Early Paleogene wildfires in peat-forming environments at Schöningen, Germany.

Wild ﬁ re activity in early Paleogene greenhouse conditions can be used as an analogue to gauge the effect of fu- turewarmingtrendsonwild ﬁ reinthecurrentclimatesystem.Inertinite(fossilcharcoalincoal)from11autoch-thonous early Paleogene lignite seams from the Schöningen mine (Germany) was quanti ﬁ ed using macerations, insitu pillars and industry standard crushedsamples.A newthreetransect methodwas developed toquantify in situ charcoal.Thecombinationof insitu pillarsandcrushedsamplesaccountsfortemporalandspatialvariationin charcoal through a stratigraphically oriented pillar, whilst maintaining comparability with industry standards and previous work. Charcoal occurs as a range of randomly distributed particle sizes, indicating that ﬁ res were burning locally intheSchöningenpeat-forming environmentandinthesurrounding areas,but accordingto petrological data, not in an episodic or periodic pattern. Although charcoal abundance is low (relative to previous high ﬁ reworldssuchastheCretaceous),threequantitativeandsemi-quantitativemethodsshowincreasedwild-ﬁ re activity (relative to the modern world) in the warmest parts of the early Paleogene. As atmospheric oxygen levelsstabilisedtomodernvaluesandprecipitationandhumiditybecamethemaincontrolonwild ﬁ re,increased rainfall followed by drier intervals would have created an environment rich in dry fuel in which wild ﬁ res could easilypropagateifhumiditywaslowenough.InthelaterpartoftheEarlyEocene(Ypresian)charcoalabundance fell to levels similar to those found in modern peats. This indicates that the transition to the modern low ﬁ re world occurred within the Early Eocene, earlier than previous records suggest.


Introduction
Wildfire has always been an important part of the natural environment (Bowman et al., 2009;Pausas and Keeley, 2009;Scott et al., 2014), but the role of wildfire in the early Paleogene ecosystems is not yet fully understood (Scott, 2000;Belcher et al., 2013). One of the key sources of evidence for ancient wildfires is the presence of charcoal (Potonié, 1929;Harris, 1958;Komarek, 1972;Scott, 1989Scott, , 2000Scott, , 2010Glasspool and Scott, 2013;Scott et al., 2014), an inert substance with a high potential for survival in the fossil record (Scott, 1989(Scott, , 2010Figueiral, 1999). A 400 million year record of charcoal in coals (Glasspool and Scott, 2010; also replotted by Bond, 2015) showed a transition in the Cenozoic from moderate (17% in the Paleocene) to low (3.5% in the Middle to Late Eocene) charcoal percentages. However, that study was limited by the use of 10 million year bins. Paleogene warm climates (e.g. the Early Eocene Climatic Optimum, EECO) are not direct analogues for future warming trends in the current climate system because they occur during a time when global temperatures were significantly warmer than now. However, improved temporal resolution of early Paleogene wildfire activity can lead to better understanding of the modern fire world and contribute to better predictions of wildfire activity in future warm climates.
Charcoal can accumulate in any depositional environment, but peat deposits are particularly valuable archives of past wildfire activity. Peat is an accumulation of partially decayed plant material which, after peatification and burial diagenesis, may transform to lignite. Lignite is less altered than more mature, higher rank, coals, preserving a record of the peat-forming environment and vegetation (Teichmüller, 1989;Scott, 1991a,b;Wüst et al., 2001;Moore and Shearer, 2003) sometimes to the molecular level (Pancost et al., 2007;Fabbri et al., 2009). Charcoal in the peat-forming environment is likely to be preserved closer to its point of formation than in other settings, as it is not subjected to variables of fluvial transport including differential saturation (Vaughan and Nichols, 1995;Nichols et al., 2000), concentration, floatation, size sorting and fragmentation (Nichols, 1999;Scott et al., 2000). Moreover, peat-forming environments require hydrological stability (Holden, 2005), which mitigates the impact of different climate regimes when comparing records of wildfire (Glasspool and Scott, 2010). Therefore, the charcoal record in lignites offers the best option for comparison of the wildfire record between past time intervals. Riegel et al. (2012) have undertaken a long term stratigraphic, sedimentological and palaeoenvironmental study of the Schöningen mine ( Fig. 1), the sole north German locality where early Paleogene lignites can be studied in succession. Their field surveys document charcoal in multiple seams in different quantities. The current work uses three methods to quantify charcoal in the Schöningen sequence of early Paleogene peat-forming environments, enabling high resolution reconstruction of long term changes in early Paleogene wildfire and interpretation of the timing of the transition to modern levels of fire activity.

Regional geology and palaeoenvironments
The Schöningen Südfeld open cast mine in Lower Saxony, Northern Germany (52.1333°N, 10.9500°E) exposes a sequence of 170 m that includes 11 continuous (occasionally splitting) lignite seams, the oldest (and thickest) of which is Main seam followed by seams 1-9 (supplementary Figs. 1.1 and 1.2). Locally, the lignite seams of minor thickness are intercalated, including "seam L" (between seam 6 and 7, supplementary Fig. 1.1B), which was also sampled for this study. The mine is located on one of the rim synclines of the Helmstedt-Stassfurt salt wall, a prominent structure that runs NW-SE within the Subhercynian basin (Brandes et al., 2012Osman et al., 2013) for more than 70 km (Riegel et al., 2012). Rim synclines formed around the domed salt wall (Manger, 1952), and although there is some regional deformation due to the salt dome there is no obvious faulting in the Schöningen area.
Rooting is generally prevalent in nearly all lignite seams at Schöningen and is frequently visible from the lignite beds into the underlying interbeds, indicating autochthonous peat accumulation (supplementary material herein; Riegel et al., 2012). Between each seam is an interbed composed of clastic sediments, these sediments are crucial indicators of the depositional environment (Riegel et al., 2012). Characteristic features include bimodal cross bedding, coarsening and fining upward sequences, Ophiomorpha burrows and infilled burrows in tops of seams. Taken together these indicate a nearshore coastal setting subjected to tidal influence, changes in energy regime and episodic submergence or emergence (Riegel et al., 2012). Interbed 5 (above seam 4) also contains rooted seagrasses, which clearly indicate marine influence (supplementary material herein; Riegel et al., 2012). Thick, flat lying root branches immediately underlying seam 6 were interpreted by Riegel et al. (2012) as indication of a tall mire forest growing on a marine substrate. Interbed 9 (above seam 8) shows the least marine influence at Schöningen as indicated by in situ palm stumps (supplementary Fig. 2.10.3;Riegel et al., 2012).
Combined evidence from lignite seams and interbeds at Schöningen reveals three main facies. Terrestrial peat-formation is obvious from the many lignite seams, shore-line deposition is indicated by palm stumps and evidence of intertidal sedimentation is visible in most interbeds. It has commonly been suggested that Schöningen was part of an estuarine environment (Standke, 2008;Krutzsch, 2011;Riegel et al., 2012) in which the peat-forming environment was cyclically flooded by marine transgressions. The sequence, therefore, is a suitable study site to examine temporal variations in charcoal occurrence in the peat-forming environments of a low-lying coastal setting.

Dating
Based on palynology, Pflug (1952) regarded the lower group of lignite seams (Schöningen Formation sensu; Riegel et al., 2012) as Early Eocene and the upper seam group (Helmstedt Formation including the Victoria seam) as Middle Eocene. Lenz (2005) also concluded that the Helmstedt Formation is Middle Eocene based on palynological evidence. The palynology at Schöningen has been well documented by Riegel et al. (2012) who found that pollen and spore assemblages are very similar in all seams apart from an increase in palm pollen in seam 4.
Evidence for dating the lowest part (Main seam and seam 1) of the Schöningen Formation comes from Ahrendt et al. (1995) who identified dinoflagellate zone D5b, as defined by Köthe (1990), in the interbed directly above Main seam in the opposite syncline in the Emmerstedt area. If the Main seam in the Emmerstedt area is coeval with the Main seam at Schöningen, then this implies that interbed 1 and seam 1 at Schöningen are Early Eocene and the Paleocene-Eocene (Thanetian-Ypresian) boundary would be within Main seam or below. However, it must be noted that the dinoflagellate marker species for base zone D5b (LO Apectodinium augustum) is not present in the onshore sections in northern Germany (Köthe, 1990(Köthe, , 2012 and the marker for top zone D5b (FO Wetzeliella astra) is very rare (Köthe, 1990). Furthermore Köthe (1990) stated that an Apectodinium acme persisted in zone D5b but that zone D5a was absent. By the original definitions (Costa and Manum, 1988) the Apectodinium acme occurs in D5a so it is unclear why Köthe (1990) considered D5a to be absent. Köthe (2012 p. 207) stated "the exact position of the Apectodinium acme in the study area [German Sector, SE North Sea Basin] has become uncertain".
Although there are few dinoflagellates present in the interbeds at Schöningen, probably due to the nearshore depositional settings, Riegel et al. (2012) recorded an abundance of Apectodinium commencing within interbed 2. If this represents the Apectodinium acme of zone D5b, and associated with the Paleocene-Eocene thermal maximum, then Main seam, interbed 1, seam 1 and the lowermost part of interbed 2 are Paleocene.
Evidence for dating the top seam in the Schöningen succession also comes from Ahrendt et al. (1995) who identified dinoflagellate zone D9a, as defined by Köthe (1990), in boreholes from the Emmerstedt area at levels both below and above a horizon they correlated with the Emmerstedter Grünsand at the top of the lower seam group. Zone D9a is calibrated to Chron C22r by Gradstein (2012) and is within the late Ypresian. Assuming that the lower seam group in the Emmerstedt and Helmstedt area (some 12 km north east of Schöningen) is coeval with the lower seam group at Schöningen (with the Emmerstedter Grünsand either being missing or replaced by interbed 9 at Schöningen. Riegel et al., 2012) then the uppermost seam (seam 9) studied in this paper is likely to be within the late Early Eocene.
Although there are uncertainties we treat seam 9 as late Early Eocene (Ypresian) in this paper. Based on this, we tentatively suggest that seam 5 was associated with the EECO. Main seam and seam 1 may be either earliest Eocene or latest Paleocene.

Field collections
Collecting opportunities at Schöningen are dependent on mining activities. In 2012 and 2013 samples were collected specifically for this charcoal study from Main seam to seam 9. All of the samples were collected in the north area of the mine except for seam 9 which was sampled in the south east. Between 2005 and 2010 samples were collected from seams 1 to 9 during the research leading to publication by Riegel et al. (2012). Seam 3 was sampled in both the north and south west areas of the mine. Seams 4 and 6 were sampled in the north area of the mine. Seams 1, 2, 5, L, 7 and 8 were sampled in the south west area of the mine. Seam 9 was sampled in the south east of the mine. The samples collected were archived in the Senckenberg Museum Palaeobotany Section and sub-sampled for use in this study in 2014.

In situ pillars, crushed blocks and macerations
To preserve intact stratigraphy from various parts of the seam, two adjacent stratigraphically oriented intact lignite pillars were excavated in the field (2012 and 2013 field season only) and tightly wrapped in extra strong aluminium foil to prevent breakage. The supplementary materials show the exact position of pillars within the seams. In the laboratory one pillar was dried and embedded in resin. The impregnated block was then polished following the method of Collinson et al. (2007) to produce an "in situ pillar". The second pillar was preserved in cold storage for future subsampling. Bulk samples were collected in the field from each bed within the seam. Beds were defined by a change in lithotype and, for the 2012 and 2013 field seasons, are shown on the logs in the supplementary material. Logs for the 2005-2010 field seasons are archived in the Senckenberg Museum Palaeobotany Section. Samples representing a whole seam or part of a seam (upper/middle/ lower), composed of 5 g of each lignite bed, were produced for each seam. The number of parts is based on seam thickness. Thin seams 3, 4, 5, L, 7, 8 and 9 were subdivided into 2 parts (lower and upper); thick seams Main, 6 and 1 were subdivided into 6, 4 and 3 parts respectively. Thicker seams with more beds are therefore represented by a larger whole seam sample and more part seam samples. For example, the 'part seam' sample 32Scho1.XXXIII.2-5 includes samples from each of the beds 2 to 5 as shown on the log of seam 1 (supplementary material Fig 2.2.3). Information on the beds incorporated into each part seam sample can be obtained in this manner using the supplementary material.
Whole and part seam samples were prepared to industry standard by Jim Hower and colleagues at the Centre for Applied Energy Research, University of Kentucky, where possible samples were crushed to the standard 20 mesh size, ≤ 840 μm, using a grinder. Soft samples were crushed manually to a top particle size of c. 1000 μm. A subsample was embedded in epoxy resin. The resultant small short cylindrical blocks were polished using 60-, 240-, 400-, and 600-grit SiC papers followed by 0.3 micron alumina on Buehler TexMet paper and 0.05 micron alumina on silk. The final product is a polished block (essentially equivalent to a petrographic pellet in some author's terminology) prepared to industry standards containing a crushed lignite sample. The resin embedded crushed samples and in situ pillars both allow for repeated or future study of the same sample. Polished blocks containing crushed samples were viewed in reflected light under immersion oil (Cargille type A, density 0.923 g/cc at 23˚C, RI of 1.514) using a Leica reflected light microscope and either ×20 or ×50 oil immersion objectives. Photomicrographs were gathered as colour jpeg using a 5 megapixel ProgRes Capture Pro 2.7 camera. 20 g lignite samples, from the same bed from which each in situ pillar was obtained, were treated with 15% H 2 O 2 for 4 hours before sieving in water into size factions retained on 125 μm, 250 μm, 500 μm and 1 mm sieve meshes. Particles released were studied using light microscopy suspended in distilled water so as to reduce breakage. Charcoal particle abundance was estimated for the 250-500 μm size fraction. An ACFOR (abundant, common, frequent, occasional, rare) Domin scale, as used in ecology, was modified by combining the abundant and common categories.

Petrography and inertinite quantification
Some authors separate charcoal into categories based on its perceived origin (Teichmüller, 1989;Moore et al., 1996;Valentim et al., 2013) but this separation has been questioned extensively (Winston, 1993;Scott, 1989Scott, , 2000Scott and Glasspool, 2007). Multiple charring experiments have been performed on various plants and fungi (e.g. Jones et al., 1993Jones et al., , 1997Guo and Bustin, 1998;McParland et al., 2007;Scott, 2010 and references therein) all of which demonstrate that increasing charring temperature leads to increasing reflectance. In the petrographic study of coals charcoal is referred to as inertinite (ICCP, 1963(ICCP, , 2001Sýkorová et al., 2005). Increased reflectance (compared to other components) characterises inertinite, thus, in the absence of macrinite, micrinite and secretinite which are restricted to hard coals, charcoal and inertinite can be considered synonymous.
Lignite components, i.e. macerals, were identified according to the International Committee for Coal and Organic Petrology (ICCP) standard allowing for comparability between disciplines and practitioners. Four inertinite macerals were studied in this paper, fusinite, semifusinite, inertodetrinite and funginite. Fusinite ( Fig. 2A-D) has the highest reflectance of the inertinite macerals (ICCP, 2001) so it appears brighter than all of the other macerals under reflected light. Pieces usually show well preserved cellular structure but a homogeneous particle with high reflectance can still be classified as fusinite as decomposition before charring can result in a bright relatively homogenous maceral (Taylor et al., 1998). Semifusinite ( Fig. 2B and D) is a product of low temperature charring, consequently it has lower reflectance than fusinite but higher than huminite (ICCP, 2001). Semifusinite particles typically have less clear cellular structure than fusinite particles (ICCP, 2001). Inertodetrinite ( Fig. 2C and E) is classified by the ICCP (2001) as a charcoal maceral which is not larger than 10 μm. Particles are formed by desiccation, attrition or crushing of a larger particle (ICCP, 2001). Scott and Glasspool (2007, see also Scott, 2010) argued that fragmentation during combustion forms small ash-like inertinite particles that are often transported by wind and are therefore indicative of regional wildfire activity. Fungal spores and sclerotia (Fig. 2F) are often preserved in lignite. Charred fungal remains are referred to as funginite (previously sclerotinite) (ICCP, 2001). Although not all agree that these remains represent charred material (Hower et al, 2009), it has been proven that reflectance increases with charring temperature in fungal sclerotia (Figs. S14 e and f in Scott et al., 2010 auxiliary material). For the purposes of this study uncharred fungal remains were included in the huminite category.
Inertinite percentages have been obtained from in situ pillars and crushed samples representing lignite seams sampled at Schöningen. Previous work has quantified inertinite in in situ pillars using eyepiece graticule grids (Belcher et al., 2005;Collinson et al., 2007;Hudspith et al., 2012), but there is no standard quantification method and those studies were very time consuming and labour intensive. In this study a new line transect method, counting three transects of 504 points each (Robson et al., 2014), using the ×20 objective, was devised and tested to quantify macerals as defined by the ICCP standard classification system. Twelve points, 50 μm apart, were counted per field of view (600 μm). Tests on variation around the mean percentage inertinite with different numbers of points counted showed that, for 500 points, fluctuations are limited to within ±0.60%. The new method is less time-consuming allowing for more samples to be studied, and it accounts for variation in inertinite particle size and distribution through the in situ pillar. Crucially, the method used is comparable with previous work and industry methodologies. The same counting method was applied to crushed samples but only one transect of 504 points was counted due to the smaller size of the sample and the lack of stratigraphic information (e.g. bedding) preserved. The same ±0.60% uncertainty applies to all counts.
To facilitate comparison with previous work, where counting methods are often not stated, tests were undertaken using different counting methods (Whipple grid, Kötter graticule, Cross hairs). These tests revealed that there can be up to 5% variation between inertinite percentages obtained from identical transects using the different methods. This potential variation is taken into account when comparing new Schöningen results with those from the Glasspool and Scott (2010) database.
This work uses complete seams (either in their entirety for thinner seams or in parts for thick seams) to obtain overall fire activity, a measure that allows comparison with the database of Glasspool and Scott (2010); we are not addressing fire frequency here. However it should be borne in mind that both, high and low fire frequencies could yield the same 'seam-average' and 'part seam average' inertinite percentages.

Crushed samples
The most common inertinite maceral is inertodetrinite (Fig. 2C and E), followed by semifusinite (Fig. 2B and D). Fusinite ( Fig. 2A-C) and funginite (Fig. 2F) are the least common sub-macerals. Other than noting the occurrence of some large particles, inertinite size range and distribution cannot be addressed in crushed samples. There is no clear relationship between inertinite percentages or percentages of the huminite macerals, attrinite, textinite and ulminite (maceral data in supplementary spreadsheet pages 1 + 3).
Crushed samples from part seams show that the variation within some seams is as high as the variation between seams (Fig. 3B and D). As might be expected, there is more spatial and temporal variation in inertinite in thick seams (Main, 1, 2, 6, 8, 9 range 2%-21%) than in thin seams (3, 4, 5, L, 7 range 1%-15%). In one set of part seam samples from thick seam 6, inertinite content varies by 17% (2-19%, Fig. 3B) but in another it varies by only 2% (4-6%, Fig. 3D). In one set of part seam samples from thin seam L, inertinite content does not vary (both 2%, Fig. 3B). In seams with overall high inertinite percentages, some part seam samples have percentages that are as low as those in the low inertinite content seams (4, L-9). The opposite is not observed; none of the part seam samples in low inertinite content seams (seam 4, 4%; seam L, 2% seam 7, 6%; seam 8, 5% seam 9, 8%; Fig. 3B and D) reach the high values of the high inertinite content seams (Main-2, 5 all contain beds with N 12% inertinite). In summary, high inertinite content seams have both higher mean percentages and also larger variability. Therefore, two striking features are revealed by the novel examination of whole seam and part seam samples. First, a long-term overall decrease in inertinite percentages is visible from Main seam to seam 9. Second, imposed on this trend, are dramatic fluctuations in charcoal percentages between seams, but even more dramatically, within individual seams.

In situ pillars
Semifusinite ( Fig. 2B and D centre) is the most common maceral in all pillars and inertodetrinite ( Fig. 2C and E) is the second most common. Fusinite ( Fig. 2A, C and D right) is uncommon in the Schöningen samples and funginite (Fig. 2F) is the least common maceral, although funginite percentages increase relative to other macerals in the samples from the younger seams L, 7, 8 and 9. Discrete bedding planes covered with charcoal (seams 5 and 6) and large (≥ several mm) charcoal particles distributed in discrete layers and lenses (Main seam and seam 1) which were observed in the field were interpreted as indicating intermittent fire activity by Riegel et al. (2012). Inertinite particles of different shapes and sizes are randomly distributed (Fig. 2C) in the majority of Schöningen material. However, some pillars contain areas with high proportions of inertodetrinite particles (Fig. 2E) and laterally discontinuous charcoal horizons.

Macerations
Charcoal abundance within macerated samples was estimated using a semi-quantitative modified ACFOR Domin scale from common to rare (Fig. 4B). The score of zero for seam 3 is explained by the absence of particles greater than 250 μm in that seam. The overall pattern of charcoal distribution through the succession is similar to that in the in situ pillars (Fig. 4A) and crushed samples (Fig. 3). The highest abundances occur in seams Main and 5 with the lowest abundances in seams L-9.

Value of combined approaches for the study of charcoal
Different information can be obtained by various methods of studying charcoal but there are caveats applicable to each method. By choosing a specific size faction (250 μm-500 μm in this study), comparison of charcoal abundance between macerated samples is easier, although results can be misleading when a sample consists solely of small charcoal (seam 3, Fig. 4B). Low temperature charcoals with low lustre can be hard to recognise in macerated samples using light microscopy. By contrast, petrographic study of inertinite, in polished blocks or in situ pillars, can identify charcoal across a wide range of charring temperatures (e.g. fusinite/semifusinite) (Glasspool and Scott, 2013).
Due to the preparation method, crushed samples and macerated samples preserve no temporal or spatial distribution data and only a restricted perspective on particle size. In contrast, in situ pillars retain evidence of charcoal particle size and distribution (e.g. banded, clumped or irregular).
However, sampling constraints mean that in situ pillars only represent part of the seam, and therefore a limited window of fire activity, whereas a crushed sample can represent either the whole seam or part of it. Inertinite relative abundance in a high-resolution (every c. 5 cm) study using crushed samples of seam 1 (Inglis et al., in press) largely matches the lower resolution (part seam) results herein (Fig. 3B), with higher percentages at the top and lower percentages at the base of the seam. This indicates that inertinite variation can be adequately documented by lower resolution sampling.
There are benefits and limitations to each individual method. The combined study of in situ pillars, crushed samples and macerated samples provides a powerful tool for analysis of charcoal.

Inertinite macerals, size and distribution: wildfire implications
A mix of particle sizes from N500 μm to less than 10 μm ( Fig. 2A, B, D, E) is visible in the in situ pillars. This indicates that the Schöningen material is a record of local wildfire activity (Innes and Simmons, 2000;Nichols et al., 2000) as well as regional activity Scott, 2010). Semifusinite and inertodetrinite are the most common inertinite macerals quantified using both in situ pillars and crushed samples. Since inertodetrinite is considered to be predominantly wind-  Individual parts of seams can contain higher or lower inertinite percentages than the whole seam sample but seams 4 and L-9 never contain more than 8% inertinite. Although inertinite content is variable, both whole and part seam samples, collected from two areas of the mine in different field seasons, show a similar pattern, with higher inertinite percentages in Main seam, seams 1,2 and 5 and lower inertinite in seam 4 and in the later part of the Early Eocene (seams L-9). Fig. 4. Inertinite (charcoal) percentages from in situ pillars and charcoal abundance from macerated samples from Schöningen, adjacent to simplified log (with seams to scale) adapted from Riegel et al. (2012). (A) Inertinite (charcoal) percentages obtained from in situ pillars. Uncertainty of 0.60% applies to all values (see Section 3.3) and the maximum error on the mean is an additional ±0.24% (refer to supplementary spreadsheet page 4 for individual values for each pillar). Due to the thickness of Main seam and seam 6 and the nature of access in the mine, pillars were collected from the upper and lower parts of the seam and results were averaged (Main seam 6%-23%, seam 6 1%-18%). (B) Charcoal particles, released from samples representing the same bed as the in situ pillars, relative abundance estimated using a semi-quantitative method (modified ACFOR Domin scale). The 250 μm-500 μm size fraction was chosen as it is most likely to represent mainly local wildfire. The pattern of charcoal relative abundance from both methods is very similar to those shown in Fig. 3A and C. blown due to its small size (Scott, 2010), it can be used as an indicator of local and regional high temperature crown fire activity (Clark, 1988;Clark et al., 1998;Scott, 2010). In situ pillars contain all particle sizes as all macerals are preserved in their original spatial and temporal context. The single plane of section studied might mean that some particles were cut at an angle or in a position that did not reflect their maximum size, but the effects would be minimal in this study of three transects in each pillar.
Inertinite in in situ pillars shows no distinctive distribution pattern unlike the repeated, multiple scale banding that characterises immediately pre-PETM fires at Cobham, UK (Collinson et al., , 2009. At Schöningen inertinite is mostly irregularly distributed in polished pillars, with the occasional discontinuous band in some seams. This relative homogeneity may explain why pillars representing part of a seam (Fig. 4A) show a similar inertinite distribution pattern to that in crushed samples representing whole seams (Fig. 3A).
Areas of rare inertinite suggest that fuel would have accumulated between periods of fire activity, much as it does in modern fire prone areas (Pierce et al., 2004). Partially charred particles (Fig. 2D) in the in situ pillars indicate a brief heating duration or high levels of fuel moisture (Hudspith et al., 2012;Marynowski et al., 2014) as might be expected for wildfires in peat-forming settings.

Early Paleogene charcoal from Schöningen and the global Phanerozoic charcoal record
The high (11-19%, Figs. 3A, C and 4A) early Paleogene levels of inertinite (charcoal) in some seams at Schöningen are low by comparison with periods of mid to high fire activity recorded for intervals in the Paleozoic and Mesozoic by Glasspool and Scott (2010) i.e. the Permian = mean 44.4%, Late Jurassic = mean 26.1%, Middle Cretaceous = mean 42.8%. These high inertinite levels in the Paleozoic and Mesozoic are thought to be a consequence of higher atmospheric oxygen content which enabled wetter plant material to burn more easily (Glasspool and Scott, 2010). In the early Paleogene, models suggest that atmospheric oxygen levels began to stabilise to modern levels, so the effects of precipitation on wildfires may have increased (Belcher et al., 2013;Scott et al., 2014). Consequently, in spite of the overall warm early Paleogene climate, fire activity was reduced compared to that in the Paleozoic and Mesozoic.
Data from the three early Paleogene bins (Middle to Late Paleocene; Early Eocene; Middle to Late Eocene) in the Glasspool and Scott (2010) database, were re-plotted and Schöningen whole seam crushed sample results were superimposed upon them (Fig. 5). The Schöningen Main seam and seam 1 may be either earliest Eocene or latest Paleocene (Section 2.2) but for simplicity Fig. 5 shows their charcoal abundance superimposed on the Late Paleocene bin. Seam 9 is treated here as late Early Eocene (Section 2.2).
The database compiled by Glasspool and Scott (2010) displayed a stepped reduction in inertinite percentages from the middle Cretaceous (mean 42.8%) to the present day (mean 4.5%). The Middle to Late Paleocene bin has high inertinite percentages (mean 17%), the Early Eocene bin has lower inertinite percentages (mean 8.6%) and the Middle to Late Eocene bin has lower percentages still (mean 3.5%). When Main seam and seam 1 are treated as Paleocene (Fig. 5), inertinite percentages from Schöningen are slightly lower than mean values previously reported for the Paleocene to Paleocene-Eocene transition interval. Using the alternative age model, where all seams are Early Eocene, the high inertinite percentages for Main seam and seam 1 are higher than previously reported for the earliest Eocene.
Subsequently the inertinite percentages from Schöningen differ from those in Glasspool and Scott (2010), irrespective of possible alternative stratigraphic interpretations for seam 9. Inertinite percentages fall in seams 3 and 4, with seam 4 differing significantly from Glasspool and Scott's (2010) dataset (Fig. 5). Seam 4 palynology shows an increase in palm pollen (Riegel et al., 2012) compared to all other seams. This might imply that local vegetation was less flammable, however low inertinite values also occur higher in the Schöningen sequence where there are typical palynological (Riegel et al., 2012) and huminite maceral assemblages (maceral data in supplementary spreadsheet pages 1 and 3). High inertinite percentages characterise seam 5 in the middle Early Eocene (possibly at the time of the EECO). These higher abundances are followed in the middle Early Eocene by a fall in inertinite in seams L-9 to low percentages characteristic of those reported for the Middle Eocene at Helmstedt (and other sites, unpublished Diploma thesis, Bode, 1994, University of Göttingen-Riegel et al., 1999, 2012, later Cenozoic records (Glasspool and Scott, 2010) and modern peat forming environments (Eble and Grady, 1993;Scott et al., 2014) the average being~4.3%.
An increase in peat accumulation rates could result in a decrease in inertinite percentages but without any change in fire activity. This possibility is very unlikely to explain the decrease in inertinite percentages in seams L-9 at Schöningen because increased peat growth (hence higher peat accumulation rate) is more likely during accumulation of seams Main to 6 when climates were equally wet and included the warmest intervals (Section 5.4). Furthermore, this explanation is very unlikely to account for the sustained lower percentages of inertinite through the remaining Cenozoic, a time of climate cooling .  Glasspool and Scott (2010) database (pale grey background area with black bars showing 5% variation to account for different quantification methods (see Section 3.3)). Where samples were collected from the same area of the mine in different field seasons, the inertinite percentages were averaged (seams 3, 5%-9%; seam 4, 1%-2%; seam 6, 5%-6%; seam 9, 2%-4%). Schöningen inertinite percentages from Main seam and seam 1 are high, consistent with data from the Glasspool and Scott (2010) database for the Paleocene. Inertinite percentages remain high in the early Early Eocene (seam 2) but fall to a very low value (seam 4), and then rise again in the middle of the Early Eocene (seam 5). Subsequently, (seams 6-9) later Early Eocene inertinite percentages fall to values similar to those of the Middle and Late Eocene as reported by Glasspool and Scott (2010).
Today, only lowland tropical rain forests and montane rain forests with an annual precipitation of 2000-3000 mm, evenly distributed over the year, are inflammable despite high thunderstorm activity (Goldammer, 1993). High precipitation and humidity have been inferred for the Middle Eocene in Central Europe (Riegel, 2001), allowing tropical elements to be sustained in the vegetation even though the climatic optimum had already passed. This suggests that low inertinite in seams L-9 may be linked to uninterrupted highly per-humid conditions preventing the flammability of forest litter.
The varied inertinite pattern at Schöningen suggests that fire activity fluctuated in the warmest parts of the Early Eocene (including high and low levels) before reaching consistently low levels from the mid to late Early Eocene onwards. The Schöningen data indicate that the transition to the modern low fire world occurred earlier than previously thought i.e. in the mid-late part of the Early Eocene, not in the Middle Eocene. Furthermore, the long term Cenozoic trend of falling inertinite levels (Glasspool and Scott, 2010) is supported by the data from Schöningen. This is evidence against any major influence of sampling bias in the Cenozoic part of the Glasspool and Scott (2010) global dataset.

Climate effects on early Paleogene wildfire
Models suggest that atmospheric oxygen fell to modern levels (21%) in the Early Eocene (Berner et al., 2003;Zachos et al., 2008;Berner, 2009). Oxygen is a key part of the wildfire triangle (Scott, 2000) with fire in wet vegetation unsustainable at atmospheric oxygen levels lower than 23% . The stabilisation of oxygen levels in the Early Eocene heralded the first time in Earth's history when oxygen was not the controlling factor on wildfire ; consequently the impact of precipitation (Belcher et al., 2013) and temperature on wildfire would have increased. Changes in lightning (due to increased CO 2 , Belcher et al., 2010), temperature (Westerling et al., 2006;Marlon et al., 2009;Scott et al., 2014), fuel formation (Scott, 2000) and moisture (Dimitrakopoulos and Papaioannou, 2001) can all affect wildfire activity. Increased rainfall would lead to wetter fuel but also to increased plant growth, which, when coupled with increased temperatures, could increase the burn potential of biomes previously little affected by wildfire.
Global climate patterns from the Paleocene to the middle Early Eocene show a long term warming trend on which transient hyperthermals were superimposed Bijl et al., 2013). The Paleocene-Eocene thermal maximum (PETM) is one shortlived hyperthermal. It is marked by (i) a global increase in temperatures of~5˚C (Dunkley Jones et al., 2013) and (ii) irregularities in the hydrologic cycle (Bowen et al., 2004;Sluijs et al., 2006;Tipple et al., 2011;Handley et al., 2012;Krishnan et al., 2014). Similarly, the warmest climates of the EECO (c. 49 Ma) were marked by high global temperatures and changes in global monsoonal activity that affected the distribution of precipitation (Huber and Goldner, 2012).
It is suggested that the intensity and distribution of precipitation changed as a result of Paleogene warming (Huber and Goldner, 2012), with greater moisture transport to high latitudes  and increased precipitation at mid to high latitudes, as evidenced by Azolla blooms in the Arctic and Nordic seas as far south as Denmark (Barke et al., 2012). Some areas may also have experienced enhanced seasonal extremes (John et al., 2008). Plink-Björklund et al. (2014) suggested that extremes in the hydrological cycle were manifested as increased precipitation 'peakedness' in the subtropics and mid-latitudes during hyperthermals, where high intensity rainfall events were followed by drier intervals. A large scale arid/semi-arid, humid, arid/semi-arid pattern across the PETM is recognisable in clay rich sequences from Egypt (Khozyem et al., 2013), the Tethys and the Atlantic (Bolle and Adatte, 2000), continental flood-plain sediments in the Spanish Pyrenees (Schmitz and Pujalte, 2007) and complex sedimentological and isotopic changes in Tanzanian sediments (Handley et al., 2012), suggesting that variability in rainfall was not restricted to one locality.
Main seam to seam 5 at Schöningen, deposited during the warmest intervals of the early Paleogene (probably including the PETM and including the EECO according to the current age model), contains the highest charcoal relative abundances based on the new petrological evidence from crushed samples (Fig. 3) and in situ pillars (Fig. 4A). The variability in inertinite in seams 3 and 6 ( Fig. 3B and D) may record transitional periods from higher wildfire activity (Main-2, 5) to lower wildfire activity (4, L-9), at least locally and possibly globally. This suggests that brief drier intervals, following wet periods with increased growth and fuel load, may have led to increased wildfire activity during the warmest climates of the Paleogene. One possible modern analogue for this effect is in the Amazon rainforest (Cochrane, 2003). High fuel loads and temperatures coupled with lower humidity in periods of drought caused by modern climate change have increased the number of high-intensity wildfires (Brando et al., 2014). It is therefore possible that, although wetter, the warmest climates of the Early Eocene could have been more fire prone than 'background' Eocene greenhouse conditions, hence the variation in charcoal abundance between seams Main to 9 at Schöningen (Fig. 5).

Conclusions
Petrological analysis of lignites using crushed samples and in situ pillars provides detailed quantitative documentation of inertinite (charcoal) abundance in the Schöningen lignites whilst allowing for comparability with previous work and industry standards. Large charcoal particles (≥500 μm) in the rooted autochthonous early Paleogene lignites at Schöningen are unlikely to have been washed in and therefore represent locally occurring wildfires. Small particles, such as inertodetrinite, which can be easily carried by the wind, may be derived from both local and regional fires. The most common inertinite macerals quantified using both methods are semi-fusinite and inertodetrinite. In spite of their varied advantages and disadvantages, all three methods of analysis (crushed samples, in situ pillars and macerations) show very similar patterns of relative charcoal abundance through the early Paleogene of Schöningen.
Despite the early Paleogene greenhouse climate, the highest charcoal percentages at Schöningen are low in comparison to those recorded in Paleozoic and Mesozoic intervals. By the Eocene, atmospheric oxygen had stabilised to modern levels, increasing the effect of precipitation on wildfire. The highest charcoal abundances and greater within seam variability occur during the warmest intervals of the Early Eocene which also experience episodic precipitation. In the subsequent cooler periods, low charcoal percentages are recorded which are similar to those of the later Cenozoic and present day, suggesting that the onset of the modern low-fire world began earlier than previously thought, in the later part of the Early Eocene (Ypresian).