climatic influence upon early to mid‐Holocene fire regimes within temperate woodlands: a multi‐proxy reconstruction from the New Forest, southern England

A combined pollen, charcoal and climatic record is presented from Cranes Moor, southern England, covering the period c. 10 500–5850 cal a BP. It is shown that the occurrence of burning is closely related to natural processes, including prevailing climatic conditions and vegetation composition. These burning events are often linked to an increase in the summer moisture deficit, implying that the timing of burning events is linked to periods of warmer/drier climate during the Holocene Thermal Maximum (c. 11 000–5000 cal a BP). These events play an important role in the vegetation composition and succession around the site. The nature of the burning recorded at the site shows strong similarities with other records from northern Europe. This study throws caution on suggestions that fire in the Holocene record from areas such as the British Isles is linked only to human activity, and enhances the possibility that natural fire incidence played an important role in natural woodland structure dynamics.


Introduction
The occurrence of burning during the early to middle Holocene is widely recognized yet interpretations of its causes are still heavily debated. In contrast to studies from North America where a fire-climate relationship is widely discussed (e.g. Long et al., 1998;Carcaillet and Richard, 2000;Millsaugh et al., 2000;Paduano et al., 2003;Marlon et al., 2006;Whitlock et al., 2007), few European studies make the same connection, and human activity is often proposed as the primary cause of fire. However, a series of recent studies encompassing a range of different ecosystems (e.g. Power et al., 2008;Olsson et al., 2010;Rius et al., 2012;Marlon et al., 2013) have suggested a climatic control during the early to middle Holocene within a European context. This is in contrast to the view, notably among archaeologists from Britain, that burning is largely anthropogenic in origin. In this region evidence of Mesolithic activity is frequently found together with evidence for burning; for example, at lowland sites such as Thatcham (Barnett, 2009), Star Carr (Mellars and Dark, 1998), Three Ways Wharf (Grant et al., 2014) and along the Severn Estuary (Bell et al., 2002), and at upland sites such as the North York Moors (Innes and Simmons, 2000;Innes et al., 2004), Dartmoor (Blackford et al., 2006) and Waun-Fignen-Felen (Smith and Cloutman, 1988; Barton et al., 1995). Consequently, natural causes of ignition are often downplayed or dismissed (e.g. Rackham, 1980;Simmons, 1996;Barnett, 2009). However, because these study sites are intrinsically linked to local Mesolithic human occupation, and burning of wetland and woodland edge environments, they may not be representative of the wider landscape processes. Brown (1997), Moore (2000) and Tipping (1996) have cautioned that evidence for anthropogenic impact requires a more balanced understanding of the role of other natural factors and ecological processes.
The prevalence of the anthropogenic burning view in Britain arises in part because of the assumption that Holocene woodland is non-flammable and natural sources of ignition (e.g. lightning) are too sparse temporally and spatially. However, these assumptions do not take full account of past climatic conditions, changes in woodland composition or indeed woodland structure (see Clifford and Booth, 2013). For example, the climate of the early Holocene in Northwest Europe differed substantially from modern conditions. Across Northwest Europe large terrestrial landscapes were exposed by reduced sea levels (e.g. Doggerbank; Shennan et al., 2000) and coupled with the precession-driven insolation peak at 10 ka BP, much of the region experienced more continental climatic conditions than present (Briffa and Atkinson, 1997). This regime changed substantially throughout the early to middle Holocene with the flooding of the southern North Sea basin by c. 7.8 ka BP when a more maritime climate became established. The Holocene climatic regime also became more stable with the end of the major meltwater pulses into the North Atlantic by c. 6.8 ka BP (Carlson et al., 2008;Tö rnqvist and Hijma, 2012).
Discussions over the possible structure of mid-Holocene temperate woodland of Northwest Europe have often suggested fire played some role in natural woodland dynamics (e.g. Svenning, 2002;Bradshaw and Hannon, 2004;Hodder et al., 2005), such as the maintenance of light-demanding or short-stature woody species within closed forests (Svenning, 2002). Similarly, investigations in Ireland have suggested an extensive role for fire in maintaining Pinus sylvestris within primarily deciduous forest (Bradshaw and Browne, 1987;Dodson and Bradshaw, 1987;Bradshaw, 1993). Studies from Scandinavia (Lindbladh and Bradshaw, 1998;Lindbladh et al., 2000;Bradshaw and Hannon, 2004) have also suggested that fire may have been an important natural disturbance factor.
Given the debate outlined above, there is a need to better understand the baseline interactions between vegetation, fire and climate. The early-to mid-Holocene provides a suitable study period to explore how vegetation communities respond to burning and climate change over millennial timescales because of its high climate variability. This will provide a basis for understanding the drivers of long-term vegetation stability and change. A site in southern England was selected because this region is situated in a particularly sensitive area for registering North Atlantic-wide climate changes, as well as important regional changes driven by the reorganization of the palaeogeography of the Northwest European coastal zone. This paper presents parallel plant macrofossil, pollen and charcoal archives, preserved within a single core from a mire in southern England. This approach allows direct comparisons to be made between the records, eliminating the problem of comparison between proxies when derived from different locations (Charman et al., 2006).

Study Site
Cranes Moor forms part of one of the largest mire systems in southern England, located on the western edge of the New Forest (50.824˚N, 1.726˚W; NGR SU194028) (Fig. 1). The local geological setting is largely composed of Palaeogene Chama and Becton Sand Formations, which display strong podsol development under extensive areas of surrounding heathland. Local areas of woodland on the heath are dominated by Pinus sylvestris originating from 19th-century plantations, with Betula pendula forming some localized scrub. Pteridium aquilinum, Ulex europaeus and Erica cinerea are confined to the dry heath, whereas Calluna vulgaris, Erica tetralix and occasionally Ulex minor are also found in the mire communities. The current mire vegetation comprises peripheral wet heath, flushed areas along the northern and southern margins with Schoenus nigricans, and a well-developed Sphagnum lawn across the central area (see Newbould, 1960) containing large regularly spaced pools (up to 5 Â 8 m), the latter being the result of past peat cutting. The site is also located in an area with limited evidence of Mesolithic occupation.
Cranes Moor was selected for this study because it contains the deepest Sphagnum peat deposits in the New Forest and began accumulating in the early Holocene, unlike most other Northwest European mires, which were in an early successional swamp or sedge fen stage at this time (Hughes et al., 2000). Detailed stratigraphy by Newbould (1953) identified deep, Sphagnum-dominated peats, focused predominantly towards the eastern side of Cranes Moor, which shallow towards the western margins where they are replaced by deeper reedswamp peat to the margins along zones of flushing. The presence of extensive oligotrophic Sphagnum deposits lying between two zones of flushing and immediately overlying the sand geology (Newbould, 1953) suggests that water movement through the peat strata was minimal in the eastern and central part of the bog from the early Holocene onwards. Although recent peat cutting has removed evidence of the surface profile of the bog, the deeper stratigraphy suggests and that the surface vegetation was likely to have been largely or solely ombrogenous from the early Holocene. Under these conditions the peat archive may contain a sensitive indicator of palaeoclimatic change sensu Barber (1981).

Lithostratigraphy and chronology
Following the results of Newbould's (1953) stratigraphic survey, subsequent work has focused upon the eastern part of the site (Barber and Clarke, 1987;Grant, 2005). In this study a core was sampled from the location that contained the deepest (and thickest) Sphagnum-peat deposits and multiple visible pool layers. The upper 40 cm of peat was sampled using a monolith tin, with lower levels recovered using a modified Russian corer (Barber, 1984). Grant et al. (2009) have shown that there is a clear break in this sequence above 80 cm, the result of past peat extraction from the site (Barber and Clarke, 1987). Consequently, results from the upper 80 cm of this sequence are omitted in this paper. A chronological framework for the sequence is provided by seven accelerator mass spectrometry radiocarbon measurements on Sphagnum samples (Table 1). Radiocarbon dates were calibrated against the IntCal09 dataset (Reimer et al., 2009) and age-depth modelled using CLAM software, with linear interpolation between the weighted averages of each calibrated date (Blaauw, 2010) (Fig. 2). The radiocarbon dates show that the sequence spans c. 10 500-5850 cal a BP.     Radiocarbon date SUERC-5246 is clearly anomalous because it represents a clear outlier in the age-depth model.

Pollen analysis
Sub-samples for pollen analysis were processed using standard techniques (Moore et al., 1991), with Lycopodium tablets added to enable the calculation of pollen and microcharcoal concentrations (Stockmarr, 1971). The pollen sum was standardized between depth levels using a minimum count of 400 total land pollen (TLP) grains, excluding aquatics and pteridophytes. Zonation was undertaken in Psimpoll (version 4.25;Bennett, 1992), using the method 'Optimal Splitting by Information Content'. Five local pollen assemblage zones (LPAZs) were designated and they are shown in Fig. 3 and described in Table 2. Pollen nomenclature follows Bennett (1994) and Bennett et al. (1994), except for Ericaceae which follows Moore et al. (1991), with plant nomenclature following Stace (1997).

Charcoal analysis
Microscopic charcoal was quantified on the pollen slides using a method adapted from Clark (1982) whereby a minimum of 200 random fields of view were applied to a slide, and the number, and size, of each charcoal particle observed recorded. The number of Lycopodium spores observed in each field of view was also recorded, with a minimum of 50 spores counted from each level. Both area and abundance of charcoal were measured for the microscopic charcoal, with the correlation between these two parameters high (Pearson's r ¼ 0.933). Macroscopic charcoal (>125 mm) was prepared and counted following a modified method from Rhodes (1998), using 1 cm 3 of sediment. To estimate the number of fire events, the charcoal accumulation rate (CHAR) was calculated using the computer program CharAnalysis (Higuera et al., 2008) for both the micro-and the macroscopic charcoal size fractions. This program was originally developed for modelling of macroscopic charcoal in lake sediments in North America. However, modelling of the results of both size fractions, to compare local and extra-local/regional burning patterns, was deemed appropriate to identify changes in fire occurrence and allow comparison with vegetation change within a wider catchment area. CharAnalysis interpolates the actual charcoal counts, sample volume and sample depths by using the median sample  (2014) resolution (calculated as 29 years) before calculating CHAR. No transformations were used. CharAnalysis divides CHAR into two components: C background and C peak . The background component illustrates low-frequency trends reflecting changes in the rates of charcoal production and transportation (Higuera et al., 2008). It also correlates well with long-distance transport from the entire charcoal source area (Higuera et al., 2007). The C peak component identifies the highfrequency variations representing local fire events or episodes in the source area and may correspond to more than one fire.
C background was estimated using a locally weighted Lowess smoother, robust to outliers. The selected smoothing window was calculated based upon a test between the empirical C noise values and the modelled C noise distribution, which provides an index of how well the C noise model fits the empirical data (Higuera et al., 2009). A comparison of both the signal-tonoise index (SNI) and the noise distribution goodness-of-fit (GOF) against different smoothing window widths (in 100year intervals) identified that a 600-year smoothing window was most appropriate for the two datasets. The C peak component was calculated as residuals, i.e. C background subtracted from C interpolated . The threshold value separating fire-related from non-fire-related variability in the peak component was set at the 99th percentile of a Gaussian mixture model. Identified fire episodes are marked with closed circles, with fire return interval (FRI, number of years between fire events) and fire frequency (number of fires 1000 a À1 ) plotted as a function of time (Fig. 4). The results of the charcoal analyses are shown in Figs 3 and 4 and described in Table 2.

Correlation analysis
Correlation coefficients were calculated to investigate whether a relationship existed between the pollen and charcoal abundances (microscopic and macroscopic). Pollen taxa were included where pollen percentages exceeded 1% TLP and were present at a minimum of five occurrences within the full dataset/subgroup upon which correlation analysis was being undertaken. A t-test was used to determine whether correlation coefficients r are significantly different from 0 (r 6 ¼ 0, P ¼ 0.05, two-sided, Bahrenbeg et al., 1985; those significantly different are identified on Fig. 5 with close circles). Analysis was undertaken for the whole dataset, individual LPAZs and amalgamated zones reflecting the main vegetation phases ( Fig. 5; Supplementary Table S1).

Plant macrofossil analysis
Samples for plant macrofossil analysis were prepared following the methods of Barber et al. (1994). Peat slices were sieved through a standard 5 L of water using a 125-mm sieve. The unstained plant remains were quantified using the quadrat and leaf count method (QLC) for the main vegetative components of the peat. Fruits and seeds were quantified using a five-point scale of abundance where 1 ¼ rare, 2 ¼ occasional, 3 ¼ frequent, 4 ¼ very frequent and 5 ¼ abundant (Walker and Walker, 1961). Only selected macrofossils are displayed in Fig. 6 for clarity. Plant remains were identified with reference to the extensive modern type collections held in the Palaeoecology Laboratory at Southampton and with reference to Daniels and Eddy (1990) for Sphagnum mosses, Smith (2004) for other bryophytes and Grosse-Brauckmann (1968, 1972 and Katz et al. (1977) for the vegetative remains of vascular plants. Nomenclature follows Stace (1997) for vascular plants and Smith (2004) for bryophytes. Local macrofossil assemblage zones (LMAZs) were designated using CONISS and are described in Table 3. The plant macrofossil data were analysed using detrended correspondence analysis (DCA;ter Braak, 1988) in the program CANOCO (version 4.5) to investigate latent environmental gradients. Only the main constituents of the peat matrix, quantified using the QLC system, were included in the ordination with downweighting of rare species. The Axis 1 score was found to contain a strong hydrological gradient (Fig. 7a). The plant macrofossil results were used to reconstruct bog surface wetness (BSW; see Daley and Barber, 2012) using the indicator-weighted hydroclimatic index (IHCI) devised by Mallon (2012; modified from Dupont, 1986) (Fig. 7b).

Plant macrofossils
Cranes Moor is one of the few sites in lowland Britain that presents relatively few difficulties for the reconstruction of palaeoclimate records from the surviving peat deposits. The mire complex has developed on base-poor sands in the lee of a sand ridge (Barber and Clarke, 1987) where the area named 'Sphagnum Bog' is protected from flush lines that flow to the north and the south-west of the coring site (Fig. 1). The area of 'Sphagnum Bog' investigated dates from c. 10 500 cal a BP, with the basal deposits indicative of an oligotrophic Sphagnum-dominated community composed principally of S. papillosum and members of the Acutifolia section, together with small quantities of S. austinii and S. tenellum. This basal peat stratum may not have been entirely protected from flushing as the macrofossil assemblage contains traces of S. palustre, Phragmites australis leaves and Juncus species seeds.
Most traces of poor fen species disappear from the record above 320 cm (c. 9550 cal a BP). The establishment of strongly oligotrophic conditions means that the site became isolated from the surrounding seepages and it is highly probable that this area of the mire became an ombrotrophic centre. In such bogs the water table can be principally related to the length and severity of the summer water deficit (Charman, 2007;Charman et al., 2009). This deficit is a function of the balance between precipitation and evapotranspiration. BSW, as reflected in plant macrofossil assemblages and resultant IHCI (Fig. 7b), can therefore provide a reconstruction of past surface water balance conditions, which is closely related to the atmospheric water balance, in the absence of major human impacts to the bog or catchment.
Phases of pool development can be detected in the macrofossil record commencing at c 9500 (but see above), 7500 and 6400-5900 cal a BP. The wet phase recorded at 7500 cal a BP parallels an increase in wetness observed far to the north in the raised bogs at Bolton Fell Moss (Barber et al., 2003) and Walton Moss (Hughes et al., 2000). These three phases also correspond to known humid episodes in other palaeoclimatic records, such as continental lake level data (e.g. Magny et al., 2003) and maritime glacier advances (Nesje et al., 2000), although note that these types of records are likely to be strongly related to winter precipitation rather than summer water deficit. The well-known climatic shift at c. 8200 cal a BP (Alley et al., 1997;Magny et al., 2003;Baker, 2012) is clearly expressed here by a wet-shift just before 8200 cal a BP, a dry episode at 8200 cal a BP and a definite wetting trend commencing c. 7800 cal a BP. This agrees with other palaeoclimate records from the UK which show a cool, dry climatic anomaly in the region (e.g. Rousseau et al., 1998;Lang et al., 2010), and is clearly recorded in the GISP2 oxygen isotope record (Alley et al., 1997). The broad agreement of the Cranes Moor BSW data with other regional climatic proxies implies that the mire is climatically sensitive through this period.

Interpretation of the pollen and charcoal sequences
Corylus avellana-type and Pinus sylvestris are dominant between c. 10 500 and c. 9650 cal a BP, with pollen percentages closely inter-linked. This may be a result of interdependence within the pollen sum, but also an effect of local burning events, as shown through a statistically significant negative correlation of macroscopic charcoal with P.   Table S1 for tubulized results. sylvestris and positive correlation with C. avellana-type (Fig. 5). The first fire events recorded, at 10 260 and 10 180 cal a BP for microscopic charcoal and 10 230 cal a BP for macroscopic CHAR, suggest that these events are likely to be broadly contemporary with each other. Fire event(s) coincide with reducing P. sylvestris values and small increases in Quercus, Ulmus, Calluna vulgaris and Pteridium aquilinum. Intermittent increases in P. aquilinum are also found associated with the charcoal record, although there is no overall consistent statistical correlation in LPAZ CRMp-1. However, it is possible that burning may have led to increased openness and some partial canopy reduction, leading to understorey development, such as a field layer of P. aquilinum and increased flowering/re-sprouting of C. avellana-type. A notable reduction in Pinus sylvestris (30-10%) occurs at the LPAZ CRMp-1/2 boundary at c. 9650 cal a BP. This is coupled with an increase in Quercus and Betula and a reduction in Ulmus, implying a notable change in the surrounding woodland. FRI is notably longer, extending from 180 years in the previous zone to 340 years in this one. Statistically significant positive correlations are found between the pollen of Cyperaceae and Pteridium aquilinum with the microscopic charcoal, and Betula and Pteropsida (monolete) indet. with the macroscopic charcoal. This may reflect slightly different spatial source areas, with birch and ferns reflecting very localized burning events (the larger charcoal size fraction), while the changes in the sedges and bracken may be responses to burning occurring at a greater distance. There is also an increase in Calluna vulgaris, although this may be related to the onsite of mire vegetation. A low presence of Plantago lanceolata and P. aquilinum indicates some disturbance probably associated with areas of open ground after burning. The peak in pollen concentrations at the end of this zone might reflect a slow-down in peat accumulation or short-lived change in pollen input because of the changes in the local vegetation.
During LPAZ CRMp-3, Pinus sylvestris and Corlyus avellana-type decrease, with increases in Alnus glutinosa and Quercus; Quercus was found to have a significant negative correlation with microscopic charcoal. Increases in Pteridium aquilinum and Melampyrum coincide with peaks in both charcoal size fractions. Melampyrum has been found in several studies associated with burning, perhaps of the ground layer, in woodland or at woodland edges (Innes and Simmons, 1988;Caseldine and Hatton, 1993;Innes and Blackford, 2003). For LPAZ CRMp-3 there is no significant correlation between Melampyrum and charcoal concentrations. However, when the data from LPAZ CRMp-3 and -4 are combined (along with the total dataset) there is a strong significant positive correlation shown between Melampyrum and both charcoal fractions, representing a direct fire response. The peak in macroscopic charcoal at c. 8750 cal a BP is the largest recorded from this site, although there was no evidence from the plant macrofossil analyses (e.g. charred plant remains) suggesting burning at the sample location.
Another fire episode, at c. 8200 cal a BP, is associated with Melampyrum and immediately precedes the expansion of Alnus glutinosa at c. 8150 cal a BP. Although not statistically significant for LPAZ CRMp-4 alone, when the data from LPAZ  There is also a reduction in the microscopic CHAR FRI to 190 years coupled with charcoal values that show consistency in the size of the peaks recorded in both size fractions.

Interaction of fire and vegetation
The data presented here show that there are significant interactions between occurrence of fire and the composition of early and mid-Holocene coniferous and deciduous vegetation communities. Correlation coefficients show several positive correlations with taxa known to be fire-responsive (e.g. Melampyrum and Cypercaeae), with negative correlations with fire-sensitive species (such as Tilia and Fraxinus; Delarze et al., 1992). Kaltenrieder et al. (2010) also found correlations between fire-responsive and fire-sensitive species, although notably the Cranes Moor dataset differs with Alnus having a negative relationship and Corylus a positive one. Tinner and Lotter (2006) also found some statistically significant correlations between pollen and microscopic charcoal, such as positive relationships for Corylus and Urtica and negative for Ulmus. This led the authors to suggest low to moderate effects of fires on forest ecosystems and differences in taxa responses to the occurrence of fire (Tinner and Lotter, 2006, p. 534). Some of these differences may also be related to differences in the speed of first flowering in different taxa, with Corylus avellana occurring rapidly within a few years, whereas other arboreal taxa often take longer (e.g. Schü tt et al., 2006;Matthias et al., 2012;Waller et al., 2012). Strong correlations between the pollen and charcoal data had not been expected due to pollen being continuously produced from a large pollen source area (extra-local and regional pollen; Jacobson and Bradshaw, 1981), whereas burning is a stochastic event variable spatially and temporally, therefore possibly acting on different scales to the pollen source. This would result in pollen deriving from both affected and unaffected areas simultaneously. Moreover, the incorporation of charcoal into a basin, particularly a mire, could lead to dampening or amalgamation of charcoal peaks which could distort the record, particularly as it is known that charcoal deposition may peak several years after the maximum of forest fires (e.g. Whitlock and Millspaugh, 1996;Tinner et al., 1998).

Relationship between vegetation composition and occurrence of burning
The genus Pinus is inextricably linked over space and time with fire (Agee, 1998). Pinus sylvestris woodland has a moderate-severity fire regime, including a wider range of fire frequencies and less intense burning than the boreal pines of North America. This can range from an apparent absence of fire near the treeline, as shown from the Scandes Mountains of central Sweden (e.g. Kullman, 1986), to regular FRIs as low as 20 years in Siberia (Sannikov, 1985). Fire plays an important role in the maintenance and regeneration of P. sylvestris as it can consume the humus layer exposing mineral soil to allow the establishment of seedlings, which otherwise would be inhibited by the carpet of needles (Sannikov, 1983;Hille and den Ouden, 2004). Vera (2000) suggested that grazing was also important in the removal of the build-up of the ground litter layer, although studies have shown that a substantial increase in browsing can lead to increased seedling mortality through predation (e.g. Scott et al., 2000;Weber et al., 2008). The natural occurrence of small-scale fires in the litter layer would therefore provide a means by which P. sylvestris could actively regenerate during the early Holocene. In a study from Thorne and Hatfield Moors, Lincolnshire, Whitehouse (2000) found charcoal and pyrophilous insect fauna associated with repeated burning of Pinus during the mid-Holocene. Hodder et al. (2005) also demonstrated that mid-Holocene insect assemblages often contained the presence of pyrophilic species, implying that fire was an important process. Studies in Ireland have found that persistence of Pinus was closely correlated with the presence of charcoal (Bradshaw and Browne, 1987;Bradshaw, 1993), with the suggestion that a lessening fire regime led to a reduction in the size of Pinus populations. Similar results have been derived from Hockham Mere and Quidenham Mere, Norfolk , Cothill Fen, Oxfordshire (Day, 1991), and Pannel Bridge, Sussex (Grant and Waller, 2010), where increased quantities of microscopic charcoal were associated with the main Pinus phase.
It is notable in the Cranes Moor sequence that a succession from Pinus to Quercus coincides with a change in the FRI, extending from 180 to 340 years between LPAZ CRMp-1 and -2, and a notable increase in the size of CHAR peaks in both size fractions. This would imply a change from regular, lowintensity fires to less regular, possibly higher intensity ones. It is unclear whether this change is the cause of the reduction in Pinus or simply a consequence of it, with a notable high peak in the microscopic charcoal fraction at c. 9500 cal a BP, possibly the result of mortality in Pinus stands and a greater abundance of dead wood. Tinner and Lotter (2006) found that Pinus had significant correlation coefficients with charcoal at negative lags which they suggested could be vegetational changes preceding fires, supporting the idea of increases in fuel availability enabling initial burning events.
It is unlikely that the reduction in Pinus values to <20% TLP (the threshold commonly taken to imply local presence; Bennett, 1984) in LPAZ CRMp-2 indicates its local disappearance. Binney et al. (2005) showed from modern studies within areas of alder carr that local small stands of Pinus were only recorded at c. 3% TLP, which they attributed to its pollen representation being suppressed by the in situ presence of relatively high pollen-productive vegetation. In lowland southern Britain, Groves et al. (2012) found that Pinus was competitively excluded by Corylus avellana and Quercus in many areas with more fertile soils by 9000 cal a BP, and only persisted at floodplain/valley mire sites after c. 7500 cal a BP, where it was later replaced by Alnus glutinosa. Fire may have also played a role in this succession. Higher ground light levels following fire may have promoted the germination and early growth of A. glutinosa (McVean, 1956). An association between charcoal and expansion of Alnus has been noted at several sites across Britain (e.g. Bennett et al., 1990;Edwards, 1990;Edwards and MacDonald, 1991;Grant and Waller, 2010), suggesting a causal relationship. Tinner et al. (2000) also identified in the southern Alps that A. glutinosa (along with Corylus avellana) was favoured by fire.
Suggestions have been made concerning the possible role of fire in deciduous woodland dynamics. In eastern North America, Quercus dominance at the beginning of the Holocene may have been increased by the incidence of fire (Abrams, 1992;Abrams and Sieschab, 1997). Similar trends also occur in New England where fire was important for oak forests and, in part, controlled by climate (e.g. Foster et al., 2002;Clifford and Booth, 2013). Subsequent suppression of fire by European settlers is associated with a decline in Quercus (Crow, 1988;Abrams, 1992;Crow et al., 1994;Foster et al., 2002). A similar situation has been noted in Sweden (Niklasson et al., 2002). However Clark and Royall (1996) and Clark (1997) have demonstrated the persistence of Quercus in areas with limited evidence of burning, implying that fire is not the only process sustaining Quercus populations. Mason (2000) has suggested that fire may have led to an increase in acorn production in Quercus, further promoting its expansion. At Nissatorp, south-eastern Sweden, a close relationship between the occurrence of charcoal and Quercus pollen has also been suggested, which was attributed to a fire regime, probably light ground fires, that would have helped maintain open conditions and permit Quercus regeneration (Lindbladh and Bradshaw, 1998). Tipping (1995Tipping ( , 1996 suggested a similar type of fire regime for Quercus-Ulmus woodland in Scotland during drier periods before 9000 cal a BP and after 8000 cal a BP, with Clark et al. (1989) suggesting that fire may have facilitated a transition from Corylus-dominated woodland to mixed-oak forest during the Mesolithic at a site in south-western Germany. The record from Cranes Moor fails to show a positive correlation between Quercus and charcoal, although this may in part be related to Quercus' long maturation to first flowering (30 years;Schü tt et al., 2006) in new growth. The positive correlation between Corylus avellana-type and charcoal does imply that open conditions existed. Canopy openness is required for the growth and flowering of this species, and these same conditions would have also favoured Quercus growth and competition. However, the low fire frequency evident in the Cranes Moor record does suggest that burning is simply one of many contributing factors.
Cranes Moor today is situated within a large area of lowland heathland, which is regarded as being a cultural landscape. With the exception of sites subject to climatic stress (e.g. coastal heathlands), it has been shown across north-western Europe that burning appears to have been important in the establishment and subsequent maintenance of heathland communities (see Groves et al., 2012). The identification of possible heathland formation and persistence in the Cranes Moor record is, however, difficult. The main taxa associated with this vegetation community are likely to be Calluna vulgaris and Erica tetralix, although both are shown to be part of the in situ mire vegetation (Fig. 6). Huntley and Birks (1983) have suggested that in areas where C. vulgaris is abundant, pollen values >25% TLP are frequent, although many of their sites consisted of large ombrotrophic mires where C. vulgaris was part of the in situ vegetation. Recent modern studies have suggested that the local presence of C. vulgaris can only be separated from the background component at distances of 4 m or less (Bunting, 2003;Bunting and Hjelle, 2010). DCA of the pollen data provided no meaningful associations for C. vulgaris and E. tetralix to enable any separation between local wetland or dryland sources.
The present landscape surrounding Cranes Moor contains large tracts of heathland, yet pollen samples from the upper 80 cm of the mire (above the hiatus) only reach a maximum of 10% C. vulgaris (Grant et al., 2009). The consistent low values imply that it is not possible to distinguish the presence of surrounding heathland at this site from the in situ mire vegetation, as suggested by Groves et al. (2012). Much of the lowland heathland in north-western Europe dates from the Neolithic onwards, therefore post-dating the Cranes Moor sequence. It can be implied that heathland may have been present locally during the early to middle Holocene around Cranes Moor, and may account for the presence of Ulex-type pollen throughout the upper part of the record, particularly when associated with the recorded increase in BSW, which could suggest a climatic factor contributing to the development of early tracts of heathland. The correlation analysis indicated no statistically significant relationship between C. vulgaris and E. tetralix and microcharcoal, although a positive statistical relationship between macrocharcoal and C. vulgaris was found across the entire dataset, which may again be a reflection of the limited distance that C. vulgaris pollen can be separated from the background component. Fyfe and Woodbridge (2012) found statistically significant correlations between Poaceae and Calluna with microcharcoal from early Holocene contexts in the upland moorlands of Dartmoor, which they suggested might indicate that fire played a role in at least initially promoting open vegetation, particularly Poaceae communities. The evidence for local burning and the persistent record of Betula, Ulex-type and Pteridium aquilinum at Cranes Moor make it enticing to suggest local heath formation and persistence throughout much of the Holocene. However, it is probably safe to suggest only that any heathland present was restricted to small localized patches around the study site at this time, rather than the large-scale tracts of heathland that are associated with the modern cultural landscape.

Variations in fire frequency and local prevailing climate
The results from the plant macrofossils imply a sustained period of predominantly drier BSW conditions between c. 10 500 and 7800 cal a BP, which coincides with the Holocene Thermal Maximum (HTM; c 11 000-5000 cal a BP; Renssen et al., 2012). With the exception of the notable pool layer development at 9500 cal a BP, because of local hydrological disturbances, fire events during this period predominantly coincide with positive (drier) peaks in the BSW (Fig. 4). Charman (2007) and Charman et al. (2009) have shown that the BSW climate signal derived from ombrotrophic mires is primarily driven by precipitation, and the resultant signal reflects the length and severity of the summer moisture deficit, reinforced by the summer temperature. As BSW is largely the interplay between precipitation and evapotranspiration occurring over different timescales, it provides a useful indication of moisture balance, which would have provided a key component in controlling fire incidence. This implies that the timing of these fire events is associated with drier local summer conditions. Marlon et al. (2013) linked an early Holocene warming trend to increased burning in the British Isles, suggesting that climate was the most important driver of fire. In southern Sweden, Greisman and Gaillard (2009) also found increases in fire activity related to dry and warm climatic conditions by correlation with local proxy climate records.
One of the prevalent climatic systems operating upon Northwest Europe is the North Atlantic Oscillation (NAO). Fluctuations in the NAO are recorded in sea surface temperature (SST) (Andersen et al., 2004) and glacier fluctuations (Nesje et al., 2000). The NAO index quantifies the pressure difference between the Icelandic Low and the Azores High and is dominant over the winter weather system. A deep and northerly located Icelandic Low during the early Holocene (Harrison et al., 1992) could thereby imply climatic conditions with a positive NAO signature. In addition to the NAO, the polar vortex is also suggested to have a strong influence and, when strong, the NAO tends to be positive (Baldwin and Dunkerton, 2001;Walter and Graf, 2005). During periods with strong polar vortex regimes, higher pressures at mid-latitudes drive ocean storms further north, and changes in the circulation pattern bring humid winter weather to Alaska, Scotland and Scandinavia, as well as drier conditions to the western United States and the Mediterranean (Walter and Graf, 2005). De Pablo and Soriano (2007) have suggested, albeit upon a limited dataset, that positive NAO index values will result in increased winter lightning strikes in northern areas of western Europe. The sustained positive NAO indexes, high solar insolation and strong polar vortex observed during the HTM may therefore indicate increased lightning strikes over areas of the British Isles, increasing the potential for natural burns to occur. The continuation of warmer, drier conditions may also be related to periods of persistence in anticyclonic weather patterns over areas of north-western Europe, causing increased incidence of lightning, typical of continental-type climates.
From c. 7800 cal a BP the Cranes Moor record demonstrates the beginning of a shift towards a more oceanic climatic regime. This accords with the known timing of the breaching of the Dover Strait (just before 8000 cal a BP; Shennan et al., 2000) and also correlates with pedological evidence from upland regions for accelerated hydromorphism (cf. Bell and Walker, 2005, p. 130) and evidence of major wet shifts recorded in mire records (Hughes et al., 2000). The expansion of Alnus at Cranes Moor coincides with this shift towards wetter conditions and is often taken as an indication of increased oceanicity. However, the expansion of Alnus in the British record does have significantly variable timings, suggesting that local site factors may principally determine its expansion (Bennett, 1990). The increased oceanicity would be expected to decrease fire incidence after 7500 cal a BP and indeed this is the pattern recorded by Marlon et al. (2013, p. 12) who found between c. 7000 and 5000 cal a BP low fire activity regionally for the British Isles.
A number of Holocene fire frequency studies are now available from Europe. In northern Sweden, Carcaillet et al. (2007) found that in areas dominated by Pinus sylvestris, the mean FRI during the Holocene was long and of a similar duration of c. 320 years in several different sites. This similarity was suggested as indicating that the ecological processes controlling fire ignition and spread were the same, with shorter FRIs relating to the dominant controlling factor being climatic. Olsson et al. (2010) and Rius et al. (2012) found periods of increased fire incidence which they attributed to a climatic influence during the early and middle Holocene. Power et al. (2008) also suggested that European fire activity was greater than present during the earlier Holocene (8500-6500 cal a BP), regulated by increased seasonality and biomass, whereas a reduction in seasonality coupled with increased anthropogenic activity was an important regulator of fire during the later Holocene. The results from Cranes Moor indicate three phases of increased fire incidence recorded in the microscopic charcoal: 10 500-9500 cal a BP (200 years), 8700-7500 cal a BP (205 years) and 7000-6000 cal a BP (125 years), coinciding with phases of more conducive climatic conditions (Fig. 4).

Conclusions
The presence of periods with warmer/drier summers coincidental with burning episodes recorded at Cranes Moor indicates that climate is an important natural control on fire incidence during the early Holocene in the British Isles, playing an important role in natural woodland dynamics. The possible role of anthropogenic activity cannot be ruled out at this site, although the paucity of evidence for local Mesolithic activity does suggest any activity would have been limited. What is apparent from the data is that the timing of burning does appear to have a climatic control, and the FRI with which burning was occurring was moderately high, suggesting that burning was not a regular long-term landscape maintenance technique employed by humans (average FRI of 190 years for microscopic charcoal). Changes in vegetation are also shown to coincide with adjustments to the occurrence of burning, with similar trends identified in several other contemporary records. However, there are likely to be a range of temporal and regional variations in the specific nature and timing of burning that can be attributed to local prevailing conditions (hydrography, topography, vegetation patchiness and type), which ultimately modulate the fire-climate relationship (Carcaillet et al., 2001;Heyerdahl et al., 2001;Rius et al., 2012). The identification of a series of natural controls on burning in this study should be heeded when discussing Mesolithic use of fire in the wider landscape.
The palaeoclimate record from Cranes Moor itself is also important for understanding the early to middle Holocene in southern Britain as there is a general paucity of information from this region at this time. Although only based upon plant macrofossils, the record demonstrates clear affinities with other palaeoclimatic records that attach a certain level of reliability to the record from this site. The Cranes Moor record offers the possibility for studying precipitation-evaporation records from mid-European latitudes. This will improve the relationship between similar precipitation-evaporation records found at more northern latitudes and the predominantly lake-based proxies from mid-European locations (e.g. Magny et al., 2003).

Supporting Information
Additional supporting information can be found in the online version of this article: Table S1. Correlation coefficients between selected pollen types and microscopic/macroscopic charcoal data.