Wildfires in northern Siberian larch dominated communities

The fire history of the northern larch forests within the permafrost zone in a portion of northern Siberia (∼66°N, 100°E) was studied. Since there is little to no human activity in this area, fires within the study area were mostly caused by lightning. Fire return intervals (FRI) were estimated on the basis of burn marks on tree stems and dates of tree natality. FRI values varied from 130 to 350 yr with a 200 ± 50 yr mean. For southerly larch dominated communities, FRI was found to be shorter (77 ± 20 yr at ∼ 61°N, and 82 ± 7 at 64°N), and it was longer at the northern boundary (∼71°) of larch stands (320 ± 50 yr). During the Little Ice Age period in the 16th–18th centuries, FRI was approximately twice as long those as recorded in this study. Fire caused changes in the soil including increases in soil drainage and permafrost thawing depth, and a radial growth increase to about twice the background value (with more than six times observed in extreme cases). This effect may simulate the predicted warming impact on the larch growth in the permafrost zone.


Introduction
Larch (Larix spp.) forests comprise about 43% of Russian forests. Larch stands are dominant from the Yenisei ridge on the west (∼92 • E longitude) to the Pacific Ocean, and from Baikal Lake to the south to the 73rd parallel to the north. Wildfires are typical for this area, with the majority occurring as ground fires due to low forest crown closure. Larch is a pyrophytic species, since fires promote the establishment of larch regeneration and reduce betweenspecies competition. Mineralized burned surfaces enriched with nutrients are favorable for germination of the small and low weight larch seeds. The expanse of larch forests is at present considered to be a carbon sink (Shvidenko et al 2007). However, an increase in fire frequency in response to observed climate changes in the area may result in conversion of this area to a source for greenhouse gases (IPCC 2007). Changes in air temperature, permafrost depth and extent may affect wildfire frequency (Kharuk et al 2008). In spite of the fact that larch dominated forests occupy about 70% of permafrost areas in Siberia, data on fire occurrence in larch forests are presented in only a few publications (Vaganov and Arbatskaya 1996, Kovacs et al 2004, Kharuk et al 2005, 2008, Schepaschenko et al 2008, Wallenius et al 2011. For larch dominated areas of central Siberia, the average fire return interval (FRI) was found to be 82 ± 7 yr (∼64 • N latitude), and 77 ± 20 yr for the southward 'larch-mixed taiga' ecotone. For the northern boundary of larch forests (∼71 • N) the FRI value was estimated to be 320 ± 50 yr (Kharuk et al 2011). For the north-east larch forests of Siberia, FRI was found to be, depending on the site, 50-80 and 80-120 yr (Schepaschenko et al 2008). For southern central Siberia, Wallenius et al (2011) reported a gradual increase of FRI from 52 yr in the 18th century to 164 yr in the 20th century.
The purpose of this work is to investigate wildfire occurrence in the central part of larch dominated communities (figure 1).

The study area
The test sites were located within the Kochechum River watershed. The Kochechum River is a tributary of the Nizhnyaya (Lower) Tunguska River which turns northwest and  (Kharuk et al 2007(Kharuk et al , 2008(Kharuk et al , 2011. Inset: Landsat image showing locations (1-9) of test sites along the Kochechum River. Measurements of older trees from a study of long-term wildfire trends were acquired in areas denoted by boxes. flows into the Yenisei River. This area is the northern part of the central Siberian plateau, with gentle hills with elevations up to 1000 m (figure 1). This is a permafrost area with a severe continental climate. Mean summer (JJA) and winter (DJF) temperatures are +12 • C and −35 • C respectively. Mean summer and annual precipitation is 180 and 390 mm, respectively (these values are means over the years 1997-2006 averaged for 1.5 • × 2.5 • grid cells and covering all test sites (Climate Explorer 2011)).
The forests are composed of larch (Larix gmelinii Rupr.) with a mixture of birch (Betula pendula Roth). Typical ground cover is composed of lichen and moss. Bushes were represented by Betula nana, Salix sp, Ribes sp, Rosa sp, Juniperus sp, Vaccinium sp and Ledum palustre L. (Labrador tea).
The wildfires across this landscape occur as ground fires due to low crown closure. The seasonal fire distribution is single-mode (late May-June) with rare late (i.e., August-early September) fires (Sofronov et al 1999, Kharuk et al 2007. Periodic stand-replacing fires cause a mosaic of (semi-)evenage stands, embedded with older trees that survived the fire.

Tree sampling
The samples were collected in 2001 and 2007 within the burned areas of larch forests along the river. There are no roads within the study area, and rivers provide the best access (figure 1). Temporary test sites were established within burned areas within about 1.0 km from the river and within a 160-420 m elevation range. Disks of larch bole cross sections for tree ring analysis were cut at the root neck level. Sampled trees included specimens of the dominant (even-age) trees, as well as older trees that survived stand-replacing fires.
Trees within test sites 1-9 (figure 1) were sampled and used for the FRI analysis. In addition to this, older trees on supplementary sites were sampled (figure 1) for the purpose of estimation of long-term trends in fire frequency. The sample consisted of 58 trees.

Sample analysis
Tree ring widths were measured with a precision of 0.01 mm using the well-known LINTAB-III instrument. The dates of fires were estimated on the basis of a master chronology constructed for northern larch forests (Naurzbaev et al 2004). The mean correlation with the master chronology was 0.54 which is satisfactory for our purposes. The COFECHA (Holmes 1983) and TSAP (Rinn 1996) programs were used to detect double-counted and missing rings. Fire-caused tree ring deletions were found only for three cases (out of about 75 analyzed). Relevant dates of fire events were adjusted on the basis of the master chronology.

Fire return interval calculation
FRI were routinely estimated by tree ring calculation, between consecutive fire scars: D i − D i−1 , where D i and D i−1 are the dates of fires i and i − 1. Since many sampled trees have only a single burn mark, the dates of tree natality were included in the FRI calculation where appropriate, as described below.
It is known that within larch dominated communities, fires are mostly stand-replacing, and promote the formation of evenaged stands. Fresh burns with mineralized soil are quickly occupied by dense regeneration. Over time, the development of a thick moss and lichen cover limits larch regeneration (Kharuk et al 2008). Larch produce seeds annually, with good harvests occurring on 5-7 yr cycles (Pleshikov and Novosibirsk 2002). Furthermore, cones 2-3 yr old are known to be viable seed sources (Sofronov et al 1999). Importantly, ground fires regularly do not damage cones, leaving fire-killed stands as a source of seed. Thus both tree natality dates and the dates of fires were used for FRI calculations. FRI was determined for each individual tree within the test sites and then the mean FRI for a given site was calculated.
The long-term history was based on fire events were for the period spanning the 17th century to 2007. To exclude the impact of a decreasing sample size for older stands, i.e., the 'fading effect', the sample size was adjusted by tree age within groups. Stand-replacing and non-stand-replacing fires were investigated.
Along the burn marks, 'tree ring width growth accelerations' were also determined. 'Growth accelerations' were identified by the following procedure. (1) Expert visual analysis of increase in tree ring width.
(2) Calculation of the mean tree ring width 20 yr before and 20 yr after the Overall mean 200 ± 51 ( p 0.05) increases began; this reference period (20 yr) approximately corresponds to the period of post-fire tree ring increment increases.
(3) If the ratio (mean tree ring width after beginning of tree width increase/mean tree ring width before tree width increase) was >2.0, the observed 'growth acceleration' was considered significant. (4) The date of the 'growth increase' was compared with the air temperature record. It is known that for the boreal-forest zone, radial growth has been closely connected with temperature variability (Esper et al 2010). If that date coincided with air temperature increase, that particular 'growth increment increase' was excluded from further analysis (because of the possibility of climatedriven tree ring increase). The above-mentioned 'growth acceleration' dates were considered as a possible additional indirect sign of wildfire impact. It is known that trees that survive fires (including those which do not have burn marks) experienced a period of growth increase (see, e.g., Sofronov et al 1999).

Error analysis
Including the tree natality date into FRI analysis increased errors due to the 0-5 yr lag of the post-fire tree establishment. The natality date also has to be adjusted to the 'stump age', i.e., the difference of real and measured stump height tree age. Even if a tree was cut at the root neck level, this difference can be 2-5 yr. In some cases disks sampled at the root neck level were not suitable for analysis; in these cases disks were sampled higher up the bole. This procedure entailed about five additional years of uncertainty. The post-fire regeneration had a high growth increment the first 15-20 yr, which gradually decreased due to increasing competition and decreasing active root layer depth (Sofronov et al 1999). In summary, the maximum total error was estimated to be 15 yr.

Fire return intervals
Dates of tree natality and fire events are presented in figure 2. FRI values for different test sites vary considerably, i.e., from 131 to 349 yr (with mean of 200 ± 51 yr; table 1).

Long-term fire event history
The long-term fire event history (for the 17th through 20th centuries) was developed on the basis of data for trees with natality dates before 1800, 1700 and 1600 yr, respectively (figure 3, table 2). To exclude the fading effect, data were adjusted for tree natality dates: >200 yr, >300 yr and >400 yr for 19th-20th, 18th-20th, and 17th-20th century comparisons, respectively.
The number of wildfires in the 20th century increased relative to that for the 19th century (13 versus 8) (table 2; N = 58). For the 18th, 19th and 20th centuries (N = 33), the fire numbers were 1, 6 and 9, respectively (table 2; N = 33). Results for the period of 17th-20th centuries also showed an

FRI
The majority of the fires in the larch communities that we studied were stand-replacing. This was revealed by the fine-grained mosaic of (semi-)even-age stands within forested landscapes. For the kinds of forest communities the mean FRI can only be reconstructed using age-class analysis or fire records (Sannikov andGoldammer 1996, Johnson andMiyanishi 2001). However, in our case the presence of fire scars indicated that non-stand-replacing fires also occurred (figure 2). FRI data for different sites within the study area (figure 1) differed by more than a factor of 2, from 131 to 349 yr with a mean of 200 ± 51 yr (table 1). The reason for this is that fires are rare events within the area investigated. The other cause is the non-uniformity of the topography and land cover within the study area. It is known that fire frequency is different for sunlit and shadowed slopes, as well as for bog areas (Beaty and Taylor 2001, Rollins et al 2002, Kharuk et al 2005, 2008 (60-150 yr: Payette 1992, Larsen 1997, Swetnam 1996, Sannikov and Goldammer 1996. Very long FRIs (up to 300 yr) were reported for fire-protected forests in Europe and North America (Weir et al 2000, Heyerdahl et al 2001, Bergeron et al 2004, Buechling and Baker 2004. Evidently this is not the case here, because within our study area fires were never suppressed. Low fire frequency is not favorable for the larch forests, because fires promote larch regeneration growth, i.e., larch is a 'pyrophytic' species. The main tree growth constraints are permafrost thawing depth and soil drainage. The depth of the seasonal thawing is dependent on the exposure, moss and lichen layer thickness, and fire history. Fires not only increase the permafrost thawing depth but also increase the soil drainage, which is very important for larch growth. With time, an increase in the thermal insulator layer composed of the on-ground moss and lichen cover caused upward migration of the permafrost layer, and compression of the active root zone within a progressively decreasing upper layer. Fires also thin regeneration, decreasing within-species competition, and, thus, promoting tree growth because larch is an extremely shade-intolerant species.

Long-term trends in wildfire number
The advanced age of some sampled trees (>400 yr; figure 3) allows estimation of the fire history within the study area back to the Little Ice Age period. Dendrochronology data showed that cooling during the Little Ice Age period caused depression in the tree annual radial growth (figure 4). Fire event numbers since the 17th century approximately coincided with air temperature deviations, increasing with warming since the second half of 19th century ( figure 4, table 2). For example, fire numbers increased from 8 in the 19th century to 13 in the 20th century. This phenomenon could not be attributed to decreased samples of older trees, i.e., a 'fading effect', since sample sizes were adjusted by natality date.
The local asynchrony of the radial increment growth and temperature deviations in figure 4 could be attributed to the fire-induced increase of the radial growth, as well as to the incrementation decrease during the lag between the fire event and incrementation increases beginning (which is about 5-7 yr; figure 5). Thus, in the middle of 20th century wildfires were observed within the majority of the test sites (figure 2). The other reason is a 'divergence phenomenon', i.e. growthversus-temperature divergence (D'Arrigo et al 2008, Esper et al 2010. These estimates coincide with earlier reported data on fire frequency increase in the 20th versus 19th centuries for the southerly larch and mixed forests (sites 2 and 1 in figure 1, respectively; Kharuk et al 2008). The causes of this trend can be both natural and anthropogenic. Earlier it was shown that for site 1 (figure 1) the FRI decrease was caused by both natural (warming) and anthropogenic causes. For site 2 (figure 1) the FRI decrease was attributed to temperature increase mainly, because the leading factor in fire ignition in the remote northern forests is lightning (>90% of cases in the northern forests, whereas within southern forests >80% of fires are anthropogenically caused; Kovacs et al 2004). This is also true for our study site because of the remoteness of the area, as well as the low population density in general; most of the fires are of natural origin. Similar observations (FRI increase since the Little Ice Age) were made for the northern boundary of the larch forests (site 3 in figure 1; Kharuk et al 2011). All these data support the suggestion that observed climate change will lead to an increase in fire frequency (Gillett et al 2004, Bergeron et al 2004, Girardin et al 2009. Comparison of FRI along the meridian (figure 1) showed a northward increase of FRI: 77 ± 20 yr at ∼61 • N (site 1), 82 ± 7 at 64 • N (site 2), 200 ± 51 yr at about 66 • N, and 320 ± 50 yr at the northern boundary of the larch dominant communities (71 • N, site 3). The main reason for the northward FRI increase is less incoming solar radiation and, consequently, shortening of the fire-danger period. (3) mean tree ring widths before and after fire; and (2) tree ring width for a specimen with an extremely high post-fire growth increment (inset). Data were compiled on the basis of fires in the 20th century (figure 2). Dates of fires were set to the 'zero' point. Note that post-fire growth increase has a lag of about 5-7 yr.

Burns as a 'simulator' of future warming
Areas experiencing wildfires may be considered as a simulator of predicted warming impacts on northern forests. Indeed, in addition to soil enrichment with nutrients and decreased competition, fires cause an increase of permafrost thawing depth by a factor of 3-5 (Kharuk et al 2008), and increased soil drainage. Trees that survived wildfires showed an approximately double (1.93; p > 0.95) increase of the radial increment (figure 5). Comparison was made for the period for 25 yr before and after the fire (periods of ±5 yr around zero were deleted to exclude the effects of direct fire damage on trees). Some trees showed an extremely high response to fire effects. For example, the tree cross section shown in figure 5 showed an increase of radial growth by about a factor of 10 in comparison with background observations. This tree was sampled at a latitude near the Polar Circle.
Generally speaking, growth increases following fire scars should be measured along radii very much distant from the wound itself. But in our case we compare 37 specimens with the same pattern of fire damage, one of which shows outstanding increment growth (about ten times that of the background set; figure 5). The basic difference of this specimen from the others sampled was in the depth of soil thawing (about 1.5 m versus <0.3-0.5 m for the other specimens), and good drainage, since that tree was growing on the southern river bank. It is known that larch prefer drained soils (Schepaschenko et al 2008). The observed radial growth increase (and, consequently, increased carbon sequestration) may be an alternative to the scenario of forests in climateinduced permafrost transformation areas becoming greenhouse gas sources (IPCC 2007). Thus, the vegetation dynamics and productivity of the burned areas deserve future investigation.

Indirect signs of wildfires
The tree ring growth history showed periods of radial growth accelerations (i.e., tree ring width increase; figures 2-4). These increases are commonly considered to be climate driven (e.g., Shiyatov 2003) and may also contain information on fire events. The observed growth accelerations may also be caused by the above-mentioned fire-caused soil melioration. Differentiation of these effects could be carried out on the basis of the comparison with air temperature anomalies. The fireinduced origin of the acceleration is supported by the fact that in some cases the date of accelerations coincides with the burn marks on the trees from the same test site (figure 2). These effects deserve future investigation based on larger sample sizes.

Conclusions
FRI in the study area is considerably longer than in southerly territories. Along the 100 • meridian (figure 1) FRI values increased northward, 77 ± 20 yr at ∼61 • N (site 1), 82 ± 7 yr at 64 • N (site 2), 200 ± 51 yr at about 66 • N, and 320 ± 50 yr at the northern boundary of the larch stands (∼71 • N). The number of fire events during the Little Ice Age period (17th-18th centuries) was approximately half the number observed in the 19th-20th centuries. Fire-caused soil melioration (basically soil drainage and thawing depth increases) caused a radial growth increase to about twice the background value (up to >6 times at extremes). This effect may simulate the predicted impact of warming on the larch growth in the permafrost zone.